Linux Audio

Check our new training course

Loading...
v3.5.6
   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/vmalloc.h>
  28#include <linux/netlink.h>
  29#include <linux/syscalls.h>
  30#include <linux/audit.h>
  31#include <linux/signal.h>
  32#include <linux/mutex.h>
  33#include <linux/nsproxy.h>
  34#include <linux/pid.h>
  35#include <linux/ipc_namespace.h>
  36#include <linux/user_namespace.h>
  37#include <linux/slab.h>
 
 
 
  38
  39#include <net/sock.h>
  40#include "util.h"
  41
 
 
 
 
  42#define MQUEUE_MAGIC	0x19800202
  43#define DIRENT_SIZE	20
  44#define FILENT_SIZE	80
  45
  46#define SEND		0
  47#define RECV		1
  48
  49#define STATE_NONE	0
  50#define STATE_PENDING	1
  51#define STATE_READY	2
  52
  53struct posix_msg_tree_node {
  54	struct rb_node		rb_node;
  55	struct list_head	msg_list;
  56	int			priority;
  57};
  58
  59struct 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
  66struct mqueue_inode_info {
  67	spinlock_t lock;
  68	struct inode vfs_inode;
  69	wait_queue_head_t wait_q;
  70
  71	struct rb_root msg_tree;
 
  72	struct posix_msg_tree_node *node_cache;
  73	struct mq_attr attr;
  74
  75	struct sigevent notify;
  76	struct pid* notify_owner;
  77	struct user_namespace *notify_user_ns;
  78	struct user_struct *user;	/* user who created, for accounting */
  79	struct sock *notify_sock;
  80	struct sk_buff *notify_cookie;
  81
  82	/* for tasks waiting for free space and messages, respectively */
  83	struct ext_wait_queue e_wait_q[2];
  84
  85	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  86};
  87
 
  88static const struct inode_operations mqueue_dir_inode_operations;
  89static const struct file_operations mqueue_file_operations;
  90static const struct super_operations mqueue_super_ops;
 
  91static void remove_notification(struct mqueue_inode_info *info);
  92
  93static struct kmem_cache *mqueue_inode_cachep;
  94
  95static struct ctl_table_header * mq_sysctl_table;
  96
  97static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
  98{
  99	return container_of(inode, struct mqueue_inode_info, vfs_inode);
 100}
 101
 102/*
 103 * This routine should be called with the mq_lock held.
 104 */
 105static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
 106{
 107	return get_ipc_ns(inode->i_sb->s_fs_info);
 108}
 109
 110static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 111{
 112	struct ipc_namespace *ns;
 113
 114	spin_lock(&mq_lock);
 115	ns = __get_ns_from_inode(inode);
 116	spin_unlock(&mq_lock);
 117	return ns;
 118}
 119
 120/* Auxiliary functions to manipulate messages' list */
 121static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
 122{
 123	struct rb_node **p, *parent = NULL;
 124	struct posix_msg_tree_node *leaf;
 
 125
 126	p = &info->msg_tree.rb_node;
 127	while (*p) {
 128		parent = *p;
 129		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
 130
 131		if (likely(leaf->priority == msg->m_type))
 132			goto insert_msg;
 133		else if (msg->m_type < leaf->priority)
 134			p = &(*p)->rb_left;
 135		else
 
 136			p = &(*p)->rb_right;
 137	}
 138	if (info->node_cache) {
 139		leaf = info->node_cache;
 140		info->node_cache = NULL;
 141	} else {
 142		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
 143		if (!leaf)
 144			return -ENOMEM;
 145		rb_init_node(&leaf->rb_node);
 146		INIT_LIST_HEAD(&leaf->msg_list);
 147		info->qsize += sizeof(*leaf);
 148	}
 149	leaf->priority = msg->m_type;
 
 
 
 
 150	rb_link_node(&leaf->rb_node, parent, p);
 151	rb_insert_color(&leaf->rb_node, &info->msg_tree);
 152insert_msg:
 153	info->attr.mq_curmsgs++;
 154	info->qsize += msg->m_ts;
 155	list_add_tail(&msg->m_list, &leaf->msg_list);
 156	return 0;
 157}
 158
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 159static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 160{
 161	struct rb_node **p, *parent = NULL;
 162	struct posix_msg_tree_node *leaf;
 163	struct msg_msg *msg;
 164
 165try_again:
 166	p = &info->msg_tree.rb_node;
 167	while (*p) {
 168		parent = *p;
 169		/*
 170		 * During insert, low priorities go to the left and high to the
 171		 * right.  On receive, we want the highest priorities first, so
 172		 * walk all the way to the right.
 173		 */
 174		p = &(*p)->rb_right;
 175	}
 176	if (!parent) {
 177		if (info->attr.mq_curmsgs) {
 178			pr_warn_once("Inconsistency in POSIX message queue, "
 179				     "no tree element, but supposedly messages "
 180				     "should exist!\n");
 181			info->attr.mq_curmsgs = 0;
 182		}
 183		return NULL;
 184	}
 185	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
 186	if (unlikely(list_empty(&leaf->msg_list))) {
 187		pr_warn_once("Inconsistency in POSIX message queue, "
 188			     "empty leaf node but we haven't implemented "
 189			     "lazy leaf delete!\n");
 190		rb_erase(&leaf->rb_node, &info->msg_tree);
 191		if (info->node_cache) {
 192			info->qsize -= sizeof(*leaf);
 193			kfree(leaf);
 194		} else {
 195			info->node_cache = leaf;
 196		}
 197		goto try_again;
 198	} else {
 199		msg = list_first_entry(&leaf->msg_list,
 200				       struct msg_msg, m_list);
 201		list_del(&msg->m_list);
 202		if (list_empty(&leaf->msg_list)) {
 203			rb_erase(&leaf->rb_node, &info->msg_tree);
 204			if (info->node_cache) {
 205				info->qsize -= sizeof(*leaf);
 206				kfree(leaf);
 207			} else {
 208				info->node_cache = leaf;
 209			}
 210		}
 211	}
 212	info->attr.mq_curmsgs--;
 213	info->qsize -= msg->m_ts;
 214	return msg;
 215}
 216
 217static struct inode *mqueue_get_inode(struct super_block *sb,
 218		struct ipc_namespace *ipc_ns, umode_t mode,
 219		struct mq_attr *attr)
 220{
 221	struct user_struct *u = current_user();
 222	struct inode *inode;
 223	int ret = -ENOMEM;
 224
 225	inode = new_inode(sb);
 226	if (!inode)
 227		goto err;
 228
 229	inode->i_ino = get_next_ino();
 230	inode->i_mode = mode;
 231	inode->i_uid = current_fsuid();
 232	inode->i_gid = current_fsgid();
 233	inode->i_mtime = inode->i_ctime = inode->i_atime = CURRENT_TIME;
 234
 235	if (S_ISREG(mode)) {
 236		struct mqueue_inode_info *info;
 237		unsigned long mq_bytes, mq_treesize;
 238
 239		inode->i_fop = &mqueue_file_operations;
 240		inode->i_size = FILENT_SIZE;
 241		/* mqueue specific info */
 242		info = MQUEUE_I(inode);
 243		spin_lock_init(&info->lock);
 244		init_waitqueue_head(&info->wait_q);
 245		INIT_LIST_HEAD(&info->e_wait_q[0].list);
 246		INIT_LIST_HEAD(&info->e_wait_q[1].list);
 247		info->notify_owner = NULL;
 248		info->notify_user_ns = NULL;
 249		info->qsize = 0;
 250		info->user = NULL;	/* set when all is ok */
 251		info->msg_tree = RB_ROOT;
 
 252		info->node_cache = NULL;
 253		memset(&info->attr, 0, sizeof(info->attr));
 254		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
 255					   ipc_ns->mq_msg_default);
 256		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
 257					    ipc_ns->mq_msgsize_default);
 258		if (attr) {
 259			info->attr.mq_maxmsg = attr->mq_maxmsg;
 260			info->attr.mq_msgsize = attr->mq_msgsize;
 261		}
 262		/*
 263		 * We used to allocate a static array of pointers and account
 264		 * the size of that array as well as one msg_msg struct per
 265		 * possible message into the queue size. That's no longer
 266		 * accurate as the queue is now an rbtree and will grow and
 267		 * shrink depending on usage patterns.  We can, however, still
 268		 * account one msg_msg struct per message, but the nodes are
 269		 * allocated depending on priority usage, and most programs
 270		 * only use one, or a handful, of priorities.  However, since
 271		 * this is pinned memory, we need to assume worst case, so
 272		 * that means the min(mq_maxmsg, max_priorities) * struct
 273		 * posix_msg_tree_node.
 274		 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 275		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 276			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 277			sizeof(struct posix_msg_tree_node);
 278
 279		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
 280					  info->attr.mq_msgsize);
 281
 282		spin_lock(&mq_lock);
 283		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 284		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
 285			spin_unlock(&mq_lock);
 286			/* mqueue_evict_inode() releases info->messages */
 287			ret = -EMFILE;
 288			goto out_inode;
 289		}
 290		u->mq_bytes += mq_bytes;
 291		spin_unlock(&mq_lock);
 292
 293		/* all is ok */
 294		info->user = get_uid(u);
 295	} else if (S_ISDIR(mode)) {
 296		inc_nlink(inode);
 297		/* Some things misbehave if size == 0 on a directory */
 298		inode->i_size = 2 * DIRENT_SIZE;
 299		inode->i_op = &mqueue_dir_inode_operations;
 300		inode->i_fop = &simple_dir_operations;
 301	}
 302
 303	return inode;
 304out_inode:
 305	iput(inode);
 306err:
 307	return ERR_PTR(ret);
 308}
 309
 310static int mqueue_fill_super(struct super_block *sb, void *data, int silent)
 311{
 312	struct inode *inode;
 313	struct ipc_namespace *ns = data;
 314
 315	sb->s_blocksize = PAGE_CACHE_SIZE;
 316	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 
 317	sb->s_magic = MQUEUE_MAGIC;
 318	sb->s_op = &mqueue_super_ops;
 319
 320	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 321	if (IS_ERR(inode))
 322		return PTR_ERR(inode);
 323
 324	sb->s_root = d_make_root(inode);
 325	if (!sb->s_root)
 326		return -ENOMEM;
 327	return 0;
 328}
 329
 330static struct dentry *mqueue_mount(struct file_system_type *fs_type,
 331			 int flags, const char *dev_name,
 332			 void *data)
 333{
 334	if (!(flags & MS_KERNMOUNT))
 335		data = current->nsproxy->ipc_ns;
 336	return mount_ns(fs_type, flags, data, mqueue_fill_super);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 337}
 338
 339static void init_once(void *foo)
 340{
 341	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 342
 343	inode_init_once(&p->vfs_inode);
 344}
 345
 346static struct inode *mqueue_alloc_inode(struct super_block *sb)
 347{
 348	struct mqueue_inode_info *ei;
 349
 350	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 351	if (!ei)
 352		return NULL;
 353	return &ei->vfs_inode;
 354}
 355
 356static void mqueue_i_callback(struct rcu_head *head)
 357{
 358	struct inode *inode = container_of(head, struct inode, i_rcu);
 359	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 360}
 361
 362static void mqueue_destroy_inode(struct inode *inode)
 363{
 364	call_rcu(&inode->i_rcu, mqueue_i_callback);
 365}
 366
 367static void mqueue_evict_inode(struct inode *inode)
 368{
 369	struct mqueue_inode_info *info;
 370	struct user_struct *user;
 371	unsigned long mq_bytes, mq_treesize;
 372	struct ipc_namespace *ipc_ns;
 373	struct msg_msg *msg;
 
 374
 375	clear_inode(inode);
 376
 377	if (S_ISDIR(inode->i_mode))
 378		return;
 379
 380	ipc_ns = get_ns_from_inode(inode);
 381	info = MQUEUE_I(inode);
 382	spin_lock(&info->lock);
 383	while ((msg = msg_get(info)) != NULL)
 384		free_msg(msg);
 385	kfree(info->node_cache);
 386	spin_unlock(&info->lock);
 387
 388	/* Total amount of bytes accounted for the mqueue */
 389	mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 390		min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 391		sizeof(struct posix_msg_tree_node);
 392
 393	mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
 394				  info->attr.mq_msgsize);
 395
 396	user = info->user;
 397	if (user) {
 
 
 
 
 
 
 
 
 
 
 398		spin_lock(&mq_lock);
 399		user->mq_bytes -= mq_bytes;
 400		/*
 401		 * get_ns_from_inode() ensures that the
 402		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 403		 * to which we now hold a reference, or it is NULL.
 404		 * We can't put it here under mq_lock, though.
 405		 */
 406		if (ipc_ns)
 407			ipc_ns->mq_queues_count--;
 408		spin_unlock(&mq_lock);
 409		free_uid(user);
 410	}
 411	if (ipc_ns)
 412		put_ipc_ns(ipc_ns);
 413}
 414
 415static int mqueue_create(struct inode *dir, struct dentry *dentry,
 416				umode_t mode, struct nameidata *nd)
 417{
 
