fs/dlm/dlm_internal.h

   1 /******************************************************************************
   2 *******************************************************************************
   3 **
   4 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   5 **  Copyright (C) 2004-2005 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 <asm/semaphore.h>
  39 #include <asm/uaccess.h>
  40
  41 #include <linux/dlm.h>
  42
  43 #define DLM_LOCKSPACE_LEN       64
  44
  45 #ifndef TRUE
  46 #define TRUE 1
  47 #endif
  48
  49 #ifndef FALSE
  50 #define FALSE 0
  51 #endif
  52
  53 #if (BITS_PER_LONG == 64)
  54 #define PRIx64 "lx"
  55 #else
  56 #define PRIx64 "Lx"
  57 #endif
  58
  59 /* Size of the temp buffer midcomms allocates on the stack.
  60    We try to make this large enough so most messages fit.
  61    FIXME: should sctp make this unnecessary? */
  62
  63 #define DLM_INBUF_LEN           148
  64
  65 struct dlm_ls;
  66 struct dlm_lkb;
  67 struct dlm_rsb;
  68 struct dlm_member;
  69 struct dlm_lkbtable;
  70 struct dlm_rsbtable;
  71 struct dlm_dirtable;
  72 struct dlm_direntry;
  73 struct dlm_recover;
  74 struct dlm_header;
  75 struct dlm_message;
  76 struct dlm_rcom;
  77 struct dlm_mhandle;
  78
  79 #define log_print(fmt, args...) \
  80         printk(KERN_ERR "dlm: "fmt"\n" , ##args)
  81 #define log_error(ls, fmt, args...) \
  82         printk(KERN_ERR "dlm: %s: " fmt "\n", (ls)->ls_name , ##args)
  83
  84 #ifdef DLM_LOG_DEBUG
  85 #define log_debug(ls, fmt, args...) log_error(ls, fmt, ##args)
  86 #else
  87 #define log_debug(ls, fmt, args...)
  88 #endif
  89
  90 #define DLM_ASSERT(x, do) \
  91 { \
  92   if (!(x)) \
  93   { \
  94     printk(KERN_ERR "\nDLM:  Assertion failed on line %d of file %s\n" \
  95                "DLM:  assertion:  \"%s\"\n" \
  96                "DLM:  time = %lu\n", \
  97                __LINE__, __FILE__, #x, jiffies); \
  98     {do} \
  99     printk("\n"); \
 100     BUG(); \
 101     panic("DLM:  Record message above and reboot.\n"); \
 102   } \
 103 }
 104
 105
 106 struct dlm_direntry {
 107         struct list_head        list;
 108         uint32_t                master_nodeid;
 109         uint16_t                length;
 110         char                    name[1];
 111 };
 112
 113 struct dlm_dirtable {
 114         struct list_head        list;
 115         rwlock_t                lock;
 116 };
 117
 118 struct dlm_rsbtable {
 119         struct list_head        list;
 120         struct list_head        toss;
 121         rwlock_t                lock;
 122 };
 123
 124 struct dlm_lkbtable {
 125         struct list_head        list;
 126         rwlock_t                lock;
 127         uint16_t                counter;
 128 };
 129
 130 /*
 131  * Lockspace member (per node in a ls)
 132  */
 133
 134 struct dlm_member {
 135         struct list_head        list;
 136         int                     nodeid;
 137         int                     weight;
 138 };
 139
 140 /*
 141  * Save and manage recovery state for a lockspace.
 142  */
 143
 144 struct dlm_recover {
 145         struct list_head        list;
 146         int                     *nodeids;
 147         int                     node_count;
 148         uint64_t                seq;
 149 };
 150
 151 /*
 152  * Pass input args to second stage locking function.
 153  */
 154
 155 struct dlm_args {
 156         uint32_t                flags;
 157         void                    *astaddr;
 158         long                    astparam;
 159         void                    *bastaddr;
 160         int                     mode;
 161         struct dlm_lksb         *lksb;
 162         struct dlm_range        *range;
 163 };
 164
 165
 166 /*
 167  * Lock block
 168  *
 169  * A lock can be one of three types:
 170  *
 171  * local copy      lock is mastered locally
 172  *                 (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set)
 173  * process copy    lock is mastered on a remote node
 174  *                 (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set)
 175  * master copy     master node's copy of a lock owned by remote node
 176  *                 (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set)
 177  *
 178  * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or
 179  * dlm_unlock.  The dlm does not modify these or use any private flags in
 180  * this field; it only contains DLM_LKF_ flags from dlm.h.  These flags
 181  * are sent as-is to the remote master when the lock is remote.
 182  *
 183  * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h.
