ipc/mqueue.c

   1 /*
   2  * POSIX message queues filesystem for Linux.
   3  *
   4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
   5  *                          Michal Wronski          (michal.wronski@gmail.com)
   6  *
   7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
   8  * Lockless receive & send, fd based notify:
   9  *                          Manfred Spraul          (manfred@colorfullife.com)
  10  *
  11  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
  12  *
  13  * This file is released under the GPL.
  14  */
  15
  16 #include <linux/capability.h>
  17 #include <linux/init.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/file.h>
  20 #include <linux/mount.h>
  21 #include <linux/namei.h>
  22 #include <linux/sysctl.h>
  23 #include <linux/poll.h>
  24 #include <linux/mqueue.h>
  25 #include <linux/msg.h>
  26 #include <linux/skbuff.h>
  27 #include <linux/netlink.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/audit.h>
  30 #include <linux/signal.h>
  31 #include <linux/mutex.h>
  32
  33 #include <net/sock.h>
  34 #include "util.h"
  35
  36 #define MQUEUE_MAGIC    0x19800202
  37 #define DIRENT_SIZE     20
  38 #define FILENT_SIZE     80
  39
  40 #define SEND            0
  41 #define RECV            1
  42
  43 #define STATE_NONE      0
  44 #define STATE_PENDING   1
  45 #define STATE_READY     2
  46
  47 /* 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         struct pid* 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 const struct inode_operations mqueue_dir_inode_operations;
  88 static const 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 struct kmem_cache *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 = NULL;
 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, struct kmem_cache * cachep, unsigned long flags)
 215 {
 216         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 217
 218         if (flags & SLAB_CTOR_CONSTRUCTOR)
 219                 inode_init_once(&p->vfs_inode);
 220 }
 221
 222 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 223 {
 224         struct mqueue_inode_info *ei;
 225
 226         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 227         if (!ei)
 228                 return NULL;
 229         return &ei->vfs_inode;
 230 }
 231
 232 static void mqueue_destroy_inode(struct inode *inode)
 233 {
 234         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 235 }
 236
 237 static void mqueue_delete_inode(struct inode *inode)
 238 {
 239         struct mqueue_inode_info *info;
 240         struct user_struct *user;
 241         unsigned long mq_bytes;
 242         int i;
 243
 244         if (S_ISDIR(inode->i_mode)) {
 245                 clear_inode(inode);
 246                 return;
 247         }
 248         info = MQUEUE_I(inode);
 249         spin_lock(&info->lock);
 250         for (i = 0; i < info->attr.mq_curmsgs; i++)
 251                 free_msg(info->messages[i]);
 252         kfree(info->messages);
 253         spin_unlock(&info->lock);
 254
 255         clear_inode(inode);
 256
 257         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 258                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 259         user = info->user;
 260         if (user) {
 261                 spin_lock(&mq_lock);
 262                 user->mq_bytes -= mq_bytes;
 263                 queues_count--;
 264                 spin_unlock(&mq_lock);
 265                 free_uid(user);
 266         }
 267 }
 268
 269 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 270                                 int mode, struct nameidata *nd)
 271 {
 272         struct inode *inode;
 273         struct mq_attr *attr = dentry->d_fsdata;
 274         int error;
 275
 276         spin_lock(&mq_lock);
 277         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 278                 error = -ENOSPC;
 279                 goto out_lock;
 280         }
 281         queues_count++;
 282         spin_unlock(&mq_lock);
 283
 284         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 285         if (!inode) {
 286                 error = -ENOMEM;
 287                 spin_lock(&mq_lock);
 288                 queues_count--;
 289                 goto out_lock;
 290         }
 291
 292         dir->i_size += DIRENT_SIZE;
 293         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 294
 295         d_instantiate(dentry, inode);
 296         dget(dentry);
 297         return 0;
 298 out_lock:
 299         spin_unlock(&mq_lock);
 300         return error;
 301 }
 302
 303 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 304 {
 305         struct inode *inode = dentry->d_inode;
 306
 307         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 308         dir->i_size -= DIRENT_SIZE;
 309         drop_nlink(inode);
 310         dput(dentry);
 311         return 0;
 312 }
 313
 314 /*
 315 *       This is routine for system read from queue file.
