ipc/mqueue.c

   1 /*
   2  * POSIX message queues filesystem for Linux.
   3  *
   4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
   5  *                          Michal Wronski          (Michal.Wronski@motorola.com)
   6  *
   7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
   8  * Lockless receive & send, fd based notify:
   9  *                          Manfred Spraul          (manfred@colorfullife.com)
  10  *
  11  * This file is released under the GPL.
  12  */
  13
  14 #include <linux/capability.h>
  15 #include <linux/init.h>
  16 #include <linux/pagemap.h>
  17 #include <linux/file.h>
  18 #include <linux/mount.h>
  19 #include <linux/namei.h>
  20 #include <linux/sysctl.h>
  21 #include <linux/poll.h>
  22 #include <linux/mqueue.h>
  23 #include <linux/msg.h>
  24 #include <linux/skbuff.h>
  25 #include <linux/netlink.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/signal.h>
  28 #include <linux/mutex.h>
  29
  30 #include <net/sock.h>
  31 #include "util.h"
  32
  33 #define MQUEUE_MAGIC    0x19800202
  34 #define DIRENT_SIZE     20
  35 #define FILENT_SIZE     80
  36
  37 #define SEND            0
  38 #define RECV            1
  39
  40 #define STATE_NONE      0
  41 #define STATE_PENDING   1
  42 #define STATE_READY     2
  43
  44 /* used by sysctl */
  45 #define FS_MQUEUE       1
  46 #define CTL_QUEUESMAX   2
  47 #define CTL_MSGMAX      3
  48 #define CTL_MSGSIZEMAX  4
  49
  50 /* default values */
  51 #define DFLT_QUEUESMAX  256     /* max number of message queues */
  52 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
  53 #define HARD_MSGMAX     (131072/sizeof(void*))
  54 #define DFLT_MSGSIZEMAX 8192    /* max message size */
  55
  56
  57 struct ext_wait_queue {         /* queue of sleeping tasks */
  58         struct task_struct *task;
  59         struct list_head list;
  60         struct msg_msg *msg;    /* ptr of loaded message */
  61         int state;              /* one of STATE_* values */
  62 };
  63
  64 struct mqueue_inode_info {
  65         spinlock_t lock;
  66         struct inode vfs_inode;
  67         wait_queue_head_t wait_q;
  68
  69         struct msg_msg **messages;
  70         struct mq_attr attr;
  71
  72         struct sigevent notify;
  73         pid_t notify_owner;
  74         struct user_struct *user;       /* user who created, for accounting */
  75         struct sock *notify_sock;
  76         struct sk_buff *notify_cookie;
  77
  78         /* for tasks waiting for free space and messages, respectively */
  79         struct ext_wait_queue e_wait_q[2];
  80
  81         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  82 };
  83
  84 static struct inode_operations mqueue_dir_inode_operations;
  85 static struct file_operations mqueue_file_operations;
  86 static struct super_operations mqueue_super_ops;
  87 static void remove_notification(struct mqueue_inode_info *info);
  88
  89 static spinlock_t mq_lock;
  90 static kmem_cache_t *mqueue_inode_cachep;
  91 static struct vfsmount *mqueue_mnt;
  92
  93 static unsigned int queues_count;
  94 static unsigned int queues_max  = DFLT_QUEUESMAX;
  95 static unsigned int msg_max     = DFLT_MSGMAX;
  96 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
  97
  98 static struct ctl_table_header * mq_sysctl_table;
  99
 100 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
 101 {
 102         return container_of(inode, struct mqueue_inode_info, vfs_inode);
 103 }
 104
 105 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 106                                                         struct mq_attr *attr)
 107 {
 108         struct inode *inode;
 109
 110         inode = new_inode(sb);
 111         if (inode) {
 112                 inode->i_mode = mode;
 113                 inode->i_uid = current->fsuid;
 114                 inode->i_gid = current->fsgid;
 115                 inode->i_blksize = PAGE_CACHE_SIZE;
 116                 inode->i_blocks = 0;
 117                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 118                                 CURRENT_TIME;
 119
 120                 if (S_ISREG(mode)) {
 121                         struct mqueue_inode_info *info;
 122                         struct task_struct *p = current;
 123                         struct user_struct *u = p->user;
 124                         unsigned long mq_bytes, mq_msg_tblsz;
 125
 126                         inode->i_fop = &mqueue_file_operations;
 127                         inode->i_size = FILENT_SIZE;
 128                         /* mqueue specific info */
 129                         info = MQUEUE_I(inode);
 130                         spin_lock_init(&info->lock);
 131                         init_waitqueue_head(&info->wait_q);
 132                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 133                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 134                         info->messages = NULL;
 135                         info->notify_owner = 0;
 136                         info->qsize = 0;
 137                         info->user = NULL;      /* set when all is ok */
 138                         memset(&info->attr, 0, sizeof(info->attr));
 139                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 140                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 141                         if (attr) {
 142                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 143                                 info->attr.mq_msgsize = attr->mq_msgsize;
 144                         }
 145                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 146                         mq_bytes = (mq_msg_tblsz +
 147                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 148
