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
  32 #include "stackglue.h"
  33
  34 struct o2dlm_private {
  35         struct dlm_eviction_cb op_eviction_cb;
  36 };
  37
  38 static struct ocfs2_stack_plugin o2cb_stack;
  39
  40 /* These should be identical */
  41 #if (DLM_LOCK_IV != LKM_IVMODE)
  42 # error Lock modes do not match
  43 #endif
  44 #if (DLM_LOCK_NL != LKM_NLMODE)
  45 # error Lock modes do not match
  46 #endif
  47 #if (DLM_LOCK_CR != LKM_CRMODE)
  48 # error Lock modes do not match
  49 #endif
  50 #if (DLM_LOCK_CW != LKM_CWMODE)
  51 # error Lock modes do not match
  52 #endif
  53 #if (DLM_LOCK_PR != LKM_PRMODE)
  54 # error Lock modes do not match
  55 #endif
  56 #if (DLM_LOCK_PW != LKM_PWMODE)
  57 # error Lock modes do not match
  58 #endif
  59 #if (DLM_LOCK_EX != LKM_EXMODE)
  60 # error Lock modes do not match
  61 #endif
  62 static inline int mode_to_o2dlm(int mode)
  63 {
  64         BUG_ON(mode > LKM_MAXMODE);
  65
  66         return mode;
  67 }
  68
  69 #define map_flag(_generic, _o2dlm)              \
  70         if (flags & (_generic)) {               \
  71                 flags &= ~(_generic);           \
  72                 o2dlm_flags |= (_o2dlm);        \
  73         }
  74 static int flags_to_o2dlm(u32 flags)
  75 {
  76         int o2dlm_flags = 0;
  77
  78         map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE);
  79         map_flag(DLM_LKF_CANCEL, LKM_CANCEL);
  80         map_flag(DLM_LKF_CONVERT, LKM_CONVERT);
  81         map_flag(DLM_LKF_VALBLK, LKM_VALBLK);
  82         map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK);
  83         map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN);
  84         map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE);
  85         map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT);
  86         map_flag(DLM_LKF_LOCAL, LKM_LOCAL);
  87
  88         /* map_flag() should have cleared every flag passed in */
  89         BUG_ON(flags != 0);
  90
  91         return o2dlm_flags;
  92 }
  93 #undef map_flag
  94
  95 /*
  96  * Map an o2dlm status to standard errno values.
  97  *
  98  * o2dlm only uses a handful of these, and returns even fewer to the
  99  * caller. Still, we try to assign sane values to each error.
 100  *
 101  * The following value pairs have special meanings to dlmglue, thus
 102  * the right hand side needs to stay unique - never duplicate the
 103  * mapping elsewhere in the table!
 104  *
 105  * DLM_NORMAL:          0
 106  * DLM_NOTQUEUED:       -EAGAIN
 107  * DLM_CANCELGRANT:     -EBUSY
 108  * DLM_CANCEL:          -DLM_ECANCEL
 109  */
 110 /* Keep in sync with dlmapi.h */
 111 static int status_map[] = {
 112         [DLM_NORMAL]                    = 0,            /* Success */
 113         [DLM_GRANTED]                   = -EINVAL,
 114         [DLM_DENIED]                    = -EACCES,
 115         [DLM_DENIED_NOLOCKS]            = -EACCES,
 116         [DLM_WORKING]                   = -EACCES,
 117         [DLM_BLOCKED]                   = -EINVAL,
 118         [DLM_BLOCKED_ORPHAN]            = -EINVAL,
 119         [DLM_DENIED_GRACE_PERIOD]       = -EACCES,
 120         [DLM_SYSERR]                    = -ENOMEM,      /* It is what it is */
 121         [DLM_NOSUPPORT]                 = -EPROTO,
 122         [DLM_CANCELGRANT]               = -EBUSY,       /* Cancel after grant */
 123         [DLM_IVLOCKID]                  = -EINVAL,
 124         [DLM_SYNC]                      = -EINVAL,
 125         [DLM_BADTYPE]                   = -EINVAL,
 126         [DLM_BADRESOURCE]               = -EINVAL,
 127         [DLM_MAXHANDLES]                = -ENOMEM,
 128         [DLM_NOCLINFO]                  = -EINVAL,
 129         [DLM_NOLOCKMGR]                 = -EINVAL,
 130         [DLM_NOPURGED]                  = -EINVAL,
 131         [DLM_BADARGS]                   = -EINVAL,
 132         [DLM_VOID]                      = -EINVAL,
 133         [DLM_NOTQUEUED]                 = -EAGAIN,      /* Trylock failed */
 134         [DLM_IVBUFLEN]                  = -EINVAL,
 135         [DLM_CVTUNGRANT]                = -EPERM,
