ipc/mqueue.c

   1 /*
   2  * POSIX message queues filesystem for Linux.
   3  *
   4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
   5  *                          Michal Wronski          (michal.wronski@gmail.com)
   6  *
   7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
   8  * Lockless receive & send, fd based notify:
   9  *                          Manfred Spraul          (manfred@colorfullife.com)
  10  *
  11  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
  12  *
  13  * This file is released under the GPL.
  14  */
  15
  16 #include <linux/capability.h>
  17 #include <linux/init.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/file.h>
  20 #include <linux/mount.h>
  21 #include <linux/namei.h>
  22 #include <linux/sysctl.h>
  23 #include <linux/poll.h>
  24 #include <linux/mqueue.h>
  25 #include <linux/msg.h>
  26 #include <linux/skbuff.h>
  27 #include <linux/netlink.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/audit.h>
  30 #include <linux/signal.h>
  31 #include <linux/mutex.h>
  32 #include <linux/nsproxy.h>
  33 #include <linux/pid.h>
  34 #include <linux/ipc_namespace.h>
  35 #include <linux/user_namespace.h>
  36 #include <linux/slab.h>
  37
  38 #include <net/sock.h>
  39 #include "util.h"
  40
  41 #define MQUEUE_MAGIC    0x19800202
  42 #define DIRENT_SIZE     20
  43 #define FILENT_SIZE     80
  44
  45 #define SEND            0
  46 #define RECV            1
  47
  48 #define STATE_NONE      0
  49 #define STATE_PENDING   1
  50 #define STATE_READY     2
  51
  52 struct ext_wait_queue {         /* queue of sleeping tasks */
  53         struct task_struct *task;
  54         struct list_head list;
  55         struct msg_msg *msg;    /* ptr of loaded message */
  56         int state;              /* one of STATE_* values */
  57 };
  58
  59 struct mqueue_inode_info {
  60         spinlock_t lock;
  61         struct inode vfs_inode;
  62         wait_queue_head_t wait_q;
  63
  64         struct msg_msg **messages;
  65         struct mq_attr attr;
  66
  67         struct sigevent notify;
  68         struct pid* notify_owner;
  69         struct user_namespace *notify_user_ns;
  70         struct user_struct *user;       /* user who created, for accounting */
  71         struct sock *notify_sock;
  72         struct sk_buff *notify_cookie;
  73
  74         /* for tasks waiting for free space and messages, respectively */
  75         struct ext_wait_queue e_wait_q[2];
  76
  77         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  78 };
  79
  80 static const struct inode_operations mqueue_dir_inode_operations;
  81 static const struct file_operations mqueue_file_operations;
  82 static const struct super_operations mqueue_super_ops;
  83 static void remove_notification(struct mqueue_inode_info *info);
  84
  85 static struct kmem_cache *mqueue_inode_cachep;
  86
  87 static struct ctl_table_header * mq_sysctl_table;
  88
  89 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  90 {
  91         return container_of(inode, struct mqueue_inode_info, vfs_inode);
  92 }
  93
  94 /*
  95  * This routine should be called with the mq_lock held.
  96  */
  97 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
  98 {
  99         return get_ipc_ns(inode->i_sb->s_fs_info);
 100 }
 101
 102 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 103 {
 104         struct ipc_namespace *ns;
 105
 106         spin_lock(&mq_lock);
 107         ns = __get_ns_from_inode(inode);
 108         spin_unlock(&mq_lock);
 109         return ns;
 110 }
 111
 112 static struct inode *mqueue_get_inode(struct super_block *sb,
 113                 struct ipc_namespace *ipc_ns, umode_t mode,
 114                 struct mq_attr *attr)
 115 {
 116         struct user_struct *u = current_user();
 117         struct inode *inode;
 118         int ret = -ENOMEM;
 119
 120         inode = new_inode(sb);
 121         if (!inode)
 122                 goto err;
 123
 124         inode->i_ino = get_next_ino();
 125         inode->i_mode = mode;
 126         inode->i_uid = current_fsuid();
 127         inode->i_gid = current_fsgid();
 128         inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
 129
 130         if (S_ISREG(mode)) {
 131                 struct mqueue_inode_info *info;
 132                 unsigned long mq_bytes, mq_msg_tblsz;
 133
 134                 inode->i_fop = &mqueue_file_operations;
 135                 inode->i_size = FILENT_SIZE;
 136                 /* mqueue specific info */
 137                 info = MQUEUE_I(inode);
 138                 spin_lock_init(&info->lock);
 139                 init_waitqueue_head(&info->wait_q);
 140                 INIT_LIST_HEAD(&info->e_wait_q[0].list);
 141                 INIT_LIST_HEAD(&info->e_wait_q[1].list);
 142                 info->notify_owner = NULL;
 143                 info->notify_user_ns = NULL;
 144                 info->qsize = 0;
 145                 info->user = NULL;      /* set when all is ok */
