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  * 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 #include <linux/vs_base.h>
  33 #include <linux/vs_context.h>
  34 #include <linux/vs_limit.h>
  35
  36 #include <net/sock.h>
  37 #include "util.h"
  38
  39 #define MQUEUE_MAGIC    0x19800202
  40 #define DIRENT_SIZE     20
  41 #define FILENT_SIZE     80
  42
  43 #define SEND            0
  44 #define RECV            1
  45
  46 #define STATE_NONE      0
  47 #define STATE_PENDING   1
  48 #define STATE_READY     2
  49
  50 /* used by sysctl */
  51 #define FS_MQUEUE       1
  52 #define CTL_QUEUESMAX   2
  53 #define CTL_MSGMAX      3
  54 #define CTL_MSGSIZEMAX  4
  55
  56 /* default values */
  57 #define DFLT_QUEUESMAX  256     /* max number of message queues */
  58 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
  59 #define HARD_MSGMAX     (131072/sizeof(void*))
  60 #define DFLT_MSGSIZEMAX 8192    /* max message size */
  61
  62
  63 struct ext_wait_queue {         /* queue of sleeping tasks */
  64         struct task_struct *task;
  65         struct list_head list;
  66         struct msg_msg *msg;    /* ptr of loaded message */
  67         int state;              /* one of STATE_* values */
  68 };
  69
  70 struct mqueue_inode_info {
  71         spinlock_t lock;
  72         struct inode vfs_inode;
  73         wait_queue_head_t wait_q;
  74
  75         struct msg_msg **messages;
  76         struct mq_attr attr;
  77
  78         struct sigevent notify;
  79         pid_t notify_owner;
  80         struct user_struct *user;       /* user who created, for accounting */
  81         struct sock *notify_sock;
  82         struct sk_buff *notify_cookie;
  83
  84         /* for tasks waiting for free space and messages, respectively */
  85         struct ext_wait_queue e_wait_q[2];
  86
  87         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  88 };
  89
  90 static struct inode_operations mqueue_dir_inode_operations;
  91 static struct file_operations mqueue_file_operations;
  92 static struct super_operations mqueue_super_ops;
  93 static void remove_notification(struct mqueue_inode_info *info);
  94
  95 static spinlock_t mq_lock;
  96 static kmem_cache_t *mqueue_inode_cachep;
  97 static struct vfsmount *mqueue_mnt;
  98
  99 static unsigned int queues_count;
 100 static unsigned int queues_max  = DFLT_QUEUESMAX;
 101 static unsigned int msg_max     = DFLT_MSGMAX;
 102 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
 103
 104 static struct ctl_table_header * mq_sysctl_table;
 105
 106 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
 107 {
 108         return container_of(inode, struct mqueue_inode_info, vfs_inode);
 109 }
 110
 111 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 112                                                         struct mq_attr *attr)
 113 {
 114         struct inode *inode;
 115
 116         inode = new_inode(sb);
 117         if (inode) {
 118                 inode->i_mode = mode;
 119                 inode->i_uid = current->fsuid;
 120                 inode->i_gid = current->fsgid;
 121                 inode->i_blocks = 0;
 122                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 123                                 CURRENT_TIME;
 124
 125                 if (S_ISREG(mode)) {
 126                         struct mqueue_inode_info *info;
 127                         struct task_struct *p = current;
 128                         struct user_struct *u = p->user;
 129                         unsigned long mq_bytes, mq_msg_tblsz;
 130
 131                         inode->i_fop = &mqueue_file_operations;
 132                         inode->i_size = FILENT_SIZE;
 133                         /* mqueue specific info */
 134                         info = MQUEUE_I(inode);
 135                         spin_lock_init(&info->lock);
 136                         init_waitqueue_head(&info->wait_q);
 137                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 138                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 139                         info->messages = NULL;
 140                         info->notify_owner = 0;
 141                         info->qsize = 0;
 142                         info->user = NULL;      /* set when all is ok */
 143                         memset(&info->attr, 0, sizeof(info->attr));
 144                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 145                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 146                         if (attr) {
 147                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 148                                 info->attr.mq_msgsize = attr->mq_msgsize;
 149                         }
 150                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 151                         mq_bytes = (mq_msg_tblsz +
 152                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 153
 154                         spin_lock(&mq_lock);
 155                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 156                             u->mq_bytes + mq_bytes >
 157                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur ||
 158                             !vx_ipcmsg_avail(p->vx_info, mq_bytes)) {
 159                                 spin_unlock(&mq_lock);
 160                                 goto out_inode;
 161                         }
 162                         u->mq_bytes += mq_bytes;
 163                         vx_ipcmsg_add(p->vx_info, u, mq_bytes);
