git.oblomov.eu Git - linux-2.6/blob - 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@gmail.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  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
  12  *
  13  * This file is released under the GPL.
  14  */
  15
  16 #include <linux/capability.h>
  17 #include <linux/init.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/file.h>
  20 #include <linux/mount.h>
  21 #include <linux/namei.h>
  22 #include <linux/sysctl.h>
  23 #include <linux/poll.h>
  24 #include <linux/mqueue.h>
  25 #include <linux/msg.h>
  26 #include <linux/skbuff.h>
  27 #include <linux/netlink.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/audit.h>
  30 #include <linux/signal.h>
  31 #include <linux/mutex.h>
  32
  33 #include <net/sock.h>
  34 #include "util.h"
  35
  36 #define MQUEUE_MAGIC    0x19800202
  37 #define DIRENT_SIZE     20
  38 #define FILENT_SIZE     80
  39
  40 #define SEND            0
  41 #define RECV            1
  42
  43 #define STATE_NONE      0
  44 #define STATE_PENDING   1
  45 #define STATE_READY     2
  46
  47 /* default values */
  48 #define DFLT_QUEUESMAX  256     /* max number of message queues */
  49 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
  50 #define HARD_MSGMAX     (131072/sizeof(void*))
  51 #define DFLT_MSGSIZEMAX 8192    /* max message size */
  52
  53
  54 struct ext_wait_queue {         /* queue of sleeping tasks */
  55         struct task_struct *task;
  56         struct list_head list;
  57         struct msg_msg *msg;    /* ptr of loaded message */
  58         int state;              /* one of STATE_* values */
  59 };
  60
  61 struct mqueue_inode_info {
  62         spinlock_t lock;
  63         struct inode vfs_inode;
  64         wait_queue_head_t wait_q;
  65
  66         struct msg_msg **messages;
  67         struct mq_attr attr;
  68
  69         struct sigevent notify;
  70         struct pid* notify_owner;
  71         struct user_struct *user;       /* user who created, for accounting */
  72         struct sock *notify_sock;
  73         struct sk_buff *notify_cookie;
  74
  75         /* for tasks waiting for free space and messages, respectively */
  76         struct ext_wait_queue e_wait_q[2];
  77
  78         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  79 };
  80
  81 static const struct inode_operations mqueue_dir_inode_operations;
  82 static const struct file_operations mqueue_file_operations;
  83 static struct super_operations mqueue_super_ops;
  84 static void remove_notification(struct mqueue_inode_info *info);
  85
  86 static spinlock_t mq_lock;
  87 static struct kmem_cache *mqueue_inode_cachep;
  88 static struct vfsmount *mqueue_mnt;
  89
  90 static unsigned int queues_count;
  91 static unsigned int queues_max  = DFLT_QUEUESMAX;
  92 static unsigned int msg_max     = DFLT_MSGMAX;
  93 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
  94
  95 static struct ctl_table_header * mq_sysctl_table;
  96
  97 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  98 {
  99         return container_of(inode, struct mqueue_inode_info, vfs_inode);
 100 }
 101
 102 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 103                                                         struct mq_attr *attr)
 104 {
 105         struct inode *inode;
 106
 107         inode = new_inode(sb);
 108         if (inode) {
 109                 inode->i_mode = mode;
 110                 inode->i_uid = current->fsuid;
 111                 inode->i_gid = current->fsgid;
 112                 inode->i_blocks = 0;
 113                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 114                                 CURRENT_TIME;
 115
 116                 if (S_ISREG(mode)) {
 117                         struct mqueue_inode_info *info;
 118                         struct task_struct *p = current;
 119                         struct user_struct *u = p->user;
 120                         unsigned long mq_bytes, mq_msg_tblsz;
 121
 122                         inode->i_fop = &mqueue_file_operations;
 123                         inode->i_size = FILENT_SIZE;
 124                         /* mqueue specific info */
 125                         info = MQUEUE_I(inode);
 126                         spin_lock_init(&info->lock);
 127                         init_waitqueue_head(&info->wait_q);
 128                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 129                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 130                         info->messages = NULL;
 131                         info->notify_owner = NULL;
 132                         info->qsize = 0;
 133                         info->user = NULL;      /* set when all is ok */
 134                         memset(&info->attr, 0, sizeof(info->attr));
 135                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 136                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 137                         if (attr) {
 138                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 139                                 info->attr.mq_msgsize = attr->mq_msgsize;
 140                         }
 141                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 142                         mq_bytes = (mq_msg_tblsz +
 143                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 144
 145                         spin_lock(&mq_lock);
 146                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 147                             u->mq_bytes + mq_bytes >