 149                         spin_lock(&mq_lock);
 150                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 151                             u->mq_bytes + mq_bytes >
 152                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
 153                                 spin_unlock(&mq_lock);
 154                                 goto out_inode;
 155                         }
 156                         u->mq_bytes += mq_bytes;
 157                         spin_unlock(&mq_lock);
 158
 159                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 160                         if (!info->messages) {
 161                                 spin_lock(&mq_lock);
 162                                 u->mq_bytes -= mq_bytes;
 163                                 spin_unlock(&mq_lock);
 164                                 goto out_inode;
 165                         }
 166                         /* all is ok */
 167                         info->user = get_uid(u);
 168                 } else if (S_ISDIR(mode)) {
 169                         inode->i_nlink++;
 170                         /* Some things misbehave if size == 0 on a directory */
 171                         inode->i_size = 2 * DIRENT_SIZE;
 172                         inode->i_op = &mqueue_dir_inode_operations;
 173                         inode->i_fop = &simple_dir_operations;
 174                 }
 175         }
 176         return inode;
 177 out_inode:
 178         make_bad_inode(inode);
 179         iput(inode);
 180         return NULL;
 181 }
 182
 183 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 184 {
 185         struct inode *inode;
 186
 187         sb->s_blocksize = PAGE_CACHE_SIZE;
 188         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 189         sb->s_magic = MQUEUE_MAGIC;
 190         sb->s_op = &mqueue_super_ops;
 191
 192         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 193         if (!inode)
 194                 return -ENOMEM;
 195
 196         sb->s_root = d_alloc_root(inode);
 197         if (!sb->s_root) {
 198                 iput(inode);
 199                 return -ENOMEM;
 200         }
 201
 202         return 0;
 203 }
 204
 205 static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
 206                                          int flags, const char *dev_name,
 207                                          void *data)
 208 {
 209         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
 210 }
 211
 212 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 213 {
 214         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 215
 216         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 217                 SLAB_CTOR_CONSTRUCTOR)
 218                 inode_init_once(&p->vfs_inode);
 219 }
 220
 221 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 222 {
 223         struct mqueue_inode_info *ei;
 224
 225         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
 226         if (!ei)
 227                 return NULL;
 228         return &ei->vfs_inode;
 229 }
 230
 231 static void mqueue_destroy_inode(struct inode *inode)
 232 {
 233         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 234 }
 235
 236 static void mqueue_delete_inode(struct inode *inode)
 237 {
 238         struct mqueue_inode_info *info;
 239         struct user_struct *user;
 240         unsigned long mq_bytes;
 241         int i;
 242
 243         if (S_ISDIR(inode->i_mode)) {
 244                 clear_inode(inode);
 245                 return;
 246         }
 247         info = MQUEUE_I(inode);
 248         spin_lock(&info->lock);
 249         for (i = 0; i < info->attr.mq_curmsgs; i++)
 250                 free_msg(info->messages[i]);
 251         kfree(info->messages);
 252         spin_unlock(&info->lock);
 253
 254         clear_inode(inode);
 255
 256         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 257                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 258         user = info->user;
 259         if (user) {
 260                 spin_lock(&mq_lock);
 261                 user->mq_bytes -= mq_bytes;
 262                 queues_count--;
 263                 spin_unlock(&mq_lock);
 264                 free_uid(user);
 265         }
 266 }
 267
 268 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 269                                 int mode, struct nameidata *nd)
 270 {
 271         struct inode *inode;
 272         struct mq_attr *attr = dentry->d_fsdata;
 273         int error;
 274
 275         spin_lock(&mq_lock);
 276         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 277                 error = -ENOSPC;
 278                 goto out_lock;
 279         }
 280         queues_count++;
 281         spin_unlock(&mq_lock);
 282
 283         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 284         if (!inode) {
 285                 error = -ENOMEM;
 286                 spin_lock(&mq_lock);
 287                 queues_count--;
 288                 goto out_lock;
 289         }
 290
 291         dir->i_size += DIRENT_SIZE;
 292         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 293
 294         d_instantiate(dentry, inode);
 295         dget(dentry);
 296         return 0;
 297 out_lock:
 298         spin_unlock(&mq_lock);
 299         return error;
 300 }
 301
 302 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 303 {
 304         struct inode *inode = dentry->d_inode;
 305
 306         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 307         dir->i_size -= DIRENT_SIZE;
 308         inode->i_nlink--;
 309         dput(dentry);
 310         return 0;
 311 }
 312
 313 /*
 314 *       This is routine for system read from queue file.