 164                         spin_unlock(&mq_lock);
 165
 166                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 167                         if (!info->messages) {
 168                                 spin_lock(&mq_lock);
 169                                 u->mq_bytes -= mq_bytes;
 170                                 vx_ipcmsg_sub(p->vx_info, u, mq_bytes);
 171                                 spin_unlock(&mq_lock);
 172                                 goto out_inode;
 173                         }
 174                         /* all is ok */
 175                         info->user = get_uid(u);
 176                 } else if (S_ISDIR(mode)) {
 177                         inode->i_nlink++;
 178                         /* Some things misbehave if size == 0 on a directory */
 179                         inode->i_size = 2 * DIRENT_SIZE;
 180                         inode->i_op = &mqueue_dir_inode_operations;
 181                         inode->i_fop = &simple_dir_operations;
 182                 }
 183         }
 184         return inode;
 185 out_inode:
 186         make_bad_inode(inode);
 187         iput(inode);
 188         return NULL;
 189 }
 190
 191 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 192 {
 193         struct inode *inode;
 194
 195         sb->s_blocksize = PAGE_CACHE_SIZE;
 196         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 197         sb->s_magic = MQUEUE_MAGIC;
 198         sb->s_op = &mqueue_super_ops;
 199
 200         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 201         if (!inode)
 202                 return -ENOMEM;
 203
 204         sb->s_root = d_alloc_root(inode);
 205         if (!sb->s_root) {
 206                 iput(inode);
 207                 return -ENOMEM;
 208         }
 209
 210         return 0;
 211 }
 212
 213 static int mqueue_get_sb(struct file_system_type *fs_type,
 214                          int flags, const char *dev_name,
 215                          void *data, struct vfsmount *mnt)
 216 {
 217         return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
 218 }
 219
 220 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 221 {
 222         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 223
 224         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 225                 SLAB_CTOR_CONSTRUCTOR)
 226                 inode_init_once(&p->vfs_inode);
 227 }
 228
 229 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 230 {
 231         struct mqueue_inode_info *ei;
 232
 233         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
 234         if (!ei)
 235                 return NULL;
 236         return &ei->vfs_inode;
 237 }
 238
 239 static void mqueue_destroy_inode(struct inode *inode)
 240 {
 241         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 242 }
 243
 244 static void mqueue_delete_inode(struct inode *inode)
 245 {
 246         struct mqueue_inode_info *info;
 247         struct user_struct *user;
 248         unsigned long mq_bytes;
 249         int i;
 250
 251         if (S_ISDIR(inode->i_mode)) {
 252                 clear_inode(inode);
 253                 return;
 254         }
 255         info = MQUEUE_I(inode);
 256         spin_lock(&info->lock);
 257         for (i = 0; i < info->attr.mq_curmsgs; i++)
 258                 free_msg(info->messages[i]);
 259         kfree(info->messages);
 260         spin_unlock(&info->lock);
 261
 262         clear_inode(inode);
 263
 264         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 265                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 266         user = info->user;
 267         if (user) {
 268                 struct vx_info *vxi = lookup_vx_info(user->xid);
 269
 270                 spin_lock(&mq_lock);
 271                 user->mq_bytes -= mq_bytes;
 272                 vx_ipcmsg_sub(vxi, user, mq_bytes);
 273                 queues_count--;
 274                 spin_unlock(&mq_lock);
 275                 put_vx_info(vxi);
 276                 free_uid(user);
 277         }
 278 }
 279
 280 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 281                                 int mode, struct nameidata *nd)
 282 {
 283         struct inode *inode;
 284         struct mq_attr *attr = dentry->d_fsdata;
 285         int error;
 286
 287         spin_lock(&mq_lock);
 288         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 289                 error = -ENOSPC;
 290                 goto out_lock;
 291         }
 292         queues_count++;
 293         spin_unlock(&mq_lock);
 294
 295         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 296         if (!inode) {
 297                 error = -ENOMEM;
 298                 spin_lock(&mq_lock);
 299                 queues_count--;
 300                 goto out_lock;
 301         }
 302
 303         dir->i_size += DIRENT_SIZE;
 304         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 305
 306         d_instantiate(dentry, inode);
 307         dget(dentry);
 308         return 0;
 309 out_lock:
 310         spin_unlock(&mq_lock);
 311         return error;
 312 }
 313
 314 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 315 {
 316         struct inode *inode = dentry->d_inode;
 317
 318         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 319         dir->i_size -= DIRENT_SIZE;
 320         inode->i_nlink--;
 321         dput(dentry);
 322         return 0;
 323 }
 324
 325 /*
 326 *       This is routine for system read from queue file.
