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/vmalloc.h>
  28 #include <linux/netlink.h>
  29 #include <linux/syscalls.h>
  30 #include <linux/audit.h>
  31 #include <linux/signal.h>
  32 #include <linux/mutex.h>
  33 #include <linux/nsproxy.h>
  34 #include <linux/pid.h>
  35 #include <linux/ipc_namespace.h>
  36 #include <linux/user_namespace.h>
  37 #include <linux/slab.h>
  38
  39 #include <net/sock.h>
  40 #include "util.h"
  41
  42 #define MQUEUE_MAGIC    0x19800202
  43 #define DIRENT_SIZE     20
  44 #define FILENT_SIZE     80
  45
  46 #define SEND            0
  47 #define RECV            1
  48
  49 #define STATE_NONE      0
  50 #define STATE_PENDING   1
  51 #define STATE_READY     2
  52
  53 struct ext_wait_queue {         /* queue of sleeping tasks */
  54         struct task_struct *task;
  55         struct list_head list;
  56         struct msg_msg *msg;    /* ptr of loaded message */
  57         int state;              /* one of STATE_* values */
  58 };
  59
  60 struct mqueue_inode_info {
  61         spinlock_t lock;
  62         struct inode vfs_inode;
  63         wait_queue_head_t wait_q;
  64
  65         struct msg_msg **messages;
  66         struct mq_attr attr;
  67
  68         struct sigevent notify;
  69         struct pid* notify_owner;
  70         struct user_namespace *notify_user_ns;
  71         struct user_struct *user;       /* user who created, for accounting */
  72         struct sock *notify_sock;
  73         struct sk_buff *notify_cookie;
  74
  75         /* for tasks waiting for free space and messages, respectively */
  76         struct ext_wait_queue e_wait_q[2];
  77
  78         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  79 };
  80
  81 static const struct inode_operations mqueue_dir_inode_operations;
  82 static const struct file_operations mqueue_file_operations;
  83 static const struct super_operations mqueue_super_ops;
  84 static void remove_notification(struct mqueue_inode_info *info);
  85
  86 static struct kmem_cache *mqueue_inode_cachep;
  87
  88 static struct ctl_table_header * mq_sysctl_table;
  89
  90 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  91 {
  92         return container_of(inode, struct mqueue_inode_info, vfs_inode);
  93 }
  94
  95 /*
  96  * This routine should be called with the mq_lock held.
  97  */
  98 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
  99 {
 100         return get_ipc_ns(inode->i_sb->s_fs_info);
 101 }
 102
 103 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 104 {
 105         struct ipc_namespace *ns;
 106
 107         spin_lock(&mq_lock);
 108         ns = __get_ns_from_inode(inode);
 109         spin_unlock(&mq_lock);
 110         return ns;
 111 }
 112
 113 static struct inode *mqueue_get_inode(struct super_block *sb,
 114                 struct ipc_namespace *ipc_ns, umode_t mode,
 115                 struct mq_attr *attr)
 116 {
 117         struct user_struct *u = current_user();
 118         struct inode *inode;
 119         int ret = -ENOMEM;
 120
 121         inode = new_inode(sb);
 122         if (!inode)
 123                 goto err;
 124
 125         inode->i_ino = get_next_ino();
 126         inode->i_mode = mode;
 127         inode->i_uid = current_fsuid();
 128         inode->i_gid = current_fsgid();
 129         inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
 130
 131         if (S_ISREG(mode)) {
 132                 struct mqueue_inode_info *info;
 133                 unsigned long mq_bytes, mq_msg_tblsz;
 134
 135                 inode->i_fop = &mqueue_file_operations;
 136                 inode->i_size = FILENT_SIZE;
 137                 /* mqueue specific info */
 138                 info = MQUEUE_I(inode);
 139                 spin_lock_init(&info->lock);
 140                 init_waitqueue_head(&info->wait_q);
 141                 INIT_LIST_HEAD(&info->e_wait_q[0].list);
 142                 INIT_LIST_HEAD(&info->e_wait_q[1].list);
 143                 info->notify_owner = NULL;
 144                 info->notify_user_ns = NULL;
 145                 info->qsize = 0;
 146                 info->user = NULL;      /* set when all is ok */
 147                 memset(&info->attr, 0, sizeof(info->attr));
 148                 info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max, DFLT_MSG);
 149                 info->attr.mq_msgsize =
 150                         min(ipc_ns->mq_msgsize_max, DFLT_MSGSIZE);
 151                 if (attr) {
 152                         info->attr.mq_maxmsg = attr->mq_maxmsg;
 153                         info->attr.mq_msgsize = attr->mq_msgsize;
 154                 }
 155                 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 156                 if (mq_msg_tblsz > PAGE_SIZE)
 157                         info->messages = vmalloc(mq_msg_tblsz);
 158                 else
 159                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 160                 if (!info->messages)
 161                         goto out_inode;
 162
 163                 mq_bytes = (mq_msg_tblsz +
 164                         (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 165
 166                 spin_lock(&mq_lock);
 167                 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 168                     u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
 169                         spin_unlock(&mq_lock);
 170                         /* mqueue_evict_inode() releases info->messages */
 171                         ret = -EMFILE;
 172                         goto out_inode;
 173                 }
 174                 u->mq_bytes += mq_bytes;
