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