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  * This file is released under the GPL.
  12  */
  13
  14 #include <linux/capability.h>
  15 #include <linux/init.h>
  16 #include <linux/pagemap.h>
  17 #include <linux/file.h>
  18 #include <linux/mount.h>
  19 #include <linux/namei.h>
  20 #include <linux/sysctl.h>
  21 #include <linux/poll.h>
  22 #include <linux/mqueue.h>
  23 #include <linux/msg.h>
  24 #include <linux/skbuff.h>
  25 #include <linux/netlink.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/signal.h>
  28 #include <linux/mutex.h>
  29 #include <linux/vs_base.h>
  30 #include <linux/vs_context.h>
  31 #include <linux/vs_limit.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_blksize = PAGE_CACHE_SIZE;
 119                 inode->i_blocks = 0;
 120                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 121                                 CURRENT_TIME;
 122
 123                 if (S_ISREG(mode)) {
 124                         struct mqueue_inode_info *info;
 125                         struct task_struct *p = current;
 126                         struct user_struct *u = p->user;
 127                         unsigned long mq_bytes, mq_msg_tblsz;
 128
 129                         inode->i_fop = &mqueue_file_operations;
 130                         inode->i_size = FILENT_SIZE;
 131                         /* mqueue specific info */
 132                         info = MQUEUE_I(inode);
 133                         spin_lock_init(&info->lock);
 134                         init_waitqueue_head(&info->wait_q);
 135                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 136                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 137                         info->messages = NULL;
 138                         info->notify_owner = 0;
 139                         info->qsize = 0;
 140                         info->user = NULL;      /* set when all is ok */
 141                         memset(&info->attr, 0, sizeof(info->attr));
 142                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 143                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 144                         if (attr) {
 145                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 146                                 info->attr.mq_msgsize = attr->mq_msgsize;
 147                         }
 148                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 149                         mq_bytes = (mq_msg_tblsz +
 150                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 151
 152                         spin_lock(&mq_lock);
 153                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 154                             u->mq_bytes + mq_bytes >
 155                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur ||
 156                             !vx_ipcmsg_avail(p->vx_info, mq_bytes)) {
 157                                 spin_unlock(&mq_lock);
 158                                 goto out_inode;
 159                         }
 160                         u->mq_bytes += mq_bytes;
 161                         vx_ipcmsg_add(p->vx_info, u, mq_bytes);
 162                         spin_unlock(&mq_lock);
 163
 164                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 165                         if (!info->messages) {
 166                                 spin_lock(&mq_lock);
 167                                 u->mq_bytes -= mq_bytes;
 168                                 vx_ipcmsg_sub(p->vx_info, u, mq_bytes);
 169                                 spin_unlock(&mq_lock);
 170                                 goto out_inode;
 171                         }
 172                         /* all is ok */
 173                         info->user = get_uid(u);
 174                 } else if (S_ISDIR(mode)) {
 175                         inode->i_nlink++;
 176                         /* Some things misbehave if size == 0 on a directory */
 177                         inode->i_size = 2 * DIRENT_SIZE;
 178                         inode->i_op = &mqueue_dir_inode_operations;
 179                         inode->i_fop = &simple_dir_operations;
 180                 }
 181         }
 182         return inode;
 183 out_inode:
 184         make_bad_inode(inode);
 185         iput(inode);
 186         return NULL;
 187 }
 188
 189 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 190 {
 191         struct inode *inode;
 192
 193         sb->s_blocksize = PAGE_CACHE_SIZE;
 194         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 195         sb->s_magic = MQUEUE_MAGIC;
 196         sb->s_op = &mqueue_super_ops;
 197
 198         inode = mqueue_get_inode(sb, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 199         if (!inode)
 200                 return -ENOMEM;
 201
 202         sb->s_root = d_alloc_root(inode);
 203         if (!sb->s_root) {
 204                 iput(inode);
 205                 return -ENOMEM;
 206         }
 207
 208         return 0;
 209 }
 210
 211 static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
 212                                          int flags, const char *dev_name,
 213                                          void *data)
 214 {
 215         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
 216 }
 217
 218 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 219 {
 220         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 221
 222         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 223                 SLAB_CTOR_CONSTRUCTOR)
 224                 inode_init_once(&p->vfs_inode);
 225 }
 226
 227 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 228 {
 229         struct mqueue_inode_info *ei;
 230
 231         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
 232         if (!ei)
 233                 return NULL;
 234         return &ei->vfs_inode;
 235 }
 236
 237 static void mqueue_destroy_inode(struct inode *inode)
 238 {
 239         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 240 }
