ipc/mqueue.c

   1 /*
   2  * POSIX message queues filesystem for Linux.
   3  *
   4  * Copyright (C) 2003,2004  Krzysztof Benedyczak    (golbi@mat.uni.torun.pl)
   5  *                          Michal Wronski          (michal.wronski@gmail.com)
   6  *
   7  * Spinlocks:               Mohamed Abbas           (abbas.mohamed@intel.com)
   8  * Lockless receive & send, fd based notify:
   9  *                          Manfred Spraul          (manfred@colorfullife.com)
  10  *
  11  * Audit:                   George Wilson           (ltcgcw@us.ibm.com)
  12  *
  13  * This file is released under the GPL.
  14  */
  15
  16 #include <linux/capability.h>
  17 #include <linux/init.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/file.h>
  20 #include <linux/mount.h>
  21 #include <linux/namei.h>
  22 #include <linux/sysctl.h>
  23 #include <linux/poll.h>
  24 #include <linux/mqueue.h>
  25 #include <linux/msg.h>
  26 #include <linux/skbuff.h>
  27 #include <linux/netlink.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/audit.h>
  30 #include <linux/signal.h>
  31 #include <linux/mutex.h>
  32 #include <linux/nsproxy.h>
  33 #include <linux/pid.h>
  34 #include <linux/ipc_namespace.h>
  35 #include <linux/user_namespace.h>
  36 #include <linux/slab.h>
  37
  38 #include <net/sock.h>
  39 #include "util.h"
  40
  41 #define MQUEUE_MAGIC    0x19800202
  42 #define DIRENT_SIZE     20
  43 #define FILENT_SIZE     80
  44
  45 #define SEND            0
  46 #define RECV            1
  47
  48 #define STATE_NONE      0
  49 #define STATE_PENDING   1
  50 #define STATE_READY     2
  51
  52 struct ext_wait_queue {         /* queue of sleeping tasks */
  53         struct task_struct *task;
  54         struct list_head list;
  55         struct msg_msg *msg;    /* ptr of loaded message */
  56         int state;              /* one of STATE_* values */
  57 };
  58
  59 struct mqueue_inode_info {
  60         spinlock_t lock;
  61         struct inode vfs_inode;
  62         wait_queue_head_t wait_q;
  63
  64         struct msg_msg **messages;
  65         struct mq_attr attr;
  66
  67         struct sigevent notify;
  68         struct pid* notify_owner;
  69         struct user_struct *user;       /* user who created, for accounting */
  70         struct sock *notify_sock;
  71         struct sk_buff *notify_cookie;
  72
  73         /* for tasks waiting for free space and messages, respectively */
  74         struct ext_wait_queue e_wait_q[2];
  75
  76         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  77 };
  78
  79 static const struct inode_operations mqueue_dir_inode_operations;
  80 static const struct file_operations mqueue_file_operations;
  81 static const struct super_operations mqueue_super_ops;
  82 static void remove_notification(struct mqueue_inode_info *info);
  83
  84 static struct kmem_cache *mqueue_inode_cachep;
  85
  86 static struct ctl_table_header * mq_sysctl_table;
  87
  88 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  89 {
  90         return container_of(inode, struct mqueue_inode_info, vfs_inode);
  91 }
  92
  93 /*
  94  * This routine should be called with the mq_lock held.
  95  */
  96 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
  97 {
  98         return get_ipc_ns(inode->i_sb->s_fs_info);
  99 }
 100
 101 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 102 {
 103         struct ipc_namespace *ns;
 104
 105         spin_lock(&mq_lock);
 106         ns = __get_ns_from_inode(inode);
 107         spin_unlock(&mq_lock);
 108         return ns;
 109 }
 110
 111 static struct inode *mqueue_get_inode(struct super_block *sb,
 112                 struct ipc_namespace *ipc_ns, umode_t mode,
 113                 struct mq_attr *attr)
 114 {
 115         struct user_struct *u = current_user();
 116         struct inode *inode;
 117         int ret = -ENOMEM;
 118
 119         inode = new_inode(sb);
 120         if (!inode)
 121                 goto err;
 122
 123         inode->i_ino = get_next_ino();
 124         inode->i_mode = mode;
 125         inode->i_uid = current_fsuid();
 126         inode->i_gid = current_fsgid();
 127         inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
 128
 129         if (S_ISREG(mode)) {
 130                 struct mqueue_inode_info *info;
 131                 unsigned long mq_bytes, mq_msg_tblsz;
 132
 133                 inode->i_fop = &mqueue_file_operations;
 134                 inode->i_size = FILENT_SIZE;
 135                 /* mqueue specific info */
 136                 info = MQUEUE_I(inode);
 137                 spin_lock_init(&info->lock);
 138                 init_waitqueue_head(&info->wait_q);
 139                 INIT_LIST_HEAD(&info->e_wait_q[0].list);
 140                 INIT_LIST_HEAD(&info->e_wait_q[1].list);
 141                 info->notify_owner = NULL;
 142                 info->qsize = 0;