 316 *       To avoid mess with doing here some sort of mq_receive we allow
 317 *       to read only queue size & notification info (the only values
 318 *       that are interesting from user point of view and aren't accessible
 319 *       through std routines)
 320 */
 321 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 322                                 size_t count, loff_t * off)
 323 {
 324         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 325         char buffer[FILENT_SIZE];
 326         size_t slen;
 327         loff_t o;
 328
 329         if (!count)
 330                 return 0;
 331
 332         spin_lock(&info->lock);
 333         snprintf(buffer, sizeof(buffer),
 334                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 335                         info->qsize,
 336                         info->notify_owner ? info->notify.sigev_notify : 0,
 337                         (info->notify_owner &&
 338                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 339                                 info->notify.sigev_signo : 0,
 340                         pid_nr(info->notify_owner));
 341         spin_unlock(&info->lock);
 342         buffer[sizeof(buffer)-1] = '\0';
 343         slen = strlen(buffer)+1;
 344
 345         o = *off;
 346         if (o > slen)
 347                 return 0;
 348
 349         if (o + count > slen)
 350                 count = slen - o;
 351
 352         if (copy_to_user(u_data, buffer + o, count))
 353                 return -EFAULT;
 354
 355         *off = o + count;
 356         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 357         return count;
 358 }
 359
 360 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 361 {
 362         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 363
 364         spin_lock(&info->lock);
 365         if (task_tgid(current) == info->notify_owner)
 366                 remove_notification(info);
 367
 368         spin_unlock(&info->lock);
 369         return 0;
 370 }
 371
 372 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 373 {
 374         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 375         int retval = 0;
 376
 377         poll_wait(filp, &info->wait_q, poll_tab);
 378
 379         spin_lock(&info->lock);
 380         if (info->attr.mq_curmsgs)
 381                 retval = POLLIN | POLLRDNORM;
 382
 383         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 384                 retval |= POLLOUT | POLLWRNORM;
 385         spin_unlock(&info->lock);
 386
 387         return retval;
 388 }
 389
 390 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 391 static void wq_add(struct mqueue_inode_info *info, int sr,
 392                         struct ext_wait_queue *ewp)
 393 {
 394         struct ext_wait_queue *walk;
 395
 396         ewp->task = current;
 397
 398         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 399                 if (walk->task->static_prio <= current->static_prio) {
 400                         list_add_tail(&ewp->list, &walk->list);
 401                         return;
 402                 }
 403         }
 404         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 405 }
 406
 407 /*
 408  * Puts current task to sleep. Caller must hold queue lock. After return
 409  * lock isn't held.
 410  * sr: SEND or RECV
 411  */
 412 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 413                         long timeout, struct ext_wait_queue *ewp)
 414 {
 415         int retval;
 416         signed long time;
 417
 418         wq_add(info, sr, ewp);
 419
 420         for (;;) {
 421                 set_current_state(TASK_INTERRUPTIBLE);
 422
 423                 spin_unlock(&info->lock);
 424                 time = schedule_timeout(timeout);
 425
 426                 while (ewp->state == STATE_PENDING)
 427                         cpu_relax();
 428
 429                 if (ewp->state == STATE_READY) {
 430                         retval = 0;
 431                         goto out;
 432                 }
 433                 spin_lock(&info->lock);
 434                 if (ewp->state == STATE_READY) {
 435                         retval = 0;
 436                         goto out_unlock;
 437                 }
 438                 if (signal_pending(current)) {
 439                         retval = -ERESTARTSYS;
 440                         break;
 441                 }
 442                 if (time == 0) {
 443                         retval = -ETIMEDOUT;
 444                         break;
 445                 }
 446         }
 447         list_del(&ewp->list);
 448 out_unlock:
 449         spin_unlock(&info->lock);
 450 out:
 451         return retval;
 452 }
 453
 454 /*
 455  * Returns waiting task that should be serviced first or NULL if none exists
 456  */
 457 static struct ext_wait_queue *wq_get_first_waiter(
 458                 struct mqueue_inode_info *info, int sr)
 459 {
 460         struct list_head *ptr;
 461
