fs/dlm/plock.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2005-2008 Red Hat, Inc.  All rights reserved.
   4  */
   5
   6 #include <linux/fs.h>
   7 #include <linux/filelock.h>
   8 #include <linux/miscdevice.h>
   9 #include <linux/poll.h>
  10 #include <linux/dlm.h>
  11 #include <linux/dlm_plock.h>
  12 #include <linux/slab.h>
  13
  14 #include "dlm_internal.h"
  15 #include "lockspace.h"
  16
  17 static DEFINE_SPINLOCK(ops_lock);
  18 static LIST_HEAD(send_list);
  19 static LIST_HEAD(recv_list);
  20 static DECLARE_WAIT_QUEUE_HEAD(send_wq);
  21 static DECLARE_WAIT_QUEUE_HEAD(recv_wq);
  22
  23 struct plock_async_data {
  24         void *fl;
  25         void *file;
  26         struct file_lock flc;
  27         int (*callback)(struct file_lock *fl, int result);
  28 };
  29
  30 struct plock_op {
  31         struct list_head list;
  32         int done;
  33         struct dlm_plock_info info;
  34         /* if set indicates async handling */
  35         struct plock_async_data *data;
  36 };
  37
  38 static inline void set_version(struct dlm_plock_info *info)
  39 {
  40         info->version[0] = DLM_PLOCK_VERSION_MAJOR;
  41         info->version[1] = DLM_PLOCK_VERSION_MINOR;
  42         info->version[2] = DLM_PLOCK_VERSION_PATCH;
  43 }
  44
  45 static int check_version(struct dlm_plock_info *info)
  46 {
  47         if ((DLM_PLOCK_VERSION_MAJOR != info->version[0]) ||
  48             (DLM_PLOCK_VERSION_MINOR < info->version[1])) {
  49                 log_print("plock device version mismatch: "
  50                           "kernel (%u.%u.%u), user (%u.%u.%u)",
  51                           DLM_PLOCK_VERSION_MAJOR,
  52                           DLM_PLOCK_VERSION_MINOR,
  53                           DLM_PLOCK_VERSION_PATCH,
  54                           info->version[0],
  55                           info->version[1],
  56                           info->version[2]);
  57                 return -EINVAL;
  58         }
  59         return 0;
  60 }
  61
  62 static void dlm_release_plock_op(struct plock_op *op)
  63 {
  64         kfree(op->data);
  65         kfree(op);
  66 }
  67
  68 static void send_op(struct plock_op *op)
  69 {
  70         set_version(&op->info);
  71         spin_lock(&ops_lock);
  72         list_add_tail(&op->list, &send_list);
  73         spin_unlock(&ops_lock);
  74         wake_up(&send_wq);
  75 }
  76
  77 /* If a process was killed while waiting for the only plock on a file,
  78    locks_remove_posix will not see any lock on the file so it won't
  79    send an unlock-close to us to pass on to userspace to clean up the
  80    abandoned waiter.  So, we have to insert the unlock-close when the
  81    lock call is interrupted. */
  82
  83 static void do_unlock_close(const struct dlm_plock_info *info)
  84 {
  85         struct plock_op *op;
  86
  87         op = kzalloc(sizeof(*op), GFP_NOFS);
  88         if (!op)
  89                 return;
  90
  91         op->info.optype         = DLM_PLOCK_OP_UNLOCK;
  92         op->info.pid            = info->pid;
  93         op->info.fsid           = info->fsid;
  94         op->info.number         = info->number;
  95         op->info.start          = 0;
  96         op->info.end            = OFFSET_MAX;
  97         op->info.owner          = info->owner;
  98
  99         op->info.flags |= DLM_PLOCK_FL_CLOSE;
 100         send_op(op);
 101 }
 102
 103 int dlm_posix_lock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
 104                    int cmd, struct file_lock *fl)
 105 {
 106         struct plock_async_data *op_data;
 107         struct dlm_ls *ls;
 108         struct plock_op *op;
 109         int rv;
 110
 111         ls = dlm_find_lockspace_local(lockspace);
 112         if (!ls)
 113                 return -EINVAL;
 114
 115         op = kzalloc(sizeof(*op), GFP_NOFS);
 116         if (!op) {
 117                 rv = -ENOMEM;
 118                 goto out;
 119         }
 120
 121         op->info.optype         = DLM_PLOCK_OP_LOCK;
 122         op->info.pid            = fl->fl_pid;
 123         op->info.ex             = (fl->fl_type == F_WRLCK);
 124         op->info.wait           = IS_SETLKW(cmd);
 125         op->info.fsid           = ls->ls_global_id;
 126         op->info.number         = number;
 127         op->info.start          = fl->fl_start;
