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