LXR linux/fs/dlm/dlm

   1/******************************************************************************
   2*******************************************************************************
   3**
   4**  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   5**  Copyright (C) 2004-2008 Red Hat, Inc.  All rights reserved.
   6**
   7**  This copyrighted material is made available to anyone wishing to use,
   8**  modify, copy, or redistribute it subject to the terms and conditions
   9**  of the GNU General Public License v.2.
  10**
  11*******************************************************************************
  12******************************************************************************/
  13
  14#ifndef __DLM_INTERNAL_DOT_H__
  15#define __DLM_INTERNAL_DOT_H__
  16
  17/*
  18 * This is the main header file to be included in each DLM source file.
  19 */
  20
  21#include <linux/module.h>
  22#include <linux/slab.h>
  23#include <linux/sched.h>
  24#include <linux/types.h>
  25#include <linux/ctype.h>
  26#include <linux/spinlock.h>
  27#include <linux/vmalloc.h>
  28#include <linux/list.h>
  29#include <linux/errno.h>
  30#include <linux/random.h>
  31#include <linux/delay.h>
  32#include <linux/socket.h>
  33#include <linux/kthread.h>
  34#include <linux/kobject.h>
  35#include <linux/kref.h>
  36#include <linux/kernel.h>
  37#include <linux/jhash.h>
  38#include <linux/miscdevice.h>
  39#include <linux/mutex.h>
  40#include <asm/uaccess.h>
  41
  42#include <linux/dlm.h>
  43#include "config.h"
  44
  45/* Size of the temp buffer midcomms allocates on the stack.
  46   We try to make this large enough so most messages fit.
  47   FIXME: should sctp make this unnecessary? */
  48
  49#define DLM_INBUF_LEN           148
  50
  51struct dlm_ls;
  52struct dlm_lkb;
  53struct dlm_rsb;
  54struct dlm_member;
  55struct dlm_lkbtable;
  56struct dlm_rsbtable;
  57struct dlm_dirtable;
  58struct dlm_direntry;
  59struct dlm_recover;
  60struct dlm_header;
  61struct dlm_message;
  62struct dlm_rcom;
  63struct dlm_mhandle;
  64
  65#define log_print(fmt, args...) \
  66        printk(KERN_ERR "dlm: "fmt"\n" , ##args)
  67#define log_error(ls, fmt, args...) \
  68        printk(KERN_ERR "dlm: %s: " fmt "\n", (ls)->ls_name , ##args)
  69
  70#define log_debug(ls, fmt, args...) \
  71do { \
  72        if (dlm_config.ci_log_debug) \
  73                printk(KERN_DEBUG "dlm: %s: " fmt "\n", \
  74                       (ls)->ls_name , ##args); \
  75} while (0)
  76
  77#define DLM_ASSERT(x, do) \
  78{ \
  79  if (!(x)) \
  80  { \
  81    printk(KERN_ERR "\nDLM:  Assertion failed on line %d of file %s\n" \
  82               "DLM:  assertion:  \"%s\"\n" \
  83               "DLM:  time = %lu\n", \
  84               __LINE__, __FILE__, #x, jiffies); \
  85    {do} \
  86    printk("\n"); \
  87    BUG(); \
  88    panic("DLM:  Record message above and reboot.\n"); \
  89  } \
  90}
  91
  92
  93struct dlm_direntry {
  94        struct list_head        list;
  95        uint32_t                master_nodeid;
  96        uint16_t                length;
  97        char                    name[1];
  98};
  99
 100struct dlm_dirtable {
 101        struct list_head        list;
 102        spinlock_t              lock;
 103};
 104
 105struct dlm_rsbtable {
 106        struct list_head        list;
 107        struct list_head        toss;
 108        spinlock_t              lock;
 109};
 110
 111struct dlm_lkbtable {
 112        struct list_head        list;
 113        rwlock_t                lock;
 114        uint16_t                counter;
 115};
 116
 117/*
 118 * Lockspace member (per node in a ls)
 119 */
 120
 121struct dlm_member {
 122        struct list_head        list;
 123        int                     nodeid;
 124        int                     weight;
 125};
 126
 127/*
 128 * Save and manage recovery state for a lockspace.