 418	struct inode *inode;
 419	struct mq_attr *attr = dentry->d_fsdata;
 420	int error;
 421	struct ipc_namespace *ipc_ns;
 422
 423	spin_lock(&mq_lock);
 424	ipc_ns = __get_ns_from_inode(dir);
 425	if (!ipc_ns) {
 426		error = -EACCES;
 427		goto out_unlock;
 428	}
 429	if (ipc_ns->mq_queues_count >= HARD_QUEUESMAX ||
 430	    (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 431	     !capable(CAP_SYS_RESOURCE))) {
 432		error = -ENOSPC;
 433		goto out_unlock;
 434	}
 435	ipc_ns->mq_queues_count++;
 436	spin_unlock(&mq_lock);
 437
 438	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 439	if (IS_ERR(inode)) {
 440		error = PTR_ERR(inode);
 441		spin_lock(&mq_lock);
 442		ipc_ns->mq_queues_count--;
 443		goto out_unlock;
 444	}
 445
 446	put_ipc_ns(ipc_ns);
 447	dir->i_size += DIRENT_SIZE;
 448	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 449
 450	d_instantiate(dentry, inode);
 451	dget(dentry);
 452	return 0;
 453out_unlock:
 454	spin_unlock(&mq_lock);
 455	if (ipc_ns)
 456		put_ipc_ns(ipc_ns);
 457	return error;
 458}
 459
 
 
 
 
 
 
 460static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 461{
 462  	struct inode *inode = dentry->d_inode;
 463
 464	dir->i_ctime = dir->i_mtime = dir->i_atime = CURRENT_TIME;
 465	dir->i_size -= DIRENT_SIZE;
 466  	drop_nlink(inode);
 467  	dput(dentry);
 468  	return 0;
 469}
 470
 471/*
 472*	This is routine for system read from queue file.
 473*	To avoid mess with doing here some sort of mq_receive we allow
 474*	to read only queue size & notification info (the only values
 475*	that are interesting from user point of view and aren't accessible
 476*	through std routines)
 477*/
 478static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 479				size_t count, loff_t *off)
 480{
 481	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 482	char buffer[FILENT_SIZE];
 483	ssize_t ret;
 484
 485	spin_lock(&info->lock);
 486	snprintf(buffer, sizeof(buffer),
 487			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 488			info->qsize,
 489			info->notify_owner ? info->notify.sigev_notify : 0,
 490			(info->notify_owner &&
 491			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
 492				info->notify.sigev_signo : 0,
 493			pid_vnr(info->notify_owner));
 494	spin_unlock(&info->lock);
 495	buffer[sizeof(buffer)-1] = '\0';
 496
 497	ret = simple_read_from_buffer(u_data, count, off, buffer,
 498				strlen(buffer));
 499	if (ret <= 0)
 500		return ret;
 501
 502	filp->f_path.dentry->d_inode->i_atime = filp->f_path.dentry->d_inode->i_ctime = CURRENT_TIME;
 503	return ret;
 504}
 505
 506static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 507{
 508	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 509
 510	spin_lock(&info->lock);
 511	if (task_tgid(current) == info->notify_owner)
 512		remove_notification(info);
 513
 514	spin_unlock(&info->lock);
 515	return 0;
 516}
 517
 518static unsigned int mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 519{
 520	struct mqueue_inode_info *info = MQUEUE_I(filp->f_path.dentry->d_inode);
 521	int retval = 0;
 522
 523	poll_wait(filp, &info->wait_q, poll_tab);
 524
 525	spin_lock(&info->lock);
 526	if (info->attr.mq_curmsgs)
 527		retval = POLLIN | POLLRDNORM;
 528
 529	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 530		retval |= POLLOUT | POLLWRNORM;
 531	spin_unlock(&info->lock);
 532
 533	return retval;
 534}
 535
 536/* Adds current to info->e_wait_q[sr] before element with smaller prio */
 537static void wq_add(struct mqueue_inode_info *info, int sr,
 538			struct ext_wait_queue *ewp)
 539{
 540	struct ext_wait_queue *walk;
 541
 542	ewp->task = current;
 543
 544	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 545		if (walk->task->static_prio <= current->static_prio) {
 546			list_add_tail(&ewp->list, &walk->list);
 547			return;
 548		}
 549	}
 550	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 551}
 552
 553/*
 554 * Puts current task to sleep. Caller must hold queue lock. After return
 555 * lock isn't held.
 556 * sr: SEND or RECV
 557 */
 558static int wq_sleep(struct mqueue_inode_info *info, int sr,
 559		    ktime_t *timeout, struct ext_wait_queue *ewp)
 
 560{
 561	int retval;
 562	signed long time;
 563
 564	wq_add(info, sr, ewp);
 565
 566	for (;;) {
 567		set_current_state(TASK_INTERRUPTIBLE);
 568
 569		spin_unlock(&info->lock);
 570		time = schedule_hrtimeout_range_clock(timeout, 0,
 571			HRTIMER_MODE_ABS, CLOCK_REALTIME);
 572
 573		while (ewp->state == STATE_PENDING)
 574			cpu_relax();
 575
 576		if (ewp->state == STATE_READY) {
 577			retval = 0;
 578			goto out;
 579		}
 580		spin_lock(&info->lock);
 581		if (ewp->state == STATE_READY) {
 582			retval = 0;
 583			goto out_unlock;
 584		}
 585		if (signal_pending(current)) {
 586			retval = -ERESTARTSYS;
 587			break;
 588		}
 589		if (time == 0) {
 590			retval = -ETIMEDOUT;
 591			break;
 592		}
 593	}
 594	list_del(&ewp->list);
 595out_unlock:
 596	spin_unlock(&info->lock);
 597out:
 598	return retval;
 599}
 600
 601/*
 602 * Returns waiting task that should be serviced first or NULL if none exists
 603 */
 604static struct ext_wait_queue *wq_get_first_waiter(
 605		struct mqueue_inode_info *info, int sr)
 606{
 607	struct list_head *ptr;
 608
 609	ptr = info->e_wait_q[sr].list.prev;
 610	if (ptr == &info->e_wait_q[sr].list)
 611		return NULL;
 612	return list_entry(ptr, struct ext_wait_queue, list);
 613}
 614
 615
 616static inline void set_cookie(struct sk_buff *skb, char code)
 617{
 618	((char*)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 619}
 620
 621/*
 622 * The next function is only to split too long sys_mq_timedsend
 623 */
 624static void __do_notify(struct mqueue_inode_info *info)
 625{
 626	/* notification
 627	 * invoked when there is registered process and there isn't process
 628	 * waiting synchronously for message AND state of queue changed from
 629	 * empty to not empty. Here we are sure that no one is waiting
 630	 * synchronously. */
 631	if (info->notify_owner &&
 632	    info->attr.mq_curmsgs == 1) {
 633		struct siginfo sig_i;
 634		switch (info->notify.sigev_notify) {
 635		case SIGEV_NONE:
 636			break;
 637		case SIGEV_SIGNAL:
 638			/* sends signal */
 639
 
 640			sig_i.si_signo = info->notify.sigev_signo;
 641			sig_i.si_errno = 0;
 642			sig_i.si_code = SI_MESGQ;
 643			sig_i.si_value = info->notify.sigev_value;
 644			/* map current pid/uid into info->owner's namespaces */
 645			rcu_read_lock();
 646			sig_i.si_pid = task_tgid_nr_ns(current,
 647						ns_of_pid(info->notify_owner));
 648			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
 649			rcu_read_unlock();
 650
 651			kill_pid_info(info->notify.sigev_signo,
 652				      &sig_i, info->notify_owner);
 653			break;
 654		case SIGEV_THREAD:
 655			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 656			netlink_sendskb(info->notify_sock, info->notify_cookie);
 657			break;
 658		}
 659		/* after notification unregisters process */
 660		put_pid(info->notify_owner);
 661		put_user_ns(info->notify_user_ns);
 662		info->notify_owner = NULL;
 663		info->notify_user_ns = NULL;
 664	}
 665	wake_up(&info->wait_q);
 666}
 667
 668static int prepare_timeout(const struct timespec __user *u_abs_timeout,
 669			   ktime_t *expires, struct timespec *ts)
 670{
 671	if (copy_from_user(ts, u_abs_timeout, sizeof(struct timespec)))
 672		return -EFAULT;
 673	if (!timespec_valid(ts))
 674		return -EINVAL;
 675
 676	*expires = timespec_to_ktime(*ts);
 677	return 0;
 678}
 679
 680static void remove_notification(struct mqueue_inode_info *info)
 681{
 682	if (info->notify_owner != NULL &&
 683	    info->notify.sigev_notify == SIGEV_THREAD) {
 684		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 685		netlink_sendskb(info->notify_sock, info->notify_cookie);
 686	}
 687	put_pid(info->notify_owner);
 688	put_user_ns(info->notify_user_ns);
 689	info->notify_owner = NULL;
 690	info->notify_user_ns = NULL;
 691}
 692
 693static int mq_attr_ok(struct ipc_namespace *ipc_ns, struct mq_attr *attr)
 694{
 695	int mq_treesize;
 696	unsigned long total_size;
 697
 698	if (attr->mq_maxmsg <= 0 || attr->mq_msgsize <= 0)
 699		return -EINVAL;
 700	if (capable(CAP_SYS_RESOURCE)) {
 701		if (attr->mq_maxmsg > HARD_MSGMAX ||
 702		    attr->mq_msgsize > HARD_MSGSIZEMAX)
 703			return -EINVAL;
 704	} else {
 705		if (attr->mq_maxmsg > ipc_ns->mq_msg_max ||
 706				attr->mq_msgsize > ipc_ns->mq_msgsize_max)
 707			return -EINVAL;
 708	}
 709	/* check for overflow */
 710	if (attr->mq_msgsize > ULONG_MAX/attr->mq_maxmsg)
 711		return -EOVERFLOW;
 712	mq_treesize = attr->mq_maxmsg * sizeof(struct msg_msg) +
 713		min_t(unsigned int, attr->mq_maxmsg, MQ_PRIO_MAX) *
 714		sizeof(struct posix_msg_tree_node);
 715	total_size = attr->mq_maxmsg * attr->mq_msgsize;
 716	if (total_size + mq_treesize < total_size)
 717		return -EOVERFLOW;
 718	return 0;
 719}
 720
 721/*
 722 * Invoked when creating a new queue via sys_mq_open
 723 */
 724static struct file *do_create(struct ipc_namespace *ipc_ns, struct dentry *dir,
 725			struct dentry *dentry, int oflag, umode_t mode,
 726			struct mq_attr *attr)
 727{
 728	const struct cred *cred = current_cred();
 729	struct file *result;
 730	int ret;
 731
 732	if (attr) {
 733		ret = mq_attr_ok(ipc_ns, attr);
 734		if (ret)
 735			goto out;
 736		/* store for use during create */
 737		dentry->d_fsdata = attr;
 738	} else {
 739		struct mq_attr def_attr;
 740
 741		def_attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
 742					 ipc_ns->mq_msg_default);
 743		def_attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
 744					  ipc_ns->mq_msgsize_default);
 745		ret = mq_attr_ok(ipc_ns, &def_attr);
 746		if (ret)
 747			goto out;
 748	}
 749
 750	mode &= ~current_umask();
 751	ret = mnt_want_write(ipc_ns->mq_mnt);
 752	if (ret)
 753		goto out;
 754	ret = vfs_create(dir->d_inode, dentry, mode, NULL);
 755	dentry->d_fsdata = NULL;
 756	if (ret)
 757		goto out_drop_write;
 758
 759	result = dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 760	/*
 761	 * dentry_open() took a persistent mnt_want_write(),
 762	 * so we can now drop this one.
 763	 */
 764	mnt_drop_write(ipc_ns->mq_mnt);
 765	return result;
 766
 767out_drop_write:
 768	mnt_drop_write(ipc_ns->mq_mnt);
 769out:
 770	dput(dentry);
 771	mntput(ipc_ns->mq_mnt);
 772	return ERR_PTR(ret);
 773}
 774
 775/* Opens existing queue */
 776static struct file *do_open(struct ipc_namespace *ipc_ns,
 777				struct dentry *dentry, int oflag)
 778{
 779	int ret;
 780	const struct cred *cred = current_cred();
 781
 782	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 783						  MAY_READ | MAY_WRITE };
 
 784
 785	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY)) {
 786		ret = -EINVAL;
 787		goto err;
 788	}
 789
 790	if (inode_permission(dentry->d_inode, oflag2acc[oflag & O_ACCMODE])) {
 791		ret = -EACCES;
 792		goto err;
 793	}
 794
 795	return dentry_open(dentry, ipc_ns->mq_mnt, oflag, cred);
 796
 797err:
 798	dput(dentry);
 799	mntput(ipc_ns->mq_mnt);
 800	return ERR_PTR(ret);
 
 801}
 802
 803SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
 804		struct mq_attr __user *, u_attr)
 805{
 806	struct dentry *dentry;
 807	struct file *filp;
 808	char *name;
 809	struct mq_attr attr;
 810	int fd, error;
 811	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 812
 813	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 814		return -EFAULT;
 815
 816	audit_mq_open(oflag, mode, u_attr ? &attr : NULL);
 817
 818	if (IS_ERR(name = getname(u_name)))
 819		return PTR_ERR(name);
 820
 821	fd = get_unused_fd_flags(O_CLOEXEC);
 822	if (fd < 0)
 823		goto out_putname;
 824
 825	mutex_lock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 826	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 827	if (IS_ERR(dentry)) {
 828		error = PTR_ERR(dentry);
 
