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@motorola.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  * This file is released under the GPL.
  12  */
  13
  14 #include <linux/capability.h>
  15 #include <linux/init.h>
  16 #include <linux/pagemap.h>
  17 #include <linux/file.h>
  18 #include <linux/mount.h>
  19 #include <linux/namei.h>
  20 #include <linux/sysctl.h>
  21 #include <linux/poll.h>
  22 #include <linux/mqueue.h>
  23 #include <linux/msg.h>
  24 #include <linux/skbuff.h>
  25 #include <linux/netlink.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/signal.h>
  28 #include <linux/mutex.h>
  29 #include <linux/vs_context.h>
  30 #include <linux/vs_limit.h>
  31
  32 #include <net/sock.h>
  33 #include "util.h"
  34
  35 #define MQUEUE_MAGIC    0x19800202
  36 #define DIRENT_SIZE     20
  37 #define FILENT_SIZE     80
  38
  39 #define SEND            0
  40 #define RECV            1
  41
  42 #define STATE_NONE      0
  43 #define STATE_PENDING   1
  44 #define STATE_READY     2
  45
  46 /* used by sysctl */
  47 #define FS_MQUEUE       1
  48 #define CTL_QUEUESMAX   2
  49 #define CTL_MSGMAX      3
  50 #define CTL_MSGSIZEMAX  4
  51
  52 /* default values */
  53 #define DFLT_QUEUESMAX  256     /* max number of message queues */
  54 #define DFLT_MSGMAX     10      /* max number of messages in each queue */
  55 #define HARD_MSGMAX     (131072/sizeof(void*))
  56 #define DFLT_MSGSIZEMAX 8192    /* max message size */
  57
  58
  59 struct ext_wait_queue {         /* queue of sleeping tasks */
  60         struct task_struct *task;
  61         struct list_head list;
  62         struct msg_msg *msg;    /* ptr of loaded message */
  63         int state;              /* one of STATE_* values */
  64 };
  65
  66 struct mqueue_inode_info {
  67         spinlock_t lock;
  68         struct inode vfs_inode;
  69         wait_queue_head_t wait_q;
  70
  71         struct msg_msg **messages;
  72         struct mq_attr attr;
  73
  74         struct sigevent notify;
  75         pid_t notify_owner;
  76         struct user_struct *user;       /* user who created, for accounting */
  77         struct sock *notify_sock;
  78         struct sk_buff *notify_cookie;
  79
  80         /* for tasks waiting for free space and messages, respectively */
  81         struct ext_wait_queue e_wait_q[2];
  82
  83         unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  84 };
  85
  86 static struct inode_operations mqueue_dir_inode_operations;
  87 static struct file_operations mqueue_file_operations;
  88 static struct super_operations mqueue_super_ops;
  89 static void remove_notification(struct mqueue_inode_info *info);
  90
  91 static spinlock_t mq_lock;
  92 static kmem_cache_t *mqueue_inode_cachep;
  93 static struct vfsmount *mqueue_mnt;
  94
  95 static unsigned int queues_count;
  96 static unsigned int queues_max  = DFLT_QUEUESMAX;
  97 static unsigned int msg_max     = DFLT_MSGMAX;
  98 static unsigned int msgsize_max = DFLT_MSGSIZEMAX;
  99
 100 static struct ctl_table_header * mq_sysctl_table;
 101
 102 static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
 103 {
 104         return container_of(inode, struct mqueue_inode_info, vfs_inode);
 105 }
 106
 107 static struct inode *mqueue_get_inode(struct super_block *sb, int mode,
 108                                                         struct mq_attr *attr)
 109 {
 110         struct inode *inode;
 111
 112         inode = new_inode(sb);
 113         if (inode) {
 114                 inode->i_mode = mode;
 115                 inode->i_uid = current->fsuid;
 116                 inode->i_gid = current->fsgid;
 117                 inode->i_blksize = PAGE_CACHE_SIZE;
 118                 inode->i_blocks = 0;
 119                 inode->i_mtime = inode->i_ctime = inode->i_atime =
 120                                 CURRENT_TIME;
 121
 122                 if (S_ISREG(mode)) {
 123                         struct mqueue_inode_info *info;
 124                         struct task_struct *p = current;
 125                         struct user_struct *u = p->user;
 126                         unsigned long mq_bytes, mq_msg_tblsz;
 127
 128                         inode->i_fop = &mqueue_file_operations;
 129                         inode->i_size = FILENT_SIZE;
 130                         /* mqueue specific info */
 131                         info = MQUEUE_I(inode);
 132                         spin_lock_init(&info->lock);
 133                         init_waitqueue_head(&info->wait_q);
 134                         INIT_LIST_HEAD(&info->e_wait_q[0].list);
 135                         INIT_LIST_HEAD(&info->e_wait_q[1].list);
 136                         info->messages = NULL;
 137                         info->notify_owner = 0;
 138                         info->qsize = 0;
 139                         info->user = NULL;      /* set when all is ok */
 140                         memset(&info->attr, 0, sizeof(info->attr));
 141                         info->attr.mq_maxmsg = DFLT_MSGMAX;
 142                         info->attr.mq_msgsize = DFLT_MSGSIZEMAX;
 143                         if (attr) {
 144                                 info->attr.mq_maxmsg = attr->mq_maxmsg;
 145                                 info->attr.mq_msgsize = attr->mq_msgsize;
 146                         }
 147                         mq_msg_tblsz = info->attr.mq_maxmsg * sizeof(struct msg_msg *);
 148                         mq_bytes = (mq_msg_tblsz +
 149                                 (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 150
 151                         spin_lock(&mq_lock);
 152                         if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 153                             u->mq_bytes + mq_bytes >