 143                 info->user = NULL;      /* set when all is ok */
 144                 memset(&info->attr, 0, sizeof(info->attr));
 145                 info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
 146                 info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
 147                 if (attr) {
 148                         info->attr.mq_maxmsg = attr->mq_maxmsg;
 149                         info->attr.mq_msgsize = attr->mq_msgsize;
 150                 }
 151                 mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 152                 info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 153                 if (!info->messages)
 154                         goto out_inode;
 155
 156                 mq_bytes = (mq_msg_tblsz +
 157                         (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 158
 159                 spin_lock(&mq_lock);
 160                 if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 161                     u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
 162                         spin_unlock(&mq_lock);
 163                         /* mqueue_evict_inode() releases info->messages */
 164                         ret = -EMFILE;
 165                         goto out_inode;
 166                 }
 167                 u->mq_bytes += mq_bytes;
 168                 spin_unlock(&mq_lock);
 169
 170                 /* all is ok */
 171                 info->user = get_uid(u);
 172         } else if (S_ISDIR(mode)) {
 173                 inc_nlink(inode);
 174                 /* Some things misbehave if size == 0 on a directory */
 175                 inode->i_size = 2 * DIRENT_SIZE;
 176                 inode->i_op = &mqueue_dir_inode_operations;
 177                 inode->i_fop = &simple_dir_operations;
 178         }
 179
 180         return inode;
 181 out_inode:
 182         iput(inode);
 183 err:
 184         return ERR_PTR(ret);
 185 }
 186
 187 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 188 {
 189         struct inode *inode;
 190         struct ipc_namespace *ns = data;
 191
 192         sb->s_blocksize = PAGE_CACHE_SIZE;
 193         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 194         sb->s_magic = MQUEUE_MAGIC;
 195         sb->s_op = &mqueue_super_ops;
 196
 197         inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 198         if (IS_ERR(inode))
 199                 return PTR_ERR(inode);
 200
 201         sb->s_root = d_make_root(inode);
 202         if (!sb->s_root)
 203                 return -ENOMEM;
 204         return 0;
 205 }
 206
 207 static struct dentry *mqueue_mount(struct file_system_type *fs_type,
 208                          int flags, const char *dev_name,
 209                          void *data)
 210 {
 211         if (!(flags & MS_KERNMOUNT))
 212                 data = current->nsproxy->ipc_ns;
 213         return mount_ns(fs_type, flags, data, mqueue_fill_super);
 214 }
 215
 216 static void init_once(void *foo)
 217 {
 218         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 219
 220         inode_init_once(&p->vfs_inode);
 221 }
 222
 223 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 224 {
 225         struct mqueue_inode_info *ei;
 226
 227         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 228         if (!ei)
 229                 return NULL;
 230         return &ei->vfs_inode;
 231 }
 232
 233 static void mqueue_i_callback(struct rcu_head *head)
 234 {
 235         struct inode *inode = container_of(head, struct inode, i_rcu);
 236         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 237 }
 238
 239 static void mqueue_destroy_inode(struct inode *inode)
 240 {
 241         call_rcu(&inode->i_rcu, mqueue_i_callback);
 242 }
 243
 244 static void mqueue_evict_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         end_writeback(inode);
 253
 254         if (S_ISDIR(inode->i_mode))
 255                 return;
 256
 257         ipc_ns = get_ns_from_inode(inode);
 258         info = MQUEUE_I(inode);
 259         spin_lock(&info->lock);
 260         for (i = 0; i < info->attr.mq_curmsgs; i++)
 261                 free_msg(info->messages[i]);
 262         kfree(info->messages);
 263         spin_unlock(&info->lock);
 264
 265         /* Total amount of bytes accounted for the mqueue */
 266         mq_bytes = info->attr.mq_maxmsg * (sizeof(struct msg_msg *)
 267             + info->attr.mq_msgsize);
 268         user = info->user;
 269         if (user) {
 270                 spin_lock(&mq_lock);
 271                 user->mq_bytes -= mq_bytes;
 272                 /*
 273                  * get_ns_from_inode() ensures that the
 274                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 275                  * to which we now hold a reference, or it is NULL.
 276                  * We can't put it here under mq_lock, though.