 327 *       To avoid mess with doing here some sort of mq_receive we allow
 328 *       to read only queue size & notification info (the only values
 329 *       that are interesting from user point of view and aren't accessible
 330 *       through std routines)
 331 */
 332 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 333                                 size_t count, loff_t * off)
 334 {
 335         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 336         char buffer[FILENT_SIZE];
 337         size_t slen;
 338         loff_t o;
 339
 340         if (!count)
 341                 return 0;
 342
 343         spin_lock(&info->lock);
 344         snprintf(buffer, sizeof(buffer),
 345                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 346                         info->qsize,
 347                         info->notify_owner ? info->notify.sigev_notify : 0,
 348                         (info->notify_owner &&
 349                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 350                                 info->notify.sigev_signo : 0,
 351                         info->notify_owner);
 352         spin_unlock(&info->lock);
 353         buffer[sizeof(buffer)-1] = '\0';
 354         slen = strlen(buffer)+1;
 355
 356         o = *off;
 357         if (o > slen)
 358                 return 0;
 359
 360         if (o + count > slen)
 361                 count = slen - o;
 362
 363         if (copy_to_user(u_data, buffer + o, count))
 364                 return -EFAULT;
 365
 366         *off = o + count;
 367         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
 368         return count;
 369 }
 370
 371 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 372 {
 373         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 374
 375         spin_lock(&info->lock);
 376         if (current->tgid == info->notify_owner)
 377                 remove_notification(info);
 378
 379         spin_unlock(&info->lock);
 380         return 0;
 381 }
 382
 383 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 384 {
 385         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 386         int retval = 0;
 387
 388         poll_wait(filp, &info->wait_q, poll_tab);
 389
 390         spin_lock(&info->lock);
 391         if (info->attr.mq_curmsgs)
 392                 retval = POLLIN | POLLRDNORM;
 393
 394         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 395                 retval |= POLLOUT | POLLWRNORM;
 396         spin_unlock(&info->lock);
 397
 398         return retval;
 399 }
 400
 401 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 402 static void wq_add(struct mqueue_inode_info *info, int sr,
 403                         struct ext_wait_queue *ewp)
 404 {
 405         struct ext_wait_queue *walk;
 406
 407         ewp->task = current;
 408
 409         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 410                 if (walk->task->static_prio <= current->static_prio) {
 411                         list_add_tail(&ewp->list, &walk->list);
 412                         return;
 413                 }
 414         }
 415         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 416 }
 417
 418 /*
 419  * Puts current task to sleep. Caller must hold queue lock. After return
 420  * lock isn't held.
 421  * sr: SEND or RECV
 422  */
 423 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 424                         long timeout, struct ext_wait_queue *ewp)
 425 {
 426         int retval;
 427         signed long time;
 428
 429         wq_add(info, sr, ewp);
 430
 431         for (;;) {
 432                 set_current_state(TASK_INTERRUPTIBLE);
 433
 434                 spin_unlock(&info->lock);
 435                 time = schedule_timeout(timeout);
 436
 437                 while (ewp->state == STATE_PENDING)
 438                         cpu_relax();
 439
 440                 if (ewp->state == STATE_READY) {
 441                         retval = 0;
 442                         goto out;
 443                 }
 444                 spin_lock(&info->lock);
 445                 if (ewp->state == STATE_READY) {
 446                         retval = 0;
 447                         goto out_unlock;
 448                 }
 449                 if (signal_pending(current)) {
 450                         retval = -ERESTARTSYS;
 451                         break;
 452                 }
 453                 if (time == 0) {
 454                         retval = -ETIMEDOUT;
 455                         break;
 456                 }
 457         }
 458         list_del(&ewp->list);
 459 out_unlock:
 460         spin_unlock(&info->lock);
 461 out:
 462         return retval;
 463 }
 464
 465 /*
 466  * Returns waiting task that should be serviced first or NULL if none exists
 467  */
 468 static struct ext_wait_queue *wq_get_first_waiter(
 469                 struct mqueue_inode_info *info, int sr)
 470 {
 471         struct list_head *ptr;
 472
 473         ptr = info->e_wait_q[sr].list.prev;
 474         if (ptr == &info->e_wait_q[sr].list)
 475                 return NULL;
 476         return list_entry(ptr, struct ext_wait_queue, list);
 477 }
 478
 479 /* Auxiliary functions to manipulate messages' list */
 480 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 481 {
 482         int k;
 483
 484         k = info->attr.mq_curmsgs - 1;
 485         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 486                 info->messages[k + 1] = info->messages[k];
 487                 k--;
 488         }
 489         info->attr.mq_curmsgs++;
 490         info->qsize += ptr->m_ts;
 491         info->messages[k + 1] = ptr;
 492 }
 493
 494 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 495 {
 496         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 497         return info->messages[info->attr.mq_curmsgs];
 498 }
 499
 500 static inline void set_cookie(struct sk_buff *skb, char code)