 184  * Some internal flags are shared between the master and process nodes;
 185  * these shared flags are kept in the lower two bytes.  One of these
 186  * flags set on the master copy will be propagated to the process copy
 187  * and v.v.  Other internal flags are private to the master or process
 188  * node (e.g. DLM_IFL_MSTCPY).  These are kept in the high two bytes.
 189  *
 190  * lkb_sbflags: status block flags.  These flags are copied directly into
 191  * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion
 192  * ast.  All defined in dlm.h with DLM_SBF_ prefix.
 193  *
 194  * lkb_status: the lock status indicates which rsb queue the lock is
 195  * on, grant, convert, or wait.  DLM_LKSTS_ WAITING/GRANTED/CONVERT
 196  *
 197  * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a
 198  * reply is needed.  Only set when the lkb is on the lockspace waiters
 199  * list awaiting a reply from a remote node.
 200  *
 201  * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb
 202  * is a master copy, nodeid specifies the remote lock holder, when the
 203  * lkb is a process copy, the nodeid specifies the lock master.
 204  */
 205
 206 /* lkb_ast_type */
 207
 208 #define AST_COMP                1
 209 #define AST_BAST                2
 210
 211 /* lkb_range[] */
 212
 213 #define GR_RANGE_START          0
 214 #define GR_RANGE_END            1
 215 #define RQ_RANGE_START          2
 216 #define RQ_RANGE_END            3
 217
 218 /* lkb_status */
 219
 220 #define DLM_LKSTS_WAITING       1
 221 #define DLM_LKSTS_GRANTED       2
 222 #define DLM_LKSTS_CONVERT       3
 223
 224 /* lkb_flags */
 225
 226 #define DLM_IFL_MSTCPY          0x00010000
 227 #define DLM_IFL_RESEND          0x00020000
 228 #define DLM_IFL_RANGE           0x00000001
 229
 230 struct dlm_lkb {
 231         struct dlm_rsb          *lkb_resource;  /* the rsb */
 232         struct kref             lkb_ref;
 233         int                     lkb_nodeid;     /* copied from rsb */
 234         int                     lkb_ownpid;     /* pid of lock owner */
 235         uint32_t                lkb_id;         /* our lock ID */
 236         uint32_t                lkb_remid;      /* lock ID on remote partner */
 237         uint32_t                lkb_exflags;    /* external flags from caller */
 238         uint32_t                lkb_sbflags;    /* lksb flags */
 239         uint32_t                lkb_flags;      /* internal flags */
 240         uint32_t                lkb_lvbseq;     /* lvb sequence number */
 241
 242         int8_t                  lkb_status;     /* granted, waiting, convert */
 243         int8_t                  lkb_rqmode;     /* requested lock mode */
 244         int8_t                  lkb_grmode;     /* granted lock mode */
 245         int8_t                  lkb_bastmode;   /* requested mode */
 246         int8_t                  lkb_highbast;   /* highest mode bast sent for */
 247
 248         int8_t                  lkb_wait_type;  /* type of reply waiting for */
 249         int8_t                  lkb_ast_type;   /* type of ast queued for */
 250
 251         struct list_head        lkb_idtbl_list; /* lockspace lkbtbl */
 252         struct list_head        lkb_statequeue; /* rsb g/c/w list */
 253         struct list_head        lkb_rsb_lookup; /* waiting for rsb lookup */
 254         struct list_head        lkb_wait_reply; /* waiting for remote reply */
 255         struct list_head        lkb_astqueue;   /* need ast to be sent */
 256
 257         uint64_t                *lkb_range;     /* array of gr/rq ranges */
 258         char                    *lkb_lvbptr;
 259         struct dlm_lksb         *lkb_lksb;      /* caller's status block */
 260         void                    *lkb_astaddr;   /* caller's ast function */
 261         void                    *lkb_bastaddr;  /* caller's bast function */
 262         long                    lkb_astparam;   /* caller's ast arg */
 263 };
 264
 265
 266 struct dlm_rsb {
 267         struct dlm_ls           *res_ls;        /* the lockspace */
 268         struct kref             res_ref;
 269         struct semaphore        res_sem;
 270         unsigned long           res_flags;
 271         int                     res_length;     /* length of rsb name */
 272         int                     res_nodeid;
 273         uint32_t                res_lvbseq;
 274         uint32_t                res_hash;
 275         uint32_t                res_bucket;     /* rsbtbl */
 276         unsigned long           res_toss_time;
 277         uint32_t                res_first_lkid;
 278         struct list_head        res_lookup;     /* lkbs waiting on first */
 279         struct list_head        res_hashchain;  /* rsbtbl */