 277                  */
 278                 if (ipc_ns)
 279                         ipc_ns->mq_queues_count--;
 280                 spin_unlock(&mq_lock);
 281                 free_uid(user);
 282         }
 283         if (ipc_ns)
 284                 put_ipc_ns(ipc_ns);
 285 }
 286
 287 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 288                                 umode_t mode, struct nameidata *nd)
 289 {
 290         struct inode *inode;
 291         struct mq_attr *attr = dentry->d_fsdata;
 292         int error;
 293         struct ipc_namespace *ipc_ns;
 294
 295         spin_lock(&mq_lock);
 296         ipc_ns = __get_ns_from_inode(dir);
 297         if (!ipc_ns) {
 298                 error = -EACCES;
 299                 goto out_unlock;
 300         }
 301         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 302                         !capable(CAP_SYS_RESOURCE)) {
 303                 error = -ENOSPC;
 304                 goto out_unlock;
 305         }
 306         ipc_ns->mq_queues_count++;
 307         spin_unlock(&mq_lock);
 308
 309         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 310         if (IS_ERR(inode)) {
 311                 error = PTR_ERR(inode);
 312                 spin_lock(&mq_lock);
 313                 ipc_ns->mq_queues_count--;
 314                 goto out_unlock;
 315         }
 316
 317         put_ipc_ns(ipc_ns);
 318         dir->i_size += DIRENT_SIZE;
 319         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 320
 321         d_instantiate(dentry, inode);
 322         dget(dentry);
 323         return 0;
 324 out_unlock:
 325         spin_unlock(&mq_lock);
 326         if (ipc_ns)
 327                 put_ipc_ns(ipc_ns);
 328         return error;
 329 }
 330
 331 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 332 {
 333         struct inode *inode = dentry->d_inode;
 334
 335         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 336         dir->i_size -= DIRENT_SIZE;
 337         drop_nlink(inode);
 338         dput(dentry);
 339         return 0;
 340 }
 341
 342 /*
 343 *       This is routine for system read from queue file.
 344 *       To avoid mess with doing here some sort of mq_receive we allow
 345 *       to read only queue size & notification info (the only values
 346 *       that are interesting from user point of view and aren't accessible
 347 *       through std routines)
 348 */
 349 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 350                                 size_t count, loff_t *off)
 351 {
 352         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 353         char buffer[FILENT_SIZE];
 354         ssize_t ret;
 355
 356         spin_lock(&info->lock);
 357         snprintf(buffer, sizeof(buffer),
 358                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 359                         info->qsize,
 360                         info->notify_owner ? info->notify.sigev_notify : 0,
 361                         (info->notify_owner &&
 362                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 363                                 info->notify.sigev_signo : 0,
 364                         pid_vnr(info->notify_owner));
 365         spin_unlock(&info->lock);
 366         buffer[sizeof(buffer)-1] = '\0';
 367
 368         ret = simple_read_from_buffer(u_data, count, off, buffer,
 369                                 strlen(buffer));
 370         if (ret <= 0)
 371                 return ret;
 372
 373         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 374         return ret;
 375 }
 376
 377 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 378 {
 379         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 380
 381         spin_lock(&info->lock);
 382         if (task_tgid(current) == info->notify_owner)
 383                 remove_notification(info);
 384
 385         spin_unlock(&info->lock);
 386         return 0;
 387 }
 388
 389 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 390 {
 391         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 392         int retval = 0;
 393
 394         poll_wait(filp, &info->wait_q, poll_tab);
 395
 396         spin_lock(&info->lock);
 397         if (info->attr.mq_curmsgs)
 398                 retval = POLLIN | POLLRDNORM;
 399
 400         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 401                 retval |= POLLOUT | POLLWRNORM;
 402         spin_unlock(&info->lock);
 403
 404         return retval;
 405 }
 406
 407 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 408 static void wq_add(struct mqueue_inode_info *info, int sr,
 409                         struct ext_wait_queue *ewp)
 410 {
 411         struct ext_wait_queue *walk;
 412
 413         ewp->task = current;
 414
 415         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 416                 if (walk->task->static_prio <= current->static_prio) {
 417                         list_add_tail(&ewp->list, &walk->list);
 418                         return;
 419                 }
 420         }
 421         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 422 }
 423
 424 /*
 425  * Puts current task to sleep. Caller must hold queue lock. After return
 426  * lock isn't held.
