Linux Audio

Check our new training course

Loading...
v6.13.7
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * This file contains the procedures for the handling of select and poll
   4 *
   5 * Created for Linux based loosely upon Mathius Lattner's minix
   6 * patches by Peter MacDonald. Heavily edited by Linus.
   7 *
   8 *  4 February 1994
   9 *     COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS
  10 *     flag set in its personality we do *not* modify the given timeout
  11 *     parameter to reflect time remaining.
  12 *
  13 *  24 January 2000
  14 *     Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation 
  15 *     of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
  16 */
  17
  18#include <linux/compat.h>
  19#include <linux/kernel.h>
  20#include <linux/sched/signal.h>
  21#include <linux/sched/rt.h>
  22#include <linux/syscalls.h>
  23#include <linux/export.h>
  24#include <linux/slab.h>
  25#include <linux/poll.h>
  26#include <linux/personality.h> /* for STICKY_TIMEOUTS */
  27#include <linux/file.h>
  28#include <linux/fdtable.h>
  29#include <linux/fs.h>
  30#include <linux/rcupdate.h>
  31#include <linux/hrtimer.h>
  32#include <linux/freezer.h>
  33#include <net/busy_poll.h>
  34#include <linux/vmalloc.h>
  35
  36#include <linux/uaccess.h>
  37
  38
  39/*
  40 * Estimate expected accuracy in ns from a timeval.
  41 *
  42 * After quite a bit of churning around, we've settled on
  43 * a simple thing of taking 0.1% of the timeout as the
  44 * slack, with a cap of 100 msec.
  45 * "nice" tasks get a 0.5% slack instead.
  46 *
  47 * Consider this comment an open invitation to come up with even
  48 * better solutions..
  49 */
  50
  51#define MAX_SLACK	(100 * NSEC_PER_MSEC)
  52
  53static long __estimate_accuracy(struct timespec64 *tv)
  54{
  55	long slack;
  56	int divfactor = 1000;
  57
  58	if (tv->tv_sec < 0)
  59		return 0;
  60
  61	if (task_nice(current) > 0)
  62		divfactor = divfactor / 5;
  63
  64	if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
  65		return MAX_SLACK;
  66
  67	slack = tv->tv_nsec / divfactor;
  68	slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
  69
  70	if (slack > MAX_SLACK)
  71		return MAX_SLACK;
  72
  73	return slack;
  74}
  75
  76u64 select_estimate_accuracy(struct timespec64 *tv)
  77{
  78	u64 ret;
  79	struct timespec64 now;
  80	u64 slack = current->timer_slack_ns;
  81
  82	if (slack == 0)
 
 
 
