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/ima.h>
  36
  37 #include <net/sock.h>
  38 #include "util.h"
  39
  40 #define MQUEUE_MAGIC    0x19800202
  41 #define DIRENT_SIZE     20
  42 #define FILENT_SIZE     80
  43
  44 #define SEND            0
  45 #define RECV            1
  46
  47 #define STATE_NONE      0
  48 #define STATE_PENDING   1
  49 #define STATE_READY     2
  50
  51 struct ext_wait_queue {         /* queue of sleeping tasks */
  52         struct task_struct *task;
  53         struct list_head list;
  54         struct msg_msg *msg;    /* ptr of loaded message */
  55         int state;              /* one of STATE_* values */
  56 };
  57
  58 struct mqueue_inode_info {
  59         spinlock_t lock;
  60         struct inode vfs_inode;
  61         wait_queue_head_t wait_q;
  62
  63         struct msg_msg **messages;
  64         struct mq_attr attr;
  65
  66         struct sigevent notify;
  67         struct pid* notify_owner;
  68         struct user_struct *user;       /* user who created, for accounting */
  69         struct sock *notify_sock;
  70         struct sk_buff *notify_cookie;
  71
  72         /* for tasks waiting for free space and messages, respectively */
  73         struct ext_wait_queue e_wait_q[2];
  74
  75         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  76 };
  77
  78 static const struct inode_operations mqueue_dir_inode_operations;
  79 static const struct file_operations mqueue_file_operations;
  80 static const struct super_operations mqueue_super_ops;
  81 static void remove_notification(struct mqueue_inode_info *info);
  82
  83 static struct kmem_cache *mqueue_inode_cachep;
  84
  85 static struct ctl_table_header * mq_sysctl_table;
  86
  87 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  88 {
  89         return container_of(inode, struct mqueue_inode_info, vfs_inode);
  90 }
  91
  92 /*
  93  * This routine should be called with the mq_lock held.
  94  */
  95 static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
  96 {
  97         return get_ipc_ns(inode->i_sb->s_fs_info);
  98 }
  99
 100 static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 101 {
 102         struct ipc_namespace *ns;
 103
 104         spin_lock(&mq_lock);
 105         ns = __get_ns_from_inode(inode);
 106         spin_unlock(&mq_lock);
 107         return ns;
 108 }
 109
 110 static struct inode *mqueue_get_inode(struct super_block *sb,
 111                 struct ipc_namespace *ipc_ns, int mode,
 112                 struct mq_attr *attr)
 113 {
 114         struct user_struct *u = current_user();
 115         struct inode *inode;
 116
 117         inode = new_inode(sb);
 118         if (inode) {
 119                 inode->i_mode = mode;
 120                 inode->i_uid = current_fsuid();
 121                 inode->i_gid = current_fsgid();
 122                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 123                                 CURRENT_TIME;
 124
 125                 if (S_ISREG(mode)) {
 126                         struct mqueue_inode_info *info;
 127                         struct task_struct *p = current;
 128                         unsigned long mq_bytes, mq_msg_tblsz;
 129
 130                         inode->i_fop = &mqueue_file_operations;
 131                         inode->i_size = FILENT_SIZE;
 132                         /* mqueue specific info */
 133                         info = MQUEUE_I(inode);
 134                         spin_lock_init(&info->lock);
 135                         init_waitqueue_head(&info->wait_q);
 136                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 137                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 138                         info->messages = NULL;
 139                         info->notify_owner = NULL;
 140                         info->qsize = 0;
 141                         info->user = NULL;      /* set when all is ok */
 142                         memset(&info->attr, 0, sizeof(info->attr));
 143                         info->attr.mq_maxmsg = ipc_ns->mq_msg_max;
 144                         info->attr.mq_msgsize = ipc_ns->mq_msgsize_max;
 145                         if (attr) {
 146                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 147                                 info->attr.mq_msgsize = attr->mq_msgsize;
 148                         }
 149                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 150                         mq_bytes = (mq_msg_tblsz +
 151                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 152
 153                         spin_lock(&mq_lock);
 154                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 155                             u->mq_bytes + mq_bytes >
 156                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur) {
 157                                 spin_unlock(&mq_lock);
 158                                 goto out_inode;
 159                         }
 160                         u->mq_bytes += mq_bytes;
 161                         spin_unlock(&mq_lock);
 162
 163                         info->messages = kmalloc(mq_msg_tblsz, GFP_KERNEL);
 164                         if (!info->messages) {
 165                                 spin_lock(&mq_lock);
 166                                 u->mq_bytes -= mq_bytes;