 154                             p->signal->rlim[RLIMIT_MSGQUEUE].rlim_cur ||
 155                             !vx_ipcmsg_avail(p->vx_info, mq_bytes)) {
 156                                 spin_unlock(&mq_lock);
 157                                 goto out_inode;
 158                         }
 159                         u->mq_bytes += mq_bytes;
 160                         vx_ipcmsg_add(p->vx_info, u, 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                                 vx_ipcmsg_sub(p->vx_info, u, mq_bytes);
 168                                 spin_unlock(&mq_lock);
 169                                 goto out_inode;
 170                         }
 171                         /* all is ok */
 172                         info->user = get_uid(u);
 173                 } else if (S_ISDIR(mode)) {
 174                         inode->i_nlink++;
 175                         /* Some things misbehave if size == 0 on a directory */
 176                         inode->i_size = 2 * DIRENT_SIZE;
 177                         inode->i_op = &mqueue_dir_inode_operations;
 178                         inode->i_fop = &simple_dir_operations;
 179                 }
 180         }
 181         return inode;
 182 out_inode:
 183         make_bad_inode(inode);
 184         iput(inode);
 185         return NULL;
 186 }
 187
 188 static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 189 {
 190         struct inode *inode;
 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, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 198         if (!inode)
 199                 return -ENOMEM;
 200
 201         sb->s_root = d_alloc_root(inode);
 202         if (!sb->s_root) {
 203                 iput(inode);
 204                 return -ENOMEM;
 205         }
 206
 207         return 0;
 208 }
 209
 210 static struct super_block *mqueue_get_sb(struct file_system_type *fs_type,
 211                                          int flags, const char *dev_name,
 212                                          void *data)
 213 {
 214         return get_sb_single(fs_type, flags, data, mqueue_fill_super);
 215 }
 216
 217 static void init_once(void *foo, kmem_cache_t * cachep, unsigned long flags)
 218 {
 219         struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 220
 221         if ((flags & (SLAB_CTOR_VERIFY | SLAB_CTOR_CONSTRUCTOR)) ==
 222                 SLAB_CTOR_CONSTRUCTOR)
 223                 inode_init_once(&p->vfs_inode);
 224 }
 225
 226 static struct inode *mqueue_alloc_inode(struct super_block *sb)
 227 {
 228         struct mqueue_inode_info *ei;
 229
 230         ei = kmem_cache_alloc(mqueue_inode_cachep, SLAB_KERNEL);
 231         if (!ei)
 232                 return NULL;
 233         return &ei->vfs_inode;
 234 }
 235
 236 static void mqueue_destroy_inode(struct inode *inode)
 237 {
 238         kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 239 }
 240
 241 static void mqueue_delete_inode(struct inode *inode)
 242 {
 243         struct mqueue_inode_info *info;
 244         struct user_struct *user;
 245         unsigned long mq_bytes;
 246         int i;
 247
 248         if (S_ISDIR(inode->i_mode)) {
 249                 clear_inode(inode);
 250                 return;
 251         }
 252         info = MQUEUE_I(inode);
 253         spin_lock(&info->lock);
 254         for (i = 0; i < info->attr.mq_curmsgs; i++)
 255                 free_msg(info->messages[i]);
 256         kfree(info->messages);
 257         spin_unlock(&info->lock);
 258
 259         clear_inode(inode);
 260
 261         mq_bytes = (info->attr.mq_maxmsg * sizeof(struct msg_msg *) +
 262                    (info->attr.mq_maxmsg * info->attr.mq_msgsize));
 263         user = info->user;
 264         if (user) {
 265                 struct vx_info *vxi = lookup_vx_info(user->xid);
 266
 267                 spin_lock(&mq_lock);
 268                 user->mq_bytes -= mq_bytes;
 269                 vx_ipcmsg_sub(vxi, user, mq_bytes);
 270                 queues_count--;
 271                 spin_unlock(&mq_lock);
 272                 put_vx_info(vxi);
 273                 free_uid(user);
 274         }
 275 }
 276
 277 static int mqueue_create(struct inode *dir, struct dentry *dentry,
 278                                 int mode, struct nameidata *nd)
 279 {
 280         struct inode *inode;
 281         struct mq_attr *attr = dentry->d_fsdata;
 282         int error;
 283
 284         spin_lock(&mq_lock);
 285         if (queues_count >= queues_max && !capable(CAP_SYS_RESOURCE)) {
 286                 error = -ENOSPC;
 287                 goto out_lock;
 288         }
 289         queues_count++;
 290         spin_unlock(&mq_lock);
 291
 292         inode = mqueue_get_inode(dir->i_sb, mode, attr);
 293         if (!inode) {
 294                 error = -ENOMEM;
 295                 spin_lock(&mq_lock);
 296                 queues_count--;
 297                 goto out_lock;
 298         }
 299
 300         dir->i_size += DIRENT_SIZE;
 301         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 302
 303         d_instantiate(dentry, inode);
 304         dget(dentry);
 305         return 0;
 306 out_lock:
 307         spin_unlock(&mq_lock);
 308         return error;
 309 }
 310
 311 static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 312 {
 313         struct inode *inode = dentry->d_inode;
 314
 315         dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 316         dir->i_size -= DIRENT_SIZE;
 317         inode->i_nlink--;
 318         dput(dentry);
 319         return 0;
 320 }
 321
 322 /*
 323 *       This is routine for system read from queue file.