 175                 spin_unlock(&mq_lock);
 176
 177                 /* all is ok */
 178                 info->user = get_uid(u);
 179         } else if (S_ISDIR(mode)) {
 180                 inc_nlink(inode);
 181                 /* Some things misbehave if size == 0 on a directory */
 182                 inode->i_size = 2 * DIRENT_SIZE;
 183                 inode->i_op = &mqueue_dir_inode_operations;
 184                 inode->i_fop = &simple_dir_operations;
 185         }
 186
 187         return inode;
 188 out_inode:
 189         iput(inode);
 190 err:
 191         return ERR_PTR(ret);
 192 }
 193
 194 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 195 {
 196         struct inode *inode;
 197         struct ipc_namespace *ns = data;
 198
 199         sb->s_blocksize = PAGE_CACHE_SIZE;
 200         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 201         sb->s_magic = MQUEUE_MAGIC;
 202         sb->s_op = &mqueue_super_ops;
 203
 204         inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 205         if (IS_ERR(inode))
 206                 return PTR_ERR(inode);
 207
 208         sb->s_root = d_make_root(inode);
 209         if (!sb->s_root)
 210                 return -ENOMEM;
 211         return 0;
 212 }
 213
 214 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
 215                          int flags, const char *dev_name,
 216                          void *data)
 217 {
 218         if (!(flags & MS_KERNMOUNT))
 219                 data = current->nsproxy->ipc_ns;
 220         return mount_ns(fs_type, flags, data, mqueue_fill_super);
 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_i_callback(struct rcu_head *head)
 241 {
 242         struct inode *inode = container_of(head, struct inode, i_rcu);
 243         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 244 }
 245
 246 static void mqueue_destroy_inode(struct inode *inode)
 247 {
 248         call_rcu(&inode->i_rcu, mqueue_i_callback);
 249 }
 250
 251 static void mqueue_evict_inode(struct inode *inode)
 252 {
 253         struct mqueue_inode_info *info;
 254         struct user_struct *user;
 255         unsigned long mq_bytes;
 256         int i;
 257         struct ipc_namespace *ipc_ns;
 258
 259         clear_inode(inode);
 260
 261         if (S_ISDIR(inode->i_mode))
 262                 return;
 263
 264         ipc_ns = get_ns_from_inode(inode);
 265         info = MQUEUE_I(inode);
 266         spin_lock(&info->lock);
 267         for (i = 0; i < info->attr.mq_curmsgs; i++)
 268                 free_msg(info->messages[i]);
 269         if (is_vmalloc_addr(info->messages))
 270                 vfree(info->messages);
 271         else
 272                 kfree(info->messages);
 273         spin_unlock(&info->lock);
 274
 275         /* Total amount of bytes accounted for the mqueue */
 276         mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
 277             + info->attr.mq_msgsize);
 278         user = info->user;
 279         if (user) {
 280                 spin_lock(&mq_lock);
 281                 user->mq_bytes -= mq_bytes;
 282                 /*
 283                  * get_ns_from_inode() ensures that the
 284                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 285                  * to which we now hold a reference, or it is NULL.
 286                  * We can't put it here under mq_lock, though.
 287                  */
 288                 if (ipc_ns)
 289                         ipc_ns->mq_queues_count--;
 290                 spin_unlock(&mq_lock);
 291                 free_uid(user);
 292         }
 293         if (ipc_ns)
 294                 put_ipc_ns(ipc_ns);
 295 }
 296
 297 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 298                                 umode_t mode, struct nameidata *nd)
 299 {
 300         struct inode *inode;
 301         struct mq_attr *attr = dentry->d_fsdata;
 302         int error;
 303         struct ipc_namespace *ipc_ns;
 304
 305         spin_lock(&mq_lock);
 306         ipc_ns = __get_ns_from_inode(dir);
 307         if (!ipc_ns) {
 308                 error = -EACCES;
 309                 goto out_unlock;
 310         }
 311         if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX ||
 312             (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 313              !capable(CAP_SYS_RESOURCE))) {
 314                 error = -ENOSPC;
 315                 goto out_unlock;
 316         }
 317         ipc_ns->mq_queues_count++;
 318         spin_unlock(&mq_lock);
 319
 320         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 321         if (IS_ERR(inode)) {
 322                 error = PTR_ERR(inode);
 323                 spin_lock(&mq_lock);
 324                 ipc_ns->mq_queues_count--;
 325                 goto out_unlock;
 326         }
 327
 328         put_ipc_ns(ipc_ns);
 329         dir->i_size += DIRENT_SIZE;
 330         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 331
 332         d_instantiate(dentry, inode);
 333         dget(dentry);
 334         return 0;
 335 out_unlock:
 336         spin_unlock(&mq_lock);
 337         if (ipc_ns)
 338                 put_ipc_ns(ipc_ns);
 339         return error;
 340 }
 341
 342 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 343 {
 344         struct inode *inode = dentry->d_inode;
 345
 346         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 347         dir->i_size -= DIRENT_SIZE;
 348         drop_nlink(inode);
 349         dput(dentry);
 350         return 0;
 351 }
 352
 353 /*
 354 *       This is routine for system read from queue file.
