SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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
  36 #include <net/sock.h>
  37 #include "util.h"
  38
  39 #define MQUEUE_MAGIC    0x19800202
  40 #define DIRENT_SIZE     20
  41 #define FILENT_SIZE     80
  42
  43 #define SEND            0
  44 #define RECV            1
  45
  46 #define STATE_NONE      0
  47 #define STATE_PENDING   1
  48 #define STATE_READY     2
  49
  50 struct ext_wait_queue {         /* queue of sleeping tasks */
  51         struct task_struct *task;
  52         struct list_head list;
  53         struct msg_msg *msg;    /* ptr of loaded message */
  54         int state;              /* one of STATE_* values */
  55 };
  56
  57 struct mqueue_inode_info {
  58         spinlock_t lock;
  59         struct inode vfs_inode;
  60         wait_queue_head_t wait_q;
  61
  62         struct msg_msg **messages;
  63         struct mq_attr attr;
  64
  65         struct sigevent notify;
  66         struct pid* notify_owner;
  67         struct user_struct *user;       /* user who created, for accounting */
  68         struct sock *notify_sock;
  69         struct sk_buff *notify_cookie;
  70
  71         /* for tasks waiting for free space and messages, respectively */
  72         struct ext_wait_queue e_wait_q[2];
  73
  74         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  75 };
  76
  77 static const struct inode_operations mqueue_dir_inode_operations;
  78 static const struct file_operations mqueue_file_operations;
  79 static const struct super_operations mqueue_super_ops;
  80 static void remove_notification(struct mqueue_inode_info *info);
  81
  82 static struct kmem_cache *mqueue_inode_cachep;
  83
  84 static struct ctl_table_header * mq_sysctl_table;
  85
  86 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  87 {
  88         return container_of(inode, struct mqueue_inode_info, vfs_inode);
  89 }
  90
  91 /*
  92  * This routine should be called with the mq_lock held.
  93  */
  94 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
  95 {
  96         return get_ipc_ns(inode->i_sb->s_fs_info);
  97 }
  98
  99 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 100 {
 101         struct ipc_namespace *ns;
 102
 103         spin_lock(&mq_lock);
 104         ns = __get_ns_from_inode(inode);
 105         spin_unlock(&mq_lock);
 106         return ns;
 107 }
 108
 109 static struct inode *mqueue_get_inode(struct super_block *sb,
 110                 struct ipc_namespace *ipc_ns, int mode,
 111                 struct mq_attr *attr)
 112 {
 113         struct user_struct *u = current_user();
 114         struct inode *inode;
 115
 116         inode = new_inode(sb);
 117         if (inode) {
 118                 inode->i_mode = mode;
 119                 inode->i_uid = current_fsuid();
 120                 inode->i_gid = current_fsgid();
 121                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 122                                 CURRENT_TIME;
 123
 124                 if (S_ISREG(mode)) {
 125                         struct mqueue_inode_info *info;
 126                         struct task_struct *p = current;
 127                         unsigned long mq_bytes, mq_msg_tblsz;
 128
 129                         inode->i_fop = &mqueue_file_operations;
 130                         inode->i_size = FILENT_SIZE;
 131                         /* mqueue specific info */
 132                         info = MQUEUE_I(inode);
 133                         spin_lock_init(&info->lock);
 134                         init_waitqueue_head(&info->wait_q);
 135                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 136                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 137                         info->notify_owner = NULL;
 138                         info->qsize = 0;
 139                         info->user = NULL;      /* set when all is ok */
 140                         memset(&info->attr, 0, sizeof(info->attr));
 141                         info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
 142                         info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
 143                         if (attr) {
 144                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 145                                 info->attr.mq_msgsize = attr->mq_msgsize;
 146                         }
 147                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 148                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 149                         if (!info->messages)
 150                                 goto out_inode;
 151
 152                         mq_bytes = (mq_msg_tblsz +
 153                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 154
 155                         spin_lock(&mq_lock);
 156                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 157                             u->mq_bytes + mq_bytes >
 158                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
 159                                 spin_unlock(&mq_lock);
 160                                 kfree(info->messages);
 161                                 goto out_inode;
 162                         }
 163                         u->mq_bytes += mq_bytes;