 324 *       To avoid mess with doing here some sort of mq_receive we allow
 325 *       to read only queue size & notification info (the only values
 326 *       that are interesting from user point of view and aren't accessible
 327 *       through std routines)
 328 */
 329 static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 330                                 size_t count, loff_t * off)
 331 {
 332         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 333         char buffer[FILENT_SIZE];
 334         size_t slen;
 335         loff_t o;
 336
 337         if (!count)
 338                 return 0;
 339
 340         spin_lock(&info->lock);
 341         snprintf(buffer, sizeof(buffer),
 342                         "QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 343                         info->qsize,
 344                         info->notify_owner ? info->notify.sigev_notify : 0,
 345                         (info->notify_owner &&
 346                          info->notify.sigev_notify == SIGEV_SIGNAL) ?
 347                                 info->notify.sigev_signo : 0,
 348                         info->notify_owner);
 349         spin_unlock(&info->lock);
 350         buffer[sizeof(buffer)-1] = '\0';
 351         slen = strlen(buffer)+1;
 352
 353         o = *off;
 354         if (o > slen)
 355                 return 0;
 356
 357         if (o + count > slen)
 358                 count = slen - o;
 359
 360         if (copy_to_user(u_data, buffer + o, count))
 361                 return -EFAULT;
 362
 363         *off = o + count;
 364         filp->f_dentry->d_inode->i_atime = filp->f_dentry->d_inode->i_ctime = CURRENT_TIME;
 365         return count;
 366 }
 367
 368 static int mqueue_flush_file(struct file *filp)
 369 {
 370         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 371
 372         spin_lock(&info->lock);
 373         if (current->tgid == info->notify_owner)
 374                 remove_notification(info);
 375
 376         spin_unlock(&info->lock);
 377         return 0;
 378 }
 379
 380 static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 381 {
 382         struct mqueue_inode_info *info = MQUEUE_I(filp->f_dentry->d_inode);
 383         int retval = 0;
 384
 385         poll_wait(filp, &info->wait_q, poll_tab);
 386
 387         spin_lock(&info->lock);
 388         if (info->attr.mq_curmsgs)
 389                 retval = POLLIN | POLLRDNORM;
 390
 391         if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 392                 retval |= POLLOUT | POLLWRNORM;
 393         spin_unlock(&info->lock);
 394
 395         return retval;
 396 }
 397
 398 /* Adds current to info->e_wait_q[sr] before element with smaller prio */
 399 static void wq_add(struct mqueue_inode_info *info, int sr,
 400                         struct ext_wait_queue *ewp)
 401 {
 402         struct ext_wait_queue *walk;
 403
 404         ewp->task = current;
 405
 406         list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 407                 if (walk->task->static_prio <= current->static_prio) {
 408                         list_add_tail(&ewp->list, &walk->list);
 409                         return;
 410                 }
 411         }
 412         list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 413 }
 414
 415 /*
 416  * Puts current task to sleep. Caller must hold queue lock. After return
 417  * lock isn't held.
 418  * sr: SEND or RECV
 419  */
 420 static int wq_sleep(struct mqueue_inode_info *info, int sr,
 421                         long timeout, struct ext_wait_queue *ewp)
 422 {
 423         int retval;
 424         signed long time;
 425
 426         wq_add(info, sr, ewp);
 427
 428         for (;;) {
 429                 set_current_state(TASK_INTERRUPTIBLE);
 430
 431                 spin_unlock(&info->lock);
 432                 time = schedule_timeout(timeout);
 433
 434                 while (ewp->state == STATE_PENDING)
 435                         cpu_relax();
 436
 437                 if (ewp->state == STATE_READY) {
 438                         retval = 0;
 439                         goto out;
 440                 }
 441                 spin_lock(&info->lock);
 442                 if (ewp->state == STATE_READY) {
 443                         retval = 0;
 444                         goto out_unlock;
 445                 }
 446                 if (signal_pending(current)) {
 447                         retval = -ERESTARTSYS;
 448                         break;
 449                 }
 450                 if (time == 0) {
 451                         retval = -ETIMEDOUT;
 452                         break;
 453                 }
 454         }
 455         list_del(&ewp->list);
 456 out_unlock:
 457         spin_unlock(&info->lock);
 458 out:
 459         return retval;
 460 }
 461
 462 /*
 463  * Returns waiting task that should be serviced first or NULL if none exists
 464  */
 465 static struct ext_wait_queue *wq_get_first_waiter(
 466                 struct mqueue_inode_info *info, int sr)
 467 {
 468         struct list_head *ptr;
 469
 470         ptr = info->e_wait_q[sr].list.prev;
 471         if (ptr == &info->e_wait_q[sr].list)
 472                 return NULL;
 473         return list_entry(ptr, struct ext_wait_queue, list);
 474 }
 475
 476 /* Auxiliary functions to manipulate messages' list */
 477 static void msg_insert(struct msg_msg *ptr, struct mqueue_inode_info *info)
 478 {
 479         int k;
 480