 829		goto out_putfd;
 830	}
 831	mntget(ipc_ns->mq_mnt);
 832
 833	if (oflag & O_CREAT) {
 834		if (dentry->d_inode) {	/* entry already exists */
 835			audit_inode(name, dentry);
 836			if (oflag & O_EXCL) {
 837				error = -EEXIST;
 838				goto out;
 839			}
 840			filp = do_open(ipc_ns, dentry, oflag);
 841		} else {
 842			filp = do_create(ipc_ns, ipc_ns->mq_mnt->mnt_root,
 843						dentry, oflag, mode,
 844						u_attr ? &attr : NULL);
 845		}
 846	} else {
 847		if (!dentry->d_inode) {
 848			error = -ENOENT;
 849			goto out;
 850		}
 851		audit_inode(name, dentry);
 852		filp = do_open(ipc_ns, dentry, oflag);
 853	}
 854
 855	if (IS_ERR(filp)) {
 856		error = PTR_ERR(filp);
 857		goto out_putfd;
 858	}
 859
 860	fd_install(fd, filp);
 861	goto out_upsem;
 862
 863out:
 864	dput(dentry);
 865	mntput(ipc_ns->mq_mnt);
 866out_putfd:
 867	put_unused_fd(fd);
 868	fd = error;
 869out_upsem:
 870	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 
 
 
 871out_putname:
 872	putname(name);
 873	return fd;
 874}
 875
 
 
 
 
 
 
 
 
 
 
 876SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 877{
 878	int err;
 879	char *name;
 880	struct dentry *dentry;
 881	struct inode *inode = NULL;
 882	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 
 883
 884	name = getname(u_name);
 885	if (IS_ERR(name))
 886		return PTR_ERR(name);
 887
 888	mutex_lock_nested(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex,
 889			I_MUTEX_PARENT);
 890	dentry = lookup_one_len(name, ipc_ns->mq_mnt->mnt_root, strlen(name));
 
 
 
 
 891	if (IS_ERR(dentry)) {
 892		err = PTR_ERR(dentry);
 893		goto out_unlock;
 894	}
 895
 896	if (!dentry->d_inode) {
 
 897		err = -ENOENT;
 898		goto out_err;
 899	}
 900
 901	inode = dentry->d_inode;
 902	if (inode)
 903		ihold(inode);
 904	err = mnt_want_write(ipc_ns->mq_mnt);
 905	if (err)
 906		goto out_err;
 907	err = vfs_unlink(dentry->d_parent->d_inode, dentry);
 908	mnt_drop_write(ipc_ns->mq_mnt);
 909out_err:
 910	dput(dentry);
 911
 912out_unlock:
 913	mutex_unlock(&ipc_ns->mq_mnt->mnt_root->d_inode->i_mutex);
 914	putname(name);
 915	if (inode)
 916		iput(inode);
 
 
 
 917
 918	return err;
 919}
 920
 921/* Pipelined send and receive functions.
 922 *
 923 * If a receiver finds no waiting message, then it registers itself in the
 924 * list of waiting receivers. A sender checks that list before adding the new
 925 * message into the message array. If there is a waiting receiver, then it
 926 * bypasses the message array and directly hands the message over to the
 927 * receiver.
 928 * The receiver accepts the message and returns without grabbing the queue
 929 * spinlock. Therefore an intermediate STATE_PENDING state and memory barriers
 930 * are necessary. The same algorithm is used for sysv semaphores, see
 931 * ipc/sem.c for more details.
 
 
 
 
 932 *
 933 * The same algorithm is used for senders.
 934 */
 935
 936/* pipelined_send() - send a message directly to the task waiting in
 937 * sys_mq_timedreceive() (without inserting message into a queue).
 938 */
 939static inline void pipelined_send(struct mqueue_inode_info *info,
 
 940				  struct msg_msg *message,
 941				  struct ext_wait_queue *receiver)
 942{
 943	receiver->msg = message;
 944	list_del(&receiver->list);
 945	receiver->state = STATE_PENDING;
 946	wake_up_process(receiver->task);
 947	smp_wmb();
 
 
 
 
 
 
 948	receiver->state = STATE_READY;
 949}
 950
 951/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 952 * gets its message and put to the queue (we have one free place for sure). */
 953static inline void pipelined_receive(struct mqueue_inode_info *info)
 
 954{
 955	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 956
 957	if (!sender) {
 958		/* for poll */
 959		wake_up_interruptible(&info->wait_q);
 960		return;
 961	}
 962	if (msg_insert(sender->msg, info))
 963		return;
 
 964	list_del(&sender->list);
 965	sender->state = STATE_PENDING;
 966	wake_up_process(sender->task);
 967	smp_wmb();
 968	sender->state = STATE_READY;
 969}
 970
 971SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
 972		size_t, msg_len, unsigned int, msg_prio,
 973		const struct timespec __user *, u_abs_timeout)
 974{
 975	struct file *filp;
 976	struct inode *inode;
 977	struct ext_wait_queue wait;
 978	struct ext_wait_queue *receiver;
 979	struct msg_msg *msg_ptr;
 980	struct mqueue_inode_info *info;
 981	ktime_t expires, *timeout = NULL;
 982	struct timespec ts;
 983	struct posix_msg_tree_node *new_leaf = NULL;
 984	int ret = 0;
 985
 986	if (u_abs_timeout) {
 987		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
 988		if (res)
 989			return res;
 990		timeout = &expires;
 991	}
 992
 993	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 994		return -EINVAL;
 995
 996	audit_mq_sendrecv(mqdes, msg_len, msg_prio, timeout ? &ts : NULL);
 
 
 
 
 
 997
 998	filp = fget(mqdes);
 999	if (unlikely(!filp)) {
1000		ret = -EBADF;
1001		goto out;
1002	}
1003
1004	inode = filp->f_path.dentry->d_inode;
1005	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1006		ret = -EBADF;
1007		goto out_fput;
1008	}
1009	info = MQUEUE_I(inode);
1010	audit_inode(NULL, filp->f_path.dentry);
1011
1012	if (unlikely(!(filp->f_mode & FMODE_WRITE))) {
1013		ret = -EBADF;
1014		goto out_fput;
1015	}
1016
1017	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1018		ret = -EMSGSIZE;
1019		goto out_fput;
1020	}
1021
1022	/* First try to allocate memory, before doing anything with
1023	 * existing queues. */
1024	msg_ptr = load_msg(u_msg_ptr, msg_len);
1025	if (IS_ERR(msg_ptr)) {
1026		ret = PTR_ERR(msg_ptr);
1027		goto out_fput;
1028	}
1029	msg_ptr->m_ts = msg_len;
1030	msg_ptr->m_type = msg_prio;
1031
1032	/*
1033	 * msg_insert really wants us to have a valid, spare node struct so
1034	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1035	 * fall back to that if necessary.
1036	 */
1037	if (!info->node_cache)
1038		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1039
1040	spin_lock(&info->lock);
1041
1042	if (!info->node_cache && new_leaf) {
1043		/* Save our speculative allocation into the cache */
1044		rb_init_node(&new_leaf->rb_node);
1045		INIT_LIST_HEAD(&new_leaf->msg_list);
1046		info->node_cache = new_leaf;
1047		info->qsize += sizeof(*new_leaf);
1048		new_leaf = NULL;
1049	} else {
1050		kfree(new_leaf);
1051	}
1052
1053	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1054		if (filp->f_flags & O_NONBLOCK) {
1055			ret = -EAGAIN;
1056		} else {
1057			wait.task = current;
1058			wait.msg = (void *) msg_ptr;
1059			wait.state = STATE_NONE;
1060			ret = wq_sleep(info, SEND, timeout, &wait);
1061			/*
1062			 * wq_sleep must be called with info->lock held, and
1063			 * returns with the lock released
1064			 */
1065			goto out_free;
1066		}
1067	} else {
1068		receiver = wq_get_first_waiter(info, RECV);
1069		if (receiver) {
1070			pipelined_send(info, msg_ptr, receiver);
1071		} else {
1072			/* adds message to the queue */
1073			ret = msg_insert(msg_ptr, info);
1074			if (ret)
1075				goto out_unlock;
1076			__do_notify(info);
1077		}
1078		inode->i_atime = inode->i_mtime = inode->i_ctime =
1079				CURRENT_TIME;
1080	}
1081out_unlock:
1082	spin_unlock(&info->lock);
 
1083out_free:
1084	if (ret)
1085		free_msg(msg_ptr);
1086out_fput:
1087	fput(filp);
1088out:
1089	return ret;
1090}
1091
1092SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1093		size_t, msg_len, unsigned int __user *, u_msg_prio,
1094		const struct timespec __user *, u_abs_timeout)
1095{
1096	ssize_t ret;
1097	struct msg_msg *msg_ptr;
1098	struct file *filp;
1099	struct inode *inode;
1100	struct mqueue_inode_info *info;
1101	struct ext_wait_queue wait;
1102	ktime_t expires, *timeout = NULL;
1103	struct timespec ts;
1104	struct posix_msg_tree_node *new_leaf = NULL;
1105
1106	if (u_abs_timeout) {
1107		int res = prepare_timeout(u_abs_timeout, &expires, &ts);
1108		if (res)
1109			return res;
1110		timeout = &expires;
1111	}
1112
1113	audit_mq_sendrecv(mqdes, msg_len, 0, timeout ? &ts : NULL);
1114
1115	filp = fget(mqdes);
1116	if (unlikely(!filp)) {
1117		ret = -EBADF;
1118		goto out;
1119	}
1120
1121	inode = filp->f_path.dentry->d_inode;
1122	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1123		ret = -EBADF;
1124		goto out_fput;
1125	}
1126	info = MQUEUE_I(inode);
1127	audit_inode(NULL, filp->f_path.dentry);
1128
1129	if (unlikely(!(filp->f_mode & FMODE_READ))) {
1130		ret = -EBADF;
1131		goto out_fput;
1132	}
1133
1134	/* checks if buffer is big enough */
1135	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1136		ret = -EMSGSIZE;
1137		goto out_fput;
1138	}
1139
1140	/*
1141	 * msg_insert really wants us to have a valid, spare node struct so
1142	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1143	 * fall back to that if necessary.
1144	 */
1145	if (!info->node_cache)
1146		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1147
1148	spin_lock(&info->lock);
1149
1150	if (!info->node_cache && new_leaf) {
1151		/* Save our speculative allocation into the cache */
1152		rb_init_node(&new_leaf->rb_node);
1153		INIT_LIST_HEAD(&new_leaf->msg_list);
1154		info->node_cache = new_leaf;
1155		info->qsize += sizeof(*new_leaf);
1156	} else {
1157		kfree(new_leaf);
1158	}
1159
1160	if (info->attr.mq_curmsgs == 0) {
1161		if (filp->f_flags & O_NONBLOCK) {
1162			spin_unlock(&info->lock);
1163			ret = -EAGAIN;
1164		} else {
1165			wait.task = current;
1166			wait.state = STATE_NONE;
1167			ret = wq_sleep(info, RECV, timeout, &wait);
1168			msg_ptr = wait.msg;
1169		}
1170	} else {
 
 
1171		msg_ptr = msg_get(info);
1172
1173		inode->i_atime = inode->i_mtime = inode->i_ctime =
1174				CURRENT_TIME;
1175
1176		/* There is now free space in queue. */
1177		pipelined_receive(info);
1178		spin_unlock(&info->lock);
 
1179		ret = 0;
1180	}
1181	if (ret == 0) {
1182		ret = msg_ptr->m_ts;
1183
1184		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1185			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1186			ret = -EFAULT;
1187		}
1188		free_msg(msg_ptr);
1189	}
1190out_fput:
1191	fput(filp);
1192out:
1193	return ret;
1194}
1195
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1196/*
1197 * Notes: the case when user wants us to deregister (with NULL as pointer)
1198 * and he isn't currently owner of notification, will be silently discarded.
1199 * It isn't explicitly defined in the POSIX.
1200 */
1201SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1202		const struct sigevent __user *, u_notification)
1203{
1204	int ret;
1205	struct file *filp;
1206	struct sock *sock;
1207	struct inode *inode;
1208	struct sigevent notification;
1209	struct mqueue_inode_info *info;
1210	struct sk_buff *nc;
1211
1212	if (u_notification) {
1213		if (copy_from_user(&notification, u_notification,
1214					sizeof(struct sigevent)))
1215			return -EFAULT;
1216	}
1217
1218	audit_mq_notify(mqdes, u_notification ? &notification : NULL);
1219
1220	nc = NULL;
1221	sock = NULL;
1222	if (u_notification != NULL) {
1223		if (unlikely(notification.sigev_notify != SIGEV_NONE &&
1224			     notification.sigev_notify != SIGEV_SIGNAL &&
1225			     notification.sigev_notify != SIGEV_THREAD))
1226			return -EINVAL;
1227		if (notification.sigev_notify == SIGEV_SIGNAL &&
1228			!valid_signal(notification.sigev_signo)) {
1229			return -EINVAL;
1230		}
1231		if (notification.sigev_notify == SIGEV_THREAD) {
1232			long timeo;
1233
1234			/* create the notify skb */
1235			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1236			if (!nc) {
1237				ret = -ENOMEM;
1238				goto out;
1239			}
1240			if (copy_from_user(nc->data,
1241					notification.sigev_value.sival_ptr,
1242					NOTIFY_COOKIE_LEN)) {
1243				ret = -EFAULT;
1244				goto out;
1245			}
1246
1247			/* TODO: add a header? */
1248			skb_put(nc, NOTIFY_COOKIE_LEN);
1249			/* and attach it to the socket */
1250retry:
1251			filp = fget(notification.sigev_signo);
1252			if (!filp) {
1253				ret = -EBADF;
1254				goto out;
1255			}
1256			sock = netlink_getsockbyfilp(filp);
1257			fput(filp);
1258			if (IS_ERR(sock)) {
1259				ret = PTR_ERR(sock);
1260				sock = NULL;
1261				goto out;
1262			}
1263
1264			timeo = MAX_SCHEDULE_TIMEOUT;
1265			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1266			if (ret == 1)
1267				goto retry;
1268			if (ret) {
1269				sock = NULL;
1270				nc = NULL;
1271				goto out;
1272			}
 