 427  * sr: SEND or RECV
 428  */
 429 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 430                     ktime_t *timeout, struct ext_wait_queue *ewp)
 431 {
 432         int retval;
 433         signed long time;
 434
 435         wq_add(info, sr, ewp);
 436
 437         for (;;) {
 438                 set_current_state(TASK_INTERRUPTIBLE);
 439
 440                 spin_unlock(&info->lock);
 441                 time = schedule_hrtimeout_range_clock(timeout, 0,
 442                         HRTIMER_MODE_ABS, CLOCK_REALTIME);
 443
 444                 while (ewp->state == STATE_PENDING)
 445                         cpu_relax();
 446
 447                 if (ewp->state == STATE_READY) {
 448                         retval = 0;
 449                         goto out;
 450                 }
 451                 spin_lock(&info->lock);
 452                 if (ewp->state == STATE_READY) {
 453                         retval = 0;
 454                         goto out_unlock;
 455                 }
 456                 if (signal_pending(current)) {
 457                         retval = -ERESTARTSYS;
 458                         break;
 459                 }
 460                 if (time == 0) {
 461                         retval = -ETIMEDOUT;
 462                         break;
 463                 }
 464         }
 465         list_del(&ewp->list);
 466 out_unlock:
 467         spin_unlock(&info->lock);
 468 out:
 469         return retval;
 470 }
 471
 472 /*
 473  * Returns waiting task that should be serviced first or NULL if none exists
 474  */
 475 static struct ext_wait_queue *wq_get_first_waiter(
 476                 struct mqueue_inode_info *info, int sr)
 477 {
 478         struct list_head *ptr;
 479
 480         ptr = info->e_wait_q[sr].list.prev;
 481         if (ptr == &info->e_wait_q[sr].list)
 482                 return NULL;
 483         return list_entry(ptr, struct ext_wait_queue, list);
 484 }
 485
 486 /* Auxiliary functions to manipulate messages' list */
 487 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 488 {
 489         int k;
 490
 491         k = info->attr.mq_curmsgs - 1;
 492         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 493                 info->messages[k + 1] = info->messages[k];
 494                 k--;
 495         }
 496         info->attr.mq_curmsgs++;
 497         info->qsize += ptr->m_ts;
 498         info->messages[k + 1] = ptr;
 499 }
 500
 501 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 502 {
 503         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 504         return info->messages[info->attr.mq_curmsgs];
 505 }
 506
 507 static inline void set_cookie(struct sk_buff *skb, char code)
 508 {
 509         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 510 }
 511
 512 /*
 513  * The next function is only to split too long sys_mq_timedsend
 514  */
 515 static void __do_notify(struct mqueue_inode_info *info)
 516 {
 517         /* notification
 518          * invoked when there is registered process and there isn't process
 519          * waiting synchronously for message AND state of queue changed from
 520          * empty to not empty. Here we are sure that no one is waiting
 521          * synchronously. */
 522         if (info->notify_owner &&
 523             info->attr.mq_curmsgs == 1) {
 524                 struct siginfo sig_i;
 525                 switch (info->notify.sigev_notify) {
 526                 case SIGEV_NONE:
 527                         break;
 528                 case SIGEV_SIGNAL:
 529                         /* sends signal */
 530
 531                         sig_i.si_signo = info->notify.sigev_signo;
 532                         sig_i.si_errno = 0;
 533                         sig_i.si_code = SI_MESGQ;
 534                         sig_i.si_value = info->notify.sigev_value;
 535                         /* map current pid/uid into info->owner's namespaces */
 536                         rcu_read_lock();
 537                         sig_i.si_pid = task_tgid_nr_ns(current,
 538                                                 ns_of_pid(info->notify_owner));
 539                         sig_i.si_uid = user_ns_map_uid(info->user->user_ns,
 540                                                 current_cred(), current_uid());
 541                         rcu_read_unlock();
 542
 543                         kill_pid_info(info->notify.sigev_signo,
 544                                       &sig_i, info->notify_owner);
 545                         break;
 546                 case SIGEV_THREAD:
 547                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 548                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 549                         break;
 550                 }
 551                 /* after notification unregisters process */
 552                 put_pid(info->notify_owner);
 553                 info->notify_owner = NULL;
 554         }
 555         wake_up(&info->wait_q);
 556 }
 557
 558 static int prepare_timeout(const struct timespec __user *u_abs_timeout,
 559                            ktime_t *expires, struct timespec *ts)
 560 {
 561         if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
 562                 return -EFAULT;
 563         if (!timespec_valid(ts))
 564                 return -EINVAL;
 565
 566         *expires = timespec_to_ktime(*ts);
 567         return 0;
 568 }
 569
 570 static void remove_notification(struct mqueue_inode_info *info)
 571 {
 572         if (info->notify_owner != NULL &&
 573             info->notify.sigev_notify == SIGEV_THREAD) {
 574                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 575                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 576         }
 577         put_pid(info->notify_owner);
 578         info->notify_owner = NULL;
 579 }
 580
 581 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
 582 {
 583         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 584                 return 0;
 585         if (capable(CAP_SYS_RESOURCE)) {
 586                 if (attr->mq_maxmsg > HARD_MSGMAX)
 587                         return 0;
 588         } else {
 589                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
 590                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
 591                         return 0;
 592         }
 593         /* check for overflow */
 594         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 595                 return 0;
 596         if ((unsigned long)(attr->mq_maxmsg * (attr->mq_msgsize
 597             + sizeof (struct msg_msg *))) <
 598             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 599                 return 0;
 600         return 1;
 601 }
 602
 603 /*
 604  * Invoked when creating a new queue via sys_mq_open
 605  */
 606 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
 607                         struct dentry *dentry, int oflag, umode_t mode,
 608                         struct mq_attr *attr)
 609 {
 610         const struct cred *cred = current_cred();
 611         struct file *result;
 612         int ret;
 613
 614         if (attr) {
 615                 if (!mq_attr_ok(ipc_ns, attr)) {
 616                         ret = -EINVAL;
 617                         goto out;
 618                 }
 619                 /* store for use during create */
 620                 dentry->d_fsdata = attr;
 621         }
 622
 623         mode &= ~current_umask();
 624         ret = mnt_want_write(ipc_ns->mq_mnt);
 625         if (ret)
 626                 goto out;
 627         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 628         dentry->d_fsdata = NULL;
 629         if (ret)
 630                 goto out_drop_write;
 631
 632         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 633         /*
 634          * dentry_open() took a persistent mnt_want_write(),
 635          * so we can now drop this one.
