ipc/mqueue.c

   1 /*
   2  * POSIX message queues filesystem for Linux.
   3  *
   4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    ([email protected])
   5  *                          Michal Wronski          ([email protected])
   6  *
   7  * Spinlocks:               Mohamed Abbas           ([email protected])
   8  * Lockless receive & send, fd based notify:
   9  *                          Manfred Spraul          ([email protected])
  10  *
  11  * Audit:                   George Wilson           ([email protected])
  12  *
  13  * This file is released under the GPL.
  14  */
  15
  16 #include <linux/capability.h>
  17 #include <linux/init.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/file.h>
  20 #include <linux/mount.h>
  21 #include <linux/namei.h>
  22 #include <linux/sysctl.h>
  23 #include <linux/poll.h>
  24 #include <linux/mqueue.h>
  25 #include <linux/msg.h>
  26 #include <linux/skbuff.h>
  27 #include <linux/netlink.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/audit.h>
  30 #include <linux/signal.h>
  31 #include <linux/mutex.h>
  32 #include <linux/nsproxy.h>
  33 #include <linux/pid.h>
  34 #include <linux/ipc_namespace.h>
  35 #include <linux/slab.h>
  36
  37 #include <net/sock.h>
  38 #include "util.h"
  39
  40 #define MQUEUE_MAGIC    0x19800202
  41 #define DIRENT_SIZE     20
  42 #define FILENT_SIZE     80
  43
  44 #define SEND            0
  45 #define RECV            1
  46
  47 #define STATE_NONE      0
  48 #define STATE_PENDING   1
  49 #define STATE_READY     2
  50
  51 struct ext_wait_queue {         /* queue of sleeping tasks */
  52         struct task_struct *task;
  53         struct list_head list;
  54         struct msg_msg *msg;    /* ptr of loaded message */
  55         int state;              /* one of STATE_* values */
  56 };
  57
  58 struct mqueue_inode_info {
  59         spinlock_t lock;
  60         struct inode vfs_inode;
  61         wait_queue_head_t wait_q;
  62
  63         struct msg_msg **messages;
  64         struct mq_attr attr;
  65
  66         struct sigevent notify;
  67         struct pid* notify_owner;
  68         struct user_struct *user;       /* user who created, for accounting */
  69         struct sock *notify_sock;
  70         struct sk_buff *notify_cookie;
  71
  72         /* for tasks waiting for free space and messages, respectively */
  73         struct ext_wait_queue e_wait_q[2];
  74
  75         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  76 };
  77
  78 static const struct inode_operations mqueue_dir_inode_operations;
  79 static const struct file_operations mqueue_file_operations;
  80 static const struct super_operations mqueue_super_ops;
  81 static void remove_notification(struct mqueue_inode_info *info);
  82
  83 static struct kmem_cache *mqueue_inode_cachep;
  84
  85 static struct ctl_table_header * mq_sysctl_table;
  86
  87 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  88 {
  89         return container_of(inode, struct mqueue_inode_info, vfs_inode);
  90 }
  91
  92 /*
  93  * This routine should be called with the mq_lock held.
  94  */
  95 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
  96 {
  97         return get_ipc_ns(inode->i_sb->s_fs_info);
  98 }
  99
 100 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 101 {
 102         struct ipc_namespace *ns;
 103
 104         spin_lock(&mq_lock);
 105         ns = __get_ns_from_inode(inode);
 106         spin_unlock(&mq_lock);
 107         return ns;
 108 }
 109
 110 static struct inode *mqueue_get_inode(struct super_block *sb,
 111                 struct ipc_namespace *ipc_ns, int mode,
 112                 struct mq_attr *attr)
 113 {
 114         struct user_struct *u = current_user();
 115         struct inode *inode;
 116
 117         inode = new_inode(sb);
 118         if (inode) {
 119                 inode->i_mode = mode;
 120                 inode->i_uid = current_fsuid();
 121                 inode->i_gid = current_fsgid();
 122                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 123                                 CURRENT_TIME;
 124
 125                 if (S_ISREG(mode)) {
 126                         struct mqueue_inode_info *info;
 127                         struct task_struct *p = current;
 128                         unsigned long mq_bytes, mq_msg_tblsz;
 129
 130                         inode->i_fop = &mqueue_file_operations;
 131                         inode->i_size = FILENT_SIZE;
 132                         /* mqueue specific info */
 133                         info = MQUEUE_I(inode);
 134                         spin_lock_init(&info->lock);
 135                         init_waitqueue_head(&info->wait_q);
 136                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 137                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 138                         info->notify_owner = NULL;
 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 = ipc_ns->mq_msg_max;
 143                         info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
 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                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 150                         if (!info->messages)
 151                                 goto out_inode;
 152
 153                         mq_bytes = (mq_msg_tblsz +
 154                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 155
 156                         spin_lock(&mq_lock);
 157                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 158                             u->mq_bytes + mq_bytes >
 159                             task_rlimit(p, RLIMIT_MSGQUEUE)) {
 160                                 spin_unlock(&mq_lock);
 161                                 /* mqueue_delete_inode() releases info->messages */
 162                                 goto out_inode;
 163                         }
 164                         u->mq_bytes += mq_bytes;
 165                         spin_unlock(&mq_lock);