 
1273		}
1274	}
1275
1276	filp = fget(mqdes);
1277	if (!filp) {
1278		ret = -EBADF;
1279		goto out;
1280	}
1281
1282	inode = filp->f_path.dentry->d_inode;
1283	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1284		ret = -EBADF;
1285		goto out_fput;
1286	}
1287	info = MQUEUE_I(inode);
1288
1289	ret = 0;
1290	spin_lock(&info->lock);
1291	if (u_notification == NULL) {
1292		if (info->notify_owner == task_tgid(current)) {
1293			remove_notification(info);
1294			inode->i_atime = inode->i_ctime = CURRENT_TIME;
1295		}
1296	} else if (info->notify_owner != NULL) {
1297		ret = -EBUSY;
1298	} else {
1299		switch (notification.sigev_notify) {
1300		case SIGEV_NONE:
1301			info->notify.sigev_notify = SIGEV_NONE;
1302			break;
1303		case SIGEV_THREAD:
1304			info->notify_sock = sock;
1305			info->notify_cookie = nc;
1306			sock = NULL;
1307			nc = NULL;
1308			info->notify.sigev_notify = SIGEV_THREAD;
1309			break;
1310		case SIGEV_SIGNAL:
1311			info->notify.sigev_signo = notification.sigev_signo;
1312			info->notify.sigev_value = notification.sigev_value;
1313			info->notify.sigev_notify = SIGEV_SIGNAL;
1314			break;
1315		}
1316
1317		info->notify_owner = get_pid(task_tgid(current));
1318		info->notify_user_ns = get_user_ns(current_user_ns());
1319		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1320	}
1321	spin_unlock(&info->lock);
1322out_fput:
1323	fput(filp);
1324out:
1325	if (sock) {
1326		netlink_detachskb(sock, nc);
1327	} else if (nc) {
 
1328		dev_kfree_skb(nc);
1329	}
1330	return ret;
1331}
1332
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1333SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1334		const struct mq_attr __user *, u_mqstat,
1335		struct mq_attr __user *, u_omqstat)
1336{
1337	int ret;
1338	struct mq_attr mqstat, omqstat;
1339	struct file *filp;
1340	struct inode *inode;
1341	struct mqueue_inode_info *info;
1342
1343	if (u_mqstat != NULL) {
1344		if (copy_from_user(&mqstat, u_mqstat, sizeof(struct mq_attr)))
 
1345			return -EFAULT;
1346		if (mqstat.mq_flags & (~O_NONBLOCK))
1347			return -EINVAL;
1348	}
 
 
1349
1350	filp = fget(mqdes);
1351	if (!filp) {
1352		ret = -EBADF;
1353		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1354	}
 
 
1355
1356	inode = filp->f_path.dentry->d_inode;
1357	if (unlikely(filp->f_op != &mqueue_file_operations)) {
1358		ret = -EBADF;
1359		goto out_fput;
 
 
 
 
 
 
1360	}
1361	info = MQUEUE_I(inode);
 
1362
1363	spin_lock(&info->lock);
 
 
 
 
 
 
1364
1365	omqstat = info->attr;
1366	omqstat.mq_flags = filp->f_flags & O_NONBLOCK;
1367	if (u_mqstat) {
1368		audit_mq_getsetattr(mqdes, &mqstat);
1369		spin_lock(&filp->f_lock);
1370		if (mqstat.mq_flags & O_NONBLOCK)
1371			filp->f_flags |= O_NONBLOCK;
1372		else
1373			filp->f_flags &= ~O_NONBLOCK;
1374		spin_unlock(&filp->f_lock);
1375
1376		inode->i_atime = inode->i_ctime = CURRENT_TIME;
1377	}
 
 
1378
1379	spin_unlock(&info->lock);
 
 
1380
1381	ret = 0;
1382	if (u_omqstat != NULL && copy_to_user(u_omqstat, &omqstat,
1383						sizeof(struct mq_attr)))
1384		ret = -EFAULT;
 
1385
1386out_fput:
1387	fput(filp);
1388out:
1389	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1390}
 
1391
1392static const struct inode_operations mqueue_dir_inode_operations = {
1393	.lookup = simple_lookup,
1394	.create = mqueue_create,
1395	.unlink = mqueue_unlink,
1396};
1397
1398static const struct file_operations mqueue_file_operations = {
1399	.flush = mqueue_flush_file,
1400	.poll = mqueue_poll_file,
1401	.read = mqueue_read_file,
1402	.llseek = default_llseek,
1403};
1404
1405static const struct super_operations mqueue_super_ops = {
1406	.alloc_inode = mqueue_alloc_inode,
1407	.destroy_inode = mqueue_destroy_inode,
1408	.evict_inode = mqueue_evict_inode,
1409	.statfs = simple_statfs,
1410};
1411
 
 
 
 
 
1412static struct file_system_type mqueue_fs_type = {
1413	.name = "mqueue",
1414	.mount = mqueue_mount,
1415	.kill_sb = kill_litter_super,
 
1416};
1417
1418int mq_init_ns(struct ipc_namespace *ns)
1419{
 
 
1420	ns->mq_queues_count  = 0;
1421	ns->mq_queues_max    = DFLT_QUEUESMAX;
1422	ns->mq_msg_max       = DFLT_MSGMAX;
1423	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1424	ns->mq_msg_default   = DFLT_MSG;
1425	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1426
1427	ns->mq_mnt = kern_mount_data(&mqueue_fs_type, ns);
1428	if (IS_ERR(ns->mq_mnt)) {
1429		int err = PTR_ERR(ns->mq_mnt);
1430		ns->mq_mnt = NULL;
1431		return err;
1432	}
1433	return 0;
1434}
1435
1436void mq_clear_sbinfo(struct ipc_namespace *ns)
1437{
1438	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1439}
1440
1441void mq_put_mnt(struct ipc_namespace *ns)
1442{
1443	kern_unmount(ns->mq_mnt);
1444}
1445
1446static int __init init_mqueue_fs(void)
1447{
1448	int error;
1449
1450	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1451				sizeof(struct mqueue_inode_info), 0,
1452				SLAB_HWCACHE_ALIGN, init_once);
1453	if (mqueue_inode_cachep == NULL)
1454		return -ENOMEM;
1455
1456	/* ignore failures - they are not fatal */
1457	mq_sysctl_table = mq_register_sysctl_table();
1458
1459	error = register_filesystem(&mqueue_fs_type);
1460	if (error)
1461		goto out_sysctl;
1462
1463	spin_lock_init(&mq_lock);
1464
1465	error = mq_init_ns(&init_ipc_ns);
1466	if (error)
1467		goto out_filesystem;
1468
1469	return 0;
1470
1471out_filesystem:
1472	unregister_filesystem(&mqueue_fs_type);
1473out_sysctl:
1474	if (mq_sysctl_table)
1475		unregister_sysctl_table(mq_sysctl_table);
1476	kmem_cache_destroy(mqueue_inode_cachep);
1477	return error;
1478}
1479
1480__initcall(init_mqueue_fs);
v5.4
   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/fs_context.h>
  22#include <linux/namei.h>
  23#include <linux/sysctl.h>
  24#include <linux/poll.h>
  25#include <linux/mqueue.h>
  26#include <linux/msg.h>
  27#include <linux/skbuff.h>
  28#include <linux/vmalloc.h>
  29#include <linux/netlink.h>
  30#include <linux/syscalls.h>
  31#include <linux/audit.h>
  32#include <linux/signal.h>
  33#include <linux/mutex.h>
  34#include <linux/nsproxy.h>
  35#include <linux/pid.h>
  36#include <linux/ipc_namespace.h>
  37#include <linux/user_namespace.h>
  38#include <linux/slab.h>
  39#include <linux/sched/wake_q.h>
  40#include <linux/sched/signal.h>
  41#include <linux/sched/user.h>
  42
  43#include <net/sock.h>
  44#include "util.h"
  45
  46struct mqueue_fs_context {
  47	struct ipc_namespace	*ipc_ns;
  48};
  49
  50#define MQUEUE_MAGIC	0x19800202
  51#define DIRENT_SIZE	20
  52#define FILENT_SIZE	80
  53
  54#define SEND		0
  55#define RECV		1
  56
  57#define STATE_NONE	0
  58#define STATE_READY	1
 
  59
  60struct posix_msg_tree_node {
  61	struct rb_node		rb_node;
  62	struct list_head	msg_list;
  63	int			priority;
  64};
  65
  66struct ext_wait_queue {		/* queue of sleeping tasks */
  67	struct task_struct *task;
  68	struct list_head list;
  69	struct msg_msg *msg;	/* ptr of loaded message */
  70	int state;		/* one of STATE_* values */
  71};
  72
  73struct mqueue_inode_info {
  74	spinlock_t lock;
  75	struct inode vfs_inode;
  76	wait_queue_head_t wait_q;
  77
  78	struct rb_root msg_tree;
  79	struct rb_node *msg_tree_rightmost;
  80	struct posix_msg_tree_node *node_cache;
  81	struct mq_attr attr;
  82
  83	struct sigevent notify;
  84	struct pid *notify_owner;
  85	struct user_namespace *notify_user_ns;
  86	struct user_struct *user;	/* user who created, for accounting */
  87	struct sock *notify_sock;
  88	struct sk_buff *notify_cookie;
  89
  90	/* for tasks waiting for free space and messages, respectively */
  91	struct ext_wait_queue e_wait_q[2];
  92
  93	unsigned long qsize; /* size of queue in memory (sum of all msgs) */
  94};
  95
  96static struct file_system_type mqueue_fs_type;
  97static const struct inode_operations mqueue_dir_inode_operations;
  98static const struct file_operations mqueue_file_operations;
  99static const struct super_operations mqueue_super_ops;
 100static const struct fs_context_operations mqueue_fs_context_ops;
 101static void remove_notification(struct mqueue_inode_info *info);
 102
 103static struct kmem_cache *mqueue_inode_cachep;
 104
 105static struct ctl_table_header *mq_sysctl_table;
 106
 107static inline struct mqueue_inode_info *MQUEUE_I(struct inode *inode)
 108{
 109	return container_of(inode, struct mqueue_inode_info, vfs_inode);
 110}
 111
 112/*
 113 * This routine should be called with the mq_lock held.
 114 */
 115static inline struct ipc_namespace *__get_ns_from_inode(struct inode *inode)
 116{
 117	return get_ipc_ns(inode->i_sb->s_fs_info);
 118}
 119
 120static struct ipc_namespace *get_ns_from_inode(struct inode *inode)
 121{
 122	struct ipc_namespace *ns;
 123
 124	spin_lock(&mq_lock);
 125	ns = __get_ns_from_inode(inode);
 126	spin_unlock(&mq_lock);
 127	return ns;
 128}
 129
 130/* Auxiliary functions to manipulate messages' list */
 131static int msg_insert(struct msg_msg *msg, struct mqueue_inode_info *info)
 132{
 133	struct rb_node **p, *parent = NULL;
 134	struct posix_msg_tree_node *leaf;
 135	bool rightmost = true;
 136
 137	p = &info->msg_tree.rb_node;
 138	while (*p) {
 139		parent = *p;
 140		leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
 141
 142		if (likely(leaf->priority == msg->m_type))
 143			goto insert_msg;
 144		else if (msg->m_type < leaf->priority) {
 145			p = &(*p)->rb_left;
 146			rightmost = false;
 147		} else
 148			p = &(*p)->rb_right;
 149	}
 150	if (info->node_cache) {
 151		leaf = info->node_cache;
 152		info->node_cache = NULL;
 153	} else {
 154		leaf = kmalloc(sizeof(*leaf), GFP_ATOMIC);
 155		if (!leaf)
 156			return -ENOMEM;
 
 157		INIT_LIST_HEAD(&leaf->msg_list);
 
 158	}
 159	leaf->priority = msg->m_type;
 160
 161	if (rightmost)
 162		info->msg_tree_rightmost = &leaf->rb_node;
 163
 164	rb_link_node(&leaf->rb_node, parent, p);
 165	rb_insert_color(&leaf->rb_node, &info->msg_tree);
 166insert_msg:
 167	info->attr.mq_curmsgs++;
 168	info->qsize += msg->m_ts;
 169	list_add_tail(&msg->m_list, &leaf->msg_list);
 170	return 0;
 171}
 172
 173static inline void msg_tree_erase(struct posix_msg_tree_node *leaf,
 174				  struct mqueue_inode_info *info)
 175{
 176	struct rb_node *node = &leaf->rb_node;
 177
 178	if (info->msg_tree_rightmost == node)
 179		info->msg_tree_rightmost = rb_prev(node);
 180
 181	rb_erase(node, &info->msg_tree);
 182	if (info->node_cache) {
 183		kfree(leaf);
 184	} else {
 185		info->node_cache = leaf;
 186	}
 187}
 188
 189static inline struct msg_msg *msg_get(struct mqueue_inode_info *info)
 190{
 191	struct rb_node *parent = NULL;
 192	struct posix_msg_tree_node *leaf;
 193	struct msg_msg *msg;
 194
 195try_again:
 196	/*
 197	 * During insert, low priorities go to the left and high to the
 198	 * right.  On receive, we want the highest priorities first, so
 199	 * walk all the way to the right.
 200	 */
 201	parent = info->msg_tree_rightmost;
 