 
  83		return 0;
  84
  85	ktime_get_ts64(&now);
  86	now = timespec64_sub(*tv, now);
  87	ret = __estimate_accuracy(&now);
  88	if (ret < slack)
  89		return slack;
  90	return ret;
  91}
  92
  93
  94
  95struct poll_table_page {
  96	struct poll_table_page * next;
  97	struct poll_table_entry * entry;
  98	struct poll_table_entry entries[];
  99};
 100
 101#define POLL_TABLE_FULL(table) \
 102	((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
 103
 104/*
 105 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
 106 * I have rewritten this, taking some shortcuts: This code may not be easy to
 107 * follow, but it should be free of race-conditions, and it's practical. If you
 108 * understand what I'm doing here, then you understand how the linux
 109 * sleep/wakeup mechanism works.
 110 *
 111 * Two very simple procedures, poll_wait() and poll_freewait() make all the
 112 * work.  poll_wait() is an inline-function defined in <linux/poll.h>,
 113 * as all select/poll functions have to call it to add an entry to the
 114 * poll table.
 115 */
 116static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
 117		       poll_table *p);
 118
 119void poll_initwait(struct poll_wqueues *pwq)
 120{
 121	init_poll_funcptr(&pwq->pt, __pollwait);
 122	pwq->polling_task = current;
 123	pwq->triggered = 0;
 124	pwq->error = 0;
 125	pwq->table = NULL;
 126	pwq->inline_index = 0;
 127}
 128EXPORT_SYMBOL(poll_initwait);
 129
 130static void free_poll_entry(struct poll_table_entry *entry)
 131{
 132	remove_wait_queue(entry->wait_address, &entry->wait);
 133	fput(entry->filp);
 134}
 135
 136void poll_freewait(struct poll_wqueues *pwq)
 137{
 138	struct poll_table_page * p = pwq->table;
 139	int i;
 140	for (i = 0; i < pwq->inline_index; i++)
 141		free_poll_entry(pwq->inline_entries + i);
 142	while (p) {
 143		struct poll_table_entry * entry;
 144		struct poll_table_page *old;
 145
 146		entry = p->entry;
 147		do {
 148			entry--;
 149			free_poll_entry(entry);
 150		} while (entry > p->entries);
 151		old = p;
 152		p = p->next;
 153		free_page((unsigned long) old);
 154	}
 155}
 156EXPORT_SYMBOL(poll_freewait);
 157
 158static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
 159{
 160	struct poll_table_page *table = p->table;
 161
 162	if (p->inline_index < N_INLINE_POLL_ENTRIES)
 163		return p->inline_entries + p->inline_index++;
 164
 165	if (!table || POLL_TABLE_FULL(table)) {
 166		struct poll_table_page *new_table;
 167
 168		new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
 169		if (!new_table) {
 170			p->error = -ENOMEM;
 171			return NULL;
 172		}
 173		new_table->entry = new_table->entries;
 174		new_table->next = table;
 175		p->table = new_table;
 176		table = new_table;
 177	}
 178
 179	return table->entry++;
 180}
 181
 182static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
 183{
 184	struct poll_wqueues *pwq = wait->private;
 185	DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
 186
 187	/*
 188	 * Although this function is called under waitqueue lock, LOCK
 189	 * doesn't imply write barrier and the users expect write
 190	 * barrier semantics on wakeup functions.  The following
 191	 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
 192	 * and is paired with smp_store_mb() in poll_schedule_timeout.
 193	 */
 194	smp_wmb();
 195	pwq->triggered = 1;
 196
 197	/*
 198	 * Perform the default wake up operation using a dummy
 199	 * waitqueue.
 200	 *
 201	 * TODO: This is hacky but there currently is no interface to
 202	 * pass in @sync.  @sync is scheduled to be removed and once
 203	 * that happens, wake_up_process() can be used directly.
 204	 */
 205	return default_wake_function(&dummy_wait, mode, sync, key);
 206}
 207
 208static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
 209{
 210	struct poll_table_entry *entry;
 211
 212	entry = container_of(wait, struct poll_table_entry, wait);
 213	if (key && !(key_to_poll(key) & entry->key))
 214		return 0;
 215	return __pollwake(wait, mode, sync, key);
 216}
 217
 218/* Add a new entry */
 219static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
 220				poll_table *p)
 221{
 222	struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
 223	struct poll_table_entry *entry = poll_get_entry(pwq);
 224	if (!entry)
 225		return;
 226	entry->filp = get_file(filp);
 227	entry->wait_address = wait_address;
 228	entry->key = p->_key;
 229	init_waitqueue_func_entry(&entry->wait, pollwake);
 230	entry->wait.private = pwq;
 231	add_wait_queue(wait_address, &entry->wait);
 232}
 233
 234static int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
 235			  ktime_t *expires, unsigned long slack)
 236{
 237	int rc = -EINTR;
 238
 239	set_current_state(state);
 240	if (!pwq->triggered)
 241		rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
 242	__set_current_state(TASK_RUNNING);
 243
 244	/*
 245	 * Prepare for the next iteration.
 246	 *
 247	 * The following smp_store_mb() serves two purposes.  First, it's
 248	 * the counterpart rmb of the wmb in pollwake() such that data
 249	 * written before wake up is always visible after wake up.
 250	 * Second, the full barrier guarantees that triggered clearing
 251	 * doesn't pass event check of the next iteration.  Note that
 252	 * this problem doesn't exist for the first iteration as
 253	 * add_wait_queue() has full barrier semantics.
 254	 */
 255	smp_store_mb(pwq->triggered, 0);
 256
 257	return rc;
 258}
 259
 260/**
 261 * poll_select_set_timeout - helper function to setup the timeout value
 262 * @to:		pointer to timespec64 variable for the final timeout
 263 * @sec:	seconds (from user space)
 264 * @nsec:	nanoseconds (from user space)
 265 *
 266 * Note, we do not use a timespec for the user space value here, That
 267 * way we can use the function for timeval and compat interfaces as well.
 268 *
 269 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
 270 */
 271int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
 272{
 273	struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
 274
 275	if (!timespec64_valid(&ts))
 276		return -EINVAL;
 277
 278	/* Optimize for the zero timeout value here */
 279	if (!sec && !nsec) {
 280		to->tv_sec = to->tv_nsec = 0;
 281	} else {
 282		ktime_get_ts64(to);
 283		*to = timespec64_add_safe(*to, ts);
 284	}
 285	return 0;
 286}
 287
 288enum poll_time_type {
 289	PT_TIMEVAL = 0,
 290	PT_OLD_TIMEVAL = 1,
 291	PT_TIMESPEC = 2,
 292	PT_OLD_TIMESPEC = 3,
 293};
 294
 295static int poll_select_finish(struct timespec64 *end_time,
 296			      void __user *p,
 297			      enum poll_time_type pt_type, int ret)
 298{
 299	struct timespec64 rts;
 300
 301	restore_saved_sigmask_unless(ret == -ERESTARTNOHAND);
 302
 303	if (!p)
 304		return ret;
 305
 306	if (current->personality & STICKY_TIMEOUTS)
 307		goto sticky;
 308
 309	/* No update for zero timeout */
 310	if (!end_time->tv_sec && !end_time->tv_nsec)
 311		return ret;
 312
 313	ktime_get_ts64(&rts);
 314	rts = timespec64_sub(*end_time, rts);
 315	if (rts.tv_sec < 0)
 316		rts.tv_sec = rts.tv_nsec = 0;
 317
 318
 319	switch (pt_type) {
 320	case PT_TIMEVAL:
 321		{
 322			struct __kernel_old_timeval rtv;
 323
 324			if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
 325				memset(&rtv, 0, sizeof(rtv));
 326			rtv.tv_sec = rts.tv_sec;
 327			rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
 328			if (!copy_to_user(p, &rtv, sizeof(rtv)))
 329				return ret;
 330		}
 331		break;
 332	case PT_OLD_TIMEVAL:
 333		{
 334			struct old_timeval32 rtv;
 335
 336			rtv.tv_sec = rts.tv_sec;
 337			rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
 338			if (!copy_to_user(p, &rtv, sizeof(rtv)))
 339				return ret;
 340		}
 341		break;
 342	case PT_TIMESPEC:
 343		if (!put_timespec64(&rts, p))
 344			return ret;
 345		break;
 346	case PT_OLD_TIMESPEC:
 347		if (!put_old_timespec32(&rts, p))
 348			return ret;
 349		break;
 350	default:
 351		BUG();
 352	}
 353	/*
 354	 * If an application puts its timeval in read-only memory, we
 355	 * don't want the Linux-specific update to the timeval to
 356	 * cause a fault after the select has completed
 357	 * successfully. However, because we're not updating the
 358	 * timeval, we can't restart the system call.
 359	 */
 360
 361sticky:
 362	if (ret == -ERESTARTNOHAND)
 363		ret = -EINTR;
 364	return ret;
 365}
 366
 367/*
 368 * Scalable version of the fd_set.
 369 */
 370
 371typedef struct {
 372	unsigned long *in, *out, *ex;
 373	unsigned long *res_in, *res_out, *res_ex;
 374} fd_set_bits;
 375
 376/*
 377 * How many longwords for "nr" bits?
 378 */
 379#define FDS_BITPERLONG	(8*sizeof(long))
 380#define FDS_LONGS(nr)	(((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
 381#define FDS_BYTES(nr)	(FDS_LONGS(nr)*sizeof(long))
 382
 383/*
 384 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
 385 */
 386static inline
 387int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
 388{
 389	nr = FDS_BYTES(nr);
 390	if (ufdset)
 391		return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0;
 392
 393	memset(fdset, 0, nr);
 394	return 0;
 395}
 396
 397static inline unsigned long __must_check
 398set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
 399{
 400	if (ufdset)
 401		return __copy_to_user(ufdset, fdset, FDS_BYTES(nr));
 402	return 0;
 403}
 404
 405static inline
 406void zero_fd_set(unsigned long nr, unsigned long *fdset)
 407{
 408	memset(fdset, 0, FDS_BYTES(nr));
 409}
 410
 411#define FDS_IN(fds, n)		(fds->in + n)
 412#define FDS_OUT(fds, n)		(fds->out + n)
 413#define FDS_EX(fds, n)		(fds->ex + n)
 414
 415#define BITS(fds, n)	(*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
 416
 417static int max_select_fd(unsigned long n, fd_set_bits *fds)
 418{
 419	unsigned long *open_fds;
 420	unsigned long set;
 421	int max;
 422	struct fdtable *fdt;
 423
 424	/* handle last in-complete long-word first */
 425	set = ~(~0UL << (n & (BITS_PER_LONG-1)));
 426	n /= BITS_PER_LONG;
 427	fdt = files_fdtable(current->files);
 428	open_fds = fdt->open_fds + n;
 429	max = 0;
 430	if (set) {
 431		set &= BITS(fds, n);
 432		if (set) {
 433			if (!(set & ~*open_fds))
 434				goto get_max;
 435			return -EBADF;
 436		}
 437	}
 438	while (n) {
 439		open_fds--;
 440		n--;
 441		set = BITS(fds, n);
 442		if (!set)
 443			continue;
 444		if (set & ~*open_fds)
 445			return -EBADF;
 446		if (max)
 447			continue;
 448get_max:
 449		do {
 450			max++;
 451			set >>= 1;
 452		} while (set);
 453		max += n * BITS_PER_LONG;
 454	}
 455
 456	return max;
 457}
 458
 459#define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR |\
 460			EPOLLNVAL)
 461#define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR |\
 462			 EPOLLNVAL)
 463#define POLLEX_SET (EPOLLPRI | EPOLLNVAL)
 464
 465static inline __poll_t select_poll_one(int fd, poll_table *wait, unsigned long in,
 466				unsigned long out, unsigned long bit,
 467				__poll_t ll_flag)
 468{
 469	CLASS(fd, f)(fd);
 470
 471	if (fd_empty(f))
 472		return EPOLLNVAL;
 473
 474	wait->_key = POLLEX_SET | ll_flag;
 475	if (in & bit)
 476		wait->_key |= POLLIN_SET;
 477	if (out & bit)
 478		wait->_key |= POLLOUT_SET;
 479
 480	return vfs_poll(fd_file(f), wait);
 481}
 482
 483static noinline_for_stack int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
 484{
 485	ktime_t expire, *to = NULL;
 486	struct poll_wqueues table;
 487	poll_table *wait;
 488	int retval, i, timed_out = 0;
 489	u64 slack = 0;
 490	__poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
 491	unsigned long busy_start = 0;
 492
 493	rcu_read_lock();
 494	retval = max_select_fd(n, fds);
 495	rcu_read_unlock();
 496
 497	if (retval < 0)
 498		return retval;
 499	n = retval;
 500
 501	poll_initwait(&table);
 502	wait = &table.pt;
 503	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
 504		wait->_qproc = NULL;
 505		timed_out = 1;
 506	}
 507
 508	if (end_time && !timed_out)
 509		slack = select_estimate_accuracy(end_time);
 510
 511	retval = 0;
 512	for (;;) {
 513		unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
 514		bool can_busy_loop = false;
 515
 516		inp = fds->in; outp = fds->out; exp = fds->ex;
 517		rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
 518
 519		for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
 520			unsigned long in, out, ex, all_bits, bit = 1, j;
 521			unsigned long res_in = 0, res_out = 0, res_ex = 0;
 522			__poll_t mask;
 523
 524			in = *inp++; out = *outp++; ex = *exp++;
 525			all_bits = in | out | ex;
 526			if (all_bits == 0) {
 527				i += BITS_PER_LONG;
 528				continue;
 529			}
 530
 531			for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
 