 501 {
 502         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 503 }
 504
 505 /*
 506  * The next function is only to split too long sys_mq_timedsend
 507  */
 508 static void __do_notify(struct mqueue_inode_info *info)
 509 {
 510         /* notification
 511          * invoked when there is registered process and there isn't process
 512          * waiting synchronously for message AND state of queue changed from
 513          * empty to not empty. Here we are sure that no one is waiting
 514          * synchronously. */
 515         if (info->notify_owner &&
 516             info->attr.mq_curmsgs == 1) {
 517                 struct siginfo sig_i;
 518                 switch (info->notify.sigev_notify) {
 519                 case SIGEV_NONE:
 520                         break;
 521                 case SIGEV_SIGNAL:
 522                         /* sends signal */
 523
 524                         sig_i.si_signo = info->notify.sigev_signo;
 525                         sig_i.si_errno = 0;
 526                         sig_i.si_code = SI_MESGQ;
 527                         sig_i.si_value = info->notify.sigev_value;
 528                         sig_i.si_pid = current->tgid;
 529                         sig_i.si_uid = current->uid;
 530
 531                         kill_proc_info(info->notify.sigev_signo,
 532                                        &sig_i, info->notify_owner);
 533                         break;
 534                 case SIGEV_THREAD:
 535                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 536                         netlink_sendskb(info->notify_sock,
 537                                         info->notify_cookie, 0);
 538                         break;
 539                 }
 540                 /* after notification unregisters process */
 541                 info->notify_owner = 0;
 542         }
 543         wake_up(&info->wait_q);
 544 }
 545
 546 static long prepare_timeout(const struct timespec __user *u_arg)
 547 {
 548         struct timespec ts, nowts;
 549         long timeout;
 550
 551         if (u_arg) {
 552                 if (unlikely(copy_from_user(&ts, u_arg,
 553                                         sizeof(struct timespec))))
 554                         return -EFAULT;
 555
 556                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 557                         || ts.tv_nsec >= NSEC_PER_SEC))
 558                         return -EINVAL;
 559                 nowts = CURRENT_TIME;
 560                 /* first subtract as jiffies can't be too big */
 561                 ts.tv_sec -= nowts.tv_sec;
 562                 if (ts.tv_nsec < nowts.tv_nsec) {
 563                         ts.tv_nsec += NSEC_PER_SEC;
 564                         ts.tv_sec--;
 565                 }
 566                 ts.tv_nsec -= nowts.tv_nsec;
 567                 if (ts.tv_sec < 0)
 568                         return 0;
 569
 570                 timeout = timespec_to_jiffies(&ts) + 1;
 571         } else
 572                 return MAX_SCHEDULE_TIMEOUT;
 573
 574         return timeout;
 575 }
 576
 577 static void remove_notification(struct mqueue_inode_info *info)
 578 {
 579         if (info->notify_owner != 0 &&
 580             info->notify.sigev_notify == SIGEV_THREAD) {
 581                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 582                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
 583         }
 584         info->notify_owner = 0;
 585 }
 586
 587 static int mq_attr_ok(struct mq_attr *attr)
 588 {
 589         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 590                 return 0;
 591         if (capable(CAP_SYS_RESOURCE)) {
 592                 if (attr->mq_maxmsg > HARD_MSGMAX)
 593                         return 0;
 594         } else {
 595                 if (attr->mq_maxmsg > msg_max ||
 596                                 attr->mq_msgsize > msgsize_max)
 597                         return 0;
 598         }
 599         /* check for overflow */
 600         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 601                 return 0;
 602         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 603             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 604             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 605                 return 0;
 606         return 1;
 607 }
 608
 609 /*
 610  * Invoked when creating a new queue via sys_mq_open
 611  */
 612 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 613                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 614 {
 615         struct mq_attr attr;
 616         int ret;
 617
 618         if (u_attr) {
 619                 ret = -EFAULT;
 620                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 621                         goto out;
 622                 ret = -EINVAL;
 623                 if (!mq_attr_ok(&attr))
 624                         goto out;
 625                 /* store for use during create */
 626                 dentry->d_fsdata = &attr;
 627         }
 628
 629         mode &= ~current->fs->umask;
 630         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 631         dentry->d_fsdata = NULL;
 632         if (ret)
 633                 goto out;
 634
 635         return dentry_open(dentry, mqueue_mnt, oflag);
 636
 637 out:
 638         dput(dentry);
 639         mntput(mqueue_mnt);
 640         return ERR_PTR(ret);
 641 }
 642
 643 /* Opens existing queue */
 644 static struct file *do_open(struct dentry *dentry, int oflag)
 645 {
 646 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 647                                         MAY_READ | MAY_WRITE };
 648
 649         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 650                 dput(dentry);
 651                 mntput(mqueue_mnt);
 652                 return ERR_PTR(-EINVAL);
 653         }
 654