 164                         spin_unlock(&mq_lock);
 165
 166                         /* all is ok */
 167                         info->user = get_uid(u);
 168                 } else if (S_ISDIR(mode)) {
 169                         inc_nlink(inode);
 170                         /* Some things misbehave if size == 0 on a directory */
 171                         inode->i_size = 2 * DIRENT_SIZE;
 172                         inode->i_op = &mqueue_dir_inode_operations;
 173                         inode->i_fop = &simple_dir_operations;
 174                 }
 175         }
 176         return inode;
 177 out_inode:
 178         make_bad_inode(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 = 0;
 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         }
 206
 207 out:
 208         return error;
 209 }
 210
 211 static int mqueue_get_sb(struct file_system_type *fs_type,
 212                          int flags, const char *dev_name,
 213                          void *data, struct vfsmount *mnt)
 214 {
 215         if (!(flags & MS_KERNMOUNT))
 216                 data = current->nsproxy->ipc_ns;
 217         return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
 218 }
 219
 220 static void init_once(void *foo)
 221 {
 222         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 223
 224         inode_init_once(&p->vfs_inode);
 225 }
 226
 227 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 228 {
 229         struct mqueue_inode_info *ei;
 230
 231         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 232         if (!ei)
 233                 return NULL;
 234         return &ei->vfs_inode;
 235 }
 236
 237 static void mqueue_destroy_inode(struct inode *inode)
 238 {
 239         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 240 }
 241
 242 static void mqueue_delete_inode(struct inode *inode)
 243 {
 244         struct mqueue_inode_info *info;
 245         struct user_struct *user;
 246         unsigned long mq_bytes;
 247         int i;
 248         struct ipc_namespace *ipc_ns;
 249
 250         if (S_ISDIR(inode->i_mode)) {
 251                 clear_inode(inode);
 252                 return;
 253         }
 254         ipc_ns = get_ns_from_inode(inode);
 255         info = MQUEUE_I(inode);
 256         spin_lock(&info->lock);
 257         for (i = 0; i < info->attr.mq_curmsgs; i++)
 258                 free_msg(info->messages[i]);
 259         kfree(info->messages);
 260         spin_unlock(&info->lock);
 261
 262         clear_inode(inode);
 263
 264         /* Total amount of bytes accounted for the mqueue */
 265         mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
 266             + info->attr.mq_msgsize);
 267         user = info->user;
 268         if (user) {
 269                 spin_lock(&mq_lock);
 270                 user->mq_bytes -= mq_bytes;
 271                 /*
 272                  * get_ns_from_inode() ensures that the
 273                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 274                  * to which we now hold a reference, or it is NULL.
 275                  * We can't put it here under mq_lock, though.
 276                  */
 277                 if (ipc_ns)
 278                         ipc_ns->mq_queues_count--;
 279                 spin_unlock(&mq_lock);
 280                 free_uid(user);
 281         }
 282         if (ipc_ns)
 283                 put_ipc_ns(ipc_ns);
 284 }
 285
 286 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 287                                 int mode, struct nameidata *nd)
 288 {
 289         struct inode *inode;
 290         struct mq_attr *attr = dentry->d_fsdata;
 291         int error;
 292         struct ipc_namespace *ipc_ns;
 293
 294         spin_lock(&mq_lock);
 295         ipc_ns = __get_ns_from_inode(dir);
 296         if (!ipc_ns) {
 297                 error = -EACCES;
 298                 goto out_unlock;
 299         }
 300         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 301                         !capable(CAP_SYS_RESOURCE)) {
 302                 error = -ENOSPC;
 303                 goto out_unlock;
 304         }
 305         ipc_ns->mq_queues_count++;
 306         spin_unlock(&mq_lock);
 307
 308         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 309         if (!inode) {
 310                 error = -ENOMEM;
 311                 spin_lock(&mq_lock);
 312                 ipc_ns->mq_queues_count--;
 313                 goto out_unlock;
 314         }
 315
 316         put_ipc_ns(ipc_ns);
 317         dir->i_size += DIRENT_SIZE;
 318         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 319
 320         d_instantiate(dentry, inode);
 321         dget(dentry);
 322         return 0;
 323 out_unlock:
 324         spin_unlock(&mq_lock);
 325         if (ipc_ns)
 326                 put_ipc_ns(ipc_ns);
 327         return error;
 328 }
 329
 330 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 331 {
 332         struct inode *inode = dentry->d_inode;
 333
 334         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 335         dir->i_size -= DIRENT_SIZE;
 336         drop_nlink(inode);
 337         dput(dentry);
 338         return 0;
 339 }
 340
 341 /*
 342 *       This is routine for system read from queue file.