 532				if (i >= n)
 533					break;
 534				if (!(bit & all_bits))
 535					continue;
 536				mask = select_poll_one(i, wait, in, out, bit,
 537						       busy_flag);
 538				if ((mask & POLLIN_SET) && (in & bit)) {
 539					res_in |= bit;
 540					retval++;
 541					wait->_qproc = NULL;
 542				}
 543				if ((mask & POLLOUT_SET) && (out & bit)) {
 544					res_out |= bit;
 545					retval++;
 546					wait->_qproc = NULL;
 547				}
 548				if ((mask & POLLEX_SET) && (ex & bit)) {
 549					res_ex |= bit;
 550					retval++;
 551					wait->_qproc = NULL;
 552				}
 553				/* got something, stop busy polling */
 554				if (retval) {
 555					can_busy_loop = false;
 556					busy_flag = 0;
 557
 558				/*
 559				 * only remember a returned
 560				 * POLL_BUSY_LOOP if we asked for it
 561				 */
 562				} else if (busy_flag & mask)
 563					can_busy_loop = true;
 
 
 
 
 
 564
 
 565			}
 566			if (res_in)
 567				*rinp = res_in;
 568			if (res_out)
 569				*routp = res_out;
 570			if (res_ex)
 571				*rexp = res_ex;
 572			cond_resched();
 573		}
 574		wait->_qproc = NULL;
 575		if (retval || timed_out || signal_pending(current))
 576			break;
 577		if (table.error) {
 578			retval = table.error;
 579			break;
 580		}
 581
 582		/* only if found POLL_BUSY_LOOP sockets && not out of time */
 583		if (can_busy_loop && !need_resched()) {
 584			if (!busy_start) {
 585				busy_start = busy_loop_current_time();
 586				continue;
 587			}
 588			if (!busy_loop_timeout(busy_start))
 589				continue;
 590		}
 591		busy_flag = 0;
 592
 593		/*
 594		 * If this is the first loop and we have a timeout
 595		 * given, then we convert to ktime_t and set the to
 596		 * pointer to the expiry value.
 597		 */
 598		if (end_time && !to) {
 599			expire = timespec64_to_ktime(*end_time);
 600			to = &expire;
 601		}
 602
 603		if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
 604					   to, slack))
 605			timed_out = 1;
 606	}
 607
 608	poll_freewait(&table);
 609
 610	return retval;
 611}
 612
 613/*
 614 * We can actually return ERESTARTSYS instead of EINTR, but I'd
 615 * like to be certain this leads to no problems. So I return
 616 * EINTR just for safety.
 617 *
 618 * Update: ERESTARTSYS breaks at least the xview clock binary, so
 619 * I'm trying ERESTARTNOHAND which restart only when you want to.
 620 */
 621int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
 622			   fd_set __user *exp, struct timespec64 *end_time)
 623{
 624	fd_set_bits fds;
 625	void *bits;
 626	int ret, max_fds;
 627	size_t size, alloc_size;
 628	struct fdtable *fdt;
 629	/* Allocate small arguments on the stack to save memory and be faster */
 630	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
 631
 632	ret = -EINVAL;
 633	if (n < 0)
 634		goto out_nofds;
 635
 636	/* max_fds can increase, so grab it once to avoid race */
 637	rcu_read_lock();
 638	fdt = files_fdtable(current->files);
 639	max_fds = fdt->max_fds;
 640	rcu_read_unlock();
 641	if (n > max_fds)
 642		n = max_fds;
 643
 644	/*
 645	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
 646	 * since we used fdset we need to allocate memory in units of
 647	 * long-words. 
 648	 */
 649	size = FDS_BYTES(n);
 650	bits = stack_fds;
 651	if (size > sizeof(stack_fds) / 6) {
 652		/* Not enough space in on-stack array; must use kmalloc */
 653		ret = -ENOMEM;
 654		if (size > (SIZE_MAX / 6))
 655			goto out_nofds;
 656
 657		alloc_size = 6 * size;
 658		bits = kvmalloc(alloc_size, GFP_KERNEL);
 659		if (!bits)
 660			goto out_nofds;
 661	}
 662	fds.in      = bits;
 663	fds.out     = bits +   size;
 664	fds.ex      = bits + 2*size;
 665	fds.res_in  = bits + 3*size;
 666	fds.res_out = bits + 4*size;
 667	fds.res_ex  = bits + 5*size;
 668
 669	if ((ret = get_fd_set(n, inp, fds.in)) ||
 670	    (ret = get_fd_set(n, outp, fds.out)) ||
 671	    (ret = get_fd_set(n, exp, fds.ex)))
 672		goto out;
 673	zero_fd_set(n, fds.res_in);
 674	zero_fd_set(n, fds.res_out);
 675	zero_fd_set(n, fds.res_ex);
 676
 677	ret = do_select(n, &fds, end_time);
 678
 679	if (ret < 0)
 680		goto out;
 681	if (!ret) {
 682		ret = -ERESTARTNOHAND;
 683		if (signal_pending(current))
 684			goto out;
 685		ret = 0;
 686	}
 687
 688	if (set_fd_set(n, inp, fds.res_in) ||
 689	    set_fd_set(n, outp, fds.res_out) ||
 690	    set_fd_set(n, exp, fds.res_ex))
 691		ret = -EFAULT;
 692
 693out:
 694	if (bits != stack_fds)
 695		kvfree(bits);
 696out_nofds:
 697	return ret;
 698}
 699
 700static int kern_select(int n, fd_set __user *inp, fd_set __user *outp,
 701		       fd_set __user *exp, struct __kernel_old_timeval __user *tvp)
 702{
 703	struct timespec64 end_time, *to = NULL;
 704	struct __kernel_old_timeval tv;
 705	int ret;
 706
 707	if (tvp) {
 708		if (copy_from_user(&tv, tvp, sizeof(tv)))
 709			return -EFAULT;
 710
 711		to = &end_time;
 712		if (poll_select_set_timeout(to,
 713				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
 714				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
 715			return -EINVAL;
 716	}
 717
 718	ret = core_sys_select(n, inp, outp, exp, to);
 719	return poll_select_finish(&end_time, tvp, PT_TIMEVAL, ret);
 720}
 721
 722SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
 723		fd_set __user *, exp, struct __kernel_old_timeval __user *, tvp)
 724{
 725	return kern_select(n, inp, outp, exp, tvp);
 726}
 727
 728static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
 729		       fd_set __user *exp, void __user *tsp,
 730		       const sigset_t __user *sigmask, size_t sigsetsize,
 731		       enum poll_time_type type)
 732{
 733	struct timespec64 ts, end_time, *to = NULL;
 734	int ret;
 735
 736	if (tsp) {
 737		switch (type) {
 738		case PT_TIMESPEC:
 739			if (get_timespec64(&ts, tsp))
 740				return -EFAULT;
 741			break;
 742		case PT_OLD_TIMESPEC:
 743			if (get_old_timespec32(&ts, tsp))
 744				return -EFAULT;
 745			break;
 746		default:
 747			BUG();
 748		}
 749
 750		to = &end_time;
 751		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
 752			return -EINVAL;
 753	}
 754
 755	ret = set_user_sigmask(sigmask, sigsetsize);
 756	if (ret)
 757		return ret;
 758
 759	ret = core_sys_select(n, inp, outp, exp, to);
 760	return poll_select_finish(&end_time, tsp, type, ret);
 761}
 762
 763/*
 764 * Most architectures can't handle 7-argument syscalls. So we provide a
 765 * 6-argument version where the sixth argument is a pointer to a structure
 766 * which has a pointer to the sigset_t itself followed by a size_t containing
 767 * the sigset size.
 768 */
 769struct sigset_argpack {
 770	sigset_t __user *p;
 771	size_t size;
 772};
 773
 774static inline int get_sigset_argpack(struct sigset_argpack *to,
 775				     struct sigset_argpack __user *from)
 776{
 777	// the path is hot enough for overhead of copy_from_user() to matter
 778	if (from) {
 779		if (can_do_masked_user_access())
 780			from = masked_user_access_begin(from);
 781		else if (!user_read_access_begin(from, sizeof(*from)))
 782			return -EFAULT;
 783		unsafe_get_user(to->p, &from->p, Efault);
 784		unsafe_get_user(to->size, &from->size, Efault);
 785		user_read_access_end();
 786	}
 787	return 0;
 788Efault:
 789	user_read_access_end();
 790	return -EFAULT;
 791}
 792
 793SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
 794		fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
 795		void __user *, sig)
 796{
 797	struct sigset_argpack x = {NULL, 0};
 798
 799	if (get_sigset_argpack(&x, sig))
 800		return -EFAULT;
 801
 802	return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_TIMESPEC);
 803}
 804
 805#if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
 806
 807SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
 808		fd_set __user *, exp, struct old_timespec32 __user *, tsp,
 809		void __user *, sig)
 810{
 811	struct sigset_argpack x = {NULL, 0};
 812
 813	if (get_sigset_argpack(&x, sig))
 814		return -EFAULT;
 815
 816	return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC);
 817}
 818
 819#endif
 820
 821#ifdef __ARCH_WANT_SYS_OLD_SELECT
 822struct sel_arg_struct {
 823	unsigned long n;
 824	fd_set __user *inp, *outp, *exp;
 825	struct __kernel_old_timeval __user *tvp;
 826};
 827
 828SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
 829{
 830	struct sel_arg_struct a;
 831
 832	if (copy_from_user(&a, arg, sizeof(a)))
 833		return -EFAULT;
 834	return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
 835}
 836#endif
 837
 838struct poll_list {
 839	struct poll_list *next;
 840	unsigned int len;
 841	struct pollfd entries[] __counted_by(len);
 842};
 843
 844#define POLLFD_PER_PAGE  ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
 845
 846/*
 847 * Fish for pollable events on the pollfd->fd file descriptor. We're only
 848 * interested in events matching the pollfd->events mask, and the result
 849 * matching that mask is both recorded in pollfd->revents and returned. The
 850 * pwait poll_table will be used by the fd-provided poll handler for waiting,
 851 * if pwait->_qproc is non-NULL.
 852 */
 853static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
 854				     bool *can_busy_poll,
 855				     __poll_t busy_flag)
 856{
 857	int fd = pollfd->fd;
 858	__poll_t mask, filter;
 