 481         k = info->attr.mq_curmsgs - 1;
 482         while (k >= 0 && info->messages[k]->m_type >= ptr->m_type) {
 483                 info->messages[k + 1] = info->messages[k];
 484                 k--;
 485         }
 486         info->attr.mq_curmsgs++;
 487         info->qsize += ptr->m_ts;
 488         info->messages[k + 1] = ptr;
 489 }
 490
 491 static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 492 {
 493         info->qsize -= info->messages[--info->attr.mq_curmsgs]->m_ts;
 494         return info->messages[info->attr.mq_curmsgs];
 495 }
 496
 497 static inline void set_cookie(struct sk_buff *skb, char code)
 498 {
 499         ((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 500 }
 501
 502 /*
 503  * The next function is only to split too long sys_mq_timedsend
 504  */
 505 static void __do_notify(struct mqueue_inode_info *info)
 506 {
 507         /* notification
 508          * invoked when there is registered process and there isn't process
 509          * waiting synchronously for message AND state of queue changed from
 510          * empty to not empty. Here we are sure that no one is waiting
 511          * synchronously. */
 512         if (info->notify_owner &&
 513             info->attr.mq_curmsgs == 1) {
 514                 struct siginfo sig_i;
 515                 switch (info->notify.sigev_notify) {
 516                 case SIGEV_NONE:
 517                         break;
 518                 case SIGEV_SIGNAL:
 519                         /* sends signal */
 520
 521                         sig_i.si_signo = info->notify.sigev_signo;
 522                         sig_i.si_errno = 0;
 523                         sig_i.si_code = SI_MESGQ;
 524                         sig_i.si_value = info->notify.sigev_value;
 525                         sig_i.si_pid = current->tgid;
 526                         sig_i.si_uid = current->uid;
 527
 528                         kill_proc_info(info->notify.sigev_signo,
 529                                        &sig_i, info->notify_owner);
 530                         break;
 531                 case SIGEV_THREAD:
 532                         set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 533                         netlink_sendskb(info->notify_sock,
 534                                         info->notify_cookie, 0);
 535                         break;
 536                 }
 537                 /* after notification unregisters process */
 538                 info->notify_owner = 0;
 539         }
 540         wake_up(&info->wait_q);
 541 }
 542
 543 static long prepare_timeout(const struct timespec __user *u_arg)
 544 {
 545         struct timespec ts, nowts;
 546         long timeout;
 547
 548         if (u_arg) {
 549                 if (unlikely(copy_from_user(&ts, u_arg,
 550                                         sizeof(struct timespec))))
 551                         return -EFAULT;
 552
 553                 if (unlikely(ts.tv_nsec < 0 || ts.tv_sec < 0
 554                         || ts.tv_nsec >= NSEC_PER_SEC))
 555                         return -EINVAL;
 556                 nowts = CURRENT_TIME;
 557                 /* first subtract as jiffies can't be too big */
 558                 ts.tv_sec -= nowts.tv_sec;
 559                 if (ts.tv_nsec < nowts.tv_nsec) {
 560                         ts.tv_nsec += NSEC_PER_SEC;
 561                         ts.tv_sec--;
 562                 }
 563                 ts.tv_nsec -= nowts.tv_nsec;
 564                 if (ts.tv_sec < 0)
 565                         return 0;
 566
 567                 timeout = timespec_to_jiffies(&ts) + 1;
 568         } else
 569                 return MAX_SCHEDULE_TIMEOUT;
 570
 571         return timeout;
 572 }
 573
 574 static void remove_notification(struct mqueue_inode_info *info)
 575 {
 576         if (info->notify_owner != 0 &&
 577             info->notify.sigev_notify == SIGEV_THREAD) {
 578                 set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 579                 netlink_sendskb(info->notify_sock, info->notify_cookie, 0);
 580         }
 581         info->notify_owner = 0;
 582 }
 583
 584 static int mq_attr_ok(struct mq_attr *attr)
 585 {
 586         if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 587                 return 0;
 588         if (capable(CAP_SYS_RESOURCE)) {
 589                 if (attr->mq_maxmsg > HARD_MSGMAX)
 590                         return 0;
 591         } else {
 592                 if (attr->mq_maxmsg > msg_max ||
 593                                 attr->mq_msgsize > msgsize_max)
 594                         return 0;
 595         }
 596         /* check for overflow */
 597         if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 598                 return 0;
 599         if ((unsigned long)(attr->mq_maxmsg * attr->mq_msgsize) +
 600             (attr->mq_maxmsg * sizeof (struct msg_msg *)) <
 601             (unsigned long)(attr->mq_maxmsg * attr->mq_msgsize))
 602                 return 0;
 603         return 1;
 604 }
 605
 606 /*
 607  * Invoked when creating a new queue via sys_mq_open
 608  */
 609 static struct file *do_create(struct dentry *dir, struct dentry *dentry,
 610                         int oflag, mode_t mode, struct mq_attr __user *u_attr)
 611 {
 612         struct mq_attr attr;
 613         int ret;
 614
 615         if (u_attr) {
 616                 ret = -EFAULT;
 617                 if (copy_from_user(&attr, u_attr, sizeof(attr)))
 618                         goto out;
