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         int (*callback)(struct file_lock *fl, int result);
  30 };
  31
  32 struct plock_xop {
  33         struct plock_op xop;
  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 /* If a process was killed while waiting for the only plock on a file,
  75    locks_remove_posix will not see any lock on the file so it won't
  76    send an unlock-close to us to pass on to userspace to clean up the
  77    abandoned waiter.  So, we have to insert the unlock-close when the
  78    lock call is interrupted. */
  79
  80 static void do_unlock_close(struct dlm_ls *ls, u64 number,
  81                             struct file *file, struct file_lock *fl)
  82 {
  83         struct plock_op *op;
  84
  85         op = kzalloc(sizeof(*op), GFP_NOFS);
  86         if (!op)
  87                 return;
  88
  89         op->info.optype         = DLM_PLOCK_OP_UNLOCK;
  90         op->info.pid            = fl->fl_pid;
  91         op->info.fsid           = ls->ls_global_id;
  92         op->info.number         = number;
  93         op->info.start          = 0;
  94         op->info.end            = OFFSET_MAX;
  95         if (fl->fl_lmops && fl->fl_lmops->lm_grant)
  96                 op->info.owner  = (__u64) fl->fl_pid;
  97         else
  98                 op->info.owner  = (__u64)(long) fl->fl_owner;
  99
 100         op->info.flags |= DLM_PLOCK_FL_CLOSE;
 101         send_op(op);
 102 }
 103
 104 int dlm_posix_lock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
 105                    int cmd, struct file_lock *fl)
 106 {
 107         struct dlm_ls *ls;
 108         struct plock_op *op;
 109         struct plock_xop *xop;
 110         int rv;
 111
 112         ls = dlm_find_lockspace_local(lockspace);
 113         if (!ls)
 114                 return -EINVAL;
 115
 116         xop = kzalloc(sizeof(*xop), GFP_NOFS);
 117         if (!xop) {
 118                 rv = -ENOMEM;
 119                 goto out;
 120         }
 121
 122         op = &xop->xop;
 123         op->info.optype         = DLM_PLOCK_OP_LOCK;
 124         op->info.pid            = fl->fl_pid;
 125         op->info.ex             = (fl->fl_type == F_WRLCK);
 126         op->info.wait           = IS_SETLKW(cmd);
 127         op->info.fsid           = ls->ls_global_id;
 128         op->info.number         = number;
 129         op->info.start          = fl->fl_start;
 130         op->info.end            = fl->fl_end;
 131         if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
 132                 /* fl_owner is lockd which doesn't distinguish
 133                    processes on the nfs client */
 134                 op->info.owner  = (__u64) fl->fl_pid;
 135                 op->callback    = fl->fl_lmops->lm_grant;
 136                 locks_init_lock(&xop->flc);
 137                 locks_copy_lock(&xop->flc, fl);
 138                 xop->fl         = fl;
 139                 xop->file       = file;
 140         } else {
 141                 op->info.owner  = (__u64)(long) fl->fl_owner;
 142         }
 143
 144         send_op(op);
 145
 146         if (!op->callback) {
 147                 rv = wait_event_interruptible(recv_wq, (op->done != 0));
 148                 if (rv == -ERESTARTSYS) {
 149                         log_debug(ls, "dlm_posix_lock: wait killed %llx",
 150                                   (unsigned long long)number);
 151                         spin_lock(&ops_lock);
 152                         list_del(&op->list);
 153                         spin_unlock(&ops_lock);
 154                         kfree(xop);
 155                         do_unlock_close(ls, number, file, fl);
 156                         goto out;
 157                 }
 158         } else {
 159                 rv = FILE_LOCK_DEFERRED;
 160                 goto out;
 161         }
 162
 163         spin_lock(&ops_lock);
 164         if (!list_empty(&op->list)) {
 165                 log_error(ls, "dlm_posix_lock: op on list %llx",