 859
 860	if (fd < 0)
 861		return 0;
 862
 863	CLASS(fd, f)(fd);
 864	if (fd_empty(f))
 865		return EPOLLNVAL;
 866
 867	/* userland u16 ->events contains POLL... bitmap */
 868	filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
 869	pwait->_key = filter | busy_flag;
 870	mask = vfs_poll(fd_file(f), pwait);
 871	if (mask & busy_flag)
 872		*can_busy_poll = true;
 873	return mask & filter;		/* Mask out unneeded events. */
 
 
 
 
 
 
 874}
 875
 876static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
 877		   struct timespec64 *end_time)
 878{
 879	poll_table* pt = &wait->pt;
 880	ktime_t expire, *to = NULL;
 881	int timed_out = 0, count = 0;
 882	u64 slack = 0;
 883	__poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
 884	unsigned long busy_start = 0;
 885
 886	/* Optimise the no-wait case */
 887	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
 888		pt->_qproc = NULL;
 889		timed_out = 1;
 890	}
 891
 892	if (end_time && !timed_out)
 893		slack = select_estimate_accuracy(end_time);
 894
 895	for (;;) {
 896		struct poll_list *walk;
 897		bool can_busy_loop = false;
 898
 899		for (walk = list; walk != NULL; walk = walk->next) {
 900			struct pollfd * pfd, * pfd_end;
 901
 902			pfd = walk->entries;
 903			pfd_end = pfd + walk->len;
 904			for (; pfd != pfd_end; pfd++) {
 905				__poll_t mask;
 906				/*
 907				 * Fish for events. If we found one, record it
 908				 * and kill poll_table->_qproc, so we don't
 909				 * needlessly register any other waiters after
 910				 * this. They'll get immediately deregistered
 911				 * when we break out and return.
 912				 */
 913				mask = do_pollfd(pfd, pt, &can_busy_loop, busy_flag);
 914				pfd->revents = mangle_poll(mask);
 915				if (mask) {
 916					count++;
 917					pt->_qproc = NULL;
 918					/* found something, stop busy polling */
 919					busy_flag = 0;
 920					can_busy_loop = false;
 921				}
 922			}
 923		}
 924		/*
 925		 * All waiters have already been registered, so don't provide
 926		 * a poll_table->_qproc to them on the next loop iteration.
 927		 */
 928		pt->_qproc = NULL;
 929		if (!count) {
 930			count = wait->error;
 931			if (signal_pending(current))
 932				count = -ERESTARTNOHAND;
 933		}
 934		if (count || timed_out)
 935			break;
 936
 937		/* only if found POLL_BUSY_LOOP sockets && not out of time */
 938		if (can_busy_loop && !need_resched()) {
 939			if (!busy_start) {
 940				busy_start = busy_loop_current_time();
 941				continue;
 942			}
 943			if (!busy_loop_timeout(busy_start))
 944				continue;
 945		}
 946		busy_flag = 0;
 947
 948		/*
 949		 * If this is the first loop and we have a timeout
 950		 * given, then we convert to ktime_t and set the to
 951		 * pointer to the expiry value.
 952		 */
 953		if (end_time && !to) {
 954			expire = timespec64_to_ktime(*end_time);
 955			to = &expire;
 956		}
 957
 958		if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
 959			timed_out = 1;
 960	}
 961	return count;
 962}
 963
 964#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list))  / \
 965			sizeof(struct pollfd))
 966
 967static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
 968		struct timespec64 *end_time)
 969{
 970	struct poll_wqueues table;
 971	int err = -EFAULT, fdcount;
 972	/* Allocate small arguments on the stack to save memory and be
 973	   faster - use long to make sure the buffer is aligned properly
 974	   on 64 bit archs to avoid unaligned access */
 975	long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
 976	struct poll_list *const head = (struct poll_list *)stack_pps;
 977 	struct poll_list *walk = head;
 978	unsigned int todo = nfds;
 979	unsigned int len;
 980
 981	if (nfds > rlimit(RLIMIT_NOFILE))
 982		return -EINVAL;
 983
 984	len = min_t(unsigned int, nfds, N_STACK_PPS);
 985	for (;;) {
 986		walk->next = NULL;
 987		walk->len = len;
 988		if (!len)
 989			break;
 990
 991		if (copy_from_user(walk->entries, ufds + nfds-todo,
 992					sizeof(struct pollfd) * walk->len))
 993			goto out_fds;
 994
 995		if (walk->len >= todo)
 996			break;
 997		todo -= walk->len;
 