 241
 242 static void mqueue_delete_inode(struct inode *inode)
 243 {
 244         struct mqueue_inode_info *info;
 245         struct user_struct *user;
 246         unsigned long mq_bytes;
 247         int i;
 248
 249         if (S_ISDIR(inode->i_mode)) {
 250                 clear_inode(inode);
 251                 return;
 252         }
 253         info = MQUEUE_I(inode);
 254         spin_lock(&info->lock);
 255         for (i = 0; i < info->attr.mq_curmsgs; i++)
 256                 free_msg(info->messages[i]);
 257         kfree(info->messages);
 258         spin_unlock(&info->lock);
 259
 260         clear_inode(inode);
 261
 262         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 263                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 264         user = info->user;
 265         if (user) {
 266                 struct vx_info *vxi = lookup_vx_info(user->xid);
 267
 268                 spin_lock(&mq_lock);
 269                 user->mq_bytes -= mq_bytes;
 270                 vx_ipcmsg_sub(vxi, user, mq_bytes);
 271                 queues_count--;
 272                 spin_unlock(&mq_lock);
 273                 put_vx_info(vxi);
 274                 free_uid(user);
 275         }
 276 }
 277
 278 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 279                                 int mode, struct nameidata *nd)
 280 {
 281         struct inode *inode;
 282         struct mq_attr *attr = dentry->d_fsdata;
 283         int error;
 284
 285         spin_lock(&mq_lock);
 286         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 287                 error = -ENOSPC;
 288                 goto out_lock;
 289         }
 290         queues_count++;
 291         spin_unlock(&mq_lock);
 292
 293         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 294         if (!inode) {
 295                 error = -ENOMEM;
 296                 spin_lock(&mq_lock);
 297                 queues_count--;
 298                 goto out_lock;
 299         }
 300
 301         dir->i_size += DIRENT_SIZE;
 302         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 303
 304         d_instantiate(dentry, inode);
 305         dget(dentry);
 306         return 0;
 307 out_lock:
 308         spin_unlock(&mq_lock);
 309         return error;
 310 }
 311
 312 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 313 {
 314         struct inode *inode = dentry->d_inode;
 315
 316         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 317         dir->i_size -= DIRENT_SIZE;
 318         inode->i_nlink--;
 319         dput(dentry);
 320         return 0;
 321 }
 322
 323 /*
 324 *       This is routine for system read from queue file.
 325 *       To avoid mess with doing here some sort of mq_receive we allow
 326 *       to read only queue size & notification info (the only values
 327 *       that are interesting from user point of view and aren't accessible
 328 *       through std routines)
 329 */
 330 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 331                                 size_t count, loff_t * off)
 332 {
 333         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 334         char buffer[FILENT_SIZE];
 335         size_t slen;
 336         loff_t o;
 337
 338         if (!count)
 339                 return 0;
 340
 341         spin_lock(&info->lock);
 342         snprintf(buffer, sizeof(buffer),
 343                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 344                         info->qsize,
 345                         info->notify_owner ? info->notify.sigev_notify : 0,
 346                         (info->notify_owner &&
 347                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 348                                 info->notify.sigev_signo : 0,
 349                         info->notify_owner);
 350         spin_unlock(&info->lock);
 351         buffer[sizeof(buffer)-1] = '\0';
 352         slen = strlen(buffer)+1;
 353
 354         o = *off;
 355         if (o > slen)
 356                 return 0;
 357
 358         if (o + count > slen)
 359                 count = slen - o;
 360
 361         if (copy_to_user(u_data, buffer + o, count))
 362                 return -EFAULT;
 363
 364         *off = o + count;
 365         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
 366         return count;
 367 }
 368
 369 static int mqueue_flush_file(struct file *filp)
 370 {
 371         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 372
 373         spin_lock(&info->lock);
 374         if (current->tgid == info->notify_owner)
 375                 remove_notification(info);
 376
 377         spin_unlock(&info->lock);
 378         return 0;
 379 }
 380
 381 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 382 {
 383         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 384         int retval = 0;
 385
 386         poll_wait(filp, &info->wait_q, poll_tab);
 387
 388         spin_lock(&info->lock);
 389         if (info->attr.mq_curmsgs)
 390                 retval = POLLIN | POLLRDNORM;
 391
 392         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 393                 retval |= POLLOUT | POLLWRNORM;
 394         spin_unlock(&info->lock);
 395
 396         return retval;
 397 }
 398
 399 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 400 static void wq_add(struct mqueue_inode_info *info, int sr,
 401                         struct ext_wait_queue *ewp)
 402 {
 403         struct ext_wait_queue *walk;
 404
 405         ewp->task = current;
 406
 407         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 408                 if (walk->task->static_prio <= current->static_prio) {
 409                         list_add_tail(&ewp->list, &walk->list);
 410                         return;
 411                 }
 412         }
 413         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 414 }
 415
 416 /*
 417  * Puts current task to sleep. Caller must hold queue lock. After return
 418  * lock isn't held.