 128         op->info.end            = fl->fl_end;
 129         /* async handling */
 130         if (fl->fl_lmops && fl->fl_lmops->lm_grant) {
 131                 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
 132                 if (!op_data) {
 133                         dlm_release_plock_op(op);
 134                         rv = -ENOMEM;
 135                         goto out;
 136                 }
 137
 138                 /* fl_owner is lockd which doesn't distinguish
 139                    processes on the nfs client */
 140                 op->info.owner  = (__u64) fl->fl_pid;
 141                 op_data->callback = fl->fl_lmops->lm_grant;
 142                 locks_init_lock(&op_data->flc);
 143                 locks_copy_lock(&op_data->flc, fl);
 144                 op_data->fl             = fl;
 145                 op_data->file   = file;
 146
 147                 op->data = op_data;
 148
 149                 send_op(op);
 150                 rv = FILE_LOCK_DEFERRED;
 151                 goto out;
 152         } else {
 153                 op->info.owner  = (__u64)(long) fl->fl_owner;
 154         }
 155
 156         send_op(op);
 157
 158         if (op->info.wait) {
 159                 rv = wait_event_killable(recv_wq, (op->done != 0));
 160                 if (rv == -ERESTARTSYS) {
 161                         spin_lock(&ops_lock);
 162                         /* recheck under ops_lock if we got a done != 0,
 163                          * if so this interrupt case should be ignored
 164                          */
 165                         if (op->done != 0) {
 166                                 spin_unlock(&ops_lock);
 167                                 goto do_lock_wait;
 168                         }
 169                         list_del(&op->list);
 170                         spin_unlock(&ops_lock);
 171
 172                         log_debug(ls, "%s: wait interrupted %x %llx pid %d",
 173                                   __func__, ls->ls_global_id,
 174                                   (unsigned long long)number, op->info.pid);
 175                         do_unlock_close(&op->info);
 176                         dlm_release_plock_op(op);
 177                         goto out;
 178                 }
 179         } else {
 180                 wait_event(recv_wq, (op->done != 0));
 181         }
 182
 183 do_lock_wait:
 184
 185         WARN_ON(!list_empty(&op->list));
 186
 187         rv = op->info.rv;
 188
 189         if (!rv) {
 190                 if (locks_lock_file_wait(file, fl) < 0)
 191                         log_error(ls, "dlm_posix_lock: vfs lock error %llx",
 192                                   (unsigned long long)number);
 193         }
 194
 195         dlm_release_plock_op(op);
 196 out:
 197         dlm_put_lockspace(ls);
 198         return rv;
 199 }
 200 EXPORT_SYMBOL_GPL(dlm_posix_lock);
 201
 202 /* Returns failure iff a successful lock operation should be canceled */
 203 static int dlm_plock_callback(struct plock_op *op)
 204 {
 205         struct plock_async_data *op_data = op->data;
 206         struct file *file;
 207         struct file_lock *fl;
 208         struct file_lock *flc;
 209         int (*notify)(struct file_lock *fl, int result) = NULL;
 210         int rv = 0;
 211
 212         WARN_ON(!list_empty(&op->list));
 213
 214         /* check if the following 2 are still valid or make a copy */
 215         file = op_data->file;
 216         flc = &op_data->flc;
 217         fl = op_data->fl;
 218         notify = op_data->callback;
 219
 220         if (op->info.rv) {
 221                 notify(fl, op->info.rv);
 222                 goto out;
 223         }
 224
 225         /* got fs lock; bookkeep locally as well: */
 226         flc->fl_flags &= ~FL_SLEEP;
 227         if (posix_lock_file(file, flc, NULL)) {
 228                 /*
 229                  * This can only happen in the case of kmalloc() failure.
 230                  * The filesystem's own lock is the authoritative lock,
 231                  * so a failure to get the lock locally is not a disaster.
 232                  * As long as the fs cannot reliably cancel locks (especially
 233                  * in a low-memory situation), we're better off ignoring
 234                  * this failure than trying to recover.