 655         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
 656                 dput(dentry);
 657                 mntput(mqueue_mnt);
 658                 return ERR_PTR(-EACCES);
 659         }
 660
 661         return dentry_open(dentry, mqueue_mnt, oflag);
 662 }
 663
 664 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 665                                 struct mq_attr __user *u_attr)
 666 {
 667         struct dentry *dentry;
 668         struct file *filp;
 669         char *name;
 670         int fd, error;
 671
 672         error = audit_mq_open(oflag, mode, u_attr);
 673         if (error != 0)
 674                 return error;
 675
 676         if (IS_ERR(name = getname(u_name)))
 677                 return PTR_ERR(name);
 678
 679         fd = get_unused_fd();
 680         if (fd < 0)
 681                 goto out_putname;
 682
 683         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 684         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 685         if (IS_ERR(dentry)) {
 686                 error = PTR_ERR(dentry);
 687                 goto out_err;
 688         }
 689         mntget(mqueue_mnt);
 690
 691         if (oflag & O_CREAT) {
 692                 if (dentry->d_inode) {  /* entry already exists */
 693                         error = -EEXIST;
 694                         if (oflag & O_EXCL)
 695                                 goto out;
 696                         filp = do_open(dentry, oflag);
 697                 } else {
 698                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 699                                                 oflag, mode, u_attr);
 700                 }
 701         } else {
 702                 error = -ENOENT;
 703                 if (!dentry->d_inode)
 704                         goto out;
 705                 filp = do_open(dentry, oflag);
 706         }
 707
 708         if (IS_ERR(filp)) {
 709                 error = PTR_ERR(filp);
 710                 goto out_putfd;
 711         }
 712
 713         set_close_on_exec(fd, 1);
 714         fd_install(fd, filp);
 715         goto out_upsem;
 716
 717 out:
 718         dput(dentry);
 719         mntput(mqueue_mnt);
 720 out_putfd:
 721         put_unused_fd(fd);
 722 out_err:
 723         fd = error;
 724 out_upsem:
 725         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 726 out_putname:
 727         putname(name);
 728         return fd;
 729 }
 730
 731 asmlinkage long sys_mq_unlink(const char __user *u_name)
 732 {
 733         int err;
 734         char *name;
 735         struct dentry *dentry;
 736         struct inode *inode = NULL;
 737
 738         name = getname(u_name);
 739         if (IS_ERR(name))
 740                 return PTR_ERR(name);
 741
 742         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 743         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 744         if (IS_ERR(dentry)) {
 745                 err = PTR_ERR(dentry);
 746                 goto out_unlock;
 747         }
 748
 749         if (!dentry->d_inode) {
 750                 err = -ENOENT;
 751                 goto out_err;
 752         }
 753
 754         inode = dentry->d_inode;
 755         if (inode)
 756                 atomic_inc(&inode->i_count);
 757
 758         err = vfs_unlink(dentry->d_parent->d_inode, dentry, NULL);
 759 out_err:
 760         dput(dentry);
 761
 762 out_unlock:
 763         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 764         putname(name);
 765         if (inode)
 766                 iput(inode);
 767
 768         return err;
 769 }
 770
 771 /* Pipelined send and receive functions.
 772  *
 773  * If a receiver finds no waiting message, then it registers itself in the
 774  * list of waiting receivers. A sender checks that list before adding the new
 775  * message into the message array. If there is a waiting receiver, then it
 776  * bypasses the message array and directly hands the message over to the
 777  * receiver.
 778  * The receiver accepts the message and returns without grabbing the queue
 779  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 780  * are necessary. The same algorithm is used for sysv semaphores, see
 781  * ipc/sem.c for more details.
 782  *
 783  * The same algorithm is used for senders.
 784  */
 785
 786 /* pipelined_send() - send a message directly to the task waiting in
 787  * sys_mq_timedreceive() (without inserting message into a queue).
 788  */
 789 static inline void pipelined_send(struct mqueue_inode_info *info,
 790                                   struct msg_msg *message,
 791                                   struct ext_wait_queue *receiver)
 792 {
 793         receiver->msg = message;
 794         list_del(&receiver->list);
 795         receiver->state = STATE_PENDING;
 796         wake_up_process(receiver->task);
 797         smp_wmb();
 798         receiver->state = STATE_READY;
 799 }
 800
 801 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 802  * gets its message and put to the queue (we have one free place for sure). */
 803 static inline void pipelined_receive(struct mqueue_inode_info *info)
 804 {
 805         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 806
 807         if (!sender) {
 808                 /* for poll */
 809                 wake_up_interruptible(&info->wait_q);
 810                 return;
 811         }
 812         msg_insert(sender->msg, info);
 813         list_del(&sender->list);
 814         sender->state = STATE_PENDING;
 815         wake_up_process(sender->task);
 816         smp_wmb();
 817         sender->state = STATE_READY;
 818 }
 819
 820 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 821         size_t msg_len, unsigned int msg_prio,
 822         const struct timespec __user *u_abs_timeout)
 823 {
 824         struct file *filp;
 825         struct inode *inode;
 826         struct ext_wait_queue wait;
 827         struct ext_wait_queue *receiver;
