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