 355 *       To avoid mess with doing here some sort of mq_receive we allow
 356 *       to read only queue size & notification info (the only values
 357 *       that are interesting from user point of view and aren't accessible
 358 *       through std routines)
 359 */
 360 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 361                                 size_t count, loff_t *off)
 362 {
 363         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 364         char buffer[FILENT_SIZE];
 365         ssize_t ret;
 366
 367         spin_lock(&info->lock);
 368         snprintf(buffer, sizeof(buffer),
 369                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 370                         info->qsize,
 371                         info->notify_owner ? info->notify.sigev_notify : 0,
 372                         (info->notify_owner &&
 373                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 374                                 info->notify.sigev_signo : 0,
 375                         pid_vnr(info->notify_owner));
 376         spin_unlock(&info->lock);
 377         buffer[sizeof(buffer)-1] = '\0';
 378
 379         ret = simple_read_from_buffer(u_data, count, off, buffer,
 380                                 strlen(buffer));
 381         if (ret <= 0)
 382                 return ret;
 383
 384         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 385         return ret;
 386 }
 387
 388 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 389 {
 390         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 391
 392         spin_lock(&info->lock);
 393         if (task_tgid(current) == info->notify_owner)
 394                 remove_notification(info);
 395
 396         spin_unlock(&info->lock);
 397         return 0;
 398 }
 399
 400 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 401 {
 402         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 403         int retval = 0;
 404
 405         poll_wait(filp, &info->wait_q, poll_tab);
 406
 407         spin_lock(&info->lock);
 408         if (info->attr.mq_curmsgs)
 409                 retval = POLLIN | POLLRDNORM;
 410
 411         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 412                 retval |= POLLOUT | POLLWRNORM;
 413         spin_unlock(&info->lock);
 414
 415         return retval;
 416 }
 417
 418 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 419 static void wq_add(struct mqueue_inode_info *info, int sr,
 420                         struct ext_wait_queue *ewp)
 421 {
 422         struct ext_wait_queue *walk;
 423
 424         ewp->task = current;
 425
 426         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 427                 if (walk->task->static_prio <= current->static_prio) {
 428                         list_add_tail(&ewp->list, &walk->list);
 429                         return;
 430                 }
 431         }
 432         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 433 }
 434
 435 /*
 436  * Puts current task to sleep. Caller must hold queue lock. After return
 437  * lock isn't held.
 438  * sr: SEND or RECV
 439  */
 440 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 441                     ktime_t *timeout, struct ext_wait_queue *ewp)
 442 {
 443         int retval;
 444         signed long time;
 445
 446         wq_add(info, sr, ewp);
 447
 448         for (;;) {
 449                 set_current_state(TASK_INTERRUPTIBLE);
 450
 451                 spin_unlock(&info->lock);
 452                 time = schedule_hrtimeout_range_clock(timeout, 0,
 453                         HRTIMER_MODE_ABS, CLOCK_REALTIME);
 454
 455                 while (ewp->state == STATE_PENDING)
 456                         cpu_relax();
 457
 458                 if (ewp->state == STATE_READY) {
 459                         retval = 0;
 460                         goto out;
 461                 }
 462                 spin_lock(&info->lock);
 463                 if (ewp->state == STATE_READY) {
 464                         retval = 0;
 465                         goto out_unlock;
 466                 }
 467                 if (signal_pending(current)) {
 468                         retval = -ERESTARTSYS;
 469                         break;
 470                 }
 471                 if (time == 0) {
 472                         retval = -ETIMEDOUT;
 473                         break;
 474                 }
 475         }
 476         list_del(&ewp->list);
 477 out_unlock:
 478         spin_unlock(&info->lock);
 479 out:
 480         return retval;
 481 }
 482
 483 /*
 484  * Returns waiting task that should be serviced first or NULL if none exists
 485  */
 486 static struct ext_wait_queue *wq_get_first_waiter(
 487                 struct mqueue_inode_info *info, int sr)
 488 {
 489         struct list_head *ptr;
 490
 491         ptr = info->e_wait_q[sr].list.prev;
 492         if (ptr == &info->e_wait_q[sr].list)
 493                 return NULL;
 494         return list_entry(ptr, struct ext_wait_queue, list);
 495 }
 496
 497 /* Auxiliary functions to manipulate messages' list */
 498 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 499 {
 500         int k;
 501
 502         k = info->attr.mq_curmsgs - 1;
 503         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 504                 info->messages[k + 1] = info->messages[k];
 505                 k--;
 506         }
 507         info->attr.mq_curmsgs++;
 508         info->qsize += ptr->m_ts;
 509         info->messages[k + 1] = ptr;
 510 }
 511
 512 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 513 {