 462         ptr = info->e_wait_q[sr].list.prev;
 463         if (ptr == &info->e_wait_q[sr].list)
 464                 return NULL;
 465         return list_entry(ptr, struct ext_wait_queue, list);
 466 }
 467
 468 /* Auxiliary functions to manipulate messages' list */
 469 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 470 {
 471         int k;
 472
 473         k = info->attr.mq_curmsgs - 1;
 474         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 475                 info->messages[k + 1] = info->messages[k];
 476                 k--;
 477         }
 478         info->attr.mq_curmsgs++;
 479         info->qsize += ptr->m_ts;
 480         info->messages[k + 1] = ptr;
 481 }
 482
 483 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 484 {
 485         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 486         return info->messages[info->attr.mq_curmsgs];
 487 }
 488
 489 static inline void set_cookie(struct sk_buff *skb, char code)
 490 {
 491         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 492 }
 493
 494 /*
 495  * The next function is only to split too long sys_mq_timedsend
 496  */
 497 static void __do_notify(struct mqueue_inode_info *info)
 498 {
 499         /* notification
 500          * invoked when there is registered process and there isn't process
 501          * waiting synchronously for message AND state of queue changed from
 502          * empty to not empty. Here we are sure that no one is waiting
 503          * synchronously. */
 504         if (info->notify_owner &&
 505             info->attr.mq_curmsgs == 1) {
 506                 struct siginfo sig_i;
 507                 switch (info->notify.sigev_notify) {
 508                 case SIGEV_NONE:
 509                         break;
 510                 case SIGEV_SIGNAL:
 511                         /* sends signal */
 512
 513                         sig_i.si_signo = info->notify.sigev_signo;
 514                         sig_i.si_errno = 0;
 515                         sig_i.si_code = SI_MESGQ;
 516                         sig_i.si_value = info->notify.sigev_value;
 517                         sig_i.si_pid = current->tgid;
 518                         sig_i.si_uid = current->uid;
 519
 520                         kill_pid_info(info->notify.sigev_signo,
 521                                       &sig_i, info->notify_owner);
 522                         break;
 523                 case SIGEV_THREAD:
 524                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 525                         netlink_sendskb(info->notify_sock,
 526                                         info->notify_cookie, 0);
 527                         break;
 528                 }
 529                 /* after notification unregisters process */
 530                 put_pid(info->notify_owner);
 531                 info->notify_owner = NULL;
 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 != NULL &&
 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         put_pid(info->notify_owner);
 575         info->notify_owner = NULL;
 576 }
 577
 578 static int mq_attr_ok(struct mq_attr *attr)
 579 {
 580         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 581                 return 0;
 582         if (capable(CAP_SYS_RESOURCE)) {
 583                 if (attr->mq_maxmsg > HARD_MSGMAX)
 584                         return 0;
 585         } else {
 586                 if (attr->mq_maxmsg > msg_max ||
 587                                 attr->mq_msgsize > msgsize_max)
 588                         return 0;
 589         }
 590         /* check for overflow */
 591         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 592                 return 0;
 593         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 594             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 595             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 596                 return 0;
 597         return 1;
 598 }
 599
 600 /*
 601  * Invoked when creating a new queue via sys_mq_open
 602  */
 603 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 604                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 605 {
 606         struct mq_attr attr;
 607         int ret;
 608
 609         if (u_attr) {
 610                 ret = -EFAULT;
 611                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 612                         goto out;
 613                 ret = -EINVAL;
 614                 if (!mq_attr_ok(&attr))
 615                         goto out;
 616                 /* store for use during create */
 617                 dentry->d_fsdata = &attr;
 618         }
 619
 620         mode &= ~current->fs->umask;
 621         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 622         dentry->d_fsdata = NULL;
 623         if (ret)
 624                 goto out;
 625
 626         return dentry_open(dentry, mqueue_mnt, oflag);
 627
 628 out:
 629         dput(dentry);
 630         mntput(mqueue_mnt);