 167                                 spin_unlock(&mq_lock);
 168                                 goto out_inode;
 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         make_bad_inode(inode);
 183         iput(inode);
 184         return NULL;
 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         int error = 0;
 192
 193         sb->s_blocksize = PAGE_CACHE_SIZE;
 194         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 195         sb->s_magic = MQUEUE_MAGIC;
 196         sb->s_op = &mqueue_super_ops;
 197
 198         inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO,
 199                                 NULL);
 200         if (!inode) {
 201                 error = -ENOMEM;
 202                 goto out;
 203         }
 204
 205         sb->s_root = d_alloc_root(inode);
 206         if (!sb->s_root) {
 207                 iput(inode);
 208                 error = -ENOMEM;
 209         }
 210
 211 out:
 212         return error;
 213 }
 214
 215 static int mqueue_get_sb(struct file_system_type *fs_type,
 216                          int flags, const char *dev_name,
 217                          void *data, struct vfsmount *mnt)
 218 {
 219         if (!(flags & MS_KERNMOUNT))
 220                 data = current->nsproxy->ipc_ns;
 221         return get_sb_ns(fs_type, flags, data, mqueue_fill_super, mnt);
 222 }
 223
 224 static void init_once(void *foo)
 225 {
 226         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 227
 228         inode_init_once(&p->vfs_inode);
 229 }
 230
 231 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 232 {
 233         struct mqueue_inode_info *ei;
 234
 235         ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 236         if (!ei)
 237                 return NULL;
 238         return &ei->vfs_inode;
 239 }
 240
 241 static void mqueue_destroy_inode(struct inode *inode)
 242 {
 243         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 244 }
 245
 246 static void mqueue_delete_inode(struct inode *inode)
 247 {
 248         struct mqueue_inode_info *info;
 249         struct user_struct *user;
 250         unsigned long mq_bytes;
 251         int i;
 252         struct ipc_namespace *ipc_ns;
 253
 254         if (S_ISDIR(inode->i_mode)) {
 255                 clear_inode(inode);
 256                 return;
 257         }
 258         ipc_ns = get_ns_from_inode(inode);
 259         info = MQUEUE_I(inode);
 260         spin_lock(&info->lock);
 261         for (i = 0; i < info->attr.mq_curmsgs; i++)
 262                 free_msg(info->messages[i]);
 263         kfree(info->messages);
 264         spin_unlock(&info->lock);
 265
 266         clear_inode(inode);
 267
 268         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 269                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 270         user = info->user;
 271         if (user) {
 272                 spin_lock(&mq_lock);
 273                 user->mq_bytes -= mq_bytes;
 274                 /*
 275                  * get_ns_from_inode() ensures that the
 276                  * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 277                  * to which we now hold a reference, or it is NULL.
 278                  * We can't put it here under mq_lock, though.
 279                  */
 280                 if (ipc_ns)
 281                         ipc_ns->mq_queues_count--;
 282                 spin_unlock(&mq_lock);
 283                 free_uid(user);
 284         }
 285         if (ipc_ns)
 286                 put_ipc_ns(ipc_ns);
 287 }
 288
 289 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 290                                 int mode, struct nameidata *nd)
 291 {
 292         struct inode *inode;
 293         struct mq_attr *attr = dentry->d_fsdata;
 294         int error;
 295         struct ipc_namespace *ipc_ns;
 296
 297         spin_lock(&mq_lock);
 298         ipc_ns = __get_ns_from_inode(dir);
 299         if (!ipc_ns) {
 300                 error = -EACCES;
 301                 goto out_unlock;
 302         }
 303         if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 304                         !capable(CAP_SYS_RESOURCE)) {
 305                 error = -ENOSPC;
 306                 goto out_unlock;
 307         }
 308         ipc_ns->mq_queues_count++;
 309         spin_unlock(&mq_lock);
 310
 311         inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 312         if (!inode) {
 313                 error = -ENOMEM;
 314                 spin_lock(&mq_lock);
 315                 ipc_ns->mq_queues_count--;
 316                 goto out_unlock;
 317         }
 318
 319         put_ipc_ns(ipc_ns);
 320         dir->i_size += DIRENT_SIZE;
 321         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 322
 323         d_instantiate(dentry, inode);
 324         dget(dentry);
 325         return 0;
 326 out_unlock:
 327         spin_unlock(&mq_lock);
 328         if (ipc_ns)
 329                 put_ipc_ns(ipc_ns);
 330         return error;
 331 }
 332
 333 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 334 {
 335         struct inode *inode = dentry->d_inode;
 336
 337         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 338         dir->i_size -= DIRENT_SIZE;
 339         drop_nlink(inode);
 340         dput(dentry);
 341         return 0;
 342 }
 343
 344 /*
 345 *       This is routine for system read from queue file.
