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