 631         return ERR_PTR(ret);
 632 }
 633
 634 /* Opens existing queue */
 635 static struct file *do_open(struct dentry *dentry, int oflag)
 636 {
 637 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 638                                         MAY_READ | MAY_WRITE };
 639
 640         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 641                 dput(dentry);
 642                 mntput(mqueue_mnt);
 643                 return ERR_PTR(-EINVAL);
 644         }
 645
 646         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
 647                 dput(dentry);
 648                 mntput(mqueue_mnt);
 649                 return ERR_PTR(-EACCES);
 650         }
 651
 652         return dentry_open(dentry, mqueue_mnt, oflag);
 653 }
 654
 655 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 656                                 struct mq_attr __user *u_attr)
 657 {
 658         struct dentry *dentry;
 659         struct file *filp;
 660         char *name;
 661         int fd, error;
 662
 663         error = audit_mq_open(oflag, mode, u_attr);
 664         if (error != 0)
 665                 return error;
 666
 667         if (IS_ERR(name = getname(u_name)))
 668                 return PTR_ERR(name);
 669
 670         fd = get_unused_fd();
 671         if (fd < 0)
 672                 goto out_putname;
 673
 674         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 675         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 676         if (IS_ERR(dentry)) {
 677                 error = PTR_ERR(dentry);
 678                 goto out_err;
 679         }
 680         mntget(mqueue_mnt);
 681
 682         if (oflag & O_CREAT) {
 683                 if (dentry->d_inode) {  /* entry already exists */
 684                         error = -EEXIST;
 685                         if (oflag & O_EXCL)
 686                                 goto out;
 687                         filp = do_open(dentry, oflag);
 688                 } else {
 689                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 690                                                 oflag, mode, u_attr);
 691                 }
 692         } else {
 693                 error = -ENOENT;
 694                 if (!dentry->d_inode)
 695                         goto out;
 696                 filp = do_open(dentry, oflag);
 697         }
 698
 699         if (IS_ERR(filp)) {
 700                 error = PTR_ERR(filp);
 701                 goto out_putfd;
 702         }
 703
 704         set_close_on_exec(fd, 1);
 705         fd_install(fd, filp);
 706         goto out_upsem;
 707
 708 out:
 709         dput(dentry);
 710         mntput(mqueue_mnt);
 711 out_putfd:
 712         put_unused_fd(fd);
 713 out_err:
 714         fd = error;
 715 out_upsem:
 716         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 717 out_putname:
 718         putname(name);
 719         return fd;
 720 }
 721
 722 asmlinkage long sys_mq_unlink(const char __user *u_name)
 723 {
 724         int err;
 725         char *name;
 726         struct dentry *dentry;
 727         struct inode *inode = NULL;
 728
 729         name = getname(u_name);
 730         if (IS_ERR(name))
 731                 return PTR_ERR(name);
 732
 733         mutex_lock_nested(&mqueue_mnt->mnt_root->d_inode->i_mutex,
 734                         I_MUTEX_PARENT);
 735         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 736         if (IS_ERR(dentry)) {
 737                 err = PTR_ERR(dentry);
 738                 goto out_unlock;
 739         }
 740
 741         if (!dentry->d_inode) {
 742                 err = -ENOENT;
 743                 goto out_err;
 744         }
 745
 746         inode = dentry->d_inode;
 747         if (inode)
 748                 atomic_inc(&inode->i_count);
 749
 750         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 751 out_err:
 752         dput(dentry);
 753
 754 out_unlock:
 755         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 756         putname(name);
 757         if (inode)
 758                 iput(inode);
 759
 760         return err;
 761 }
 762
 763 /* Pipelined send and receive functions.
 764  *
 765  * If a receiver finds no waiting message, then it registers itself in the
 766  * list of waiting receivers. A sender checks that list before adding the new
 767  * message into the message array. If there is a waiting receiver, then it
 768  * bypasses the message array and directly hands the message over to the
 769  * receiver.
 770  * The receiver accepts the message and returns without grabbing the queue
 771  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 772  * are necessary. The same algorithm is used for sysv semaphores, see
 773  * ipc/sem.c for more details.
 774  *
 775  * The same algorithm is used for senders.
 776  */
 777
 778 /* pipelined_send() - send a message directly to the task waiting in
 779  * sys_mq_timedreceive() (without inserting message into a queue).