 166
 167                         /* all is ok */
 168                         info->user = get_uid(u);
 169                 } else if (S_ISDIR(mode)) {
 170                         inc_nlink(inode);
 171                         /* Some things misbehave if size == 0 on a directory */
 172                         inode->i_size = 2 * DIRENT_SIZE;
 173                         inode->i_op = &mqueue_dir_inode_operations;
 174                         inode->i_fop = &simple_dir_operations;
 175                 }
 176         }
 177         return inode;
 178 out_inode:
 179         make_bad_inode(inode);
 180         iput(inode);
 181         return NULL;
 182 }
 183
 184 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 185 {
 186         struct inode *inode;
 187         struct ipc_namespace *ns = data;
 188         int error;
 189
 190         sb->s_blocksize = PAGE_CACHE_SIZE;
 191         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 192         sb->s_magic = MQUEUE_MAGIC;
 193         sb->s_op = &mqueue_super_ops;
 194
 195         inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
 196                                 NULL);
 197         if (!inode) {
 198                 error = -ENOMEM;
 199                 goto out;
 200         }
 201
 202         sb->s_root = d_alloc_root(inode);
 203         if (!sb->s_root) {
 204                 iput(inode);
 205                 error = -ENOMEM;
 206                 goto out;
 207         }
 208         error = 0;
 209
 210 out:
 211         return error;
 212 }
 213
 214 static int mqueue_get_sb(struct file_system_type *fs_type,
 215                          int flags, const char *dev_name,
 216                          void *data, struct vfsmount *mnt)
 217 {
 218         if (!(flags & MS_KERNMOUNT))
 219                 data = current->nsproxy->ipc_ns;
 220         return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
 221 }
 222
 223 static void init_once(void *foo)
 224 {
 225         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 226
 227         inode_init_once(&p->vfs_inode);
 228 }
 229
 230 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 231 {
 232         struct mqueue_inode_info *ei;
 233
 234         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 235         if (!ei)
 236                 return NULL;
 237         return &ei->vfs_inode;
 238 }
 239
 240 static void mqueue_destroy_inode(struct inode *inode)
 241 {
 242         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 243 }
 244
 245 static void mqueue_delete_inode(struct inode *inode)
 246 {
 247         struct mqueue_inode_info *info;
 248         struct user_struct *user;
 249         unsigned long mq_bytes;
 250         int i;
 251         struct ipc_namespace *ipc_ns;
 252
 253         if (S_ISDIR(inode->i_mode)) {
 254                 clear_inode(inode);
 255                 return;
 256         }
 257         ipc_ns = get_ns_from_inode(inode);
 258         info = MQUEUE_I(inode);
 259         spin_lock(&info->lock);
 260         for (i = 0; i < info->attr.mq_curmsgs; i++)
 261                 free_msg(info->messages[i]);
 262         kfree(info->messages);
 263         spin_unlock(&info->lock);
 264
 265         clear_inode(inode);
 266
 267         /* Total amount of bytes accounted for the mqueue */
 268         mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
 269             + info->attr.mq_msgsize);
 270         user = info->user;
 271         if (user) {
 272                 spin_lock(&mq_lock);
 273                 user->mq_bytes -= mq_bytes;
 274                 /*
 275                  * get_ns_from_inode() ensures that the
 276                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 277                  * to which we now hold a reference, or it is NULL.
 278                  * We can't put it here under mq_lock, though.
 279                  */
 280                 if (ipc_ns)
 281                         ipc_ns->mq_queues_count--;
 282                 spin_unlock(&mq_lock);
 283                 free_uid(user);
 284         }
 285         if (ipc_ns)
 286                 put_ipc_ns(ipc_ns);
 287 }
 288
 289 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 290                                 int mode, struct nameidata *nd)
 291 {
 292         struct inode *inode;
 293         struct mq_attr *attr = dentry->d_fsdata;
 294         int error;
 295         struct ipc_namespace *ipc_ns;
 296
 297         spin_lock(&mq_lock);
 298         ipc_ns = __get_ns_from_inode(dir);
 299         if (!ipc_ns) {
 300                 error = -EACCES;
 301                 goto out_unlock;
 302         }
 303         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 304                         !capable(CAP_SYS_RESOURCE)) {
 305                 error = -ENOSPC;
 306                 goto out_unlock;
 307         }
 308         ipc_ns->mq_queues_count++;
 309         spin_unlock(&mq_lock);
 310
 311         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 312         if (!inode) {
 313                 error = -ENOMEM;
 314                 spin_lock(&mq_lock);
 315                 ipc_ns->mq_queues_count--;
 316                 goto out_unlock;
 317         }
 318
 319         put_ipc_ns(ipc_ns);
 320         dir->i_size += DIRENT_SIZE;
 321         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 322
 323         d_instantiate(dentry, inode);
 324         dget(dentry);
 325         return 0;
 326 out_unlock:
 327         spin_unlock(&mq_lock);
 328         if (ipc_ns)
 329                 put_ipc_ns(ipc_ns);
 330         return error;
 331 }
 332
 333 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 334 {
 335         struct inode *inode = dentry->d_inode;
 336
 337         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 338         dir->i_size -= DIRENT_SIZE;
 339         drop_nlink(inode);
 340         dput(dentry);
 341         return 0;
 342 }
 343
 344 /*
 345 *       This is routine for system read from queue file.
