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                 /* XXX: We need to cancel the fs lock here: */
 194                 log_print("dlm_plock_callback: lock granted after lock request "
 195                           "failed; dangling lock!\n");
 196                 goto out;
 197         }
 198
 199 out:
 200         kfree(xop);
 201         return rv;
 202 }
 203
 204 int dlm_posix_unlock(dlm_lockspace_t *lockspace, u64 number, struct file *file,
 205                      struct file_lock *fl)
 206 {
 207         struct dlm_ls *ls;
 208         struct plock_op *op;
 209         int rv;
 210
 211         ls = dlm_find_lockspace_local(lockspace);
 212         if (!ls)
 213                 return -EINVAL;
 214
 215         op = kzalloc(sizeof(*op), GFP_NOFS);
 216         if (!op) {
 217                 rv = -ENOMEM;
 218                 goto out;
 219         }
 220
 221         if (posix_lock_file_wait(file, fl) < 0)
 222                 log_error(ls, "dlm_posix_unlock: vfs unlock error %llx",
 223                           (unsigned long long)number);
 224
 225         op->info.optype         = DLM_PLOCK_OP_UNLOCK;
 226         op->info.pid            = fl->fl_pid;
 227         op->info.fsid           = ls->ls_global_id;
 228         op->info.number         = number;
 229         op->info.start          = fl->fl_start;
 230         op->info.end            = fl->fl_end;
 231         if (fl->fl_lmops && fl->fl_lmops->fl_grant)
 232                 op->info.owner  = (__u64) fl->fl_pid;
 233         else
 234                 op->info.owner  = (__u64)(long) fl->fl_owner;
 235
 236         send_op(op);
 237         wait_event(recv_wq, (op->done != 0));
 238
 239         spin_lock(&ops_lock);
 240         if (!list_empty(&op->list)) {
 241                 log_error(ls, "dlm_posix_unlock: op on list %llx",
 242                           (unsigned long long)number);
 243                 list_del(&op->list);
 244         }
 245         spin_unlock(&ops_lock);
 246
 247         rv = op->info.rv;
 248
 249         if (rv == -ENOENT)
 250                 rv = 0;
 251
 252         kfree(op);
 253 out:
 254         dlm_put_lockspace(ls);
 255         return rv;
 256 }
 257 EXPORT_SYMBOL_GPL(dlm_posix_unlock);
 258
 259 int dlm_posix_get(dlm_lockspace_t *lockspace, u64 number, struct file *file,
 260                   struct file_lock *fl)
 261 {
 262         struct dlm_ls *ls;
 263         struct plock_op *op;
 264         int rv;
 265
 266         ls = dlm_find_lockspace_local(lockspace);
 267         if (!ls)
 268                 return -EINVAL;
 269
 270         op = kzalloc(sizeof(*op), GFP_NOFS);
 271         if (!op) {
 272                 rv = -ENOMEM;
 273                 goto out;
 274         }
 275
 276         op->info.optype         = DLM_PLOCK_OP_GET;
 277         op->info.pid            = fl->fl_pid;
 278         op->info.ex             = (fl->fl_type == F_WRLCK);
 279         op->info.fsid           = ls->ls_global_id;
 280         op->info.number         = number;
 281         op->info.start          = fl->fl_start;
 282         op->info.end            = fl->fl_end;
 283         if (fl->fl_lmops && fl->fl_lmops->fl_grant)
 284                 op->info.owner  = (__u64) fl->fl_pid;
 285         else
 286                 op->info.owner  = (__u64)(long) fl->fl_owner;
 287
 288         send_op(op);
 289         wait_event(recv_wq, (op->done != 0));
 290
 291         spin_lock(&ops_lock);
 292         if (!list_empty(&op->list)) {
 293                 log_error(ls, "dlm_posix_get: op on list %llx",
 294                           (unsigned long long)number);
 295                 list_del(&op->list);
 296         }
 297         spin_unlock(&ops_lock);
 298
 299         /* info.rv from userspace is 1 for conflict, 0 for no-conflict,
 300            -ENOENT if there are no locks on the file */
 301
 302         rv = op->info.rv;
 303
 304         fl->fl_type = F_UNLCK;
 305         if (rv == -ENOENT)
 306                 rv = 0;
 307         else if (rv > 0) {
 308                 locks_init_lock(fl);
 309                 fl->fl_type = (op->info.ex) ? F_WRLCK : F_RDLCK;
 310                 fl->fl_flags = FL_POSIX;
 311                 fl->fl_pid = op->info.pid;
 312                 fl->fl_start = op->info.start;
 313                 fl->fl_end = op->info.end;