 514         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 515         return info->messages[info->attr.mq_curmsgs];
 516 }
 517
 518 static inline void set_cookie(struct sk_buff *skb, char code)
 519 {
 520         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 521 }
 522
 523 /*
 524  * The next function is only to split too long sys_mq_timedsend
 525  */
 526 static void __do_notify(struct mqueue_inode_info *info)
 527 {
 528         /* notification
 529          * invoked when there is registered process and there isn't process
 530          * waiting synchronously for message AND state of queue changed from
 531          * empty to not empty. Here we are sure that no one is waiting
 532          * synchronously. */
 533         if (info->notify_owner &&
 534             info->attr.mq_curmsgs == 1) {
 535                 struct siginfo sig_i;
 536                 switch (info->notify.sigev_notify) {
 537                 case SIGEV_NONE:
 538                         break;
 539                 case SIGEV_SIGNAL:
 540                         /* sends signal */
 541
 542                         sig_i.si_signo = info->notify.sigev_signo;
 543                         sig_i.si_errno = 0;
 544                         sig_i.si_code = SI_MESGQ;
 545                         sig_i.si_value = info->notify.sigev_value;
 546                         /* map current pid/uid into info->owner's namespaces */
 547                         rcu_read_lock();
 548                         sig_i.si_pid = task_tgid_nr_ns(current,
 549                                                 ns_of_pid(info->notify_owner));
 550                         sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
 551                         rcu_read_unlock();
 552
 553                         kill_pid_info(info->notify.sigev_signo,
 554                                       &sig_i, info->notify_owner);
 555                         break;
 556                 case SIGEV_THREAD:
 557                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 558                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 559                         break;
 560                 }
 561                 /* after notification unregisters process */
 562                 put_pid(info->notify_owner);
 563                 put_user_ns(info->notify_user_ns);
 564                 info->notify_owner = NULL;
 565                 info->notify_user_ns = NULL;
 566         }
 567         wake_up(&info->wait_q);
 568 }
 569
 570 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
 571                            ktime_t *expires, struct timespec *ts)
 572 {
 573         if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
 574                 return -EFAULT;
 575         if (!timespec_valid(ts))
 576                 return -EINVAL;
 577
 578         *expires = timespec_to_ktime(*ts);
 579         return 0;
 580 }
 581
 582 static void remove_notification(struct mqueue_inode_info *info)
 583 {
 584         if (info->notify_owner != NULL &&
 585             info->notify.sigev_notify == SIGEV_THREAD) {
 586                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 587                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 588         }
 589         put_pid(info->notify_owner);
 590         put_user_ns(info->notify_user_ns);
 591         info->notify_owner = NULL;
 592         info->notify_user_ns = NULL;
 593 }
 594
 595 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
 596 {
 597         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 598                 return 0;
 599         if (capable(CAP_SYS_RESOURCE)) {
 600                 if (attr->mq_maxmsg > HARD_MSGMAX ||
 601                     attr->mq_msgsize > HARD_MSGSIZEMAX)
 602                         return 0;
 603         } else {
 604                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
 605                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
 606                         return 0;
 607         }
 608         /* check for overflow */
 609         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 610                 return 0;
 611         if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
 612             + sizeof (struct msg_msg *))) <
 613             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 614                 return 0;
 615         return 1;
 616 }
 617
 618 /*
 619  * Invoked when creating a new queue via sys_mq_open
 620  */
 621 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
 622                         struct dentry *dentry, int oflag, umode_t mode,
 623                         struct mq_attr *attr)
 624 {
 625         const struct cred *cred = current_cred();
 626         struct file *result;
 627         int ret;
 628
 629         if (attr) {
 630                 if (!mq_attr_ok(ipc_ns, attr)) {
 631                         ret = -EINVAL;
 632                         goto out;
 633                 }
 634                 /* store for use during create */
 635                 dentry->d_fsdata = attr;
 636         }
 637
 638         mode &= ~current_umask();
 639         ret = mnt_want_write(ipc_ns->mq_mnt);
 640         if (ret)
 641                 goto out;
 642         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 643         dentry->d_fsdata = NULL;
 644         if (ret)
 645                 goto out_drop_write;
 646
 647         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 648         /*
 649          * dentry_open() took a persistent mnt_want_write(),
 650          * so we can now drop this one.