 
 
 
 202	if (!parent) {
 203		if (info->attr.mq_curmsgs) {
 204			pr_warn_once("Inconsistency in POSIX message queue, "
 205				     "no tree element, but supposedly messages "
 206				     "should exist!\n");
 207			info->attr.mq_curmsgs = 0;
 208		}
 209		return NULL;
 210	}
 211	leaf = rb_entry(parent, struct posix_msg_tree_node, rb_node);
 212	if (unlikely(list_empty(&leaf->msg_list))) {
 213		pr_warn_once("Inconsistency in POSIX message queue, "
 214			     "empty leaf node but we haven't implemented "
 215			     "lazy leaf delete!\n");
 216		msg_tree_erase(leaf, info);
 
 
 
 
 
 
 217		goto try_again;
 218	} else {
 219		msg = list_first_entry(&leaf->msg_list,
 220				       struct msg_msg, m_list);
 221		list_del(&msg->m_list);
 222		if (list_empty(&leaf->msg_list)) {
 223			msg_tree_erase(leaf, info);
 
 
 
 
 
 
 224		}
 225	}
 226	info->attr.mq_curmsgs--;
 227	info->qsize -= msg->m_ts;
 228	return msg;
 229}
 230
 231static struct inode *mqueue_get_inode(struct super_block *sb,
 232		struct ipc_namespace *ipc_ns, umode_t mode,
 233		struct mq_attr *attr)
 234{
 235	struct user_struct *u = current_user();
 236	struct inode *inode;
 237	int ret = -ENOMEM;
 238
 239	inode = new_inode(sb);
 240	if (!inode)
 241		goto err;
 242
 243	inode->i_ino = get_next_ino();
 244	inode->i_mode = mode;
 245	inode->i_uid = current_fsuid();
 246	inode->i_gid = current_fsgid();
 247	inode->i_mtime = inode->i_ctime = inode->i_atime = current_time(inode);
 248
 249	if (S_ISREG(mode)) {
 250		struct mqueue_inode_info *info;
 251		unsigned long mq_bytes, mq_treesize;
 252
 253		inode->i_fop = &mqueue_file_operations;
 254		inode->i_size = FILENT_SIZE;
 255		/* mqueue specific info */
 256		info = MQUEUE_I(inode);
 257		spin_lock_init(&info->lock);
 258		init_waitqueue_head(&info->wait_q);
 259		INIT_LIST_HEAD(&info->e_wait_q[0].list);
 260		INIT_LIST_HEAD(&info->e_wait_q[1].list);
 261		info->notify_owner = NULL;
 262		info->notify_user_ns = NULL;
 263		info->qsize = 0;
 264		info->user = NULL;	/* set when all is ok */
 265		info->msg_tree = RB_ROOT;
 266		info->msg_tree_rightmost = NULL;
 267		info->node_cache = NULL;
 268		memset(&info->attr, 0, sizeof(info->attr));
 269		info->attr.mq_maxmsg = min(ipc_ns->mq_msg_max,
 270					   ipc_ns->mq_msg_default);
 271		info->attr.mq_msgsize = min(ipc_ns->mq_msgsize_max,
 272					    ipc_ns->mq_msgsize_default);
 273		if (attr) {
 274			info->attr.mq_maxmsg = attr->mq_maxmsg;
 275			info->attr.mq_msgsize = attr->mq_msgsize;
 276		}
 277		/*
 278		 * We used to allocate a static array of pointers and account
 279		 * the size of that array as well as one msg_msg struct per
 280		 * possible message into the queue size. That's no longer
 281		 * accurate as the queue is now an rbtree and will grow and
 282		 * shrink depending on usage patterns.  We can, however, still
 283		 * account one msg_msg struct per message, but the nodes are
 284		 * allocated depending on priority usage, and most programs
 285		 * only use one, or a handful, of priorities.  However, since
 286		 * this is pinned memory, we need to assume worst case, so
 287		 * that means the min(mq_maxmsg, max_priorities) * struct
 288		 * posix_msg_tree_node.
 289		 */
 290
 291		ret = -EINVAL;
 292		if (info->attr.mq_maxmsg <= 0 || info->attr.mq_msgsize <= 0)
 293			goto out_inode;
 294		if (capable(CAP_SYS_RESOURCE)) {
 295			if (info->attr.mq_maxmsg > HARD_MSGMAX ||
 296			    info->attr.mq_msgsize > HARD_MSGSIZEMAX)
 297				goto out_inode;
 298		} else {
 299			if (info->attr.mq_maxmsg > ipc_ns->mq_msg_max ||
 300					info->attr.mq_msgsize > ipc_ns->mq_msgsize_max)
 301				goto out_inode;
 302		}
 303		ret = -EOVERFLOW;
 304		/* check for overflow */
 305		if (info->attr.mq_msgsize > ULONG_MAX/info->attr.mq_maxmsg)
 306			goto out_inode;
 307		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 308			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 309			sizeof(struct posix_msg_tree_node);
 310		mq_bytes = info->attr.mq_maxmsg * info->attr.mq_msgsize;
 311		if (mq_bytes + mq_treesize < mq_bytes)
 312			goto out_inode;
 313		mq_bytes += mq_treesize;
 314		spin_lock(&mq_lock);
 315		if (u->mq_bytes + mq_bytes < u->mq_bytes ||
 316		    u->mq_bytes + mq_bytes > rlimit(RLIMIT_MSGQUEUE)) {
 317			spin_unlock(&mq_lock);
 318			/* mqueue_evict_inode() releases info->messages */
 319			ret = -EMFILE;
 320			goto out_inode;
 321		}
 322		u->mq_bytes += mq_bytes;
 323		spin_unlock(&mq_lock);
 324
 325		/* all is ok */
 326		info->user = get_uid(u);
 327	} else if (S_ISDIR(mode)) {
 328		inc_nlink(inode);
 329		/* Some things misbehave if size == 0 on a directory */
 330		inode->i_size = 2 * DIRENT_SIZE;
 331		inode->i_op = &mqueue_dir_inode_operations;
 332		inode->i_fop = &simple_dir_operations;
 333	}
 334
 335	return inode;
 336out_inode:
 337	iput(inode);
 338err:
 339	return ERR_PTR(ret);
 340}
 341
 342static int mqueue_fill_super(struct super_block *sb, struct fs_context *fc)
 343{
 344	struct inode *inode;
 345	struct ipc_namespace *ns = sb->s_fs_info;
 346
 347	sb->s_iflags |= SB_I_NOEXEC | SB_I_NODEV;
 348	sb->s_blocksize = PAGE_SIZE;
 349	sb->s_blocksize_bits = PAGE_SHIFT;
 350	sb->s_magic = MQUEUE_MAGIC;
 351	sb->s_op = &mqueue_super_ops;
 352
 353	inode = mqueue_get_inode(sb, ns, S_IFDIR | S_ISVTX | S_IRWXUGO, NULL);
 354	if (IS_ERR(inode))
 355		return PTR_ERR(inode);
 356
 357	sb->s_root = d_make_root(inode);
 358	if (!sb->s_root)
 359		return -ENOMEM;
 360	return 0;
 361}
 362
 363static int mqueue_get_tree(struct fs_context *fc)
 364{
 365	struct mqueue_fs_context *ctx = fc->fs_private;
 366
 367	return get_tree_keyed(fc, mqueue_fill_super, ctx->ipc_ns);
 368}
 369
 370static void mqueue_fs_context_free(struct fs_context *fc)
 371{
 372	struct mqueue_fs_context *ctx = fc->fs_private;
 373
 374	put_ipc_ns(ctx->ipc_ns);
 375	kfree(ctx);
 376}
 377
 378static int mqueue_init_fs_context(struct fs_context *fc)
 379{
 380	struct mqueue_fs_context *ctx;
 381
 382	ctx = kzalloc(sizeof(struct mqueue_fs_context), GFP_KERNEL);
 383	if (!ctx)
 384		return -ENOMEM;
 385
 386	ctx->ipc_ns = get_ipc_ns(current->nsproxy->ipc_ns);
 387	put_user_ns(fc->user_ns);
 388	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
 389	fc->fs_private = ctx;
 390	fc->ops = &mqueue_fs_context_ops;
 391	return 0;
 392}
 393
 394static struct vfsmount *mq_create_mount(struct ipc_namespace *ns)
 395{
 396	struct mqueue_fs_context *ctx;
 397	struct fs_context *fc;
 398	struct vfsmount *mnt;
 399
 400	fc = fs_context_for_mount(&mqueue_fs_type, SB_KERNMOUNT);
 401	if (IS_ERR(fc))
 402		return ERR_CAST(fc);
 403
 404	ctx = fc->fs_private;
 405	put_ipc_ns(ctx->ipc_ns);
 406	ctx->ipc_ns = get_ipc_ns(ns);
 407	put_user_ns(fc->user_ns);
 408	fc->user_ns = get_user_ns(ctx->ipc_ns->user_ns);
 409
 410	mnt = fc_mount(fc);
 411	put_fs_context(fc);
 412	return mnt;
 413}
 414
 415static void init_once(void *foo)
 416{
 417	struct mqueue_inode_info *p = (struct mqueue_inode_info *) foo;
 418
 419	inode_init_once(&p->vfs_inode);
 420}
 421
 422static struct inode *mqueue_alloc_inode(struct super_block *sb)
 423{
 424	struct mqueue_inode_info *ei;
 425
 426	ei = kmem_cache_alloc(mqueue_inode_cachep, GFP_KERNEL);
 427	if (!ei)
 428		return NULL;
 429	return &ei->vfs_inode;
 430}
 431
 432static void mqueue_free_inode(struct inode *inode)
 433{
 
 434	kmem_cache_free(mqueue_inode_cachep, MQUEUE_I(inode));
 435}
 436
 
 
 
 
 
 437static void mqueue_evict_inode(struct inode *inode)
 438{
 439	struct mqueue_inode_info *info;
 440	struct user_struct *user;
 
 441	struct ipc_namespace *ipc_ns;
 442	struct msg_msg *msg, *nmsg;
 443	LIST_HEAD(tmp_msg);
 444
 445	clear_inode(inode);
 446
 447	if (S_ISDIR(inode->i_mode))
 448		return;
 449
 450	ipc_ns = get_ns_from_inode(inode);
 451	info = MQUEUE_I(inode);
 452	spin_lock(&info->lock);
 453	while ((msg = msg_get(info)) != NULL)
 454		list_add_tail(&msg->m_list, &tmp_msg);
 455	kfree(info->node_cache);
 456	spin_unlock(&info->lock);
 457
 458	list_for_each_entry_safe(msg, nmsg, &tmp_msg, m_list) {
 459		list_del(&msg->m_list);
 460		free_msg(msg);
 461	}
 
 
 
 462
 463	user = info->user;
 464	if (user) {
 465		unsigned long mq_bytes, mq_treesize;
 466
 467		/* Total amount of bytes accounted for the mqueue */
 468		mq_treesize = info->attr.mq_maxmsg * sizeof(struct msg_msg) +
 469			min_t(unsigned int, info->attr.mq_maxmsg, MQ_PRIO_MAX) *
 470			sizeof(struct posix_msg_tree_node);
 471
 472		mq_bytes = mq_treesize + (info->attr.mq_maxmsg *
 473					  info->attr.mq_msgsize);
 474
 475		spin_lock(&mq_lock);
 476		user->mq_bytes -= mq_bytes;
 477		/*
 478		 * get_ns_from_inode() ensures that the
 479		 * (ipc_ns = sb->s_fs_info) is either a valid ipc_ns
 480		 * to which we now hold a reference, or it is NULL.
 481		 * We can't put it here under mq_lock, though.
 482		 */
 483		if (ipc_ns)
 484			ipc_ns->mq_queues_count--;
 485		spin_unlock(&mq_lock);
 486		free_uid(user);
 487	}
 488	if (ipc_ns)
 489		put_ipc_ns(ipc_ns);
 490}
 491
 492static int mqueue_create_attr(struct dentry *dentry, umode_t mode, void *arg)
 
