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/crc32.h>
  21 #include <linux/kmod.h>
  22
  23 /* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
  24 #include <linux/fs.h>
  25
  26 #include "cluster/masklog.h"
  27 #include "cluster/nodemanager.h"
  28 #include "cluster/heartbeat.h"
  29
  30 #include "stackglue.h"
  31
  32 struct o2dlm_private {
  33         struct dlm_eviction_cb op_eviction_cb;
  34 };
  35
  36 /* These should be identical */
  37 #if (DLM_LOCK_IV != LKM_IVMODE)
  38 # error Lock modes do not match
  39 #endif
  40 #if (DLM_LOCK_NL != LKM_NLMODE)
  41 # error Lock modes do not match
  42 #endif
  43 #if (DLM_LOCK_CR != LKM_CRMODE)
  44 # error Lock modes do not match
  45 #endif
  46 #if (DLM_LOCK_CW != LKM_CWMODE)
  47 # error Lock modes do not match
  48 #endif
  49 #if (DLM_LOCK_PR != LKM_PRMODE)
  50 # error Lock modes do not match
  51 #endif
  52 #if (DLM_LOCK_PW != LKM_PWMODE)
  53 # error Lock modes do not match
  54 #endif
  55 #if (DLM_LOCK_EX != LKM_EXMODE)
  56 # error Lock modes do not match
  57 #endif
  58 static inline int mode_to_o2dlm(int mode)
  59 {
  60         BUG_ON(mode > LKM_MAXMODE);
  61
  62         return mode;
  63 }
  64
  65 #define map_flag(_generic, _o2dlm)              \
  66         if (flags & (_generic)) {               \
  67                 flags &= ~(_generic);           \
  68                 o2dlm_flags |= (_o2dlm);        \
  69         }
  70 static int flags_to_o2dlm(u32 flags)
  71 {
  72         int o2dlm_flags = 0;
  73
  74         map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE);
  75         map_flag(DLM_LKF_CANCEL, LKM_CANCEL);
  76         map_flag(DLM_LKF_CONVERT, LKM_CONVERT);
  77         map_flag(DLM_LKF_VALBLK, LKM_VALBLK);
  78         map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK);
  79         map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN);
  80         map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE);
  81         map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT);
  82         map_flag(DLM_LKF_LOCAL, LKM_LOCAL);
  83
  84         /* map_flag() should have cleared every flag passed in */
  85         BUG_ON(flags != 0);
  86
  87         return o2dlm_flags;
  88 }
  89 #undef map_flag
  90
  91 /*
  92  * Map an o2dlm status to standard errno values.
  93  *
  94  * o2dlm only uses a handful of these, and returns even fewer to the
  95  * caller. Still, we try to assign sane values to each error.
  96  *
  97  * The following value pairs have special meanings to dlmglue, thus
  98  * the right hand side needs to stay unique - never duplicate the
  99  * mapping elsewhere in the table!
 100  *
 101  * DLM_NORMAL:          0
 102  * DLM_NOTQUEUED:       -EAGAIN
 103  * DLM_CANCELGRANT:     -EBUSY
 104  * DLM_CANCEL:          -DLM_ECANCEL
 105  */
 106 /* Keep in sync with dlmapi.h */
 107 static int status_map[] = {
 108         [DLM_NORMAL]                    = 0,            /* Success */
 109         [DLM_GRANTED]                   = -EINVAL,
 110         [DLM_DENIED]                    = -EACCES,
 111         [DLM_DENIED_NOLOCKS]            = -EACCES,
 112         [DLM_WORKING]                   = -EACCES,
 113         [DLM_BLOCKED]                   = -EINVAL,
 114         [DLM_BLOCKED_ORPHAN]            = -EINVAL,
 115         [DLM_DENIED_GRACE_PERIOD]       = -EACCES,
 116         [DLM_SYSERR]                    = -ENOMEM,      /* It is what it is */
 117         [DLM_NOSUPPORT]                 = -EPROTO,
 118         [DLM_CANCELGRANT]               = -EBUSY,       /* Cancel after grant */
 119         [DLM_IVLOCKID]                  = -EINVAL,
 120         [DLM_SYNC]                      = -EINVAL,
 121         [DLM_BADTYPE]                   = -EINVAL,
 122         [DLM_BADRESOURCE]               = -EINVAL,
 123         [DLM_MAXHANDLES]                = -ENOMEM,
 124         [DLM_NOCLINFO]                  = -EINVAL,
 125         [DLM_NOLOCKMGR]                 = -EINVAL,
 126         [DLM_NOPURGED]                  = -EINVAL,
 127         [DLM_BADARGS]                   = -EINVAL,
 128         [DLM_VOID]                      = -EINVAL,
 129         [DLM_NOTQUEUED]                 = -EAGAIN,      /* Trylock failed */
 130         [DLM_IVBUFLEN]                  = -EINVAL,
 131         [DLM_CVTUNGRANT]                = -EPERM,