 346 *       To avoid mess with doing here some sort of mq_receive we allow
 347 *       to read only queue size & notification info (the only values
 348 *       that are interesting from user point of view and aren't accessible
 349 *       through std routines)
 350 */
 351 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 352                                 size_t count, loff_t *off)
 353 {
 354         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 355         char buffer[FILENT_SIZE];
 356         ssize_t ret;
 357
 358         spin_lock(&info->lock);
 359         snprintf(buffer, sizeof(buffer),
 360                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 361                         info->qsize,
 362                         info->notify_owner ? info->notify.sigev_notify : 0,
 363                         (info->notify_owner &&
 364                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 365                                 info->notify.sigev_signo : 0,
 366                         pid_vnr(info->notify_owner));
 367         spin_unlock(&info->lock);
 368         buffer[sizeof(buffer)-1] = '\0';
 369
 370         ret = simple_read_from_buffer(u_data, count, off, buffer,
 371                                 strlen(buffer));
 372         if (ret <= 0)
 373                 return ret;
 374
 375         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 376         return ret;
 377 }
 378
 379 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 380 {
 381         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 382
 383         spin_lock(&info->lock);
 384         if (task_tgid(current) == info->notify_owner)
 385                 remove_notification(info);
 386
 387         spin_unlock(&info->lock);
 388         return 0;
 389 }
 390
 391 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 392 {
 393         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 394         int retval = 0;
 395
 396         poll_wait(filp, &info->wait_q, poll_tab);
 397
 398         spin_lock(&info->lock);
 399         if (info->attr.mq_curmsgs)
 400                 retval = POLLIN | POLLRDNORM;
 401
 402         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 403                 retval |= POLLOUT | POLLWRNORM;
 404         spin_unlock(&info->lock);
 405
 406         return retval;
 407 }
 408
 409 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 410 static void wq_add(struct mqueue_inode_info *info, int sr,
 411                         struct ext_wait_queue *ewp)
 412 {
 413         struct ext_wait_queue *walk;
 414
 415         ewp->task = current;
 416
 417         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 418                 if (walk->task->static_prio <= current->static_prio) {
 419                         list_add_tail(&ewp->list, &walk->list);
 420                         return;
 421                 }
 422         }
 423         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 424 }
 425
 426 /*
 427  * Puts current task to sleep. Caller must hold queue lock. After return
 428  * lock isn't held.
 429  * sr: SEND or RECV
 430  */
 431 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 432                     ktime_t *timeout, struct ext_wait_queue *ewp)
 433 {
 434         int retval;
 435         signed long time;
 436
 437         wq_add(info, sr, ewp);
 438
 439         for (;;) {
 440                 set_current_state(TASK_INTERRUPTIBLE);
 441
 442                 spin_unlock(&info->lock);
 443                 time = schedule_hrtimeout_range_clock(timeout,
 444                     HRTIMER_MODE_ABS, 0, CLOCK_REALTIME);
 445
 446                 while (ewp->state == STATE_PENDING)
 447                         cpu_relax();
 448
 449                 if (ewp->state == STATE_READY) {
 450                         retval = 0;
 451                         goto out;
 452                 }
 453                 spin_lock(&info->lock);
 454                 if (ewp->state == STATE_READY) {
 455                         retval = 0;
 456                         goto out_unlock;
 457                 }
 458                 if (signal_pending(current)) {
 459                         retval = -ERESTARTSYS;
 460                         break;
 461                 }
 462                 if (time == 0) {
 463                         retval = -ETIMEDOUT;
 464                         break;
 465                 }
 466         }
 467         list_del(&ewp->list);
 468 out_unlock:
 469         spin_unlock(&info->lock);
 470 out:
 471         return retval;
 472 }
 473
 474 /*
 475  * Returns waiting task that should be serviced first or NULL if none exists
 476  */
 477 static struct ext_wait_queue *wq_get_first_waiter(
 478                 struct mqueue_inode_info *info, int sr)
 479 {
 480         struct list_head *ptr;
 481
 482         ptr = info->e_wait_q[sr].list.prev;
 483         if (ptr == &info->e_wait_q[sr].list)
 484                 return NULL;
 485         return list_entry(ptr, struct ext_wait_queue, list);
 486 }
 487
 488 /* Auxiliary functions to manipulate messages' list */
 489 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 490 {
 491         int k;
 492
 493         k = info->attr.mq_curmsgs - 1;
 494         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 495                 info->messages[k + 1] = info->messages[k];