 651          */
 652         mnt_drop_write(ipc_ns->mq_mnt);
 653         return result;
 654
 655 out_drop_write:
 656         mnt_drop_write(ipc_ns->mq_mnt);
 657 out:
 658         dput(dentry);
 659         mntput(ipc_ns->mq_mnt);
 660         return ERR_PTR(ret);
 661 }
 662
 663 /* Opens existing queue */
 664 static struct file *do_open(struct ipc_namespace *ipc_ns,
 665                                 struct dentry *dentry, int oflag)
 666 {
 667         int ret;
 668         const struct cred *cred = current_cred();
 669
 670         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 671                                                   MAY_READ | MAY_WRITE };
 672
 673         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 674                 ret = -EINVAL;
 675                 goto err;
 676         }
 677
 678         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 679                 ret = -EACCES;
 680                 goto err;
 681         }
 682
 683         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 684
 685 err:
 686         dput(dentry);
 687         mntput(ipc_ns->mq_mnt);
 688         return ERR_PTR(ret);
 689 }
 690
 691 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
 692                 struct mq_attr __user *, u_attr)
 693 {
 694         struct dentry *dentry;
 695         struct file *filp;
 696         char *name;
 697         struct mq_attr attr;
 698         int fd, error;
 699         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 700
 701         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 702                 return -EFAULT;
 703
 704         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
 705
 706         if (IS_ERR(name = getname(u_name)))
 707                 return PTR_ERR(name);
 708
 709         fd = get_unused_fd_flags(O_CLOEXEC);
 710         if (fd < 0)
 711                 goto out_putname;
 712
 713         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 714         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 715         if (IS_ERR(dentry)) {
 716                 error = PTR_ERR(dentry);
 717                 goto out_putfd;
 718         }
 719         mntget(ipc_ns->mq_mnt);
 720
 721         if (oflag & O_CREAT) {
 722                 if (dentry->d_inode) {  /* entry already exists */
 723                         audit_inode(name, dentry);
 724                         if (oflag & O_EXCL) {
 725                                 error = -EEXIST;
 726                                 goto out;
 727                         }
 728                         filp = do_open(ipc_ns, dentry, oflag);
 729                 } else {
 730                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
 731                                                 dentry, oflag, mode,
 732                                                 u_attr ? &attr : NULL);
 733                 }
 734         } else {
 735                 if (!dentry->d_inode) {
 736                         error = -ENOENT;
 737                         goto out;
 738                 }
 739                 audit_inode(name, dentry);
 740                 filp = do_open(ipc_ns, dentry, oflag);
 741         }
 742
 743         if (IS_ERR(filp)) {
 744                 error = PTR_ERR(filp);
 745                 goto out_putfd;
 746         }
 747
 748         fd_install(fd, filp);
 749         goto out_upsem;
 750
 751 out:
 752         dput(dentry);
 753         mntput(ipc_ns->mq_mnt);
 754 out_putfd:
 755         put_unused_fd(fd);
 756         fd = error;
 757 out_upsem:
 758         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 759 out_putname:
 760         putname(name);
 761         return fd;
 762 }
 763
 764 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 765 {
 766         int err;
 767         char *name;
 768         struct dentry *dentry;
 769         struct inode *inode = NULL;
 770         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 771
 772         name = getname(u_name);
 773         if (IS_ERR(name))
 774                 return PTR_ERR(name);
 775
 776         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
 777                         I_MUTEX_PARENT);
 778         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 779         if (IS_ERR(dentry)) {
 780                 err = PTR_ERR(dentry);
 781                 goto out_unlock;
 782         }
 783
 784         if (!dentry->d_inode) {
 785                 err = -ENOENT;
 786                 goto out_err;
 787         }
 788
 789         inode = dentry->d_inode;
 790         if (inode)
 791                 ihold(inode);
 792         err = mnt_want_write(ipc_ns->mq_mnt);
 793         if (err)
 794                 goto out_err;
 795         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 796         mnt_drop_write(ipc_ns->mq_mnt);
 797 out_err:
 798         dput(dentry);
 799
 800 out_unlock:
 801         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 802         putname(name);
 803         if (inode)
 804                 iput(inode);
 805
 806         return err;
 807 }
 808
 809 /* Pipelined send and receive functions.
 810  *
 811  * If a receiver finds no waiting message, then it registers itself in the
 812  * list of waiting receivers. A sender checks that list before adding the new
 813  * message into the message array. If there is a waiting receiver, then it
 814  * bypasses the message array and directly hands the message over to the
 815  * receiver.
 816  * The receiver accepts the message and returns without grabbing the queue
 817  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 818  * are necessary. The same algorithm is used for sysv semaphores, see
 819  * ipc/sem.c for more details.
 820  *
 821  * The same algorithm is used for senders.
 822  */
 823
 824 /* pipelined_send() - send a message directly to the task waiting in
 825  * sys_mq_timedreceive() (without inserting message into a queue).