 780  */
 781 static inline void pipelined_send(struct mqueue_inode_info *info,
 782                                   struct msg_msg *message,
 783                                   struct ext_wait_queue *receiver)
 784 {
 785         receiver->msg = message;
 786         list_del(&receiver->list);
 787         receiver->state = STATE_PENDING;
 788         wake_up_process(receiver->task);
 789         smp_wmb();
 790         receiver->state = STATE_READY;
 791 }
 792
 793 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 794  * gets its message and put to the queue (we have one free place for sure). */
 795 static inline void pipelined_receive(struct mqueue_inode_info *info)
 796 {
 797         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 798
 799         if (!sender) {
 800                 /* for poll */
 801                 wake_up_interruptible(&info->wait_q);
 802                 return;
 803         }
 804         msg_insert(sender->msg, info);
 805         list_del(&sender->list);
 806         sender->state = STATE_PENDING;
 807         wake_up_process(sender->task);
 808         smp_wmb();
 809         sender->state = STATE_READY;
 810 }
 811
 812 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 813         size_t msg_len, unsigned int msg_prio,
 814         const struct timespec __user *u_abs_timeout)
 815 {
 816         struct file *filp;
 817         struct inode *inode;
 818         struct ext_wait_queue wait;
 819         struct ext_wait_queue *receiver;
 820         struct msg_msg *msg_ptr;
 821         struct mqueue_inode_info *info;
 822         long timeout;
 823         int ret;
 824
 825         ret = audit_mq_timedsend(mqdes, msg_len, msg_prio, u_abs_timeout);
 826         if (ret != 0)
 827                 return ret;
 828
 829         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 830                 return -EINVAL;
 831
 832         timeout = prepare_timeout(u_abs_timeout);
 833
 834         ret = -EBADF;
 835         filp = fget(mqdes);
 836         if (unlikely(!filp))
 837                 goto out;
 838
 839         inode = filp->f_path.dentry->d_inode;
 840         if (unlikely(filp->f_op != &mqueue_file_operations))
 841                 goto out_fput;
 842         info = MQUEUE_I(inode);
 843
 844         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 845                 goto out_fput;
 846
 847         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 848                 ret = -EMSGSIZE;
 849                 goto out_fput;
 850         }
 851
 852         /* First try to allocate memory, before doing anything with
 853          * existing queues. */
 854         msg_ptr = load_msg(u_msg_ptr, msg_len);
 855         if (IS_ERR(msg_ptr)) {
 856                 ret = PTR_ERR(msg_ptr);
 857                 goto out_fput;
 858         }
 859         msg_ptr->m_ts = msg_len;
 860         msg_ptr->m_type = msg_prio;
 861
 862         spin_lock(&info->lock);
 863
 864         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 865                 if (filp->f_flags & O_NONBLOCK) {
 866                         spin_unlock(&info->lock);
 867                         ret = -EAGAIN;
 868                 } else if (unlikely(timeout < 0)) {
 869                         spin_unlock(&info->lock);
 870                         ret = timeout;
 871                 } else {
 872                         wait.task = current;
 873                         wait.msg = (void *) msg_ptr;
 874                         wait.state = STATE_NONE;
 875                         ret = wq_sleep(info, SEND, timeout, &wait);
 876                 }
 877                 if (ret < 0)
 878                         free_msg(msg_ptr);
 879         } else {
 880                 receiver = wq_get_first_waiter(info, RECV);
 881                 if (receiver) {
 882                         pipelined_send(info, msg_ptr, receiver);
 883                 } else {
 884                         /* adds message to the queue */
 885                         msg_insert(msg_ptr, info);
 886                         __do_notify(info);
 887                 }
 888                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 889                                 CURRENT_TIME;
 890                 spin_unlock(&info->lock);
 891                 ret = 0;
 892         }
 893 out_fput:
 894         fput(filp);
 895 out:
 896         return ret;
 897 }
 898
 899 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 900         size_t msg_len, unsigned int __user *u_msg_prio,
 901         const struct timespec __user *u_abs_timeout)
 902 {
 903         long timeout;
 904         ssize_t ret;
 905         struct msg_msg *msg_ptr;
 906         struct file *filp;
 907         struct inode *inode;
