ipc/mqueue.c

   1 /*
   2  * POSIX message queues filesystem for Linux.
   3  *
   4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
   5  *                          Michal Wronski          (michal.wronski@gmail.com)
   6  *
   7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
   8  * Lockless receive & send, fd based notify:
   9  *                          Manfred Spraul          (manfred@colorfullife.com)
  10  *
  11  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
  12  *
  13  * This file is released under the GPL.
  14  */
  15
  16 #include <linux/capability.h>
  17 #include <linux/init.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/file.h>
  20 #include <linux/mount.h>
  21 #include <linux/namei.h>
  22 #include <linux/sysctl.h>
  23 #include <linux/poll.h>
  24 #include <linux/mqueue.h>
  25 #include <linux/msg.h>
  26 #include <linux/skbuff.h>
  27 #include <linux/netlink.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/audit.h>
  30 #include <linux/signal.h>
  31 #include <linux/mutex.h>
  32 #include <linux/nsproxy.h>
  33 #include <linux/pid.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 /* default values */
  50 #define DFLT_QUEUESMAX  256     /* max number of message queues */
  51 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
  52 #define HARD_MSGMAX     (131072/sizeof(void*))
  53 #define DFLT_MSGSIZEMAX 8192    /* max message size */
  54
  55 /*
  56  * Define the ranges various user-specified maximum values can
  57  * be set to.
  58  */
  59 #define MIN_MSGMAX      1               /* min value for msg_max */
  60 #define MAX_MSGMAX      HARD_MSGMAX     /* max value for msg_max */
  61 #define MIN_MSGSIZEMAX  128             /* min value for msgsize_max */
  62 #define MAX_MSGSIZEMAX  (8192*128)      /* max value for msgsize_max */
  63
  64 struct ext_wait_queue {         /* queue of sleeping tasks */
  65         struct task_struct *task;
  66         struct list_head list;
  67         struct msg_msg *msg;    /* ptr of loaded message */
  68         int state;              /* one of STATE_* values */
  69 };
  70
  71 struct mqueue_inode_info {
  72         spinlock_t lock;
  73         struct inode vfs_inode;
  74         wait_queue_head_t wait_q;
  75
  76         struct msg_msg **messages;
  77         struct mq_attr attr;
  78
  79         struct sigevent notify;
  80         struct pid* notify_owner;
  81         struct user_struct *user;       /* user who created, for accounting */
  82         struct sock *notify_sock;
  83         struct sk_buff *notify_cookie;
  84
  85         /* for tasks waiting for free space and messages, respectively */
  86         struct ext_wait_queue e_wait_q[2];
  87
  88         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  89 };
  90
  91 static const struct inode_operations mqueue_dir_inode_operations;
  92 static const struct file_operations mqueue_file_operations;
  93 static struct super_operations mqueue_super_ops;
  94 static void remove_notification(struct mqueue_inode_info *info);
  95
  96 static spinlock_t mq_lock;
  97 static struct kmem_cache *mqueue_inode_cachep;
  98 static struct vfsmount *mqueue_mnt;
  99
 100 static unsigned int queues_count;
 101 static unsigned int queues_max  = DFLT_QUEUESMAX;
 102 static unsigned int msg_max     = DFLT_MSGMAX;
 103 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
 104
 105 static struct ctl_table_header * mq_sysctl_table;
 106
 107 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
 108 {
 109         return container_of(inode, struct mqueue_inode_info, vfs_inode);
 110 }
 111
 112 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 113                                                         struct mq_attr *attr)
 114 {
 115         struct inode *inode;
 116
 117         inode = new_inode(sb);
 118         if (inode) {
 119                 inode->i_mode = mode;
 120                 inode->i_uid = current->fsuid;
 121                 inode->i_gid = current->fsgid;
 122                 inode->i_blocks = 0;
 123                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 124                                 CURRENT_TIME;
 125
 126                 if (S_ISREG(mode)) {
 127                         struct mqueue_inode_info *info;
 128                         struct task_struct *p = current;
 129                         struct user_struct *u = p->user;
 130                         unsigned long mq_bytes, mq_msg_tblsz;
 131
 132                         inode->i_fop = &mqueue_file_operations;
 133                         inode->i_size = FILENT_SIZE;
 134                         /* mqueue specific info */
 135                         info = MQUEUE_I(inode);
 136                         spin_lock_init(&info->lock);
 137                         init_waitqueue_head(&info->wait_q);
 138                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 139                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 140                         info->messages = NULL;
 141                         info->notify_owner = NULL;
 142                         info->qsize = 0;
 143                         info->user = NULL;      /* set when all is ok */
 144                         memset(&info->attr, 0, sizeof(info->attr));
 145                         info->attr.mq_maxmsg = msg_max;
 146                         info->attr.mq_msgsize = msgsize_max;
 147                         if (attr) {
 148                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 149                                 info->attr.mq_msgsize = attr->mq_msgsize;