 129 */
 130
 131struct dlm_recover {
 132        struct list_head        list;
 133        int                     *nodeids;   /* nodeids of all members */
 134        int                     node_count;
 135        int                     *new;       /* nodeids of new members */
 136        int                     new_count;
 137        uint64_t                seq;
 138};
 139
 140/*
 141 * Pass input args to second stage locking function.
 142 */
 143
 144struct dlm_args {
 145        uint32_t                flags;
 146        void                    (*astfn) (void *astparam);
 147        void                    *astparam;
 148        void                    (*bastfn) (void *astparam, int mode);
 149        int                     mode;
 150        struct dlm_lksb         *lksb;
 151        unsigned long           timeout;
 152};
 153
 154
 155/*
 156 * Lock block
 157 *
 158 * A lock can be one of three types:
 159 *
 160 * local copy      lock is mastered locally
 161 *                 (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set)
 162 * process copy    lock is mastered on a remote node
 163 *                 (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set)
 164 * master copy     master node's copy of a lock owned by remote node
 165 *                 (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set)
 166 *
 167 * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or
 168 * dlm_unlock.  The dlm does not modify these or use any private flags in
 169 * this field; it only contains DLM_LKF_ flags from dlm.h.  These flags
 170 * are sent as-is to the remote master when the lock is remote.
 171 *
 172 * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h.
 173 * Some internal flags are shared between the master and process nodes;
 174 * these shared flags are kept in the lower two bytes.  One of these
 175 * flags set on the master copy will be propagated to the process copy
 176 * and v.v.  Other internal flags are private to the master or process
 177 * node (e.g. DLM_IFL_MSTCPY).  These are kept in the high two bytes.
 178 *
 179 * lkb_sbflags: status block flags.  These flags are copied directly into
 180 * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion
 181 * ast.  All defined in dlm.h with DLM_SBF_ prefix.
 182 *
 183 * lkb_status: the lock status indicates which rsb queue the lock is
 184 * on, grant, convert, or wait.  DLM_LKSTS_ WAITING/GRANTED/CONVERT
 185 *
 186 * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a
 187 * reply is needed.  Only set when the lkb is on the lockspace waiters
 188 * list awaiting a reply from a remote node.
 189 *
 190 * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb
 191 * is a master copy, nodeid specifies the remote lock holder, when the
 192 * lkb is a process copy, the nodeid specifies the lock master.
 193 */
 194
 195/* lkb_ast_type */
 196
 197#define AST_COMP                1
 198#define AST_BAST                2
 199
 200/* lkb_status */
 201
 202#define DLM_LKSTS_WAITING       1
 203#define DLM_LKSTS_GRANTED       2
 204#define DLM_LKSTS_CONVERT       3
 205
 206/* lkb_flags */
 207
 208#define DLM_IFL_MSTCPY          0x00010000
 209#define DLM_IFL_RESEND          0x00020000
 210#define DLM_IFL_DEAD            0x00040000
 211#define DLM_IFL_OVERLAP_UNLOCK  0x00080000
 212#define DLM_IFL_OVERLAP_CANCEL  0x00100000
 213#define DLM_IFL_ENDOFLIFE       0x00200000
 214#define DLM_IFL_WATCH_TIMEWARN  0x00400000
 215#define DLM_IFL_TIMEOUT_CANCEL  0x00800000
 216#define DLM_IFL_DEADLOCK_CANCEL 0x01000000
 217#define DLM_IFL_USER            0x00000001
 218#define DLM_IFL_ORPHAN          0x00000002
 219
 220struct dlm_lkb {
 221        struct dlm_rsb          *lkb_resource;  /* the rsb */
 222        struct kref             lkb_ref;
 223        int                     lkb_nodeid;     /* copied from rsb */
 224        int                     lkb_ownpid;     /* pid of lock owner */
 225        uint32_t                lkb_id;         /* our lock ID */
