release/src-rt-6.x.4708/linux/linux-2.6.36/fs/dlm/plock.c

   1 /*
   2  * Copyright (C) 2005-2008 Red Hat, Inc.  All rights reserved.
   3  *
   4  * This copyrighted material is made available to anyone wishing to use,
   5  * modify, copy, or redistribute it subject to the terms and conditions
   6  * of the GNU General Public License version 2.
   7  */
   8
   9 #include <linux/fs.h>
  10 #include <linux/miscdevice.h>
  11 #include <linux/poll.h>
  12 #include <linux/dlm.h>
  13 #include <linux/dlm_plock.h>
  14 #include <linux/slab.h>
  15
  16 #include "dlm_internal.h"
  17 #include "lockspace.h"
  18
  19 static spinlock_t ops_lock;
  20 static struct list_head send_list;
  21 static struct list_head recv_list;
  22 static wait_queue_head_t send_wq;
  23 static wait_queue_head_t recv_wq;
  24
  25 struct plock_op {
  26         struct list_head list;
  27         int done;
  28         struct dlm_plock_info info;
  29 };
  30
  31 struct plock_xop {
  32         struct plock_op xop;
  33         void *callback;
  34         void *fl;
  35         void *file;
  36         struct file_lock flc;
  37 };
  38
  39
  40 static inline void set_version(struct dlm_plock_info *info)
  41 {
  42         info->version[0] = DLM_PLOCK_VERSION_MAJOR;
  43         info->version[1] = DLM_PLOCK_VERSION_MINOR;
  44         info->version[2] = DLM_PLOCK_VERSION_PATCH;
  45 }
  46
  47 static int check_version(struct dlm_plock_info *info)
  48 {
  49         if ((DLM_PLOCK_VERSION_MAJOR != info->version[0]) ||
  50             (DLM_PLOCK_VERSION_MINOR < info->version[1])) {
  51                 log_print("plock device version mismatch: "
  52                           "kernel (%u.%u.%u), user (%u.%u.%u)",
  53                           DLM_PLOCK_VERSION_MAJOR,
  54                           DLM_PLOCK_VERSION_MINOR,
  55                           DLM_PLOCK_VERSION_PATCH,
  56                           info->version[0],
  57                           info->version[1],
  58                           info->version[2]);
  59                 return -EINVAL;
  60         }
  61         return 0;
  62 }
  63
  64 static void send_op(struct plock_op *op)
  65 {
  66         set_version(&op->info);
  67         INIT_LIST_HEAD(&op->list);
  68         spin_lock(&ops_lock);
  69         list_add_tail(&op->list, &send_list);
  70         spin_unlock(&ops_lock);
  71         wake_up(&send_wq);
  72 }
  73
  74 int dlm_posix_lock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
  75                    int cmd, struct file_lock *fl)
  76 {
  77         struct dlm_ls *ls;
  78         struct plock_op *op;
  79         struct plock_xop *xop;
  80         int rv;
  81
  82         ls = dlm_find_lockspace_local(lockspace);
  83         if (!ls)
  84                 return -EINVAL;
  85
  86         xop = kzalloc(sizeof(*xop), GFP_NOFS);
  87         if (!xop) {
  88                 rv = -ENOMEM;
  89                 goto out;
  90         }
  91
  92         op = &xop->xop;
  93         op->info.optype         = DLM_PLOCK_OP_LOCK;
  94         op->info.pid            = fl->fl_pid;
  95         op->info.ex             = (fl->fl_type == F_WRLCK);
  96         op->info.wait           = IS_SETLKW(cmd);
  97         op->info.fsid           = ls->ls_global_id;
  98         op->info.number         = number;
  99         op->info.start          = fl->fl_start;