 315 *       To avoid mess with doing here some sort of mq_receive we allow
 316 *       to read only queue size & notification info (the only values
 317 *       that are interesting from user point of view and aren't accessible
 318 *       through std routines)
 319 */
 320 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 321                                 size_t count, loff_t * off)
 322 {
 323         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 324         char buffer[FILENT_SIZE];
 325         size_t slen;
 326         loff_t o;
 327
 328         if (!count)
 329                 return 0;
 330
 331         spin_lock(&info->lock);
 332         snprintf(buffer, sizeof(buffer),
 333                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 334                         info->qsize,
 335                         info->notify_owner ? info->notify.sigev_notify : 0,
 336                         (info->notify_owner &&
 337                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 338                                 info->notify.sigev_signo : 0,
 339                         info->notify_owner);
 340         spin_unlock(&info->lock);
 341         buffer[sizeof(buffer)-1] = '\0';
 342         slen = strlen(buffer)+1;
 343
 344         o = *off;
 345         if (o > slen)
 346                 return 0;
 347
 348         if (o + count > slen)
 349                 count = slen - o;
 350
 351         if (copy_to_user(u_data, buffer + o, count))
 352                 return -EFAULT;
 353
 354         *off = o + count;
 355         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
 356         return count;
 357 }
 358
 359 static int mqueue_flush_file(struct file *filp)
 360 {
 361         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 362
 363         spin_lock(&info->lock);
 364         if (current->tgid == info->notify_owner)
 365                 remove_notification(info);
 366
 367         spin_unlock(&info->lock);
 368         return 0;
 369 }
 370
 371 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 372 {
 373         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 374         int retval = 0;
 375
 376         poll_wait(filp, &info->wait_q, poll_tab);
 377
 378         spin_lock(&info->lock);
 379         if (info->attr.mq_curmsgs)
 380                 retval = POLLIN | POLLRDNORM;
 381
 382         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 383                 retval |= POLLOUT | POLLWRNORM;
 384         spin_unlock(&info->lock);
 385
 386         return retval;
 387 }
 388
 389 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 390 static void wq_add(struct mqueue_inode_info *info, int sr,
 391                         struct ext_wait_queue *ewp)
 392 {
 393         struct ext_wait_queue *walk;
 394
 395         ewp->task = current;
 396
 397         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 398                 if (walk->task->static_prio <= current->static_prio) {
 399                         list_add_tail(&ewp->list, &walk->list);
 400                         return;
 401                 }
 402         }
 403         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 404 }
 405
 406 /*
 407  * Puts current task to sleep. Caller must hold queue lock. After return
 408  * lock isn't held.
 409  * sr: SEND or RECV
 410  */
 411 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 412                         long timeout, struct ext_wait_queue *ewp)
 413 {
 414         int retval;
 415         signed long time;
 416
 417         wq_add(info, sr, ewp);
 418
 419         for (;;) {
 420                 set_current_state(TASK_INTERRUPTIBLE);
 421
 422                 spin_unlock(&info->lock);
 423                 time = schedule_timeout(timeout);
 424
 425                 while (ewp->state == STATE_PENDING)
 426                         cpu_relax();
 427
 428                 if (ewp->state == STATE_READY) {
 429                         retval = 0;
 430                         goto out;
 431                 }
 432                 spin_lock(&info->lock);
 433                 if (ewp->state == STATE_READY) {
 434                         retval = 0;
 435                         goto out_unlock;
 436                 }
 437                 if (signal_pending(current)) {
 438                         retval = -ERESTARTSYS;
 439                         break;
 440                 }
 441                 if (time == 0) {
 442                         retval = -ETIMEDOUT;
 443                         break;
 444                 }
 445         }
 446         list_del(&ewp->list);
 447 out_unlock:
 448         spin_unlock(&info->lock);
 449 out:
 450         return retval;
 451 }
 452
 453 /*
 454  * Returns waiting task that should be serviced first or NULL if none exists
 455  */
 456 static struct ext_wait_queue *wq_get_first_waiter(
 457                 struct mqueue_inode_info *info, int sr)
 458 {
 459         struct list_head *ptr;
 460
 461         ptr = info->e_wait_q[sr].list.prev;
 462         if (ptr == &info->e_wait_q[sr].list)
 463                 return NULL;
 464         return list_entry(ptr, struct ext_wait_queue, list);
 465 }
 466
 467 /* Auxiliary functions to manipulate messages' list */
 468 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 469 {
 470         int k;
 471
 472         k = info->attr.mq_curmsgs - 1;
 473         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 474                 info->messages[k + 1] = info->messages[k];
 475                 k--;
 476         }
 477         info->attr.mq_curmsgs++;
 478         info->qsize += ptr->m_ts;
 479         info->messages[k + 1] = ptr;
 480 }
 481
 482 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 483 {
 484         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 485         return info->messages[info->attr.mq_curmsgs];