 419  * sr: SEND or RECV
 420  */
 421 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 422                         long timeout, struct ext_wait_queue *ewp)
 423 {
 424         int retval;
 425         signed long time;
 426
 427         wq_add(info, sr, ewp);
 428
 429         for (;;) {
 430                 set_current_state(TASK_INTERRUPTIBLE);
 431
 432                 spin_unlock(&info->lock);
 433                 time = schedule_timeout(timeout);
 434
 435                 while (ewp->state == STATE_PENDING)
 436                         cpu_relax();
 437
 438                 if (ewp->state == STATE_READY) {
 439                         retval = 0;
 440                         goto out;
 441                 }
 442                 spin_lock(&info->lock);
 443                 if (ewp->state == STATE_READY) {
 444                         retval = 0;
 445                         goto out_unlock;
 446                 }
 447                 if (signal_pending(current)) {
 448                         retval = -ERESTARTSYS;
 449                         break;
 450                 }
 451                 if (time == 0) {
 452                         retval = -ETIMEDOUT;
 453                         break;
 454                 }
 455         }
 456         list_del(&ewp->list);
 457 out_unlock:
 458         spin_unlock(&info->lock);
 459 out:
 460         return retval;
 461 }
 462
 463 /*
 464  * Returns waiting task that should be serviced first or NULL if none exists
 465  */
 466 static struct ext_wait_queue *wq_get_first_waiter(
 467                 struct mqueue_inode_info *info, int sr)
 468 {
 469         struct list_head *ptr;
 470
 471         ptr = info->e_wait_q[sr].list.prev;
 472         if (ptr == &info->e_wait_q[sr].list)
 473                 return NULL;
 474         return list_entry(ptr, struct ext_wait_queue, list);
 475 }
 476
 477 /* Auxiliary functions to manipulate messages' list */
 478 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 479 {
 480         int k;
 481
 482         k = info->attr.mq_curmsgs - 1;
 483         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 484                 info->messages[k + 1] = info->messages[k];
 485                 k--;
 486         }
 487         info->attr.mq_curmsgs++;
 488         info->qsize += ptr->m_ts;
 489         info->messages[k + 1] = ptr;
 490 }
 491
 492 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 493 {
 494         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 495         return info->messages[info->attr.mq_curmsgs];
 496 }
 497
 498 static inline void set_cookie(struct sk_buff *skb, char code)
 499 {
 500         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 501 }
 502
 503 /*
 504  * The next function is only to split too long sys_mq_timedsend
 505  */
 506 static void __do_notify(struct mqueue_inode_info *info)
 507 {
 508         /* notification
 509          * invoked when there is registered process and there isn't process
 510          * waiting synchronously for message AND state of queue changed from
 511          * empty to not empty. Here we are sure that no one is waiting
 512          * synchronously. */
 513         if (info->notify_owner &&
 514             info->attr.mq_curmsgs == 1) {
 515                 struct siginfo sig_i;
 516                 switch (info->notify.sigev_notify) {
 517                 case SIGEV_NONE:
 518                         break;
 519                 case SIGEV_SIGNAL:
 520                         /* sends signal */
 521
 522                         sig_i.si_signo = info->notify.sigev_signo;
 523                         sig_i.si_errno = 0;
 524                         sig_i.si_code = SI_MESGQ;
 525                         sig_i.si_value = info->notify.sigev_value;
 526                         sig_i.si_pid = current->tgid;
 527                         sig_i.si_uid = current->uid;
 528
 529                         kill_proc_info(info->notify.sigev_signo,
 530                                        &sig_i, info->notify_owner);
 531                         break;
 532                 case SIGEV_THREAD:
 533                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 534                         netlink_sendskb(info->notify_sock,
 535                                         info->notify_cookie, 0);