 100         op->info.end            = fl->fl_end;
 101         if (fl->fl_lmops && fl->fl_lmops->fl_grant) {
 102                 /* fl_owner is lockd which doesn't distinguish
 103                    processes on the nfs client */
 104                 op->info.owner  = (__u64) fl->fl_pid;
 105                 xop->callback   = fl->fl_lmops->fl_grant;
 106                 locks_init_lock(&xop->flc);
 107                 locks_copy_lock(&xop->flc, fl);
 108                 xop->fl         = fl;
 109                 xop->file       = file;
 110         } else {
 111                 op->info.owner  = (__u64)(long) fl->fl_owner;
 112                 xop->callback   = NULL;
 113         }
 114
 115         send_op(op);
 116
 117         if (xop->callback == NULL)
 118                 wait_event(recv_wq, (op->done != 0));
 119         else {
 120                 rv = FILE_LOCK_DEFERRED;
 121                 goto out;
 122         }
 123
 124         spin_lock(&ops_lock);
 125         if (!list_empty(&op->list)) {
 126                 log_error(ls, "dlm_posix_lock: op on list %llx",
 127                           (unsigned long long)number);
 128                 list_del(&op->list);
 129         }
 130         spin_unlock(&ops_lock);
 131
 132         rv = op->info.rv;
 133
 134         if (!rv) {
 135                 if (posix_lock_file_wait(file, fl) < 0)
 136                         log_error(ls, "dlm_posix_lock: vfs lock error %llx",
 137                                   (unsigned long long)number);
 138         }
 139
 140         kfree(xop);
 141 out:
 142         dlm_put_lockspace(ls);
 143         return rv;
 144 }
 145 EXPORT_SYMBOL_GPL(dlm_posix_lock);
 146
 147 /* Returns failure iff a successful lock operation should be canceled */
 148 static int dlm_plock_callback(struct plock_op *op)
 149 {
 150         struct file *file;
 151         struct file_lock *fl;
 152         struct file_lock *flc;
 153         int (*notify)(void *, void *, int) = NULL;
 154         struct plock_xop *xop = (struct plock_xop *)op;
 155         int rv = 0;
 156
 157         spin_lock(&ops_lock);
 158         if (!list_empty(&op->list)) {
 159                 log_print("dlm_plock_callback: op on list %llx",
 160                           (unsigned long long)op->info.number);
 161                 list_del(&op->list);
 162         }
 163         spin_unlock(&ops_lock);
 164
 165         /* check if the following 2 are still valid or make a copy */
 166         file = xop->file;
 167         flc = &xop->flc;
 168         fl = xop->fl;
 169         notify = xop->callback;
 170
 171         if (op->info.rv) {
 172                 notify(fl, NULL, op->info.rv);
 173                 goto out;
 174         }
 175
 176         /* got fs lock; bookkeep locally as well: */
 177         flc->fl_flags &= ~FL_SLEEP;
 178         if (posix_lock_file(file, flc, NULL)) {
 179                 /*
 180                  * This can only happen in the case of kmalloc() failure.
 181                  * The filesystem's own lock is the authoritative lock,
 182                  * so a failure to get the lock locally is not a disaster.
 183                  * As long as the fs cannot reliably cancel locks (especially
 184                  * in a low-memory situation), we're better off ignoring
 185                  * this failure than trying to recover.