 
 998
 999		len = min(todo, POLLFD_PER_PAGE);
1000		walk = walk->next = kmalloc(struct_size(walk, entries, len),
1001					    GFP_KERNEL);
1002		if (!walk) {
1003			err = -ENOMEM;
1004			goto out_fds;
1005		}
1006	}
1007
1008	poll_initwait(&table);
1009	fdcount = do_poll(head, &table, end_time);
1010	poll_freewait(&table);
1011
1012	if (!user_write_access_begin(ufds, nfds * sizeof(*ufds)))
1013		goto out_fds;
1014
1015	for (walk = head; walk; walk = walk->next) {
1016		struct pollfd *fds = walk->entries;
1017		unsigned int j;
1018
1019		for (j = walk->len; j; fds++, ufds++, j--)
1020			unsafe_put_user(fds->revents, &ufds->revents, Efault);
 
1021  	}
1022	user_write_access_end();
1023
1024	err = fdcount;
1025out_fds:
1026	walk = head->next;
1027	while (walk) {
1028		struct poll_list *pos = walk;
1029		walk = walk->next;
1030		kfree(pos);
1031	}
1032
1033	return err;
1034
1035Efault:
1036	user_write_access_end();
1037	err = -EFAULT;
1038	goto out_fds;
1039}
1040
1041static long do_restart_poll(struct restart_block *restart_block)
1042{
1043	struct pollfd __user *ufds = restart_block->poll.ufds;
1044	int nfds = restart_block->poll.nfds;
1045	struct timespec64 *to = NULL, end_time;
1046	int ret;
1047
1048	if (restart_block->poll.has_timeout) {
1049		end_time.tv_sec = restart_block->poll.tv_sec;
1050		end_time.tv_nsec = restart_block->poll.tv_nsec;
1051		to = &end_time;
1052	}
1053
1054	ret = do_sys_poll(ufds, nfds, to);
1055
1056	if (ret == -ERESTARTNOHAND)
1057		ret = set_restart_fn(restart_block, do_restart_poll);
1058
 
1059	return ret;
1060}
1061
1062SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1063		int, timeout_msecs)
1064{
1065	struct timespec64 end_time, *to = NULL;
1066	int ret;
1067
1068	if (timeout_msecs >= 0) {
1069		to = &end_time;
1070		poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1071			NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1072	}
1073
1074	ret = do_sys_poll(ufds, nfds, to);
1075
1076	if (ret == -ERESTARTNOHAND) {
1077		struct restart_block *restart_block;
1078
1079		restart_block = &current->restart_block;
 