 343 *       To avoid mess with doing here some sort of mq_receive we allow
 344 *       to read only queue size & notification info (the only values
 345 *       that are interesting from user point of view and aren't accessible
 346 *       through std routines)
 347 */
 348 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 349                                 size_t count, loff_t *off)
 350 {
 351         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 352         char buffer[FILENT_SIZE];
 353         ssize_t ret;
 354
 355         spin_lock(&info->lock);
 356         snprintf(buffer, sizeof(buffer),
 357                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 358                         info->qsize,
 359                         info->notify_owner ? info->notify.sigev_notify : 0,
 360                         (info->notify_owner &&
 361                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 362                                 info->notify.sigev_signo : 0,
 363                         pid_vnr(info->notify_owner));
 364         spin_unlock(&info->lock);
 365         buffer[sizeof(buffer)-1] = '\0';
 366
 367         ret = simple_read_from_buffer(u_data, count, off, buffer,
 368                                 strlen(buffer));
 369         if (ret <= 0)
 370                 return ret;
 371
 372         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 373         return ret;
 374 }
 375
 376 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 377 {
 378         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 379
 380         spin_lock(&info->lock);
 381         if (task_tgid(current) == info->notify_owner)
 382                 remove_notification(info);
 383
 384         spin_unlock(&info->lock);
 385         return 0;
 386 }
 387
 388 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 389 {
 390         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 391         int retval = 0;
 392
 393         poll_wait(filp, &info->wait_q, poll_tab);
 394
 395         spin_lock(&info->lock);
 396         if (info->attr.mq_curmsgs)
 397                 retval = POLLIN | POLLRDNORM;
 398
 399         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 400                 retval |= POLLOUT | POLLWRNORM;
 401         spin_unlock(&info->lock);
 402
 403         return retval;
 404 }
 405
 406 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 407 static void wq_add(struct mqueue_inode_info *info, int sr,
 408                         struct ext_wait_queue *ewp)
 409 {
 410         struct ext_wait_queue *walk;
 411
 412         ewp->task = current;
 413
 414         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 415                 if (walk->task->static_prio <= current->static_prio) {
 416                         list_add_tail(&ewp->list, &walk->list);
 417                         return;
 418                 }
 419         }
 420         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 421 }
 422
 423 /*
 424  * Puts current task to sleep. Caller must hold queue lock. After return
 425  * lock isn't held.
 426  * sr: SEND or RECV
 427  */
 428 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 429                         long timeout, struct ext_wait_queue *ewp)
 430 {
 431         int retval;
 432         signed long time;
 433
 434         wq_add(info, sr, ewp);
 435
 436         for (;;) {
 437                 set_current_state(TASK_INTERRUPTIBLE);
 438
 439                 spin_unlock(&info->lock);
 440                 time = schedule_timeout(timeout);
 441
 442                 while (ewp->state == STATE_PENDING)
 443                         cpu_relax();
 444
 445                 if (ewp->state == STATE_READY) {
 446                         retval = 0;
 447                         goto out;
 448                 }
 449                 spin_lock(&info->lock);
 450                 if (ewp->state == STATE_READY) {
 451                         retval = 0;
 452                         goto out_unlock;
 453                 }
 454                 if (signal_pending(current)) {
 455                         retval = -ERESTARTSYS;
 456                         break;
 457                 }
 458                 if (time == 0) {
 459                         retval = -ETIMEDOUT;
 460                         break;
 461                 }
 462         }
 463         list_del(&ewp->list);
 464 out_unlock:
 465         spin_unlock(&info->lock);
 466 out:
 467         return retval;
 468 }
 469
 470 /*
 471  * Returns waiting task that should be serviced first or NULL if none exists
 472  */
 473 static struct ext_wait_queue *wq_get_first_waiter(
 474                 struct mqueue_inode_info *info, int sr)
 475 {
 476         struct list_head *ptr;
 477
 478         ptr = info->e_wait_q[sr].list.prev;
 479         if (ptr == &info->e_wait_q[sr].list)
 480                 return NULL;
 481         return list_entry(ptr, struct ext_wait_queue, list);
 482 }
 483
 484 /* Auxiliary functions to manipulate messages' list */
 485 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 486 {
 487         int k;
 488
 489         k = info->attr.mq_curmsgs - 1;
 490         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 491                 info->messages[k + 1] = info->messages[k];
 492                 k--;
 493         }
 494         info->attr.mq_curmsgs++;
 495         info->qsize += ptr->m_ts;
 496         info->messages[k + 1] = ptr;
 497 }
 498
 499 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 500 {
 501         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 502         return info->messages[info->attr.mq_curmsgs];