 536                         break;
 537                 }
 538                 /* after notification unregisters process */
 539                 info->notify_owner = 0;
 540         }
 541         wake_up(&info->wait_q);
 542 }
 543
 544 static long prepare_timeout(const struct timespec __user *u_arg)
 545 {
 546         struct timespec ts, nowts;
 547         long timeout;
 548
 549         if (u_arg) {
 550                 if (unlikely(copy_from_user(&ts, u_arg,
 551                                         sizeof(struct timespec))))
 552                         return -EFAULT;
 553
 554                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 555                         || ts.tv_nsec >= NSEC_PER_SEC))
 556                         return -EINVAL;
 557                 nowts = CURRENT_TIME;
 558                 /* first subtract as jiffies can't be too big */
 559                 ts.tv_sec -= nowts.tv_sec;
 560                 if (ts.tv_nsec < nowts.tv_nsec) {
 561                         ts.tv_nsec += NSEC_PER_SEC;
 562                         ts.tv_sec--;
 563                 }
 564                 ts.tv_nsec -= nowts.tv_nsec;
 565                 if (ts.tv_sec < 0)
 566                         return 0;
 567
 568                 timeout = timespec_to_jiffies(&ts) + 1;
 569         } else
 570                 return MAX_SCHEDULE_TIMEOUT;
 571
 572         return timeout;
 573 }
 574
 575 static void remove_notification(struct mqueue_inode_info *info)
 576 {
 577         if (info->notify_owner != 0 &&
 578             info->notify.sigev_notify == SIGEV_THREAD) {
 579                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 580                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
 581         }
 582         info->notify_owner = 0;
 583 }
 584
 585 static int mq_attr_ok(struct mq_attr *attr)
 586 {
 587         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 588                 return 0;
 589         if (capable(CAP_SYS_RESOURCE)) {
 590                 if (attr->mq_maxmsg > HARD_MSGMAX)
 591                         return 0;
 592         } else {
 593                 if (attr->mq_maxmsg > msg_max ||
 594                                 attr->mq_msgsize > msgsize_max)
 595                         return 0;
 596         }
 597         /* check for overflow */
 598         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 599                 return 0;
 600         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 601             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 602             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 603                 return 0;
 604         return 1;
 605 }
 606
 607 /*
 608  * Invoked when creating a new queue via sys_mq_open
 609  */
 610 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 611                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 612 {
 613         struct mq_attr attr;
 614         int ret;
 615
 616         if (u_attr) {
 617                 ret = -EFAULT;
 618                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 619                         goto out;
 620                 ret = -EINVAL;
 621                 if (!mq_attr_ok(&attr))
 622                         goto out;
 623                 /* store for use during create */
 624                 dentry->d_fsdata = &attr;
 625         }
 626
 627         mode &= ~current->fs->umask;
 628         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 629         dentry->d_fsdata = NULL;
 630         if (ret)
 631                 goto out;
 632
 633         return dentry_open(dentry, mqueue_mnt, oflag);
 634
 635 out:
 636         dput(dentry);
 637         mntput(mqueue_mnt);
 638         return ERR_PTR(ret);
 639 }
 640
 641 /* Opens existing queue */
 642 static struct file *do_open(struct dentry *dentry, int oflag)
 643 {
 644 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 645                                         MAY_READ | MAY_WRITE };
 646
 647         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 648                 dput(dentry);
 649                 mntput(mqueue_mnt);
 650                 return ERR_PTR(-EINVAL);
 651         }
 652
 653         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