 235                  */
 236                 log_print("dlm_plock_callback: vfs lock error %llx file %p fl %p",
 237                           (unsigned long long)op->info.number, file, fl);
 238         }
 239
 240         rv = notify(fl, 0);
 241         if (rv) {
 242                 /* XXX: We need to cancel the fs lock here: */
 243                 log_print("dlm_plock_callback: lock granted after lock request "
 244                           "failed; dangling lock!\n");
 245                 goto out;
 246         }
 247
 248 out:
 249         dlm_release_plock_op(op);
 250         return rv;
 251 }
 252
 253 int dlm_posix_unlock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
 254                      struct file_lock *fl)
 255 {
 256         struct dlm_ls *ls;
 257         struct plock_op *op;
 258         int rv;
 259         unsigned char fl_flags = fl->fl_flags;
 260
 261         ls = dlm_find_lockspace_local(lockspace);
 262         if (!ls)
 263                 return -EINVAL;
 264
 265         op = kzalloc(sizeof(*op), GFP_NOFS);
 266         if (!op) {
 267                 rv = -ENOMEM;
 268                 goto out;
 269         }
 270
 271         /* cause the vfs unlock to return ENOENT if lock is not found */
 272         fl->fl_flags |= FL_EXISTS;
 273
 274         rv = locks_lock_file_wait(file, fl);
 275         if (rv == -ENOENT) {
 276                 rv = 0;
 277                 goto out_free;
 278         }
 279         if (rv < 0) {
 280                 log_error(ls, "dlm_posix_unlock: vfs unlock error %d %llx",
 281                           rv, (unsigned long long)number);
 282         }
 283
 284         op->info.optype         = DLM_PLOCK_OP_UNLOCK;
 285         op->info.pid            = fl->fl_pid;
 286         op->info.fsid           = ls->ls_global_id;
 287         op->info.number         = number;
 288         op->info.start          = fl->fl_start;
 289         op->info.end            = fl->fl_end;
 290         if (fl->fl_lmops && fl->fl_lmops->lm_grant)
 291                 op->info.owner  = (__u64) fl->fl_pid;
 292         else
 293                 op->info.owner  = (__u64)(long) fl->fl_owner;
 294
 295         if (fl->fl_flags & FL_CLOSE) {
 296                 op->info.flags |= DLM_PLOCK_FL_CLOSE;
 297                 send_op(op);
 298                 rv = 0;
 299                 goto out;
 300         }
 301
 302         send_op(op);
 303         wait_event(recv_wq, (op->done != 0));
 304
 305         WARN_ON(!list_empty(&op->list));
 306
 307         rv = op->info.rv;
 308
 309         if (rv == -ENOENT)
 310                 rv = 0;
 311
 312 out_free:
 313         dlm_release_plock_op(op);
 314 out:
 315         dlm_put_lockspace(ls);
 316         fl->fl_flags = fl_flags;
 317         return rv;
 318 }
 319 EXPORT_SYMBOL_GPL(dlm_posix_unlock);
 320
 321 int dlm_posix_get(dlm_lockspace_t *lockspace, u64 number, struct file *file,
 322                   struct file_lock *fl)
 323 {
 324         struct dlm_ls *ls;
 325         struct plock_op *op;
 326         int rv;
 327
 328         ls = dlm_find_lockspace_local(lockspace);
 329         if (!ls)
 330                 return -EINVAL;
 331
 332         op = kzalloc(sizeof(*op), GFP_NOFS);
 333         if (!op) {
 334                 rv = -ENOMEM;
 335                 goto out;
 336         }
 337
 338         op->info.optype         = DLM_PLOCK_OP_GET;
 339         op->info.pid            = fl->fl_pid;
 340         op->info.ex             = (fl->fl_type == F_WRLCK);
 341         op->info.fsid           = ls->ls_global_id;
 342         op->info.number         = number;
 343         op->info.start          = fl->fl_start;
 344         op->info.end            = fl->fl_end;
 345         if (fl->fl_lmops && fl->fl_lmops->lm_grant)
 346                 op->info.owner  = (__u64) fl->fl_pid;
 347         else
 348                 op->info.owner  = (__u64)(long) fl->fl_owner;
 349
 350         send_op(op);
 351         wait_event(recv_wq, (op->done != 0));
 352
 353         WARN_ON(!list_empty(&op->list));
 354
 355         /* info.rv from userspace is 1 for conflict, 0 for no-conflict,
 356            -ENOENT if there are no locks on the file */
 357
 358         rv = op->info.rv;
 359
 360         fl->fl_type = F_UNLCK;
 361         if (rv == -ENOENT)
 362                 rv = 0;
 363         else if (rv > 0) {
 364                 locks_init_lock(fl);
 365                 fl->fl_type = (op->info.ex) ? F_WRLCK : F_RDLCK;
 366                 fl->fl_flags = FL_POSIX;
 367                 fl->fl_pid = op->info.pid;
 368                 if (op->info.nodeid != dlm_our_nodeid())
 369                         fl->fl_pid = -fl->fl_pid;
 370                 fl->fl_start = op->info.start;
 371                 fl->fl_end = op->info.end;
 372                 rv = 0;
 373         }
 374
 375         dlm_release_plock_op(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_first_entry(&send_list, struct plock_op, list);
 395                 if (op->info.flags & DLM_PLOCK_FL_CLOSE)
 396                         list_del(&op->list);
 397                 else
 398                         list_move_tail(&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                 dlm_release_plock_op(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 plock_op *op = NULL, *iter;
 424         struct dlm_plock_info info;
 425         int 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         /*
 437          * The results for waiting ops (SETLKW) can be returned in any
 438          * order, so match all fields to find the op.  The results for
 439          * non-waiting ops are returned in the order that they were sent
 440          * to userspace, so match the result with the first non-waiting op.
