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         struct ocfs2_dlm_lksb *lksb = astarg;
 165
 166         lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
 167 }
 168
 169 static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
 170 {
 171         struct ocfs2_dlm_lksb *lksb = astarg;
 172
 173         lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
 174 }
 175
 176 static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
 177 {
 178         struct ocfs2_dlm_lksb *lksb = astarg;
 179         int error = dlm_status_to_errno(status);
 180
 181         /*
 182          * In o2dlm, you can get both the lock_ast() for the lock being
 183          * granted and the unlock_ast() for the CANCEL failing.  A
 184          * successful cancel sends DLM_NORMAL here.  If the
 185          * lock grant happened before the cancel arrived, you get
 186          * DLM_CANCELGRANT.
 187          *
 188          * There's no need for the double-ast.  If we see DLM_CANCELGRANT,
 189          * we just ignore it.  We expect the lock_ast() to handle the
 190          * granted lock.
 191          */
 192         if (status == DLM_CANCELGRANT)
 193                 return;
 194
 195         lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, error);
 196 }
 197
 198 static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
 199                          int mode,
 200                          struct ocfs2_dlm_lksb *lksb,
 201                          u32 flags,
 202                          void *name,
 203                          unsigned int namelen)
 204 {
 205         enum dlm_status status;
 206         int o2dlm_mode = mode_to_o2dlm(mode);
 207         int o2dlm_flags = flags_to_o2dlm(flags);
 208         int ret;
 209
 210         status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
 211                          o2dlm_flags, name, namelen,
 212                          o2dlm_lock_ast_wrapper, lksb,
 213                          o2dlm_blocking_ast_wrapper);
 214         ret = dlm_status_to_errno(status);
 215         return ret;
 216 }
 217
 218 static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn,
 219                            struct ocfs2_dlm_lksb *lksb,
 220                            u32 flags)
 221 {
 222         enum dlm_status status;
 223         int o2dlm_flags = flags_to_o2dlm(flags);
 224         int ret;
 225
 226         status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm,
 227                            o2dlm_flags, o2dlm_unlock_ast_wrapper, lksb);
 228         ret = dlm_status_to_errno(status);
 229         return ret;
 230 }
 231
 232 static int o2cb_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
 233 {
 234         return dlm_status_to_errno(lksb->lksb_o2dlm.status);
 235 }
 236
 237 /*
 238  * o2dlm aways has a "valid" LVB. If the dlm loses track of the LVB
 239  * contents, it will zero out the LVB.  Thus the caller can always trust
 240  * the contents.
 241  */
 242 static int o2cb_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
 243 {
 244         return 1;
 245 }
 246
 247 static void *o2cb_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
 248 {
 249         return (void *)(lksb->lksb_o2dlm.lvb);
 250 }
 251
 252 static void o2cb_dump_lksb(struct ocfs2_dlm_lksb *lksb)
 253 {
 254         dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
 255 }
 256
 257 /*
 258  * Called from the dlm when it's about to evict a node. This is how the
 259  * classic stack signals node death.
 260  */
 261 static void o2dlm_eviction_cb(int node_num, void *data)
 262 {
 263         struct ocfs2_cluster_connection *conn = data;
 264
 265         mlog(ML_NOTICE, "o2dlm has evicted node %d from group %.*s\n",
 266              node_num, conn->cc_namelen, conn->cc_name);
 267
 268         conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
 269 }
 270
 271 static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
 272 {
 273         int rc = 0;
 274         u32 dlm_key;
 275         struct dlm_ctxt *dlm;
 276         struct o2dlm_private *priv;
 277         struct dlm_protocol_version fs_version;
 278
 279         BUG_ON(conn == NULL);
 280         BUG_ON(conn->cc_proto == NULL);
 281
 282         /* for now we only have one cluster/node, make sure we see it
 283          * in the heartbeat universe */
 284         if (!o2hb_check_local_node_heartbeating()) {
 285                 rc = -EINVAL;
 286                 goto out;
 287         }
 288
 289         priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
 290         if (!priv) {
 291                 rc = -ENOMEM;
 292                 goto out_free;
 293         }
 294
 295         /* This just fills the structure in.  It is safe to pass conn. */
 296         dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
 297                               conn);
 298
 299         conn->cc_private = priv;
 300
 301         /* used by the dlm code to make message headers unique, each
 302          * node in this domain must agree on this. */
 303         dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
 304         fs_version.pv_major = conn->cc_version.pv_major;
 305         fs_version.pv_minor = conn->cc_version.pv_minor;
 306
 307         dlm = dlm_register_domain(conn->cc_name, dlm_key, &fs_version);
 308         if (IS_ERR(dlm)) {
 309                 rc = PTR_ERR(dlm);
 310                 mlog_errno(rc);
 311                 goto out_free;
 312         }
 313
 314         conn->cc_version.pv_major = fs_version.pv_major;
 315         conn->cc_version.pv_minor = fs_version.pv_minor;
 316         conn->cc_lockspace = dlm;
 317
 318         dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
 319
 320 out_free:
 321         if (rc && conn->cc_private)
 322                 kfree(conn->cc_private);
 323
 324 out:
 325         return rc;
 326 }
 327
 328 static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
 329 {
 330         struct dlm_ctxt *dlm = conn->cc_lockspace;
 331         struct o2dlm_private *priv = conn->cc_private;
 332
 333         dlm_unregister_eviction_cb(&priv->op_eviction_cb);
 334         conn->cc_private = NULL;
 335         kfree(priv);
 336
 337         dlm_unregister_domain(dlm);
 338         conn->cc_lockspace = NULL;
 339
 340         return 0;
 341 }
 342
 343 static int o2cb_cluster_this_node(unsigned int *node)
 344 {
 345         int node_num;
 346
 347         node_num = o2nm_this_node();
 348         if (node_num == O2NM_INVALID_NODE_NUM)
 349                 return -ENOENT;
 350
 351         if (node_num >= O2NM_MAX_NODES)
 352                 return -EOVERFLOW;
 353
 354         *node = node_num;
 355         return 0;
 356 }
 357
 358 static struct ocfs2_stack_operations o2cb_stack_ops = {
 359         .connect        = o2cb_cluster_connect,
 360         .disconnect     = o2cb_cluster_disconnect,
 361         .this_node      = o2cb_cluster_this_node,
 362         .dlm_lock       = o2cb_dlm_lock,
 363         .dlm_unlock     = o2cb_dlm_unlock,
 364         .lock_status    = o2cb_dlm_lock_status,
 365         .lvb_valid      = o2cb_dlm_lvb_valid,
 366         .lock_lvb       = o2cb_dlm_lvb,
 367         .dump_lksb      = o2cb_dump_lksb,
 368 };
 369
 370 static struct ocfs2_stack_plugin o2cb_stack = {
 371         .sp_name        = "o2cb",
 372         .sp_ops         = &o2cb_stack_ops,
 373         .sp_owner       = THIS_MODULE,
 374 };
 375
 376 static int __init o2cb_stack_init(void)
 377 {
 378         return ocfs2_stack_glue_register(&o2cb_stack);
 379 }
 380
 381 static void __exit o2cb_stack_exit(void)
 382 {
 383         ocfs2_stack_glue_unregister(&o2cb_stack);
 384 }
 385
 386 MODULE_AUTHOR("Oracle");
 387 MODULE_DESCRIPTION("ocfs2 driver for the classic o2cb stack");
 388 MODULE_LICENSE("GPL");
 389 module_init(o2cb_stack_init);
 390 module_exit(o2cb_stack_exit);