 186                  */
 187                 log_print("dlm_plock_callback: vfs lock error %llx file %p fl %p",
 188                           (unsigned long long)op->info.number, file, fl);
 189         }
 190
 191         rv = notify(fl, NULL, 0);
 192         if (rv) {
 193                 log_print("dlm_plock_callback: lock granted after lock request "
 194                           "failed; dangling lock!\n");
 195                 goto out;
 196         }
 197
 198 out:
 199         kfree(xop);
 200         return rv;
 201 }
 202
 203 int dlm_posix_unlock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
 204                      struct file_lock *fl)
 205 {
 206         struct dlm_ls *ls;
 207         struct plock_op *op;
 208         int rv;
 209
 210         ls = dlm_find_lockspace_local(lockspace);
 211         if (!ls)
 212                 return -EINVAL;
 213
 214         op = kzalloc(sizeof(*op), GFP_NOFS);
 215         if (!op) {
 216                 rv = -ENOMEM;
 217                 goto out;
 218         }
 219
 220         if (posix_lock_file_wait(file, fl) < 0)
 221                 log_error(ls, "dlm_posix_unlock: vfs unlock error %llx",
 222                           (unsigned long long)number);
 223
 224         op->info.optype         = DLM_PLOCK_OP_UNLOCK;
 225         op->info.pid            = fl->fl_pid;
 226         op->info.fsid           = ls->ls_global_id;
 227         op->info.number         = number;
 228         op->info.start          = fl->fl_start;
 229         op->info.end            = fl->fl_end;
 230         if (fl->fl_lmops && fl->fl_lmops->fl_grant)
 231                 op->info.owner  = (__u64) fl->fl_pid;
 232         else
 233                 op->info.owner  = (__u64)(long) fl->fl_owner;
 234
 235         send_op(op);
 236         wait_event(recv_wq, (op->done != 0));
 237
 238         spin_lock(&ops_lock);
 239         if (!list_empty(&op->list)) {
 240                 log_error(ls, "dlm_posix_unlock: op on list %llx",
 241                           (unsigned long long)number);
 242                 list_del(&op->list);
 243         }
 244         spin_unlock(&ops_lock);
 245
 246         rv = op->info.rv;
 247
 248         if (rv == -ENOENT)
 249                 rv = 0;
 250
 251         kfree(op);
 252 out:
 253         dlm_put_lockspace(ls);
 254         return rv;
 255 }
 256 EXPORT_SYMBOL_GPL(dlm_posix_unlock);
 257
 258 int dlm_posix_get(dlm_lockspace_t *lockspace, u64 number, struct file *file,
 259                   struct file_lock *fl)
 260 {
 261         struct dlm_ls *ls;
 262         struct plock_op *op;
 263         int rv;
 264
 265         ls = dlm_find_lockspace_local(lockspace);
 266         if (!ls)
 267                 return -EINVAL;
 268
 269         op = kzalloc(sizeof(*op), GFP_NOFS);
 270         if (!op) {
 271                 rv = -ENOMEM;
 272                 goto out;
 273         }
 274
 275         op->info.optype         = DLM_PLOCK_OP_GET;
 276         op->info.pid            = fl->fl_pid;
 277         op->info.ex             = (fl->fl_type == F_WRLCK);
 278         op->info.fsid           = ls->ls_global_id;
 279         op->info.number         = number;
 280         op->info.start          = fl->fl_start;
 281         op->info.end            = fl->fl_end;
 282         if (fl->fl_lmops && fl->fl_lmops->fl_grant)
 283                 op->info.owner  = (__u64) fl->fl_pid;
 284         else
 285                 op->info.owner  = (__u64)(long) fl->fl_owner;
 286
 287         send_op(op);
 288         wait_event(recv_wq, (op->done != 0));
 289
 290         spin_lock(&ops_lock);
 291         if (!list_empty(&op->list)) {
 292                 log_error(ls, "dlm_posix_get: op on list %llx",
 293                           (unsigned long long)number);
 294                 list_del(&op->list);
 295         }
 296         spin_unlock(&ops_lock);
 297
 298         /* info.rv from userspace is 1 for conflict, 0 for no-conflict,
 299            -ENOENT if there are no locks on the file */
 300
 301         rv = op->info.rv;
 302
 303         fl->fl_type = F_UNLCK;
 304         if (rv == -ENOENT)
 305                 rv = 0;
 306         else if (rv > 0) {
 307                 locks_init_lock(fl);
 308                 fl->fl_type = (op->info.ex) ? F_WRLCK : F_RDLCK;
 309                 fl->fl_flags = FL_POSIX;
 310                 fl->fl_pid = op->info.pid;