 346 *       To avoid mess with doing here some sort of mq_receive we allow
 347 *       to read only queue size & notification info (the only values
 348 *       that are interesting from user point of view and aren't accessible
 349 *       through std routines)
 350 */
 351 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 352                                 size_t count, loff_t *off)
 353 {
 354         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 355         char buffer[FILENT_SIZE];
 356         ssize_t ret;
 357
 358         spin_lock(&info->lock);
 359         snprintf(buffer, sizeof(buffer),
 360                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 361                         info->qsize,
 362                         info->notify_owner ? info->notify.sigev_notify : 0,
 363                         (info->notify_owner &&
 364                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 365                                 info->notify.sigev_signo : 0,
 366                         pid_vnr(info->notify_owner));
 367         spin_unlock(&info->lock);
 368         buffer[sizeof(buffer)-1] = '\0';
 369
 370         ret = simple_read_from_buffer(u_data, count, off, buffer,
 371                                 strlen(buffer));
 372         if (ret <= 0)
 373                 return ret;
 374
 375         filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 376         return ret;
 377 }
 378
 379 static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 380 {
 381         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 382
 383         spin_lock(&info->lock);
 384         if (task_tgid(current) == info->notify_owner)
 385                 remove_notification(info);
 386
 387         spin_unlock(&info->lock);
 388         return 0;
 389 }
 390
 391 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 392 {
 393         struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 394         int retval = 0;
 395
 396         poll_wait(filp, &info->wait_q, poll_tab);
 397
 398         spin_lock(&info->lock);
 399         if (info->attr.mq_curmsgs)
 400                 retval = POLLIN | POLLRDNORM;
 401
 402         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 403                 retval |= POLLOUT | POLLWRNORM;
 404         spin_unlock(&info->lock);
 405
 406         return retval;
 407 }
 408
 409 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 410 static void wq_add(struct mqueue_inode_info *info, int sr,
 411                         struct ext_wait_queue *ewp)
 412 {
 413         struct ext_wait_queue *walk;
 414
 415         ewp->task = current;
 416
 417         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 418                 if (walk->task->static_prio <= current->static_prio) {
 419                         list_add_tail(&ewp->list, &walk->list);
 420                         return;
 421                 }
 422         }
 423         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 424 }
 425
 426 /*
 427  * Puts current task to sleep. Caller must hold queue lock. After return
 428  * lock isn't held.
 429  * sr: SEND or RECV
 430  */
 431 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 432                         long timeout, struct ext_wait_queue *ewp)
 433 {
 434         int retval;
 435         signed long time;
 436
 437         wq_add(info, sr, ewp);
 438
 439         for (;;) {
 440                 set_current_state(TASK_INTERRUPTIBLE);
 441
 442                 spin_unlock(&info->lock);
 443                 time = schedule_timeout(timeout);
 444
 445                 while (ewp->state == STATE_PENDING)
 446                         cpu_relax();
 447
 448                 if (ewp->state == STATE_READY) {
 449                         retval = 0;
 450                         goto out;
 451                 }
 452                 spin_lock(&info->lock);
 453                 if (ewp->state == STATE_READY) {
 454                         retval = 0;
 455                         goto out_unlock;
 456                 }
 457                 if (signal_pending(current)) {
 458                         retval = -ERESTARTSYS;
 459                         break;
 460                 }
 461                 if (time == 0) {
 462                         retval = -ETIMEDOUT;
 463                         break;
 464                 }
 465         }
 466         list_del(&ewp->list);
 467 out_unlock:
 468         spin_unlock(&info->lock);
 469 out:
 470         return retval;
 471 }
 472
 473 /*
 474  * Returns waiting task that should be serviced first or NULL if none exists
 475  */
 476 static struct ext_wait_queue *wq_get_first_waiter(
 477                 struct mqueue_inode_info *info, int sr)
 478 {
 479         struct list_head *ptr;
 480
 481         ptr = info->e_wait_q[sr].list.prev;
 482         if (ptr == &info->e_wait_q[sr].list)
 483                 return NULL;
 484         return list_entry(ptr, struct ext_wait_queue, list);
 485 }
 486
 487 /* Auxiliary functions to manipulate messages' list */
 488 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 489 {
 490         int k;
 491
 492         k = info->attr.mq_curmsgs - 1;
 493         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 494                 info->messages[k + 1] = info->messages[k];
 495                 k--;
 496         }
 497         info->attr.mq_curmsgs++;
 498         info->qsize += ptr->m_ts;
 499         info->messages[k + 1] = ptr;
 500 }
 501
 502 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 503 {
 504         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 505         return info->messages[info->attr.mq_curmsgs];