 496                 k--;
 497         }
 498         info->attr.mq_curmsgs++;
 499         info->qsize += ptr->m_ts;
 500         info->messages[k + 1] = ptr;
 501 }
 502
 503 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 504 {
 505         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 506         return info->messages[info->attr.mq_curmsgs];
 507 }
 508
 509 static inline void set_cookie(struct sk_buff *skb, char code)
 510 {
 511         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 512 }
 513
 514 /*
 515  * The next function is only to split too long sys_mq_timedsend
 516  */
 517 static void __do_notify(struct mqueue_inode_info *info)
 518 {
 519         /* notification
 520          * invoked when there is registered process and there isn't process
 521          * waiting synchronously for message AND state of queue changed from
 522          * empty to not empty. Here we are sure that no one is waiting
 523          * synchronously. */
 524         if (info->notify_owner &&
 525             info->attr.mq_curmsgs == 1) {
 526                 struct siginfo sig_i;
 527                 switch (info->notify.sigev_notify) {
 528                 case SIGEV_NONE:
 529                         break;
 530                 case SIGEV_SIGNAL:
 531                         /* sends signal */
 532
 533                         sig_i.si_signo = info->notify.sigev_signo;
 534                         sig_i.si_errno = 0;
 535                         sig_i.si_code = SI_MESGQ;
 536                         sig_i.si_value = info->notify.sigev_value;
 537                         sig_i.si_pid = task_tgid_nr_ns(current,
 538                                                 ns_of_pid(info->notify_owner));
 539                         sig_i.si_uid = current_uid();
 540
 541                         kill_pid_info(info->notify.sigev_signo,
 542                                       &sig_i, info->notify_owner);
 543                         break;
 544                 case SIGEV_THREAD:
 545                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 546                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 547                         break;
 548                 }
 549                 /* after notification unregisters process */
 550                 put_pid(info->notify_owner);
 551                 info->notify_owner = NULL;
 552         }
 553         wake_up(&info->wait_q);
 554 }
 555
 556 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
 557                            ktime_t *expires, struct timespec *ts)
 558 {
 559         if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
 560                 return -EFAULT;
 561         if (!timespec_valid(ts))
 562                 return -EINVAL;
 563
 564         *expires = timespec_to_ktime(*ts);
 565         return 0;
 566 }
 567
 568 static void remove_notification(struct mqueue_inode_info *info)
 569 {
 570         if (info->notify_owner != NULL &&
 571             info->notify.sigev_notify == SIGEV_THREAD) {
 572                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 573                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 574         }
 575         put_pid(info->notify_owner);
 576         info->notify_owner = NULL;
 577 }
 578
 579 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
 580 {
 581         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 582                 return 0;
 583         if (capable(CAP_SYS_RESOURCE)) {
 584                 if (attr->mq_maxmsg > HARD_MSGMAX)
 585                         return 0;
 586         } else {
 587                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
 588                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
 589                         return 0;
 590         }
 591         /* check for overflow */
 592         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 593                 return 0;
 594         if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
 595             + sizeof (struct msg_msg *))) <
 596             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 597                 return 0;
 598         return 1;
 599 }
 600
 601 /*
 602  * Invoked when creating a new queue via sys_mq_open
 603  */
 604 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
 605                         struct dentry *dentry, int oflag, mode_t mode,
 606                         struct mq_attr *attr)
 607 {
 608         const struct cred *cred = current_cred();
 609         struct file *result;
 610         int ret;
 611
 612         if (attr) {
 613                 if (!mq_attr_ok(ipc_ns, attr)) {
 614                         ret = -EINVAL;
 615                         goto out;
 616                 }
 617                 /* store for use during create */
 618                 dentry->d_fsdata = attr;
 619         }
 620
 621         mode &= ~current_umask();
 622         ret = mnt_want_write(ipc_ns->mq_mnt);
 623         if (ret)
 624                 goto out;
 625         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 626         dentry->d_fsdata = NULL;
 627         if (ret)
 628                 goto out_drop_write;
 629
 630         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 631         /*
 632          * dentry_open() took a persistent mnt_want_write(),
 633          * so we can now drop this one.