 619                 ret = -EINVAL;
 620                 if (!mq_attr_ok(&attr))
 621                         goto out;
 622                 /* store for use during create */
 623                 dentry->d_fsdata = &attr;
 624         }
 625
 626         mode &= ~current->fs->umask;
 627         ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 628         dentry->d_fsdata = NULL;
 629         if (ret)
 630                 goto out;
 631
 632         return dentry_open(dentry, mqueue_mnt, oflag);
 633
 634 out:
 635         dput(dentry);
 636         mntput(mqueue_mnt);
 637         return ERR_PTR(ret);
 638 }
 639
 640 /* Opens existing queue */
 641 static struct file *do_open(struct dentry *dentry, int oflag)
 642 {
 643 static int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 644                                         MAY_READ | MAY_WRITE };
 645
 646         if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 647                 dput(dentry);
 648                 mntput(mqueue_mnt);
 649                 return ERR_PTR(-EINVAL);
 650         }
 651
 652         if (permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE], NULL)) {
 653                 dput(dentry);
 654                 mntput(mqueue_mnt);
 655                 return ERR_PTR(-EACCES);
 656         }
 657
 658         return dentry_open(dentry, mqueue_mnt, oflag);
 659 }
 660
 661 asmlinkage long sys_mq_open(const char __user *u_name, int oflag, mode_t mode,
 662                                 struct mq_attr __user *u_attr)
 663 {
 664         struct dentry *dentry;
 665         struct file *filp;
 666         char *name;
 667         int fd, error;
 668
 669         if (IS_ERR(name = getname(u_name)))
 670                 return PTR_ERR(name);
 671
 672         fd = get_unused_fd();
 673         if (fd < 0)
 674                 goto out_putname;
 675
 676         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 677         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 678         if (IS_ERR(dentry)) {
 679                 error = PTR_ERR(dentry);
 680                 goto out_err;
 681         }
 682         mntget(mqueue_mnt);
 683
 684         if (oflag & O_CREAT) {
 685                 if (dentry->d_inode) {  /* entry already exists */
 686                         error = -EEXIST;
 687                         if (oflag & O_EXCL)
 688                                 goto out;
 689                         filp = do_open(dentry, oflag);
 690                 } else {
 691                         filp = do_create(mqueue_mnt->mnt_root, dentry,
 692                                                 oflag, mode, u_attr);
 693                 }
 694         } else {
 695                 error = -ENOENT;
 696                 if (!dentry->d_inode)
 697                         goto out;
 698                 filp = do_open(dentry, oflag);
 699         }
 700
 701         if (IS_ERR(filp)) {
 702                 error = PTR_ERR(filp);
 703                 goto out_putfd;
 704         }
 705
 706         set_close_on_exec(fd, 1);
 707         fd_install(fd, filp);
 708         goto out_upsem;
 709
 710 out:
 711         dput(dentry);
 712         mntput(mqueue_mnt);
 713 out_putfd:
 714         put_unused_fd(fd);
 715 out_err:
 716         fd = error;
 717 out_upsem:
 718         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 719 out_putname:
 720         putname(name);
 721         return fd;
 722 }
 723
 724 asmlinkage long sys_mq_unlink(const char __user *u_name)
 725 {
 726         int err;
 727         char *name;
 728         struct dentry *dentry;
 729         struct inode *inode = NULL;
 730
 731         name = getname(u_name);
 732         if (IS_ERR(name))
 733                 return PTR_ERR(name);
 734
 735         mutex_lock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 736         dentry = lookup_one_len(name, mqueue_mnt->mnt_root, strlen(name));
 737         if (IS_ERR(dentry)) {
 738                 err = PTR_ERR(dentry);
 739                 goto out_unlock;
 740         }
 741
 742         if (!dentry->d_inode) {
 743                 err = -ENOENT;
 744                 goto out_err;
 745         }
 746
 747         inode = dentry->d_inode;
 748         if (inode)
 749                 atomic_inc(&inode->i_count);
 750
 751         err = vfs_unlink(dentry->d_parent->d_inode, dentry, NULL);
 752 out_err:
 753         dput(dentry);
 754
 755 out_unlock:
 756         mutex_unlock(&mqueue_mnt->mnt_root->d_inode->i_mutex);
 757         putname(name);
 758         if (inode)
 759                 iput(inode);
 760
 761         return err;
 762 }
 763
 764 /* Pipelined send and receive functions.
 765  *
 766  * If a receiver finds no waiting message, then it registers itself in the
 767  * list of waiting receivers. A sender checks that list before adding the new
 768  * message into the message array. If there is a waiting receiver, then it
 769  * bypasses the message array and directly hands the message over to the
 770  * receiver.
 771  * The receiver accepts the message and returns without grabbing the queue
 772  * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 773  * are necessary. The same algorithm is used for sysv semaphores, see
 774  * ipc/sem.c for more details.
 775  *
 776  * The same algorithm is used for senders.
 777  */
 778
 779 /* pipelined_send() - send a message directly to the task waiting in
 780  * sys_mq_timedreceive() (without inserting message into a queue).