 150                         }
 151                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 152                         mq_bytes = (mq_msg_tblsz +
 153                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 154
 155                         spin_lock(&mq_lock);
 156                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 157                             u->mq_bytes + mq_bytes >
 158                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
 159                                 spin_unlock(&mq_lock);
 160                                 goto out_inode;
 161                         }
 162                         u->mq_bytes += 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                                 spin_unlock(&mq_lock);
 170                                 goto out_inode;
 171                         }
 172                         /* all is ok */
 173                         info->user = get_uid(u);
 174                 } else if (S_ISDIR(mode)) {
 175                         inc_nlink(inode);
 176                         /* Some things misbehave if size == 0 on a directory */
 177                         inode->i_size = 2 * DIRENT_SIZE;
 178                         inode->i_op = &mqueue_dir_inode_operations;
 179                         inode->i_fop = &simple_dir_operations;
 180                 }
 181         }
 182         return inode;
 183 out_inode:
 184         make_bad_inode(inode);
 185         iput(inode);
 186         return NULL;
 187 }
 188
 189 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 190 {
 191         struct inode *inode;
 192
 193         sb->s_blocksize = PAGE_CACHE_SIZE;
 194         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 195         sb->s_magic = MQUEUE_MAGIC;
 196         sb->s_op = &mqueue_super_ops;
 197
 198         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 199         if (!inode)
 200                 return -ENOMEM;
 201
 202         sb->s_root = d_alloc_root(inode);
 203         if (!sb->s_root) {
 204                 iput(inode);
 205                 return -ENOMEM;
 206         }
 207
 208         return 0;
 209 }
 210
 211 static int mqueue_get_sb(struct file_system_type *fs_type,
 212                          int flags, const char *dev_name,
 213                          void *data, struct vfsmount *mnt)
 214 {
 215         return get_sb_single(fs_type, flags, data, mqueue_fill_super, mnt);
 216 }
 217
 218 static void init_once(void *foo)
 219 {
 220         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 221
 222         inode_init_once(&p->vfs_inode);
 223 }
 224
 225 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 226 {
 227         struct mqueue_inode_info *ei;
 228
 229         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 230         if (!ei)
 231                 return NULL;
 232         return &ei->vfs_inode;
 233 }
 234
 235 static void mqueue_destroy_inode(struct inode *inode)
 236 {
 237         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 238 }
 239
 240 static void mqueue_delete_inode(struct inode *inode)
 241 {
 242         struct mqueue_inode_info *info;
 243         struct user_struct *user;
 244         unsigned long mq_bytes;
 245         int i;
 246
 247         if (S_ISDIR(inode->i_mode)) {
 248                 clear_inode(inode);
 249                 return;
 250         }
 251         info = MQUEUE_I(inode);
 252         spin_lock(&info->lock);
 253         for (i = 0; i < info->attr.mq_curmsgs; i++)
 254                 free_msg(info->messages[i]);
 255         kfree(info->messages);
 256         spin_unlock(&info->lock);
 257
 258         clear_inode(inode);
 259
 260         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 261                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 262         user = info->user;
 263         if (user) {
 264                 spin_lock(&mq_lock);
 265                 user->mq_bytes -= mq_bytes;
 266                 queues_count--;
 267                 spin_unlock(&mq_lock);
 268                 free_uid(user);
 269         }
 270 }
 271
 272 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 273                                 int mode, struct nameidata *nd)
 274 {
 275         struct inode *inode;
 276         struct mq_attr *attr = dentry->d_fsdata;
 277         int error;
 278
 279         spin_lock(&mq_lock);
 280         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 281                 error = -ENOSPC;
 282                 goto out_lock;
 283         }
 284         queues_count++;
 285         spin_unlock(&mq_lock);
 286
 287         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 288         if (!inode) {
 289                 error = -ENOMEM;
 290                 spin_lock(&mq_lock);
 291                 queues_count--;
 292                 goto out_lock;
 293         }
 294
 295         dir->i_size += DIRENT_SIZE;
 296         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 297
 298         d_instantiate(dentry, inode);
 299         dget(dentry);
 300         return 0;
 301 out_lock:
 302         spin_unlock(&mq_lock);
 303         return error;
 304 }
 305
 306 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 307 {
 308         struct inode *inode = dentry->d_inode;
 309
 310         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 311         dir->i_size -= DIRENT_SIZE;
 312         drop_nlink(inode);
 313         dput(dentry);
 314         return 0;
 315 }
 316
 317 /*
 318 *       This is routine for system read from queue file.