 148                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
 149                                 spin_unlock(&mq_lock);
 150                                 goto out_inode;
 151                         }
 152                         u->mq_bytes += mq_bytes;
 153                         spin_unlock(&mq_lock);
 154
 155                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 156                         if (!info->messages) {
 157                                 spin_lock(&mq_lock);
 158                                 u->mq_bytes -= mq_bytes;
 159                                 spin_unlock(&mq_lock);
 160                                 goto out_inode;
 161                         }
 162                         /* all is ok */
 163                         info->user = get_uid(u);
 164                 } else if (S_ISDIR(mode)) {
 165                         inc_nlink(inode);
 166                         /* Some things misbehave if size == 0 on a directory */
 167                         inode->i_size = 2 * DIRENT_SIZE;
 168                         inode->i_op = &mqueue_dir_inode_operations;
 169                         inode->i_fop = &simple_dir_operations;
 170                 }
 171         }
 172         return inode;
 173 out_inode:
 174         make_bad_inode(inode);
 175         iput(inode);
 176         return NULL;
 177 }
 178
 179 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 180 {
 181         struct inode *inode;
 182
 183         sb->s_blocksize = PAGE_CACHE_SIZE;
 184         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 185         sb->s_magic = MQUEUE_MAGIC;
 186         sb->s_op = &mqueue_super_ops;
 187
 188         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 189         if (!inode)
 190                 return -ENOMEM;
 191
 192         sb->s_root = d_alloc_root(inode);
 193         if (!sb->s_root) {
 194                 iput(inode);
 195                 return -ENOMEM;
 196         }
 197
 198         return 0;
 199 }
 200
 201 static int mqueue_get_sb(struct file_system_type *fs_type,
 202                          int flags, const char *dev_name,
 203                          void *data, struct vfsmount *mnt)
 204 {
 205         return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
 206 }
 207
 208 static void init_once(struct kmem_cache *cachep, void *foo)
 209 {
 210         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 211
 212         inode_init_once(&p->vfs_inode);
 213 }
 214
 215 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 216 {
 217         struct mqueue_inode_info *ei;
 218
 219         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 220         if (!ei)
 221                 return NULL;
 222         return &ei->vfs_inode;
 223 }
 224
 225 static void mqueue_destroy_inode(struct inode *inode)
 226 {
 227         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 228 }
 229
 230 static void mqueue_delete_inode(struct inode *inode)
 231 {
 232         struct mqueue_inode_info *info;
 233         struct user_struct *user;
 234         unsigned long mq_bytes;
 235         int i;
 236
 237         if (S_ISDIR(inode->i_mode)) {
 238                 clear_inode(inode);
 239                 return;
 240         }
 241         info = MQUEUE_I(inode);
 242         spin_lock(&info->lock);
 243         for (i = 0; i < info->attr.mq_curmsgs; i++)
 244                 free_msg(info->messages[i]);
 245         kfree(info->messages);
 246         spin_unlock(&info->lock);
 247
 248         clear_inode(inode);
 249
 250         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 251                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 252         user = info->user;
 253         if (user) {
 254                 spin_lock(&mq_lock);
 255                 user->mq_bytes -= mq_bytes;
 256                 queues_count--;
 257                 spin_unlock(&mq_lock);
 258                 free_uid(user);
 259         }
 260 }
 261
 262 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 263                                 int mode, struct nameidata *nd)
 264 {
 265         struct inode *inode;
 266         struct mq_attr *attr = dentry->d_fsdata;
 267         int error;
 268
 269         spin_lock(&mq_lock);
 270         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 271                 error = -ENOSPC;
 272                 goto out_lock;
 273         }
 274         queues_count++;
 275         spin_unlock(&mq_lock);
 276
 277         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 278         if (!inode) {
 279                 error = -ENOMEM;
 280                 spin_lock(&mq_lock);
 281                 queues_count--;
 282                 goto out_lock;
 283         }
 284
 285         dir->i_size += DIRENT_SIZE;
 286         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 287
 288         d_instantiate(dentry, inode);
 289         dget(dentry);
 290         return 0;
 291 out_lock:
 292         spin_unlock(&mq_lock);
 293         return error;
 294 }
 295
 296 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 297 {
 298         struct inode *inode = dentry->d_inode;
 299
 300         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 301         dir->i_size -= DIRENT_SIZE;
 302         drop_nlink(inode);
 303         dput(dentry);
 304         return 0;
 305 }
 306
 307 /*
 308 *       This is routine for system read from queue file.
