fs/gfs2/locking/dlm/plock.c

   1 /*
   2  * Copyright (C) 2005 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/miscdevice.h>
  10 #include <linux/lock_dlm_plock.h>
  11 #include <linux/poll.h>
  12
  13 #include "lock_dlm.h"
  14
  15
  16 static spinlock_t ops_lock;
  17 static struct list_head send_list;
  18 static struct list_head recv_list;
  19 static wait_queue_head_t send_wq;
  20 static wait_queue_head_t recv_wq;
  21
  22 struct plock_op {
  23         struct list_head list;
  24         int done;
  25         struct gdlm_plock_info info;
  26 };
  27
  28 struct plock_xop {
  29         struct plock_op xop;
  30         void *callback;
  31         void *fl;
  32         void *file;
  33         struct file_lock flc;
  34 };
  35
  36
  37 static inline void set_version(struct gdlm_plock_info *info)
  38 {
  39         info->version[0] = GDLM_PLOCK_VERSION_MAJOR;
  40         info->version[1] = GDLM_PLOCK_VERSION_MINOR;
  41         info->version[2] = GDLM_PLOCK_VERSION_PATCH;
  42 }
  43
  44 static int check_version(struct gdlm_plock_info *info)
  45 {
  46         if ((GDLM_PLOCK_VERSION_MAJOR != info->version[0]) ||
  47             (GDLM_PLOCK_VERSION_MINOR < info->version[1])) {
  48                 log_error("plock device version mismatch: "
  49                           "kernel (%u.%u.%u), user (%u.%u.%u)",
  50                           GDLM_PLOCK_VERSION_MAJOR,
  51                           GDLM_PLOCK_VERSION_MINOR,
  52                           GDLM_PLOCK_VERSION_PATCH,
  53                           info->version[0],
  54                           info->version[1],
  55                           info->version[2]);
  56                 return -EINVAL;
  57         }
  58         return 0;
  59 }
  60
  61 static void send_op(struct plock_op *op)
  62 {
  63         set_version(&op->info);
  64         INIT_LIST_HEAD(&op->list);
  65         spin_lock(&ops_lock);
  66         list_add_tail(&op->list, &send_list);
  67         spin_unlock(&ops_lock);
  68         wake_up(&send_wq);
  69 }
  70
  71 int gdlm_plock(void *lockspace, struct lm_lockname *name,
  72                struct file *file, int cmd, struct file_lock *fl)
  73 {
  74         struct gdlm_ls *ls = lockspace;
  75         struct plock_op *op;
  76         struct plock_xop *xop;
  77         int rv;
  78
  79         xop = kzalloc(sizeof(*xop), GFP_KERNEL);
  80         if (!xop)
  81                 return -ENOMEM;
  82
  83         op = &xop->xop;
  84         op->info.optype         = GDLM_PLOCK_OP_LOCK;
  85         op->info.pid            = fl->fl_pid;
  86         op->info.ex             = (fl->fl_type == F_WRLCK);
  87         op->info.wait           = IS_SETLKW(cmd);
  88         op->info.fsid           = ls->id;
  89         op->info.number         = name->ln_number;
  90         op->info.start          = fl->fl_start;
  91         op->info.end            = fl->fl_end;
  92         op->info.owner          = (__u64)(long) fl->fl_owner;
  93         if (fl->fl_lmops && fl->fl_lmops->fl_grant) {
  94                 xop->callback   = fl->fl_lmops->fl_grant;
  95                 locks_init_lock(&xop->flc);
  96                 locks_copy_lock(&xop->flc, fl);
  97                 xop->fl         = fl;
  98                 xop->file       = file;
  99         } else
 100                 xop->callback   = NULL;
 101
 102         send_op(op);
 103
 104         if (xop->callback == NULL)
 105                 wait_event(recv_wq, (op->done != 0));
 106         else
 107                 return -EINPROGRESS;
 108
 109         spin_lock(&ops_lock);
 110         if (!list_empty(&op->list)) {
 111                 printk(KERN_INFO "plock op on list\n");
 112                 list_del(&op->list);
 113         }
 114         spin_unlock(&ops_lock);
 115
 116         rv = op->info.rv;
 117
 118         if (!rv) {
 119                 if (posix_lock_file_wait(file, fl) < 0)
 120                         log_error("gdlm_plock: vfs lock error %x,%llx",
 121                                   name->ln_type,
 122                                   (unsigned long long)name->ln_number);
 123         }
 124
 125         kfree(xop);
 126         return rv;
 127 }
 128
 129 /* Returns failure iff a succesful lock operation should be canceled */
 130 static int gdlm_plock_callback(struct plock_op *op)
 131 {
 132         struct file *file;
 133         struct file_lock *fl;
 134         struct file_lock *flc;
 135         int (*notify)(void *, void *, int) = NULL;
 136         struct plock_xop *xop = (struct plock_xop *)op;
 137         int rv = 0;
 138
 139         spin_lock(&ops_lock);
 140         if (!list_empty(&op->list)) {
 141                 printk(KERN_INFO "plock op on list\n");
 142                 list_del(&op->list);
 143         }
 144         spin_unlock(&ops_lock);
 145
 146         /* check if the following 2 are still valid or make a copy */
 147         file = xop->file;
 148         flc = &xop->flc;
 149         fl = xop->fl;
 150         notify = xop->callback;
 151
 152         if (op->info.rv) {
 153                 notify(flc, NULL, op->info.rv);
 154                 goto out;
 155         }
 156
 157         /* got fs lock; bookkeep locally as well: */
 158         flc->fl_flags &= ~FL_SLEEP;
 159         if (posix_lock_file(file, flc, NULL)) {
 160                 /*
 161                  * This can only happen in the case of kmalloc() failure.