 166                           (unsigned long long)number);
 167                 list_del(&op->list);
 168         }
 169         spin_unlock(&ops_lock);
 170
 171         rv = op->info.rv;
 172
 173         if (!rv) {
 174                 if (locks_lock_file_wait(file, fl) < 0)
 175                         log_error(ls, "dlm_posix_lock: vfs lock error %llx",
 176                                   (unsigned long long)number);
 177         }
 178
 179         kfree(xop);
 180 out:
 181         dlm_put_lockspace(ls);
 182         return rv;
 183 }
 184 EXPORT_SYMBOL_GPL(dlm_posix_lock);
 185
 186 /* Returns failure iff a successful lock operation should be canceled */
 187 static int dlm_plock_callback(struct plock_op *op)
 188 {
 189         struct file *file;
 190         struct file_lock *fl;
 191         struct file_lock *flc;
 192         int (*notify)(struct file_lock *fl, int result) = NULL;
 193         struct plock_xop *xop = (struct plock_xop *)op;
 194         int rv = 0;
 195
 196         spin_lock(&ops_lock);
 197         if (!list_empty(&op->list)) {
 198                 log_print("dlm_plock_callback: op on list %llx",
 199                           (unsigned long long)op->info.number);
 200                 list_del(&op->list);
 201         }
 202         spin_unlock(&ops_lock);
 203
 204         /* check if the following 2 are still valid or make a copy */
 205         file = xop->file;
 206         flc = &xop->flc;
 207         fl = xop->fl;
 208         notify = op->callback;
 209
 210         if (op->info.rv) {
 211                 notify(fl, op->info.rv);
 212                 goto out;
 213         }
 214
 215         /* got fs lock; bookkeep locally as well: */
 216         flc->fl_flags &= ~FL_SLEEP;
 217         if (posix_lock_file(file, flc, NULL)) {
 218                 /*
 219                  * This can only happen in the case of kmalloc() failure.
 220                  * The filesystem's own lock is the authoritative lock,
 221                  * so a failure to get the lock locally is not a disaster.
 222                  * As long as the fs cannot reliably cancel locks (especially
 223                  * in a low-memory situation), we're better off ignoring
 224                  * this failure than trying to recover.
 225                  */
 226                 log_print("dlm_plock_callback: vfs lock error %llx file %p fl %p",
 227                           (unsigned long long)op->info.number, file, fl);
 228         }
 229
 230         rv = notify(fl, 0);
 231         if (rv) {
 232                 /* XXX: We need to cancel the fs lock here: */
 233                 log_print("dlm_plock_callback: lock granted after lock request "
 234                           "failed; dangling lock!\n");
 235                 goto out;
 236         }
 237
 238 out:
 239         kfree(xop);
 240         return rv;
 241 }
 242
 243 int dlm_posix_unlock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
 244                      struct file_lock *fl)
 245 {
 246         struct dlm_ls *ls;
 247         struct plock_op *op;
 248         int rv;
 249         unsigned char fl_flags = fl->fl_flags;
 250
 251         ls = dlm_find_lockspace_local(lockspace);
 252         if (!ls)
 253                 return -EINVAL;
 254
 255         op = kzalloc(sizeof(*op), GFP_NOFS);
 256         if (!op) {
 257                 rv = -ENOMEM;
 258                 goto out;
 259         }
 260
 261         /* cause the vfs unlock to return ENOENT if lock is not found */
 262         fl->fl_flags |= FL_EXISTS;
 263
 264         rv = locks_lock_file_wait(file, fl);
 265         if (rv == -ENOENT) {
 266                 rv = 0;
 267                 goto out_free;
 268         }
 269         if (rv < 0) {
 270                 log_error(ls, "dlm_posix_unlock: vfs unlock error %d %llx",
 271                           rv, (unsigned long long)number);
 272         }
 273
 274         op->info.optype         = DLM_PLOCK_OP_UNLOCK;
 275         op->info.pid            = fl->fl_pid;