 146                 memset(&info->attr, 0, sizeof(info->attr));
 147                 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max, DFLT_MSG);
 148                 info->attr.mq_msgsize =
 149                         min(ipc_ns->mq_msgsize_max, DFLT_MSGSIZE);
 150                 if (attr) {
 151                         info->attr.mq_maxmsg = attr->mq_maxmsg;
 152                         info->attr.mq_msgsize = attr->mq_msgsize;
 153                 }
 154                 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 155                 info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 156                 if (!info->messages)
 157                         goto out_inode;
 158
 159                 mq_bytes = (mq_msg_tblsz +
 160                         (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 161
 162                 spin_lock(&mq_lock);
 163                 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 164                     u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
 165                         spin_unlock(&mq_lock);
 166                         /* mqueue_evict_inode() releases info->messages */
 167                         ret = -EMFILE;
 168                         goto out_inode;
 169                 }
 170                 u->mq_bytes += mq_bytes;
 171                 spin_unlock(&mq_lock);
 172
 173                 /* all is ok */
 174                 info->user = get_uid(u);
 175         } else if (S_ISDIR(mode)) {
 176                 inc_nlink(inode);
 177                 /* Some things misbehave if size == 0 on a directory */
 178                 inode->i_size = 2 * DIRENT_SIZE;
 179                 inode->i_op = &mqueue_dir_inode_operations;
 180                 inode->i_fop = &simple_dir_operations;
 181         }
 182
 183         return inode;
 184 out_inode:
 185         iput(inode);
 186 err:
 187         return ERR_PTR(ret);
 188 }
 189
 190 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 191 {
 192         struct inode *inode;
 193         struct ipc_namespace *ns = data;
 194
 195         sb->s_blocksize = PAGE_CACHE_SIZE;
 196         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 197         sb->s_magic = MQUEUE_MAGIC;
 198         sb->s_op = &mqueue_super_ops;
 199
 200         inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 201         if (IS_ERR(inode))
 202                 return PTR_ERR(inode);
 203
 204         sb->s_root = d_make_root(inode);
 205         if (!sb->s_root)
 206                 return -ENOMEM;
 207         return 0;
 208 }
 209
 210 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
 211                          int flags, const char *dev_name,
 212                          void *data)
 213 {
 214         if (!(flags & MS_KERNMOUNT))
 215                 data = current->nsproxy->ipc_ns;
 216         return mount_ns(fs_type, flags, data, mqueue_fill_super);
 217 }
 218
 219 static void init_once(void *foo)
 220 {
 221         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 222
 223         inode_init_once(&p->vfs_inode);
 224 }
 225
 226 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 227 {
 228         struct mqueue_inode_info *ei;
 229
 230         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 231         if (!ei)
 232                 return NULL;
 233         return &ei->vfs_inode;
 234 }
 235
 236 static void mqueue_i_callback(struct rcu_head *head)
 237 {
 238         struct inode *inode = container_of(head, struct inode, i_rcu);
 239         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 240 }
 241
 242 static void mqueue_destroy_inode(struct inode *inode)
 243 {
 244         call_rcu(&inode->i_rcu, mqueue_i_callback);
 245 }
 246
 247 static void mqueue_evict_inode(struct inode *inode)
 248 {
 249         struct mqueue_inode_info *info;
 250         struct user_struct *user;
 251         unsigned long mq_bytes;
 252         int i;
 253         struct ipc_namespace *ipc_ns;
 254
 255         clear_inode(inode);
 256
 257         if (S_ISDIR(inode->i_mode))
 258                 return;
 259
 260         ipc_ns = get_ns_from_inode(inode);
 261         info = MQUEUE_I(inode);
 262         spin_lock(&info->lock);
 263         for (i = 0; i < info->attr.mq_curmsgs; i++)
 264                 free_msg(info->messages[i]);
 265         kfree(info->messages);
 266         spin_unlock(&info->lock);
 267
 268         /* Total amount of bytes accounted for the mqueue */
 269         mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
 270             + info->attr.mq_msgsize);
 271         user = info->user;
 272         if (user) {
 273                 spin_lock(&mq_lock);
 274                 user->mq_bytes -= mq_bytes;
 275                 /*
 276                  * get_ns_from_inode() ensures that the
 277                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 278                  * to which we now hold a reference, or it is NULL.
 279                  * We can't put it here under mq_lock, though.