 486 }
 487
 488 static inline void set_cookie(struct sk_buff *skb, char code)
 489 {
 490         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 491 }
 492
 493 /*
 494  * The next function is only to split too long sys_mq_timedsend
 495  */
 496 static void __do_notify(struct mqueue_inode_info *info)
 497 {
 498         /* notification
 499          * invoked when there is registered process and there isn't process
 500          * waiting synchronously for message AND state of queue changed from
 501          * empty to not empty. Here we are sure that no one is waiting
 502          * synchronously. */
 503         if (info->notify_owner &&
 504             info->attr.mq_curmsgs == 1) {
 505                 struct siginfo sig_i;
 506                 switch (info->notify.sigev_notify) {
 507                 case SIGEV_NONE:
 508                         break;
 509                 case SIGEV_SIGNAL:
 510                         /* sends signal */
 511
 512                         sig_i.si_signo = info->notify.sigev_signo;
 513                         sig_i.si_errno = 0;
 514                         sig_i.si_code = SI_MESGQ;
 515                         sig_i.si_value = info->notify.sigev_value;
 516                         sig_i.si_pid = current->tgid;
 517                         sig_i.si_uid = current->uid;
 518
 519                         kill_proc_info(info->notify.sigev_signo,
 520                                        &sig_i, info->notify_owner);
 521                         break;
 522                 case SIGEV_THREAD:
 523                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 524                         netlink_sendskb(info->notify_sock,
 525                                         info->notify_cookie, 0);
 526                         break;
 527                 }
 528                 /* after notification unregisters process */
 529                 info->notify_owner = 0;
 530         }
 531         wake_up(&info->wait_q);
 532 }
 533
 534 static long prepare_timeout(const struct timespec __user *u_arg)
 535 {
 536         struct timespec ts, nowts;
 537         long timeout;
 538
 539         if (u_arg) {
 540                 if (unlikely(copy_from_user(&ts, u_arg,
 541                                         sizeof(struct timespec))))
 542                         return -EFAULT;
 543
 544                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 545                         || ts.tv_nsec >= NSEC_PER_SEC))
 546                         return -EINVAL;
 547                 nowts = CURRENT_TIME;
 548                 /* first subtract as jiffies can't be too big */
 549                 ts.tv_sec -= nowts.tv_sec;
 550                 if (ts.tv_nsec < nowts.tv_nsec) {
 551                         ts.tv_nsec += NSEC_PER_SEC;
 552                         ts.tv_sec--;
 553                 }
 554                 ts.tv_nsec -= nowts.tv_nsec;
 555                 if (ts.tv_sec < 0)
 556                         return 0;
 557
 558                 timeout = timespec_to_jiffies(&ts) + 1;
 559         } else
 560                 return MAX_SCHEDULE_TIMEOUT;
 561
 562         return timeout;
 563 }
 564
 565 static void remove_notification(struct mqueue_inode_info *info)
 566 {
 567         if (info->notify_owner != 0 &&
 568             info->notify.sigev_notify == SIGEV_THREAD) {
 569                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 570                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
 571         }
 572         info->notify_owner = 0;
 573 }
 574
 575 static int mq_attr_ok(struct mq_attr *attr)
 576 {
 577         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 578                 return 0;
 579         if (capable(CAP_SYS_RESOURCE)) {
 580                 if (attr->mq_maxmsg > HARD_MSGMAX)
 581                         return 0;
 582         } else {
 583                 if (attr->mq_maxmsg > msg_max ||
 584                                 attr->mq_msgsize > msgsize_max)
 585                         return 0;
 586         }
 587         /* check for overflow */
 588         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 589                 return 0;
 590         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 591             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 592             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 593                 return 0;
 594         return 1;
 595 }
 596
 597 /*
 598  * Invoked when creating a new queue via sys_mq_open
 599  */
 600 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 601                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 602 {
 603         struct mq_attr attr;
 604         int ret;
 605
 606         if (u_attr) {
 607                 ret = -EFAULT;
 608                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 609                         goto out;
 610                 ret = -EINVAL;
 611                 if (!mq_attr_ok(&attr))
 612                         goto out;
 613                 /* store for use during create */
 614                 dentry->d_fsdata = &attr;
 615         }
 616
 617         mode &= ~current->fs->umask;
 618         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 619         dentry->d_fsdata = NULL;
 620         if (ret)
 621                 goto out;
 622
 623         return dentry_open(dentry, mqueue_mnt, oflag);
 624
 625 out:
 626         dput(dentry);
 627         mntput(mqueue_mnt);
 628         return ERR_PTR(ret);
 629 }
 630
 631 /* Opens existing queue */
 632 static struct file *do_open(struct dentry *dentry, int oflag)
 633 {
 634 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 635                                         MAY_READ | MAY_WRITE };
 636