 828         struct msg_msg *msg_ptr;
 829         struct mqueue_inode_info *info;
 830         long timeout;
 831         int ret;
 832
 833         ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
 834         if (ret != 0)
 835                 return ret;
 836
 837         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 838                 return -EINVAL;
 839
 840         timeout = prepare_timeout(u_abs_timeout);
 841
 842         ret = -EBADF;
 843         filp = fget(mqdes);
 844         if (unlikely(!filp))
 845                 goto out;
 846
 847         inode = filp->f_dentry->d_inode;
 848         if (unlikely(filp->f_op != &mqueue_file_operations))
 849                 goto out_fput;
 850         info = MQUEUE_I(inode);
 851
 852         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 853                 goto out_fput;
 854
 855         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 856                 ret = -EMSGSIZE;
 857                 goto out_fput;
 858         }
 859
 860         /* First try to allocate memory, before doing anything with
 861          * existing queues. */
 862         msg_ptr = load_msg(u_msg_ptr, msg_len);
 863         if (IS_ERR(msg_ptr)) {
 864                 ret = PTR_ERR(msg_ptr);
 865                 goto out_fput;
 866         }
 867         msg_ptr->m_ts = msg_len;
 868         msg_ptr->m_type = msg_prio;
 869
 870         spin_lock(&info->lock);
 871
 872         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 873                 if (filp->f_flags & O_NONBLOCK) {
 874                         spin_unlock(&info->lock);
 875                         ret = -EAGAIN;
 876                 } else if (unlikely(timeout < 0)) {
 877                         spin_unlock(&info->lock);
 878                         ret = timeout;
 879                 } else {
 880                         wait.task = current;
 881                         wait.msg = (void *) msg_ptr;
 882                         wait.state = STATE_NONE;
 883                         ret = wq_sleep(info, SEND, timeout, &wait);
 884                 }
 885                 if (ret < 0)
 886                         free_msg(msg_ptr);
 887         } else {
 888                 receiver = wq_get_first_waiter(info, RECV);
 889                 if (receiver) {
 890                         pipelined_send(info, msg_ptr, receiver);
 891                 } else {
 892                         /* adds message to the queue */
 893                         msg_insert(msg_ptr, info);
 894                         __do_notify(info);
 895                 }
 896                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 897                                 CURRENT_TIME;
 898                 spin_unlock(&info->lock);
 899                 ret = 0;
 900         }
 901 out_fput:
 902         fput(filp);
 903 out:
 904         return ret;
 905 }
 906
 907 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 908         size_t msg_len, unsigned int __user *u_msg_prio,
 909         const struct timespec __user *u_abs_timeout)
 910 {
 911         long timeout;
 912         ssize_t ret;
 913         struct msg_msg *msg_ptr;
 914         struct file *filp;
 915         struct inode *inode;
 916         struct mqueue_inode_info *info;
 917         struct ext_wait_queue wait;
 918
 919         ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
 920         if (ret != 0)
 921                 return ret;
 922
 923         timeout = prepare_timeout(u_abs_timeout);
 924
 925         ret = -EBADF;
 926         filp = fget(mqdes);
 927         if (unlikely(!filp))
 928                 goto out;
 929
 930         inode = filp->f_dentry->d_inode;
 931         if (unlikely(filp->f_op != &mqueue_file_operations))
 932                 goto out_fput;
 933         info = MQUEUE_I(inode);
 934
 935         if (unlikely(!(filp->f_mode & FMODE_READ)))
 936                 goto out_fput;
 937
 938         /* checks if buffer is big enough */
 939         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 940                 ret = -EMSGSIZE;
 941                 goto out_fput;
 942         }
 943
 944         spin_lock(&info->lock);
 945         if (info->attr.mq_curmsgs == 0) {
 946                 if (filp->f_flags & O_NONBLOCK) {
 947                         spin_unlock(&info->lock);
 948                         ret = -EAGAIN;
 949                         msg_ptr = NULL;
 950                 } else if (unlikely(timeout < 0)) {
 951                         spin_unlock(&info->lock);
 952                         ret = timeout;
 953                         msg_ptr = NULL;
 954                 } else {
 955                         wait.task = current;
 956                         wait.state = STATE_NONE;
 957                         ret = wq_sleep(info, RECV, timeout, &wait);
 958                         msg_ptr = wait.msg;
 959                 }
 960         } else {
 961                 msg_ptr = msg_get(info);
 962
 963                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 964                                 CURRENT_TIME;
 965
 966                 /* There is now free space in queue. */
 967                 pipelined_receive(info);
 968                 spin_unlock(&info->lock);
 969                 ret = 0;
 970         }
 971         if (ret == 0) {
 972                 ret = msg_ptr->m_ts;
 973
 974                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 975                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 976                         ret = -EFAULT;
 977                 }
 978                 free_msg(msg_ptr);
 979         }
 980 out_fput:
 981         fput(filp);
 982 out:
 983         return ret;
 984 }
 985
 986 /*
 987  * Notes: the case when user wants us to deregister (with NULL as pointer)
 988  * and he isn't currently owner of notification, will be silently discarded.
 989  * It isn't explicitly defined in the POSIX.