 319 *       To avoid mess with doing here some sort of mq_receive we allow
 320 *       to read only queue size & notification info (the only values
 321 *       that are interesting from user point of view and aren't accessible
 322 *       through std routines)
 323 */
 324 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 325                                 size_t count, loff_t *off)
 326 {
 327         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 328         char buffer[FILENT_SIZE];
 329         ssize_t ret;
 330
 331         spin_lock(&info->lock);
 332         snprintf(buffer, sizeof(buffer),
 333                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 334                         info->qsize,
 335                         info->notify_owner ? info->notify.sigev_notify : 0,
 336                         (info->notify_owner &&
 337                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 338                                 info->notify.sigev_signo : 0,
 339                         pid_vnr(info->notify_owner));
 340         spin_unlock(&info->lock);
 341         buffer[sizeof(buffer)-1] = '\0';
 342
 343         ret = simple_read_from_buffer(u_data, count, off, buffer,
 344                                 strlen(buffer));
 345         if (ret <= 0)
 346                 return ret;
 347
 348         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 349         return ret;
 350 }
 351
 352 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 353 {
 354         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 355
 356         spin_lock(&info->lock);
 357         if (task_tgid(current) == info->notify_owner)
 358                 remove_notification(info);
 359
 360         spin_unlock(&info->lock);
 361         return 0;
 362 }
 363
 364 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 365 {
 366         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 367         int retval = 0;
 368
 369         poll_wait(filp, &info->wait_q, poll_tab);
 370
 371         spin_lock(&info->lock);
 372         if (info->attr.mq_curmsgs)
 373                 retval = POLLIN | POLLRDNORM;
 374
 375         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 376                 retval |= POLLOUT | POLLWRNORM;
 377         spin_unlock(&info->lock);
 378
 379         return retval;
 380 }
 381
 382 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 383 static void wq_add(struct mqueue_inode_info *info, int sr,
 384                         struct ext_wait_queue *ewp)
 385 {
 386         struct ext_wait_queue *walk;
 387
 388         ewp->task = current;
 389
 390         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 391                 if (walk->task->static_prio <= current->static_prio) {
 392                         list_add_tail(&ewp->list, &walk->list);
 393                         return;
 394                 }
 395         }
 396         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 397 }
 398
 399 /*
 400  * Puts current task to sleep. Caller must hold queue lock. After return
 401  * lock isn't held.
 402  * sr: SEND or RECV
 403  */
 404 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 405                         long timeout, struct ext_wait_queue *ewp)
 406 {
 407         int retval;
 408         signed long time;
 409
 410         wq_add(info, sr, ewp);
 411
 412         for (;;) {
 413                 set_current_state(TASK_INTERRUPTIBLE);
 414
 415                 spin_unlock(&info->lock);
 416                 time = schedule_timeout(timeout);
 417
 418                 while (ewp->state == STATE_PENDING)
 419                         cpu_relax();
 420
 421                 if (ewp->state == STATE_READY) {
 422                         retval = 0;
 423                         goto out;
 424                 }
 425                 spin_lock(&info->lock);
 426                 if (ewp->state == STATE_READY) {
 427                         retval = 0;
 428                         goto out_unlock;
 429                 }
 430                 if (signal_pending(current)) {
 431                         retval = -ERESTARTSYS;
 432                         break;
 433                 }
 434                 if (time == 0) {
 435                         retval = -ETIMEDOUT;
 436                         break;
 437                 }
 438         }
 439         list_del(&ewp->list);
 440 out_unlock:
 441         spin_unlock(&info->lock);
 442 out:
 443         return retval;
 444 }
 445
 446 /*
 447  * Returns waiting task that should be serviced first or NULL if none exists
 448  */
 449 static struct ext_wait_queue *wq_get_first_waiter(
 450                 struct mqueue_inode_info *info, int sr)
 451 {
 452         struct list_head *ptr;
 453
 454         ptr = info->e_wait_q[sr].list.prev;
 455         if (ptr == &info->e_wait_q[sr].list)
 456                 return NULL;
 457         return list_entry(ptr, struct ext_wait_queue, list);
 458 }
 459
 460 /* Auxiliary functions to manipulate messages' list */
 461 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 462 {
 463         int k;
 464
 465         k = info->attr.mq_curmsgs - 1;
 466         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 467                 info->messages[k + 1] = info->messages[k];
 468                 k--;
 469         }
 470         info->attr.mq_curmsgs++;
 471         info->qsize += ptr->m_ts;
 472         info->messages[k + 1] = ptr;
 473 }
 474
 475 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 476 {
 477         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 478         return info->messages[info->attr.mq_curmsgs];
 479 }
 480