 654                 dput(dentry);
 655                 mntput(mqueue_mnt);
 656                 return ERR_PTR(-EACCES);
 657         }
 658
 659         return dentry_open(dentry, mqueue_mnt, oflag);
 660 }
 661
 662 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 663                                 struct mq_attr __user *u_attr)
 664 {
 665         struct dentry *dentry;
 666         struct file *filp;
 667         char *name;
 668         int fd, error;
 669
 670         if (IS_ERR(name = getname(u_name)))
 671                 return PTR_ERR(name);
 672
 673         fd = get_unused_fd();
 674         if (fd < 0)
 675                 goto out_putname;
 676
 677         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 678         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 679         if (IS_ERR(dentry)) {
 680                 error = PTR_ERR(dentry);
 681                 goto out_err;
 682         }
 683         mntget(mqueue_mnt);
 684
 685         if (oflag & O_CREAT) {
 686                 if (dentry->d_inode) {  /* entry already exists */
 687                         error = -EEXIST;
 688                         if (oflag & O_EXCL)
 689                                 goto out;
 690                         filp = do_open(dentry, oflag);
 691                 } else {
 692                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 693                                                 oflag, mode, u_attr);
 694                 }
 695         } else {
 696                 error = -ENOENT;
 697                 if (!dentry->d_inode)
 698                         goto out;
 699                 filp = do_open(dentry, oflag);
 700         }
 701
 702         if (IS_ERR(filp)) {
 703                 error = PTR_ERR(filp);
 704                 goto out_putfd;
 705         }
 706
 707         set_close_on_exec(fd, 1);
 708         fd_install(fd, filp);
 709         goto out_upsem;
 710
 711 out:
 712         dput(dentry);
 713         mntput(mqueue_mnt);
 714 out_putfd:
 715         put_unused_fd(fd);
 716 out_err:
 717         fd = error;
 718 out_upsem:
 719         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 720 out_putname:
 721         putname(name);
 722         return fd;
 723 }
 724
 725 asmlinkage long sys_mq_unlink(const char __user *u_name)
 726 {
 727         int err;
 728         char *name;
 729         struct dentry *dentry;
 730         struct inode *inode = NULL;
 731
 732         name = getname(u_name);
 733         if (IS_ERR(name))
 734                 return PTR_ERR(name);
 735
 736         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 737         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 738         if (IS_ERR(dentry)) {
 739                 err = PTR_ERR(dentry);
 740                 goto out_unlock;
 741         }
 742
 743         if (!dentry->d_inode) {
 744                 err = -ENOENT;
 745                 goto out_err;
 746         }
 747
 748         inode = dentry->d_inode;
 749         if (inode)
 750                 atomic_inc(&inode->i_count);
 751
 752         err = vfs_unlink(dentry->d_parent->d_inode, dentry, NULL);
 753 out_err:
 754         dput(dentry);
 755
 756 out_unlock:
 757         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 758         putname(name);
 759         if (inode)
 760                 iput(inode);
 761
 762         return err;
 763 }
 764
 765 /* Pipelined send and receive functions.
 766  *
 767  * If a receiver finds no waiting message, then it registers itself in the
 768  * list of waiting receivers. A sender checks that list before adding the new
 769  * message into the message array. If there is a waiting receiver, then it
 770  * bypasses the message array and directly hands the message over to the
 771  * receiver.
 772  * The receiver accepts the message and returns without grabbing the queue
 773  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 774  * are necessary. The same algorithm is used for sysv semaphores, see
 775  * ipc/sem.c for more details.
 776  *
 777  * The same algorithm is used for senders.
 778  */
 779
 780 /* pipelined_send() - send a message directly to the task waiting in
 781  * sys_mq_timedreceive() (without inserting message into a queue).