 441          */
 442         spin_lock(&ops_lock);
 443         if (info.wait) {
 444                 list_for_each_entry(iter, &recv_list, list) {
 445                         if (iter->info.fsid == info.fsid &&
 446                             iter->info.number == info.number &&
 447                             iter->info.owner == info.owner &&
 448                             iter->info.pid == info.pid &&
 449                             iter->info.start == info.start &&
 450                             iter->info.end == info.end &&
 451                             iter->info.ex == info.ex &&
 452                             iter->info.wait) {
 453                                 op = iter;
 454                                 break;
 455                         }
 456                 }
 457         } else {
 458                 list_for_each_entry(iter, &recv_list, list) {
 459                         if (!iter->info.wait) {
 460                                 op = iter;
 461                                 break;
 462                         }
 463                 }
 464         }
 465
 466         if (op) {
 467                 /* Sanity check that op and info match. */
 468                 if (info.wait)
 469                         WARN_ON(op->info.optype != DLM_PLOCK_OP_LOCK);
 470                 else
 471                         WARN_ON(op->info.fsid != info.fsid ||
 472                                 op->info.number != info.number ||
 473                                 op->info.owner != info.owner ||
 474                                 op->info.optype != info.optype);
 475
 476                 list_del_init(&op->list);
 477                 memcpy(&op->info, &info, sizeof(info));
 478                 if (op->data)
 479                         do_callback = 1;
 480                 else
 481                         op->done = 1;
 482         }
 483         spin_unlock(&ops_lock);
 484
 485         if (op) {
 486                 if (do_callback)
 487                         dlm_plock_callback(op);
 488                 else
 489                         wake_up(&recv_wq);
 490         } else
 491                 pr_debug("%s: no op %x %llx", __func__,
 492                          info.fsid, (unsigned long long)info.number);
 493         return count;
 494 }
 495
 496 static __poll_t dev_poll(struct file *file, poll_table *wait)
 497 {
 498         __poll_t mask = 0;
 499
 500         poll_wait(file, &send_wq, wait);
 501
 502         spin_lock(&ops_lock);
 503         if (!list_empty(&send_list))
 504                 mask = EPOLLIN | EPOLLRDNORM;
 505         spin_unlock(&ops_lock);
 506
 507         return mask;
 508 }
 509
 510 static const struct file_operations dev_fops = {
 511         .read    = dev_read,
 512         .write   = dev_write,
 513         .poll    = dev_poll,
 514         .owner   = THIS_MODULE,
 515         .llseek  = noop_llseek,
 516 };
 517
 518 static struct miscdevice plock_dev_misc = {
 519         .minor = MISC_DYNAMIC_MINOR,
 520         .name = DLM_PLOCK_MISC_NAME,
 521         .fops = &dev_fops
 522 };
 523
 524 int dlm_plock_init(void)
 525 {
 526         int rv;
 527
 528         rv = misc_register(&plock_dev_misc);
 529         if (rv)
 530                 log_print("dlm_plock_init: misc_register failed %d", rv);
 531         return rv;
 532 }
 533
 534 void dlm_plock_exit(void)
 535 {
 536         misc_deregister(&plock_dev_misc);
 537 }
 538