 493{
 494	struct inode *dir = dentry->d_parent->d_inode;
 495	struct inode *inode;
 496	struct mq_attr *attr = arg;
 497	int error;
 498	struct ipc_namespace *ipc_ns;
 499
 500	spin_lock(&mq_lock);
 501	ipc_ns = __get_ns_from_inode(dir);
 502	if (!ipc_ns) {
 503		error = -EACCES;
 504		goto out_unlock;
 505	}
 506
 507	if (ipc_ns->mq_queues_count >= ipc_ns->mq_queues_max &&
 508	    !capable(CAP_SYS_RESOURCE)) {
 509		error = -ENOSPC;
 510		goto out_unlock;
 511	}
 512	ipc_ns->mq_queues_count++;
 513	spin_unlock(&mq_lock);
 514
 515	inode = mqueue_get_inode(dir->i_sb, ipc_ns, mode, attr);
 516	if (IS_ERR(inode)) {
 517		error = PTR_ERR(inode);
 518		spin_lock(&mq_lock);
 519		ipc_ns->mq_queues_count--;
 520		goto out_unlock;
 521	}
 522
 523	put_ipc_ns(ipc_ns);
 524	dir->i_size += DIRENT_SIZE;
 525	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
 526
 527	d_instantiate(dentry, inode);
 528	dget(dentry);
 529	return 0;
 530out_unlock:
 531	spin_unlock(&mq_lock);
 532	if (ipc_ns)
 533		put_ipc_ns(ipc_ns);
 534	return error;
 535}
 536
 537static int mqueue_create(struct inode *dir, struct dentry *dentry,
 538				umode_t mode, bool excl)
 539{
 540	return mqueue_create_attr(dentry, mode, NULL);
 541}
 542
 543static int mqueue_unlink(struct inode *dir, struct dentry *dentry)
 544{
 545	struct inode *inode = d_inode(dentry);
 546
 547	dir->i_ctime = dir->i_mtime = dir->i_atime = current_time(dir);
 548	dir->i_size -= DIRENT_SIZE;
 549	drop_nlink(inode);
 550	dput(dentry);
 551	return 0;
 552}
 553
 554/*
 555*	This is routine for system read from queue file.
 556*	To avoid mess with doing here some sort of mq_receive we allow
 557*	to read only queue size & notification info (the only values
 558*	that are interesting from user point of view and aren't accessible
 559*	through std routines)
 560*/
 561static ssize_t mqueue_read_file(struct file *filp, char __user *u_data,
 562				size_t count, loff_t *off)
 563{
 564	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 565	char buffer[FILENT_SIZE];
 566	ssize_t ret;
 567
 568	spin_lock(&info->lock);
 569	snprintf(buffer, sizeof(buffer),
 570			"QSIZE:%-10lu NOTIFY:%-5d SIGNO:%-5d NOTIFY_PID:%-6d\n",
 571			info->qsize,
 572			info->notify_owner ? info->notify.sigev_notify : 0,
 573			(info->notify_owner &&
 574			 info->notify.sigev_notify == SIGEV_SIGNAL) ?
 575				info->notify.sigev_signo : 0,
 576			pid_vnr(info->notify_owner));
 577	spin_unlock(&info->lock);
 578	buffer[sizeof(buffer)-1] = '\0';
 579
 580	ret = simple_read_from_buffer(u_data, count, off, buffer,
 581				strlen(buffer));
 582	if (ret <= 0)
 583		return ret;
 584
 585	file_inode(filp)->i_atime = file_inode(filp)->i_ctime = current_time(file_inode(filp));
 586	return ret;
 587}
 588
 589static int mqueue_flush_file(struct file *filp, fl_owner_t id)
 590{
 591	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 592
 593	spin_lock(&info->lock);
 594	if (task_tgid(current) == info->notify_owner)
 595		remove_notification(info);
 596
 597	spin_unlock(&info->lock);
 598	return 0;
 599}
 600
 601static __poll_t mqueue_poll_file(struct file *filp, struct poll_table_struct *poll_tab)
 602{
 603	struct mqueue_inode_info *info = MQUEUE_I(file_inode(filp));
 604	__poll_t retval = 0;
 605
 606	poll_wait(filp, &info->wait_q, poll_tab);
 607
 608	spin_lock(&info->lock);
 609	if (info->attr.mq_curmsgs)
 610		retval = EPOLLIN | EPOLLRDNORM;
 611
 612	if (info->attr.mq_curmsgs < info->attr.mq_maxmsg)
 613		retval |= EPOLLOUT | EPOLLWRNORM;
 614	spin_unlock(&info->lock);
 615
 616	return retval;
 617}
 618
 619/* Adds current to info->e_wait_q[sr] before element with smaller prio */
 620static void wq_add(struct mqueue_inode_info *info, int sr,
 621			struct ext_wait_queue *ewp)
 622{
 623	struct ext_wait_queue *walk;
 624
 
 
 625	list_for_each_entry(walk, &info->e_wait_q[sr].list, list) {
 626		if (walk->task->prio <= current->prio) {
 627			list_add_tail(&ewp->list, &walk->list);
 628			return;
 629		}
 630	}
 631	list_add_tail(&ewp->list, &info->e_wait_q[sr].list);
 632}
 633
 634/*
 635 * Puts current task to sleep. Caller must hold queue lock. After return
 636 * lock isn't held.
 637 * sr: SEND or RECV
 638 */
 639static int wq_sleep(struct mqueue_inode_info *info, int sr,
 640		    ktime_t *timeout, struct ext_wait_queue *ewp)
 641	__releases(&info->lock)
 642{
 643	int retval;
 644	signed long time;
 645
 646	wq_add(info, sr, ewp);
 647
 648	for (;;) {
 649		__set_current_state(TASK_INTERRUPTIBLE);
 650
 651		spin_unlock(&info->lock);
 652		time = schedule_hrtimeout_range_clock(timeout, 0,
 653			HRTIMER_MODE_ABS, CLOCK_REALTIME);
 654
 
 
 
 655		if (ewp->state == STATE_READY) {
 656			retval = 0;
 657			goto out;
 658		}
 659		spin_lock(&info->lock);
 660		if (ewp->state == STATE_READY) {
 661			retval = 0;
 662			goto out_unlock;
 663		}
 664		if (signal_pending(current)) {
 665			retval = -ERESTARTSYS;
 666			break;
 667		}
 668		if (time == 0) {
 669			retval = -ETIMEDOUT;
 670			break;
 671		}
 672	}
 673	list_del(&ewp->list);
 674out_unlock:
 675	spin_unlock(&info->lock);
 676out:
 677	return retval;
 678}
 679
 680/*
 681 * Returns waiting task that should be serviced first or NULL if none exists
 682 */
 683static struct ext_wait_queue *wq_get_first_waiter(
 684		struct mqueue_inode_info *info, int sr)
 685{
 686	struct list_head *ptr;
 687
 688	ptr = info->e_wait_q[sr].list.prev;
 689	if (ptr == &info->e_wait_q[sr].list)
 690		return NULL;
 691	return list_entry(ptr, struct ext_wait_queue, list);
 692}
 693
 694
 695static inline void set_cookie(struct sk_buff *skb, char code)
 696{
 697	((char *)skb->data)[NOTIFY_COOKIE_LEN-1] = code;
 698}
 699
 700/*
 701 * The next function is only to split too long sys_mq_timedsend
 702 */
 703static void __do_notify(struct mqueue_inode_info *info)
 704{
 705	/* notification
 706	 * invoked when there is registered process and there isn't process
 707	 * waiting synchronously for message AND state of queue changed from
 708	 * empty to not empty. Here we are sure that no one is waiting
 709	 * synchronously. */
 710	if (info->notify_owner &&
 711	    info->attr.mq_curmsgs == 1) {
 712		struct kernel_siginfo sig_i;
 713		switch (info->notify.sigev_notify) {
 714		case SIGEV_NONE:
 715			break;
 716		case SIGEV_SIGNAL:
 717			/* sends signal */
 718
 719			clear_siginfo(&sig_i);
 720			sig_i.si_signo = info->notify.sigev_signo;
 721			sig_i.si_errno = 0;
 722			sig_i.si_code = SI_MESGQ;
 723			sig_i.si_value = info->notify.sigev_value;
 724			/* map current pid/uid into info->owner's namespaces */
 725			rcu_read_lock();
 726			sig_i.si_pid = task_tgid_nr_ns(current,
 727						ns_of_pid(info->notify_owner));
 728			sig_i.si_uid = from_kuid_munged(info->notify_user_ns, current_uid());
 729			rcu_read_unlock();
 730
 731			kill_pid_info(info->notify.sigev_signo,
 732				      &sig_i, info->notify_owner);
 733			break;
 734		case SIGEV_THREAD:
 735			set_cookie(info->notify_cookie, NOTIFY_WOKENUP);
 736			netlink_sendskb(info->notify_sock, info->notify_cookie);
 737			break;
 738		}
 739		/* after notification unregisters process */
 740		put_pid(info->notify_owner);
 741		put_user_ns(info->notify_user_ns);
 742		info->notify_owner = NULL;
 743		info->notify_user_ns = NULL;
 744	}
 745	wake_up(&info->wait_q);
 746}
 747
 748static int prepare_timeout(const struct __kernel_timespec __user *u_abs_timeout,
 749			   struct timespec64 *ts)
 750{
 751	if (get_timespec64(ts, u_abs_timeout))
 752		return -EFAULT;
 753	if (!timespec64_valid(ts))
 754		return -EINVAL;
 
 
 755	return 0;
 756}
 757
 758static void remove_notification(struct mqueue_inode_info *info)
 759{
 760	if (info->notify_owner != NULL &&
 761	    info->notify.sigev_notify == SIGEV_THREAD) {
 762		set_cookie(info->notify_cookie, NOTIFY_REMOVED);
 763		netlink_sendskb(info->notify_sock, info->notify_cookie);
 764	}
 765	put_pid(info->notify_owner);
 766	put_user_ns(info->notify_user_ns);
 767	info->notify_owner = NULL;
 768	info->notify_user_ns = NULL;
 769}
 770
 771static int prepare_open(struct dentry *dentry, int oflag, int ro,
 772			umode_t mode, struct filename *name,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 773			struct mq_attr *attr)
 774{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 775	static const int oflag2acc[O_ACCMODE] = { MAY_READ, MAY_WRITE,
 776						  MAY_READ | MAY_WRITE };
 777	int acc;
 778
 779	if (d_really_is_negative(dentry)) {
 780		if (!(oflag & O_CREAT))
 781			return -ENOENT;
 782		if (ro)
 783			return ro;
 784		audit_inode_parent_hidden(name, dentry->d_parent);
 785		return vfs_mkobj(dentry, mode & ~current_umask(),
 786				  mqueue_create_attr, attr);
 787	}
 788	/* it already existed */
 789	audit_inode(name, dentry, 0);
 790	if ((oflag & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
 791		return -EEXIST;
 792	if ((oflag & O_ACCMODE) == (O_RDWR | O_WRONLY))
 793		return -EINVAL;
 794	acc = oflag2acc[oflag & O_ACCMODE];
 795	return inode_permission(d_inode(dentry), acc);
 796}
 797
 798static int do_mq_open(const char __user *u_name, int oflag, umode_t mode,
 799		      struct mq_attr *attr)
 800{
 801	struct vfsmount *mnt = current->nsproxy->ipc_ns->mq_mnt;
 802	struct dentry *root = mnt->mnt_root;
 803	struct filename *name;
 804	struct path path;
 805	int fd, error;
 806	int ro;
 807
 808	audit_mq_open(oflag, mode, attr);
 
 
 
 809
 810	if (IS_ERR(name = getname(u_name)))
 811		return PTR_ERR(name);
 812
 813	fd = get_unused_fd_flags(O_CLOEXEC);
 814	if (fd < 0)
 815		goto out_putname;
 816
 817	ro = mnt_want_write(mnt);	/* we'll drop it in any case */
 818	inode_lock(d_inode(root));
 819	path.dentry = lookup_one_len(name->name, root, strlen(name->name));
 820	if (IS_ERR(path.dentry)) {
 821		error = PTR_ERR(path.dentry);
 822		goto out_putfd;
 823	}
 824	path.mnt = mntget(mnt);
 825	error = prepare_open(path.dentry, oflag, ro, mode, name, attr);
 826	if (!error) {
 827		struct file *file = dentry_open(&path, oflag, current_cred());
 828		if (!IS_ERR(file))
 829			fd_install(fd, file);
 830		else
 831			error = PTR_ERR(file);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 832	}
 833	path_put(&path);
 
 
 
 
 
 
 834out_putfd:
 835	if (error) {
 836		put_unused_fd(fd);
 837		fd = error;
 838	}
 839	inode_unlock(d_inode(root));
 840	if (!ro)
 841		mnt_drop_write(mnt);
 842out_putname:
 843	putname(name);
 844	return fd;
 845}
 846
 847SYSCALL_DEFINE4(mq_open, const char __user *, u_name, int, oflag, umode_t, mode,
 848		struct mq_attr __user *, u_attr)
 849{
 850	struct mq_attr attr;
 851	if (u_attr && copy_from_user(&attr, u_attr, sizeof(struct mq_attr)))
 852		return -EFAULT;
 853
 854	return do_mq_open(u_name, oflag, mode, u_attr ? &attr : NULL);
 855}
 856
 857SYSCALL_DEFINE1(mq_unlink, const char __user *, u_name)
 858{
 859	int err;
 860	struct filename *name;
 861	struct dentry *dentry;
 862	struct inode *inode = NULL;
 863	struct ipc_namespace *ipc_ns = current->nsproxy->ipc_ns;
 864	struct vfsmount *mnt = ipc_ns->mq_mnt;
 865
 866	name = getname(u_name);
 867	if (IS_ERR(name))
 868		return PTR_ERR(name);
 869
 870	audit_inode_parent_hidden(name, mnt->mnt_root);
 871	err = mnt_want_write(mnt);
 872	if (err)
 873		goto out_name;
 874	inode_lock_nested(d_inode(mnt->mnt_root), I_MUTEX_PARENT);
 875	dentry = lookup_one_len(name->name, mnt->mnt_root,
 876				strlen(name->name));
 877	if (IS_ERR(dentry)) {
 878		err = PTR_ERR(dentry);
 879		goto out_unlock;
 880	}
 881
 882	inode = d_inode(dentry);
 883	if (!inode) {
 884		err = -ENOENT;
 885	} else {
 
 
 
 
 886		ihold(inode);
 887		err = vfs_unlink(d_inode(dentry->d_parent), dentry, NULL);
 888	}
 
 
 
 
 889	dput(dentry);
 890
 891out_unlock:
 892	inode_unlock(d_inode(mnt->mnt_root));
 