 309 *       To avoid mess with doing here some sort of mq_receive we allow
 310 *       to read only queue size & notification info (the only values
 311 *       that are interesting from user point of view and aren't accessible
 312 *       through std routines)
 313 */
 314 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 315                                 size_t count, loff_t * off)
 316 {
 317         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 318         char buffer[FILENT_SIZE];
 319         size_t slen;
 320         loff_t o;
 321
 322         if (!count)
 323                 return 0;
 324
 325         spin_lock(&info->lock);
 326         snprintf(buffer, sizeof(buffer),
 327                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 328                         info->qsize,
 329                         info->notify_owner ? info->notify.sigev_notify : 0,
 330                         (info->notify_owner &&
 331                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 332                                 info->notify.sigev_signo : 0,
 333                         pid_nr(info->notify_owner));
 334         spin_unlock(&info->lock);
 335         buffer[sizeof(buffer)-1] = '\0';
 336         slen = strlen(buffer)+1;
 337
 338         o = *off;
 339         if (o > slen)
 340                 return 0;
 341
 342         if (o + count > slen)
 343                 count = slen - o;
 344
 345         if (copy_to_user(u_data, buffer + o, count))
 346                 return -EFAULT;
 347
 348         *off = o + count;
 349         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 350         return count;
 351 }
 352
 353 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 354 {
 355         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 356
 357         spin_lock(&info->lock);
 358         if (task_tgid(current) == info->notify_owner)
 359                 remove_notification(info);
 360
 361         spin_unlock(&info->lock);
 362         return 0;
 363 }
 364
 365 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 366 {
 367         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 368         int retval = 0;
 369
 370         poll_wait(filp, &info->wait_q, poll_tab);
 371
 372         spin_lock(&info->lock);
 373         if (info->attr.mq_curmsgs)
 374                 retval = POLLIN | POLLRDNORM;
 375
 376         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 377                 retval |= POLLOUT | POLLWRNORM;
 378         spin_unlock(&info->lock);
 379
 380         return retval;
 381 }
 382
 383 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 384 static void wq_add(struct mqueue_inode_info *info, int sr,
 385                         struct ext_wait_queue *ewp)
 386 {
 387         struct ext_wait_queue *walk;
 388
 389         ewp->task = current;
 390
 391         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 392                 if (walk->task->static_prio <= current->static_prio) {
 393                         list_add_tail(&ewp->list, &walk->list);
 394                         return;
 395                 }
 396         }
 397         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 398 }
 399
 400 /*
 401  * Puts current task to sleep. Caller must hold queue lock. After return
 402  * lock isn't held.
 403  * sr: SEND or RECV
 404  */
 405 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 406                         long timeout, struct ext_wait_queue *ewp)
 407 {
 408         int retval;
 409         signed long time;
 410
 411         wq_add(info, sr, ewp);
 412
 413         for (;;) {
 414                 set_current_state(TASK_INTERRUPTIBLE);
 415
 416                 spin_unlock(&info->lock);
 417                 time = schedule_timeout(timeout);
 418
 419                 while (ewp->state == STATE_PENDING)
 420                         cpu_relax();
 421
 422                 if (ewp->state == STATE_READY) {
 423                         retval = 0;
 424                         goto out;
 425                 }
 426                 spin_lock(&info->lock);
 427                 if (ewp->state == STATE_READY) {
 428                         retval = 0;
 429                         goto out_unlock;
 430                 }
 431                 if (signal_pending(current)) {
 432                         retval = -ERESTARTSYS;
 433                         break;
 434                 }
 435                 if (time == 0) {
 436                         retval = -ETIMEDOUT;
 437                         break;
 438                 }
 439         }
 440         list_del(&ewp->list);
 441 out_unlock:
 442         spin_unlock(&info->lock);
 443 out:
 444         return retval;
 445 }
 446
 447 /*
 448  * Returns waiting task that should be serviced first or NULL if none exists
 449  */
 450 static struct ext_wait_queue *wq_get_first_waiter(
 451                 struct mqueue_inode_info *info, int sr)
 452 {
 453         struct list_head *ptr;
 454
 455         ptr = info->e_wait_q[sr].list.prev;
 456         if (ptr == &info->e_wait_q[sr].list)
 457                 return NULL;
 458         return list_entry(ptr, struct ext_wait_queue, list);
 459 }
 460
 461 /* Auxiliary functions to manipulate messages' list */
 462 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 463 {
 464         int k;
 465
 466         k = info->attr.mq_curmsgs - 1;
 467         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 468                 info->messages[k + 1] = info->messages[k];
 469                 k--;
 470         }
 471         info->attr.mq_curmsgs++;
 472         info->qsize += ptr->m_ts;
 473         info->messages[k + 1] = ptr;
 474 }
 475
 476 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 477 {
 478         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 479         return info->messages[info->attr.mq_curmsgs];
 480 }