 634          */
 635         mnt_drop_write(ipc_ns->mq_mnt);
 636         return result;
 637
 638 out_drop_write:
 639         mnt_drop_write(ipc_ns->mq_mnt);
 640 out:
 641         dput(dentry);
 642         mntput(ipc_ns->mq_mnt);
 643         return ERR_PTR(ret);
 644 }
 645
 646 /* Opens existing queue */
 647 static struct file *do_open(struct ipc_namespace *ipc_ns,
 648                                 struct dentry *dentry, int oflag)
 649 {
 650         int ret;
 651         const struct cred *cred = current_cred();
 652
 653         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 654                                                   MAY_READ | MAY_WRITE };
 655
 656         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 657                 ret = -EINVAL;
 658                 goto err;
 659         }
 660
 661         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 662                 ret = -EACCES;
 663                 goto err;
 664         }
 665
 666         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 667
 668 err:
 669         dput(dentry);
 670         mntput(ipc_ns->mq_mnt);
 671         return ERR_PTR(ret);
 672 }
 673
 674 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
 675                 struct mq_attr __user *, u_attr)
 676 {
 677         struct dentry *dentry;
 678         struct file *filp;
 679         char *name;
 680         struct mq_attr attr;
 681         int fd, error;
 682         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 683
 684         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 685                 return -EFAULT;
 686
 687         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
 688
 689         if (IS_ERR(name = getname(u_name)))
 690                 return PTR_ERR(name);
 691
 692         fd = get_unused_fd_flags(O_CLOEXEC);
 693         if (fd < 0)
 694                 goto out_putname;
 695
 696         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 697         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 698         if (IS_ERR(dentry)) {
 699                 error = PTR_ERR(dentry);
 700                 goto out_putfd;
 701         }
 702         mntget(ipc_ns->mq_mnt);
 703
 704         if (oflag & O_CREAT) {
 705                 if (dentry->d_inode) {  /* entry already exists */
 706                         audit_inode(name, dentry);
 707                         if (oflag & O_EXCL) {
 708                                 error = -EEXIST;
 709                                 goto out;
 710                         }
 711                         filp = do_open(ipc_ns, dentry, oflag);
 712                 } else {
 713                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
 714                                                 dentry, oflag, mode,
 715                                                 u_attr ? &attr : NULL);
 716                 }
 717         } else {
 718                 if (!dentry->d_inode) {
 719                         error = -ENOENT;
 720                         goto out;
 721                 }
 722                 audit_inode(name, dentry);
 723                 filp = do_open(ipc_ns, dentry, oflag);
 724         }
 725
 726         if (IS_ERR(filp)) {
 727                 error = PTR_ERR(filp);
 728                 goto out_putfd;
 729         }
 730
 731         fd_install(fd, filp);
 732         goto out_upsem;
 733
 734 out:
 735         dput(dentry);
 736         mntput(ipc_ns->mq_mnt);
 737 out_putfd:
 738         put_unused_fd(fd);
 739         fd = error;
 740 out_upsem:
 741         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 742 out_putname:
 743         putname(name);
 744         return fd;
 745 }
 746
 747 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 748 {
 749         int err;
 750         char *name;
 751         struct dentry *dentry;
 752         struct inode *inode = NULL;
 753         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 754
 755         name = getname(u_name);
 756         if (IS_ERR(name))
 757                 return PTR_ERR(name);
 758
 759         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
 760                         I_MUTEX_PARENT);
 761         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 762         if (IS_ERR(dentry)) {
 763                 err = PTR_ERR(dentry);
 764                 goto out_unlock;
 765         }
 766
 767         if (!dentry->d_inode) {
 768                 err = -ENOENT;
 769                 goto out_err;
 770         }
 771
 772         inode = dentry->d_inode;
 773         if (inode)
 774                 atomic_inc(&inode->i_count);
 775         err = mnt_want_write(ipc_ns->mq_mnt);
 776         if (err)
 777                 goto out_err;
 778         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 779         mnt_drop_write(ipc_ns->mq_mnt);
 780 out_err:
 781         dput(dentry);
 782
 783 out_unlock:
 784         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 785         putname(name);
 786         if (inode)
 787                 iput(inode);
 788
 789         return err;
 790 }
 791
 792 /* Pipelined send and receive functions.
 793  *
 794  * If a receiver finds no waiting message, then it registers itself in the
 795  * list of waiting receivers. A sender checks that list before adding the new
 796  * message into the message array. If there is a waiting receiver, then it
 797  * bypasses the message array and directly hands the message over to the
 798  * receiver.
 799  * The receiver accepts the message and returns without grabbing the queue
 800  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 801  * are necessary. The same algorithm is used for sysv semaphores, see
 802  * ipc/sem.c for more details.
 803  *
 804  * The same algorithm is used for senders.
 805  */
 806
 807 /* pipelined_send() - send a message directly to the task waiting in
 808  * sys_mq_timedreceive() (without inserting message into a queue).