 503 }
 504
 505 static inline void set_cookie(struct sk_buff *skb, char code)
 506 {
 507         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 508 }
 509
 510 /*
 511  * The next function is only to split too long sys_mq_timedsend
 512  */
 513 static void __do_notify(struct mqueue_inode_info *info)
 514 {
 515         /* notification
 516          * invoked when there is registered process and there isn't process
 517          * waiting synchronously for message AND state of queue changed from
 518          * empty to not empty. Here we are sure that no one is waiting
 519          * synchronously. */
 520         if (info->notify_owner &&
 521             info->attr.mq_curmsgs == 1) {
 522                 struct siginfo sig_i;
 523                 switch (info->notify.sigev_notify) {
 524                 case SIGEV_NONE:
 525                         break;
 526                 case SIGEV_SIGNAL:
 527                         /* sends signal */
 528
 529                         sig_i.si_signo = info->notify.sigev_signo;
 530                         sig_i.si_errno = 0;
 531                         sig_i.si_code = SI_MESGQ;
 532                         sig_i.si_value = info->notify.sigev_value;
 533                         sig_i.si_pid = task_tgid_nr_ns(current,
 534                                                 ns_of_pid(info->notify_owner));
 535                         sig_i.si_uid = current_uid();
 536
 537                         kill_pid_info(info->notify.sigev_signo,
 538                                       &sig_i, info->notify_owner);
 539                         break;
 540                 case SIGEV_THREAD:
 541                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 542                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 543                         break;
 544                 }
 545                 /* after notification unregisters process */
 546                 put_pid(info->notify_owner);
 547                 info->notify_owner = NULL;
 548         }
 549         wake_up(&info->wait_q);
 550 }
 551
 552 static long prepare_timeout(struct timespec *p)
 553 {
 554         struct timespec nowts;
 555         long timeout;
 556
 557         if (p) {
 558                 if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
 559                         || p->tv_nsec >= NSEC_PER_SEC))
 560                         return -EINVAL;
 561                 nowts = CURRENT_TIME;
 562                 /* first subtract as jiffies can't be too big */
 563                 p->tv_sec -= nowts.tv_sec;
 564                 if (p->tv_nsec < nowts.tv_nsec) {
 565                         p->tv_nsec += NSEC_PER_SEC;
 566                         p->tv_sec--;
 567                 }
 568                 p->tv_nsec -= nowts.tv_nsec;
 569                 if (p->tv_sec < 0)
 570                         return 0;
 571
 572                 timeout = timespec_to_jiffies(p) + 1;
 573         } else
 574                 return MAX_SCHEDULE_TIMEOUT;
 575
 576         return timeout;
 577 }
 578
 579 static void remove_notification(struct mqueue_inode_info *info)
 580 {
 581         if (info->notify_owner != NULL &&
 582             info->notify.sigev_notify == SIGEV_THREAD) {
 583                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 584                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 585         }
 586         put_pid(info->notify_owner);
 587         info->notify_owner = NULL;
 588 }
 589
 590 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
 591 {
 592         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 593                 return 0;
 594         if (capable(CAP_SYS_RESOURCE)) {
 595                 if (attr->mq_maxmsg > HARD_MSGMAX)
 596                         return 0;
 597         } else {
 598                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
 599                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
 600                         return 0;
 601         }
 602         /* check for overflow */
 603         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 604                 return 0;
 605         if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
 606             + sizeof (struct msg_msg *))) <
 607             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 608                 return 0;
 609         return 1;
 610 }
 611
 612 /*
 613  * Invoked when creating a new queue via sys_mq_open
 614  */
 615 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
 616                         struct dentry *dentry, int oflag, mode_t mode,
 617                         struct mq_attr *attr)
 618 {
 619         const struct cred *cred = current_cred();
 620         struct file *result;
 621         int ret;
 622
 623         if (attr) {
 624                 ret = -EINVAL;
 625                 if (!mq_attr_ok(ipc_ns, attr))
 626                         goto out;
 627                 /* store for use during create */
 628                 dentry->d_fsdata = attr;
 629         }
 630
 631         mode &= ~current_umask();
 632         ret = mnt_want_write(ipc_ns->mq_mnt);
 633         if (ret)
 634                 goto out;
 635         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 636         dentry->d_fsdata = NULL;
 637         if (ret)
 638                 goto out_drop_write;
 639
 640         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 641         /*
 642          * dentry_open() took a persistent mnt_want_write(),
 643          * so we can now drop this one.