 276         op->info.fsid           = ls->ls_global_id;
 277         op->info.number         = number;
 278         op->info.start          = fl->fl_start;
 279         op->info.end            = fl->fl_end;
 280         if (fl->fl_lmops && fl->fl_lmops->lm_grant)
 281                 op->info.owner  = (__u64) fl->fl_pid;
 282         else
 283                 op->info.owner  = (__u64)(long) fl->fl_owner;
 284
 285         if (fl->fl_flags & FL_CLOSE) {
 286                 op->info.flags |= DLM_PLOCK_FL_CLOSE;
 287                 send_op(op);
 288                 rv = 0;
 289                 goto out;
 290         }
 291
 292         send_op(op);
 293         wait_event(recv_wq, (op->done != 0));
 294
 295         spin_lock(&ops_lock);
 296         if (!list_empty(&op->list)) {
 297                 log_error(ls, "dlm_posix_unlock: op on list %llx",
 298                           (unsigned long long)number);
 299                 list_del(&op->list);
 300         }
 301         spin_unlock(&ops_lock);
 302
 303         rv = op->info.rv;
 304
 305         if (rv == -ENOENT)
 306                 rv = 0;
 307
 308 out_free:
 309         kfree(op);
 310 out:
 311         dlm_put_lockspace(ls);
 312         fl->fl_flags = fl_flags;
 313         return rv;
 314 }
 315 EXPORT_SYMBOL_GPL(dlm_posix_unlock);
 316
 317 int dlm_posix_get(dlm_lockspace_t *lockspace, u64 number, struct file *file,
 318                   struct file_lock *fl)
 319 {
 320         struct dlm_ls *ls;
 321         struct plock_op *op;
 322         int rv;
 323
 324         ls = dlm_find_lockspace_local(lockspace);
 325         if (!ls)
 326                 return -EINVAL;
 327
 328         op = kzalloc(sizeof(*op), GFP_NOFS);
 329         if (!op) {
 330                 rv = -ENOMEM;
 331                 goto out;
 332         }
 333
 334         op->info.optype         = DLM_PLOCK_OP_GET;
 335         op->info.pid            = fl->fl_pid;
 336         op->info.ex             = (fl->fl_type == F_WRLCK);
 337         op->info.fsid           = ls->ls_global_id;
 338         op->info.number         = number;
 339         op->info.start          = fl->fl_start;
 340         op->info.end            = fl->fl_end;
 341         if (fl->fl_lmops && fl->fl_lmops->lm_grant)
 342                 op->info.owner  = (__u64) fl->fl_pid;
 343         else
 344                 op->info.owner  = (__u64)(long) fl->fl_owner;
 345
 346         send_op(op);
 347         wait_event(recv_wq, (op->done != 0));
 348
 349         spin_lock(&ops_lock);
 350         if (!list_empty(&op->list)) {
 351                 log_error(ls, "dlm_posix_get: op on list %llx",
 352                           (unsigned long long)number);
 353                 list_del(&op->list);
 354         }
 355         spin_unlock(&ops_lock);
 356
 357         /* info.rv from userspace is 1 for conflict, 0 for no-conflict,
 358            -ENOENT if there are no locks on the file */
 359
 360         rv = op->info.rv;
 361
 362         fl->fl_type = F_UNLCK;
 363         if (rv == -ENOENT)
 364                 rv = 0;
 365         else if (rv > 0) {
 366                 locks_init_lock(fl);
 367                 fl->fl_type = (op->info.ex) ? F_WRLCK : F_RDLCK;
 368                 fl->fl_flags = FL_POSIX;
 369                 fl->fl_pid = op->info.pid;
 370                 fl->fl_start = op->info.start;
 371                 fl->fl_end = op->info.end;
 372                 rv = 0;
 373         }
 374
 375         kfree(op);
 376 out:
 377         dlm_put_lockspace(ls);
 378         return rv;
 379 }
 380 EXPORT_SYMBOL_GPL(dlm_posix_get);
 381
 382 /* a read copies out one plock request from the send list */
 383 static ssize_t dev_read(struct file *file, char __user *u, size_t count,
 384                         loff_t *ppos)
 385 {
 386         struct dlm_plock_info info;
 387         struct plock_op *op = NULL;
 388
 389         if (count < sizeof(info))
 390                 return -EINVAL;
 391
 392         spin_lock(&ops_lock);