 481
 482 static inline void set_cookie(struct sk_buff *skb, char code)
 483 {
 484         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 485 }
 486
 487 /*
 488  * The next function is only to split too long sys_mq_timedsend
 489  */
 490 static void __do_notify(struct mqueue_inode_info *info)
 491 {
 492         /* notification
 493          * invoked when there is registered process and there isn't process
 494          * waiting synchronously for message AND state of queue changed from
 495          * empty to not empty. Here we are sure that no one is waiting
 496          * synchronously. */
 497         if (info->notify_owner &&
 498             info->attr.mq_curmsgs == 1) {
 499                 struct siginfo sig_i;
 500                 switch (info->notify.sigev_notify) {
 501                 case SIGEV_NONE:
 502                         break;
 503                 case SIGEV_SIGNAL:
 504                         /* sends signal */
 505
 506                         sig_i.si_signo = info->notify.sigev_signo;
 507                         sig_i.si_errno = 0;
 508                         sig_i.si_code = SI_MESGQ;
 509                         sig_i.si_value = info->notify.sigev_value;
 510                         sig_i.si_pid = current->tgid;
 511                         sig_i.si_uid = current->uid;
 512
 513                         kill_pid_info(info->notify.sigev_signo,
 514                                       &sig_i, info->notify_owner);
 515                         break;
 516                 case SIGEV_THREAD:
 517                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 518                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 519                         break;
 520                 }
 521                 /* after notification unregisters process */
 522                 put_pid(info->notify_owner);
 523                 info->notify_owner = NULL;
 524         }
 525         wake_up(&info->wait_q);
 526 }
 527
 528 static long prepare_timeout(const struct timespec __user *u_arg)
 529 {
 530         struct timespec ts, nowts;
 531         long timeout;
 532
 533         if (u_arg) {
 534                 if (unlikely(copy_from_user(&ts, u_arg,
 535                                         sizeof(struct timespec))))
 536                         return -EFAULT;
 537
 538                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 539                         || ts.tv_nsec >= NSEC_PER_SEC))
 540                         return -EINVAL;
 541                 nowts = CURRENT_TIME;
 542                 /* first subtract as jiffies can't be too big */
 543                 ts.tv_sec -= nowts.tv_sec;
 544                 if (ts.tv_nsec < nowts.tv_nsec) {
 545                         ts.tv_nsec += NSEC_PER_SEC;
 546                         ts.tv_sec--;
 547                 }
 548                 ts.tv_nsec -= nowts.tv_nsec;
 549                 if (ts.tv_sec < 0)
 550                         return 0;
 551
 552                 timeout = timespec_to_jiffies(&ts) + 1;
 553         } else
 554                 return MAX_SCHEDULE_TIMEOUT;
 555
 556         return timeout;
 557 }
 558
 559 static void remove_notification(struct mqueue_inode_info *info)
 560 {
 561         if (info->notify_owner != NULL &&
 562             info->notify.sigev_notify == SIGEV_THREAD) {
 563                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 564                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 565         }
 566         put_pid(info->notify_owner);
 567         info->notify_owner = NULL;
 568 }
 569
 570 static int mq_attr_ok(struct mq_attr *attr)
 571 {
 572         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 573                 return 0;
 574         if (capable(CAP_SYS_RESOURCE)) {
 575                 if (attr->mq_maxmsg > HARD_MSGMAX)
 576                         return 0;
 577         } else {
 578                 if (attr->mq_maxmsg > msg_max ||
 579                                 attr->mq_msgsize > msgsize_max)
 580                         return 0;
 581         }
 582         /* check for overflow */
 583         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 584                 return 0;
 585         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 586             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 587             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 588                 return 0;
 589         return 1;
 590 }
 591
 592 /*
 593  * Invoked when creating a new queue via sys_mq_open
 594  */
 595 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 596                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 597 {
 598         struct mq_attr attr;
 599         int ret;
 600
 601         if (u_attr) {
 602                 ret = -EFAULT;
 603                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 604                         goto out;
 605                 ret = -EINVAL;
 606                 if (!mq_attr_ok(&attr))
 607                         goto out;
 608                 /* store for use during create */
 609                 dentry->d_fsdata = &attr;
 610         }
 611
 612         mode &= ~current->fs->umask;
 613         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 614         dentry->d_fsdata = NULL;
 615         if (ret)
 616                 goto out;
 617
 618         return dentry_open(dentry, mqueue_mnt, oflag);
 619
 620 out:
 621         dput(dentry);
 622         mntput(mqueue_mnt);
 623         return ERR_PTR(ret);
 624 }
 625
 626 /* Opens existing queue */
 627 static struct file *do_open(struct dentry *dentry, int oflag)
 628 {
 629 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 630                                         MAY_READ | MAY_WRITE };
 631
 632         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 633                 dput(dentry);