 644          */
 645         mnt_drop_write(ipc_ns->mq_mnt);
 646         return result;
 647
 648 out_drop_write:
 649         mnt_drop_write(ipc_ns->mq_mnt);
 650 out:
 651         dput(dentry);
 652         mntput(ipc_ns->mq_mnt);
 653         return ERR_PTR(ret);
 654 }
 655
 656 /* Opens existing queue */
 657 static struct file *do_open(struct ipc_namespace *ipc_ns,
 658                                 struct dentry *dentry, int oflag)
 659 {
 660         const struct cred *cred = current_cred();
 661
 662         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 663                                                   MAY_READ | MAY_WRITE };
 664
 665         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 666                 dput(dentry);
 667                 mntput(ipc_ns->mq_mnt);
 668                 return ERR_PTR(-EINVAL);
 669         }
 670
 671         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 672                 dput(dentry);
 673                 mntput(ipc_ns->mq_mnt);
 674                 return ERR_PTR(-EACCES);
 675         }
 676
 677         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 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                         error = -EEXIST;
 714                         if (oflag & O_EXCL)
 715                                 goto out;
 716                         filp = do_open(ipc_ns, dentry, oflag);
 717                 } else {
 718                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
 719                                                 dentry, oflag, mode,
 720                                                 u_attr ? &attr : NULL);
 721                 }
 722         } else {
 723                 error = -ENOENT;
 724                 if (!dentry->d_inode)
 725                         goto out;
 726                 audit_inode(name, dentry);
 727                 filp = do_open(ipc_ns, dentry, oflag);
 728         }
 729
 730         if (IS_ERR(filp)) {
 731                 error = PTR_ERR(filp);
 732                 goto out_putfd;
 733         }
 734
 735         fd_install(fd, filp);
 736         goto out_upsem;
 737
 738 out:
 739         dput(dentry);
 740         mntput(ipc_ns->mq_mnt);
 741 out_putfd:
 742         put_unused_fd(fd);
 743         fd = error;
 744 out_upsem:
 745         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 746 out_putname:
 747         putname(name);
 748         return fd;
 749 }
 750
 751 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 752 {
 753         int err;
 754         char *name;
 755         struct dentry *dentry;
 756         struct inode *inode = NULL;
 757         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 758
 759         name = getname(u_name);
 760         if (IS_ERR(name))
 761                 return PTR_ERR(name);
 762
 763         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
 764                         I_MUTEX_PARENT);
 765         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 766         if (IS_ERR(dentry)) {
 767                 err = PTR_ERR(dentry);
 768                 goto out_unlock;
 769         }
 770
 771         if (!dentry->d_inode) {
 772                 err = -ENOENT;
 773                 goto out_err;
 774         }
 775
 776         inode = dentry->d_inode;
 777         if (inode)
 778                 atomic_inc(&inode->i_count);
 779         err = mnt_want_write(ipc_ns->mq_mnt);
 780         if (err)
 781                 goto out_err;
 782         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 783         mnt_drop_write(ipc_ns->mq_mnt);
 784 out_err:
 785         dput(dentry);
 786
 787 out_unlock:
 788         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 789         putname(name);
 790         if (inode)
 791                 iput(inode);
 792
 793         return err;
 794 }
 795
 796 /* Pipelined send and receive functions.
 797  *
 798  * If a receiver finds no waiting message, then it registers itself in the
 799  * list of waiting receivers. A sender checks that list before adding the new
 800  * message into the message array. If there is a waiting receiver, then it
 801  * bypasses the message array and directly hands the message over to the
 802  * receiver.
 803  * The receiver accepts the message and returns without grabbing the queue
 804  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 805  * are necessary. The same algorithm is used for sysv semaphores, see
 806  * ipc/sem.c for more details.
 807  *
 808  * The same algorithm is used for senders.
 809  */
 810
 811 /* pipelined_send() - send a message directly to the task waiting in
 812  * sys_mq_timedreceive() (without inserting message into a queue).