 636          */
 637         mnt_drop_write(ipc_ns->mq_mnt);
 638         return result;
 639
 640 out_drop_write:
 641         mnt_drop_write(ipc_ns->mq_mnt);
 642 out:
 643         dput(dentry);
 644         mntput(ipc_ns->mq_mnt);
 645         return ERR_PTR(ret);
 646 }
 647
 648 /* Opens existing queue */
 649 static struct file *do_open(struct ipc_namespace *ipc_ns,
 650                                 struct dentry *dentry, int oflag)
 651 {
 652         int ret;
 653         const struct cred *cred = current_cred();
 654
 655         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 656                                                   MAY_READ | MAY_WRITE };
 657
 658         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 659                 ret = -EINVAL;
 660                 goto err;
 661         }
 662
 663         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 664                 ret = -EACCES;
 665                 goto err;
 666         }
 667
 668         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 669
 670 err:
 671         dput(dentry);
 672         mntput(ipc_ns->mq_mnt);
 673         return ERR_PTR(ret);
 674 }
 675
 676 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
 677                 struct mq_attr __user *, u_attr)
 678 {
 679         struct dentry *dentry;
 680         struct file *filp;
 681         char *name;
 682         struct mq_attr attr;
 683         int fd, error;
 684         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 685
 686         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 687                 return -EFAULT;
 688
 689         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
 690
 691         if (IS_ERR(name = getname(u_name)))
 692                 return PTR_ERR(name);
 693
 694         fd = get_unused_fd_flags(O_CLOEXEC);
 695         if (fd < 0)
 696                 goto out_putname;
 697
 698         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 699         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 700         if (IS_ERR(dentry)) {
 701                 error = PTR_ERR(dentry);
 702                 goto out_putfd;
 703         }
 704         mntget(ipc_ns->mq_mnt);
 705
 706         if (oflag & O_CREAT) {
 707                 if (dentry->d_inode) {  /* entry already exists */
 708                         audit_inode(name, dentry);
 709                         if (oflag & O_EXCL) {
 710                                 error = -EEXIST;
 711                                 goto out;
 712                         }
 713                         filp = do_open(ipc_ns, dentry, oflag);
 714                 } else {
 715                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
 716                                                 dentry, oflag, mode,
 717                                                 u_attr ? &attr : NULL);
 718                 }
 719         } else {
 720                 if (!dentry->d_inode) {
 721                         error = -ENOENT;
 722                         goto out;
 723                 }
 724                 audit_inode(name, dentry);
 725                 filp = do_open(ipc_ns, dentry, oflag);
 726         }
 727
 728         if (IS_ERR(filp)) {
 729                 error = PTR_ERR(filp);
 730                 goto out_putfd;
 731         }
 732
 733         fd_install(fd, filp);
 734         goto out_upsem;
 735
 736 out:
 737         dput(dentry);
 738         mntput(ipc_ns->mq_mnt);
 739 out_putfd:
 740         put_unused_fd(fd);
 741         fd = error;
 742 out_upsem:
 743         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 744 out_putname:
 745         putname(name);
 746         return fd;
 747 }
 748
 749 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 750 {
 751         int err;
 752         char *name;
 753         struct dentry *dentry;
 754         struct inode *inode = NULL;
 755         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 756
 757         name = getname(u_name);
 758         if (IS_ERR(name))
 759                 return PTR_ERR(name);
 760
 761         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
 762                         I_MUTEX_PARENT);
 763         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 764         if (IS_ERR(dentry)) {
 765                 err = PTR_ERR(dentry);
 766                 goto out_unlock;
 767         }
 768
 769         if (!dentry->d_inode) {
 770                 err = -ENOENT;
 771                 goto out_err;
 772         }
 773
 774         inode = dentry->d_inode;
 775         if (inode)
 776                 ihold(inode);
 777         err = mnt_want_write(ipc_ns->mq_mnt);
 778         if (err)
 779                 goto out_err;
 780         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 781         mnt_drop_write(ipc_ns->mq_mnt);
 782 out_err:
 783         dput(dentry);
 784
 785 out_unlock:
 786         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 787         putname(name);
 788         if (inode)
 789                 iput(inode);
 790
 791         return err;
 792 }
 793
 794 /* Pipelined send and receive functions.
 795  *
 796  * If a receiver finds no waiting message, then it registers itself in the
 797  * list of waiting receivers. A sender checks that list before adding the new
 798  * message into the message array. If there is a waiting receiver, then it
 799  * bypasses the message array and directly hands the message over to the
 800  * receiver.
 801  * The receiver accepts the message and returns without grabbing the queue
 802  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 803  * are necessary. The same algorithm is used for sysv semaphores, see
 804  * ipc/sem.c for more details.
 805  *
 806  * The same algorithm is used for senders.
 807  */
 808
 809 /* pipelined_send() - send a message directly to the task waiting in
 810  * sys_mq_timedreceive() (without inserting message into a queue).