 226        uint32_t                lkb_remid;      /* lock ID on remote partner */
 227        uint32_t                lkb_exflags;    /* external flags from caller */
 228        uint32_t                lkb_sbflags;    /* lksb flags */
 229        uint32_t                lkb_flags;      /* internal flags */
 230        uint32_t                lkb_lvbseq;     /* lvb sequence number */
 231
 232        int8_t                  lkb_status;     /* granted, waiting, convert */
 233        int8_t                  lkb_rqmode;     /* requested lock mode */
 234        int8_t                  lkb_grmode;     /* granted lock mode */
 235        int8_t                  lkb_bastmode;   /* requested mode */
 236        int8_t                  lkb_highbast;   /* highest mode bast sent for */
 237        int8_t                  lkb_wait_type;  /* type of reply waiting for */
 238        int8_t                  lkb_wait_count;
 239        int8_t                  lkb_ast_type;   /* type of ast queued for */
 240
 241        struct list_head        lkb_idtbl_list; /* lockspace lkbtbl */
 242        struct list_head        lkb_statequeue; /* rsb g/c/w list */
 243        struct list_head        lkb_rsb_lookup; /* waiting for rsb lookup */
 244        struct list_head        lkb_wait_reply; /* waiting for remote reply */
 245        struct list_head        lkb_astqueue;   /* need ast to be sent */
 246        struct list_head        lkb_ownqueue;   /* list of locks for a process */
 247        struct list_head        lkb_time_list;
 248        ktime_t                 lkb_time_bast;  /* for debugging */
 249        ktime_t                 lkb_timestamp;
 250        unsigned long           lkb_timeout_cs;
 251
 252        char                    *lkb_lvbptr;
 253        struct dlm_lksb         *lkb_lksb;      /* caller's status block */
 254        void                    (*lkb_astfn) (void *astparam);
 255        void                    (*lkb_bastfn) (void *astparam, int mode);
 256        union {
 257                void                    *lkb_astparam;  /* caller's ast arg */
 258                struct dlm_user_args    *lkb_ua;
 259        };
 260};
 261
 262
 263struct dlm_rsb {
 264        struct dlm_ls           *res_ls;        /* the lockspace */
 265        struct kref             res_ref;
 266        struct mutex            res_mutex;
 267        unsigned long           res_flags;
 268        int                     res_length;     /* length of rsb name */
 269        int                     res_nodeid;
 270        uint32_t                res_lvbseq;
 271        uint32_t                res_hash;
 272        uint32_t                res_bucket;     /* rsbtbl */
 273        unsigned long           res_toss_time;
 274        uint32_t                res_first_lkid;
 275        struct list_head        res_lookup;     /* lkbs waiting on first */
 276        struct list_head        res_hashchain;  /* rsbtbl */
 277        struct list_head        res_grantqueue;
 278        struct list_head        res_convertqueue;
 279        struct list_head        res_waitqueue;
 280
 281        struct list_head        res_root_list;      /* used for recovery */
 282        struct list_head        res_recover_list;   /* used for recovery */
 283        int                     res_recover_locks_count;
 284
 285        char                    *res_lvbptr;
 286        char                    res_name[1];
 287};
 288
 289/* find_rsb() flags */
 290
 291#define R_MASTER                1       /* only return rsb if it's a master */
 292#define R_CREATE                2       /* create/add rsb if not found */
 293
 294/* rsb_flags */
 295
 296enum rsb_flags {
 297        RSB_MASTER_UNCERTAIN,
 298        RSB_VALNOTVALID,
 299        RSB_VALNOTVALID_PREV,
 300        RSB_NEW_MASTER,
 301        RSB_NEW_MASTER2,
 302        RSB_RECOVER_CONVERT,
 303        RSB_LOCKS_PURGED,
 304};
 305
 306static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag)
 307{
 308        __set_bit(flag, &r->res_flags);
 309}
 310
 311static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag)
