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
  15 #include "dlm_internal.h"
  16 #include "lockspace.h"
  17
  18 static spinlock_t ops_lock;
  19 static struct list_head send_list;
  20 static struct list_head recv_list;
  21 static wait_queue_head_t send_wq;
  22 static wait_queue_head_t recv_wq;
  23
  24 struct plock_op {
  25         struct list_head list;
  26         int done;
  27         struct dlm_plock_info info;
  28 };
  29
  30 struct plock_xop {
  31         struct plock_op xop;
  32         void *callback;
  33         void *fl;
  34         void *file;
  35         struct file_lock flc;
  36 };
  37
  38
  39 static inline void set_version(struct dlm_plock_info *info)
  40 {
  41         info->version[0] = DLM_PLOCK_VERSION_MAJOR;
  42         info->version[1] = DLM_PLOCK_VERSION_MINOR;
  43         info->version[2] = DLM_PLOCK_VERSION_PATCH;
  44 }
  45
  46 static int check_version(struct dlm_plock_info *info)
  47 {
  48         if ((DLM_PLOCK_VERSION_MAJOR != info->version[0]) ||
  49             (DLM_PLOCK_VERSION_MINOR < info->version[1])) {
  50                 log_print("plock device version mismatch: "
  51                           "kernel (%u.%u.%u), user (%u.%u.%u)",
  52                           DLM_PLOCK_VERSION_MAJOR,
  53                           DLM_PLOCK_VERSION_MINOR,
  54                           DLM_PLOCK_VERSION_PATCH,
  55                           info->version[0],
  56                           info->version[1],
  57                           info->version[2]);
  58                 return -EINVAL;
  59         }
  60         return 0;
  61 }
  62
  63 static void send_op(struct plock_op *op)
  64 {
  65         set_version(&op->info);
  66         INIT_LIST_HEAD(&op->list);
  67         spin_lock(&ops_lock);
  68         list_add_tail(&op->list, &send_list);
  69         spin_unlock(&ops_lock);
  70         wake_up(&send_wq);
  71 }
  72
  73 int dlm_posix_lock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
  74                    int cmd, struct file_lock *fl)
  75 {
  76         struct dlm_ls *ls;
  77         struct plock_op *op;
  78         struct plock_xop *xop;
  79         int rv;
  80
  81         ls = dlm_find_lockspace_local(lockspace);
  82         if (!ls)
  83                 return -EINVAL;
  84
  85         xop = kzalloc(sizeof(*xop), GFP_KERNEL);
  86         if (!xop) {
  87                 rv = -ENOMEM;
  88                 goto out;
  89         }
  90
  91         op = &xop->xop;
  92         op->info.optype         = DLM_PLOCK_OP_LOCK;
  93         op->info.pid            = fl->fl_pid;
  94         op->info.ex             = (fl->fl_type == F_WRLCK);
  95         op->info.wait           = IS_SETLKW(cmd);
  96         op->info.fsid           = ls->ls_global_id;
  97         op->info.number         = number;
  98         op->info.start          = fl->fl_start;
  99         op->info.end            = fl->fl_end;
 100         if (fl->fl_lmops && fl->fl_lmops->fl_grant) {
 101                 /* fl_owner is lockd which doesn't distinguish
 102                    processes on the nfs client */
 103                 op->info.owner  = (__u64) fl->fl_pid;
 104                 xop->callback   = fl->fl_lmops->fl_grant;
 105                 locks_init_lock(&xop->flc);
 106                 locks_copy_lock(&xop->flc, fl);
 107                 xop->fl         = fl;
 108                 xop->file       = file;
 109         } else {
 110                 op->info.owner  = (__u64)(long) fl->fl_owner;
 111                 xop->callback   = NULL;
 112         }
 113
 114         send_op(op);
 115
 116         if (xop->callback == NULL)
 117                 wait_event(recv_wq, (op->done != 0));
 118         else {
 119                 rv = FILE_LOCK_DEFERRED;
 120                 goto out;
 121         }
 122
 123         spin_lock(&ops_lock);
 124         if (!list_empty(&op->list)) {
 125                 log_error(ls, "dlm_posix_lock: op on list %llx",
 126                           (unsigned long long)number);
 127                 list_del(&op->list);
 128         }
 129         spin_unlock(&ops_lock);
 130
 131         rv = op->info.rv;
 132
 133         if (!rv) {
 134                 if (posix_lock_file_wait(file, fl) < 0)
 135                         log_error(ls, "dlm_posix_lock: vfs lock error %llx",
 136                                   (unsigned long long)number);
 137         }
 138
 139         kfree(xop);
 140 out:
 141         dlm_put_lockspace(ls);
 142         return rv;
 143 }
 144 EXPORT_SYMBOL_GPL(dlm_posix_lock);
 145
 146 /* Returns failure iff a succesful lock operation should be canceled */
 147 static int dlm_plock_callback(struct plock_op *op)
 148 {
 149         struct file *file;
 150         struct file_lock *fl;
 151         struct file_lock *flc;
 152         int (*notify)(void *, void *, int) = NULL;
 153         struct plock_xop *xop = (struct plock_xop *)op;
 154         int rv = 0;
 155
 156         spin_lock(&ops_lock);
 157         if (!list_empty(&op->list)) {
 158                 log_print("dlm_plock_callback: op on list %llx",
 159                           (unsigned long long)op->info.number);
 160                 list_del(&op->list);
 161         }
 162         spin_unlock(&ops_lock);
 163
 164         /* check if the following 2 are still valid or make a copy */
 165         file = xop->file;
 166         flc = &xop->flc;
 167         fl = xop->fl;
 168         notify = xop->callback;
 169
 170         if (op->info.rv) {
 171                 notify(flc, NULL, op->info.rv);
 172                 goto out;
 173         }
 174
 175         /* got fs lock; bookkeep locally as well: */
 176         flc->fl_flags &= ~FL_SLEEP;
 177         if (posix_lock_file(file, flc, NULL)) {
 178                 /*
 179                  * This can only happen in the case of kmalloc() failure.