 893	if (inode)
 894		iput(inode);
 895	mnt_drop_write(mnt);
 896out_name:
 897	putname(name);
 898
 899	return err;
 900}
 901
 902/* Pipelined send and receive functions.
 903 *
 904 * If a receiver finds no waiting message, then it registers itself in the
 905 * list of waiting receivers. A sender checks that list before adding the new
 906 * message into the message array. If there is a waiting receiver, then it
 907 * bypasses the message array and directly hands the message over to the
 908 * receiver. The receiver accepts the message and returns without grabbing the
 909 * queue spinlock:
 910 *
 911 * - Set pointer to message.
 912 * - Queue the receiver task for later wakeup (without the info->lock).
 913 * - Update its state to STATE_READY. Now the receiver can continue.
 914 * - Wake up the process after the lock is dropped. Should the process wake up
 915 *   before this wakeup (due to a timeout or a signal) it will either see
 916 *   STATE_READY and continue or acquire the lock to check the state again.
 917 *
 918 * The same algorithm is used for senders.
 919 */
 920
 921/* pipelined_send() - send a message directly to the task waiting in
 922 * sys_mq_timedreceive() (without inserting message into a queue).
 923 */
 924static inline void pipelined_send(struct wake_q_head *wake_q,
 925				  struct mqueue_inode_info *info,
 926				  struct msg_msg *message,
 927				  struct ext_wait_queue *receiver)
 928{
 929	receiver->msg = message;
 930	list_del(&receiver->list);
 931	wake_q_add(wake_q, receiver->task);
 932	/*
 933	 * Rely on the implicit cmpxchg barrier from wake_q_add such
 934	 * that we can ensure that updating receiver->state is the last
 935	 * write operation: As once set, the receiver can continue,
 936	 * and if we don't have the reference count from the wake_q,
 937	 * yet, at that point we can later have a use-after-free
 938	 * condition and bogus wakeup.
 939	 */
 940	receiver->state = STATE_READY;
 941}
 942
 943/* pipelined_receive() - if there is task waiting in sys_mq_timedsend()
 944 * gets its message and put to the queue (we have one free place for sure). */
 945static inline void pipelined_receive(struct wake_q_head *wake_q,
 946				     struct mqueue_inode_info *info)
 947{
 948	struct ext_wait_queue *sender = wq_get_first_waiter(info, SEND);
 949
 950	if (!sender) {
 951		/* for poll */
 952		wake_up_interruptible(&info->wait_q);
 953		return;
 954	}
 955	if (msg_insert(sender->msg, info))
 956		return;
 957
 958	list_del(&sender->list);
 959	wake_q_add(wake_q, sender->task);
 
 
 960	sender->state = STATE_READY;
 961}
 962
 963static int do_mq_timedsend(mqd_t mqdes, const char __user *u_msg_ptr,
 964		size_t msg_len, unsigned int msg_prio,
 965		struct timespec64 *ts)
 966{
 967	struct fd f;
 968	struct inode *inode;
 969	struct ext_wait_queue wait;
 970	struct ext_wait_queue *receiver;
 971	struct msg_msg *msg_ptr;
 972	struct mqueue_inode_info *info;
 973	ktime_t expires, *timeout = NULL;
 
 974	struct posix_msg_tree_node *new_leaf = NULL;
 975	int ret = 0;
 976	DEFINE_WAKE_Q(wake_q);
 
 
 
 
 
 
 977
 978	if (unlikely(msg_prio >= (unsigned long) MQ_PRIO_MAX))
 979		return -EINVAL;
 980
 981	if (ts) {
 982		expires = timespec64_to_ktime(*ts);
 983		timeout = &expires;
 984	}
 985
 986	audit_mq_sendrecv(mqdes, msg_len, msg_prio, ts);
 987
 988	f = fdget(mqdes);
 989	if (unlikely(!f.file)) {
 990		ret = -EBADF;
 991		goto out;
 992	}
 993
 994	inode = file_inode(f.file);
 995	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
 996		ret = -EBADF;
 997		goto out_fput;
 998	}
 999	info = MQUEUE_I(inode);
1000	audit_file(f.file);
1001
1002	if (unlikely(!(f.file->f_mode & FMODE_WRITE))) {
1003		ret = -EBADF;
1004		goto out_fput;
1005	}
1006
1007	if (unlikely(msg_len > info->attr.mq_msgsize)) {
1008		ret = -EMSGSIZE;
1009		goto out_fput;
1010	}
1011
1012	/* First try to allocate memory, before doing anything with
1013	 * existing queues. */
1014	msg_ptr = load_msg(u_msg_ptr, msg_len);
1015	if (IS_ERR(msg_ptr)) {
1016		ret = PTR_ERR(msg_ptr);
1017		goto out_fput;
1018	}
1019	msg_ptr->m_ts = msg_len;
1020	msg_ptr->m_type = msg_prio;
1021
1022	/*
1023	 * msg_insert really wants us to have a valid, spare node struct so
1024	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1025	 * fall back to that if necessary.
1026	 */
1027	if (!info->node_cache)
1028		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1029
1030	spin_lock(&info->lock);
1031
1032	if (!info->node_cache && new_leaf) {
1033		/* Save our speculative allocation into the cache */
 
1034		INIT_LIST_HEAD(&new_leaf->msg_list);
1035		info->node_cache = new_leaf;
 
1036		new_leaf = NULL;
1037	} else {
1038		kfree(new_leaf);
1039	}
1040
1041	if (info->attr.mq_curmsgs == info->attr.mq_maxmsg) {
1042		if (f.file->f_flags & O_NONBLOCK) {
1043			ret = -EAGAIN;
1044		} else {
1045			wait.task = current;
1046			wait.msg = (void *) msg_ptr;
1047			wait.state = STATE_NONE;
1048			ret = wq_sleep(info, SEND, timeout, &wait);
1049			/*
1050			 * wq_sleep must be called with info->lock held, and
1051			 * returns with the lock released
1052			 */
1053			goto out_free;
1054		}
1055	} else {
1056		receiver = wq_get_first_waiter(info, RECV);
1057		if (receiver) {
1058			pipelined_send(&wake_q, info, msg_ptr, receiver);
1059		} else {
1060			/* adds message to the queue */
1061			ret = msg_insert(msg_ptr, info);
1062			if (ret)
1063				goto out_unlock;
1064			__do_notify(info);
1065		}
1066		inode->i_atime = inode->i_mtime = inode->i_ctime =
1067				current_time(inode);
1068	}
1069out_unlock:
1070	spin_unlock(&info->lock);
1071	wake_up_q(&wake_q);
1072out_free:
1073	if (ret)
1074		free_msg(msg_ptr);
1075out_fput:
1076	fdput(f);
1077out:
1078	return ret;
1079}
1080
1081static int do_mq_timedreceive(mqd_t mqdes, char __user *u_msg_ptr,
1082		size_t msg_len, unsigned int __user *u_msg_prio,
1083		struct timespec64 *ts)
1084{
1085	ssize_t ret;
1086	struct msg_msg *msg_ptr;
1087	struct fd f;
1088	struct inode *inode;
1089	struct mqueue_inode_info *info;
1090	struct ext_wait_queue wait;
1091	ktime_t expires, *timeout = NULL;
 
1092	struct posix_msg_tree_node *new_leaf = NULL;
1093
1094	if (ts) {
1095		expires = timespec64_to_ktime(*ts);
 
 
1096		timeout = &expires;
1097	}
1098
1099	audit_mq_sendrecv(mqdes, msg_len, 0, ts);
1100
1101	f = fdget(mqdes);
1102	if (unlikely(!f.file)) {
1103		ret = -EBADF;
1104		goto out;
1105	}
1106
1107	inode = file_inode(f.file);
1108	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1109		ret = -EBADF;
1110		goto out_fput;
1111	}
1112	info = MQUEUE_I(inode);
1113	audit_file(f.file);
1114
1115	if (unlikely(!(f.file->f_mode & FMODE_READ))) {
1116		ret = -EBADF;
1117		goto out_fput;
1118	}
1119
1120	/* checks if buffer is big enough */
1121	if (unlikely(msg_len < info->attr.mq_msgsize)) {
1122		ret = -EMSGSIZE;
1123		goto out_fput;
1124	}
1125
1126	/*
1127	 * msg_insert really wants us to have a valid, spare node struct so
1128	 * it doesn't have to kmalloc a GFP_ATOMIC allocation, but it will
1129	 * fall back to that if necessary.
1130	 */
1131	if (!info->node_cache)
1132		new_leaf = kmalloc(sizeof(*new_leaf), GFP_KERNEL);
1133
1134	spin_lock(&info->lock);
1135
1136	if (!info->node_cache && new_leaf) {
1137		/* Save our speculative allocation into the cache */
 
1138		INIT_LIST_HEAD(&new_leaf->msg_list);
1139		info->node_cache = new_leaf;
 
1140	} else {
1141		kfree(new_leaf);
1142	}
1143
1144	if (info->attr.mq_curmsgs == 0) {
1145		if (f.file->f_flags & O_NONBLOCK) {
1146			spin_unlock(&info->lock);
1147			ret = -EAGAIN;
1148		} else {
1149			wait.task = current;
1150			wait.state = STATE_NONE;
1151			ret = wq_sleep(info, RECV, timeout, &wait);
1152			msg_ptr = wait.msg;
1153		}
1154	} else {
1155		DEFINE_WAKE_Q(wake_q);
1156
1157		msg_ptr = msg_get(info);
1158
1159		inode->i_atime = inode->i_mtime = inode->i_ctime =
1160				current_time(inode);
1161
1162		/* There is now free space in queue. */
1163		pipelined_receive(&wake_q, info);
1164		spin_unlock(&info->lock);
1165		wake_up_q(&wake_q);
1166		ret = 0;
1167	}
1168	if (ret == 0) {
1169		ret = msg_ptr->m_ts;
1170
1171		if ((u_msg_prio && put_user(msg_ptr->m_type, u_msg_prio)) ||
1172			store_msg(u_msg_ptr, msg_ptr, msg_ptr->m_ts)) {
1173			ret = -EFAULT;
1174		}
1175		free_msg(msg_ptr);
1176	}
1177out_fput:
1178	fdput(f);
1179out:
1180	return ret;
1181}
1182
1183SYSCALL_DEFINE5(mq_timedsend, mqd_t, mqdes, const char __user *, u_msg_ptr,
1184		size_t, msg_len, unsigned int, msg_prio,
1185		const struct __kernel_timespec __user *, u_abs_timeout)
1186{
1187	struct timespec64 ts, *p = NULL;
1188	if (u_abs_timeout) {
1189		int res = prepare_timeout(u_abs_timeout, &ts);
1190		if (res)
1191			return res;
1192		p = &ts;
1193	}
1194	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1195}
1196
1197SYSCALL_DEFINE5(mq_timedreceive, mqd_t, mqdes, char __user *, u_msg_ptr,
1198		size_t, msg_len, unsigned int __user *, u_msg_prio,
1199		const struct __kernel_timespec __user *, u_abs_timeout)
1200{
1201	struct timespec64 ts, *p = NULL;
1202	if (u_abs_timeout) {
1203		int res = prepare_timeout(u_abs_timeout, &ts);
1204		if (res)
1205			return res;
1206		p = &ts;
1207	}
1208	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1209}
1210
1211/*
1212 * Notes: the case when user wants us to deregister (with NULL as pointer)
1213 * and he isn't currently owner of notification, will be silently discarded.
1214 * It isn't explicitly defined in the POSIX.
1215 */
1216static int do_mq_notify(mqd_t mqdes, const struct sigevent *notification)
 
1217{
1218	int ret;
1219	struct fd f;
1220	struct sock *sock;
1221	struct inode *inode;
 
1222	struct mqueue_inode_info *info;
1223	struct sk_buff *nc;
1224
1225	audit_mq_notify(mqdes, notification);
 
 
 
 
 
 
1226
1227	nc = NULL;
1228	sock = NULL;
1229	if (notification != NULL) {
1230		if (unlikely(notification->sigev_notify != SIGEV_NONE &&
1231			     notification->sigev_notify != SIGEV_SIGNAL &&
1232			     notification->sigev_notify != SIGEV_THREAD))
1233			return -EINVAL;
1234		if (notification->sigev_notify == SIGEV_SIGNAL &&
1235			!valid_signal(notification->sigev_signo)) {
1236			return -EINVAL;
1237		}
1238		if (notification->sigev_notify == SIGEV_THREAD) {
1239			long timeo;
1240
1241			/* create the notify skb */
1242			nc = alloc_skb(NOTIFY_COOKIE_LEN, GFP_KERNEL);
1243			if (!nc)
1244				return -ENOMEM;
1245
 
1246			if (copy_from_user(nc->data,
1247					notification->sigev_value.sival_ptr,
1248					NOTIFY_COOKIE_LEN)) {
1249				ret = -EFAULT;
1250				goto free_skb;
1251			}
1252
1253			/* TODO: add a header? */
1254			skb_put(nc, NOTIFY_COOKIE_LEN);
1255			/* and attach it to the socket */
1256retry:
1257			f = fdget(notification->sigev_signo);
1258			if (!f.file) {
1259				ret = -EBADF;
1260				goto out;
1261			}
1262			sock = netlink_getsockbyfilp(f.file);
1263			fdput(f);
1264			if (IS_ERR(sock)) {
1265				ret = PTR_ERR(sock);
1266				goto free_skb;
 
1267			}
1268
1269			timeo = MAX_SCHEDULE_TIMEOUT;
1270			ret = netlink_attachskb(sock, nc, &timeo, NULL);
1271			if (ret == 1) {
 