 811  */
 812 static inline void pipelined_send(struct mqueue_inode_info *info,
 813                                   struct msg_msg *message,
 814                                   struct ext_wait_queue *receiver)
 815 {
 816         receiver->msg = message;
 817         list_del(&receiver->list);
 818         receiver->state = STATE_PENDING;
 819         wake_up_process(receiver->task);
 820         smp_wmb();
 821         receiver->state = STATE_READY;
 822 }
 823
 824 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 825  * gets its message and put to the queue (we have one free place for sure). */
 826 static inline void pipelined_receive(struct mqueue_inode_info *info)
 827 {
 828         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 829
 830         if (!sender) {
 831                 /* for poll */
 832                 wake_up_interruptible(&info->wait_q);
 833                 return;
 834         }
 835         msg_insert(sender->msg, info);
 836         list_del(&sender->list);
 837         sender->state = STATE_PENDING;
 838         wake_up_process(sender->task);
 839         smp_wmb();
 840         sender->state = STATE_READY;
 841 }
 842
 843 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 844                 size_t, msg_len, unsigned int, msg_prio,
 845                 const struct timespec __user *, u_abs_timeout)
 846 {
 847         struct file *filp;
 848         struct inode *inode;
 849         struct ext_wait_queue wait;
 850         struct ext_wait_queue *receiver;
 851         struct msg_msg *msg_ptr;
 852         struct mqueue_inode_info *info;
 853         ktime_t expires, *timeout = NULL;
 854         struct timespec ts;
 855         int ret;
 856
 857         if (u_abs_timeout) {
 858                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 859                 if (res)
 860                         return res;
 861                 timeout = &expires;
 862         }
 863
 864         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 865                 return -EINVAL;
 866
 867         audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
 868
 869         filp = fget(mqdes);
 870         if (unlikely(!filp)) {
 871                 ret = -EBADF;
 872                 goto out;
 873         }
 874
 875         inode = filp->f_path.dentry->d_inode;
 876         if (unlikely(filp->f_op != &mqueue_file_operations)) {
 877                 ret = -EBADF;
 878                 goto out_fput;
 879         }
 880         info = MQUEUE_I(inode);
 881         audit_inode(NULL, filp->f_path.dentry);
 882
 883         if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
 884                 ret = -EBADF;
 885                 goto out_fput;
 886         }
 887
 888         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 889                 ret = -EMSGSIZE;
 890                 goto out_fput;
 891         }
 892
 893         /* First try to allocate memory, before doing anything with
 894          * existing queues. */
 895         msg_ptr = load_msg(u_msg_ptr, msg_len);
 896         if (IS_ERR(msg_ptr)) {
 897                 ret = PTR_ERR(msg_ptr);
 898                 goto out_fput;
 899         }
 900         msg_ptr->m_ts = msg_len;
 901         msg_ptr->m_type = msg_prio;
 902
 903         spin_lock(&info->lock);
 904
 905         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 906                 if (filp->f_flags & O_NONBLOCK) {
 907                         spin_unlock(&info->lock);
 908                         ret = -EAGAIN;
 909                 } else {
 910                         wait.task = current;
 911                         wait.msg = (void *) msg_ptr;
 912                         wait.state = STATE_NONE;
 913                         ret = wq_sleep(info, SEND, timeout, &wait);
 914                 }
 915                 if (ret < 0)
 916                         free_msg(msg_ptr);
 917         } else {
 918                 receiver = wq_get_first_waiter(info, RECV);
 919                 if (receiver) {
 920                         pipelined_send(info, msg_ptr, receiver);
 921                 } else {
 922                         /* adds message to the queue */
 923                         msg_insert(msg_ptr, info);
 924                         __do_notify(info);
 925                 }
 926                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 927                                 CURRENT_TIME;
 928                 spin_unlock(&info->lock);
 929                 ret = 0;
 930         }
 931 out_fput:
 932         fput(filp);
 933 out:
 934         return ret;
 935 }
 936
 937 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
 938                 size_t, msg_len, unsigned int __user *, u_msg_prio,
 939                 const struct timespec __user *, u_abs_timeout)
 940 {
 941         ssize_t ret;
 942         struct msg_msg *msg_ptr;
 943         struct file *filp;
 944         struct inode *inode;
 945         struct mqueue_inode_info *info;
 946         struct ext_wait_queue wait;
 947         ktime_t expires, *timeout = NULL;
 948         struct timespec ts;
 949
 950         if (u_abs_timeout) {
 951                 int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 952                 if (res)
 953                         return res;
 954                 timeout = &expires;
 955         }
 956
 957         audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
 958
 959         filp = fget(mqdes);
 960         if (unlikely(!filp)) {
 961                 ret = -EBADF;
 962                 goto out;
 963         }
 964
 965         inode = filp->f_path.dentry->d_inode;
 966         if (unlikely(filp->f_op != &mqueue_file_operations)) {
 967                 ret = -EBADF;