 637         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 638                 dput(dentry);
 639                 mntput(mqueue_mnt);
 640                 return ERR_PTR(-EINVAL);
 641         }
 642
 643         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
 644                 dput(dentry);
 645                 mntput(mqueue_mnt);
 646                 return ERR_PTR(-EACCES);
 647         }
 648
 649         return dentry_open(dentry, mqueue_mnt, oflag);
 650 }
 651
 652 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 653                                 struct mq_attr __user *u_attr)
 654 {
 655         struct dentry *dentry;
 656         struct file *filp;
 657         char *name;
 658         int fd, error;
 659
 660         if (IS_ERR(name = getname(u_name)))
 661                 return PTR_ERR(name);
 662
 663         fd = get_unused_fd();
 664         if (fd < 0)
 665                 goto out_putname;
 666
 667         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 668         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 669         if (IS_ERR(dentry)) {
 670                 error = PTR_ERR(dentry);
 671                 goto out_err;
 672         }
 673         mntget(mqueue_mnt);
 674
 675         if (oflag & O_CREAT) {
 676                 if (dentry->d_inode) {  /* entry already exists */
 677                         error = -EEXIST;
 678                         if (oflag & O_EXCL)
 679                                 goto out;
 680                         filp = do_open(dentry, oflag);
 681                 } else {
 682                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 683                                                 oflag, mode, u_attr);
 684                 }
 685         } else {
 686                 error = -ENOENT;
 687                 if (!dentry->d_inode)
 688                         goto out;
 689                 filp = do_open(dentry, oflag);
 690         }
 691
 692         if (IS_ERR(filp)) {
 693                 error = PTR_ERR(filp);
 694                 goto out_putfd;
 695         }
 696
 697         set_close_on_exec(fd, 1);
 698         fd_install(fd, filp);
 699         goto out_upsem;
 700
 701 out:
 702         dput(dentry);
 703         mntput(mqueue_mnt);
 704 out_putfd:
 705         put_unused_fd(fd);
 706 out_err:
 707         fd = error;
 708 out_upsem:
 709         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 710 out_putname:
 711         putname(name);
 712         return fd;
 713 }
 714
 715 asmlinkage long sys_mq_unlink(const char __user *u_name)
 716 {
 717         int err;
 718         char *name;
 719         struct dentry *dentry;
 720         struct inode *inode = NULL;
 721
 722         name = getname(u_name);
 723         if (IS_ERR(name))
 724                 return PTR_ERR(name);
 725
 726         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 727         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 728         if (IS_ERR(dentry)) {
 729                 err = PTR_ERR(dentry);
 730                 goto out_unlock;
 731         }
 732
 733         if (!dentry->d_inode) {
 734                 err = -ENOENT;
 735                 goto out_err;
 736         }
 737
 738         inode = dentry->d_inode;
 739         if (inode)
 740                 atomic_inc(&inode->i_count);
 741
 742         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 743 out_err:
 744         dput(dentry);
 745
 746 out_unlock:
 747         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 748         putname(name);
 749         if (inode)
 750                 iput(inode);
 751
 752         return err;
 753 }
 754
 755 /* Pipelined send and receive functions.
 756  *
 757  * If a receiver finds no waiting message, then it registers itself in the
 758  * list of waiting receivers. A sender checks that list before adding the new
 759  * message into the message array. If there is a waiting receiver, then it
 760  * bypasses the message array and directly hands the message over to the
 761  * receiver.
 762  * The receiver accepts the message and returns without grabbing the queue
 763  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 764  * are necessary. The same algorithm is used for sysv semaphores, see
 765  * ipc/sem.c for more details.
 766  *
 767  * The same algorithm is used for senders.
 768  */
 769
 770 /* pipelined_send() - send a message directly to the task waiting in
 771  * sys_mq_timedreceive() (without inserting message into a queue).
 772  */
 773 static inline void pipelined_send(struct mqueue_inode_info *info,
 774                                   struct msg_msg *message,
 775                                   struct ext_wait_queue *receiver)
 776 {
 777         receiver->msg = message;
 778         list_del(&receiver->list);
 779         receiver->state = STATE_PENDING;
 780         wake_up_process(receiver->task);
 781         smp_wmb();
 782         receiver->state = STATE_READY;
 783 }
 784
 785 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 786  * gets its message and put to the queue (we have one free place for sure). */
 787 static inline void pipelined_receive(struct mqueue_inode_info *info)
 788 {
 789         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 790
 791         if (!sender) {
 792                 /* for poll */
 793                 wake_up_interruptible(&info->wait_q);
 794                 return;
 795         }
 796         msg_insert(sender->msg, info);
 797         list_del(&sender->list);
 798         sender->state = STATE_PENDING;
 799         wake_up_process(sender->task);
 800         smp_wmb();
 801         sender->state = STATE_READY;
 802 }
 803
 804 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 805         size_t msg_len, unsigned int msg_prio,
 806         const struct timespec __user *u_abs_timeout)
 807 {