 634                 mntput(mqueue_mnt);
 635                 return ERR_PTR(-EINVAL);
 636         }
 637
 638         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
 639                 dput(dentry);
 640                 mntput(mqueue_mnt);
 641                 return ERR_PTR(-EACCES);
 642         }
 643
 644         return dentry_open(dentry, mqueue_mnt, oflag);
 645 }
 646
 647 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 648                                 struct mq_attr __user *u_attr)
 649 {
 650         struct dentry *dentry;
 651         struct file *filp;
 652         char *name;
 653         int fd, error;
 654
 655         error = audit_mq_open(oflag, mode, u_attr);
 656         if (error != 0)
 657                 return error;
 658
 659         if (IS_ERR(name = getname(u_name)))
 660                 return PTR_ERR(name);
 661
 662         fd = get_unused_fd();
 663         if (fd < 0)
 664                 goto out_putname;
 665
 666         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 667         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 668         if (IS_ERR(dentry)) {
 669                 error = PTR_ERR(dentry);
 670                 goto out_err;
 671         }
 672         mntget(mqueue_mnt);
 673
 674         if (oflag & O_CREAT) {
 675                 if (dentry->d_inode) {  /* entry already exists */
 676                         audit_inode(name, dentry->d_inode);
 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                 audit_inode(name, dentry->d_inode);
 690                 filp = do_open(dentry, oflag);
 691         }
 692
 693         if (IS_ERR(filp)) {
 694                 error = PTR_ERR(filp);
 695                 goto out_putfd;
 696         }
 697
 698         set_close_on_exec(fd, 1);
 699         fd_install(fd, filp);
 700         goto out_upsem;
 701
 702 out:
 703         dput(dentry);
 704         mntput(mqueue_mnt);
 705 out_putfd:
 706         put_unused_fd(fd);
 707 out_err:
 708         fd = error;
 709 out_upsem:
 710         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 711 out_putname:
 712         putname(name);
 713         return fd;
 714 }
 715
 716 asmlinkage long sys_mq_unlink(const char __user *u_name)
 717 {
 718         int err;
 719         char *name;
 720         struct dentry *dentry;
 721         struct inode *inode = NULL;
 722
 723         name = getname(u_name);
 724         if (IS_ERR(name))
 725                 return PTR_ERR(name);
 726
 727         mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
 728                         I_MUTEX_PARENT);
 729         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 730         if (IS_ERR(dentry)) {
 731                 err = PTR_ERR(dentry);
 732                 goto out_unlock;
 733         }
 734
 735         if (!dentry->d_inode) {
 736                 err = -ENOENT;
 737                 goto out_err;
 738         }
 739
 740         inode = dentry->d_inode;
 741         if (inode)
 742                 atomic_inc(&inode->i_count);
 743
 744         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 745 out_err:
 746         dput(dentry);
 747
 748 out_unlock:
 749         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 750         putname(name);
 751         if (inode)
 752                 iput(inode);
 753
 754         return err;
 755 }
 756
 757 /* Pipelined send and receive functions.
 758  *
 759  * If a receiver finds no waiting message, then it registers itself in the
 760  * list of waiting receivers. A sender checks that list before adding the new
 761  * message into the message array. If there is a waiting receiver, then it
 762  * bypasses the message array and directly hands the message over to the
 763  * receiver.
 764  * The receiver accepts the message and returns without grabbing the queue
 765  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 766  * are necessary. The same algorithm is used for sysv semaphores, see
 767  * ipc/sem.c for more details.
 768  *
 769  * The same algorithm is used for senders.
 770  */
 771
 772 /* pipelined_send() - send a message directly to the task waiting in
 773  * sys_mq_timedreceive() (without inserting message into a queue).
 774  */
 775 static inline void pipelined_send(struct mqueue_inode_info *info,
 776                                   struct msg_msg *message,
 777                                   struct ext_wait_queue *receiver)
 778 {
 779         receiver->msg = message;
 780         list_del(&receiver->list);
 781         receiver->state = STATE_PENDING;
 782         wake_up_process(receiver->task);
 783         smp_wmb();
 784         receiver->state = STATE_READY;
 785 }
 786
 787 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 788  * gets its message and put to the queue (we have one free place for sure). */
 789 static inline void pipelined_receive(struct mqueue_inode_info *info)
 790 {
 791         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 792
 793         if (!sender) {
 794                 /* for poll */
 795                 wake_up_interruptible(&info->wait_q);
 796                 return;
 797         }
 798         msg_insert(sender->msg, info);
 799         list_del(&sender->list);
 800         sender->state = STATE_PENDING;
 801         wake_up_process(sender->task);
 802         smp_wmb();
 803         sender->state = STATE_READY;
 804 }
 805
 806 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 807         size_t msg_len, unsigned int msg_prio,
 808         const struct timespec __user *u_abs_timeout)
 809 {
 810         struct file *filp;
 811         struct inode *inode;
 812         struct ext_wait_queue wait;
 813         struct ext_wait_queue *receiver;
 814         struct msg_msg *msg_ptr;
 815         struct mqueue_inode_info *info;
 816         long timeout;
 817         int ret;
 818