 280         struct list_head        res_grantqueue;
 281         struct list_head        res_convertqueue;
 282         struct list_head        res_waitqueue;
 283
 284         struct list_head        res_root_list;      /* used for recovery */
 285         struct list_head        res_recover_list;   /* used for recovery */
 286         int                     res_recover_locks_count;
 287
 288         char                    *res_lvbptr;
 289         char                    res_name[1];
 290 };
 291
 292 /* find_rsb() flags */
 293
 294 #define R_MASTER                1       /* only return rsb if it's a master */
 295 #define R_CREATE                2       /* create/add rsb if not found */
 296
 297 /* rsb_flags */
 298
 299 enum rsb_flags {
 300         RSB_MASTER_UNCERTAIN,
 301         RSB_VALNOTVALID,
 302         RSB_VALNOTVALID_PREV,
 303         RSB_NEW_MASTER,
 304         RSB_NEW_MASTER2,
 305         RSB_RECOVER_CONVERT,
 306 };
 307
 308 static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag)
 309 {
 310         __set_bit(flag, &r->res_flags);
 311 }
 312
 313 static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag)
 314 {
 315         __clear_bit(flag, &r->res_flags);
 316 }
 317
 318 static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag)
 319 {
 320         return test_bit(flag, &r->res_flags);
 321 }
 322
 323
 324 /* dlm_header is first element of all structs sent between nodes */
 325
 326 #define DLM_HEADER_MAJOR        0x00020000
 327 #define DLM_HEADER_MINOR        0x00000001
 328
 329 #define DLM_MSG                 1
 330 #define DLM_RCOM                2
 331
 332 struct dlm_header {
 333         uint32_t                h_version;
 334         uint32_t                h_lockspace;
 335         uint32_t                h_nodeid;       /* nodeid of sender */
 336         uint16_t                h_length;
 337         uint8_t                 h_cmd;          /* DLM_MSG, DLM_RCOM */
 338         uint8_t                 h_pad;
 339 };
 340
 341
 342 #define DLM_MSG_REQUEST         1
 343 #define DLM_MSG_CONVERT         2
 344 #define DLM_MSG_UNLOCK          3
 345 #define DLM_MSG_CANCEL          4
 346 #define DLM_MSG_REQUEST_REPLY   5
 347 #define DLM_MSG_CONVERT_REPLY   6
 348 #define DLM_MSG_UNLOCK_REPLY    7
 349 #define DLM_MSG_CANCEL_REPLY    8
 350 #define DLM_MSG_GRANT           9
 351 #define DLM_MSG_BAST            10
 352 #define DLM_MSG_LOOKUP          11
 353 #define DLM_MSG_REMOVE          12
 354 #define DLM_MSG_LOOKUP_REPLY    13
 355
 356 struct dlm_message {
 357         struct dlm_header       m_header;
 358         uint32_t                m_type;         /* DLM_MSG_ */
 359         uint32_t                m_nodeid;
 360         uint32_t                m_pid;
 361         uint32_t                m_lkid;         /* lkid on sender */
 362         uint32_t                m_remid;        /* lkid on receiver */
 363         uint32_t                m_parent_lkid;
 364         uint32_t                m_parent_remid;
 365         uint32_t                m_exflags;
 366         uint32_t                m_sbflags;
 367         uint32_t                m_flags;
 368         uint32_t                m_lvbseq;
 369         uint32_t                m_hash;
 370         int                     m_status;
 371         int                     m_grmode;
 372         int                     m_rqmode;
 373         int                     m_bastmode;
 374         int                     m_asts;
 375         int                     m_result;       /* 0 or -EXXX */
 376         uint64_t                m_range[2];
 377         char                    m_extra[0];     /* name or lvb */
 378 };
 379
 380
 381 #define DLM_RS_NODES            0x00000001
 382 #define DLM_RS_NODES_ALL        0x00000002
 383 #define DLM_RS_DIR              0x00000004
 384 #define DLM_RS_DIR_ALL          0x00000008
 385 #define DLM_RS_LOCKS            0x00000010
 386 #define DLM_RS_LOCKS_ALL        0x00000020
 387 #define DLM_RS_DONE             0x00000040
 388 #define DLM_RS_DONE_ALL         0x00000080
 389
 390 #define DLM_RCOM_STATUS         1
 391 #define DLM_RCOM_NAMES          2
 392 #define DLM_RCOM_LOOKUP         3
 393 #define DLM_RCOM_LOCK           4
 394 #define DLM_RCOM_STATUS_REPLY   5
 395 #define DLM_RCOM_NAMES_REPLY    6
 396 #define DLM_RCOM_LOOKUP_REPLY   7
 397 #define DLM_RCOM_LOCK_REPLY     8
 398
 399 struct dlm_rcom {
 400         struct dlm_header       rc_header;
 401         uint32_t                rc_type;        /* DLM_RCOM_ */
 402         int                     rc_result;      /* multi-purpose */
 403         uint64_t                rc_id;          /* match reply with request */
 404         char                    rc_buf[0];
 405 };
 406
 407 struct rcom_config {