 968                 goto out_fput;
 969         }
 970         info = MQUEUE_I(inode);
 971         audit_inode(NULL, filp->f_path.dentry);
 972
 973         if (unlikely(!(filp->f_mode & FMODE_READ))) {
 974                 ret = -EBADF;
 975                 goto out_fput;
 976         }
 977
 978         /* checks if buffer is big enough */
 979         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 980                 ret = -EMSGSIZE;
 981                 goto out_fput;
 982         }
 983
 984         spin_lock(&info->lock);
 985         if (info->attr.mq_curmsgs == 0) {
 986                 if (filp->f_flags & O_NONBLOCK) {
 987                         spin_unlock(&info->lock);
 988                         ret = -EAGAIN;
 989                 } else {
 990                         wait.task = current;
 991                         wait.state = STATE_NONE;
 992                         ret = wq_sleep(info, RECV, timeout, &wait);
 993                         msg_ptr = wait.msg;
 994                 }
 995         } else {
 996                 msg_ptr = msg_get(info);
 997
 998                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 999                                 CURRENT_TIME;
1000
1001                 /* There is now free space in queue. */
1002                 pipelined_receive(info);
1003                 spin_unlock(&info->lock);
1004                 ret = 0;
1005         }
1006         if (ret == 0) {
1007                 ret = msg_ptr->m_ts;
1008
1009                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1010                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1011                         ret = -EFAULT;
1012                 }
1013                 free_msg(msg_ptr);
1014         }
1015 out_fput:
1016         fput(filp);
1017 out:
1018         return ret;
1019 }
1020
1021 /*
1022  * Notes: the case when user wants us to deregister (with NULL as pointer)
1023  * and he isn't currently owner of notification, will be silently discarded.
1024  * It isn't explicitly defined in the POSIX.
1025  */
1026 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1027                 const struct sigevent __user *, u_notification)
1028 {
1029         int ret;
1030         struct file *filp;
1031         struct sock *sock;
1032         struct inode *inode;
1033         struct sigevent notification;
1034         struct mqueue_inode_info *info;
1035         struct sk_buff *nc;
1036
1037         if (u_notification) {
1038                 if (copy_from_user(&notification, u_notification,
1039                                         sizeof(struct sigevent)))
1040                         return -EFAULT;
1041         }
1042
1043         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1044
1045         nc = NULL;
1046         sock = NULL;
1047         if (u_notification != NULL) {
1048                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1049                              notification.sigev_notify != SIGEV_SIGNAL &&
1050                              notification.sigev_notify != SIGEV_THREAD))
1051                         return -EINVAL;
1052                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1053                         !valid_signal(notification.sigev_signo)) {
1054                         return -EINVAL;
1055                 }
1056                 if (notification.sigev_notify == SIGEV_THREAD) {
1057                         long timeo;
1058
1059                         /* create the notify skb */
1060                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1061                         if (!nc) {
1062                                 ret = -ENOMEM;
1063                                 goto out;
1064                         }
1065                         if (copy_from_user(nc->data,
1066                                         notification.sigev_value.sival_ptr,
1067                                         NOTIFY_COOKIE_LEN)) {
1068                                 ret = -EFAULT;
1069                                 goto out;
1070                         }
1071
1072                         /* TODO: add a header? */
1073                         skb_put(nc, NOTIFY_COOKIE_LEN);
1074                         /* and attach it to the socket */
1075 retry:
1076                         filp = fget(notification.sigev_signo);
1077                         if (!filp) {
1078                                 ret = -EBADF;
1079                                 goto out;
1080                         }
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         filp = fget(mqdes);
1102         if (!filp) {
1103                 ret = -EBADF;
1104                 goto out;
1105         }
1106
1107         inode = filp->f_path.dentry->d_inode;
1108         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1109                 ret = -EBADF;
1110                 goto out_fput;
1111         }
1112         info = MQUEUE_I(inode);
1113
1114         ret = 0;
1115         spin_lock(&info->lock);
1116         if (u_notification == NULL) {
1117                 if (info->notify_owner == task_tgid(current)) {
1118                         remove_notification(info);
1119                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1120                 }
1121         } else if (info->notify_owner != NULL) {
1122                 ret = -EBUSY;
1123         } else {
1124                 switch (notification.sigev_notify) {
1125                 case SIGEV_NONE:
1126                         info->notify.sigev_notify = SIGEV_NONE;
1127                         break;
1128                 case SIGEV_THREAD:
1129                         info->notify_sock = sock;
1130                         info->notify_cookie = nc;
