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