 782  */
 783 static inline void pipelined_send(struct mqueue_inode_info *info,
 784                                   struct msg_msg *message,
 785                                   struct ext_wait_queue *receiver)
 786 {
 787         receiver->msg = message;
 788         list_del(&receiver->list);
 789         receiver->state = STATE_PENDING;
 790         wake_up_process(receiver->task);
 791         smp_wmb();
 792         receiver->state = STATE_READY;
 793 }
 794
 795 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 796  * gets its message and put to the queue (we have one free place for sure). */
 797 static inline void pipelined_receive(struct mqueue_inode_info *info)
 798 {
 799         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 800
 801         if (!sender) {
 802                 /* for poll */
 803                 wake_up_interruptible(&info->wait_q);
 804                 return;
 805         }
 806         msg_insert(sender->msg, info);
 807         list_del(&sender->list);
 808         sender->state = STATE_PENDING;
 809         wake_up_process(sender->task);
 810         smp_wmb();
 811         sender->state = STATE_READY;
 812 }
 813
 814 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 815         size_t msg_len, unsigned int msg_prio,
 816         const struct timespec __user *u_abs_timeout)
 817 {
 818         struct file *filp;
 819         struct inode *inode;
 820         struct ext_wait_queue wait;
 821         struct ext_wait_queue *receiver;
 822         struct msg_msg *msg_ptr;
 823         struct mqueue_inode_info *info;
 824         long timeout;
 825         int 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         timeout = prepare_timeout(u_abs_timeout);
 910
 911         ret = -EBADF;
 912         filp = fget(mqdes);
 913         if (unlikely(!filp))
 914                 goto out;
 915
 916         inode = filp->f_dentry->d_inode;
 917         if (unlikely(filp->f_op != &mqueue_file_operations))
 918                 goto out_fput;
 919         info = MQUEUE_I(inode);
 920
 921         if (unlikely(!(filp->f_mode & FMODE_READ)))
 922                 goto out_fput;
 923
 924         /* checks if buffer is big enough */
 925         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 926                 ret = -EMSGSIZE;
 927                 goto out_fput;
 928         }
 929
 930         spin_lock(&info->lock);
 931         if (info->attr.mq_curmsgs == 0) {
 932                 if (filp->f_flags & O_NONBLOCK) {
 933                         spin_unlock(&info->lock);
 934                         ret = -EAGAIN;
 935                         msg_ptr = NULL;
 936                 } else if (unlikely(timeout < 0)) {
 937                         spin_unlock(&info->lock);
 938                         ret = timeout;
 939                         msg_ptr = NULL;
 940                 } else {
 941                         wait.task = current;
 942                         wait.state = STATE_NONE;
 943                         ret = wq_sleep(info, RECV, timeout, &wait);
 944                         msg_ptr = wait.msg;
 945                 }
 946         } else {
 947                 msg_ptr = msg_get(info);
 948
 949                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 950                                 CURRENT_TIME;
 951
 952                 /* There is now free space in queue. */
 953                 pipelined_receive(info);
 954                 spin_unlock(&info->lock);
 955                 ret = 0;
 956         }
 957         if (ret == 0) {
 958                 ret = msg_ptr->m_ts;
 959
 960                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 961                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 962                         ret = -EFAULT;
 963                 }
 964                 free_msg(msg_ptr);
 965         }
 966 out_fput:
 967         fput(filp);
 968 out:
 969         return ret;
 970 }
 971
 972 /*
 973  * Notes: the case when user wants us to deregister (with NULL as pointer)
 974  * and he isn't currently owner of notification, will be silently discarded.
 975  * It isn't explicitly defined in the POSIX.
 976  */
 977 asmlinkage long sys_mq_notify(mqd_t mqdes,
 978                                 const struct sigevent __user *u_notification)
 979 {
 980         int ret;
 981         struct file *filp;
 982         struct sock *sock;
 983         struct inode *inode;
 984         struct sigevent notification;
 985         struct mqueue_inode_info *info;
 986         struct sk_buff *nc;
 987
 988         nc = NULL;
 989         sock = NULL;
 990         if (u_notification != NULL) {
 991                 if (copy_from_user(&notification, u_notification,
 992                                         sizeof(struct sigevent)))
 993                         return -EFAULT;
 994
 995                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
 996                              notification.sigev_notify != SIGEV_SIGNAL &&
 997                              notification.sigev_notify != SIGEV_THREAD))
 998                         return -EINVAL;