 819         ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
 820         if (ret != 0)
 821                 return ret;
 822
 823         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 824                 return -EINVAL;
 825
 826         timeout = prepare_timeout(u_abs_timeout);
 827
 828         ret = -EBADF;
 829         filp = fget(mqdes);
 830         if (unlikely(!filp))
 831                 goto out;
 832
 833         inode = filp->f_path.dentry->d_inode;
 834         if (unlikely(filp->f_op != &mqueue_file_operations))
 835                 goto out_fput;
 836         info = MQUEUE_I(inode);
 837         audit_inode(NULL, inode);
 838
 839         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 840                 goto out_fput;
 841
 842         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 843                 ret = -EMSGSIZE;
 844                 goto out_fput;
 845         }
 846
 847         /* First try to allocate memory, before doing anything with
 848          * existing queues. */
 849         msg_ptr = load_msg(u_msg_ptr, msg_len);
 850         if (IS_ERR(msg_ptr)) {
 851                 ret = PTR_ERR(msg_ptr);
 852                 goto out_fput;
 853         }
 854         msg_ptr->m_ts = msg_len;
 855         msg_ptr->m_type = msg_prio;
 856
 857         spin_lock(&info->lock);
 858
 859         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 860                 if (filp->f_flags & O_NONBLOCK) {
 861                         spin_unlock(&info->lock);
 862                         ret = -EAGAIN;
 863                 } else if (unlikely(timeout < 0)) {
 864                         spin_unlock(&info->lock);
 865                         ret = timeout;
 866                 } else {
 867                         wait.task = current;
 868                         wait.msg = (void *) msg_ptr;
 869                         wait.state = STATE_NONE;
 870                         ret = wq_sleep(info, SEND, timeout, &wait);
 871                 }
 872                 if (ret < 0)
 873                         free_msg(msg_ptr);
 874         } else {
 875                 receiver = wq_get_first_waiter(info, RECV);
 876                 if (receiver) {
 877                         pipelined_send(info, msg_ptr, receiver);
 878                 } else {
 879                         /* adds message to the queue */
 880                         msg_insert(msg_ptr, info);
 881                         __do_notify(info);
 882                 }
 883                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 884                                 CURRENT_TIME;
 885                 spin_unlock(&info->lock);
 886                 ret = 0;
 887         }
 888 out_fput:
 889         fput(filp);
 890 out:
 891         return ret;
 892 }
 893
 894 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 895         size_t msg_len, unsigned int __user *u_msg_prio,
 896         const struct timespec __user *u_abs_timeout)
 897 {
 898         long timeout;
 899         ssize_t ret;
 900         struct msg_msg *msg_ptr;
 901         struct file *filp;
 902         struct inode *inode;
 903         struct mqueue_inode_info *info;
 904         struct ext_wait_queue wait;
 905
 906         ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
 907         if (ret != 0)
 908                 return ret;
 909
 910         timeout = prepare_timeout(u_abs_timeout);
 911
 912         ret = -EBADF;
 913         filp = fget(mqdes);
 914         if (unlikely(!filp))
 915                 goto out;
 916
 917         inode = filp->f_path.dentry->d_inode;
 918         if (unlikely(filp->f_op != &mqueue_file_operations))
 919                 goto out_fput;
 920         info = MQUEUE_I(inode);
 921         audit_inode(NULL, inode);
 922
 923         if (unlikely(!(filp->f_mode & FMODE_READ)))
 924                 goto out_fput;
 925
 926         /* checks if buffer is big enough */
 927         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 928                 ret = -EMSGSIZE;
 929                 goto out_fput;
 930         }
 931
 932         spin_lock(&info->lock);
 933         if (info->attr.mq_curmsgs == 0) {
 934                 if (filp->f_flags & O_NONBLOCK) {
 935                         spin_unlock(&info->lock);
 936                         ret = -EAGAIN;
 937                         msg_ptr = NULL;
 938                 } else if (unlikely(timeout < 0)) {
 939                         spin_unlock(&info->lock);
 940                         ret = timeout;
 941                         msg_ptr = NULL;
 942                 } else {
 943                         wait.task = current;
 944                         wait.state = STATE_NONE;
 945                         ret = wq_sleep(info, RECV, timeout, &wait);
 946                         msg_ptr = wait.msg;
 947                 }
 948         } else {
 949                 msg_ptr = msg_get(info);
 950
 951                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 952                                 CURRENT_TIME;
 953
 954                 /* There is now free space in queue. */
 955                 pipelined_receive(info);
 956                 spin_unlock(&info->lock);
 957                 ret = 0;
 958         }
 959         if (ret == 0) {
 960                 ret = msg_ptr->m_ts;
 961
 962                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 963                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 964                         ret = -EFAULT;
 965                 }
 966                 free_msg(msg_ptr);
 967         }
 968 out_fput:
 969         fput(filp);
 970 out:
 971         return ret;
 972 }
 973
 974 /*
 975  * Notes: the case when user wants us to deregister (with NULL as pointer)
 976  * and he isn't currently owner of notification, will be silently discarded.
 977  * It isn't explicitly defined in the POSIX.