 180                  * The filesystem's own lock is the authoritative lock,
 181                  * so a failure to get the lock locally is not a disaster.
 182                  * As long as the fs cannot reliably cancel locks (especially
 183                  * in a low-memory situation), we're better off ignoring
 184                  * this failure than trying to recover.
 185                  */
 186                 log_print("dlm_plock_callback: vfs lock error %llx file %p fl %p",
 187                           (unsigned long long)op->info.number, file, fl);
 188         }
 189
 190         rv = notify(flc, NULL, 0);
 191         if (rv) {
 192                 /* XXX: We need to cancel the fs lock here: */
 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_KERNEL);
 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_KERNEL);
 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                 fl->fl_type = (op->info.ex) ? F_WRLCK : F_RDLCK;
 308                 fl->fl_pid = op->info.pid;
 309                 fl->fl_start = op->info.start;
 310                 fl->fl_end = op->info.end;
 311                 rv = 0;
 312         }
 313
 314         kfree(op);
 315 out:
 316         dlm_put_lockspace(ls);
 317         return rv;
 318 }
 319 EXPORT_SYMBOL_GPL(dlm_posix_get);
 320
 321 /* a read copies out one plock request from the send list */
 322 static ssize_t dev_read(struct file *file, char __user *u, size_t count,
 323                         loff_t *ppos)
 324 {
 325         struct dlm_plock_info info;
 326         struct plock_op *op = NULL;
 327
 328         if (count < sizeof(info))
 329                 return -EINVAL;
 330
 331         spin_lock(&ops_lock);
 332         if (!list_empty(&send_list)) {
 333                 op = list_entry(send_list.next, struct plock_op, list);
 334                 list_move(&op->list, &recv_list);
 335                 memcpy(&info, &op->info, sizeof(info));
 336         }
 337         spin_unlock(&ops_lock);
 338
 339         if (!op)
 340                 return -EAGAIN;
 341
 342         if (copy_to_user(u, &info, sizeof(info)))
 343                 return -EFAULT;
 344         return sizeof(info);
 345 }
 346
 347 /* a write copies in one plock result that should match a plock_op
 348    on the recv list */
 349 static ssize_t dev_write(struct file *file, const char __user *u, size_t count,
 350                          loff_t *ppos)
 351 {
 352         struct dlm_plock_info info;
 353         struct plock_op *op;
 354         int found = 0;
 355
 356         if (count != sizeof(info))
 357                 return -EINVAL;
 358
 359         if (copy_from_user(&info, u, sizeof(info)))
 360                 return -EFAULT;
 361
 362         if (check_version(&info))
 363                 return -EINVAL;
 364
 365         spin_lock(&ops_lock);
 366         list_for_each_entry(op, &recv_list, list) {
 367                 if (op->info.fsid == info.fsid && op->info.number == info.number &&
 368                     op->info.owner == info.owner) {
 369                         list_del_init(&op->list);
 370                         found = 1;
 371                         op->done = 1;
 372                         memcpy(&op->info, &info, sizeof(info));
 373                         break;
 374                 }
 375         }
 376         spin_unlock(&ops_lock);
 377
 378         if (found) {
 379                 struct plock_xop *xop;
 380                 xop = (struct plock_xop *)op;
 381                 if (xop->callback)
 382                         dlm_plock_callback(op);
 383                 else
 384                         wake_up(&recv_wq);
 385         } else
 386                 log_print("dev_write no op %x %llx", info.fsid,
 387                           (unsigned long long)info.number);
 388         return count;
 389 }
 390
 391 static unsigned int dev_poll(struct file *file, poll_table *wait)
 392 {
 393         unsigned int mask = 0;
 394
 395         poll_wait(file, &send_wq, wait);
 396
 397         spin_lock(&ops_lock);
 398         if (!list_empty(&send_list))
 399                 mask = POLLIN | POLLRDNORM;
 400         spin_unlock(&ops_lock);
 401
 402         return mask;
 403 }
 404
 405 static const struct file_operations dev_fops = {
 406         .read    = dev_read,
 407         .write   = dev_write,
 408         .poll    = dev_poll,
 409         .owner   = THIS_MODULE
 410 };
 411
 412 static struct miscdevice plock_dev_misc = {
 413         .minor = MISC_DYNAMIC_MINOR,
 414         .name = DLM_PLOCK_MISC_NAME,
 415         .fops = &dev_fops
 416 };
 417
 418 int dlm_plock_init(void)
 419 {
 420         int rv;
 421
 422         spin_lock_init(&ops_lock);
 423         INIT_LIST_HEAD(&send_list);
 424         INIT_LIST_HEAD(&recv_list);
 425         init_waitqueue_head(&send_wq);
 426         init_waitqueue_head(&recv_wq);
 427
 428         rv = misc_register(&plock_dev_misc);
 429         if (rv)
 430                 log_print("dlm_plock_init: misc_register failed %d", rv);
 431         return rv;
 432 }
 433
 434 void dlm_plock_exit(void)
 435 {
 436         if (misc_deregister(&plock_dev_misc) < 0)
 437                 log_print("dlm_plock_exit: misc_deregister failed");
 438 }
 439