 908         struct mqueue_inode_info *info;
 909         struct ext_wait_queue wait;
 910
 911         ret = audit_mq_timedreceive(mqdes, msg_len, u_msg_prio, u_abs_timeout);
 912         if (ret != 0)
 913                 return ret;
 914
 915         timeout = prepare_timeout(u_abs_timeout);
 916
 917         ret = -EBADF;
 918         filp = fget(mqdes);
 919         if (unlikely(!filp))
 920                 goto out;
 921
 922         inode = filp->f_path.dentry->d_inode;
 923         if (unlikely(filp->f_op != &mqueue_file_operations))
 924                 goto out_fput;
 925         info = MQUEUE_I(inode);
 926
 927         if (unlikely(!(filp->f_mode & FMODE_READ)))
 928                 goto out_fput;
 929
 930         /* checks if buffer is big enough */
 931         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 932                 ret = -EMSGSIZE;
 933                 goto out_fput;
 934         }
 935
 936         spin_lock(&info->lock);
 937         if (info->attr.mq_curmsgs == 0) {
 938                 if (filp->f_flags & O_NONBLOCK) {
 939                         spin_unlock(&info->lock);
 940                         ret = -EAGAIN;
 941                         msg_ptr = NULL;
 942                 } else if (unlikely(timeout < 0)) {
 943                         spin_unlock(&info->lock);
 944                         ret = timeout;
 945                         msg_ptr = NULL;
 946                 } else {
 947                         wait.task = current;
 948                         wait.state = STATE_NONE;
 949                         ret = wq_sleep(info, RECV, timeout, &wait);
 950                         msg_ptr = wait.msg;
 951                 }
 952         } else {
 953                 msg_ptr = msg_get(info);
 954
 955                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 956                                 CURRENT_TIME;
 957
 958                 /* There is now free space in queue. */
 959                 pipelined_receive(info);
 960                 spin_unlock(&info->lock);
 961                 ret = 0;
 962         }
 963         if (ret == 0) {
 964                 ret = msg_ptr->m_ts;
 965
 966                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 967                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 968                         ret = -EFAULT;
 969                 }
 970                 free_msg(msg_ptr);
 971         }
 972 out_fput:
 973         fput(filp);
 974 out:
 975         return ret;
 976 }
 977
 978 /*
 979  * Notes: the case when user wants us to deregister (with NULL as pointer)
 980  * and he isn't currently owner of notification, will be silently discarded.
 981  * It isn't explicitly defined in the POSIX.
 982  */
 983 asmlinkage long sys_mq_notify(mqd_t mqdes,
 984                                 const struct sigevent __user *u_notification)
 985 {
 986         int ret;
 987         struct file *filp;
 988         struct sock *sock;
 989         struct inode *inode;
 990         struct sigevent notification;
 991         struct mqueue_inode_info *info;
 992         struct sk_buff *nc;
 993
 994         ret = audit_mq_notify(mqdes, u_notification);
 995         if (ret != 0)
 996                 return ret;
 997
 998         nc = NULL;
 999         sock = NULL;
1000         if (u_notification != NULL) {
1001                 if (copy_from_user(&notification, u_notification,
1002                                         sizeof(struct sigevent)))
1003                         return -EFAULT;
1004
1005                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1006                              notification.sigev_notify != SIGEV_SIGNAL &&
1007                              notification.sigev_notify != SIGEV_THREAD))
1008                         return -EINVAL;
1009                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1010                         !valid_signal(notification.sigev_signo)) {
1011                         return -EINVAL;
1012                 }
1013                 if (notification.sigev_notify == SIGEV_THREAD) {
1014                         /* create the notify skb */
1015                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1016                         ret = -ENOMEM;
1017                         if (!nc)
1018                                 goto out;
1019                         ret = -EFAULT;
1020                         if (copy_from_user(nc->data,
1021                                         notification.sigev_value.sival_ptr,
1022                                         NOTIFY_COOKIE_LEN)) {
1023                                 goto out;
1024                         }
1025
1026                         /* TODO: add a header? */