 280                  */
 281                 if (ipc_ns)
 282                         ipc_ns->mq_queues_count--;
 283                 spin_unlock(&mq_lock);
 284                 free_uid(user);
 285         }
 286         if (ipc_ns)
 287                 put_ipc_ns(ipc_ns);
 288 }
 289
 290 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 291                                 umode_t mode, struct nameidata *nd)
 292 {
 293         struct inode *inode;
 294         struct mq_attr *attr = dentry->d_fsdata;
 295         int error;
 296         struct ipc_namespace *ipc_ns;
 297
 298         spin_lock(&mq_lock);
 299         ipc_ns = __get_ns_from_inode(dir);
 300         if (!ipc_ns) {
 301                 error = -EACCES;
 302                 goto out_unlock;
 303         }
 304         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 305                         !capable(CAP_SYS_RESOURCE)) {
 306                 error = -ENOSPC;
 307                 goto out_unlock;
 308         }
 309         ipc_ns->mq_queues_count++;
 310         spin_unlock(&mq_lock);
 311
 312         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 313         if (IS_ERR(inode)) {
 314                 error = PTR_ERR(inode);
 315                 spin_lock(&mq_lock);
 316                 ipc_ns->mq_queues_count--;
 317                 goto out_unlock;
 318         }
 319
 320         put_ipc_ns(ipc_ns);
 321         dir->i_size += DIRENT_SIZE;
 322         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 323
 324         d_instantiate(dentry, inode);
 325         dget(dentry);
 326         return 0;
 327 out_unlock:
 328         spin_unlock(&mq_lock);
 329         if (ipc_ns)
 330                 put_ipc_ns(ipc_ns);
 331         return error;
 332 }
 333
 334 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 335 {
 336         struct inode *inode = dentry->d_inode;
 337
 338         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 339         dir->i_size -= DIRENT_SIZE;
 340         drop_nlink(inode);
 341         dput(dentry);
 342         return 0;
 343 }
 344
 345 /*
 346 *       This is routine for system read from queue file.
 347 *       To avoid mess with doing here some sort of mq_receive we allow
 348 *       to read only queue size & notification info (the only values
 349 *       that are interesting from user point of view and aren't accessible
 350 *       through std routines)
 351 */
 352 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 353                                 size_t count, loff_t *off)
 354 {
 355         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 356         char buffer[FILENT_SIZE];
 357         ssize_t ret;
 358
 359         spin_lock(&info->lock);
 360         snprintf(buffer, sizeof(buffer),
 361                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 362                         info->qsize,
 363                         info->notify_owner ? info->notify.sigev_notify : 0,
 364                         (info->notify_owner &&
 365                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 366                                 info->notify.sigev_signo : 0,
 367                         pid_vnr(info->notify_owner));
 368         spin_unlock(&info->lock);
 369         buffer[sizeof(buffer)-1] = '\0';
 370
 371         ret = simple_read_from_buffer(u_data, count, off, buffer,
 372                                 strlen(buffer));
 373         if (ret <= 0)
 374                 return ret;
 375
 376         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 377         return ret;
 378 }
 379
 380 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 381 {
 382         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 383
 384         spin_lock(&info->lock);
 385         if (task_tgid(current) == info->notify_owner)
 386                 remove_notification(info);
 387
 388         spin_unlock(&info->lock);
 389         return 0;
 390 }
 391
 392 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 393 {
 394         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 395         int retval = 0;
 396
 397         poll_wait(filp, &info->wait_q, poll_tab);
 398
 399         spin_lock(&info->lock);
 400         if (info->attr.mq_curmsgs)
 401                 retval = POLLIN | POLLRDNORM;
 402
 403         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 404                 retval |= POLLOUT | POLLWRNORM;
 405         spin_unlock(&info->lock);
 406
 407         return retval;
 408 }
 409
 410 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 411 static void wq_add(struct mqueue_inode_info *info, int sr,
 412                         struct ext_wait_queue *ewp)
 413 {
 414         struct ext_wait_queue *walk;
 415
 416         ewp->task = current;
 417
 418         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 419                 if (walk->task->static_prio <= current->static_prio) {
 420                         list_add_tail(&ewp->list, &walk->list);
 421                         return;
 422                 }
 423         }
 424         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 425 }
 426
 427 /*
 428  * Puts current task to sleep. Caller must hold queue lock. After return
 429  * lock isn't held.
 430  * sr: SEND or RECV
 431  */
 432 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 433                     ktime_t *timeout, struct ext_wait_queue *ewp)
 434 {
 435         int retval;
 436         signed long time;
 437
 438         wq_add(info, sr, ewp);
 439
 440         for (;;) {
 441                 set_current_state(TASK_INTERRUPTIBLE);
 442
 443                 spin_unlock(&info->lock);
 444                 time = schedule_hrtimeout_range_clock(timeout, 0,
 445                         HRTIMER_MODE_ABS, CLOCK_REALTIME);
 446
 447                 while (ewp->state == STATE_PENDING)
 448                         cpu_relax();
 449
 450                 if (ewp->state == STATE_READY) {
 451                         retval = 0;
 452                         goto out;
 453                 }
 454                 spin_lock(&info->lock);
 455                 if (ewp->state == STATE_READY) {
 456                         retval = 0;
 457                         goto out_unlock;
 458                 }
 459                 if (signal_pending(current)) {
 460                         retval = -ERESTARTSYS;
 461                         break;
 462                 }
 463                 if (time == 0) {
 464                         retval = -ETIMEDOUT;
 465                         break;
 466                 }
 467         }
 468         list_del(&ewp->list);
 469 out_unlock:
 470         spin_unlock(&info->lock);
 471 out:
 472         return retval;
 473 }
 474
 475 /*
 476  * Returns waiting task that should be serviced first or NULL if none exists
 477  */
 478 static struct ext_wait_queue *wq_get_first_waiter(
 479                 struct mqueue_inode_info *info, int sr)
 480 {
 481         struct list_head *ptr;
 482
 483         ptr = info->e_wait_q[sr].list.prev;
 484         if (ptr == &info->e_wait_q[sr].list)
 485                 return NULL;
 486         return list_entry(ptr, struct ext_wait_queue, list);
 487 }
 488
 489 /* Auxiliary functions to manipulate messages' list */
 490 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 491 {
 492         int k;
 493
 494         k = info->attr.mq_curmsgs - 1;
 495         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 496                 info->messages[k + 1] = info->messages[k];
 497                 k--;
 498         }
 499         info->attr.mq_curmsgs++;
 500         info->qsize += ptr->m_ts;
 501         info->messages[k + 1] = ptr;
 502 }
 503
 504 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 505 {
 506         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 507         return info->messages[info->attr.mq_curmsgs];
 508 }
 509
 510 static inline void set_cookie(struct sk_buff *skb, char code)
 511 {
 512         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 513 }
 514
 515 /*
 516  * The next function is only to split too long sys_mq_timedsend
 517  */
 518 static void __do_notify(struct mqueue_inode_info *info)
 519 {
 520         /* notification
 521          * invoked when there is registered process and there isn't process
 522          * waiting synchronously for message AND state of queue changed from
 523          * empty to not empty. Here we are sure that no one is waiting
 524          * synchronously. */
 525         if (info->notify_owner &&
 526             info->attr.mq_curmsgs == 1) {
 527                 struct siginfo sig_i;
 528                 switch (info->notify.sigev_notify) {
 529                 case SIGEV_NONE:
 530                         break;
 531                 case SIGEV_SIGNAL:
 532                         /* sends signal */
 533
 534                         sig_i.si_signo = info->notify.sigev_signo;
 535                         sig_i.si_errno = 0;
 536                         sig_i.si_code = SI_MESGQ;
 537                         sig_i.si_value = info->notify.sigev_value;
 538                         /* map current pid/uid into info->owner's namespaces */
 539                         rcu_read_lock();
 540                         sig_i.si_pid = task_tgid_nr_ns(current,
 541                                                 ns_of_pid(info->notify_owner));
 542                         sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
 543                         rcu_read_unlock();
 544
 545                         kill_pid_info(info->notify.sigev_signo,
 546                                       &sig_i, info->notify_owner);
 547                         break;
 548                 case SIGEV_THREAD:
 549                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 550                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 551                         break;
 552                 }
 553                 /* after notification unregisters process */
 554                 put_pid(info->notify_owner);
 555                 put_user_ns(info->notify_user_ns);
 556                 info->notify_owner = NULL;
 557                 info->notify_user_ns = NULL;
 558         }
 559         wake_up(&info->wait_q);
 560 }
 561
 562 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
 563                            ktime_t *expires, struct timespec *ts)
 564 {
 565         if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
 566                 return -EFAULT;
 567         if (!timespec_valid(ts))
 568                 return -EINVAL;
 569
 570         *expires = timespec_to_ktime(*ts);
 571         return 0;
 572 }
 573
 574 static void remove_notification(struct mqueue_inode_info *info)
 575 {
 576         if (info->notify_owner != NULL &&
 577             info->notify.sigev_notify == SIGEV_THREAD) {
 578                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 579                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 580         }
 581         put_pid(info->notify_owner);
 582         put_user_ns(info->notify_user_ns);
 583         info->notify_owner = NULL;
 584         info->notify_user_ns = NULL;
 585 }
 586
 587 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
 588 {
 589         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 590                 return 0;
 591         if (capable(CAP_SYS_RESOURCE)) {
 592                 if (attr->mq_maxmsg > HARD_MSGMAX)
 593                         return 0;
 594         } else {
 595                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
 596                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
 597                         return 0;
 598         }
 599         /* check for overflow */
 600         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 601                 return 0;
 602         if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
 603             + sizeof (struct msg_msg *))) <
 604             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 605                 return 0;
 606         return 1;
 607 }
 608
 609 /*
 610  * Invoked when creating a new queue via sys_mq_open
 611  */
 612 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
 613                         struct dentry *dentry, int oflag, umode_t mode,
 614                         struct mq_attr *attr)
 615 {
 616         const struct cred *cred = current_cred();
 617         struct file *result;
 618         int ret;
 619
 620         if (attr) {
 621                 if (!mq_attr_ok(ipc_ns, attr)) {
 622                         ret = -EINVAL;
 623                         goto out;
 624                 }
 625                 /* store for use during create */
 626                 dentry->d_fsdata = attr;
 627         }
 628
 629         mode &= ~current_umask();
 630         ret = mnt_want_write(ipc_ns->mq_mnt);
 631         if (ret)
 632                 goto out;
 633         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 634         dentry->d_fsdata = NULL;
 635         if (ret)
 636                 goto out_drop_write;
 637
 638         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 639         /*
 640          * dentry_open() took a persistent mnt_want_write(),
 641          * so we can now drop this one.