 978  */
 979 asmlinkage long sys_mq_notify(mqd_t mqdes,
 980                                 const struct sigevent __user *u_notification)
 981 {
 982         int ret;
 983         struct file *filp;
 984         struct sock *sock;
 985         struct inode *inode;
 986         struct sigevent notification;
 987         struct mqueue_inode_info *info;
 988         struct sk_buff *nc;
 989
 990         ret = audit_mq_notify(mqdes, u_notification);
 991         if (ret != 0)
 992                 return ret;
 993
 994         nc = NULL;
 995         sock = NULL;
 996         if (u_notification != NULL) {
 997                 if (copy_from_user(&notification, u_notification,
 998                                         sizeof(struct sigevent)))
 999                         return -EFAULT;
1000
1001                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1002                              notification.sigev_notify != SIGEV_SIGNAL &&
1003                              notification.sigev_notify != SIGEV_THREAD))
1004                         return -EINVAL;
1005                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1006                         !valid_signal(notification.sigev_signo)) {
1007                         return -EINVAL;
1008                 }
1009                 if (notification.sigev_notify == SIGEV_THREAD) {
1010                         /* create the notify skb */
1011                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1012                         ret = -ENOMEM;
1013                         if (!nc)
1014                                 goto out;
1015                         ret = -EFAULT;
1016                         if (copy_from_user(nc->data,
1017                                         notification.sigev_value.sival_ptr,
1018                                         NOTIFY_COOKIE_LEN)) {
1019                                 goto out;
1020                         }
1021
1022                         /* TODO: add a header? */
1023                         skb_put(nc, NOTIFY_COOKIE_LEN);
1024                         /* and attach it to the socket */
1025 retry:
1026                         filp = fget(notification.sigev_signo);
1027                         ret = -EBADF;
1028                         if (!filp)
1029                                 goto out;
1030                         sock = netlink_getsockbyfilp(filp);
1031                         fput(filp);
1032                         if (IS_ERR(sock)) {
1033                                 ret = PTR_ERR(sock);
1034                                 sock = NULL;
1035                                 goto out;
1036                         }
1037
1038                         ret = netlink_attachskb(sock, nc, 0,
1039                                         MAX_SCHEDULE_TIMEOUT, NULL);
1040                         if (ret == 1)
1041                                 goto retry;
1042                         if (ret) {
1043                                 sock = NULL;
1044                                 nc = NULL;
1045                                 goto out;
1046                         }
1047                 }
1048         }
1049
1050         ret = -EBADF;
1051         filp = fget(mqdes);
1052         if (!filp)
1053                 goto out;
1054
1055         inode = filp->f_path.dentry->d_inode;
1056         if (unlikely(filp->f_op != &mqueue_file_operations))
1057                 goto out_fput;
1058         info = MQUEUE_I(inode);
1059
1060         ret = 0;
1061         spin_lock(&info->lock);
1062         if (u_notification == NULL) {
1063                 if (info->notify_owner == task_tgid(current)) {
1064                         remove_notification(info);
1065                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1066                 }
1067         } else if (info->notify_owner != NULL) {
1068                 ret = -EBUSY;
1069         } else {
1070                 switch (notification.sigev_notify) {
1071                 case SIGEV_NONE:
1072                         info->notify.sigev_notify = SIGEV_NONE;
1073                         break;
1074                 case SIGEV_THREAD:
1075                         info->notify_sock = sock;
1076                         info->notify_cookie = nc;
1077                         sock = NULL;
1078                         nc = NULL;
1079                         info->notify.sigev_notify = SIGEV_THREAD;
1080                         break;
1081                 case SIGEV_SIGNAL:
1082                         info->notify.sigev_signo = notification.sigev_signo;
1083                         info->notify.sigev_value = notification.sigev_value;
1084                         info->notify.sigev_notify = SIGEV_SIGNAL;
1085                         break;
1086                 }
1087
1088                 info->notify_owner = get_pid(task_tgid(current));
1089                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1090         }
1091         spin_unlock(&info->lock);
1092 out_fput:
1093         fput(filp);
1094 out:
1095         if (sock) {
1096                 netlink_detachskb(sock, nc);
1097         } else if (nc) {
1098                 dev_kfree_skb(nc);
1099         }
1100         return ret;
1101 }
1102
1103 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1104                         const struct mq_attr __user *u_mqstat,
1105                         struct mq_attr __user *u_omqstat)
1106 {
1107         int ret;
1108         struct mq_attr mqstat, omqstat;
1109         struct file *filp;
1110         struct inode *inode;
1111         struct mqueue_inode_info *info;
1112
1113         if (u_mqstat != NULL) {