1027                         skb_put(nc, NOTIFY_COOKIE_LEN);
1028                         /* and attach it to the socket */
1029 retry:
1030                         filp = fget(notification.sigev_signo);
1031                         ret = -EBADF;
1032                         if (!filp)
1033                                 goto out;
1034                         sock = netlink_getsockbyfilp(filp);
1035                         fput(filp);
1036                         if (IS_ERR(sock)) {
1037                                 ret = PTR_ERR(sock);
1038                                 sock = NULL;
1039                                 goto out;
1040                         }
1041
1042                         ret = netlink_attachskb(sock, nc, 0,
1043                                         MAX_SCHEDULE_TIMEOUT, NULL);
1044                         if (ret == 1)
1045                                 goto retry;
1046                         if (ret) {
1047                                 sock = NULL;
1048                                 nc = NULL;
1049                                 goto out;
1050                         }
1051                 }
1052         }
1053
1054         ret = -EBADF;
1055         filp = fget(mqdes);
1056         if (!filp)
1057                 goto out;
1058
1059         inode = filp->f_path.dentry->d_inode;
1060         if (unlikely(filp->f_op != &mqueue_file_operations))
1061                 goto out_fput;
1062         info = MQUEUE_I(inode);
1063
1064         ret = 0;
1065         spin_lock(&info->lock);
1066         if (u_notification == NULL) {
1067                 if (info->notify_owner == task_tgid(current)) {
1068                         remove_notification(info);
1069                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1070                 }
1071         } else if (info->notify_owner != NULL) {
1072                 ret = -EBUSY;
1073         } else {
1074                 switch (notification.sigev_notify) {
1075                 case SIGEV_NONE:
1076                         info->notify.sigev_notify = SIGEV_NONE;
1077                         break;
1078                 case SIGEV_THREAD:
1079                         info->notify_sock = sock;
1080                         info->notify_cookie = nc;
1081                         sock = NULL;
1082                         nc = NULL;
1083                         info->notify.sigev_notify = SIGEV_THREAD;
1084                         break;
1085                 case SIGEV_SIGNAL:
1086                         info->notify.sigev_signo = notification.sigev_signo;
1087                         info->notify.sigev_value = notification.sigev_value;
1088                         info->notify.sigev_notify = SIGEV_SIGNAL;
1089                         break;
1090                 }
1091
1092                 info->notify_owner = get_pid(task_tgid(current));
1093                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1094         }
1095         spin_unlock(&info->lock);
1096 out_fput:
1097         fput(filp);
1098 out:
1099         if (sock) {
1100                 netlink_detachskb(sock, nc);
1101         } else if (nc) {
1102                 dev_kfree_skb(nc);
1103         }
1104         return ret;
1105 }
1106
1107 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1108                         const struct mq_attr __user *u_mqstat,
1109                         struct mq_attr __user *u_omqstat)
1110 {
1111         int ret;
1112         struct mq_attr mqstat, omqstat;
1113         struct file *filp;
1114         struct inode *inode;
1115         struct mqueue_inode_info *info;
1116
1117         if (u_mqstat != NULL) {
1118                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1119                         return -EFAULT;
1120                 if (mqstat.mq_flags & (~O_NONBLOCK))
1121                         return -EINVAL;
1122         }
1123
1124         ret = -EBADF;
1125         filp = fget(mqdes);
1126         if (!filp)
1127                 goto out;
1128
1129         inode = filp->f_path.dentry->d_inode;
1130         if (unlikely(filp->f_op != &mqueue_file_operations))
1131                 goto out_fput;
1132         info = MQUEUE_I(inode);
1133
1134         spin_lock(&info->lock);
1135
1136         omqstat = info->attr;
1137         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1138         if (u_mqstat) {
1139                 ret = audit_mq_getsetattr(mqdes, &mqstat);
1140                 if (ret != 0)
1141                         goto out;
1142                 if (mqstat.mq_flags & O_NONBLOCK)
1143                         filp->f_flags |= O_NONBLOCK;
1144                 else
1145                         filp->f_flags &= ~O_NONBLOCK;
1146