1080		restart_block->poll.ufds = ufds;
1081		restart_block->poll.nfds = nfds;
1082
1083		if (timeout_msecs >= 0) {
1084			restart_block->poll.tv_sec = end_time.tv_sec;
1085			restart_block->poll.tv_nsec = end_time.tv_nsec;
1086			restart_block->poll.has_timeout = 1;
1087		} else
1088			restart_block->poll.has_timeout = 0;
1089
1090		ret = set_restart_fn(restart_block, do_restart_poll);
1091	}
1092	return ret;
1093}
1094
1095SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1096		struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
1097		size_t, sigsetsize)
1098{
1099	struct timespec64 ts, end_time, *to = NULL;
1100	int ret;
1101
1102	if (tsp) {
1103		if (get_timespec64(&ts, tsp))
1104			return -EFAULT;
1105
1106		to = &end_time;
1107		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1108			return -EINVAL;
1109	}
1110
1111	ret = set_user_sigmask(sigmask, sigsetsize);
1112	if (ret)
1113		return ret;
1114
1115	ret = do_sys_poll(ufds, nfds, to);
1116	return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1117}
1118
1119#if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1120
1121SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1122		struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1123		size_t, sigsetsize)
1124{
1125	struct timespec64 ts, end_time, *to = NULL;
1126	int ret;
1127
1128	if (tsp) {
1129		if (get_old_timespec32(&ts, tsp))
1130			return -EFAULT;
1131
1132		to = &end_time;
1133		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1134			return -EINVAL;
1135	}
1136
1137	ret = set_user_sigmask(sigmask, sigsetsize);
1138	if (ret)
1139		return ret;
1140
1141	ret = do_sys_poll(ufds, nfds, to);
1142	return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1143}
1144#endif
1145
1146#ifdef CONFIG_COMPAT
1147#define __COMPAT_NFDBITS       (8 * sizeof(compat_ulong_t))
1148
1149/*
1150 * Ooo, nasty.  We need here to frob 32-bit unsigned longs to
1151 * 64-bit unsigned longs.
1152 */
1153static
1154int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1155			unsigned long *fdset)
1156{
1157	if (ufdset) {
1158		return compat_get_bitmap(fdset, ufdset, nr);
1159	} else {
1160		zero_fd_set(nr, fdset);
1161		return 0;
1162	}
1163}
1164
1165static
1166int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1167		      unsigned long *fdset)
1168{
1169	if (!ufdset)
1170		return 0;
1171	return compat_put_bitmap(ufdset, fdset, nr);
1172}
1173
1174
1175/*
1176 * This is a virtual copy of sys_select from fs/select.c and probably
1177 * should be compared to it from time to time
1178 */
1179
1180/*
1181 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1182 * like to be certain this leads to no problems. So I return
1183 * EINTR just for safety.
1184 *
1185 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1186 * I'm trying ERESTARTNOHAND which restart only when you want to.
1187 */
1188static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1189	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1190	struct timespec64 *end_time)
1191{
1192	fd_set_bits fds;
1193	void *bits;
1194	int size, max_fds, ret = -EINVAL;
1195	struct fdtable *fdt;
1196	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1197
1198	if (n < 0)
1199		goto out_nofds;
1200
1201	/* max_fds can increase, so grab it once to avoid race */
1202	rcu_read_lock();
1203	fdt = files_fdtable(current->files);
1204	max_fds = fdt->max_fds;
1205	rcu_read_unlock();
1206	if (n > max_fds)
1207		n = max_fds;
1208
1209	/*
1210	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1211	 * since we used fdset we need to allocate memory in units of
1212	 * long-words.
1213	 */
1214	size = FDS_BYTES(n);
1215	bits = stack_fds;
1216	if (size > sizeof(stack_fds) / 6) {
1217		bits = kmalloc_array(6, size, GFP_KERNEL);
1218		ret = -ENOMEM;
1219		if (!bits)
1220			goto out_nofds;
1221	}
1222	fds.in      = (unsigned long *)  bits;
1223	fds.out     = (unsigned long *) (bits +   size);
1224	fds.ex      = (unsigned long *) (bits + 2*size);
1225	fds.res_in  = (unsigned long *) (bits + 3*size);
1226	fds.res_out = (unsigned long *) (bits + 4*size);
1227	fds.res_ex  = (unsigned long *) (bits + 5*size);
1228
1229	if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1230	    (ret = compat_get_fd_set(n, outp, fds.out)) ||
1231	    (ret = compat_get_fd_set(n, exp, fds.ex)))
1232		goto out;
1233	zero_fd_set(n, fds.res_in);
1234	zero_fd_set(n, fds.res_out);
1235	zero_fd_set(n, fds.res_ex);
1236
1237	ret = do_select(n, &fds, end_time);
1238
1239	if (ret < 0)
1240		goto out;
1241	if (!ret) {
1242		ret = -ERESTARTNOHAND;
1243		if (signal_pending(current))
1244			goto out;
1245		ret = 0;
1246	}
1247
1248	if (compat_set_fd_set(n, inp, fds.res_in) ||
1249	    compat_set_fd_set(n, outp, fds.res_out) ||
1250	    compat_set_fd_set(n, exp, fds.res_ex))
1251		ret = -EFAULT;
1252out:
1253	if (bits != stack_fds)
1254		kfree(bits);
1255out_nofds:
1256	return ret;
1257}
1258
1259static int do_compat_select(int n, compat_ulong_t __user *inp,
1260	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1261	struct old_timeval32 __user *tvp)
1262{
1263	struct timespec64 end_time, *to = NULL;
1264	struct old_timeval32 tv;
1265	int ret;
1266
1267	if (tvp) {
1268		if (copy_from_user(&tv, tvp, sizeof(tv)))
1269			return -EFAULT;
1270
1271		to = &end_time;
1272		if (poll_select_set_timeout(to,
1273				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1274				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1275			return -EINVAL;
1276	}
1277
1278	ret = compat_core_sys_select(n, inp, outp, exp, to);
1279	return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret);
1280}
1281
1282COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1283	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1284	struct old_timeval32 __user *, tvp)
1285{
1286	return do_compat_select(n, inp, outp, exp, tvp);
1287}
1288
1289struct compat_sel_arg_struct {
1290	compat_ulong_t n;
1291	compat_uptr_t inp;
1292	compat_uptr_t outp;
1293	compat_uptr_t exp;
1294	compat_uptr_t tvp;
1295};
1296
1297COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1298{
1299	struct compat_sel_arg_struct a;
1300
1301	if (copy_from_user(&a, arg, sizeof(a)))
1302		return -EFAULT;
1303	return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1304				compat_ptr(a.exp), compat_ptr(a.tvp));
1305}
1306
1307static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1308	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1309	void __user *tsp, compat_sigset_t __user *sigmask,
1310	compat_size_t sigsetsize, enum poll_time_type type)
1311{
1312	struct timespec64 ts, end_time, *to = NULL;
1313	int ret;
1314
1315	if (tsp) {
1316		switch (type) {
1317		case PT_OLD_TIMESPEC:
1318			if (get_old_timespec32(&ts, tsp))
1319				return -EFAULT;
1320			break;
1321		case PT_TIMESPEC:
1322			if (get_timespec64(&ts, tsp))
1323				return -EFAULT;
1324			break;
1325		default:
1326			BUG();
1327		}
1328
1329		to = &end_time;
1330		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1331			return -EINVAL;
1332	}
1333
1334	ret = set_compat_user_sigmask(sigmask, sigsetsize);
1335	if (ret)
1336		return ret;
1337
1338	ret = compat_core_sys_select(n, inp, outp, exp, to);
1339	return poll_select_finish(&end_time, tsp, type, ret);
1340}
1341
1342struct compat_sigset_argpack {
1343	compat_uptr_t p;
1344	compat_size_t size;
1345};
1346static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to,
1347					    struct compat_sigset_argpack __user *from)
1348{
1349	if (from) {
1350		if (!user_read_access_begin(from, sizeof(*from)))
1351			return -EFAULT;
1352		unsafe_get_user(to->p, &from->p, Efault);
1353		unsafe_get_user(to->size, &from->size, Efault);
1354		user_read_access_end();
1355	}
1356	return 0;
1357Efault:
1358	user_read_access_end();
1359	return -EFAULT;
1360}
1361
1362COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1363	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1364	struct __kernel_timespec __user *, tsp, void __user *, sig)
1365{
1366	struct compat_sigset_argpack x = {0, 0};
1367
1368	if (get_compat_sigset_argpack(&x, sig))
1369		return -EFAULT;
1370
1371	return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1372				 x.size, PT_TIMESPEC);
1373}
1374
1375#if defined(CONFIG_COMPAT_32BIT_TIME)
1376
1377COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp,
1378	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1379	struct old_timespec32 __user *, tsp, void __user *, sig)
1380{
1381	struct compat_sigset_argpack x = {0, 0};
1382
1383	if (get_compat_sigset_argpack(&x, sig))
1384		return -EFAULT;
1385
1386	return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1387				 x.size, PT_OLD_TIMESPEC);
1388}
1389
1390#endif
1391
1392#if defined(CONFIG_COMPAT_32BIT_TIME)
1393COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds,
1394	unsigned int,  nfds, struct old_timespec32 __user *, tsp,
1395	const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1396{
1397	struct timespec64 ts, end_time, *to = NULL;
1398	int ret;
1399
1400	if (tsp) {
1401		if (get_old_timespec32(&ts, tsp))
1402			return -EFAULT;
1403
1404		to = &end_time;
1405		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1406			return -EINVAL;
1407	}
1408
1409	ret = set_compat_user_sigmask(sigmask, sigsetsize);
1410	if (ret)
1411		return ret;
1412
1413	ret = do_sys_poll(ufds, nfds, to);
1414	return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1415}
1416#endif
1417
1418/* New compat syscall for 64 bit time_t*/
1419COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1420	unsigned int,  nfds, struct __kernel_timespec __user *, tsp,
1421	const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1422{
1423	struct timespec64 ts, end_time, *to = NULL;
1424	int ret;
1425
1426	if (tsp) {
1427		if (get_timespec64(&ts, tsp))
1428			return -EFAULT;
1429
1430		to = &end_time;
1431		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1432			return -EINVAL;
1433	}
1434
1435	ret = set_compat_user_sigmask(sigmask, sigsetsize);
1436	if (ret)
1437		return ret;
1438
1439	ret = do_sys_poll(ufds, nfds, to);
1440	return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1441}
1442
1443#endif
v5.9
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * This file contains the procedures for the handling of select and poll
   4 *
   5 * Created for Linux based loosely upon Mathius Lattner's minix
   6 * patches by Peter MacDonald. Heavily edited by Linus.
   7 *
   8 *  4 February 1994
   9 *     COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS
  10 *     flag set in its personality we do *not* modify the given timeout
  11 *     parameter to reflect time remaining.
  12 *
  13 *  24 January 2000
  14 *     Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation 
  15 *     of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
  16 */
  17
 
  18#include <linux/kernel.h>
  19#include <linux/sched/signal.h>
  20#include <linux/sched/rt.h>
  21#include <linux/syscalls.h>
  22#include <linux/export.h>
  23#include <linux/slab.h>
  24#include <linux/poll.h>
  25#include <linux/personality.h> /* for STICKY_TIMEOUTS */
  26#include <linux/file.h>
  27#include <linux/fdtable.h>
  28#include <linux/fs.h>
  29#include <linux/rcupdate.h>
  30#include <linux/hrtimer.h>
  31#include <linux/freezer.h>
  32#include <net/busy_poll.h>
  33#include <linux/vmalloc.h>
  34
  35#include <linux/uaccess.h>
  36
  37
  38/*
  39 * Estimate expected accuracy in ns from a timeval.
  40 *
  41 * After quite a bit of churning around, we've settled on
  42 * a simple thing of taking 0.1% of the timeout as the
  43 * slack, with a cap of 100 msec.
  44 * "nice" tasks get a 0.5% slack instead.
  45 *
  46 * Consider this comment an open invitation to come up with even
  47 * better solutions..
  48 */
  49
  50#define MAX_SLACK	(100 * NSEC_PER_MSEC)
  51
  52static long __estimate_accuracy(struct timespec64 *tv)
  53{
  54	long slack;
  55	int divfactor = 1000;
  56
  57	if (tv->tv_sec < 0)
  58		return 0;
  59
  60	if (task_nice(current) > 0)
  61		divfactor = divfactor / 5;
  62
  63	if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
  64		return MAX_SLACK;
  65
  66	slack = tv->tv_nsec / divfactor;
  67	slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
  68
  69	if (slack > MAX_SLACK)
  70		return MAX_SLACK;
  71
  72	return slack;
  73}
  74
  75u64 select_estimate_accuracy(struct timespec64 *tv)
  76{
  77	u64 ret;
  78	struct timespec64 now;
 