1114                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1115                         return -EFAULT;
1116                 if (mqstat.mq_flags & (~O_NONBLOCK))
1117                         return -EINVAL;
1118         }
1119
1120         ret = -EBADF;
1121         filp = fget(mqdes);
1122         if (!filp)
1123                 goto out;
1124
1125         inode = filp->f_path.dentry->d_inode;
1126         if (unlikely(filp->f_op != &mqueue_file_operations))
1127                 goto out_fput;
1128         info = MQUEUE_I(inode);
1129
1130         spin_lock(&info->lock);
1131
1132         omqstat = info->attr;
1133         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1134         if (u_mqstat) {
1135                 ret = audit_mq_getsetattr(mqdes, &mqstat);
1136                 if (ret != 0)
1137                         goto out;
1138                 if (mqstat.mq_flags & O_NONBLOCK)
1139                         filp->f_flags |= O_NONBLOCK;
1140                 else
1141                         filp->f_flags &= ~O_NONBLOCK;
1142
1143                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1144         }
1145
1146         spin_unlock(&info->lock);
1147
1148         ret = 0;
1149         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1150                                                 sizeof(struct mq_attr)))
1151                 ret = -EFAULT;
1152
1153 out_fput:
1154         fput(filp);
1155 out:
1156         return ret;
1157 }
1158
1159 static const struct inode_operations mqueue_dir_inode_operations = {
1160         .lookup = simple_lookup,
1161         .create = mqueue_create,
1162         .unlink = mqueue_unlink,
1163 };
1164
1165 static const struct file_operations mqueue_file_operations = {
1166         .flush = mqueue_flush_file,
1167         .poll = mqueue_poll_file,
1168         .read = mqueue_read_file,
1169 };
1170
1171 static struct super_operations mqueue_super_ops = {
1172         .alloc_inode = mqueue_alloc_inode,
1173         .destroy_inode = mqueue_destroy_inode,
1174         .statfs = simple_statfs,
1175         .delete_inode = mqueue_delete_inode,
1176         .drop_inode = generic_delete_inode,
1177 };
1178
1179 static struct file_system_type mqueue_fs_type = {
1180         .name = "mqueue",
1181         .get_sb = mqueue_get_sb,
1182         .kill_sb = kill_litter_super,
1183 };
1184
1185 static int msg_max_limit_min = DFLT_MSGMAX;
1186 static int msg_max_limit_max = HARD_MSGMAX;
1187
1188 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1189 static int msg_maxsize_limit_max = INT_MAX;
1190
1191 static ctl_table mq_sysctls[] = {
1192         {
1193                 .procname       = "queues_max",
1194                 .data           = &queues_max,
1195                 .maxlen         = sizeof(int),
1196                 .mode           = 0644,
1197                 .proc_handler   = &proc_dointvec,
1198         },
1199         {
1200                 .procname       = "msg_max",
1201                 .data           = &msg_max,
1202                 .maxlen         = sizeof(int),
1203                 .mode           = 0644,
1204                 .proc_handler   = &proc_dointvec_minmax,
1205                 .extra1         = &msg_max_limit_min,
1206                 .extra2         = &msg_max_limit_max,
1207         },
1208         {
1209                 .procname       = "msgsize_max",
1210                 .data           = &msgsize_max,
1211                 .maxlen         = sizeof(int),
1212                 .mode           = 0644,
1213                 .proc_handler   = &proc_dointvec_minmax,
1214                 .extra1         = &msg_maxsize_limit_min,
1215                 .extra2         = &msg_maxsize_limit_max,
1216         },
1217         { .ctl_name = 0 }
1218 };
1219
1220 static ctl_table mq_sysctl_dir[] = {
1221         {
1222                 .procname       = "mqueue",
1223                 .mode           = 0555,
1224                 .child          = mq_sysctls,
1225         },
1226         { .ctl_name = 0 }
1227 };
1228
1229 static ctl_table mq_sysctl_root[] = {
1230         {
1231                 .ctl_name       = CTL_FS,
1232                 .procname       = "fs",
1233                 .mode           = 0555,
1234                 .child          = mq_sysctl_dir,
1235         },
1236         { .ctl_name = 0 }
1237 };
1238
1239 static int __init init_mqueue_fs(void)
1240 {
1241         int error;
1242
1243         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1244                                 sizeof(struct mqueue_inode_info), 0,
1245                                 SLAB_HWCACHE_ALIGN, init_once);
1246         if (mqueue_inode_cachep == NULL)
1247                 return -ENOMEM;
1248
1249         /* ignore failues - they are not fatal */
1250         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1251
1252         error = register_filesystem(&mqueue_fs_type);
1253         if (error)
1254                 goto out_sysctl;
1255
1256         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1257                 error = PTR_ERR(mqueue_mnt);
1258                 goto out_filesystem;
1259         }
1260
1261         /* internal initialization - not common for vfs */
1262         queues_count = 0;
1263         spin_lock_init(&mq_lock);
1264
1265         return 0;
1266
1267 out_filesystem:
1268         unregister_filesystem(&mqueue_fs_type);
1269 out_sysctl:
1270         if (mq_sysctl_table)
1271                 unregister_sysctl_table(mq_sysctl_table);
1272         kmem_cache_destroy(mqueue_inode_cachep);
1273         return error;
1274 }
1275
1276 __initcall(init_mqueue_fs);