 990  */
 991 asmlinkage long sys_mq_notify(mqd_t mqdes,
 992                                 const struct sigevent __user *u_notification)
 993 {
 994         int ret;
 995         struct file *filp;
 996         struct sock *sock;
 997         struct inode *inode;
 998         struct sigevent notification;
 999         struct mqueue_inode_info *info;
1000         struct sk_buff *nc;
1001
1002         ret = audit_mq_notify(mqdes, u_notification);
1003         if (ret != 0)
1004                 return ret;
1005
1006         nc = NULL;
1007         sock = NULL;
1008         if (u_notification != NULL) {
1009                 if (copy_from_user(&notification, u_notification,
1010                                         sizeof(struct sigevent)))
1011                         return -EFAULT;
1012
1013                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1014                              notification.sigev_notify != SIGEV_SIGNAL &&
1015                              notification.sigev_notify != SIGEV_THREAD))
1016                         return -EINVAL;
1017                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1018                         !valid_signal(notification.sigev_signo)) {
1019                         return -EINVAL;
1020                 }
1021                 if (notification.sigev_notify == SIGEV_THREAD) {
1022                         /* create the notify skb */
1023                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1024                         ret = -ENOMEM;
1025                         if (!nc)
1026                                 goto out;
1027                         ret = -EFAULT;
1028                         if (copy_from_user(nc->data,
1029                                         notification.sigev_value.sival_ptr,
1030                                         NOTIFY_COOKIE_LEN)) {
1031                                 goto out;
1032                         }
1033
1034                         /* TODO: add a header? */
1035                         skb_put(nc, NOTIFY_COOKIE_LEN);
1036                         /* and attach it to the socket */
1037 retry:
1038                         filp = fget(notification.sigev_signo);
1039                         ret = -EBADF;
1040                         if (!filp)
1041                                 goto out;
1042                         sock = netlink_getsockbyfilp(filp);
1043                         fput(filp);
1044                         if (IS_ERR(sock)) {
1045                                 ret = PTR_ERR(sock);
1046                                 sock = NULL;
1047                                 goto out;
1048                         }
1049
1050                         ret = netlink_attachskb(sock, nc, 0,
1051                                         MAX_SCHEDULE_TIMEOUT, NULL);
1052                         if (ret == 1)
1053                                 goto retry;
1054                         if (ret) {
1055                                 sock = NULL;
1056                                 nc = NULL;
1057                                 goto out;
1058                         }
1059                 }
1060         }
1061
1062         ret = -EBADF;
1063         filp = fget(mqdes);
1064         if (!filp)
1065                 goto out;
1066
1067         inode = filp->f_dentry->d_inode;
1068         if (unlikely(filp->f_op != &mqueue_file_operations))
1069                 goto out_fput;
1070         info = MQUEUE_I(inode);
1071
1072         ret = 0;
1073         spin_lock(&info->lock);
1074         if (u_notification == NULL) {
1075                 if (info->notify_owner == current->tgid) {
1076                         remove_notification(info);
1077                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1078                 }
1079         } else if (info->notify_owner != 0) {
1080                 ret = -EBUSY;
1081         } else {
1082                 switch (notification.sigev_notify) {
1083                 case SIGEV_NONE:
1084                         info->notify.sigev_notify = SIGEV_NONE;
1085                         break;
1086                 case SIGEV_THREAD:
1087                         info->notify_sock = sock;
1088                         info->notify_cookie = nc;
1089                         sock = NULL;
1090                         nc = NULL;
1091                         info->notify.sigev_notify = SIGEV_THREAD;
1092                         break;
1093                 case SIGEV_SIGNAL:
1094                         info->notify.sigev_signo = notification.sigev_signo;
1095                         info->notify.sigev_value = notification.sigev_value;
1096                         info->notify.sigev_notify = SIGEV_SIGNAL;
1097                         break;
1098                 }
1099                 info->notify_owner = current->tgid;
1100                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1101         }
1102         spin_unlock(&info->lock);
1103 out_fput:
1104         fput(filp);
1105 out:
1106         if (sock) {
1107                 netlink_detachskb(sock, nc);
1108         } else if (nc) {
1109                 dev_kfree_skb(nc);
1110         }
1111         return ret;
1112 }
1113
1114 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1115                         const struct mq_attr __user *u_mqstat,
1116                         struct mq_attr __user *u_omqstat)
1117 {
1118         int ret;
1119         struct mq_attr mqstat, omqstat;
1120         struct file *filp;
1121         struct inode *inode;
1122         struct mqueue_inode_info *info;
1123
1124         if (u_mqstat != NULL) {
1125                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1126                         return -EFAULT;
1127                 if (mqstat.mq_flags & (~O_NONBLOCK))