 481 static inline void set_cookie(struct sk_buff *skb, char code)
 482 {
 483         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 484 }
 485
 486 /*
 487  * The next function is only to split too long sys_mq_timedsend
 488  */
 489 static void __do_notify(struct mqueue_inode_info *info)
 490 {
 491         /* notification
 492          * invoked when there is registered process and there isn't process
 493          * waiting synchronously for message AND state of queue changed from
 494          * empty to not empty. Here we are sure that no one is waiting
 495          * synchronously. */
 496         if (info->notify_owner &&
 497             info->attr.mq_curmsgs == 1) {
 498                 struct siginfo sig_i;
 499                 switch (info->notify.sigev_notify) {
 500                 case SIGEV_NONE:
 501                         break;
 502                 case SIGEV_SIGNAL:
 503                         /* sends signal */
 504
 505                         sig_i.si_signo = info->notify.sigev_signo;
 506                         sig_i.si_errno = 0;
 507                         sig_i.si_code = SI_MESGQ;
 508                         sig_i.si_value = info->notify.sigev_value;
 509                         sig_i.si_pid = task_tgid_vnr(current);
 510                         sig_i.si_uid = current->uid;
 511
 512                         kill_pid_info(info->notify.sigev_signo,
 513                                       &sig_i, info->notify_owner);
 514                         break;
 515                 case SIGEV_THREAD:
 516                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 517                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 518                         break;
 519                 }
 520                 /* after notification unregisters process */
 521                 put_pid(info->notify_owner);
 522                 info->notify_owner = NULL;
 523         }
 524         wake_up(&info->wait_q);
 525 }
 526
 527 static long prepare_timeout(const struct timespec __user *u_arg)
 528 {
 529         struct timespec ts, nowts;
 530         long timeout;
 531
 532         if (u_arg) {
 533                 if (unlikely(copy_from_user(&ts, u_arg,
 534                                         sizeof(struct timespec))))
 535                         return -EFAULT;
 536
 537                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 538                         || ts.tv_nsec >= NSEC_PER_SEC))
 539                         return -EINVAL;
 540                 nowts = CURRENT_TIME;
 541                 /* first subtract as jiffies can't be too big */
 542                 ts.tv_sec -= nowts.tv_sec;
 543                 if (ts.tv_nsec < nowts.tv_nsec) {
 544                         ts.tv_nsec += NSEC_PER_SEC;
 545                         ts.tv_sec--;
 546                 }
 547                 ts.tv_nsec -= nowts.tv_nsec;
 548                 if (ts.tv_sec < 0)
 549                         return 0;
 550
 551                 timeout = timespec_to_jiffies(&ts) + 1;
 552         } else
 553                 return MAX_SCHEDULE_TIMEOUT;
 554
 555         return timeout;
 556 }
 557
 558 static void remove_notification(struct mqueue_inode_info *info)
 559 {
 560         if (info->notify_owner != NULL &&
 561             info->notify.sigev_notify == SIGEV_THREAD) {
 562                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 563                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 564         }
 565         put_pid(info->notify_owner);
 566         info->notify_owner = NULL;
 567 }
 568
 569 static int mq_attr_ok(struct mq_attr *attr)
 570 {
 571         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 572                 return 0;
 573         if (capable(CAP_SYS_RESOURCE)) {
 574                 if (attr->mq_maxmsg > HARD_MSGMAX)
 575                         return 0;
 576         } else {
 577                 if (attr->mq_maxmsg > msg_max ||
 578                                 attr->mq_msgsize > msgsize_max)
 579                         return 0;
 580         }
 581         /* check for overflow */
 582         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 583                 return 0;
 584         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 585             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 586             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 587                 return 0;
 588         return 1;
 589 }
 590
 591 /*
 592  * Invoked when creating a new queue via sys_mq_open
 593  */
 594 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 595                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 596 {
 597         struct mq_attr attr;
 598         struct file *result;
 599         int ret;
 600
 601         if (u_attr) {
 602                 ret = -EFAULT;
 603                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 604                         goto out;
 605                 ret = -EINVAL;
 606                 if (!mq_attr_ok(&attr))
 607                         goto out;
 608                 /* store for use during create */
 609                 dentry->d_fsdata = &attr;
 610         }
 611
 612         mode &= ~current->fs->umask;
 613         ret = mnt_want_write(mqueue_mnt);
 614         if (ret)
 615                 goto out;
 616         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 617         dentry->d_fsdata = NULL;
 618         if (ret)
 619                 goto out_drop_write;
 620
 621         result = dentry_open(dentry, mqueue_mnt, oflag);
 622         /*
 623          * dentry_open() took a persistent mnt_want_write(),
 624          * so we can now drop this one.