 809  */
 810 static inline void pipelined_send(struct mqueue_inode_info *info,
 811                                   struct msg_msg *message,
 812                                   struct ext_wait_queue *receiver)
 813 {
 814         receiver->msg = message;
 815         list_del(&receiver->list);
 816         receiver->state = STATE_PENDING;
 817         wake_up_process(receiver->task);
 818         smp_wmb();
 819         receiver->state = STATE_READY;
 820 }
 821
 822 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 823  * gets its message and put to the queue (we have one free place for sure). */
 824 static inline void pipelined_receive(struct mqueue_inode_info *info)
 825 {
 826         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 827
 828         if (!sender) {
 829                 /* for poll */
 830                 wake_up_interruptible(&info->wait_q);
 831                 return;
 832         }
 833         msg_insert(sender->msg, info);
 834         list_del(&sender->list);
 835         sender->state = STATE_PENDING;
 836         wake_up_process(sender->task);
 837         smp_wmb();
 838         sender->state = STATE_READY;
 839 }
 840
 841 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 842                 size_t, msg_len, unsigned int, msg_prio,
 843                 const struct timespec __user *, u_abs_timeout)
 844 {
 845         struct file *filp;
 846         struct inode *inode;
 847         struct ext_wait_queue wait;
 848         struct ext_wait_queue *receiver;
 849         struct msg_msg *msg_ptr;
 850         struct mqueue_inode_info *info;
 851         ktime_t expires, *timeout = NULL;
 852         struct timespec ts;
 853         int ret;
 854
 855         if (u_abs_timeout) {
 856                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 857                 if (res)
 858                         return res;
 859                 timeout = &expires;
 860         }
 861
 862         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 863                 return -EINVAL;
 864
 865         audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
 866
 867         filp = fget(mqdes);
 868         if (unlikely(!filp)) {
 869                 ret = -EBADF;
 870                 goto out;
 871         }
 872
 873         inode = filp->f_path.dentry->d_inode;
 874         if (unlikely(filp->f_op != &mqueue_file_operations)) {
 875                 ret = -EBADF;
 876                 goto out_fput;
 877         }
 878         info = MQUEUE_I(inode);
 879         audit_inode(NULL, filp->f_path.dentry);
 880
 881         if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
 882                 ret = -EBADF;
 883                 goto out_fput;
 884         }
 885
 886         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 887                 ret = -EMSGSIZE;
 888                 goto out_fput;
 889         }
 890
 891         /* First try to allocate memory, before doing anything with
 892          * existing queues. */
 893         msg_ptr = load_msg(u_msg_ptr, msg_len);
 894         if (IS_ERR(msg_ptr)) {
 895                 ret = PTR_ERR(msg_ptr);
 896                 goto out_fput;
 897         }
 898         msg_ptr->m_ts = msg_len;
 899         msg_ptr->m_type = msg_prio;
 900
 901         spin_lock(&info->lock);
 902
 903         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 904                 if (filp->f_flags & O_NONBLOCK) {
 905                         spin_unlock(&info->lock);
 906                         ret = -EAGAIN;
 907                 } else {
 908                         wait.task = current;
 909                         wait.msg = (void *) msg_ptr;
 910                         wait.state = STATE_NONE;
 911                         ret = wq_sleep(info, SEND, timeout, &wait);
 912                 }
 913                 if (ret < 0)
 914                         free_msg(msg_ptr);
 915         } else {
 916                 receiver = wq_get_first_waiter(info, RECV);
 917                 if (receiver) {
 918                         pipelined_send(info, msg_ptr, receiver);
 919                 } else {
 920                         /* adds message to the queue */
 921                         msg_insert(msg_ptr, info);
 922                         __do_notify(info);
 923                 }
 924                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 925                                 CURRENT_TIME;
 926                 spin_unlock(&info->lock);
 927                 ret = 0;
 928         }
 929 out_fput:
 930         fput(filp);
 931 out:
 932         return ret;
 933 }
 934
 935 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
 936                 size_t, msg_len, unsigned int __user *, u_msg_prio,
 937                 const struct timespec __user *, u_abs_timeout)
 938 {
 939         ssize_t ret;
 940         struct msg_msg *msg_ptr;
 941         struct file *filp;
 942         struct inode *inode;
 943         struct mqueue_inode_info *info;
 944         struct ext_wait_queue wait;
 945         ktime_t expires, *timeout = NULL;
 946         struct timespec ts;
 947
 948         if (u_abs_timeout) {
 949                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 950                 if (res)
 951                         return res;
 952                 timeout = &expires;
 953         }
 954
 955         audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
 956
 957         filp = fget(mqdes);
 958         if (unlikely(!filp)) {
 959                 ret = -EBADF;
 960                 goto out;
 961         }
 962
 963         inode = filp->f_path.dentry->d_inode;
 964         if (unlikely(filp->f_op != &mqueue_file_operations)) {
 965                 ret = -EBADF;
 966                 goto out_fput;
 967         }
 968         info = MQUEUE_I(inode);
 969         audit_inode(NULL, filp->f_path.dentry);
 970
 971         if (unlikely(!(filp->f_mode & FMODE_READ))) {
 972                 ret = -EBADF;
 973                 goto out_fput;
 974         }
 975
 976         /* checks if buffer is big enough */
 977         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 978                 ret = -EMSGSIZE;
 979                 goto out_fput;
 980         }
 981
 982         spin_lock(&info->lock);
 983         if (info->attr.mq_curmsgs == 0) {
 984                 if (filp->f_flags & O_NONBLOCK) {
 985                         spin_unlock(&info->lock);
 986                         ret = -EAGAIN;
 987                 } else {
 988                         wait.task = current;
 989                         wait.state = STATE_NONE;
 990                         ret = wq_sleep(info, RECV, timeout, &wait);
 991                         msg_ptr = wait.msg;
 992                 }
 993         } else {
 994                 msg_ptr = msg_get(info);
 995
 996                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 997                                 CURRENT_TIME;
 998
 999                 /* There is now free space in queue. */
1000                 pipelined_receive(info);
1001                 spin_unlock(&info->lock);
1002                 ret = 0;
1003         }
1004         if (ret == 0) {
1005                 ret = msg_ptr->m_ts;
1006
1007                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1008                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1009                         ret = -EFAULT;
1010                 }
1011                 free_msg(msg_ptr);
1012         }
1013 out_fput:
1014         fput(filp);
1015 out:
1016         return ret;
1017 }
1018
1019 /*
1020  * Notes: the case when user wants us to deregister (with NULL as pointer)
1021  * and he isn't currently owner of notification, will be silently discarded.