 506 }
 507
 508 static inline void set_cookie(struct sk_buff *skb, char code)
 509 {
 510         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 511 }
 512
 513 /*
 514  * The next function is only to split too long sys_mq_timedsend
 515  */
 516 static void __do_notify(struct mqueue_inode_info *info)
 517 {
 518         /* notification
 519          * invoked when there is registered process and there isn't process
 520          * waiting synchronously for message AND state of queue changed from
 521          * empty to not empty. Here we are sure that no one is waiting
 522          * synchronously. */
 523         if (info->notify_owner &&
 524             info->attr.mq_curmsgs == 1) {
 525                 struct siginfo sig_i;
 526                 switch (info->notify.sigev_notify) {
 527                 case SIGEV_NONE:
 528                         break;
 529                 case SIGEV_SIGNAL:
 530                         /* sends signal */
 531
 532                         sig_i.si_signo = info->notify.sigev_signo;
 533                         sig_i.si_errno = 0;
 534                         sig_i.si_code = SI_MESGQ;
 535                         sig_i.si_value = info->notify.sigev_value;
 536                         sig_i.si_pid = task_tgid_nr_ns(current,
 537                                                 ns_of_pid(info->notify_owner));
 538                         sig_i.si_uid = current_uid();
 539
 540                         kill_pid_info(info->notify.sigev_signo,
 541                                       &sig_i, info->notify_owner);
 542                         break;
 543                 case SIGEV_THREAD:
 544                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 545                         netlink_sendskb(info->notify_sock, info->notify_cookie);
 546                         break;
 547                 }
 548                 /* after notification unregisters process */
 549                 put_pid(info->notify_owner);
 550                 info->notify_owner = NULL;
 551         }
 552         wake_up(&info->wait_q);
 553 }
 554
 555 static long prepare_timeout(struct timespec *p)
 556 {
 557         struct timespec nowts;
 558         long timeout;
 559
 560         if (p) {
 561                 if (unlikely(p->tv_nsec < 0 || p->tv_sec < 0
 562                         || p->tv_nsec >= NSEC_PER_SEC))
 563                         return -EINVAL;
 564                 nowts = CURRENT_TIME;
 565                 /* first subtract as jiffies can't be too big */
 566                 p->tv_sec -= nowts.tv_sec;
 567                 if (p->tv_nsec < nowts.tv_nsec) {
 568                         p->tv_nsec += NSEC_PER_SEC;
 569                         p->tv_sec--;
 570                 }
 571                 p->tv_nsec -= nowts.tv_nsec;
 572                 if (p->tv_sec < 0)
 573                         return 0;
 574
 575                 timeout = timespec_to_jiffies(p) + 1;
 576         } else
 577                 return MAX_SCHEDULE_TIMEOUT;
 578
 579         return timeout;
 580 }
 581
 582 static void remove_notification(struct mqueue_inode_info *info)
 583 {
 584         if (info->notify_owner != NULL &&
 585             info->notify.sigev_notify == SIGEV_THREAD) {
 586                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 587                 netlink_sendskb(info->notify_sock, info->notify_cookie);
 588         }
 589         put_pid(info->notify_owner);
 590         info->notify_owner = NULL;
 591 }
 592
 593 static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
 594 {
 595         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 596                 return 0;
 597         if (capable(CAP_SYS_RESOURCE)) {
 598                 if (attr->mq_maxmsg > HARD_MSGMAX)
 599                         return 0;
 600         } else {
 601                 if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
 602                                 attr->mq_msgsize > ipc_ns->mq_msgsize_max)
 603                         return 0;
 604         }
 605         /* check for overflow */
 606         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 607                 return 0;
 608         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 609             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 610             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 611                 return 0;
 612         return 1;
 613 }
 614
 615 /*
 616  * Invoked when creating a new queue via sys_mq_open
 617  */
 618 static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
 619                         struct dentry *dentry, int oflag, mode_t mode,
 620                         struct mq_attr *attr)
 621 {
 622         const struct cred *cred = current_cred();
 623         struct file *result;
 624         int ret;
 625
 626         if (attr) {
 627                 ret = -EINVAL;
 628                 if (!mq_attr_ok(ipc_ns, attr))
 629                         goto out;
 630                 /* store for use during create */
 631                 dentry->d_fsdata = attr;
 632         }
 633
 634         mode &= ~current_umask();
 635         ret = mnt_want_write(ipc_ns->mq_mnt);
 636         if (ret)
 637                 goto out;
 638         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 639         dentry->d_fsdata = NULL;
 640         if (ret)
 641                 goto out_drop_write;
 642
 643         result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 644         /*
 645          * dentry_open() took a persistent mnt_want_write(),
 646          * so we can now drop this one.