 136         [DLM_BADPARAM]                  = -EINVAL,
 137         [DLM_VALNOTVALID]               = -EINVAL,
 138         [DLM_REJECTED]                  = -EPERM,
 139         [DLM_ABORT]                     = -EINVAL,
 140         [DLM_CANCEL]                    = -DLM_ECANCEL, /* Successful cancel */
 141         [DLM_IVRESHANDLE]               = -EINVAL,
 142         [DLM_DEADLOCK]                  = -EDEADLK,
 143         [DLM_DENIED_NOASTS]             = -EINVAL,
 144         [DLM_FORWARD]                   = -EINVAL,
 145         [DLM_TIMEOUT]                   = -ETIMEDOUT,
 146         [DLM_IVGROUPID]                 = -EINVAL,
 147         [DLM_VERS_CONFLICT]             = -EOPNOTSUPP,
 148         [DLM_BAD_DEVICE_PATH]           = -ENOENT,
 149         [DLM_NO_DEVICE_PERMISSION]      = -EPERM,
 150         [DLM_NO_CONTROL_DEVICE]         = -ENOENT,
 151         [DLM_RECOVERING]                = -ENOTCONN,
 152         [DLM_MIGRATING]                 = -ERESTART,
 153         [DLM_MAXSTATS]                  = -EINVAL,
 154 };
 155
 156 static int dlm_status_to_errno(enum dlm_status status)
 157 {
 158         BUG_ON(status < 0 || status >= ARRAY_SIZE(status_map));
 159
 160         return status_map[status];
 161 }
 162
 163 static void o2dlm_lock_ast_wrapper(void *astarg)
 164 {
 165         struct ocfs2_dlm_lksb *lksb = astarg;
 166
 167         lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
 168 }
 169
 170 static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
 171 {
 172         struct ocfs2_dlm_lksb *lksb = astarg;
 173
 174         lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
 175 }
 176
 177 static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
 178 {
 179         struct ocfs2_dlm_lksb *lksb = astarg;
 180         int error = dlm_status_to_errno(status);
 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         lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, error);
 197 }
 198
 199 static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
 200                          int mode,
 201                          struct ocfs2_dlm_lksb *lksb,
 202                          u32 flags,
 203                          void *name,
 204                          unsigned int namelen)
 205 {
 206         enum dlm_status status;
 207         int o2dlm_mode = mode_to_o2dlm(mode);
 208         int o2dlm_flags = flags_to_o2dlm(flags);
 209         int ret;
 210
 211         status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
 212                          o2dlm_flags, name, namelen,
 213                          o2dlm_lock_ast_wrapper, lksb,
 214                          o2dlm_blocking_ast_wrapper);
 215         ret = dlm_status_to_errno(status);
 216         return ret;
 217 }
 218
 219 static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn,
 220                            struct ocfs2_dlm_lksb *lksb,
 221                            u32 flags)
 222 {
 223         enum dlm_status status;
 224         int o2dlm_flags = flags_to_o2dlm(flags);
 225         int ret;
 226
 227         status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm,
 228                            o2dlm_flags, o2dlm_unlock_ast_wrapper, lksb);
 229         ret = dlm_status_to_errno(status);
 230         return ret;
 231 }
 232
 233 static int o2cb_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
 234 {
 235         return dlm_status_to_errno(lksb->lksb_o2dlm.status);
 236 }
 237
 238 /*
 239  * o2dlm aways has a "valid" LVB. If the dlm loses track of the LVB
 240  * contents, it will zero out the LVB.  Thus the caller can always trust
 241  * the contents.
 242  */
 243 static int o2cb_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
 244 {
 245         return 1;
 246 }
 247
 248 static void *o2cb_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
 249 {
 250         return (void *)(lksb->lksb_o2dlm.lvb);
 251 }
 252
 253 static void o2cb_dump_lksb(struct ocfs2_dlm_lksb *lksb)
 254 {
 255         dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
 256 }
 257
 258 /*
 259  * Called from the dlm when it's about to evict a node. This is how the
 260  * classic stack signals node death.