 808         struct file *filp;
 809         struct inode *inode;
 810         struct ext_wait_queue wait;
 811         struct ext_wait_queue *receiver;
 812         struct msg_msg *msg_ptr;
 813         struct mqueue_inode_info *info;
 814         long timeout;
 815         int ret;
 816
 817         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 818                 return -EINVAL;
 819
 820         timeout = prepare_timeout(u_abs_timeout);
 821
 822         ret = -EBADF;
 823         filp = fget(mqdes);
 824         if (unlikely(!filp))
 825                 goto out;
 826
 827         inode = filp->f_dentry->d_inode;
 828         if (unlikely(filp->f_op != &mqueue_file_operations))
 829                 goto out_fput;
 830         info = MQUEUE_I(inode);
 831
 832         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 833                 goto out_fput;
 834
 835         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 836                 ret = -EMSGSIZE;
 837                 goto out_fput;
 838         }
 839
 840         /* First try to allocate memory, before doing anything with
 841          * existing queues. */
 842         msg_ptr = load_msg(u_msg_ptr, msg_len);
 843         if (IS_ERR(msg_ptr)) {
 844                 ret = PTR_ERR(msg_ptr);
 845                 goto out_fput;
 846         }
 847         msg_ptr->m_ts = msg_len;
 848         msg_ptr->m_type = msg_prio;
 849
 850         spin_lock(&info->lock);
 851
 852         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 853                 if (filp->f_flags & O_NONBLOCK) {
 854                         spin_unlock(&info->lock);
 855                         ret = -EAGAIN;
 856                 } else if (unlikely(timeout < 0)) {
 857                         spin_unlock(&info->lock);
 858                         ret = timeout;
 859                 } else {
 860                         wait.task = current;
 861                         wait.msg = (void *) msg_ptr;
 862                         wait.state = STATE_NONE;
 863                         ret = wq_sleep(info, SEND, timeout, &wait);
 864                 }
 865                 if (ret < 0)
 866                         free_msg(msg_ptr);
 867         } else {
 868                 receiver = wq_get_first_waiter(info, RECV);
 869                 if (receiver) {
 870                         pipelined_send(info, msg_ptr, receiver);
 871                 } else {
 872                         /* adds message to the queue */
 873                         msg_insert(msg_ptr, info);
 874                         __do_notify(info);
 875                 }
 876                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 877                                 CURRENT_TIME;
 878                 spin_unlock(&info->lock);
 879                 ret = 0;
 880         }
 881 out_fput:
 882         fput(filp);
 883 out:
 884         return ret;
 885 }
 886
 887 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 888         size_t msg_len, unsigned int __user *u_msg_prio,
 889         const struct timespec __user *u_abs_timeout)
 890 {
 891         long timeout;
 892         ssize_t ret;
 893         struct msg_msg *msg_ptr;
 894         struct file *filp;
 895         struct inode *inode;
 896         struct mqueue_inode_info *info;
 897         struct ext_wait_queue wait;
 898
 899         timeout = prepare_timeout(u_abs_timeout);
 900
 901         ret = -EBADF;
 902         filp = fget(mqdes);
 903         if (unlikely(!filp))
 904                 goto out;
 905
 906         inode = filp->f_dentry->d_inode;
 907         if (unlikely(filp->f_op != &mqueue_file_operations))
 908                 goto out_fput;
 909         info = MQUEUE_I(inode);
 910
 911         if (unlikely(!(filp->f_mode & FMODE_READ)))
 912                 goto out_fput;
 913
 914         /* checks if buffer is big enough */
 915         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 916                 ret = -EMSGSIZE;
 917                 goto out_fput;
 918         }
 919
 920         spin_lock(&info->lock);
 921         if (info->attr.mq_curmsgs == 0) {
 922                 if (filp->f_flags & O_NONBLOCK) {
 923                         spin_unlock(&info->lock);
 924                         ret = -EAGAIN;
 925                         msg_ptr = NULL;
 926                 } else if (unlikely(timeout < 0)) {
 927                         spin_unlock(&info->lock);
 928                         ret = timeout;
 929                         msg_ptr = NULL;
 930                 } else {
 931                         wait.task = current;
 932                         wait.state = STATE_NONE;
 933                         ret = wq_sleep(info, RECV, timeout, &wait);
 934                         msg_ptr = wait.msg;
 935                 }
 936         } else {
 937                 msg_ptr = msg_get(info);
 938
 939                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 940                                 CURRENT_TIME;
 941
 942                 /* There is now free space in queue. */
 943                 pipelined_receive(info);
 944                 spin_unlock(&info->lock);
 945                 ret = 0;
 946         }
 947         if (ret == 0) {
 948                 ret = msg_ptr->m_ts;
 949
 950                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 951                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 952                         ret = -EFAULT;
 953                 }
 954                 free_msg(msg_ptr);
 955         }
 956 out_fput:
 957         fput(filp);
 958 out:
 959         return ret;
 960 }
 961
 962 /*
 963  * Notes: the case when user wants us to deregister (with NULL as pointer)
 964  * and he isn't currently owner of notification, will be silently discarded.
 965  * It isn't explicitly defined in the POSIX.