 999                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1000                         !valid_signal(notification.sigev_signo)) {
1001                         return -EINVAL;
1002                 }
1003                 if (notification.sigev_notify == SIGEV_THREAD) {
1004                         /* create the notify skb */
1005                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1006                         ret = -ENOMEM;
1007                         if (!nc)
1008                                 goto out;
1009                         ret = -EFAULT;
1010                         if (copy_from_user(nc->data,
1011                                         notification.sigev_value.sival_ptr,
1012                                         NOTIFY_COOKIE_LEN)) {
1013                                 goto out;
1014                         }
1015
1016                         /* TODO: add a header? */
1017                         skb_put(nc, NOTIFY_COOKIE_LEN);
1018                         /* and attach it to the socket */
1019 retry:
1020                         filp = fget(notification.sigev_signo);
1021                         ret = -EBADF;
1022                         if (!filp)
1023                                 goto out;
1024                         sock = netlink_getsockbyfilp(filp);
1025                         fput(filp);
1026                         if (IS_ERR(sock)) {
1027                                 ret = PTR_ERR(sock);
1028                                 sock = NULL;
1029                                 goto out;
1030                         }
1031
1032                         ret = netlink_attachskb(sock, nc, 0,
1033                                         MAX_SCHEDULE_TIMEOUT, NULL);
1034                         if (ret == 1)
1035                                 goto retry;
1036                         if (ret) {
1037                                 sock = NULL;
1038                                 nc = NULL;
1039                                 goto out;
1040                         }
1041                 }
1042         }
1043
1044         ret = -EBADF;
1045         filp = fget(mqdes);
1046         if (!filp)
1047                 goto out;
1048
1049         inode = filp->f_dentry->d_inode;
1050         if (unlikely(filp->f_op != &mqueue_file_operations))
1051                 goto out_fput;
1052         info = MQUEUE_I(inode);
1053
1054         ret = 0;
1055         spin_lock(&info->lock);
1056         if (u_notification == NULL) {
1057                 if (info->notify_owner == current->tgid) {
1058                         remove_notification(info);
1059                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1060                 }
1061         } else if (info->notify_owner != 0) {
1062                 ret = -EBUSY;
1063         } else {
1064                 switch (notification.sigev_notify) {
1065                 case SIGEV_NONE:
1066                         info->notify.sigev_notify = SIGEV_NONE;
1067                         break;
1068                 case SIGEV_THREAD:
1069                         info->notify_sock = sock;
1070                         info->notify_cookie = nc;
1071                         sock = NULL;
1072                         nc = NULL;
1073                         info->notify.sigev_notify = SIGEV_THREAD;
1074                         break;
1075                 case SIGEV_SIGNAL:
1076                         info->notify.sigev_signo = notification.sigev_signo;
1077                         info->notify.sigev_value = notification.sigev_value;
1078                         info->notify.sigev_notify = SIGEV_SIGNAL;
1079                         break;
1080                 }
1081                 info->notify_owner = current->tgid;
1082                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1083         }
1084         spin_unlock(&info->lock);
1085 out_fput:
1086         fput(filp);
1087 out:
1088         if (sock) {
1089                 netlink_detachskb(sock, nc);
1090         } else if (nc) {
1091                 dev_kfree_skb(nc);
1092         }
1093         return ret;
1094 }
1095
1096 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1097                         const struct mq_attr __user *u_mqstat,
1098                         struct mq_attr __user *u_omqstat)
1099 {
1100         int ret;
1101         struct mq_attr mqstat, omqstat;
1102         struct file *filp;
1103         struct inode *inode;
1104         struct mqueue_inode_info *info;
1105
1106         if (u_mqstat != NULL) {
1107                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1108                         return -EFAULT;
1109                 if (mqstat.mq_flags & (~O_NONBLOCK))
1110                         return -EINVAL;
1111         }
1112
1113         ret = -EBADF;
1114         filp = fget(mqdes);
1115         if (!filp)
1116                 goto out;
1117
1118         inode = filp->f_dentry->d_inode;
1119         if (unlikely(filp->f_op != &mqueue_file_operations))
1120                 goto out_fput;
1121         info = MQUEUE_I(inode);
1122
1123         spin_lock(&info->lock);
1124
1125         omqstat = info->attr;
1126         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1127         if (u_mqstat) {