 781  */
 782 static inline void pipelined_send(struct mqueue_inode_info *info,
 783                                   struct msg_msg *message,
 784                                   struct ext_wait_queue *receiver)
 785 {
 786         receiver->msg = message;
 787         list_del(&receiver->list);
 788         receiver->state = STATE_PENDING;
 789         wake_up_process(receiver->task);
 790         smp_wmb();
 791         receiver->state = STATE_READY;
 792 }
 793
 794 /* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 795  * gets its message and put to the queue (we have one free place for sure). */
 796 static inline void pipelined_receive(struct mqueue_inode_info *info)
 797 {
 798         struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 799
 800         if (!sender) {
 801                 /* for poll */
 802                 wake_up_interruptible(&info->wait_q);
 803                 return;
 804         }
 805         msg_insert(sender->msg, info);
 806         list_del(&sender->list);
 807         sender->state = STATE_PENDING;
 808         wake_up_process(sender->task);
 809         smp_wmb();
 810         sender->state = STATE_READY;
 811 }
 812
 813 asmlinkage long sys_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 814         size_t msg_len, unsigned int msg_prio,
 815         const struct timespec __user *u_abs_timeout)
 816 {
 817         struct file *filp;
 818         struct inode *inode;
 819         struct ext_wait_queue wait;
 820         struct ext_wait_queue *receiver;
 821         struct msg_msg *msg_ptr;
 822         struct mqueue_inode_info *info;
 823         long timeout;
 824         int ret;
 825
 826         if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 827                 return -EINVAL;
 828
 829         timeout = prepare_timeout(u_abs_timeout);
 830
 831         ret = -EBADF;
 832         filp = fget(mqdes);
 833         if (unlikely(!filp))
 834                 goto out;
 835
 836         inode = filp->f_dentry->d_inode;
 837         if (unlikely(filp->f_op != &mqueue_file_operations))
 838                 goto out_fput;
 839         info = MQUEUE_I(inode);
 840
 841         if (unlikely(!(filp->f_mode & FMODE_WRITE)))
 842                 goto out_fput;
 843
 844         if (unlikely(msg_len > info->attr.mq_msgsize)) {
 845                 ret = -EMSGSIZE;
 846                 goto out_fput;
 847         }
 848
 849         /* First try to allocate memory, before doing anything with
 850          * existing queues. */
 851         msg_ptr = load_msg(u_msg_ptr, msg_len);
 852         if (IS_ERR(msg_ptr)) {
 853                 ret = PTR_ERR(msg_ptr);
 854                 goto out_fput;
 855         }
 856         msg_ptr->m_ts = msg_len;
 857         msg_ptr->m_type = msg_prio;
 858
 859         spin_lock(&info->lock);
 860
 861         if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
 862                 if (filp->f_flags & O_NONBLOCK) {
 863                         spin_unlock(&info->lock);
 864                         ret = -EAGAIN;
 865                 } else if (unlikely(timeout < 0)) {
 866                         spin_unlock(&info->lock);
 867                         ret = timeout;
 868                 } else {
 869                         wait.task = current;
 870                         wait.msg = (void *) msg_ptr;
 871                         wait.state = STATE_NONE;
 872                         ret = wq_sleep(info, SEND, timeout, &wait);
 873                 }
 874                 if (ret < 0)
 875                         free_msg(msg_ptr);
 876         } else {
 877                 receiver = wq_get_first_waiter(info, RECV);
 878                 if (receiver) {
 879                         pipelined_send(info, msg_ptr, receiver);
 880                 } else {
 881                         /* adds message to the queue */
 882                         msg_insert(msg_ptr, info);
 883                         __do_notify(info);
 884                 }
 885                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 886                                 CURRENT_TIME;
 887                 spin_unlock(&info->lock);
 888                 ret = 0;
 889         }
 890 out_fput:
 891         fput(filp);
 892 out:
 893         return ret;
 894 }
 895
 896 asmlinkage ssize_t sys_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
 897         size_t msg_len, unsigned int __user *u_msg_prio,
 898         const struct timespec __user *u_abs_timeout)
 899 {
 900         long timeout;
 901         ssize_t ret;
 902         struct msg_msg *msg_ptr;
 903         struct file *filp;
 904         struct inode *inode;
 905         struct mqueue_inode_info *info;
 906         struct ext_wait_queue wait;
 907
 908         timeout = prepare_timeout(u_abs_timeout);
 909
 910         ret = -EBADF;
 911         filp = fget(mqdes);
 912         if (unlikely(!filp))
 913                 goto out;
 914
 915         inode = filp->f_dentry->d_inode;
 916         if (unlikely(filp->f_op != &mqueue_file_operations))
 917                 goto out_fput;
 918         info = MQUEUE_I(inode);
 919
 920         if (unlikely(!(filp->f_mode & FMODE_READ)))
 921                 goto out_fput;
 922
 923         /* checks if buffer is big enough */
 924         if (unlikely(msg_len < info->attr.mq_msgsize)) {
 925                 ret = -EMSGSIZE;
 926                 goto out_fput;
 927         }
 928
 929         spin_lock(&info->lock);
 930         if (info->attr.mq_curmsgs == 0) {
 931                 if (filp->f_flags & O_NONBLOCK) {
 932                         spin_unlock(&info->lock);
 933                         ret = -EAGAIN;
 934                         msg_ptr = NULL;
 935                 } else if (unlikely(timeout < 0)) {
 936                         spin_unlock(&info->lock);
 937                         ret = timeout;
 938                         msg_ptr = NULL;
 939                 } else {
 940                         wait.task = current;
 941                         wait.state = STATE_NONE;
 942                         ret = wq_sleep(info, RECV, timeout, &wait);
 943                         msg_ptr = wait.msg;
 944                 }
 945         } else {
 946                 msg_ptr = msg_get(info);
 947
 948                 inode->i_atime = inode->i_mtime = inode->i_ctime =
 949                                 CURRENT_TIME;
 950
 951                 /* There is now free space in queue. */
 952                 pipelined_receive(info);
 953                 spin_unlock(&info->lock);
 954                 ret = 0;
 955         }
 956         if (ret == 0) {
 957                 ret = msg_ptr->m_ts;
 958
 959                 if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
 960                         store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
 961                         ret = -EFAULT;
 962                 }
 963                 free_msg(msg_ptr);
 964         }
 965 out_fput:
 966         fput(filp);
 967 out:
 968         return ret;
 969 }
 970
 971 /*
 972  * Notes: the case when user wants us to deregister (with NULL as pointer)
 973  * and he isn't currently owner of notification, will be silently discarded.