 312{
 313        __clear_bit(flag, &r->res_flags);
 314}
 315
 316static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag)
 317{
 318        return test_bit(flag, &r->res_flags);
 319}
 320
 321
 322/* dlm_header is first element of all structs sent between nodes */
 323
 324#define DLM_HEADER_MAJOR        0x00030000
 325#define DLM_HEADER_MINOR        0x00000000
 326
 327#define DLM_MSG                 1
 328#define DLM_RCOM                2
 329
 330struct dlm_header {
 331        uint32_t                h_version;
 332        uint32_t                h_lockspace;
 333        uint32_t                h_nodeid;       /* nodeid of sender */
 334        uint16_t                h_length;
 335        uint8_t                 h_cmd;          /* DLM_MSG, DLM_RCOM */
 336        uint8_t                 h_pad;
 337};
 338
 339
 340#define DLM_MSG_REQUEST         1
 341#define DLM_MSG_CONVERT         2
 342#define DLM_MSG_UNLOCK          3
 343#define DLM_MSG_CANCEL          4
 344#define DLM_MSG_REQUEST_REPLY   5
 345#define DLM_MSG_CONVERT_REPLY   6
 346#define DLM_MSG_UNLOCK_REPLY    7
 347#define DLM_MSG_CANCEL_REPLY    8
 348#define DLM_MSG_GRANT           9
 349#define DLM_MSG_BAST            10
 350#define DLM_MSG_LOOKUP          11
 351#define DLM_MSG_REMOVE          12
 352#define DLM_MSG_LOOKUP_REPLY    13
 353#define DLM_MSG_PURGE           14
 354
 355struct dlm_message {
 356        struct dlm_header       m_header;
 357        uint32_t                m_type;         /* DLM_MSG_ */
 358        uint32_t                m_nodeid;
 359        uint32_t                m_pid;
 360        uint32_t                m_lkid;         /* lkid on sender */
 361        uint32_t                m_remid;        /* lkid on receiver */
 362        uint32_t                m_parent_lkid;
 363        uint32_t                m_parent_remid;
 364        uint32_t                m_exflags;
 365        uint32_t                m_sbflags;
 366        uint32_t                m_flags;
 367        uint32_t                m_lvbseq;
 368        uint32_t                m_hash;
 369        int                     m_status;
 370        int                     m_grmode;
 371        int                     m_rqmode;
 372        int                     m_bastmode;
 373        int                     m_asts;
 374        int                     m_result;       /* 0 or -EXXX */
 375        char                    m_extra[0];     /* name or lvb */
 376};
 377
 378
 379#define DLM_RS_NODES            0x00000001
 380#define DLM_RS_NODES_ALL        0x00000002
 381#define DLM_RS_DIR              0x00000004
 382#define DLM_RS_DIR_ALL          0x00000008
 383#define DLM_RS_LOCKS            0x00000010
 384#define DLM_RS_LOCKS_ALL        0x00000020
 385#define DLM_RS_DONE             0x00000040
 386#define DLM_RS_DONE_ALL         0x00000080
 387
 388#define DLM_RCOM_STATUS         1
 389#define DLM_RCOM_NAMES          2
 390#define DLM_RCOM_LOOKUP         3
 391#define DLM_RCOM_LOCK           4
 392#define DLM_RCOM_STATUS_REPLY   5
 393#define DLM_RCOM_NAMES_REPLY    6
 394#define DLM_RCOM_LOOKUP_REPLY   7
 395#define DLM_RCOM_LOCK_REPLY     8
 396
 397struct dlm_rcom {
 398        struct dlm_header       rc_header;
 399        uint32_t                rc_type;        /* DLM_RCOM_ */
 400        int                     rc_result;      /* multi-purpose */
 401        uint64_t                rc_id;          /* match reply with request */
 402        uint64_t                rc_seq;         /* sender's ls_recover_seq */
 403        uint64_t                rc_seq_reply;   /* remote ls_recover_seq */
 404        char                    rc_buf[0];
 405};
 406
 407union dlm_packet {
 408        struct dlm_header       header;         /* common to other two */
 409        struct dlm_message      message;
 410        struct dlm_rcom         rcom;
 411};
 412
 413struct rcom_config {
 414        __le32                  rf_lvblen;
 415        __le32                  rf_lsflags;
 416        __le64                  rf_unused;