 162                  * The filesystem's own lock is the authoritative lock,
 163                  * so a failure to get the lock locally is not a disaster.
 164                  * As long as GFS cannot reliably cancel locks (especially
 165                  * in a low-memory situation), we're better off ignoring
 166                  * this failure than trying to recover.
 167                  */
 168                 log_error("gdlm_plock: vfs lock error file %p fl %p",
 169                                 file, fl);
 170         }
 171
 172         rv = notify(flc, NULL, 0);
 173         if (rv) {
 174                 /* XXX: We need to cancel the fs lock here: */
 175                 printk("gfs2 lock granted after lock request failed;"
 176                                                 " dangling lock!\n");
 177                 goto out;
 178         }
 179
 180 out:
 181         kfree(xop);
 182         return rv;
 183 }
 184
 185 int gdlm_punlock(void *lockspace, struct lm_lockname *name,
 186                  struct file *file, struct file_lock *fl)
 187 {
 188         struct gdlm_ls *ls = lockspace;
 189         struct plock_op *op;
 190         int rv;
 191
 192         op = kzalloc(sizeof(*op), GFP_KERNEL);
 193         if (!op)
 194                 return -ENOMEM;
 195
 196         if (posix_lock_file_wait(file, fl) < 0)
 197                 log_error("gdlm_punlock: vfs unlock error %x,%llx",
 198                           name->ln_type, (unsigned long long)name->ln_number);
 199
 200         op->info.optype         = GDLM_PLOCK_OP_UNLOCK;
 201         op->info.pid            = fl->fl_pid;
 202         op->info.fsid           = ls->id;
 203         op->info.number         = name->ln_number;
 204         op->info.start          = fl->fl_start;
 205         op->info.end            = fl->fl_end;
 206         op->info.owner          = (__u64)(long) fl->fl_owner;
 207
 208         send_op(op);
 209         wait_event(recv_wq, (op->done != 0));
 210
 211         spin_lock(&ops_lock);
 212         if (!list_empty(&op->list)) {
 213                 printk(KERN_INFO "punlock op on list\n");
 214                 list_del(&op->list);
 215         }
 216         spin_unlock(&ops_lock);
 217
 218         rv = op->info.rv;
 219
 220         if (rv == -ENOENT)
 221                 rv = 0;
 222
 223         kfree(op);
 224         return rv;
 225 }
 226
 227 int gdlm_plock_get(void *lockspace, struct lm_lockname *name,
 228                    struct file *file, struct file_lock *fl)
 229 {
 230         struct gdlm_ls *ls = lockspace;
 231         struct plock_op *op;
 232         int rv;
 233
 234         op = kzalloc(sizeof(*op), GFP_KERNEL);
 235         if (!op)
 236                 return -ENOMEM;
 237
 238         op->info.optype         = GDLM_PLOCK_OP_GET;
 239         op->info.pid            = fl->fl_pid;
 240         op->info.ex             = (fl->fl_type == F_WRLCK);
 241         op->info.fsid           = ls->id;
 242         op->info.number         = name->ln_number;
 243         op->info.start          = fl->fl_start;
 244         op->info.end            = fl->fl_end;
 245
 246
 247         send_op(op);
 248         wait_event(recv_wq, (op->done != 0));
 249
 250         spin_lock(&ops_lock);
 251         if (!list_empty(&op->list)) {
 252                 printk(KERN_INFO "plock_get op on list\n");
 253                 list_del(&op->list);
 254         }
 255         spin_unlock(&ops_lock);
 256
 257         rv = op->info.rv;
 258
 259         fl->fl_type = F_UNLCK;
 260         if (rv == -ENOENT)
 261                 rv = 0;
 262         else if (rv == 0 && op->info.pid != fl->fl_pid) {
 263                 fl->fl_type = (op->info.ex) ? F_WRLCK : F_RDLCK;
 264                 fl->fl_pid = op->info.pid;
 265                 fl->fl_start = op->info.start;
 266                 fl->fl_end = op->info.end;
 267         }
 268
 269         kfree(op);
 270         return rv;
 271 }
 272