 625          */
 626         mnt_drop_write(mqueue_mnt);
 627         return result;
 628
 629 out_drop_write:
 630         mnt_drop_write(mqueue_mnt);
 631 out:
 632         dput(dentry);
 633         mntput(mqueue_mnt);
 634         return ERR_PTR(ret);
 635 }
 636
 637 /* Opens existing queue */
 638 static struct file *do_open(struct dentry *dentry, int oflag)
 639 {
 640 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 641                                         MAY_READ | MAY_WRITE };
 642
 643         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 644                 dput(dentry);
 645                 mntput(mqueue_mnt);
 646                 return ERR_PTR(-EINVAL);
 647         }
 648
 649         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 650                 dput(dentry);
 651                 mntput(mqueue_mnt);
 652                 return ERR_PTR(-EACCES);
 653         }
 654
 655         return dentry_open(dentry, mqueue_mnt, oflag);
 656 }
 657
 658 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 659                                 struct mq_attr __user *u_attr)
 660 {
 661         struct dentry *dentry;
 662         struct file *filp;
 663         char *name;
 664         int fd, error;
 665
 666         error = audit_mq_open(oflag, mode, u_attr);
 667         if (error != 0)
 668                 return error;
 669
 670         if (IS_ERR(name = getname(u_name)))
 671                 return PTR_ERR(name);
 672
 673         fd = get_unused_fd_flags(O_CLOEXEC);
 674         if (fd < 0)
 675                 goto out_putname;
 676
 677         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 678         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 679         if (IS_ERR(dentry)) {
 680                 error = PTR_ERR(dentry);
 681                 goto out_err;
 682         }
 683         mntget(mqueue_mnt);
 684
 685         if (oflag & O_CREAT) {
 686                 if (dentry->d_inode) {  /* entry already exists */
 687                         audit_inode(name, dentry);
 688                         error = -EEXIST;
 689                         if (oflag & O_EXCL)
 690                                 goto out;
 691                         filp = do_open(dentry, oflag);
 692                 } else {
 693                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 694                                                 oflag, mode, u_attr);
 695                 }
 696         } else {
 697                 error = -ENOENT;
 698                 if (!dentry->d_inode)
 699                         goto out;
 700                 audit_inode(name, dentry);
 701                 filp = do_open(dentry, oflag);
 702         }
 703
 704         if (IS_ERR(filp)) {
 705                 error = PTR_ERR(filp);
 706                 goto out_putfd;
 707         }
 708
 709         fd_install(fd, filp);
 710         goto out_upsem;
 711
 712 out:
 713         dput(dentry);
 714         mntput(mqueue_mnt);
 715 out_putfd:
 716         put_unused_fd(fd);
 717 out_err:
 718         fd = error;
 719 out_upsem:
 720         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 721 out_putname:
 722         putname(name);
 723         return fd;
 724 }
 725
 726 asmlinkage long sys_mq_unlink(const char __user *u_name)
 727 {
 728         int err;
 729         char *name;
 730         struct dentry *dentry;
 731         struct inode *inode = NULL;
 732
 733         name = getname(u_name);
 734         if (IS_ERR(name))
 735                 return PTR_ERR(name);
 736
 737         mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
 738                         I_MUTEX_PARENT);
 739         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 740         if (IS_ERR(dentry)) {
 741                 err = PTR_ERR(dentry);
 742                 goto out_unlock;
 743         }
 744
 745         if (!dentry->d_inode) {
 746                 err = -ENOENT;
 747                 goto out_err;
 748         }
 749
 750         inode = dentry->d_inode;
 751         if (inode)
 752                 atomic_inc(&inode->i_count);
 753         err = mnt_want_write(mqueue_mnt);
 754         if (err)
 755                 goto out_err;
 756         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 757         mnt_drop_write(mqueue_mnt);
 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_path.dentry->d_inode;
 847         if (unlikely(filp->f_op != &mqueue_file_operations))
 848                 goto out_fput;
 849         info = MQUEUE_I(inode);
 850         audit_inode(NULL, filp->f_path.dentry);