 826  */
 827 static inline void pipelined_send(struct mqueue_inode_info *info,
 828                                   struct msg_msg *message,
 829                                   struct ext_wait_queue *receiver)
 830 {
 831         receiver->msg = message;
 832         list_del(&receiver->list);
 833         receiver->state = STATE_PENDING;
 834         wake_up_process(receiver->task);
 835         smp_wmb();
 836         receiver->state = STATE_READY;
 837 }
 838
 839 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 840  * gets its message and put to the queue (we have one free place for sure). */
 841 static inline void pipelined_receive(struct mqueue_inode_info *info)
 842 {
 843         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 844
 845         if (!sender) {
 846                 /* for poll */
 847                 wake_up_interruptible(&info->wait_q);
 848                 return;
 849         }
 850         msg_insert(sender->msg, info);
 851         list_del(&sender->list);
 852         sender->state = STATE_PENDING;
 853         wake_up_process(sender->task);
 854         smp_wmb();
 855         sender->state = STATE_READY;
 856 }
 857
 858 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 859                 size_t, msg_len, unsigned int, msg_prio,
 860                 const struct timespec __user *, u_abs_timeout)
 861 {
 862         struct file *filp;
 863         struct inode *inode;
 864         struct ext_wait_queue wait;
 865         struct ext_wait_queue *receiver;
 866         struct msg_msg *msg_ptr;
 867         struct mqueue_inode_info *info;
 868         ktime_t expires, *timeout = NULL;
 869         struct timespec ts;
 870         int ret;
 871
 872         if (u_abs_timeout) {
 873                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 874                 if (res)
 875                         return res;
 876                 timeout = &expires;
 877         }
 878
 879         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 880                 return -EINVAL;
 881
 882         audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
 883
 884         filp = fget(mqdes);
 885         if (unlikely(!filp)) {
 886                 ret = -EBADF;
 887                 goto out;
 888         }
 889
 890         inode = filp->f_path.dentry->d_inode;
 891         if (unlikely(filp->f_op != &mqueue_file_operations)) {
 892                 ret = -EBADF;
 893                 goto out_fput;
 894         }
 895         info = MQUEUE_I(inode);
 896         audit_inode(NULL, filp->f_path.dentry);
 897
 898         if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
 899                 ret = -EBADF;
 900                 goto out_fput;
 901         }
 902
 903         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 904                 ret = -EMSGSIZE;
 905                 goto out_fput;
 906         }
 907
 908         /* First try to allocate memory, before doing anything with
 909          * existing queues. */
 910         msg_ptr = load_msg(u_msg_ptr, msg_len);
 911         if (IS_ERR(msg_ptr)) {
 912                 ret = PTR_ERR(msg_ptr);
 913                 goto out_fput;
 914         }
 915         msg_ptr->m_ts = msg_len;
 916         msg_ptr->m_type = msg_prio;
 917
 918         spin_lock(&info->lock);
 919
 920         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 921                 if (filp->f_flags & O_NONBLOCK) {
 922                         spin_unlock(&info->lock);
 923                         ret = -EAGAIN;
 924                 } else {
 925                         wait.task = current;
 926                         wait.msg = (void *) msg_ptr;
 927                         wait.state = STATE_NONE;
 928                         ret = wq_sleep(info, SEND, timeout, &wait);
 929                 }
 930                 if (ret < 0)
 931                         free_msg(msg_ptr);
 932         } else {
 933                 receiver = wq_get_first_waiter(info, RECV);
 934                 if (receiver) {
 935                         pipelined_send(info, msg_ptr, receiver);
 936                 } else {
 937                         /* adds message to the queue */
 938                         msg_insert(msg_ptr, info);
 939                         __do_notify(info);
 940                 }
 941                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 942                                 CURRENT_TIME;
 943                 spin_unlock(&info->lock);
 944                 ret = 0;
 945         }
 946 out_fput:
 947         fput(filp);
 948 out:
 949         return ret;
 950 }
 951
 952 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
 953                 size_t, msg_len, unsigned int __user *, u_msg_prio,
 954                 const struct timespec __user *, u_abs_timeout)
 955 {
 956         ssize_t ret;
 957         struct msg_msg *msg_ptr;
 958         struct file *filp;
 959         struct inode *inode;
 960         struct mqueue_inode_info *info;
 961         struct ext_wait_queue wait;
 962         ktime_t expires, *timeout = NULL;
 963         struct timespec ts;
 964
 965         if (u_abs_timeout) {
 966                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 967                 if (res)
 968                         return res;
 969                 timeout = &expires;
 970         }
 971
 972         audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
 973
 974         filp = fget(mqdes);
 975         if (unlikely(!filp)) {
 976                 ret = -EBADF;
 977                 goto out;
 978         }
 979
 980         inode = filp->f_path.dentry->d_inode;
 981         if (unlikely(filp->f_op != &mqueue_file_operations)) {
 982                 ret = -EBADF;
 983                 goto out_fput;
 984         }
 985         info = MQUEUE_I(inode);
 986         audit_inode(NULL, filp->f_path.dentry);
 987
 988         if (unlikely(!(filp->f_mode & FMODE_READ))) {
 989                 ret = -EBADF;
 990                 goto out_fput;
 991         }
 992
 993         /* checks if buffer is big enough */
 994         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 995                 ret = -EMSGSIZE;
 996                 goto out_fput;
 997         }
 998
 999         spin_lock(&info->lock);
1000         if (info->attr.mq_curmsgs == 0) {
1001                 if (filp->f_flags & O_NONBLOCK) {
1002                         spin_unlock(&info->lock);
1003                         ret = -EAGAIN;
1004                 } else {
1005                         wait.task = current;
1006                         wait.state = STATE_NONE;
1007                         ret = wq_sleep(info, RECV, timeout, &wait);
1008                         msg_ptr = wait.msg;
1009                 }
1010         } else {
1011                 msg_ptr = msg_get(info);
1012
1013                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1014                                 CURRENT_TIME;
1015
1016                 /* There is now free space in queue. */
1017                 pipelined_receive(info);
1018                 spin_unlock(&info->lock);
1019                 ret = 0;
1020         }
1021         if (ret == 0) {
1022                 ret = msg_ptr->m_ts;
1023
1024                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1025                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1026                         ret = -EFAULT;
1027                 }
1028                 free_msg(msg_ptr);
1029         }
1030 out_fput:
1031         fput(filp);
1032 out:
1033         return ret;
1034 }
1035
1036 /*
1037  * Notes: the case when user wants us to deregister (with NULL as pointer)
1038  * and he isn't currently owner of notification, will be silently discarded.