 966  */
 967 asmlinkage long sys_mq_notify(mqd_t mqdes,
 968                                 const struct sigevent __user *u_notification)
 969 {
 970         int ret;
 971         struct file *filp;
 972         struct sock *sock;
 973         struct inode *inode;
 974         struct sigevent notification;
 975         struct mqueue_inode_info *info;
 976         struct sk_buff *nc;
 977
 978         nc = NULL;
 979         sock = NULL;
 980         if (u_notification != NULL) {
 981                 if (copy_from_user(&notification, u_notification,
 982                                         sizeof(struct sigevent)))
 983                         return -EFAULT;
 984
 985                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
 986                              notification.sigev_notify != SIGEV_SIGNAL &&
 987                              notification.sigev_notify != SIGEV_THREAD))
 988                         return -EINVAL;
 989                 if (notification.sigev_notify == SIGEV_SIGNAL &&
 990                         !valid_signal(notification.sigev_signo)) {
 991                         return -EINVAL;
 992                 }
 993                 if (notification.sigev_notify == SIGEV_THREAD) {
 994                         /* create the notify skb */
 995                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
 996                         ret = -ENOMEM;
 997                         if (!nc)
 998                                 goto out;
 999                         ret = -EFAULT;
1000                         if (copy_from_user(nc->data,
1001                                         notification.sigev_value.sival_ptr,
1002                                         NOTIFY_COOKIE_LEN)) {
1003                                 goto out;
1004                         }
1005
1006                         /* TODO: add a header? */
1007                         skb_put(nc, NOTIFY_COOKIE_LEN);
1008                         /* and attach it to the socket */
1009 retry:
1010                         filp = fget(notification.sigev_signo);
1011                         ret = -EBADF;
1012                         if (!filp)
1013                                 goto out;
1014                         sock = netlink_getsockbyfilp(filp);
1015                         fput(filp);
1016                         if (IS_ERR(sock)) {
1017                                 ret = PTR_ERR(sock);
1018                                 sock = NULL;
1019                                 goto out;
1020                         }
1021
1022                         ret = netlink_attachskb(sock, nc, 0,
1023                                         MAX_SCHEDULE_TIMEOUT, NULL);
1024                         if (ret == 1)
1025                                 goto retry;
1026                         if (ret) {
1027                                 sock = NULL;
1028                                 nc = NULL;
1029                                 goto out;
1030                         }
1031                 }
1032         }
1033
1034         ret = -EBADF;
1035         filp = fget(mqdes);
1036         if (!filp)
1037                 goto out;
1038
1039         inode = filp->f_dentry->d_inode;
1040         if (unlikely(filp->f_op != &mqueue_file_operations))
1041                 goto out_fput;
1042         info = MQUEUE_I(inode);
1043
1044         ret = 0;
1045         spin_lock(&info->lock);
1046         if (u_notification == NULL) {
1047                 if (info->notify_owner == current->tgid) {
1048                         remove_notification(info);
1049                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1050                 }
1051         } else if (info->notify_owner != 0) {
1052                 ret = -EBUSY;
1053         } else {
1054                 switch (notification.sigev_notify) {
1055                 case SIGEV_NONE:
1056                         info->notify.sigev_notify = SIGEV_NONE;
1057                         break;
1058                 case SIGEV_THREAD:
1059                         info->notify_sock = sock;
1060                         info->notify_cookie = nc;
1061                         sock = NULL;
1062                         nc = NULL;
1063                         info->notify.sigev_notify = SIGEV_THREAD;
1064                         break;
1065                 case SIGEV_SIGNAL:
1066                         info->notify.sigev_signo = notification.sigev_signo;
1067                         info->notify.sigev_value = notification.sigev_value;
1068                         info->notify.sigev_notify = SIGEV_SIGNAL;
1069                         break;
1070                 }
1071                 info->notify_owner = current->tgid;
1072                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1073         }
1074         spin_unlock(&info->lock);
1075 out_fput:
1076         fput(filp);
1077 out:
1078         if (sock) {
1079                 netlink_detachskb(sock, nc);
1080         } else if (nc) {
1081                 dev_kfree_skb(nc);
1082         }
1083         return ret;
1084 }
1085
1086 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1087                         const struct mq_attr __user *u_mqstat,
1088                         struct mq_attr __user *u_omqstat)
1089 {
1090         int ret;
1091         struct mq_attr mqstat, omqstat;
1092         struct file *filp;
1093         struct inode *inode;
1094         struct mqueue_inode_info *info;
1095
1096         if (u_mqstat != NULL) {
1097                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1098                         return -EFAULT;
1099                 if (mqstat.mq_flags & (~O_NONBLOCK))