1128                         return -EINVAL;
1129         }
1130
1131         ret = -EBADF;
1132         filp = fget(mqdes);
1133         if (!filp)
1134                 goto out;
1135
1136         inode = filp->f_dentry->d_inode;
1137         if (unlikely(filp->f_op != &mqueue_file_operations))
1138                 goto out_fput;
1139         info = MQUEUE_I(inode);
1140
1141         spin_lock(&info->lock);
1142
1143         omqstat = info->attr;
1144         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1145         if (u_mqstat) {
1146                 ret = audit_mq_getsetattr(mqdes, &mqstat);
1147                 if (ret != 0)
1148                         goto out;
1149                 if (mqstat.mq_flags & O_NONBLOCK)
1150                         filp->f_flags |= O_NONBLOCK;
1151                 else
1152                         filp->f_flags &= ~O_NONBLOCK;
1153
1154                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1155         }
1156
1157         spin_unlock(&info->lock);
1158
1159         ret = 0;
1160         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1161                                                 sizeof(struct mq_attr)))
1162                 ret = -EFAULT;
1163
1164 out_fput:
1165         fput(filp);
1166 out:
1167         return ret;
1168 }
1169
1170 static struct inode_operations mqueue_dir_inode_operations = {
1171         .lookup = simple_lookup,
1172         .create = mqueue_create,
1173         .unlink = mqueue_unlink,
1174 };
1175
1176 static struct file_operations mqueue_file_operations = {
1177         .flush = mqueue_flush_file,
1178         .poll = mqueue_poll_file,
1179         .read = mqueue_read_file,
1180 };
1181
1182 static struct super_operations mqueue_super_ops = {
1183         .alloc_inode = mqueue_alloc_inode,
1184         .destroy_inode = mqueue_destroy_inode,
1185         .statfs = simple_statfs,
1186         .delete_inode = mqueue_delete_inode,
1187         .drop_inode = generic_delete_inode,
1188 };
1189
1190 static struct file_system_type mqueue_fs_type = {
1191         .name = "mqueue",
1192         .get_sb = mqueue_get_sb,
1193         .kill_sb = kill_litter_super,
1194 };
1195
1196 static int msg_max_limit_min = DFLT_MSGMAX;
1197 static int msg_max_limit_max = HARD_MSGMAX;
1198
1199 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1200 static int msg_maxsize_limit_max = INT_MAX;
1201
1202 static ctl_table mq_sysctls[] = {
1203         {
1204                 .ctl_name       = CTL_QUEUESMAX,
1205                 .procname       = "queues_max",
1206                 .data           = &queues_max,
1207                 .maxlen         = sizeof(int),
1208                 .mode           = 0644,
1209                 .proc_handler   = &proc_dointvec,
1210         },
1211         {
1212                 .ctl_name       = CTL_MSGMAX,
1213                 .procname       = "msg_max",
1214                 .data           = &msg_max,
1215                 .maxlen         = sizeof(int),
1216                 .mode           = 0644,
1217                 .proc_handler   = &proc_dointvec_minmax,
1218                 .extra1         = &msg_max_limit_min,
1219                 .extra2         = &msg_max_limit_max,
1220         },
1221         {
1222                 .ctl_name       = CTL_MSGSIZEMAX,
1223                 .procname       = "msgsize_max",
1224                 .data           = &msgsize_max,
1225                 .maxlen         = sizeof(int),
1226                 .mode           = 0644,
1227                 .proc_handler   = &proc_dointvec_minmax,
1228                 .extra1         = &msg_maxsize_limit_min,
1229                 .extra2         = &msg_maxsize_limit_max,
1230         },
1231         { .ctl_name = 0 }
1232 };
1233
1234 static ctl_table mq_sysctl_dir[] = {
1235         {
1236                 .ctl_name       = FS_MQUEUE,
1237                 .procname       = "mqueue",
1238                 .mode           = 0555,
1239                 .child          = mq_sysctls,
1240         },
1241         { .ctl_name = 0 }
1242 };
1243
1244 static ctl_table mq_sysctl_root[] = {
1245         {
1246                 .ctl_name       = CTL_FS,
1247                 .procname       = "fs",
1248                 .mode           = 0555,
1249                 .child          = mq_sysctl_dir,
1250         },
1251         { .ctl_name = 0 }
1252 };
1253
1254 static int __init init_mqueue_fs(void)
1255 {
1256         int error;
1257
1258         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1259                                 sizeof(struct mqueue_inode_info), 0,
1260                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1261         if (mqueue_inode_cachep == NULL)
1262                 return -ENOMEM;
1263
1264         /* ignore failues - they are not fatal */
1265         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1266
1267         error = register_filesystem(&mqueue_fs_type);
1268         if (error)
1269                 goto out_sysctl;
1270
1271         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1272                 error = PTR_ERR(mqueue_mnt);
1273                 goto out_filesystem;
1274         }
1275
1276         /* internal initialization - not common for vfs */
1277         queues_count = 0;
1278         spin_lock_init(&mq_lock);
1279
1280         return 0;
1281
1282 out_filesystem:
1283         unregister_filesystem(&mqueue_fs_type);
1284 out_sysctl:
1285         if (mq_sysctl_table)
1286                 unregister_sysctl_table(mq_sysctl_table);
1287         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1288                 printk(KERN_INFO
1289                         "mqueue_inode_cache: not all structures were freed\n");
1290         }
1291         return error;
1292 }
1293
1294 __initcall(init_mqueue_fs);