1147                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1148         }
1149
1150         spin_unlock(&info->lock);
1151
1152         ret = 0;
1153         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1154                                                 sizeof(struct mq_attr)))
1155                 ret = -EFAULT;
1156
1157 out_fput:
1158         fput(filp);
1159 out:
1160         return ret;
1161 }
1162
1163 static const struct inode_operations mqueue_dir_inode_operations = {
1164         .lookup = simple_lookup,
1165         .create = mqueue_create,
1166         .unlink = mqueue_unlink,
1167 };
1168
1169 static const struct file_operations mqueue_file_operations = {
1170         .flush = mqueue_flush_file,
1171         .poll = mqueue_poll_file,
1172         .read = mqueue_read_file,
1173 };
1174
1175 static struct super_operations mqueue_super_ops = {
1176         .alloc_inode = mqueue_alloc_inode,
1177         .destroy_inode = mqueue_destroy_inode,
1178         .statfs = simple_statfs,
1179         .delete_inode = mqueue_delete_inode,
1180         .drop_inode = generic_delete_inode,
1181 };
1182
1183 static struct file_system_type mqueue_fs_type = {
1184         .name = "mqueue",
1185         .get_sb = mqueue_get_sb,
1186         .kill_sb = kill_litter_super,
1187 };
1188
1189 static int msg_max_limit_min = DFLT_MSGMAX;
1190 static int msg_max_limit_max = HARD_MSGMAX;
1191
1192 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1193 static int msg_maxsize_limit_max = INT_MAX;
1194
1195 static ctl_table mq_sysctls[] = {
1196         {
1197                 .ctl_name       = CTL_QUEUESMAX,
1198                 .procname       = "queues_max",
1199                 .data           = &queues_max,
1200                 .maxlen         = sizeof(int),
1201                 .mode           = 0644,
1202                 .proc_handler   = &proc_dointvec,
1203         },
1204         {
1205                 .ctl_name       = CTL_MSGMAX,
1206                 .procname       = "msg_max",
1207                 .data           = &msg_max,
1208                 .maxlen         = sizeof(int),
1209                 .mode           = 0644,
1210                 .proc_handler   = &proc_dointvec_minmax,
1211                 .extra1         = &msg_max_limit_min,
1212                 .extra2         = &msg_max_limit_max,
1213         },
1214         {
1215                 .ctl_name       = CTL_MSGSIZEMAX,
1216                 .procname       = "msgsize_max",
1217                 .data           = &msgsize_max,
1218                 .maxlen         = sizeof(int),
1219                 .mode           = 0644,
1220                 .proc_handler   = &proc_dointvec_minmax,
1221                 .extra1         = &msg_maxsize_limit_min,
1222                 .extra2         = &msg_maxsize_limit_max,
1223         },
1224         { .ctl_name = 0 }
1225 };
1226
1227 static ctl_table mq_sysctl_dir[] = {
1228         {
1229                 .ctl_name       = FS_MQUEUE,
1230                 .procname       = "mqueue",
1231                 .mode           = 0555,
1232                 .child          = mq_sysctls,
1233         },
1234         { .ctl_name = 0 }
1235 };
1236
1237 static ctl_table mq_sysctl_root[] = {
1238         {
1239                 .ctl_name       = CTL_FS,
1240                 .procname       = "fs",
1241                 .mode           = 0555,
1242                 .child          = mq_sysctl_dir,
1243         },
1244         { .ctl_name = 0 }
1245 };
1246
1247 static int __init init_mqueue_fs(void)
1248 {
1249         int error;
1250
1251         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1252                                 sizeof(struct mqueue_inode_info), 0,
1253                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1254         if (mqueue_inode_cachep == NULL)
1255                 return -ENOMEM;
1256
1257         /* ignore failues - they are not fatal */
1258         mq_sysctl_table = register_sysctl_table(mq_sysctl_root);
1259
1260         error = register_filesystem(&mqueue_fs_type);
1261         if (error)
1262                 goto out_sysctl;
1263
1264         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1265                 error = PTR_ERR(mqueue_mnt);
1266                 goto out_filesystem;
1267         }
1268
1269         /* internal initialization - not common for vfs */
1270         queues_count = 0;
1271         spin_lock_init(&mq_lock);
1272
1273         return 0;
1274
1275 out_filesystem:
1276         unregister_filesystem(&mqueue_fs_type);
1277 out_sysctl:
1278         if (mq_sysctl_table)
1279                 unregister_sysctl_table(mq_sysctl_table);
1280         kmem_cache_destroy(mqueue_inode_cachep);
1281         return error;
1282 }
1283
1284 __initcall(init_mqueue_fs);