 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_path.dentry->d_inode;
 931         if (unlikely(filp->f_op != &mqueue_file_operations))
 932                 goto out_fput;
 933         info = MQUEUE_I(inode);
 934         audit_inode(NULL, filp->f_path.dentry);
 935
 936         if (unlikely(!(filp->f_mode & FMODE_READ)))
 937                 goto out_fput;
 938
 939         /* checks if buffer is big enough */
 940         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 941                 ret = -EMSGSIZE;
 942                 goto out_fput;
 943         }
 944
 945         spin_lock(&info->lock);
 946         if (info->attr.mq_curmsgs == 0) {
 947                 if (filp->f_flags & O_NONBLOCK) {
 948                         spin_unlock(&info->lock);
 949                         ret = -EAGAIN;
 950                         msg_ptr = NULL;
 951                 } else if (unlikely(timeout < 0)) {
 952                         spin_unlock(&info->lock);
 953                         ret = timeout;
 954                         msg_ptr = NULL;
 955                 } else {
 956                         wait.task = current;
 957                         wait.state = STATE_NONE;
 958                         ret = wq_sleep(info, RECV, timeout, &wait);
 959                         msg_ptr = wait.msg;
 960                 }
 961         } else {
 962                 msg_ptr = msg_get(info);
 963
 964                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 965                                 CURRENT_TIME;
 966
 967                 /* There is now free space in queue. */
 968                 pipelined_receive(info);
 969                 spin_unlock(&info->lock);
 970                 ret = 0;
 971         }
 972         if (ret == 0) {
 973                 ret = msg_ptr->m_ts;
 974
 975                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 976                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 977                         ret = -EFAULT;
 978                 }
 979                 free_msg(msg_ptr);
 980         }
 981 out_fput:
 982         fput(filp);
 983 out:
 984         return ret;
 985 }
 986
 987 /*
 988  * Notes: the case when user wants us to deregister (with NULL as pointer)
 989  * and he isn't currently owner of notification, will be silently discarded.
 990  * It isn't explicitly defined in the POSIX.
 991  */
 992 asmlinkage long sys_mq_notify(mqd_t mqdes,
 993                                 const struct sigevent __user *u_notification)
 994 {
 995         int ret;
 996         struct file *filp;
 997         struct sock *sock;
 998         struct inode *inode;
 999         struct sigevent notification;
1000         struct mqueue_inode_info *info;
1001         struct sk_buff *nc;
1002
1003         ret = audit_mq_notify(mqdes, u_notification);
1004         if (ret != 0)
1005                 return ret;
1006
1007         nc = NULL;
1008         sock = NULL;
1009         if (u_notification != NULL) {
1010                 if (copy_from_user(&notification, u_notification,
1011                                         sizeof(struct sigevent)))
1012                         return -EFAULT;
1013
1014                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1015                              notification.sigev_notify != SIGEV_SIGNAL &&
1016                              notification.sigev_notify != SIGEV_THREAD))
1017                         return -EINVAL;
1018                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1019                         !valid_signal(notification.sigev_signo)) {
1020                         return -EINVAL;
1021                 }
1022                 if (notification.sigev_notify == SIGEV_THREAD) {
1023                         long timeo;
1024
1025                         /* create the notify skb */
1026                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1027                         ret = -ENOMEM;
1028                         if (!nc)
1029                                 goto out;
1030                         ret = -EFAULT;
1031                         if (copy_from_user(nc->data,
1032                                         notification.sigev_value.sival_ptr,
1033                                         NOTIFY_COOKIE_LEN)) {
1034                                 goto out;
1035                         }
1036
1037                         /* TODO: add a header? */
1038                         skb_put(nc, NOTIFY_COOKIE_LEN);
1039                         /* and attach it to the socket */
1040 retry:
1041                         filp = fget(notification.sigev_signo);