 642          */
 643         mnt_drop_write(ipc_ns->mq_mnt);
 644         return result;
 645
 646 out_drop_write:
 647         mnt_drop_write(ipc_ns->mq_mnt);
 648 out:
 649         dput(dentry);
 650         mntput(ipc_ns->mq_mnt);
 651         return ERR_PTR(ret);
 652 }
 653
 654 /* Opens existing queue */
 655 static struct file *do_open(struct ipc_namespace *ipc_ns,
 656                                 struct dentry *dentry, int oflag)
 657 {
 658         int ret;
 659         const struct cred *cred = current_cred();
 660
 661         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 662                                                   MAY_READ | MAY_WRITE };
 663
 664         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 665                 ret = -EINVAL;
 666                 goto err;
 667         }
 668
 669         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 670                 ret = -EACCES;
 671                 goto err;
 672         }
 673
 674         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 675
 676 err:
 677         dput(dentry);
 678         mntput(ipc_ns->mq_mnt);
 679         return ERR_PTR(ret);
 680 }
 681
 682 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
 683                 struct mq_attr __user *, u_attr)
 684 {
 685         struct dentry *dentry;
 686         struct file *filp;
 687         char *name;
 688         struct mq_attr attr;
 689         int fd, error;
 690         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 691
 692         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 693                 return -EFAULT;
 694
 695         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
 696
 697         if (IS_ERR(name = getname(u_name)))
 698                 return PTR_ERR(name);
 699
 700         fd = get_unused_fd_flags(O_CLOEXEC);
 701         if (fd < 0)
 702                 goto out_putname;
 703
 704         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 705         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 706         if (IS_ERR(dentry)) {
 707                 error = PTR_ERR(dentry);
 708                 goto out_putfd;
 709         }
 710         mntget(ipc_ns->mq_mnt);
 711
 712         if (oflag & O_CREAT) {
 713                 if (dentry->d_inode) {  /* entry already exists */
 714                         audit_inode(name, dentry);
 715                         if (oflag & O_EXCL) {
 716                                 error = -EEXIST;
 717                                 goto out;
 718                         }
 719                         filp = do_open(ipc_ns, dentry, oflag);
 720                 } else {
 721                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
 722                                                 dentry, oflag, mode,
 723                                                 u_attr ? &attr : NULL);
 724                 }
 725         } else {
 726                 if (!dentry->d_inode) {
 727                         error = -ENOENT;
 728                         goto out;
 729                 }
 730                 audit_inode(name, dentry);
 731                 filp = do_open(ipc_ns, dentry, oflag);
 732         }
 733
 734         if (IS_ERR(filp)) {
 735                 error = PTR_ERR(filp);
 736                 goto out_putfd;
 737         }
 738
 739         fd_install(fd, filp);
 740         goto out_upsem;
 741
 742 out:
 743         dput(dentry);
 744         mntput(ipc_ns->mq_mnt);
 745 out_putfd:
 746         put_unused_fd(fd);
 747         fd = error;
 748 out_upsem:
 749         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 750 out_putname:
 751         putname(name);
 752         return fd;
 753 }
 754
 755 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 756 {
 757         int err;
 758         char *name;
 759         struct dentry *dentry;
 760         struct inode *inode = NULL;
 761         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 762
 763         name = getname(u_name);
 764         if (IS_ERR(name))
 765                 return PTR_ERR(name);
 766
 767         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
 768                         I_MUTEX_PARENT);
 769         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 770         if (IS_ERR(dentry)) {
 771                 err = PTR_ERR(dentry);
 772                 goto out_unlock;
 773         }
 774
 775         if (!dentry->d_inode) {
 776                 err = -ENOENT;
 777                 goto out_err;
 778         }
 779
 780         inode = dentry->d_inode;
 781         if (inode)
 782                 ihold(inode);
 783         err = mnt_want_write(ipc_ns->mq_mnt);
 784         if (err)
 785                 goto out_err;
 786         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 787         mnt_drop_write(ipc_ns->mq_mnt);
 788 out_err:
 789         dput(dentry);
 790
 791 out_unlock:
 792         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 793         putname(name);
 794         if (inode)
 795                 iput(inode);
 796
 797         return err;
 798 }
 799
 800 /* Pipelined send and receive functions.
 801  *
 802  * If a receiver finds no waiting message, then it registers itself in the
 803  * list of waiting receivers. A sender checks that list before adding the new
 804  * message into the message array. If there is a waiting receiver, then it
 805  * bypasses the message array and directly hands the message over to the
 806  * receiver.
 807  * The receiver accepts the message and returns without grabbing the queue
 808  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 809  * are necessary. The same algorithm is used for sysv semaphores, see
 810  * ipc/sem.c for more details.
 811  *
 812  * The same algorithm is used for senders.
 813  */
 814
 815 /* pipelined_send() - send a message directly to the task waiting in
 816  * sys_mq_timedreceive() (without inserting message into a queue).