 393         if (!list_empty(&send_list)) {
 394                 op = list_entry(send_list.next, struct plock_op, list);
 395                 if (op->info.flags & DLM_PLOCK_FL_CLOSE)
 396                         list_del(&op->list);
 397                 else
 398                         list_move(&op->list, &recv_list);
 399                 memcpy(&info, &op->info, sizeof(info));
 400         }
 401         spin_unlock(&ops_lock);
 402
 403         if (!op)
 404                 return -EAGAIN;
 405
 406         /* there is no need to get a reply from userspace for unlocks
 407            that were generated by the vfs cleaning up for a close
 408            (the process did not make an unlock call). */
 409
 410         if (op->info.flags & DLM_PLOCK_FL_CLOSE)
 411                 kfree(op);
 412
 413         if (copy_to_user(u, &info, sizeof(info)))
 414                 return -EFAULT;
 415         return sizeof(info);
 416 }
 417
 418 /* a write copies in one plock result that should match a plock_op
 419    on the recv list */
 420 static ssize_t dev_write(struct file *file, const char __user *u, size_t count,
 421                          loff_t *ppos)
 422 {
 423         struct dlm_plock_info info;
 424         struct plock_op *op;
 425         int found = 0, do_callback = 0;
 426
 427         if (count != sizeof(info))
 428                 return -EINVAL;
 429
 430         if (copy_from_user(&info, u, sizeof(info)))
 431                 return -EFAULT;
 432
 433         if (check_version(&info))
 434                 return -EINVAL;
 435
 436         spin_lock(&ops_lock);
 437         list_for_each_entry(op, &recv_list, list) {
 438                 if (op->info.fsid == info.fsid &&
 439                     op->info.number == info.number &&
 440                     op->info.owner == info.owner) {
 441                         list_del_init(&op->list);
 442                         memcpy(&op->info, &info, sizeof(info));
 443                         if (op->callback)
 444                                 do_callback = 1;
 445                         else
 446                                 op->done = 1;
 447                         found = 1;
 448                         break;
 449                 }
 450         }
 451         spin_unlock(&ops_lock);
 452
 453         if (found) {
 454                 if (do_callback)
 455                         dlm_plock_callback(op);
 456                 else
 457                         wake_up(&recv_wq);
 458         } else
 459                 log_print("dev_write no op %x %llx", info.fsid,
 460                           (unsigned long long)info.number);
 461         return count;
 462 }
 463
 464 static unsigned int dev_poll(struct file *file, poll_table *wait)
 465 {
 466         unsigned int mask = 0;
 467
 468         poll_wait(file, &send_wq, wait);
 469
 470         spin_lock(&ops_lock);
 471         if (!list_empty(&send_list))
 472                 mask = POLLIN | POLLRDNORM;
 473         spin_unlock(&ops_lock);
 474
 475         return mask;
 476 }
 477
 478 static const struct file_operations dev_fops = {
 479         .read    = dev_read,
 480         .write   = dev_write,
 481         .poll    = dev_poll,
 482         .owner   = THIS_MODULE,
 483         .llseek  = noop_llseek,
 484 };
 485
 486 static struct miscdevice plock_dev_misc = {
 487         .minor = MISC_DYNAMIC_MINOR,
 488         .name = DLM_PLOCK_MISC_NAME,
 489         .fops = &dev_fops
 490 };
 491
 492 int dlm_plock_init(void)
 493 {
 494         int rv;
 495
 496         spin_lock_init(&ops_lock);
 497         INIT_LIST_HEAD(&send_list);
 498         INIT_LIST_HEAD(&recv_list);
 499         init_waitqueue_head(&send_wq);
 500         init_waitqueue_head(&recv_wq);
 501
 502         rv = misc_register(&plock_dev_misc);
 503         if (rv)
 504                 log_print("dlm_plock_init: misc_register failed %d", rv);
 505         return rv;
 506 }
 507
 508 void dlm_plock_exit(void)
 509 {
 510         misc_deregister(&plock_dev_misc);
 511 }
 512