1042                         ret = -EBADF;
1043                         if (!filp)
1044                                 goto out;
1045                         sock = netlink_getsockbyfilp(filp);
1046                         fput(filp);
1047                         if (IS_ERR(sock)) {
1048                                 ret = PTR_ERR(sock);
1049                                 sock = NULL;
1050                                 goto out;
1051                         }
1052
1053                         timeo = MAX_SCHEDULE_TIMEOUT;
1054                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1055                         if (ret == 1)
1056                                 goto retry;
1057                         if (ret) {
1058                                 sock = NULL;
1059                                 nc = NULL;
1060                                 goto out;
1061                         }
1062                 }
1063         }
1064
1065         ret = -EBADF;
1066         filp = fget(mqdes);
1067         if (!filp)
1068                 goto out;
1069
1070         inode = filp->f_path.dentry->d_inode;
1071         if (unlikely(filp->f_op != &mqueue_file_operations))
1072                 goto out_fput;
1073         info = MQUEUE_I(inode);
1074
1075         ret = 0;
1076         spin_lock(&info->lock);
1077         if (u_notification == NULL) {
1078                 if (info->notify_owner == task_tgid(current)) {
1079                         remove_notification(info);
1080                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1081                 }
1082         } else if (info->notify_owner != NULL) {
1083                 ret = -EBUSY;
1084         } else {
1085                 switch (notification.sigev_notify) {
1086                 case SIGEV_NONE:
1087                         info->notify.sigev_notify = SIGEV_NONE;
1088                         break;
1089                 case SIGEV_THREAD:
1090                         info->notify_sock = sock;
1091                         info->notify_cookie = nc;
1092                         sock = NULL;
1093                         nc = NULL;
1094                         info->notify.sigev_notify = SIGEV_THREAD;
1095                         break;
1096                 case SIGEV_SIGNAL:
1097                         info->notify.sigev_signo = notification.sigev_signo;
1098                         info->notify.sigev_value = notification.sigev_value;
1099                         info->notify.sigev_notify = SIGEV_SIGNAL;
1100                         break;
1101                 }
1102
1103                 info->notify_owner = get_pid(task_tgid(current));
1104                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1105         }
1106         spin_unlock(&info->lock);
1107 out_fput:
1108         fput(filp);
1109 out:
1110         if (sock) {
1111                 netlink_detachskb(sock, nc);
1112         } else if (nc) {
1113                 dev_kfree_skb(nc);
1114         }
1115         return ret;
1116 }
1117
1118 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1119                         const struct mq_attr __user *u_mqstat,
1120                         struct mq_attr __user *u_omqstat)
1121 {
1122         int ret;
1123         struct mq_attr mqstat, omqstat;
1124         struct file *filp;
1125         struct inode *inode;
1126         struct mqueue_inode_info *info;
1127
1128         if (u_mqstat != NULL) {
1129                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1130                         return -EFAULT;
1131                 if (mqstat.mq_flags & (~O_NONBLOCK))
1132                         return -EINVAL;
1133         }
1134
1135         ret = -EBADF;
1136         filp = fget(mqdes);
1137         if (!filp)
1138                 goto out;
1139
1140         inode = filp->f_path.dentry->d_inode;
1141         if (unlikely(filp->f_op != &mqueue_file_operations))
1142                 goto out_fput;
1143         info = MQUEUE_I(inode);
1144
1145         spin_lock(&info->lock);
1146
1147         omqstat = info->attr;
1148         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1149         if (u_mqstat) {
1150                 ret = audit_mq_getsetattr(mqdes, &mqstat);
1151                 if (ret != 0) {
1152                         spin_unlock(&info->lock);
1153                         goto out_fput;
1154                 }
1155                 if (mqstat.mq_flags & O_NONBLOCK)
1156                         filp->f_flags |= O_NONBLOCK;
1157                 else
1158                         filp->f_flags &= ~O_NONBLOCK;