 417};
 418
 419struct rcom_lock {
 420        __le32                  rl_ownpid;
 421        __le32                  rl_lkid;
 422        __le32                  rl_remid;
 423        __le32                  rl_parent_lkid;
 424        __le32                  rl_parent_remid;
 425        __le32                  rl_exflags;
 426        __le32                  rl_flags;
 427        __le32                  rl_lvbseq;
 428        __le32                  rl_result;
 429        int8_t                  rl_rqmode;
 430        int8_t                  rl_grmode;
 431        int8_t                  rl_status;
 432        int8_t                  rl_asts;
 433        __le16                  rl_wait_type;
 434        __le16                  rl_namelen;
 435        char                    rl_name[DLM_RESNAME_MAXLEN];
 436        char                    rl_lvb[0];
 437};
 438
 439struct dlm_ls {
 440        struct list_head        ls_list;        /* list of lockspaces */
 441        dlm_lockspace_t         *ls_local_handle;
 442        uint32_t                ls_global_id;   /* global unique lockspace ID */
 443        uint32_t                ls_exflags;
 444        int                     ls_lvblen;
 445        int                     ls_count;       /* refcount of processes in
 446                                                   the dlm using this ls */
 447        int                     ls_create_count; /* create/release refcount */
 448        unsigned long           ls_flags;       /* LSFL_ */
 449        unsigned long           ls_scan_time;
 450        struct kobject          ls_kobj;
 451
 452        struct dlm_rsbtable     *ls_rsbtbl;
 453        uint32_t                ls_rsbtbl_size;
 454
 455        struct dlm_lkbtable     *ls_lkbtbl;
 456        uint32_t                ls_lkbtbl_size;
 457
 458        struct dlm_dirtable     *ls_dirtbl;
 459        uint32_t                ls_dirtbl_size;
 460
 461        struct mutex            ls_waiters_mutex;
 462        struct list_head        ls_waiters;     /* lkbs needing a reply */
 463
 464        struct mutex            ls_orphans_mutex;
 465        struct list_head        ls_orphans;
 466
 467        struct mutex            ls_timeout_mutex;
 468        struct list_head        ls_timeout;
 469
 470        struct list_head        ls_nodes;       /* current nodes in ls */
 471        struct list_head        ls_nodes_gone;  /* dead node list, recovery */
 472        int                     ls_num_nodes;   /* number of nodes in ls */
 473        int                     ls_low_nodeid;
 474        int                     ls_total_weight;
 475        int                     *ls_node_array;
 476
 477        struct dlm_rsb          ls_stub_rsb;    /* for returning errors */
 478        struct dlm_lkb          ls_stub_lkb;    /* for returning errors */
 479        struct dlm_message      ls_stub_ms;     /* for faking a reply */
 480
 481        struct dentry           *ls_debug_rsb_dentry; /* debugfs */
 482        struct dentry           *ls_debug_waiters_dentry; /* debugfs */
 483        struct dentry           *ls_debug_locks_dentry; /* debugfs */
 484        struct dentry           *ls_debug_all_dentry; /* debugfs */
 485
 486        wait_queue_head_t       ls_uevent_wait; /* user part of join/leave */
 487        int                     ls_uevent_result;
 488        struct completion       ls_members_done;
 489        int                     ls_members_result;
 490
 491        struct miscdevice       ls_device;
 492
 493        /* recovery related */
 494
 495        struct timer_list       ls_timer;
 496        struct task_struct      *ls_recoverd_task;
 497        struct mutex            ls_recoverd_active;
 498        spinlock_t              ls_recover_lock;
 499        unsigned long           ls_recover_begin; /* jiffies timestamp */
 500        uint32_t                ls_recover_status; /* DLM_RS_ */
 501        uint64_t                ls_recover_seq;
 502        struct dlm_recover      *ls_recover_args;