1022  * It isn't explicitly defined in the POSIX.
1023  */
1024 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1025                 const struct sigevent __user *, u_notification)
1026 {
1027         int ret;
1028         struct file *filp;
1029         struct sock *sock;
1030         struct inode *inode;
1031         struct sigevent notification;
1032         struct mqueue_inode_info *info;
1033         struct sk_buff *nc;
1034
1035         if (u_notification) {
1036                 if (copy_from_user(&notification, u_notification,
1037                                         sizeof(struct sigevent)))
1038                         return -EFAULT;
1039         }
1040
1041         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1042
1043         nc = NULL;
1044         sock = NULL;
1045         if (u_notification != NULL) {
1046                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1047                              notification.sigev_notify != SIGEV_SIGNAL &&
1048                              notification.sigev_notify != SIGEV_THREAD))
1049                         return -EINVAL;
1050                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1051                         !valid_signal(notification.sigev_signo)) {
1052                         return -EINVAL;
1053                 }
1054                 if (notification.sigev_notify == SIGEV_THREAD) {
1055                         long timeo;
1056
1057                         /* create the notify skb */
1058                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1059                         if (!nc) {
1060                                 ret = -ENOMEM;
1061                                 goto out;
1062                         }
1063                         if (copy_from_user(nc->data,
1064                                         notification.sigev_value.sival_ptr,
1065                                         NOTIFY_COOKIE_LEN)) {
1066                                 ret = -EFAULT;
1067                                 goto out;
1068                         }
1069
1070                         /* TODO: add a header? */
1071                         skb_put(nc, NOTIFY_COOKIE_LEN);
1072                         /* and attach it to the socket */
1073 retry:
1074                         filp = fget(notification.sigev_signo);
1075                         if (!filp) {
1076                                 ret = -EBADF;
1077                                 goto out;
1078                         }
1079                         sock = netlink_getsockbyfilp(filp);
1080                         fput(filp);
1081                         if (IS_ERR(sock)) {
1082                                 ret = PTR_ERR(sock);
1083                                 sock = NULL;
1084                                 goto out;
1085                         }
1086
1087                         timeo = MAX_SCHEDULE_TIMEOUT;
1088                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1089                         if (ret == 1)
1090                                 goto retry;
1091                         if (ret) {
1092                                 sock = NULL;
1093                                 nc = NULL;
1094                                 goto out;
1095                         }
1096                 }
1097         }
1098
1099         filp = fget(mqdes);
1100         if (!filp) {
1101                 ret = -EBADF;
1102                 goto out;
1103         }
1104
1105         inode = filp->f_path.dentry->d_inode;
1106         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1107                 ret = -EBADF;
1108                 goto out_fput;
1109         }
1110         info = MQUEUE_I(inode);
1111
1112         ret = 0;
1113         spin_lock(&info->lock);
1114         if (u_notification == NULL) {
1115                 if (info->notify_owner == task_tgid(current)) {
1116                         remove_notification(info);
1117                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1118                 }
1119         } else if (info->notify_owner != NULL) {
1120                 ret = -EBUSY;
1121         } else {
1122                 switch (notification.sigev_notify) {
1123                 case SIGEV_NONE:
1124                         info->notify.sigev_notify = SIGEV_NONE;
1125                         break;
1126                 case SIGEV_THREAD:
1127                         info->notify_sock = sock;
1128                         info->notify_cookie = nc;
1129                         sock = NULL;
1130                         nc = NULL;
1131                         info->notify.sigev_notify = SIGEV_THREAD;
1132                         break;
1133                 case SIGEV_SIGNAL:
1134                         info->notify.sigev_signo = notification.sigev_signo;
1135                         info->notify.sigev_value = notification.sigev_value;
1136                         info->notify.sigev_notify = SIGEV_SIGNAL;
1137                         break;
1138                 }
1139
1140                 info->notify_owner = get_pid(task_tgid(current));
1141                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1142         }