 817  */
 818 static inline void pipelined_send(struct mqueue_inode_info *info,
 819                                   struct msg_msg *message,
 820                                   struct ext_wait_queue *receiver)
 821 {
 822         receiver->msg = message;
 823         list_del(&receiver->list);
 824         receiver->state = STATE_PENDING;
 825         wake_up_process(receiver->task);
 826         smp_wmb();
 827         receiver->state = STATE_READY;
 828 }
 829
 830 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 831  * gets its message and put to the queue (we have one free place for sure). */
 832 static inline void pipelined_receive(struct mqueue_inode_info *info)
 833 {
 834         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 835
 836         if (!sender) {
 837                 /* for poll */
 838                 wake_up_interruptible(&info->wait_q);
 839                 return;
 840         }
 841         msg_insert(sender->msg, info);
 842         list_del(&sender->list);
 843         sender->state = STATE_PENDING;
 844         wake_up_process(sender->task);
 845         smp_wmb();
 846         sender->state = STATE_READY;
 847 }
 848
 849 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 850                 size_t, msg_len, unsigned int, msg_prio,
 851                 const struct timespec __user *, u_abs_timeout)
 852 {
 853         struct file *filp;
 854         struct inode *inode;
 855         struct ext_wait_queue wait;
 856         struct ext_wait_queue *receiver;
 857         struct msg_msg *msg_ptr;
 858         struct mqueue_inode_info *info;
 859         ktime_t expires, *timeout = NULL;
 860         struct timespec ts;
 861         int ret;
 862
 863         if (u_abs_timeout) {
 864                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 865                 if (res)
 866                         return res;
 867                 timeout = &expires;
 868         }
 869
 870         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 871                 return -EINVAL;
 872
 873         audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
 874
 875         filp = fget(mqdes);
 876         if (unlikely(!filp)) {
 877                 ret = -EBADF;
 878                 goto out;
 879         }
 880
 881         inode = filp->f_path.dentry->d_inode;
 882         if (unlikely(filp->f_op != &mqueue_file_operations)) {
 883                 ret = -EBADF;
 884                 goto out_fput;
 885         }
 886         info = MQUEUE_I(inode);
 887         audit_inode(NULL, filp->f_path.dentry);
 888
 889         if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
 890                 ret = -EBADF;
 891                 goto out_fput;
 892         }
 893
 894         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 895                 ret = -EMSGSIZE;
 896                 goto out_fput;
 897         }
 898
 899         /* First try to allocate memory, before doing anything with
 900          * existing queues. */
 901         msg_ptr = load_msg(u_msg_ptr, msg_len);
 902         if (IS_ERR(msg_ptr)) {
 903                 ret = PTR_ERR(msg_ptr);
 904                 goto out_fput;
 905         }
 906         msg_ptr->m_ts = msg_len;
 907         msg_ptr->m_type = msg_prio;
 908
 909         spin_lock(&info->lock);
 910
 911         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 912                 if (filp->f_flags & O_NONBLOCK) {
 913                         spin_unlock(&info->lock);
 914                         ret = -EAGAIN;
 915                 } else {
 916                         wait.task = current;
 917                         wait.msg = (void *) msg_ptr;
 918                         wait.state = STATE_NONE;
 919                         ret = wq_sleep(info, SEND, timeout, &wait);
 920                 }
 921                 if (ret < 0)
 922                         free_msg(msg_ptr);
 923         } else {
 924                 receiver = wq_get_first_waiter(info, RECV);
 925                 if (receiver) {
 926                         pipelined_send(info, msg_ptr, receiver);
 927                 } else {
 928                         /* adds message to the queue */
 929                         msg_insert(msg_ptr, info);
 930                         __do_notify(info);
 931                 }
 932                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 933                                 CURRENT_TIME;
 934                 spin_unlock(&info->lock);
 935                 ret = 0;
 936         }
 937 out_fput:
 938         fput(filp);
 939 out:
 940         return ret;
 941 }
 942
 943 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
 944                 size_t, msg_len, unsigned int __user *, u_msg_prio,
 945                 const struct timespec __user *, u_abs_timeout)
 946 {
 947         ssize_t ret;
 948         struct msg_msg *msg_ptr;
 949         struct file *filp;
 950         struct inode *inode;
 951         struct mqueue_inode_info *info;
 952         struct ext_wait_queue wait;
 953         ktime_t expires, *timeout = NULL;
 954         struct timespec ts;
 955
 956         if (u_abs_timeout) {
 957                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 958                 if (res)
 959                         return res;
 960                 timeout = &expires;
 961         }
 962
 963         audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
 964
 965         filp = fget(mqdes);
 966         if (unlikely(!filp)) {
 967                 ret = -EBADF;
 968                 goto out;
 969         }
 970
 971         inode = filp->f_path.dentry->d_inode;
 972         if (unlikely(filp->f_op != &mqueue_file_operations)) {
 973                 ret = -EBADF;
 974                 goto out_fput;
 975         }
 976         info = MQUEUE_I(inode);
 977         audit_inode(NULL, filp->f_path.dentry);
 978
 979         if (unlikely(!(filp->f_mode & FMODE_READ))) {
 980                 ret = -EBADF;
 981                 goto out_fput;
 982         }
 983
 984         /* checks if buffer is big enough */
 985         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 986                 ret = -EMSGSIZE;
 987                 goto out_fput;
 988         }
 989
 990         spin_lock(&info->lock);
 991         if (info->attr.mq_curmsgs == 0) {
 992                 if (filp->f_flags & O_NONBLOCK) {
 993                         spin_unlock(&info->lock);
 994                         ret = -EAGAIN;
 995                 } else {
 996                         wait.task = current;
 997                         wait.state = STATE_NONE;
 998                         ret = wq_sleep(info, RECV, timeout, &wait);
 999                         msg_ptr = wait.msg;
1000                 }
1001         } else {
1002                 msg_ptr = msg_get(info);
1003
1004                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1005                                 CURRENT_TIME;
1006
1007                 /* There is now free space in queue. */
1008                 pipelined_receive(info);
1009                 spin_unlock(&info->lock);
1010                 ret = 0;
1011         }
1012         if (ret == 0) {
1013                 ret = msg_ptr->m_ts;
1014
1015                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1016                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1017                         ret = -EFAULT;
1018                 }
1019                 free_msg(msg_ptr);
1020         }
1021 out_fput:
1022         fput(filp);
1023 out:
1024         return ret;
1025 }
1026
1027 /*
1028  * Notes: the case when user wants us to deregister (with NULL as pointer)
1029  * and he isn't currently owner of notification, will be silently discarded.