 273 /* a read copies out one plock request from the send list */
 274 static ssize_t dev_read(struct file *file, char __user *u, size_t count,
 275                         loff_t *ppos)
 276 {
 277         struct gdlm_plock_info info;
 278         struct plock_op *op = NULL;
 279
 280         if (count < sizeof(info))
 281                 return -EINVAL;
 282
 283         spin_lock(&ops_lock);
 284         if (!list_empty(&send_list)) {
 285                 op = list_entry(send_list.next, struct plock_op, list);
 286                 list_move(&op->list, &recv_list);
 287                 memcpy(&info, &op->info, sizeof(info));
 288         }
 289         spin_unlock(&ops_lock);
 290
 291         if (!op)
 292                 return -EAGAIN;
 293
 294         if (copy_to_user(u, &info, sizeof(info)))
 295                 return -EFAULT;
 296         return sizeof(info);
 297 }
 298
 299 /* a write copies in one plock result that should match a plock_op
 300    on the recv list */
 301 static ssize_t dev_write(struct file *file, const char __user *u, size_t count,
 302                          loff_t *ppos)
 303 {
 304         struct gdlm_plock_info info;
 305         struct plock_op *op;
 306         int found = 0;
 307
 308         if (count != sizeof(info))
 309                 return -EINVAL;
 310
 311         if (copy_from_user(&info, u, sizeof(info)))
 312                 return -EFAULT;
 313
 314         if (check_version(&info))
 315                 return -EINVAL;
 316
 317         spin_lock(&ops_lock);
 318         list_for_each_entry(op, &recv_list, list) {
 319                 if (op->info.fsid == info.fsid && op->info.number == info.number &&
 320                     op->info.owner == info.owner) {
 321                         list_del_init(&op->list);
 322                         found = 1;
 323                         op->done = 1;
 324                         memcpy(&op->info, &info, sizeof(info));
 325                         break;
 326                 }
 327         }
 328         spin_unlock(&ops_lock);
 329
 330         if (found) {
 331                 struct plock_xop *xop;
 332                 xop = (struct plock_xop *)op;
 333                 if (xop->callback)
 334                         count = gdlm_plock_callback(op);
 335                 else
 336                         wake_up(&recv_wq);
 337         } else
 338                 printk(KERN_INFO "gdlm dev_write no op %x %llx\n", info.fsid,
 339                         (unsigned long long)info.number);
 340         return count;
 341 }
 342
 343 static unsigned int dev_poll(struct file *file, poll_table *wait)
 344 {
 345         poll_wait(file, &send_wq, wait);
 346
 347         spin_lock(&ops_lock);
 348         if (!list_empty(&send_list)) {
 349                 spin_unlock(&ops_lock);
 350                 return POLLIN | POLLRDNORM;
 351         }
 352         spin_unlock(&ops_lock);
 353         return 0;
 354 }
 355
 356 static const struct file_operations dev_fops = {
 357         .read    = dev_read,
 358         .write   = dev_write,
 359         .poll    = dev_poll,
 360         .owner   = THIS_MODULE
 361 };
 362
 363 static struct miscdevice plock_dev_misc = {
 364         .minor = MISC_DYNAMIC_MINOR,
 365         .name = GDLM_PLOCK_MISC_NAME,
 366         .fops = &dev_fops
 367 };
 368
 369 int gdlm_plock_init(void)
 370 {
 371         int rv;
 372
 373         spin_lock_init(&ops_lock);
 374         INIT_LIST_HEAD(&send_list);
 375         INIT_LIST_HEAD(&recv_list);
 376         init_waitqueue_head(&send_wq);
 377         init_waitqueue_head(&recv_wq);
 378
 379         rv = misc_register(&plock_dev_misc);
 380         if (rv)
 381                 printk(KERN_INFO "gdlm_plock_init: misc_register failed %d",
 382                        rv);
 383         return rv;
 384 }
 385
 386 void gdlm_plock_exit(void)
 387 {
 388         if (misc_deregister(&plock_dev_misc) < 0)
 389                 printk(KERN_INFO "gdlm_plock_exit: misc_deregister failed");
 390 }
 391