 314                 rv = 0;
 315         }
 316
 317         kfree(op);
 318 out:
 319         dlm_put_lockspace(ls);
 320         return rv;
 321 }
 322 EXPORT_SYMBOL_GPL(dlm_posix_get);
 323
 324 /* a read copies out one plock request from the send list */
 325 static ssize_t dev_read(struct file *file, char __user *u, size_t count,
 326                         loff_t *ppos)
 327 {
 328         struct dlm_plock_info info;
 329         struct plock_op *op = NULL;
 330
 331         if (count < sizeof(info))
 332                 return -EINVAL;
 333
 334         spin_lock(&ops_lock);
 335         if (!list_empty(&send_list)) {
 336                 op = list_entry(send_list.next, struct plock_op, list);
 337                 list_move(&op->list, &recv_list);
 338                 memcpy(&info, &op->info, sizeof(info));
 339         }
 340         spin_unlock(&ops_lock);
 341
 342         if (!op)
 343                 return -EAGAIN;
 344
 345         if (copy_to_user(u, &info, sizeof(info)))
 346                 return -EFAULT;
 347         return sizeof(info);
 348 }
 349
 350 /* a write copies in one plock result that should match a plock_op
 351    on the recv list */
 352 static ssize_t dev_write(struct file *file, const char __user *u, size_t count,
 353                          loff_t *ppos)
 354 {
 355         struct dlm_plock_info info;
 356         struct plock_op *op;
 357         int found = 0, do_callback = 0;
 358
 359         if (count != sizeof(info))
 360                 return -EINVAL;
 361
 362         if (copy_from_user(&info, u, sizeof(info)))
 363                 return -EFAULT;
 364
 365         if (check_version(&info))
 366                 return -EINVAL;
 367
 368         spin_lock(&ops_lock);
 369         list_for_each_entry(op, &recv_list, list) {
 370                 if (op->info.fsid == info.fsid &&
 371                     op->info.number == info.number &&
 372                     op->info.owner == info.owner) {
 373                         struct plock_xop *xop = (struct plock_xop *)op;
 374                         list_del_init(&op->list);
 375                         memcpy(&op->info, &info, sizeof(info));
 376                         if (xop->callback)
 377                                 do_callback = 1;
 378                         else
 379                                 op->done = 1;
 380                         found = 1;
 381                         break;
 382                 }
 383         }
 384         spin_unlock(&ops_lock);
 385
 386         if (found) {
 387                 if (do_callback)
 388                         dlm_plock_callback(op);
 389                 else
 390                         wake_up(&recv_wq);
 391         } else
 392                 log_print("dev_write no op %x %llx", info.fsid,
 393                           (unsigned long long)info.number);
 394         return count;
 395 }
 396
 397 static unsigned int dev_poll(struct file *file, poll_table *wait)
 398 {
 399         unsigned int mask = 0;
 400
 401         poll_wait(file, &send_wq, wait);
 402
 403         spin_lock(&ops_lock);
 404         if (!list_empty(&send_list))
 405                 mask = POLLIN | POLLRDNORM;
 406         spin_unlock(&ops_lock);
 407
 408         return mask;
 409 }
 410
 411 static const struct file_operations dev_fops = {
 412         .read    = dev_read,
 413         .write   = dev_write,
 414         .poll    = dev_poll,
 415         .owner   = THIS_MODULE,
 416         .llseek  = noop_llseek,
 417 };
 418
 419 static struct miscdevice plock_dev_misc = {
 420         .minor = MISC_DYNAMIC_MINOR,
 421         .name = DLM_PLOCK_MISC_NAME,
 422         .fops = &dev_fops
 423 };
 424
 425 int dlm_plock_init(void)
 426 {
 427         int rv;
 428
 429         spin_lock_init(&ops_lock);
 430         INIT_LIST_HEAD(&send_list);
 431         INIT_LIST_HEAD(&recv_list);
 432         init_waitqueue_head(&send_wq);
 433         init_waitqueue_head(&recv_wq);
 434
 435         rv = misc_register(&plock_dev_misc);
 436         if (rv)
 437                 log_print("dlm_plock_init: misc_register failed %d", rv);
 438         return rv;
 439 }
 440
 441 void dlm_plock_exit(void)
 442 {
 443         if (misc_deregister(&plock_dev_misc) < 0)
 444                 log_print("dlm_plock_exit: misc_deregister failed");
 445 }
 446