 647          */
 648         mnt_drop_write(ipc_ns->mq_mnt);
 649         return result;
 650
 651 out_drop_write:
 652         mnt_drop_write(ipc_ns->mq_mnt);
 653 out:
 654         dput(dentry);
 655         mntput(ipc_ns->mq_mnt);
 656         return ERR_PTR(ret);
 657 }
 658
 659 /* Opens existing queue */
 660 static struct file *do_open(struct ipc_namespace *ipc_ns,
 661                                 struct dentry *dentry, int oflag)
 662 {
 663         const struct cred *cred = current_cred();
 664
 665         static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 666                                                   MAY_READ | MAY_WRITE };
 667
 668         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 669                 dput(dentry);
 670                 mntput(ipc_ns->mq_mnt);
 671                 return ERR_PTR(-EINVAL);
 672         }
 673
 674         if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 675                 dput(dentry);
 676                 mntput(ipc_ns->mq_mnt);
 677                 return ERR_PTR(-EACCES);
 678         }
 679
 680         return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 681 }
 682
 683 SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, mode_t, mode,
 684                 struct mq_attr __user *, u_attr)
 685 {
 686         struct dentry *dentry;
 687         struct file *filp;
 688         char *name;
 689         struct mq_attr attr;
 690         int fd, error;
 691         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 692
 693         if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 694                 return -EFAULT;
 695
 696         audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
 697
 698         if (IS_ERR(name = getname(u_name)))
 699                 return PTR_ERR(name);
 700
 701         fd = get_unused_fd_flags(O_CLOEXEC);
 702         if (fd < 0)
 703                 goto out_putname;
 704
 705         mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 706         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 707         if (IS_ERR(dentry)) {
 708                 error = PTR_ERR(dentry);
 709                 goto out_putfd;
 710         }
 711         mntget(ipc_ns->mq_mnt);
 712
 713         if (oflag & O_CREAT) {
 714                 if (dentry->d_inode) {  /* entry already exists */
 715                         audit_inode(name, dentry);
 716                         error = -EEXIST;
 717                         if (oflag & O_EXCL)
 718                                 goto out;
 719                         filp = do_open(ipc_ns, dentry, oflag);
 720                 } else {
 721                         filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
 722                                                 dentry, oflag, mode,
 723                                                 u_attr ? &attr : NULL);
 724                 }
 725         } else {
 726                 error = -ENOENT;
 727                 if (!dentry->d_inode)
 728                         goto out;
 729                 audit_inode(name, dentry);
 730                 filp = do_open(ipc_ns, dentry, oflag);
 731         }
 732
 733         if (IS_ERR(filp)) {
 734                 error = PTR_ERR(filp);
 735                 goto out_putfd;
 736         }
 737         ima_counts_get(filp);
 738
 739         fd_install(fd, filp);
 740         goto out_upsem;
 741
 742 out:
 743         dput(dentry);
 744         mntput(ipc_ns->mq_mnt);
 745 out_putfd:
 746         put_unused_fd(fd);
 747         fd = error;
 748 out_upsem:
 749         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 750 out_putname:
 751         putname(name);
 752         return fd;
 753 }
 754
 755 SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 756 {
 757         int err;
 758         char *name;
 759         struct dentry *dentry;
 760         struct inode *inode = NULL;
 761         struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 762
 763         name = getname(u_name);
 764         if (IS_ERR(name))
 765                 return PTR_ERR(name);
 766
 767         mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
 768                         I_MUTEX_PARENT);
 769         dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 770         if (IS_ERR(dentry)) {
 771                 err = PTR_ERR(dentry);
 772                 goto out_unlock;
 773         }
 774
 775         if (!dentry->d_inode) {
 776                 err = -ENOENT;
 777                 goto out_err;
 778         }
 779
 780         inode = dentry->d_inode;
 781         if (inode)
 782                 atomic_inc(&inode->i_count);
 783         err = mnt_want_write(ipc_ns->mq_mnt);
 784         if (err)
 785                 goto out_err;
 786         err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 787         mnt_drop_write(ipc_ns->mq_mnt);
 788 out_err:
 789         dput(dentry);
 790
 791 out_unlock:
 792         mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 793         putname(name);
 794         if (inode)
 795                 iput(inode);
 796
 797         return err;
 798 }
 799
 800 /* Pipelined send and receive functions.
 801  *
 802  * If a receiver finds no waiting message, then it registers itself in the
 803  * list of waiting receivers. A sender checks that list before adding the new
 804  * message into the message array. If there is a waiting receiver, then it
 805  * bypasses the message array and directly hands the message over to the
 806  * receiver.
 807  * The receiver accepts the message and returns without grabbing the queue
 808  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 809  * are necessary. The same algorithm is used for sysv semaphores, see
 810  * ipc/sem.c for more details.
 811  *
 812  * The same algorithm is used for senders.
 813  */
 814
 815 /* pipelined_send() - send a message directly to the task waiting in
 816  * sys_mq_timedreceive() (without inserting message into a queue).