 311                 fl->fl_start = op->info.start;
 312                 fl->fl_end = op->info.end;
 313                 rv = 0;
 314         }
 315
 316         kfree(op);
 317 out:
 318         dlm_put_lockspace(ls);
 319         return rv;
 320 }
 321 EXPORT_SYMBOL_GPL(dlm_posix_get);
 322
 323 /* a read copies out one plock request from the send list */
 324 static ssize_t dev_read(struct file *file, char __user *u, size_t count,
 325                         loff_t *ppos)
 326 {
 327         struct dlm_plock_info info;
 328         struct plock_op *op = NULL;
 329
 330         if (count < sizeof(info))
 331                 return -EINVAL;
 332
 333         spin_lock(&ops_lock);
 334         if (!list_empty(&send_list)) {
 335                 op = list_entry(send_list.next, struct plock_op, list);
 336                 list_move(&op->list, &recv_list);
 337                 memcpy(&info, &op->info, sizeof(info));
 338         }
 339         spin_unlock(&ops_lock);
 340
 341         if (!op)
 342                 return -EAGAIN;
 343
 344         if (copy_to_user(u, &info, sizeof(info)))
 345                 return -EFAULT;
 346         return sizeof(info);
 347 }
 348
 349 /* a write copies in one plock result that should match a plock_op
 350    on the recv list */
 351 static ssize_t dev_write(struct file *file, const char __user *u, size_t count,
 352                          loff_t *ppos)
 353 {
 354         struct dlm_plock_info info;
 355         struct plock_op *op;
 356         int found = 0, do_callback = 0;
 357
 358         if (count != sizeof(info))
 359                 return -EINVAL;
 360
 361         if (copy_from_user(&info, u, sizeof(info)))
 362                 return -EFAULT;
 363
 364         if (check_version(&info))
 365                 return -EINVAL;
 366
 367         spin_lock(&ops_lock);
 368         list_for_each_entry(op, &recv_list, list) {
 369                 if (op->info.fsid == info.fsid &&
 370                     op->info.number == info.number &&
 371                     op->info.owner == info.owner) {
 372                         struct plock_xop *xop = (struct plock_xop *)op;
 373                         list_del_init(&op->list);
 374                         memcpy(&op->info, &info, sizeof(info));
 375                         if (xop->callback)
 376                                 do_callback = 1;
 377                         else
 378                                 op->done = 1;
 379                         found = 1;
 380                         break;
 381                 }
 382         }
 383         spin_unlock(&ops_lock);
 384
 385         if (found) {
 386                 if (do_callback)
 387                         dlm_plock_callback(op);
 388                 else
 389                         wake_up(&recv_wq);
 390         } else
 391                 log_print("dev_write no op %x %llx", info.fsid,
 392                           (unsigned long long)info.number);
 393         return count;
 394 }
 395
 396 static unsigned int dev_poll(struct file *file, poll_table *wait)
 397 {
 398         unsigned int mask = 0;
 399
 400         poll_wait(file, &send_wq, wait);
 401
 402         spin_lock(&ops_lock);
 403         if (!list_empty(&send_list))
 404                 mask = POLLIN | POLLRDNORM;
 405         spin_unlock(&ops_lock);
 406
 407         return mask;
 408 }
 409
 410 static const struct file_operations dev_fops = {
 411         .read    = dev_read,
 412         .write   = dev_write,
 413         .poll    = dev_poll,
 414         .owner   = THIS_MODULE
 415 };
 416
 417 static struct miscdevice plock_dev_misc = {
 418         .minor = MISC_DYNAMIC_MINOR,
 419         .name = DLM_PLOCK_MISC_NAME,
 420         .fops = &dev_fops
 421 };
 422
 423 int dlm_plock_init(void)
 424 {
 425         int rv;
 426
 427         spin_lock_init(&ops_lock);
 428         INIT_LIST_HEAD(&send_list);
 429         INIT_LIST_HEAD(&recv_list);
 430         init_waitqueue_head(&send_wq);
 431         init_waitqueue_head(&recv_wq);
 432
 433         rv = misc_register(&plock_dev_misc);
 434         if (rv)
 435                 log_print("dlm_plock_init: misc_register failed %d", rv);
 436         return rv;
 437 }
 438
 439 void dlm_plock_exit(void)
 440 {
 441         if (misc_deregister(&plock_dev_misc) < 0)
 442                 log_print("dlm_plock_exit: misc_deregister failed");
 443 }