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