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