 813  */
 814 static inline void pipelined_send(struct mqueue_inode_info *info,
 815                                   struct msg_msg *message,
 816                                   struct ext_wait_queue *receiver)
 817 {
 818         receiver->msg = message;
 819         list_del(&receiver->list);
 820         receiver->state = STATE_PENDING;
 821         wake_up_process(receiver->task);
 822         smp_wmb();
 823         receiver->state = STATE_READY;
 824 }
 825
 826 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 827  * gets its message and put to the queue (we have one free place for sure). */
 828 static inline void pipelined_receive(struct mqueue_inode_info *info)
 829 {
 830         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 831
 832         if (!sender) {
 833                 /* for poll */
 834                 wake_up_interruptible(&info->wait_q);
 835                 return;
 836         }
 837         msg_insert(sender->msg, info);
 838         list_del(&sender->list);
 839         sender->state = STATE_PENDING;
 840         wake_up_process(sender->task);
 841         smp_wmb();
 842         sender->state = STATE_READY;
 843 }
 844
 845 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 846                 size_t, msg_len, unsigned int, msg_prio,
 847                 const struct timespec __user *, u_abs_timeout)
 848 {
 849         struct file *filp;
 850         struct inode *inode;
 851         struct ext_wait_queue wait;
 852         struct ext_wait_queue *receiver;
 853         struct msg_msg *msg_ptr;
 854         struct mqueue_inode_info *info;
 855         struct timespec ts, *p = NULL;
 856         long timeout;
 857         int ret;
 858
 859         if (u_abs_timeout) {
 860                 if (copy_from_user(&ts, u_abs_timeout,
 861                                         sizeof(struct timespec)))
 862                         return -EFAULT;
 863                 p = &ts;
 864         }
 865
 866         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 867                 return -EINVAL;
 868
 869         audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
 870         timeout = prepare_timeout(p);
 871
 872         ret = -EBADF;
 873         filp = fget(mqdes);
 874         if (unlikely(!filp))
 875                 goto out;
 876
 877         inode = filp->f_path.dentry->d_inode;
 878         if (unlikely(filp->f_op != &mqueue_file_operations))
 879                 goto out_fput;
 880         info = MQUEUE_I(inode);
 881         audit_inode(NULL, filp->f_path.dentry);
 882
 883         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 884                 goto out_fput;
 885
 886         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 887                 ret = -EMSGSIZE;
 888                 goto out_fput;
 889         }
 890
 891         /* First try to allocate memory, before doing anything with
 892          * existing queues. */
 893         msg_ptr = load_msg(u_msg_ptr, msg_len);
 894         if (IS_ERR(msg_ptr)) {
 895                 ret = PTR_ERR(msg_ptr);
 896                 goto out_fput;
 897         }
 898         msg_ptr->m_ts = msg_len;
 899         msg_ptr->m_type = msg_prio;
 900
 901         spin_lock(&info->lock);
 902
 903         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 904                 if (filp->f_flags & O_NONBLOCK) {
 905                         spin_unlock(&info->lock);
 906                         ret = -EAGAIN;
 907                 } else if (unlikely(timeout < 0)) {
 908                         spin_unlock(&info->lock);
 909                         ret = timeout;
 910                 } else {
 911                         wait.task = current;
 912                         wait.msg = (void *) msg_ptr;
 913                         wait.state = STATE_NONE;
 914                         ret = wq_sleep(info, SEND, timeout, &wait);
 915                 }
 916                 if (ret < 0)
 917                         free_msg(msg_ptr);
 918         } else {
 919                 receiver = wq_get_first_waiter(info, RECV);
 920                 if (receiver) {
 921                         pipelined_send(info, msg_ptr, receiver);
 922                 } else {
 923                         /* adds message to the queue */
 924                         msg_insert(msg_ptr, info);
 925                         __do_notify(info);
 926                 }
 927                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 928                                 CURRENT_TIME;
 929                 spin_unlock(&info->lock);
 930                 ret = 0;
 931         }
 932 out_fput:
 933         fput(filp);
 934 out:
 935         return ret;
 936 }
 937
 938 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
 939                 size_t, msg_len, unsigned int __user *, u_msg_prio,
 940                 const struct timespec __user *, u_abs_timeout)
 941 {
 942         long timeout;
 943         ssize_t ret;
 944         struct msg_msg *msg_ptr;
 945         struct file *filp;
 946         struct inode *inode;
 947         struct mqueue_inode_info *info;