1030  * It isn't explicitly defined in the POSIX.
1031  */
1032 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1033                 const struct sigevent __user *, u_notification)
1034 {
1035         int ret;
1036         struct file *filp;
1037         struct sock *sock;
1038         struct inode *inode;
1039         struct sigevent notification;
1040         struct mqueue_inode_info *info;
1041         struct sk_buff *nc;
1042
1043         if (u_notification) {
1044                 if (copy_from_user(&notification, u_notification,
1045                                         sizeof(struct sigevent)))
1046                         return -EFAULT;
1047         }
1048
1049         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1050
1051         nc = NULL;
1052         sock = NULL;
1053         if (u_notification != NULL) {
1054                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1055                              notification.sigev_notify != SIGEV_SIGNAL &&
1056                              notification.sigev_notify != SIGEV_THREAD))
1057                         return -EINVAL;
1058                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1059                         !valid_signal(notification.sigev_signo)) {
1060                         return -EINVAL;
1061                 }
1062                 if (notification.sigev_notify == SIGEV_THREAD) {
1063                         long timeo;
1064
1065                         /* create the notify skb */
1066                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1067                         if (!nc) {
1068                                 ret = -ENOMEM;
1069                                 goto out;
1070                         }
1071                         if (copy_from_user(nc->data,
1072                                         notification.sigev_value.sival_ptr,
1073                                         NOTIFY_COOKIE_LEN)) {
1074                                 ret = -EFAULT;
1075                                 goto out;
1076                         }
1077
1078                         /* TODO: add a header? */
1079                         skb_put(nc, NOTIFY_COOKIE_LEN);
1080                         /* and attach it to the socket */
1081 retry:
1082                         filp = fget(notification.sigev_signo);
1083                         if (!filp) {
1084                                 ret = -EBADF;
1085                                 goto out;
1086                         }
1087                         sock = netlink_getsockbyfilp(filp);
1088                         fput(filp);
1089                         if (IS_ERR(sock)) {
1090                                 ret = PTR_ERR(sock);
1091                                 sock = NULL;
1092                                 goto out;
1093                         }
1094
1095                         timeo = MAX_SCHEDULE_TIMEOUT;
1096                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1097                         if (ret == 1)
1098                                 goto retry;
1099                         if (ret) {
1100                                 sock = NULL;
1101                                 nc = NULL;
1102                                 goto out;
1103                         }
1104                 }
1105         }
1106
1107         filp = fget(mqdes);
1108         if (!filp) {
1109                 ret = -EBADF;
1110                 goto out;
1111         }
1112
1113         inode = filp->f_path.dentry->d_inode;
1114         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1115                 ret = -EBADF;
1116                 goto out_fput;
1117         }
1118         info = MQUEUE_I(inode);
1119
1120         ret = 0;
1121         spin_lock(&info->lock);
1122         if (u_notification == NULL) {
1123                 if (info->notify_owner == task_tgid(current)) {
1124                         remove_notification(info);
1125                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1126                 }
1127         } else if (info->notify_owner != NULL) {
1128                 ret = -EBUSY;
1129         } else {
1130                 switch (notification.sigev_notify) {
1131                 case SIGEV_NONE:
1132                         info->notify.sigev_notify = SIGEV_NONE;
1133                         break;
1134                 case SIGEV_THREAD:
1135                         info->notify_sock = sock;
1136                         info->notify_cookie = nc;
1137                         sock = NULL;
1138                         nc = NULL;
1139                         info->notify.sigev_notify = SIGEV_THREAD;
1140                         break;
1141                 case SIGEV_SIGNAL:
1142                         info->notify.sigev_signo = notification.sigev_signo;
1143                         info->notify.sigev_value = notification.sigev_value;
1144                         info->notify.sigev_notify = SIGEV_SIGNAL;
1145                         break;
1146                 }
1147
1148                 info->notify_owner = get_pid(task_tgid(current));
1149                 info->notify_user_ns = get_user_ns(current_user_ns());