 817  */
 818 static inline void pipelined_send(struct mqueue_inode_info *info,
 819                                   struct msg_msg *message,
 820                                   struct ext_wait_queue *receiver)
 821 {
 822         receiver->msg = message;
 823         list_del(&receiver->list);
 824         receiver->state = STATE_PENDING;
 825         wake_up_process(receiver->task);
 826         smp_wmb();
 827         receiver->state = STATE_READY;
 828 }
 829
 830 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 831  * gets its message and put to the queue (we have one free place for sure). */
 832 static inline void pipelined_receive(struct mqueue_inode_info *info)
 833 {
 834         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 835
 836         if (!sender) {
 837                 /* for poll */
 838                 wake_up_interruptible(&info->wait_q);
 839                 return;
 840         }
 841         msg_insert(sender->msg, info);
 842         list_del(&sender->list);
 843         sender->state = STATE_PENDING;
 844         wake_up_process(sender->task);
 845         smp_wmb();
 846         sender->state = STATE_READY;
 847 }
 848
 849 SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 850                 size_t, msg_len, unsigned int, msg_prio,
 851                 const struct timespec __user *, u_abs_timeout)
 852 {
 853         struct file *filp;
 854         struct inode *inode;
 855         struct ext_wait_queue wait;
 856         struct ext_wait_queue *receiver;
 857         struct msg_msg *msg_ptr;
 858         struct mqueue_inode_info *info;
 859         struct timespec ts, *p = NULL;
 860         long timeout;
 861         int ret;
 862
 863         if (u_abs_timeout) {
 864                 if (copy_from_user(&ts, u_abs_timeout,
 865                                         sizeof(struct timespec)))
 866                         return -EFAULT;
 867                 p = &ts;
 868         }
 869
 870         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 871                 return -EINVAL;
 872
 873         audit_mq_sendrecv(mqdes, msg_len, msg_prio, p);
 874         timeout = prepare_timeout(p);
 875
 876         ret = -EBADF;
 877         filp = fget(mqdes);
 878         if (unlikely(!filp))
 879                 goto out;
 880
 881         inode = filp->f_path.dentry->d_inode;
 882         if (unlikely(filp->f_op != &mqueue_file_operations))
 883                 goto out_fput;
 884         info = MQUEUE_I(inode);
 885         audit_inode(NULL, filp->f_path.dentry);
 886
 887         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 888                 goto out_fput;
 889
 890         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 891                 ret = -EMSGSIZE;
 892                 goto out_fput;
 893         }
 894
 895         /* First try to allocate memory, before doing anything with
 896          * existing queues. */
 897         msg_ptr = load_msg(u_msg_ptr, msg_len);
 898         if (IS_ERR(msg_ptr)) {
 899                 ret = PTR_ERR(msg_ptr);
 900                 goto out_fput;
 901         }
 902         msg_ptr->m_ts = msg_len;
 903         msg_ptr->m_type = msg_prio;
 904
 905         spin_lock(&info->lock);
 906
 907         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 908                 if (filp->f_flags & O_NONBLOCK) {
 909                         spin_unlock(&info->lock);
 910                         ret = -EAGAIN;
 911                 } else if (unlikely(timeout < 0)) {
 912                         spin_unlock(&info->lock);
 913                         ret = timeout;
 914                 } else {
 915                         wait.task = current;
 916                         wait.msg = (void *) msg_ptr;
 917                         wait.state = STATE_NONE;
 918                         ret = wq_sleep(info, SEND, timeout, &wait);
 919                 }
 920                 if (ret < 0)
 921                         free_msg(msg_ptr);
 922         } else {
 923                 receiver = wq_get_first_waiter(info, RECV);
 924                 if (receiver) {
 925                         pipelined_send(info, msg_ptr, receiver);
 926                 } else {
 927                         /* adds message to the queue */
 928                         msg_insert(msg_ptr, info);
 929                         __do_notify(info);
 930                 }
 931                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 932                                 CURRENT_TIME;
 933                 spin_unlock(&info->lock);
 934                 ret = 0;
 935         }
 936 out_fput:
 937         fput(filp);
 938 out:
 939         return ret;
 940 }
 941
 942 SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
 943                 size_t, msg_len, unsigned int __user *, u_msg_prio,
 944                 const struct timespec __user *, u_abs_timeout)
 945 {
 946         long timeout;
 947         ssize_t ret;
 948         struct msg_msg *msg_ptr;
 949         struct file *filp;
 950         struct inode *inode;
 951         struct mqueue_inode_info *info;
 952         struct ext_wait_queue wait;
 953         struct timespec ts, *p = NULL;
 954
 955         if (u_abs_timeout) {
 956                 if (copy_from_user(&ts, u_abs_timeout,
 957                                         sizeof(struct timespec)))
 958                         return -EFAULT;
 959                 p = &ts;
 960         }
 961
 962         audit_mq_sendrecv(mqdes, msg_len, 0, p);
 963         timeout = prepare_timeout(p);
 964
 965         ret = -EBADF;
 966         filp = fget(mqdes);
 967         if (unlikely(!filp))
 968                 goto out;
 969
 970         inode = filp->f_path.dentry->d_inode;
 971         if (unlikely(filp->f_op != &mqueue_file_operations))
 972                 goto out_fput;
 973         info = MQUEUE_I(inode);
 974         audit_inode(NULL, filp->f_path.dentry);
 975
 976         if (unlikely(!(filp->f_mode & FMODE_READ)))
 977                 goto out_fput;
 978
 979         /* checks if buffer is big enough */
 980         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 981                 ret = -EMSGSIZE;
 982                 goto out_fput;
 983         }
 984
 985         spin_lock(&info->lock);
 986         if (info->attr.mq_curmsgs == 0) {
 987                 if (filp->f_flags & O_NONBLOCK) {
 988                         spin_unlock(&info->lock);
 989                         ret = -EAGAIN;
 990                         msg_ptr = NULL;
 991                 } else if (unlikely(timeout < 0)) {
 992                         spin_unlock(&info->lock);
 993                         ret = timeout;
 994                         msg_ptr = NULL;
 995                 } else {
 996                         wait.task = current;
 997                         wait.state = STATE_NONE;
 998                         ret = wq_sleep(info, RECV, timeout, &wait);
 999                         msg_ptr = wait.msg;
1000                 }
1001         } else {
1002                 msg_ptr = msg_get(info);
1003
1004                 inode->i_atime = inode->i_mtime = inode->i_ctime =
1005                                 CURRENT_TIME;
1006
1007                 /* There is now free space in queue. */
1008                 pipelined_receive(info);
1009                 spin_unlock(&info->lock);
1010                 ret = 0;
1011         }
1012         if (ret == 0) {
1013                 ret = msg_ptr->m_ts;
1014
1015                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1016                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1017                         ret = -EFAULT;
1018                 }
1019                 free_msg(msg_ptr);
1020         }
1021 out_fput:
1022         fput(filp);
1023 out:
1024         return ret;
1025 }
1026
1027 /*
1028  * Notes: the case when user wants us to deregister (with NULL as pointer)
1029  * and he isn't currently owner of notification, will be silently discarded.