1039  * It isn't explicitly defined in the POSIX.
1040  */
1041 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1042                 const struct sigevent __user *, u_notification)
1043 {
1044         int ret;
1045         struct file *filp;
1046         struct sock *sock;
1047         struct inode *inode;
1048         struct sigevent notification;
1049         struct mqueue_inode_info *info;
1050         struct sk_buff *nc;
1051
1052         if (u_notification) {
1053                 if (copy_from_user(&notification, u_notification,
1054                                         sizeof(struct sigevent)))
1055                         return -EFAULT;
1056         }
1057
1058         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1059
1060         nc = NULL;
1061         sock = NULL;
1062         if (u_notification != NULL) {
1063                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1064                              notification.sigev_notify != SIGEV_SIGNAL &&
1065                              notification.sigev_notify != SIGEV_THREAD))
1066                         return -EINVAL;
1067                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1068                         !valid_signal(notification.sigev_signo)) {
1069                         return -EINVAL;
1070                 }
1071                 if (notification.sigev_notify == SIGEV_THREAD) {
1072                         long timeo;
1073
1074                         /* create the notify skb */
1075                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1076                         if (!nc) {
1077                                 ret = -ENOMEM;
1078                                 goto out;
1079                         }
1080                         if (copy_from_user(nc->data,
1081                                         notification.sigev_value.sival_ptr,
1082                                         NOTIFY_COOKIE_LEN)) {
1083                                 ret = -EFAULT;
1084                                 goto out;
1085                         }
1086
1087                         /* TODO: add a header? */
1088                         skb_put(nc, NOTIFY_COOKIE_LEN);
1089                         /* and attach it to the socket */
1090 retry:
1091                         filp = fget(notification.sigev_signo);
1092                         if (!filp) {
1093                                 ret = -EBADF;
1094                                 goto out;
1095                         }
1096                         sock = netlink_getsockbyfilp(filp);
1097                         fput(filp);
1098                         if (IS_ERR(sock)) {
1099                                 ret = PTR_ERR(sock);
1100                                 sock = NULL;
1101                                 goto out;
1102                         }
1103
1104                         timeo = MAX_SCHEDULE_TIMEOUT;
1105                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1106                         if (ret == 1)
1107                                 goto retry;
1108                         if (ret) {
1109                                 sock = NULL;
1110                                 nc = NULL;
1111                                 goto out;
1112                         }
1113                 }
1114         }
1115
1116         filp = fget(mqdes);
1117         if (!filp) {
1118                 ret = -EBADF;
1119                 goto out;
1120         }
1121
1122         inode = filp->f_path.dentry->d_inode;
1123         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1124                 ret = -EBADF;
1125                 goto out_fput;
1126         }
1127         info = MQUEUE_I(inode);
1128
1129         ret = 0;
1130         spin_lock(&info->lock);
1131         if (u_notification == NULL) {
1132                 if (info->notify_owner == task_tgid(current)) {
1133                         remove_notification(info);
1134                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1135                 }
1136         } else if (info->notify_owner != NULL) {
1137                 ret = -EBUSY;
1138         } else {
1139                 switch (notification.sigev_notify) {
1140                 case SIGEV_NONE:
1141                         info->notify.sigev_notify = SIGEV_NONE;
1142                         break;
1143                 case SIGEV_THREAD:
1144                         info->notify_sock = sock;
1145                         info->notify_cookie = nc;
1146                         sock = NULL;
1147                         nc = NULL;
1148                         info->notify.sigev_notify = SIGEV_THREAD;
1149                         break;
1150                 case SIGEV_SIGNAL:
1151                         info->notify.sigev_signo = notification.sigev_signo;
1152                         info->notify.sigev_value = notification.sigev_value;
1153                         info->notify.sigev_notify = SIGEV_SIGNAL;
1154                         break;
1155                 }
1156
1157                 info->notify_owner = get_pid(task_tgid(current));
1158                 info->notify_user_ns = get_user_ns(current_user_ns());