 948         struct ext_wait_queue wait;
 949         struct timespec ts, *p = NULL;
 950
 951         if (u_abs_timeout) {
 952                 if (copy_from_user(&ts, u_abs_timeout,
 953                                         sizeof(struct timespec)))
 954                         return -EFAULT;
 955                 p = &ts;
 956         }
 957
 958         audit_mq_sendrecv(mqdes, msg_len, 0, p);
 959         timeout = prepare_timeout(p);
 960
 961         ret = -EBADF;
 962         filp = fget(mqdes);
 963         if (unlikely(!filp))
 964                 goto out;
 965
 966         inode = filp->f_path.dentry->d_inode;
 967         if (unlikely(filp->f_op != &mqueue_file_operations))
 968                 goto out_fput;
 969         info = MQUEUE_I(inode);
 970         audit_inode(NULL, filp->f_path.dentry);
 971
 972         if (unlikely(!(filp->f_mode & FMODE_READ)))
 973                 goto out_fput;
 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                         msg_ptr = NULL;
 987                 } else if (unlikely(timeout < 0)) {
 988                         spin_unlock(&info->lock);
 989                         ret = timeout;
 990                         msg_ptr = NULL;
 991                 } else {
 992                         wait.task = current;
 993                         wait.state = STATE_NONE;
 994                         ret = wq_sleep(info, RECV, timeout, &wait);
 995                         msg_ptr = wait.msg;
 996                 }
 997         } else {
 998                 msg_ptr = msg_get(info);
 999
1000                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1001                                 CURRENT_TIME;
1002
1003                 /* There is now free space in queue. */
1004                 pipelined_receive(info);
1005                 spin_unlock(&info->lock);
1006                 ret = 0;
1007         }
1008         if (ret == 0) {
1009                 ret = msg_ptr->m_ts;
1010
1011                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1012                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1013                         ret = -EFAULT;
1014                 }
1015                 free_msg(msg_ptr);
1016         }
1017 out_fput:
1018         fput(filp);
1019 out:
1020         return ret;
1021 }
1022
1023 /*
1024  * Notes: the case when user wants us to deregister (with NULL as pointer)
1025  * and he isn't currently owner of notification, will be silently discarded.
1026  * It isn't explicitly defined in the POSIX.
1027  */
1028 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1029                 const struct sigevent __user *, u_notification)
1030 {
1031         int ret;
1032         struct file *filp;
1033         struct sock *sock;
1034         struct inode *inode;
1035         struct sigevent notification;
1036         struct mqueue_inode_info *info;
1037         struct sk_buff *nc;
1038
1039         if (u_notification) {
1040                 if (copy_from_user(&notification, u_notification,
1041                                         sizeof(struct sigevent)))
1042                         return -EFAULT;
1043         }
1044
1045         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1046
1047         nc = NULL;
1048         sock = NULL;
1049         if (u_notification != NULL) {
1050                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1051                              notification.sigev_notify != SIGEV_SIGNAL &&
1052                              notification.sigev_notify != SIGEV_THREAD))
1053                         return -EINVAL;
1054                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1055                         !valid_signal(notification.sigev_signo)) {
1056                         return -EINVAL;
1057                 }
1058                 if (notification.sigev_notify == SIGEV_THREAD) {
1059                         long timeo;
1060
1061                         /* create the notify skb */
1062                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1063                         ret = -ENOMEM;
1064                         if (!nc)
1065                                 goto out;
1066                         ret = -EFAULT;
1067                         if (copy_from_user(nc->data,
1068                                         notification.sigev_value.sival_ptr,
1069                                         NOTIFY_COOKIE_LEN)) {
1070                                 goto out;
1071                         }
1072
1073                         /* TODO: add a header? */
1074                         skb_put(nc, NOTIFY_COOKIE_LEN);
1075                         /* and attach it to the socket */
1076 retry:
1077                         filp = fget(notification.sigev_signo);
1078                         ret = -EBADF;
1079                         if (!filp)
1080                                 goto out;
1081                         sock = netlink_getsockbyfilp(filp);
1082                         fput(filp);
1083                         if (IS_ERR(sock)) {
1084                                 ret = PTR_ERR(sock);