1128                 if (mqstat.mq_flags & O_NONBLOCK)
1129                         filp->f_flags |= O_NONBLOCK;
1130                 else
1131                         filp->f_flags &= ~O_NONBLOCK;
1132
1133                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1134         }
1135
1136         spin_unlock(&info->lock);
1137
1138         ret = 0;
1139         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1140                                                 sizeof(struct mq_attr)))
1141                 ret = -EFAULT;
1142
1143 out_fput:
1144         fput(filp);
1145 out:
1146         return ret;
1147 }
1148
1149 static struct inode_operations mqueue_dir_inode_operations = {
1150         .lookup = simple_lookup,
1151         .create = mqueue_create,
1152         .unlink = mqueue_unlink,
1153 };
1154
1155 static struct file_operations mqueue_file_operations = {
1156         .flush = mqueue_flush_file,
1157         .poll = mqueue_poll_file,
1158         .read = mqueue_read_file,
1159 };
1160
1161 static struct super_operations mqueue_super_ops = {
1162         .alloc_inode = mqueue_alloc_inode,
1163         .destroy_inode = mqueue_destroy_inode,
1164         .statfs = simple_statfs,
1165         .delete_inode = mqueue_delete_inode,
1166         .drop_inode = generic_delete_inode,
1167 };
1168
1169 static struct file_system_type mqueue_fs_type = {
1170         .name = "mqueue",
1171         .get_sb = mqueue_get_sb,
1172         .kill_sb = kill_litter_super,
1173 };
1174
1175 static int msg_max_limit_min = DFLT_MSGMAX;
1176 static int msg_max_limit_max = HARD_MSGMAX;
1177
1178 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1179 static int msg_maxsize_limit_max = INT_MAX;
1180
1181 static ctl_table mq_sysctls[] = {
1182         {
1183                 .ctl_name       = CTL_QUEUESMAX,
1184                 .procname       = "queues_max",
1185                 .data           = &queues_max,
1186                 .maxlen         = sizeof(int),
1187                 .mode           = 0644,
1188                 .proc_handler   = &proc_dointvec,
1189         },
1190         {
1191                 .ctl_name       = CTL_MSGMAX,
1192                 .procname       = "msg_max",
1193                 .data           = &msg_max,
1194                 .maxlen         = sizeof(int),
1195                 .mode           = 0644,
1196                 .proc_handler   = &proc_dointvec_minmax,
1197                 .extra1         = &msg_max_limit_min,
1198                 .extra2         = &msg_max_limit_max,
1199         },
1200         {
1201                 .ctl_name       = CTL_MSGSIZEMAX,
1202                 .procname       = "msgsize_max",
1203                 .data           = &msgsize_max,
1204                 .maxlen         = sizeof(int),
1205                 .mode           = 0644,
1206                 .proc_handler   = &proc_dointvec_minmax,
1207                 .extra1         = &msg_maxsize_limit_min,
1208                 .extra2         = &msg_maxsize_limit_max,
1209         },
1210         { .ctl_name = 0 }
1211 };
1212
1213 static ctl_table mq_sysctl_dir[] = {
1214         {
1215                 .ctl_name       = FS_MQUEUE,
1216                 .procname       = "mqueue",
1217                 .mode           = 0555,
1218                 .child          = mq_sysctls,
1219         },
1220         { .ctl_name = 0 }
1221 };
1222
1223 static ctl_table mq_sysctl_root[] = {
1224         {
1225                 .ctl_name       = CTL_FS,
1226                 .procname       = "fs",
1227                 .mode           = 0555,
1228                 .child          = mq_sysctl_dir,
1229         },
1230         { .ctl_name = 0 }
1231 };
1232
1233 static int __init init_mqueue_fs(void)
1234 {
1235         int error;
1236
1237         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1238                                 sizeof(struct mqueue_inode_info), 0,
1239                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1240         if (mqueue_inode_cachep == NULL)
1241                 return -ENOMEM;
1242
1243         /* ignore failues - they are not fatal */
1244         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1245
1246         error = register_filesystem(&mqueue_fs_type);
1247         if (error)
1248                 goto out_sysctl;
1249
1250         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1251                 error = PTR_ERR(mqueue_mnt);
1252                 goto out_filesystem;
1253         }
1254
1255         /* internal initialization - not common for vfs */
1256         queues_count = 0;
1257         spin_lock_init(&mq_lock);
1258
1259         return 0;
1260
1261 out_filesystem:
1262         unregister_filesystem(&mqueue_fs_type);
1263 out_sysctl:
1264         if (mq_sysctl_table)
1265                 unregister_sysctl_table(mq_sysctl_table);
1266         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1267                 printk(KERN_INFO
1268                         "mqueue_inode_cache: not all structures were freed\n");
1269         }
1270         return error;
1271 }
1272
1273 __initcall(init_mqueue_fs);