1143         spin_unlock(&info->lock);
1144 out_fput:
1145         fput(filp);
1146 out:
1147         if (sock) {
1148                 netlink_detachskb(sock, nc);
1149         } else if (nc) {
1150                 dev_kfree_skb(nc);
1151         }
1152         return ret;
1153 }
1154
1155 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1156                 const struct mq_attr __user *, u_mqstat,
1157                 struct mq_attr __user *, u_omqstat)
1158 {
1159         int ret;
1160         struct mq_attr mqstat, omqstat;
1161         struct file *filp;
1162         struct inode *inode;
1163         struct mqueue_inode_info *info;
1164
1165         if (u_mqstat != NULL) {
1166                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1167                         return -EFAULT;
1168                 if (mqstat.mq_flags & (~O_NONBLOCK))
1169                         return -EINVAL;
1170         }
1171
1172         filp = fget(mqdes);
1173         if (!filp) {
1174                 ret = -EBADF;
1175                 goto out;
1176         }
1177
1178         inode = filp->f_path.dentry->d_inode;
1179         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1180                 ret = -EBADF;
1181                 goto out_fput;
1182         }
1183         info = MQUEUE_I(inode);
1184
1185         spin_lock(&info->lock);
1186
1187         omqstat = info->attr;
1188         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1189         if (u_mqstat) {
1190                 audit_mq_getsetattr(mqdes, &mqstat);
1191                 spin_lock(&filp->f_lock);
1192                 if (mqstat.mq_flags & O_NONBLOCK)
1193                         filp->f_flags |= O_NONBLOCK;
1194                 else
1195                         filp->f_flags &= ~O_NONBLOCK;
1196                 spin_unlock(&filp->f_lock);
1197
1198                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1199         }
1200
1201         spin_unlock(&info->lock);
1202
1203         ret = 0;
1204         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1205                                                 sizeof(struct mq_attr)))
1206                 ret = -EFAULT;
1207
1208 out_fput:
1209         fput(filp);
1210 out:
1211         return ret;
1212 }
1213
1214 static const struct inode_operations mqueue_dir_inode_operations = {
1215         .lookup = simple_lookup,
1216         .create = mqueue_create,
1217         .unlink = mqueue_unlink,
1218 };
1219
1220 static const struct file_operations mqueue_file_operations = {
1221         .flush = mqueue_flush_file,
1222         .poll = mqueue_poll_file,
1223         .read = mqueue_read_file,
1224 };
1225
1226 static const struct super_operations mqueue_super_ops = {
1227         .alloc_inode = mqueue_alloc_inode,
1228         .destroy_inode = mqueue_destroy_inode,
1229         .statfs = simple_statfs,
1230         .delete_inode = mqueue_delete_inode,
1231         .drop_inode = generic_delete_inode,
1232 };
1233
1234 static struct file_system_type mqueue_fs_type = {
1235         .name = "mqueue",
1236         .get_sb = mqueue_get_sb,
1237         .kill_sb = kill_litter_super,
1238 };
1239
1240 int mq_init_ns(struct ipc_namespace *ns)
1241 {
1242         ns->mq_queues_count  = 0;
1243         ns->mq_queues_max    = DFLT_QUEUESMAX;
1244         ns->mq_msg_max       = DFLT_MSGMAX;
1245         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1246
1247         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1248         if (IS_ERR(ns->mq_mnt)) {
1249                 int err = PTR_ERR(ns->mq_mnt);
1250                 ns->mq_mnt = NULL;
1251                 return err;
1252         }
1253         return 0;
1254 }
1255
1256 void mq_clear_sbinfo(struct ipc_namespace *ns)
1257 {
1258         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1259 }
1260
1261 void mq_put_mnt(struct ipc_namespace *ns)
1262 {
1263         mntput(ns->mq_mnt);
1264 }
1265
1266 static int __init init_mqueue_fs(void)
1267 {
1268         int error;
1269
1270         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1271                                 sizeof(struct mqueue_inode_info), 0,
1272                                 SLAB_HWCACHE_ALIGN, init_once);
1273         if (mqueue_inode_cachep == NULL)
1274                 return -ENOMEM;
1275
1276         /* ignore failures - they are not fatal */
1277         mq_sysctl_table = mq_register_sysctl_table();
1278
1279         error = register_filesystem(&mqueue_fs_type);
1280         if (error)
1281                 goto out_sysctl;
1282
1283         spin_lock_init(&mq_lock);
1284
1285         init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1286         if (IS_ERR(init_ipc_ns.mq_mnt)) {
1287                 error = PTR_ERR(init_ipc_ns.mq_mnt);
1288                 goto out_filesystem;
1289         }
1290
1291         return 0;
1292
1293 out_filesystem:
1294         unregister_filesystem(&mqueue_fs_type);
1295 out_sysctl:
1296         if (mq_sysctl_table)
1297                 unregister_sysctl_table(mq_sysctl_table);
1298         kmem_cache_destroy(mqueue_inode_cachep);
1299         return error;
1300 }
1301
1302 __initcall(init_mqueue_fs);