1030  * It isn't explicitly defined in the POSIX.
1031  */
1032 SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1033                 const struct sigevent __user *, u_notification)
1034 {
1035         int ret;
1036         struct file *filp;
1037         struct sock *sock;
1038         struct inode *inode;
1039         struct sigevent notification;
1040         struct mqueue_inode_info *info;
1041         struct sk_buff *nc;
1042
1043         if (u_notification) {
1044                 if (copy_from_user(&notification, u_notification,
1045                                         sizeof(struct sigevent)))
1046                         return -EFAULT;
1047         }
1048
1049         audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1050
1051         nc = NULL;
1052         sock = NULL;
1053         if (u_notification != NULL) {
1054                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1055                              notification.sigev_notify != SIGEV_SIGNAL &&
1056                              notification.sigev_notify != SIGEV_THREAD))
1057                         return -EINVAL;
1058                 if (notification.sigev_notify == SIGEV_SIGNAL &&
1059                         !valid_signal(notification.sigev_signo)) {
1060                         return -EINVAL;
1061                 }
1062                 if (notification.sigev_notify == SIGEV_THREAD) {
1063                         long timeo;
1064
1065                         /* create the notify skb */
1066                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1067                         ret = -ENOMEM;
1068                         if (!nc)
1069                                 goto out;
1070                         ret = -EFAULT;
1071                         if (copy_from_user(nc->data,
1072                                         notification.sigev_value.sival_ptr,
1073                                         NOTIFY_COOKIE_LEN)) {
1074                                 goto out;
1075                         }
1076
1077                         /* TODO: add a header? */
1078                         skb_put(nc, NOTIFY_COOKIE_LEN);
1079                         /* and attach it to the socket */
1080 retry:
1081                         filp = fget(notification.sigev_signo);
1082                         ret = -EBADF;
1083                         if (!filp)
1084                                 goto out;
1085                         sock = netlink_getsockbyfilp(filp);
1086                         fput(filp);
1087                         if (IS_ERR(sock)) {
1088                                 ret = PTR_ERR(sock);
1089                                 sock = NULL;
1090                                 goto out;
1091                         }
1092
1093                         timeo = MAX_SCHEDULE_TIMEOUT;
1094                         ret = netlink_attachskb(sock, nc, &timeo, NULL);
1095                         if (ret == 1)
1096                                 goto retry;
1097                         if (ret) {
1098                                 sock = NULL;
1099                                 nc = NULL;
1100                                 goto out;
1101                         }
1102                 }
1103         }
1104
1105         ret = -EBADF;
1106         filp = fget(mqdes);
1107         if (!filp)
1108                 goto out;
1109
1110         inode = filp->f_path.dentry->d_inode;
1111         if (unlikely(filp->f_op != &mqueue_file_operations))
1112                 goto out_fput;
1113         info = MQUEUE_I(inode);
1114
1115         ret = 0;
1116         spin_lock(&info->lock);
1117         if (u_notification == NULL) {
1118                 if (info->notify_owner == task_tgid(current)) {
1119                         remove_notification(info);
1120                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1121                 }
1122         } else if (info->notify_owner != NULL) {
1123                 ret = -EBUSY;
1124         } else {
1125                 switch (notification.sigev_notify) {
1126                 case SIGEV_NONE:
1127                         info->notify.sigev_notify = SIGEV_NONE;
1128                         break;
1129                 case SIGEV_THREAD:
1130                         info->notify_sock = sock;
1131                         info->notify_cookie = nc;
1132                         sock = NULL;
1133                         nc = NULL;
1134                         info->notify.sigev_notify = SIGEV_THREAD;
1135                         break;
1136                 case SIGEV_SIGNAL:
1137                         info->notify.sigev_signo = notification.sigev_signo;
1138                         info->notify.sigev_value = notification.sigev_value;
1139                         info->notify.sigev_notify = SIGEV_SIGNAL;
1140                         break;
1141                 }
1142
1143                 info->notify_owner = get_pid(task_tgid(current));