 408         uint32_t                rf_lvblen;
 409         uint32_t                rf_lsflags;
 410         uint64_t                rf_unused;
 411 };
 412
 413 struct rcom_lock {
 414         uint32_t                rl_ownpid;
 415         uint32_t                rl_lkid;
 416         uint32_t                rl_remid;
 417         uint32_t                rl_parent_lkid;
 418         uint32_t                rl_parent_remid;
 419         uint32_t                rl_exflags;
 420         uint32_t                rl_flags;
 421         uint32_t                rl_lvbseq;
 422         int                     rl_result;
 423         int8_t                  rl_rqmode;
 424         int8_t                  rl_grmode;
 425         int8_t                  rl_status;
 426         int8_t                  rl_asts;
 427         uint16_t                rl_wait_type;
 428         uint16_t                rl_namelen;
 429         uint64_t                rl_range[4];
 430         char                    rl_name[DLM_RESNAME_MAXLEN];
 431         char                    rl_lvb[0];
 432 };
 433
 434 struct dlm_ls {
 435         struct list_head        ls_list;        /* list of lockspaces */
 436         uint32_t                ls_global_id;   /* global unique lockspace ID */
 437         uint32_t                ls_exflags;
 438         int                     ls_lvblen;
 439         int                     ls_count;       /* reference count */
 440         unsigned long           ls_flags;       /* LSFL_ */
 441         struct kobject          ls_kobj;
 442
 443         struct dlm_rsbtable     *ls_rsbtbl;
 444         uint32_t                ls_rsbtbl_size;
 445
 446         struct dlm_lkbtable     *ls_lkbtbl;
 447         uint32_t                ls_lkbtbl_size;
 448
 449         struct dlm_dirtable     *ls_dirtbl;
 450         uint32_t                ls_dirtbl_size;
 451
 452         struct semaphore        ls_waiters_sem;
 453         struct list_head        ls_waiters;     /* lkbs needing a reply */
 454
 455         struct list_head        ls_nodes;       /* current nodes in ls */
 456         struct list_head        ls_nodes_gone;  /* dead node list, recovery */
 457         int                     ls_num_nodes;   /* number of nodes in ls */
 458         int                     ls_low_nodeid;
 459         int                     ls_total_weight;
 460         int                     *ls_node_array;
 461
 462         struct dlm_rsb          ls_stub_rsb;    /* for returning errors */
 463         struct dlm_lkb          ls_stub_lkb;    /* for returning errors */
 464         struct dlm_message      ls_stub_ms;     /* for faking a reply */
 465
 466         struct dentry           *ls_debug_dentry; /* debugfs */
 467
 468         wait_queue_head_t       ls_uevent_wait; /* user part of join/leave */
 469         int                     ls_uevent_result;
 470
 471         /* recovery related */
 472
 473         struct timer_list       ls_timer;
 474         struct task_struct      *ls_recoverd_task;
 475         struct semaphore        ls_recoverd_active;
 476         spinlock_t              ls_recover_lock;
 477         uint32_t                ls_recover_status; /* DLM_RS_ */
 478         uint64_t                ls_recover_seq;
 479         struct dlm_recover      *ls_recover_args;
 480         struct rw_semaphore     ls_in_recovery; /* block local requests */
 481         struct list_head        ls_requestqueue;/* queue remote requests */
 482         struct semaphore        ls_requestqueue_lock;
 483         char                    *ls_recover_buf;
 484         struct list_head        ls_recover_list;
 485         spinlock_t              ls_recover_list_lock;
 486         int                     ls_recover_list_count;
 487         wait_queue_head_t       ls_wait_general;
 488
 489         struct list_head        ls_root_list;   /* root resources */
 490         struct rw_semaphore     ls_root_sem;    /* protect root_list */
 491
 492         int                     ls_namelen;
 493         char                    ls_name[1];
 494 };
 495
 496 #define LSFL_WORK               0
 497 #define LSFL_RUNNING            1
 498 #define LSFL_RECOVERY_STOP      2
 499 #define LSFL_RCOM_READY         3
 500 #define LSFL_UEVENT_WAIT        4
 501
 502 static inline int dlm_locking_stopped(struct dlm_ls *ls)
 503 {
 504         return !test_bit(LSFL_RUNNING, &ls->ls_flags);
 505 }
 506
 507 static inline int dlm_recovery_stopped(struct dlm_ls *ls)
 508 {
 509         return test_bit(LSFL_RECOVERY_STOP, &ls->ls_flags);
 510 }
 511
 512 static inline int dlm_no_directory(struct dlm_ls *ls)
 513 {
 514         return (ls->ls_exflags & DLM_LSFL_NODIR) ? 1 : 0;
 515 }
 516
 517 #endif                          /* __DLM_INTERNAL_DOT_H__ */
 518