1150                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1151         }
1152         spin_unlock(&info->lock);
1153 out_fput:
1154         fput(filp);
1155 out:
1156         if (sock) {
1157                 netlink_detachskb(sock, nc);
1158         } else if (nc) {
1159                 dev_kfree_skb(nc);
1160         }
1161         return ret;
1162 }
1163
1164 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1165                 const struct mq_attr __user *, u_mqstat,
1166                 struct mq_attr __user *, u_omqstat)
1167 {
1168         int ret;
1169         struct mq_attr mqstat, omqstat;
1170         struct file *filp;
1171         struct inode *inode;
1172         struct mqueue_inode_info *info;
1173
1174         if (u_mqstat != NULL) {
1175                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1176                         return -EFAULT;
1177                 if (mqstat.mq_flags & (~O_NONBLOCK))
1178                         return -EINVAL;
1179         }
1180
1181         filp = fget(mqdes);
1182         if (!filp) {
1183                 ret = -EBADF;
1184                 goto out;
1185         }
1186
1187         inode = filp->f_path.dentry->d_inode;
1188         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1189                 ret = -EBADF;
1190                 goto out_fput;
1191         }
1192         info = MQUEUE_I(inode);
1193
1194         spin_lock(&info->lock);
1195
1196         omqstat = info->attr;
1197         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1198         if (u_mqstat) {
1199                 audit_mq_getsetattr(mqdes, &mqstat);
1200                 spin_lock(&filp->f_lock);
1201                 if (mqstat.mq_flags & O_NONBLOCK)
1202                         filp->f_flags |= O_NONBLOCK;
1203                 else
1204                         filp->f_flags &= ~O_NONBLOCK;
1205                 spin_unlock(&filp->f_lock);
1206
1207                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1208         }
1209
1210         spin_unlock(&info->lock);
1211
1212         ret = 0;
1213         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1214                                                 sizeof(struct mq_attr)))
1215                 ret = -EFAULT;
1216
1217 out_fput:
1218         fput(filp);
1219 out:
1220         return ret;
1221 }
1222
1223 static const struct inode_operations mqueue_dir_inode_operations = {
1224         .lookup = simple_lookup,
1225         .create = mqueue_create,
1226         .unlink = mqueue_unlink,
1227 };
1228
1229 static const struct file_operations mqueue_file_operations = {
1230         .flush = mqueue_flush_file,
1231         .poll = mqueue_poll_file,
1232         .read = mqueue_read_file,
1233         .llseek = default_llseek,
1234 };
1235
1236 static const struct super_operations mqueue_super_ops = {
1237         .alloc_inode = mqueue_alloc_inode,
1238         .destroy_inode = mqueue_destroy_inode,
1239         .evict_inode = mqueue_evict_inode,
1240         .statfs = simple_statfs,
1241 };
1242
1243 static struct file_system_type mqueue_fs_type = {
1244         .name = "mqueue",
1245         .mount = mqueue_mount,
1246         .kill_sb = kill_litter_super,
1247 };
1248
1249 int mq_init_ns(struct ipc_namespace *ns)
1250 {
1251         ns->mq_queues_count  = 0;
1252         ns->mq_queues_max    = DFLT_QUEUESMAX;
1253         ns->mq_msg_max       = DFLT_MSGMAX;
1254         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1255
1256         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1257         if (IS_ERR(ns->mq_mnt)) {
1258                 int err = PTR_ERR(ns->mq_mnt);
1259                 ns->mq_mnt = NULL;
1260                 return err;
1261         }
1262         return 0;
1263 }
1264
1265 void mq_clear_sbinfo(struct ipc_namespace *ns)
1266 {
1267         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1268 }
1269
1270 void mq_put_mnt(struct ipc_namespace *ns)
1271 {
1272         kern_unmount(ns->mq_mnt);
1273 }
1274
1275 static int __init init_mqueue_fs(void)
1276 {
1277         int error;
1278
1279         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1280                                 sizeof(struct mqueue_inode_info), 0,
1281                                 SLAB_HWCACHE_ALIGN, init_once);
1282         if (mqueue_inode_cachep == NULL)
1283                 return -ENOMEM;
1284
1285         /* ignore failures - they are not fatal */
1286         mq_sysctl_table = mq_register_sysctl_table();
1287
1288         error = register_filesystem(&mqueue_fs_type);
1289         if (error)
1290                 goto out_sysctl;
1291
1292         spin_lock_init(&mq_lock);
1293
1294         error = mq_init_ns(&init_ipc_ns);
1295         if (error)
1296                 goto out_filesystem;
1297
1298         return 0;
1299
1300 out_filesystem:
1301         unregister_filesystem(&mqueue_fs_type);
1302 out_sysctl:
1303         if (mq_sysctl_table)
1304                 unregister_sysctl_table(mq_sysctl_table);
1305         kmem_cache_destroy(mqueue_inode_cachep);
1306         return error;
1307 }
1308
1309 __initcall(init_mqueue_fs);