1144                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1145         }
1146         spin_unlock(&info->lock);
1147 out_fput:
1148         fput(filp);
1149 out:
1150         if (sock) {
1151                 netlink_detachskb(sock, nc);
1152         } else if (nc) {
1153                 dev_kfree_skb(nc);
1154         }
1155         return ret;
1156 }
1157
1158 SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1159                 const struct mq_attr __user *, u_mqstat,
1160                 struct mq_attr __user *, u_omqstat)
1161 {
1162         int ret;
1163         struct mq_attr mqstat, omqstat;
1164         struct file *filp;
1165         struct inode *inode;
1166         struct mqueue_inode_info *info;
1167
1168         if (u_mqstat != NULL) {
1169                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1170                         return -EFAULT;
1171                 if (mqstat.mq_flags & (~O_NONBLOCK))
1172                         return -EINVAL;
1173         }
1174
1175         ret = -EBADF;
1176         filp = fget(mqdes);
1177         if (!filp)
1178                 goto out;
1179
1180         inode = filp->f_path.dentry->d_inode;
1181         if (unlikely(filp->f_op != &mqueue_file_operations))
1182                 goto out_fput;
1183         info = MQUEUE_I(inode);
1184
1185         spin_lock(&info->lock);
1186
1187         omqstat = info->attr;
1188         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1189         if (u_mqstat) {
1190                 audit_mq_getsetattr(mqdes, &mqstat);
1191                 spin_lock(&filp->f_lock);
1192                 if (mqstat.mq_flags & O_NONBLOCK)
1193                         filp->f_flags |= O_NONBLOCK;
1194                 else
1195                         filp->f_flags &= ~O_NONBLOCK;
1196                 spin_unlock(&filp->f_lock);
1197
1198                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1199         }
1200
1201         spin_unlock(&info->lock);
1202
1203         ret = 0;
1204         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1205                                                 sizeof(struct mq_attr)))
1206                 ret = -EFAULT;
1207
1208 out_fput:
1209         fput(filp);
1210 out:
1211         return ret;
1212 }
1213
1214 static const struct inode_operations mqueue_dir_inode_operations = {
1215         .lookup = simple_lookup,
1216         .create = mqueue_create,
1217         .unlink = mqueue_unlink,
1218 };
1219
1220 static const struct file_operations mqueue_file_operations = {
1221         .flush = mqueue_flush_file,
1222         .poll = mqueue_poll_file,
1223         .read = mqueue_read_file,
1224 };
1225
1226 static const struct super_operations mqueue_super_ops = {
1227         .alloc_inode = mqueue_alloc_inode,
1228         .destroy_inode = mqueue_destroy_inode,
1229         .statfs = simple_statfs,
1230         .delete_inode = mqueue_delete_inode,
1231         .drop_inode = generic_delete_inode,
1232 };
1233
1234 static struct file_system_type mqueue_fs_type = {
1235         .name = "mqueue",
1236         .get_sb = mqueue_get_sb,
1237         .kill_sb = kill_litter_super,
1238 };
1239
1240 int mq_init_ns(struct ipc_namespace *ns)
1241 {
1242         ns->mq_queues_count  = 0;
1243         ns->mq_queues_max    = DFLT_QUEUESMAX;
1244         ns->mq_msg_max       = DFLT_MSGMAX;
1245         ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1246
1247         ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1248         if (IS_ERR(ns->mq_mnt)) {
1249                 int err = PTR_ERR(ns->mq_mnt);
1250                 ns->mq_mnt = NULL;
1251                 return err;
1252         }
1253         return 0;
1254 }
1255
1256 void mq_clear_sbinfo(struct ipc_namespace *ns)
1257 {
1258         ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1259 }
1260
1261 void mq_put_mnt(struct ipc_namespace *ns)
1262 {
1263         mntput(ns->mq_mnt);
1264 }
1265
1266 static int __init init_mqueue_fs(void)
1267 {
1268         int error;
1269
1270         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1271                                 sizeof(struct mqueue_inode_info), 0,
1272                                 SLAB_HWCACHE_ALIGN, init_once);
1273         if (mqueue_inode_cachep == NULL)
1274                 return -ENOMEM;
1275
1276         /* ignore failues - they are not fatal */
1277         mq_sysctl_table = mq_register_sysctl_table();
1278
1279         error = register_filesystem(&mqueue_fs_type);
1280         if (error)
1281                 goto out_sysctl;
1282
1283         spin_lock_init(&mq_lock);
1284
1285         init_ipc_ns.mq_mnt = kern_mount_data(&mqueue_fs_type, &init_ipc_ns);
1286         if (IS_ERR(init_ipc_ns.mq_mnt)) {
1287                 error = PTR_ERR(init_ipc_ns.mq_mnt);
1288                 goto out_filesystem;
1289         }
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);