1085                                 sock = NULL;
1086                                 goto out;
1087                         }
1088
1089                         timeo = MAX_SCHEDULE_TIMEOUT;
1090                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1091                         if (ret == 1)
1092                                 goto retry;
1093                         if (ret) {
1094                                 sock = NULL;
1095                                 nc = NULL;
1096                                 goto out;
1097                         }
1098                 }
1099         }
1100
1101         ret = -EBADF;
1102         filp = fget(mqdes);
1103         if (!filp)
1104                 goto out;
1105
1106         inode = filp->f_path.dentry->d_inode;
1107         if (unlikely(filp->f_op != &mqueue_file_operations))
1108                 goto out_fput;
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         ret = -EBADF;
1172         filp = fget(mqdes);
1173         if (!filp)
1174                 goto out;
1175
1176         inode = filp->f_path.dentry->d_inode;
1177         if (unlikely(filp->f_op != &mqueue_file_operations))
1178                 goto out_fput;
1179         info = MQUEUE_I(inode);
1180
1181         spin_lock(&info->lock);
1182
1183         omqstat = info->attr;
1184         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1185         if (u_mqstat) {
1186                 audit_mq_getsetattr(mqdes, &mqstat);
1187                 spin_lock(&filp->f_lock);
1188                 if (mqstat.mq_flags & O_NONBLOCK)
1189                         filp->f_flags |= O_NONBLOCK;
1190                 else
1191                         filp->f_flags &= ~O_NONBLOCK;
1192                 spin_unlock(&filp->f_lock);
1193
1194                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1195         }
1196
1197         spin_unlock(&info->lock);
1198
1199         ret = 0;
1200         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1201                                                 sizeof(struct mq_attr)))
1202                 ret = -EFAULT;
1203
1204 out_fput:
1205         fput(filp);
1206 out:
1207         return ret;
1208 }
1209
1210 static const struct inode_operations mqueue_dir_inode_operations = {
1211         .lookup = simple_lookup,
1212         .create = mqueue_create,
1213         .unlink = mqueue_unlink,
1214 };
1215
1216 static const struct file_operations mqueue_file_operations = {
1217         .flush = mqueue_flush_file,
1218         .poll = mqueue_poll_file,
1219         .read = mqueue_read_file,
1220 };
1221
1222 static const struct super_operations mqueue_super_ops = {
1223         .alloc_inode = mqueue_alloc_inode,
1224         .destroy_inode = mqueue_destroy_inode,
1225         .statfs = simple_statfs,
1226         .delete_inode = mqueue_delete_inode,
1227         .drop_inode = generic_delete_inode,
1228 };
1229
1230 static struct file_system_type mqueue_fs_type = {
1231         .name = "mqueue",
1232         .get_sb = mqueue_get_sb,
1233         .kill_sb = kill_litter_super,
1234 };
1235
1236 int mq_init_ns(struct ipc_namespace *ns)
1237 {
1238         ns->mq_queues_count  = 0;
1239         ns->mq_queues_max    = DFLT_QUEUESMAX;
1240         ns->mq_msg_max       = DFLT_MSGMAX;
1241         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1242
1243         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1244         if (IS_ERR(ns->mq_mnt)) {
1245                 int err = PTR_ERR(ns->mq_mnt);
1246                 ns->mq_mnt = NULL;
1247                 return err;
1248         }
1249         return 0;
1250 }
1251
1252 void mq_clear_sbinfo(struct ipc_namespace *ns)
1253 {
1254         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1255 }
1256
1257 void mq_put_mnt(struct ipc_namespace *ns)
1258 {
1259         mntput(ns->mq_mnt);
1260 }
1261
1262 static int __init init_mqueue_fs(void)
1263 {
1264         int error;
1265
1266         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1267                                 sizeof(struct mqueue_inode_info), 0,
1268                                 SLAB_HWCACHE_ALIGN, init_once);
1269         if (mqueue_inode_cachep == NULL)
1270                 return -ENOMEM;
1271
1272         /* ignore failues - they are not fatal */
1273         mq_sysctl_table = mq_register_sysctl_table();
1274
1275         error = register_filesystem(&mqueue_fs_type);
1276         if (error)
1277                 goto out_sysctl;
1278
1279         spin_lock_init(&mq_lock);
1280
1281         init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1282         if (IS_ERR(init_ipc_ns.mq_mnt)) {
1283                 error = PTR_ERR(init_ipc_ns.mq_mnt);
1284                 goto out_filesystem;
1285         }
1286
1287         return 0;
1288
1289 out_filesystem:
1290         unregister_filesystem(&mqueue_fs_type);
1291 out_sysctl:
1292         if (mq_sysctl_table)
1293                 unregister_sysctl_table(mq_sysctl_table);
1294         kmem_cache_destroy(mqueue_inode_cachep);
1295         return error;
1296 }
1297
1298 __initcall(init_mqueue_fs);