  79
  80	/*
  81	 * Realtime tasks get a slack of 0 for obvious reasons.
  82	 */
  83
  84	if (rt_task(current))
  85		return 0;
  86
  87	ktime_get_ts64(&now);
  88	now = timespec64_sub(*tv, now);
  89	ret = __estimate_accuracy(&now);
  90	if (ret < current->timer_slack_ns)
  91		return current->timer_slack_ns;
  92	return ret;
  93}
  94
  95
  96
  97struct poll_table_page {
  98	struct poll_table_page * next;
  99	struct poll_table_entry * entry;
 100	struct poll_table_entry entries[0];
 101};
 102
 103#define POLL_TABLE_FULL(table) \
 104	((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
 105
 106/*
 107 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
 108 * I have rewritten this, taking some shortcuts: This code may not be easy to
 109 * follow, but it should be free of race-conditions, and it's practical. If you
 110 * understand what I'm doing here, then you understand how the linux
 111 * sleep/wakeup mechanism works.
 112 *
 113 * Two very simple procedures, poll_wait() and poll_freewait() make all the
 114 * work.  poll_wait() is an inline-function defined in <linux/poll.h>,
 115 * as all select/poll functions have to call it to add an entry to the
 116 * poll table.
 117 */
 118static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
 119		       poll_table *p);
 120
 121void poll_initwait(struct poll_wqueues *pwq)
 122{
 123	init_poll_funcptr(&pwq->pt, __pollwait);
 124	pwq->polling_task = current;
 125	pwq->triggered = 0;
 126	pwq->error = 0;
 127	pwq->table = NULL;
 128	pwq->inline_index = 0;
 129}
 130EXPORT_SYMBOL(poll_initwait);
 131
 132static void free_poll_entry(struct poll_table_entry *entry)
 133{
 134	remove_wait_queue(entry->wait_address, &entry->wait);
 135	fput(entry->filp);
 136}
 137
 138void poll_freewait(struct poll_wqueues *pwq)
 139{
 140	struct poll_table_page * p = pwq->table;
 141	int i;
 142	for (i = 0; i < pwq->inline_index; i++)
 143		free_poll_entry(pwq->inline_entries + i);
 144	while (p) {
 145		struct poll_table_entry * entry;
 146		struct poll_table_page *old;
 147
 148		entry = p->entry;
 149		do {
 150			entry--;
 151			free_poll_entry(entry);
 152		} while (entry > p->entries);
 153		old = p;
 154		p = p->next;
 155		free_page((unsigned long) old);
 156	}
 157}
 158EXPORT_SYMBOL(poll_freewait);
 159
 160static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
 161{
 162	struct poll_table_page *table = p->table;
 163
 164	if (p->inline_index < N_INLINE_POLL_ENTRIES)
 165		return p->inline_entries + p->inline_index++;
 166
 167	if (!table || POLL_TABLE_FULL(table)) {
 168		struct poll_table_page *new_table;
 169
 170		new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
 171		if (!new_table) {
 172			p->error = -ENOMEM;
 173			return NULL;
 174		}
 175		new_table->entry = new_table->entries;
 176		new_table->next = table;
 177		p->table = new_table;
 178		table = new_table;
 179	}
 180
 181	return table->entry++;
 182}
 183
 184static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
 185{
 186	struct poll_wqueues *pwq = wait->private;
 187	DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
 188
 189	/*
 190	 * Although this function is called under waitqueue lock, LOCK
 191	 * doesn't imply write barrier and the users expect write
 192	 * barrier semantics on wakeup functions.  The following
 193	 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
 194	 * and is paired with smp_store_mb() in poll_schedule_timeout.
 195	 */
 196	smp_wmb();
 197	pwq->triggered = 1;
 198
 199	/*
 200	 * Perform the default wake up operation using a dummy
 201	 * waitqueue.
 202	 *
 203	 * TODO: This is hacky but there currently is no interface to
 204	 * pass in @sync.  @sync is scheduled to be removed and once
 205	 * that happens, wake_up_process() can be used directly.
 206	 */
 207	return default_wake_function(&dummy_wait, mode, sync, key);
 208}
 209
 210static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
 211{
 212	struct poll_table_entry *entry;
 213
 214	entry = container_of(wait, struct poll_table_entry, wait);
 215	if (key && !(key_to_poll(key) & entry->key))
 216		return 0;
 217	return __pollwake(wait, mode, sync, key);
 218}
 219
 220/* Add a new entry */
 221static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
 222				poll_table *p)
 223{
 224	struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
 225	struct poll_table_entry *entry = poll_get_entry(pwq);
 226	if (!entry)
 227		return;
 228	entry->filp = get_file(filp);
 229	entry->wait_address = wait_address;
 230	entry->key = p->_key;
 231	init_waitqueue_func_entry(&entry->wait, pollwake);
 232	entry->wait.private = pwq;
 233	add_wait_queue(wait_address, &entry->wait);
 234}
 235
 236static int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
 237			  ktime_t *expires, unsigned long slack)
 238{
 239	int rc = -EINTR;
 240
 241	set_current_state(state);
 242	if (!pwq->triggered)
 243		rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
 244	__set_current_state(TASK_RUNNING);
 245
 246	/*
 247	 * Prepare for the next iteration.
 248	 *
 249	 * The following smp_store_mb() serves two purposes.  First, it's
 250	 * the counterpart rmb of the wmb in pollwake() such that data
 251	 * written before wake up is always visible after wake up.
 252	 * Second, the full barrier guarantees that triggered clearing
 253	 * doesn't pass event check of the next iteration.  Note that
 254	 * this problem doesn't exist for the first iteration as
 255	 * add_wait_queue() has full barrier semantics.
 256	 */
 257	smp_store_mb(pwq->triggered, 0);
 258
 259	return rc;
 260}
 261
 262/**
 263 * poll_select_set_timeout - helper function to setup the timeout value
 264 * @to:		pointer to timespec64 variable for the final timeout
 265 * @sec:	seconds (from user space)
 266 * @nsec:	nanoseconds (from user space)
 267 *
 268 * Note, we do not use a timespec for the user space value here, That
 269 * way we can use the function for timeval and compat interfaces as well.
 270 *
 271 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
 272 */
 273int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
 274{
 275	struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
 276
 277	if (!timespec64_valid(&ts))
 278		return -EINVAL;
 279
 280	/* Optimize for the zero timeout value here */
 281	if (!sec && !nsec) {
 282		to->tv_sec = to->tv_nsec = 0;
 283	} else {
 284		ktime_get_ts64(to);
 285		*to = timespec64_add_safe(*to, ts);
 286	}
 287	return 0;
 288}
 289
 290enum poll_time_type {
 291	PT_TIMEVAL = 0,
 292	PT_OLD_TIMEVAL = 1,
 293	PT_TIMESPEC = 2,
 294	PT_OLD_TIMESPEC = 3,
 295};
 296
 297static int poll_select_finish(struct timespec64 *end_time,
 298			      void __user *p,
 299			      enum poll_time_type pt_type, int ret)
 300{
 301	struct timespec64 rts;
 302
 303	restore_saved_sigmask_unless(ret == -ERESTARTNOHAND);
 304
 305	if (!p)
 306		return ret;
 307
 308	if (current->personality & STICKY_TIMEOUTS)
 309		goto sticky;
 310
 311	/* No update for zero timeout */
 312	if (!end_time->tv_sec && !end_time->tv_nsec)
 313		return ret;
 314
 315	ktime_get_ts64(&rts);
 316	rts = timespec64_sub(*end_time, rts);
 317	if (rts.tv_sec < 0)
 318		rts.tv_sec = rts.tv_nsec = 0;
 319
 320
 321	switch (pt_type) {
 322	case PT_TIMEVAL:
 323		{
 324			struct __kernel_old_timeval rtv;
 325
 326			if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
 327				memset(&rtv, 0, sizeof(rtv));
 328			rtv.tv_sec = rts.tv_sec;
 329			rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
 330			if (!copy_to_user(p, &rtv, sizeof(rtv)))
 331				return ret;
 332		}
 333		break;
 334	case PT_OLD_TIMEVAL:
 335		{
 336			struct old_timeval32 rtv;
 337
 338			rtv.tv_sec = rts.tv_sec;
 339			rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
 340			if (!copy_to_user(p, &rtv, sizeof(rtv)))
 341				return ret;
 342		}
 343		break;
 344	case PT_TIMESPEC:
 345		if (!put_timespec64(&rts, p))
 346			return ret;
 347		break;
 348	case PT_OLD_TIMESPEC:
 349		if (!put_old_timespec32(&rts, p))
 350			return ret;
 351		break;
 352	default:
 353		BUG();
 354	}
 355	/*
 356	 * If an application puts its timeval in read-only memory, we
 357	 * don't want the Linux-specific update to the timeval to
 358	 * cause a fault after the select has completed
 359	 * successfully. However, because we're not updating the
 360	 * timeval, we can't restart the system call.
 361	 */
 362
 363sticky:
 364	if (ret == -ERESTARTNOHAND)
 365		ret = -EINTR;
 366	return ret;
 367}
 368
 369/*
 370 * Scalable version of the fd_set.
 371 */
 372
 373typedef struct {
 374	unsigned long *in, *out, *ex;
 375	unsigned long *res_in, *res_out, *res_ex;
 376} fd_set_bits;
 377
 378/*
 379 * How many longwords for "nr" bits?
 380 */
 381#define FDS_BITPERLONG	(8*sizeof(long))
 382#define FDS_LONGS(nr)	(((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
 383#define FDS_BYTES(nr)	(FDS_LONGS(nr)*sizeof(long))
 384
 385/*
 386 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
 387 */
 388static inline
 389int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
 390{
 391	nr = FDS_BYTES(nr);
 392	if (ufdset)
 393		return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0;
 394
 395	memset(fdset, 0, nr);
 396	return 0;
 397}
 398
 399static inline unsigned long __must_check
 400set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
 401{
 402	if (ufdset)
 403		return __copy_to_user(ufdset, fdset, FDS_BYTES(nr));
 404	return 0;
 405}
 406
 407static inline
 408void zero_fd_set(unsigned long nr, unsigned long *fdset)
 409{
 410	memset(fdset, 0, FDS_BYTES(nr));
 411}
 412
 413#define FDS_IN(fds, n)		(fds->in + n)
 414#define FDS_OUT(fds, n)		(fds->out + n)
 415#define FDS_EX(fds, n)		(fds->ex + n)
 416
 417#define BITS(fds, n)	(*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
 418
 419static int max_select_fd(unsigned long n, fd_set_bits *fds)
 420{
 421	unsigned long *open_fds;
 422	unsigned long set;
 423	int max;
 424	struct fdtable *fdt;
 425
 426	/* handle last in-complete long-word first */
 427	set = ~(~0UL << (n & (BITS_PER_LONG-1)));
 428	n /= BITS_PER_LONG;
 429	fdt = files_fdtable(current->files);
 430	open_fds = fdt->open_fds + n;
 431	max = 0;
 432	if (set) {
 433		set &= BITS(fds, n);
 434		if (set) {
 435			if (!(set & ~*open_fds))
 436				goto get_max;
 437			return -EBADF;
 438		}
 439	}
 440	while (n) {
 441		open_fds--;
 442		n--;
 443		set = BITS(fds, n);
 444		if (!set)
 445			continue;
 446		if (set & ~*open_fds)
 447			return -EBADF;
 448		if (max)
 449			continue;
 450get_max:
 451		do {
 452			max++;
 453			set >>= 1;
 454		} while (set);
 455		max += n * BITS_PER_LONG;
 456	}
 457
 458	return max;
 459}
 460
 461#define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR)
 462#define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR)
 463#define POLLEX_SET (EPOLLPRI)
 