1159                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1160         }
1161         spin_unlock(&info->lock);
1162 out_fput:
1163         fput(filp);
1164 out:
1165         if (sock) {
1166                 netlink_detachskb(sock, nc);
1167         } else if (nc) {
1168                 dev_kfree_skb(nc);
1169         }
1170         return ret;
1171 }
1172
1173 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1174                 const struct mq_attr __user *, u_mqstat,
1175                 struct mq_attr __user *, u_omqstat)
1176 {
1177         int ret;
1178         struct mq_attr mqstat, omqstat;
1179         struct file *filp;
1180         struct inode *inode;
1181         struct mqueue_inode_info *info;
1182
1183         if (u_mqstat != NULL) {
1184                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1185                         return -EFAULT;
1186                 if (mqstat.mq_flags & (~O_NONBLOCK))
1187                         return -EINVAL;
1188         }
1189
1190         filp = fget(mqdes);
1191         if (!filp) {
1192                 ret = -EBADF;
1193                 goto out;
1194         }
1195
1196         inode = filp->f_path.dentry->d_inode;
1197         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1198                 ret = -EBADF;
1199                 goto out_fput;
1200         }
1201         info = MQUEUE_I(inode);
1202
1203         spin_lock(&info->lock);
1204
1205         omqstat = info->attr;
1206         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1207         if (u_mqstat) {
1208                 audit_mq_getsetattr(mqdes, &mqstat);
1209                 spin_lock(&filp->f_lock);
1210                 if (mqstat.mq_flags & O_NONBLOCK)
1211                         filp->f_flags |= O_NONBLOCK;
1212                 else
1213                         filp->f_flags &= ~O_NONBLOCK;
1214                 spin_unlock(&filp->f_lock);
1215
1216                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1217         }
1218
1219         spin_unlock(&info->lock);
1220
1221         ret = 0;
1222         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1223                                                 sizeof(struct mq_attr)))
1224                 ret = -EFAULT;
1225
1226 out_fput:
1227         fput(filp);
1228 out:
1229         return ret;
1230 }
1231
1232 static const struct inode_operations mqueue_dir_inode_operations = {
1233         .lookup = simple_lookup,
1234         .create = mqueue_create,
1235         .unlink = mqueue_unlink,
1236 };
1237
1238 static const struct file_operations mqueue_file_operations = {
1239         .flush = mqueue_flush_file,
1240         .poll = mqueue_poll_file,
1241         .read = mqueue_read_file,
1242         .llseek = default_llseek,
1243 };
1244
1245 static const struct super_operations mqueue_super_ops = {
1246         .alloc_inode = mqueue_alloc_inode,
1247         .destroy_inode = mqueue_destroy_inode,
1248         .evict_inode = mqueue_evict_inode,
1249         .statfs = simple_statfs,
1250 };
1251
1252 static struct file_system_type mqueue_fs_type = {
1253         .name = "mqueue",
1254         .mount = mqueue_mount,
1255         .kill_sb = kill_litter_super,
1256 };
1257
1258 int mq_init_ns(struct ipc_namespace *ns)
1259 {
1260         ns->mq_queues_count  = 0;
1261         ns->mq_queues_max    = DFLT_QUEUESMAX;
1262         ns->mq_msg_max       = DFLT_MSGMAX;
1263         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1264
1265         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1266         if (IS_ERR(ns->mq_mnt)) {
1267                 int err = PTR_ERR(ns->mq_mnt);
1268                 ns->mq_mnt = NULL;
1269                 return err;
1270         }
1271         return 0;
1272 }
1273
1274 void mq_clear_sbinfo(struct ipc_namespace *ns)
1275 {
1276         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1277 }
1278
1279 void mq_put_mnt(struct ipc_namespace *ns)
1280 {
1281         kern_unmount(ns->mq_mnt);
1282 }
1283
1284 static int __init init_mqueue_fs(void)
1285 {
1286         int error;
1287
1288         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1289                                 sizeof(struct mqueue_inode_info), 0,
1290                                 SLAB_HWCACHE_ALIGN, init_once);
1291         if (mqueue_inode_cachep == NULL)
1292                 return -ENOMEM;
1293
1294         /* ignore failures - they are not fatal */
1295         mq_sysctl_table = mq_register_sysctl_table();
1296
1297         error = register_filesystem(&mqueue_fs_type);
1298         if (error)
1299                 goto out_sysctl;
1300
1301         spin_lock_init(&mq_lock);
1302
1303         error = mq_init_ns(&init_ipc_ns);
1304         if (error)
1305                 goto out_filesystem;
1306
1307         return 0;
1308
1309 out_filesystem:
1310         unregister_filesystem(&mqueue_fs_type);
1311 out_sysctl:
1312         if (mq_sysctl_table)
1313                 unregister_sysctl_table(mq_sysctl_table);
1314         kmem_cache_destroy(mqueue_inode_cachep);
1315         return error;
1316 }
1317
1318 __initcall(init_mqueue_fs);