 132         [DLM_BADPARAM]                  = -EINVAL,
 133         [DLM_VALNOTVALID]               = -EINVAL,
 134         [DLM_REJECTED]                  = -EPERM,
 135         [DLM_ABORT]                     = -EINVAL,
 136         [DLM_CANCEL]                    = -DLM_ECANCEL, /* Successful cancel */
 137         [DLM_IVRESHANDLE]               = -EINVAL,
 138         [DLM_DEADLOCK]                  = -EDEADLK,
 139         [DLM_DENIED_NOASTS]             = -EINVAL,
 140         [DLM_FORWARD]                   = -EINVAL,
 141         [DLM_TIMEOUT]                   = -ETIMEDOUT,
 142         [DLM_IVGROUPID]                 = -EINVAL,
 143         [DLM_VERS_CONFLICT]             = -EOPNOTSUPP,
 144         [DLM_BAD_DEVICE_PATH]           = -ENOENT,
 145         [DLM_NO_DEVICE_PERMISSION]      = -EPERM,
 146         [DLM_NO_CONTROL_DEVICE]         = -ENOENT,
 147         [DLM_RECOVERING]                = -ENOTCONN,
 148         [DLM_MIGRATING]                 = -ERESTART,
 149         [DLM_MAXSTATS]                  = -EINVAL,
 150 };
 151
 152 static int dlm_status_to_errno(enum dlm_status status)
 153 {
 154         BUG_ON(status > (sizeof(status_map) / sizeof(status_map[0])));
 155
 156         return status_map[status];
 157 }
 158
 159 static void o2dlm_lock_ast_wrapper(void *astarg)
 160 {
 161         BUG_ON(stack_glue_lproto == NULL);
 162
 163         stack_glue_lproto->lp_lock_ast(astarg);
 164 }
 165
 166 static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
 167 {
 168         BUG_ON(stack_glue_lproto == NULL);
 169
 170         stack_glue_lproto->lp_blocking_ast(astarg, level);
 171 }
 172
 173 static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
 174 {
 175         int error = dlm_status_to_errno(status);
 176
 177         BUG_ON(stack_glue_lproto == NULL);
 178
 179         /*
 180          * In o2dlm, you can get both the lock_ast() for the lock being
 181          * granted and the unlock_ast() for the CANCEL failing.  A
 182          * successful cancel sends DLM_NORMAL here.  If the
 183          * lock grant happened before the cancel arrived, you get
 184          * DLM_CANCELGRANT.
 185          *
 186          * There's no need for the double-ast.  If we see DLM_CANCELGRANT,
 187          * we just ignore it.  We expect the lock_ast() to handle the
 188          * granted lock.
 189          */
 190         if (status == DLM_CANCELGRANT)
 191                 return;
 192
 193         stack_glue_lproto->lp_unlock_ast(astarg, error);
 194 }
 195
 196 static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
 197                          int mode,
 198                          union ocfs2_dlm_lksb *lksb,
 199                          u32 flags,
 200                          void *name,
 201                          unsigned int namelen,
 202                          void *astarg)
 203 {
 204         enum dlm_status status;
 205         int o2dlm_mode = mode_to_o2dlm(mode);
 206         int o2dlm_flags = flags_to_o2dlm(flags);
 207         int ret;
 208
 209         status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
 210                          o2dlm_flags, name, namelen,
 211                          o2dlm_lock_ast_wrapper, astarg,
 212                          o2dlm_blocking_ast_wrapper);
 213         ret = dlm_status_to_errno(status);
 214         return ret;
 215 }
 216
 217 static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn,
 218                            union ocfs2_dlm_lksb *lksb,
 219                            u32 flags,
 220                            void *astarg)
 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, astarg);
 228         ret = dlm_status_to_errno(status);
 229         return ret;
 230 }
 231
 232 static int o2cb_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
 233 {
 234         return dlm_status_to_errno(lksb->lksb_o2dlm.status);
 235 }
 236
 237 static void *o2cb_dlm_lvb(union ocfs2_dlm_lksb *lksb)
 238 {
 239         return (void *)(lksb->lksb_o2dlm.lvb);
 240 }
 241
 242 static void o2cb_dump_lksb(union ocfs2_dlm_lksb *lksb)
 243 {
 244         dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
 245 }
 246
 247 /*
 248  * Called from the dlm when it's about to evict a node. This is how the
 249  * classic stack signals node death.