 974  * It isn't explicitly defined in the POSIX.
 975  */
 976 asmlinkage long sys_mq_notify(mqd_t mqdes,
 977                                 const struct sigevent __user *u_notification)
 978 {
 979         int ret;
 980         struct file *filp;
 981         struct sock *sock;
 982         struct inode *inode;
 983         struct sigevent notification;
 984         struct mqueue_inode_info *info;
 985         struct sk_buff *nc;
 986
 987         nc = NULL;
 988         sock = NULL;
 989         if (u_notification != NULL) {
 990                 if (copy_from_user(&notification, u_notification,
 991                                         sizeof(struct sigevent)))
 992                         return -EFAULT;
 993
 994                 if (unlikely(notification.sigev_notify != SIGEV_NONE &&
 995                              notification.sigev_notify != SIGEV_SIGNAL &&
 996                              notification.sigev_notify != SIGEV_THREAD))
 997                         return -EINVAL;
 998                 if (notification.sigev_notify == SIGEV_SIGNAL &&
 999                         !valid_signal(notification.sigev_signo)) {
1000                         return -EINVAL;
1001                 }
1002                 if (notification.sigev_notify == SIGEV_THREAD) {
1003                         /* create the notify skb */
1004                         nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1005                         ret = -ENOMEM;
1006                         if (!nc)
1007                                 goto out;
1008                         ret = -EFAULT;
1009                         if (copy_from_user(nc->data,
1010                                         notification.sigev_value.sival_ptr,
1011                                         NOTIFY_COOKIE_LEN)) {
1012                                 goto out;
1013                         }
1014
1015                         /* TODO: add a header? */
1016                         skb_put(nc, NOTIFY_COOKIE_LEN);
1017                         /* and attach it to the socket */
1018 retry:
1019                         filp = fget(notification.sigev_signo);
1020                         ret = -EBADF;
1021                         if (!filp)
1022                                 goto out;
1023                         sock = netlink_getsockbyfilp(filp);
1024                         fput(filp);
1025                         if (IS_ERR(sock)) {
1026                                 ret = PTR_ERR(sock);
1027                                 sock = NULL;
1028                                 goto out;
1029                         }
1030
1031                         ret = netlink_attachskb(sock, nc, 0,
1032                                         MAX_SCHEDULE_TIMEOUT, NULL);
1033                         if (ret == 1)
1034                                 goto retry;
1035                         if (ret) {
1036                                 sock = NULL;
1037                                 nc = NULL;
1038                                 goto out;
1039                         }
1040                 }
1041         }
1042
1043         ret = -EBADF;
1044         filp = fget(mqdes);
1045         if (!filp)
1046                 goto out;
1047
1048         inode = filp->f_dentry->d_inode;
1049         if (unlikely(filp->f_op != &mqueue_file_operations))
1050                 goto out_fput;
1051         info = MQUEUE_I(inode);
1052
1053         ret = 0;
1054         spin_lock(&info->lock);
1055         if (u_notification == NULL) {
1056                 if (info->notify_owner == current->tgid) {
1057                         remove_notification(info);
1058                         inode->i_atime = inode->i_ctime = CURRENT_TIME;
1059                 }
1060         } else if (info->notify_owner != 0) {
1061                 ret = -EBUSY;
1062         } else {
1063                 switch (notification.sigev_notify) {
1064                 case SIGEV_NONE:
1065                         info->notify.sigev_notify = SIGEV_NONE;
1066                         break;
1067                 case SIGEV_THREAD:
1068                         info->notify_sock = sock;
1069                         info->notify_cookie = nc;
1070                         sock = NULL;
1071                         nc = NULL;
1072                         info->notify.sigev_notify = SIGEV_THREAD;
1073                         break;
1074                 case SIGEV_SIGNAL:
1075                         info->notify.sigev_signo = notification.sigev_signo;
1076                         info->notify.sigev_value = notification.sigev_value;
1077                         info->notify.sigev_notify = SIGEV_SIGNAL;
1078                         break;
1079                 }
1080                 info->notify_owner = current->tgid;
1081                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1082         }
1083         spin_unlock(&info->lock);
1084 out_fput:
1085         fput(filp);
1086 out:
1087         if (sock) {
1088                 netlink_detachskb(sock, nc);
1089         } else if (nc) {
1090                 dev_kfree_skb(nc);
1091         }
1092         return ret;
1093 }
1094
1095 asmlinkage long sys_mq_getsetattr(mqd_t mqdes,
1096                         const struct mq_attr __user *u_mqstat,
1097                         struct mq_attr __user *u_omqstat)
1098 {
1099         int ret;
1100         struct mq_attr mqstat, omqstat;
1101         struct file *filp;
1102         struct inode *inode;
1103         struct mqueue_inode_info *info;
1104
1105         if (u_mqstat != NULL) {
1106                 if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
1107                         return -EFAULT;
1108                 if (mqstat.mq_flags & (~O_NONBLOCK))