1100                         return -EINVAL;
1101         }
1102
1103         ret = -EBADF;
1104         filp = fget(mqdes);
1105         if (!filp)
1106                 goto out;
1107
1108         inode = filp->f_dentry->d_inode;
1109         if (unlikely(filp->f_op != &mqueue_file_operations))
1110                 goto out_fput;
1111         info = MQUEUE_I(inode);
1112
1113         spin_lock(&info->lock);
1114
1115         omqstat = info->attr;
1116         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1117         if (u_mqstat) {
1118                 if (mqstat.mq_flags & O_NONBLOCK)
1119                         filp->f_flags |= O_NONBLOCK;
1120                 else
1121                         filp->f_flags &= ~O_NONBLOCK;
1122
1123                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1124         }
1125
1126         spin_unlock(&info->lock);
1127
1128         ret = 0;
1129         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1130                                                 sizeof(struct mq_attr)))
1131                 ret = -EFAULT;
1132
1133 out_fput:
1134         fput(filp);
1135 out:
1136         return ret;
1137 }
1138
1139 static struct inode_operations mqueue_dir_inode_operations = {
1140         .lookup = simple_lookup,
1141         .create = mqueue_create,
1142         .unlink = mqueue_unlink,
1143 };
1144
1145 static struct file_operations mqueue_file_operations = {
1146         .flush = mqueue_flush_file,
1147         .poll = mqueue_poll_file,
1148         .read = mqueue_read_file,
1149 };
1150
1151 static struct super_operations mqueue_super_ops = {
1152         .alloc_inode = mqueue_alloc_inode,
1153         .destroy_inode = mqueue_destroy_inode,
1154         .statfs = simple_statfs,
1155         .delete_inode = mqueue_delete_inode,
1156         .drop_inode = generic_delete_inode,
1157 };
1158
1159 static struct file_system_type mqueue_fs_type = {
1160         .name = "mqueue",
1161         .get_sb = mqueue_get_sb,
1162         .kill_sb = kill_litter_super,
1163 };
1164
1165 static int msg_max_limit_min = DFLT_MSGMAX;
1166 static int msg_max_limit_max = HARD_MSGMAX;
1167
1168 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1169 static int msg_maxsize_limit_max = INT_MAX;
1170
1171 static ctl_table mq_sysctls[] = {
1172         {
1173                 .ctl_name       = CTL_QUEUESMAX,
1174                 .procname       = "queues_max",
1175                 .data           = &queues_max,
1176                 .maxlen         = sizeof(int),
1177                 .mode           = 0644,
1178                 .proc_handler   = &proc_dointvec,
1179         },
1180         {
1181                 .ctl_name       = CTL_MSGMAX,
1182                 .procname       = "msg_max",
1183                 .data           = &msg_max,
1184                 .maxlen         = sizeof(int),
1185                 .mode           = 0644,
1186                 .proc_handler   = &proc_dointvec_minmax,
1187                 .extra1         = &msg_max_limit_min,
1188                 .extra2         = &msg_max_limit_max,
1189         },
1190         {
1191                 .ctl_name       = CTL_MSGSIZEMAX,
1192                 .procname       = "msgsize_max",
1193                 .data           = &msgsize_max,
1194                 .maxlen         = sizeof(int),
1195                 .mode           = 0644,
1196                 .proc_handler   = &proc_dointvec_minmax,
1197                 .extra1         = &msg_maxsize_limit_min,
1198                 .extra2         = &msg_maxsize_limit_max,
1199         },
1200         { .ctl_name = 0 }
1201 };
1202
1203 static ctl_table mq_sysctl_dir[] = {
1204         {
1205                 .ctl_name       = FS_MQUEUE,
1206                 .procname       = "mqueue",
1207                 .mode           = 0555,
1208                 .child          = mq_sysctls,
1209         },
1210         { .ctl_name = 0 }
1211 };
1212
1213 static ctl_table mq_sysctl_root[] = {
1214         {
1215                 .ctl_name       = CTL_FS,
1216                 .procname       = "fs",
1217                 .mode           = 0555,
1218                 .child          = mq_sysctl_dir,
1219         },
1220         { .ctl_name = 0 }
1221 };
1222
1223 static int __init init_mqueue_fs(void)
1224 {
1225         int error;
1226
1227         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1228                                 sizeof(struct mqueue_inode_info), 0,
1229                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1230         if (mqueue_inode_cachep == NULL)
1231                 return -ENOMEM;
1232
1233         /* ignore failues - they are not fatal */
1234         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1235
1236         error = register_filesystem(&mqueue_fs_type);
1237         if (error)
1238                 goto out_sysctl;
1239
1240         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1241                 error = PTR_ERR(mqueue_mnt);
1242                 goto out_filesystem;
1243         }
1244
1245         /* internal initialization - not common for vfs */
1246         queues_count = 0;
1247         spin_lock_init(&mq_lock);
1248
1249         return 0;
1250
1251 out_filesystem:
1252         unregister_filesystem(&mqueue_fs_type);
1253 out_sysctl:
1254         if (mq_sysctl_table)
1255                 unregister_sysctl_table(mq_sysctl_table);
1256         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1257                 printk(KERN_INFO
1258                         "mqueue_inode_cache: not all structures were freed\n");
1259         }
1260         return error;
1261 }
1262
1263 __initcall(init_mqueue_fs);