 250  */
 251 static void o2dlm_eviction_cb(int node_num, void *data)
 252 {
 253         struct ocfs2_cluster_connection *conn = data;
 254
 255         mlog(ML_NOTICE, "o2dlm has evicted node %d from group %.*s\n",
 256              node_num, conn->cc_namelen, conn->cc_name);
 257
 258         conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
 259 }
 260
 261 static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
 262 {
 263         int rc = 0;
 264         u32 dlm_key;
 265         struct dlm_ctxt *dlm;
 266         struct o2dlm_private *priv;
 267         struct dlm_protocol_version dlm_version;
 268
 269         BUG_ON(conn == NULL);
 270
 271         /* for now we only have one cluster/node, make sure we see it
 272          * in the heartbeat universe */
 273         if (!o2hb_check_local_node_heartbeating()) {
 274                 rc = -EINVAL;
 275                 goto out;
 276         }
 277
 278         priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
 279         if (!priv) {
 280                 rc = -ENOMEM;
 281                 goto out_free;
 282         }
 283
 284         /* This just fills the structure in.  It is safe to pass conn. */
 285         dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
 286                               conn);
 287
 288         conn->cc_private = priv;
 289
 290         /* used by the dlm code to make message headers unique, each
 291          * node in this domain must agree on this. */
 292         dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
 293         dlm_version.pv_major = conn->cc_version.pv_major;
 294         dlm_version.pv_minor = conn->cc_version.pv_minor;
 295
 296         dlm = dlm_register_domain(conn->cc_name, dlm_key, &dlm_version);
 297         if (IS_ERR(dlm)) {
 298                 rc = PTR_ERR(dlm);
 299                 mlog_errno(rc);
 300                 goto out_free;
 301         }
 302
 303         conn->cc_version.pv_major = dlm_version.pv_major;
 304         conn->cc_version.pv_minor = dlm_version.pv_minor;
 305         conn->cc_lockspace = dlm;
 306
 307         dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
 308
 309 out_free:
 310         if (rc && conn->cc_private)
 311                 kfree(conn->cc_private);
 312
 313 out:
 314         return rc;
 315 }
 316
 317 static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
 318 {
 319         struct dlm_ctxt *dlm = conn->cc_lockspace;
 320         struct o2dlm_private *priv = conn->cc_private;
 321
 322         dlm_unregister_eviction_cb(&priv->op_eviction_cb);
 323         conn->cc_private = NULL;
 324         kfree(priv);
 325
 326         dlm_unregister_domain(dlm);
 327         conn->cc_lockspace = NULL;
 328
 329         return 0;
 330 }
 331
 332 static void o2hb_stop(const char *group)
 333 {
 334         int ret;
 335         char *argv[5], *envp[3];
 336
 337         argv[0] = (char *)o2nm_get_hb_ctl_path();
 338         argv[1] = "-K";
 339         argv[2] = "-u";
 340         argv[3] = (char *)group;
 341         argv[4] = NULL;
 342
 343         mlog(0, "Run: %s %s %s %s\n", argv[0], argv[1], argv[2], argv[3]);
 344
 345         /* minimal command environment taken from cpu_run_sbin_hotplug */
 346         envp[0] = "HOME=/";
 347         envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
 348         envp[2] = NULL;
 349
 350         ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
 351         if (ret < 0)
 352                 mlog_errno(ret);
 353 }
 354
 355 /*
 356  * Hangup is a hack for tools compatibility.  Older ocfs2-tools software
 357  * expects the filesystem to call "ocfs2_hb_ctl" during unmount.  This
 358  * happens regardless of whether the DLM got started, so we can't do it
 359  * in ocfs2_cluster_disconnect().  We bring the o2hb_stop() function into
 360  * the glue and provide a "hangup" API for super.c to call.
 361  *
 362  * Other stacks will eventually provide a NULL ->hangup() pointer.
 363  */
 364 static void o2cb_cluster_hangup(const char *group, int grouplen)
 365 {
 366         o2hb_stop(group);
 367 }
 368
 369 static int o2cb_cluster_this_node(unsigned int *node)
 370 {
 371         int node_num;
 372
 373         node_num = o2nm_this_node();
 374         if (node_num == O2NM_INVALID_NODE_NUM)
 375                 return -ENOENT;
 376
 377         if (node_num >= O2NM_MAX_NODES)
 378                 return -EOVERFLOW;
 379
 380         *node = node_num;
 381         return 0;
 382 }
 383
 384 struct ocfs2_stack_operations o2cb_stack_ops = {
 385         .connect        = o2cb_cluster_connect,
 386         .disconnect     = o2cb_cluster_disconnect,
 387         .hangup         = o2cb_cluster_hangup,
 388         .this_node      = o2cb_cluster_this_node,
 389         .dlm_lock       = o2cb_dlm_lock,
 390         .dlm_unlock     = o2cb_dlm_unlock,
 391         .lock_status    = o2cb_dlm_lock_status,
 392         .lock_lvb       = o2cb_dlm_lvb,
 393         .dump_lksb      = o2cb_dump_lksb,
 394 };
 395