1109                         return -EINVAL;
1110         }
1111
1112         ret = -EBADF;
1113         filp = fget(mqdes);
1114         if (!filp)
1115                 goto out;
1116
1117         inode = filp->f_dentry->d_inode;
1118         if (unlikely(filp->f_op != &mqueue_file_operations))
1119                 goto out_fput;
1120         info = MQUEUE_I(inode);
1121
1122         spin_lock(&info->lock);
1123
1124         omqstat = info->attr;
1125         omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1126         if (u_mqstat) {
1127                 if (mqstat.mq_flags & O_NONBLOCK)
1128                         filp->f_flags |= O_NONBLOCK;
1129                 else
1130                         filp->f_flags &= ~O_NONBLOCK;
1131
1132                 inode->i_atime = inode->i_ctime = CURRENT_TIME;
1133         }
1134
1135         spin_unlock(&info->lock);
1136
1137         ret = 0;
1138         if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1139                                                 sizeof(struct mq_attr)))
1140                 ret = -EFAULT;
1141
1142 out_fput:
1143         fput(filp);
1144 out:
1145         return ret;
1146 }
1147
1148 static struct inode_operations mqueue_dir_inode_operations = {
1149         .lookup = simple_lookup,
1150         .create = mqueue_create,
1151         .unlink = mqueue_unlink,
1152 };
1153
1154 static struct file_operations mqueue_file_operations = {
1155         .flush = mqueue_flush_file,
1156         .poll = mqueue_poll_file,
1157         .read = mqueue_read_file,
1158 };
1159
1160 static struct super_operations mqueue_super_ops = {
1161         .alloc_inode = mqueue_alloc_inode,
1162         .destroy_inode = mqueue_destroy_inode,
1163         .statfs = simple_statfs,
1164         .delete_inode = mqueue_delete_inode,
1165         .drop_inode = generic_delete_inode,
1166 };
1167
1168 static struct file_system_type mqueue_fs_type = {
1169         .name = "mqueue",
1170         .get_sb = mqueue_get_sb,
1171         .kill_sb = kill_litter_super,
1172 };
1173
1174 static int msg_max_limit_min = DFLT_MSGMAX;
1175 static int msg_max_limit_max = HARD_MSGMAX;
1176
1177 static int msg_maxsize_limit_min = DFLT_MSGSIZEMAX;
1178 static int msg_maxsize_limit_max = INT_MAX;
1179
1180 static ctl_table mq_sysctls[] = {
1181         {
1182                 .ctl_name       = CTL_QUEUESMAX,
1183                 .procname       = "queues_max",
1184                 .data           = &queues_max,
1185                 .maxlen         = sizeof(int),
1186                 .mode           = 0644,
1187                 .proc_handler   = &proc_dointvec,
1188         },
1189         {
1190                 .ctl_name       = CTL_MSGMAX,
1191                 .procname       = "msg_max",
1192                 .data           = &msg_max,
1193                 .maxlen         = sizeof(int),
1194                 .mode           = 0644,
1195                 .proc_handler   = &proc_dointvec_minmax,
1196                 .extra1         = &msg_max_limit_min,
1197                 .extra2         = &msg_max_limit_max,
1198         },
1199         {
1200                 .ctl_name       = CTL_MSGSIZEMAX,
1201                 .procname       = "msgsize_max",
1202                 .data           = &msgsize_max,
1203                 .maxlen         = sizeof(int),
1204                 .mode           = 0644,
1205                 .proc_handler   = &proc_dointvec_minmax,
1206                 .extra1         = &msg_maxsize_limit_min,
1207                 .extra2         = &msg_maxsize_limit_max,
1208         },
1209         { .ctl_name = 0 }
1210 };
1211
1212 static ctl_table mq_sysctl_dir[] = {
1213         {
1214                 .ctl_name       = FS_MQUEUE,
1215                 .procname       = "mqueue",
1216                 .mode           = 0555,
1217                 .child          = mq_sysctls,
1218         },
1219         { .ctl_name = 0 }
1220 };
1221
1222 static ctl_table mq_sysctl_root[] = {
1223         {
1224                 .ctl_name       = CTL_FS,
1225                 .procname       = "fs",
1226                 .mode           = 0555,
1227                 .child          = mq_sysctl_dir,
1228         },
1229         { .ctl_name = 0 }
1230 };
1231
1232 static int __init init_mqueue_fs(void)
1233 {
1234         int error;
1235
1236         mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1237                                 sizeof(struct mqueue_inode_info), 0,
1238                                 SLAB_HWCACHE_ALIGN, init_once, NULL);
1239         if (mqueue_inode_cachep == NULL)
1240                 return -ENOMEM;
1241
1242         /* ignore failues - they are not fatal */
1243         mq_sysctl_table = register_sysctl_table(mq_sysctl_root, 0);
1244
1245         error = register_filesystem(&mqueue_fs_type);
1246         if (error)
1247                 goto out_sysctl;
1248
1249         if (IS_ERR(mqueue_mnt = kern_mount(&mqueue_fs_type))) {
1250                 error = PTR_ERR(mqueue_mnt);
1251                 goto out_filesystem;
1252         }
1253
1254         /* internal initialization - not common for vfs */
1255         queues_count = 0;
1256         spin_lock_init(&mq_lock);
1257
1258         return 0;
1259
1260 out_filesystem:
1261         unregister_filesystem(&mqueue_fs_type);
1262 out_sysctl:
1263         if (mq_sysctl_table)
1264                 unregister_sysctl_table(mq_sysctl_table);
1265         if (kmem_cache_destroy(mqueue_inode_cachep)) {
1266                 printk(KERN_INFO
1267                         "mqueue_inode_cache: not all structures were freed\n");
1268         }
1269         return error;
1270 }
1271
1272 __initcall(init_mqueue_fs);