 503        struct rw_semaphore     ls_in_recovery; /* block local requests */
 504        struct rw_semaphore     ls_recv_active; /* block dlm_recv */
 505        struct list_head        ls_requestqueue;/* queue remote requests */
 506        struct mutex            ls_requestqueue_mutex;
 507        struct dlm_rcom         *ls_recover_buf;
 508        int                     ls_recover_nodeid; /* for debugging */
 509        uint64_t                ls_rcom_seq;
 510        spinlock_t              ls_rcom_spin;
 511        struct list_head        ls_recover_list;
 512        spinlock_t              ls_recover_list_lock;
 513        int                     ls_recover_list_count;
 514        wait_queue_head_t       ls_wait_general;
 515        struct mutex            ls_clear_proc_locks;
 516
 517        struct list_head        ls_root_list;   /* root resources */
 518        struct rw_semaphore     ls_root_sem;    /* protect root_list */
 519
 520        int                     ls_namelen;
 521        char                    ls_name[1];
 522};
 523
 524#define LSFL_WORK               0
 525#define LSFL_RUNNING            1
 526#define LSFL_RECOVERY_STOP      2
 527#define LSFL_RCOM_READY         3
 528#define LSFL_RCOM_WAIT          4
 529#define LSFL_UEVENT_WAIT        5
 530#define LSFL_TIMEWARN           6
 531
 532/* much of this is just saving user space pointers associated with the
 533   lock that we pass back to the user lib with an ast */
 534
 535struct dlm_user_args {
 536        struct dlm_user_proc    *proc; /* each process that opens the lockspace
 537                                          device has private data
 538                                          (dlm_user_proc) on the struct file,
 539                                          the process's locks point back to it*/
 540        struct dlm_lksb         lksb;
 541        int                     old_mode;
 542        int                     update_user_lvb;
 543        struct dlm_lksb __user  *user_lksb;
 544        void __user             *castparam;
 545        void __user             *castaddr;
 546        void __user             *bastparam;
 547        void __user             *bastaddr;
 548        uint64_t                xid;
 549};
 550
 551#define DLM_PROC_FLAGS_CLOSING 1
 552#define DLM_PROC_FLAGS_COMPAT  2
 553
 554/* locks list is kept so we can remove all a process's locks when it
 555   exits (or orphan those that are persistent) */
 556
 557struct dlm_user_proc {
 558        dlm_lockspace_t         *lockspace;
 559        unsigned long           flags; /* DLM_PROC_FLAGS */
 560        struct list_head        asts;
 561        spinlock_t              asts_spin;
 562        struct list_head        locks;
 563        spinlock_t              locks_spin;
 564        struct list_head        unlocking;
 565        wait_queue_head_t       wait;
 566};
 567
 568static inline int dlm_locking_stopped(struct dlm_ls *ls)
 569{
 570        return !test_bit(LSFL_RUNNING, &ls->ls_flags);
 571}
 572
 573static inline int dlm_recovery_stopped(struct dlm_ls *ls)
 574{
 575        return test_bit(LSFL_RECOVERY_STOP, &ls->ls_flags);
 576}
 577
 578static inline int dlm_no_directory(struct dlm_ls *ls)
 579{
 580        return (ls->ls_exflags & DLM_LSFL_NODIR) ? 1 : 0;
 581}
 582
 583int dlm_netlink_init(void);
 584void dlm_netlink_exit(void);
 585void dlm_timeout_warn(struct dlm_lkb *lkb);
 586int dlm_plock_init(void);
 587void dlm_plock_exit(void);
 588
 589#ifdef CONFIG_DLM_DEBUG
 590int dlm_register_debugfs(void);
 591void dlm_unregister_debugfs(void);
 592int dlm_create_debug_file(struct dlm_ls *ls);
 593void dlm_delete_debug_file(struct dlm_ls *ls);
 594#else
 595static inline int dlm_register_debugfs(void) { return 0; }
 596static inline void dlm_unregister_debugfs(void) { }
 597static inline int dlm_create_debug_file(struct dlm_ls *ls) { return 0; }
 598static inline void dlm_delete_debug_file(struct dlm_ls *ls) { }
 599#endif
 600
 601#endif                          /* __DLM_INTERNAL_DOT_H__ */
 602
 603