 
1272				sock = NULL;
1273				goto retry;
 
1274			}
1275			if (ret)
1276				return ret;
1277		}
1278	}
1279
1280	f = fdget(mqdes);
1281	if (!f.file) {
1282		ret = -EBADF;
1283		goto out;
1284	}
1285
1286	inode = file_inode(f.file);
1287	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1288		ret = -EBADF;
1289		goto out_fput;
1290	}
1291	info = MQUEUE_I(inode);
1292
1293	ret = 0;
1294	spin_lock(&info->lock);
1295	if (notification == NULL) {
1296		if (info->notify_owner == task_tgid(current)) {
1297			remove_notification(info);
1298			inode->i_atime = inode->i_ctime = current_time(inode);
1299		}
1300	} else if (info->notify_owner != NULL) {
1301		ret = -EBUSY;
1302	} else {
1303		switch (notification->sigev_notify) {
1304		case SIGEV_NONE:
1305			info->notify.sigev_notify = SIGEV_NONE;
1306			break;
1307		case SIGEV_THREAD:
1308			info->notify_sock = sock;
1309			info->notify_cookie = nc;
1310			sock = NULL;
1311			nc = NULL;
1312			info->notify.sigev_notify = SIGEV_THREAD;
1313			break;
1314		case SIGEV_SIGNAL:
1315			info->notify.sigev_signo = notification->sigev_signo;
1316			info->notify.sigev_value = notification->sigev_value;
1317			info->notify.sigev_notify = SIGEV_SIGNAL;
1318			break;
1319		}
1320
1321		info->notify_owner = get_pid(task_tgid(current));
1322		info->notify_user_ns = get_user_ns(current_user_ns());
1323		inode->i_atime = inode->i_ctime = current_time(inode);
1324	}
1325	spin_unlock(&info->lock);
1326out_fput:
1327	fdput(f);
1328out:
1329	if (sock)
1330		netlink_detachskb(sock, nc);
1331	else
1332free_skb:
1333		dev_kfree_skb(nc);
1334
1335	return ret;
1336}
1337
1338SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1339		const struct sigevent __user *, u_notification)
1340{
1341	struct sigevent n, *p = NULL;
1342	if (u_notification) {
1343		if (copy_from_user(&n, u_notification, sizeof(struct sigevent)))
1344			return -EFAULT;
1345		p = &n;
1346	}
1347	return do_mq_notify(mqdes, p);
1348}
1349
1350static int do_mq_getsetattr(int mqdes, struct mq_attr *new, struct mq_attr *old)
1351{
1352	struct fd f;
1353	struct inode *inode;
1354	struct mqueue_inode_info *info;
1355
1356	if (new && (new->mq_flags & (~O_NONBLOCK)))
1357		return -EINVAL;
1358
1359	f = fdget(mqdes);
1360	if (!f.file)
1361		return -EBADF;
1362
1363	if (unlikely(f.file->f_op != &mqueue_file_operations)) {
1364		fdput(f);
1365		return -EBADF;
1366	}
1367
1368	inode = file_inode(f.file);
1369	info = MQUEUE_I(inode);
1370
1371	spin_lock(&info->lock);
1372
1373	if (old) {
1374		*old = info->attr;
1375		old->mq_flags = f.file->f_flags & O_NONBLOCK;
1376	}
1377	if (new) {
1378		audit_mq_getsetattr(mqdes, new);
1379		spin_lock(&f.file->f_lock);
1380		if (new->mq_flags & O_NONBLOCK)
1381			f.file->f_flags |= O_NONBLOCK;
1382		else
1383			f.file->f_flags &= ~O_NONBLOCK;
1384		spin_unlock(&f.file->f_lock);
1385
1386		inode->i_atime = inode->i_ctime = current_time(inode);
1387	}
1388
1389	spin_unlock(&info->lock);
1390	fdput(f);
1391	return 0;
1392}
1393
1394SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1395		const struct mq_attr __user *, u_mqstat,
1396		struct mq_attr __user *, u_omqstat)
1397{
1398	int ret;
1399	struct mq_attr mqstat, omqstat;
1400	struct mq_attr *new = NULL, *old = NULL;
 
 
1401
1402	if (u_mqstat) {
1403		new = &mqstat;
1404		if (copy_from_user(new, u_mqstat, sizeof(struct mq_attr)))
1405			return -EFAULT;
 
 
1406	}
1407	if (u_omqstat)
1408		old = &omqstat;
1409
1410	ret = do_mq_getsetattr(mqdes, new, old);
1411	if (ret || !old)
1412		return ret;
1413
1414	if (copy_to_user(u_omqstat, old, sizeof(struct mq_attr)))
1415		return -EFAULT;
1416	return 0;
1417}
1418
1419#ifdef CONFIG_COMPAT
1420
1421struct compat_mq_attr {
1422	compat_long_t mq_flags;      /* message queue flags		     */
1423	compat_long_t mq_maxmsg;     /* maximum number of messages	     */
1424	compat_long_t mq_msgsize;    /* maximum message size		     */
1425	compat_long_t mq_curmsgs;    /* number of messages currently queued  */
1426	compat_long_t __reserved[4]; /* ignored for input, zeroed for output */
1427};
1428
1429static inline int get_compat_mq_attr(struct mq_attr *attr,
1430			const struct compat_mq_attr __user *uattr)
1431{
1432	struct compat_mq_attr v;
1433
1434	if (copy_from_user(&v, uattr, sizeof(*uattr)))
1435		return -EFAULT;
1436
1437	memset(attr, 0, sizeof(*attr));
1438	attr->mq_flags = v.mq_flags;
1439	attr->mq_maxmsg = v.mq_maxmsg;
1440	attr->mq_msgsize = v.mq_msgsize;
1441	attr->mq_curmsgs = v.mq_curmsgs;
1442	return 0;
1443}
1444
1445static inline int put_compat_mq_attr(const struct mq_attr *attr,
1446			struct compat_mq_attr __user *uattr)
1447{
1448	struct compat_mq_attr v;
1449
1450	memset(&v, 0, sizeof(v));
1451	v.mq_flags = attr->mq_flags;
1452	v.mq_maxmsg = attr->mq_maxmsg;
1453	v.mq_msgsize = attr->mq_msgsize;
1454	v.mq_curmsgs = attr->mq_curmsgs;
1455	if (copy_to_user(uattr, &v, sizeof(*uattr)))
1456		return -EFAULT;
1457	return 0;
1458}
1459
1460COMPAT_SYSCALL_DEFINE4(mq_open, const char __user *, u_name,
1461		       int, oflag, compat_mode_t, mode,
1462		       struct compat_mq_attr __user *, u_attr)
1463{
1464	struct mq_attr attr, *p = NULL;
1465	if (u_attr && oflag & O_CREAT) {
1466		p = &attr;
1467		if (get_compat_mq_attr(&attr, u_attr))
1468			return -EFAULT;
1469	}
1470	return do_mq_open(u_name, oflag, mode, p);
1471}
1472
1473COMPAT_SYSCALL_DEFINE2(mq_notify, mqd_t, mqdes,
1474		       const struct compat_sigevent __user *, u_notification)
1475{
1476	struct sigevent n, *p = NULL;
1477	if (u_notification) {
1478		if (get_compat_sigevent(&n, u_notification))
1479			return -EFAULT;
1480		if (n.sigev_notify == SIGEV_THREAD)
1481			n.sigev_value.sival_ptr = compat_ptr(n.sigev_value.sival_int);
1482		p = &n;
1483	}
1484	return do_mq_notify(mqdes, p);
1485}
1486
1487COMPAT_SYSCALL_DEFINE3(mq_getsetattr, mqd_t, mqdes,
1488		       const struct compat_mq_attr __user *, u_mqstat,
1489		       struct compat_mq_attr __user *, u_omqstat)
1490{
1491	int ret;
1492	struct mq_attr mqstat, omqstat;
1493	struct mq_attr *new = NULL, *old = NULL;
1494
 
 
1495	if (u_mqstat) {
1496		new = &mqstat;
1497		if (get_compat_mq_attr(new, u_mqstat))
1498			return -EFAULT;
 
 
 
 
 
 
1499	}
1500	if (u_omqstat)
1501		old = &omqstat;
1502
1503	ret = do_mq_getsetattr(mqdes, new, old);
1504	if (ret || !old)
1505		return ret;
1506
1507	if (put_compat_mq_attr(old, u_omqstat))
1508		return -EFAULT;
1509	return 0;
1510}
1511#endif
1512
1513#ifdef CONFIG_COMPAT_32BIT_TIME
1514static int compat_prepare_timeout(const struct old_timespec32 __user *p,
1515				   struct timespec64 *ts)
1516{
1517	if (get_old_timespec32(ts, p))
1518		return -EFAULT;
1519	if (!timespec64_valid(ts))
1520		return -EINVAL;
1521	return 0;
1522}
1523
1524SYSCALL_DEFINE5(mq_timedsend_time32, mqd_t, mqdes,
1525		const char __user *, u_msg_ptr,
1526		unsigned int, msg_len, unsigned int, msg_prio,
1527		const struct old_timespec32 __user *, u_abs_timeout)
1528{
1529	struct timespec64 ts, *p = NULL;
1530	if (u_abs_timeout) {
1531		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1532		if (res)
1533			return res;
1534		p = &ts;
1535	}
1536	return do_mq_timedsend(mqdes, u_msg_ptr, msg_len, msg_prio, p);
1537}
1538
1539SYSCALL_DEFINE5(mq_timedreceive_time32, mqd_t, mqdes,
1540		char __user *, u_msg_ptr,
1541		unsigned int, msg_len, unsigned int __user *, u_msg_prio,
1542		const struct old_timespec32 __user *, u_abs_timeout)
1543{
1544	struct timespec64 ts, *p = NULL;
1545	if (u_abs_timeout) {
1546		int res = compat_prepare_timeout(u_abs_timeout, &ts);
1547		if (res)
1548			return res;
1549		p = &ts;
1550	}
1551	return do_mq_timedreceive(mqdes, u_msg_ptr, msg_len, u_msg_prio, p);
1552}
1553#endif
1554
1555static const struct inode_operations mqueue_dir_inode_operations = {
1556	.lookup = simple_lookup,
1557	.create = mqueue_create,
1558	.unlink = mqueue_unlink,
1559};
1560
1561static const struct file_operations mqueue_file_operations = {
1562	.flush = mqueue_flush_file,
1563	.poll = mqueue_poll_file,
1564	.read = mqueue_read_file,
1565	.llseek = default_llseek,
1566};
1567
1568static const struct super_operations mqueue_super_ops = {
1569	.alloc_inode = mqueue_alloc_inode,
1570	.free_inode = mqueue_free_inode,
1571	.evict_inode = mqueue_evict_inode,
1572	.statfs = simple_statfs,
1573};
1574
1575static const struct fs_context_operations mqueue_fs_context_ops = {
1576	.free		= mqueue_fs_context_free,
1577	.get_tree	= mqueue_get_tree,
1578};
1579
1580static struct file_system_type mqueue_fs_type = {
1581	.name			= "mqueue",
1582	.init_fs_context	= mqueue_init_fs_context,
1583	.kill_sb		= kill_litter_super,
1584	.fs_flags		= FS_USERNS_MOUNT,
1585};
1586
1587int mq_init_ns(struct ipc_namespace *ns)
1588{
1589	struct vfsmount *m;
1590
1591	ns->mq_queues_count  = 0;
1592	ns->mq_queues_max    = DFLT_QUEUESMAX;
1593	ns->mq_msg_max       = DFLT_MSGMAX;
1594	ns->mq_msgsize_max   = DFLT_MSGSIZEMAX;
1595	ns->mq_msg_default   = DFLT_MSG;
1596	ns->mq_msgsize_default  = DFLT_MSGSIZE;
1597
1598	m = mq_create_mount(ns);
1599	if (IS_ERR(m))
1600		return PTR_ERR(m);
1601	ns->mq_mnt = m;
 
 
1602	return 0;
1603}
1604
1605void mq_clear_sbinfo(struct ipc_namespace *ns)
1606{
1607	ns->mq_mnt->mnt_sb->s_fs_info = NULL;
1608}
1609
1610void mq_put_mnt(struct ipc_namespace *ns)
1611{
1612	kern_unmount(ns->mq_mnt);
1613}
1614
1615static int __init init_mqueue_fs(void)
1616{
1617	int error;
1618
1619	mqueue_inode_cachep = kmem_cache_create("mqueue_inode_cache",
1620				sizeof(struct mqueue_inode_info), 0,
1621				SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT, init_once);
1622	if (mqueue_inode_cachep == NULL)
1623		return -ENOMEM;
1624
1625	/* ignore failures - they are not fatal */
1626	mq_sysctl_table = mq_register_sysctl_table();
1627
1628	error = register_filesystem(&mqueue_fs_type);
1629	if (error)
1630		goto out_sysctl;
1631
1632	spin_lock_init(&mq_lock);
1633
1634	error = mq_init_ns(&init_ipc_ns);
1635	if (error)
1636		goto out_filesystem;
1637
1638	return 0;
1639
1640out_filesystem:
1641	unregister_filesystem(&mqueue_fs_type);
1642out_sysctl:
1643	if (mq_sysctl_table)
1644		unregister_sysctl_table(mq_sysctl_table);
1645	kmem_cache_destroy(mqueue_inode_cachep);
1646	return error;
1647}
1648
1649device_initcall(init_mqueue_fs);