 261  */
 262 static void o2dlm_eviction_cb(int node_num, void *data)
 263 {
 264         struct ocfs2_cluster_connection *conn = data;
 265
 266         mlog(ML_NOTICE, "o2dlm has evicted node %d from group %.*s\n",
 267              node_num, conn->cc_namelen, conn->cc_name);
 268
 269         conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
 270 }
 271
 272 static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
 273 {
 274         int rc = 0;
 275         u32 dlm_key;
 276         struct dlm_ctxt *dlm;
 277         struct o2dlm_private *priv;
 278         struct dlm_protocol_version fs_version;
 279
 280         BUG_ON(conn == NULL);
 281         BUG_ON(conn->cc_proto == NULL);
 282
 283         /* for now we only have one cluster/node, make sure we see it
 284          * in the heartbeat universe */
 285         if (!o2hb_check_local_node_heartbeating()) {
 286                 rc = -EINVAL;
 287                 goto out;
 288         }
 289
 290         priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
 291         if (!priv) {
 292                 rc = -ENOMEM;
 293                 goto out_free;
 294         }
 295
 296         /* This just fills the structure in.  It is safe to pass conn. */
 297         dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
 298                               conn);
 299
 300         conn->cc_private = priv;
 301
 302         /* used by the dlm code to make message headers unique, each
 303          * node in this domain must agree on this. */
 304         dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
 305         fs_version.pv_major = conn->cc_version.pv_major;
 306         fs_version.pv_minor = conn->cc_version.pv_minor;
 307
 308         dlm = dlm_register_domain(conn->cc_name, dlm_key, &fs_version);
 309         if (IS_ERR(dlm)) {
 310                 rc = PTR_ERR(dlm);
 311                 mlog_errno(rc);
 312                 goto out_free;
 313         }
 314
 315         conn->cc_version.pv_major = fs_version.pv_major;
 316         conn->cc_version.pv_minor = fs_version.pv_minor;
 317         conn->cc_lockspace = dlm;
 318
 319         dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
 320
 321 out_free:
 322         if (rc && conn->cc_private)
 323                 kfree(conn->cc_private);
 324
 325 out:
 326         return rc;
 327 }
 328
 329 static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
 330 {
 331         struct dlm_ctxt *dlm = conn->cc_lockspace;
 332         struct o2dlm_private *priv = conn->cc_private;
 333
 334         dlm_unregister_eviction_cb(&priv->op_eviction_cb);
 335         conn->cc_private = NULL;
 336         kfree(priv);
 337
 338         dlm_unregister_domain(dlm);
 339         conn->cc_lockspace = NULL;
 340
 341         return 0;
 342 }
 343
 344 static int o2cb_cluster_this_node(unsigned int *node)
 345 {
 346         int node_num;
 347
 348         node_num = o2nm_this_node();
 349         if (node_num == O2NM_INVALID_NODE_NUM)
 350                 return -ENOENT;
 351
 352         if (node_num >= O2NM_MAX_NODES)
 353                 return -EOVERFLOW;
 354
 355         *node = node_num;
 356         return 0;
 357 }
 358
 359 static struct ocfs2_stack_operations o2cb_stack_ops = {
 360         .connect        = o2cb_cluster_connect,
 361         .disconnect     = o2cb_cluster_disconnect,
 362         .this_node      = o2cb_cluster_this_node,
 363         .dlm_lock       = o2cb_dlm_lock,
 364         .dlm_unlock     = o2cb_dlm_unlock,
 365         .lock_status    = o2cb_dlm_lock_status,
 366         .lvb_valid      = o2cb_dlm_lvb_valid,
 367         .lock_lvb       = o2cb_dlm_lvb,
 368         .dump_lksb      = o2cb_dump_lksb,
 369 };
 370
 371 static struct ocfs2_stack_plugin o2cb_stack = {
 372         .sp_name        = "o2cb",
 373         .sp_ops         = &o2cb_stack_ops,
 374         .sp_owner       = THIS_MODULE,
 375 };
 376
 377 static int __init o2cb_stack_init(void)
 378 {
 379         return ocfs2_stack_glue_register(&o2cb_stack);
 380 }
 381
 382 static void __exit o2cb_stack_exit(void)
 383 {
 384         ocfs2_stack_glue_unregister(&o2cb_stack);
 385 }
 386
 387 MODULE_AUTHOR("Oracle");
 388 MODULE_DESCRIPTION("ocfs2 driver for the classic o2cb stack");
 389 MODULE_LICENSE("GPL");
 390 module_init(o2cb_stack_init);
 391 module_exit(o2cb_stack_exit);