1131                         sock = NULL;
1132                         nc = NULL;
1133                         info->notify.sigev_notify = SIGEV_THREAD;
1134                         break;
1135                 case SIGEV_SIGNAL:
1136                         info->notify.sigev_signo = notification.sigev_signo;
1137                         info->notify.sigev_value = notification.sigev_value;
1138                         info->notify.sigev_notify = SIGEV_SIGNAL;
1139                         break;
1140                 }
1141
1142                 info->notify_owner = get_pid(task_tgid(current));
1143                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1144         }
1145         spin_unlock(&info->lock);
1146 out_fput:
1147         fput(filp);
1148 out:
1149         if (sock) {
1150                 netlink_detachskb(sock, nc);
1151         } else if (nc) {
1152                 dev_kfree_skb(nc);
1153         }
1154         return ret;
1155 }
1156
1157 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1158                 const struct mq_attr __user *, u_mqstat,
1159                 struct mq_attr __user *, u_omqstat)
1160 {
1161         int ret;
1162         struct mq_attr mqstat, omqstat;
1163         struct file *filp;
1164         struct inode *inode;
1165         struct mqueue_inode_info *info;
1166
1167         if (u_mqstat != NULL) {
1168                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1169                         return -EFAULT;
1170                 if (mqstat.mq_flags & (~O_NONBLOCK))
1171                         return -EINVAL;
1172         }
1173
1174         filp = fget(mqdes);
1175         if (!filp) {
1176                 ret = -EBADF;
1177                 goto out;
1178         }
1179
1180         inode = filp->f_path.dentry->d_inode;
1181         if (unlikely(filp->f_op != &mqueue_file_operations)) {
1182                 ret = -EBADF;
1183                 goto out_fput;
1184         }
1185         info = MQUEUE_I(inode);
1186
1187         spin_lock(&info->lock);
1188
1189         omqstat = info->attr;
1190         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1191         if (u_mqstat) {
1192                 audit_mq_getsetattr(mqdes, &mqstat);
1193                 spin_lock(&filp->f_lock);
1194                 if (mqstat.mq_flags & O_NONBLOCK)
1195                         filp->f_flags |= O_NONBLOCK;
1196                 else
1197                         filp->f_flags &= ~O_NONBLOCK;
1198                 spin_unlock(&filp->f_lock);
1199
1200                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1201         }
1202
1203         spin_unlock(&info->lock);
1204
1205         ret = 0;
1206         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1207                                                 sizeof(struct mq_attr)))
1208                 ret = -EFAULT;
1209
1210 out_fput:
1211         fput(filp);
1212 out:
1213         return ret;
1214 }
1215
1216 static const struct inode_operations mqueue_dir_inode_operations = {
1217         .lookup = simple_lookup,
1218         .create = mqueue_create,
1219         .unlink = mqueue_unlink,
1220 };
1221
1222 static const struct file_operations mqueue_file_operations = {
1223         .flush = mqueue_flush_file,
1224         .poll = mqueue_poll_file,
1225         .read = mqueue_read_file,
1226         .llseek = default_llseek,
1227 };
1228
1229 static const struct super_operations mqueue_super_ops = {
1230         .alloc_inode = mqueue_alloc_inode,
1231         .destroy_inode = mqueue_destroy_inode,
1232         .evict_inode = mqueue_evict_inode,
1233         .statfs = simple_statfs,
1234 };
1235
1236 static struct file_system_type mqueue_fs_type = {
1237         .name = "mqueue",
1238         .mount = mqueue_mount,
1239         .kill_sb = kill_litter_super,
1240 };
1241
1242 int mq_init_ns(struct ipc_namespace *ns)
1243 {
1244         ns->mq_queues_count  = 0;
1245         ns->mq_queues_max    = DFLT_QUEUESMAX;
1246         ns->mq_msg_max       = DFLT_MSGMAX;
1247         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1248
1249         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1250         if (IS_ERR(ns->mq_mnt)) {
1251                 int err = PTR_ERR(ns->mq_mnt);
1252                 ns->mq_mnt = NULL;
1253                 return err;
1254         }
1255         return 0;
1256 }
1257
1258 void mq_clear_sbinfo(struct ipc_namespace *ns)
1259 {
1260         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1261 }
1262
1263 void mq_put_mnt(struct ipc_namespace *ns)
1264 {
1265         kern_unmount(ns->mq_mnt);
1266 }
1267
1268 static int __init init_mqueue_fs(void)
1269 {
1270         int error;
1271
1272         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1273                                 sizeof(struct mqueue_inode_info), 0,
1274                                 SLAB_HWCACHE_ALIGN, init_once);
1275         if (mqueue_inode_cachep == NULL)
1276                 return -ENOMEM;
1277
1278         /* ignore failures - they are not fatal */
1279         mq_sysctl_table = mq_register_sysctl_table();
1280
1281         error = register_filesystem(&mqueue_fs_type);
1282         if (error)
1283                 goto out_sysctl;
1284
1285         spin_lock_init(&mq_lock);
1286
1287         error = mq_init_ns(&init_ipc_ns);
1288         if (error)
1289                 goto out_filesystem;
1290
1291         return 0;
1292
1293 out_filesystem:
1294         unregister_filesystem(&mqueue_fs_type);
1295 out_sysctl:
1296         if (mq_sysctl_table)
1297                 unregister_sysctl_table(mq_sysctl_table);
1298         kmem_cache_destroy(mqueue_inode_cachep);
1299         return error;
1300 }
1301
1302 __initcall(init_mqueue_fs);