1159
1160                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1161         }
1162
1163         spin_unlock(&info->lock);
1164
1165         ret = 0;
1166         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1167                                                 sizeof(struct mq_attr)))
1168                 ret = -EFAULT;
1169
1170 out_fput:
1171         fput(filp);
1172 out:
1173         return ret;
1174 }
1175
1176 static const struct inode_operations mqueue_dir_inode_operations = {
1177         .lookup = simple_lookup,
1178         .create = mqueue_create,
1179         .unlink = mqueue_unlink,
1180 };
1181
1182 static const struct file_operations mqueue_file_operations = {
1183         .flush = mqueue_flush_file,
1184         .poll = mqueue_poll_file,
1185         .read = mqueue_read_file,
1186 };
1187
1188 static struct super_operations mqueue_super_ops = {
1189         .alloc_inode = mqueue_alloc_inode,
1190         .destroy_inode = mqueue_destroy_inode,
1191         .statfs = simple_statfs,
1192         .delete_inode = mqueue_delete_inode,
1193         .drop_inode = generic_delete_inode,
1194 };
1195
1196 static struct file_system_type mqueue_fs_type = {
1197         .name = "mqueue",
1198         .get_sb = mqueue_get_sb,
1199         .kill_sb = kill_litter_super,
1200 };
1201
1202 static int msg_max_limit_min = MIN_MSGMAX;
1203 static int msg_max_limit_max = MAX_MSGMAX;
1204
1205 static int msg_maxsize_limit_min = MIN_MSGSIZEMAX;
1206 static int msg_maxsize_limit_max = MAX_MSGSIZEMAX;
1207
1208 static ctl_table mq_sysctls[] = {
1209         {
1210                 .procname       = "queues_max",
1211                 .data           = &queues_max,
1212                 .maxlen         = sizeof(int),
1213                 .mode           = 0644,
1214                 .proc_handler   = &proc_dointvec,
1215         },
1216         {
1217                 .procname       = "msg_max",
1218                 .data           = &msg_max,
1219                 .maxlen         = sizeof(int),
1220                 .mode           = 0644,
1221                 .proc_handler   = &proc_dointvec_minmax,
1222                 .extra1         = &msg_max_limit_min,
1223                 .extra2         = &msg_max_limit_max,
1224         },
1225         {
1226                 .procname       = "msgsize_max",
1227                 .data           = &msgsize_max,
1228                 .maxlen         = sizeof(int),
1229                 .mode           = 0644,
1230                 .proc_handler   = &proc_dointvec_minmax,
1231                 .extra1         = &msg_maxsize_limit_min,
1232                 .extra2         = &msg_maxsize_limit_max,
1233         },
1234         { .ctl_name = 0 }
1235 };
1236
1237 static ctl_table mq_sysctl_dir[] = {
1238         {
1239                 .procname       = "mqueue",
1240                 .mode           = 0555,
1241                 .child          = mq_sysctls,
1242         },
1243         { .ctl_name = 0 }
1244 };
1245
1246 static ctl_table mq_sysctl_root[] = {
1247         {
1248                 .ctl_name       = CTL_FS,
1249                 .procname       = "fs",
1250                 .mode           = 0555,
1251                 .child          = mq_sysctl_dir,
1252         },
1253         { .ctl_name = 0 }
1254 };
1255
1256 static int __init init_mqueue_fs(void)
1257 {
1258         int error;
1259
1260         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1261                                 sizeof(struct mqueue_inode_info), 0,
1262                                 SLAB_HWCACHE_ALIGN, init_once);
1263         if (mqueue_inode_cachep == NULL)
1264                 return -ENOMEM;
1265
1266         /* ignore failues - they are not fatal */
1267         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1268
1269         error = register_filesystem(&mqueue_fs_type);
1270         if (error)
1271                 goto out_sysctl;
1272
1273         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1274                 error = PTR_ERR(mqueue_mnt);
1275                 goto out_filesystem;
1276         }
1277
1278         /* internal initialization - not common for vfs */
1279         queues_count = 0;
1280         spin_lock_init(&mq_lock);
1281
1282         return 0;
1283
1284 out_filesystem:
1285         unregister_filesystem(&mqueue_fs_type);
1286 out_sysctl:
1287         if (mq_sysctl_table)
1288                 unregister_sysctl_table(mq_sysctl_table);
1289         kmem_cache_destroy(mqueue_inode_cachep);
1290         return error;
1291 }
1292
1293 __initcall(init_mqueue_fs);