 
 464
 465static inline void wait_key_set(poll_table *wait, unsigned long in,
 466				unsigned long out, unsigned long bit,
 467				__poll_t ll_flag)
 468{
 
 
 
 
 
 469	wait->_key = POLLEX_SET | ll_flag;
 470	if (in & bit)
 471		wait->_key |= POLLIN_SET;
 472	if (out & bit)
 473		wait->_key |= POLLOUT_SET;
 
 
 474}
 475
 476static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
 477{
 478	ktime_t expire, *to = NULL;
 479	struct poll_wqueues table;
 480	poll_table *wait;
 481	int retval, i, timed_out = 0;
 482	u64 slack = 0;
 483	__poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
 484	unsigned long busy_start = 0;
 485
 486	rcu_read_lock();
 487	retval = max_select_fd(n, fds);
 488	rcu_read_unlock();
 489
 490	if (retval < 0)
 491		return retval;
 492	n = retval;
 493
 494	poll_initwait(&table);
 495	wait = &table.pt;
 496	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
 497		wait->_qproc = NULL;
 498		timed_out = 1;
 499	}
 500
 501	if (end_time && !timed_out)
 502		slack = select_estimate_accuracy(end_time);
 503
 504	retval = 0;
 505	for (;;) {
 506		unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
 507		bool can_busy_loop = false;
 508
 509		inp = fds->in; outp = fds->out; exp = fds->ex;
 510		rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
 511
 512		for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
 513			unsigned long in, out, ex, all_bits, bit = 1, j;
 514			unsigned long res_in = 0, res_out = 0, res_ex = 0;
 515			__poll_t mask;
 516
 517			in = *inp++; out = *outp++; ex = *exp++;
 518			all_bits = in | out | ex;
 519			if (all_bits == 0) {
 520				i += BITS_PER_LONG;
 521				continue;
 522			}
 523
 524			for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
 525				struct fd f;
 526				if (i >= n)
 527					break;
 528				if (!(bit & all_bits))
 529					continue;
 530				f = fdget(i);
 531				if (f.file) {
 532					wait_key_set(wait, in, out, bit,
 533						     busy_flag);
 534					mask = vfs_poll(f.file, wait);
 535
 536					fdput(f);
 537					if ((mask & POLLIN_SET) && (in & bit)) {
 538						res_in |= bit;
 539						retval++;
 540						wait->_qproc = NULL;
 541					}
 542					if ((mask & POLLOUT_SET) && (out & bit)) {
 543						res_out |= bit;
 544						retval++;
 545						wait->_qproc = NULL;
 546					}
 547					if ((mask & POLLEX_SET) && (ex & bit)) {
 548						res_ex |= bit;
 549						retval++;
 550						wait->_qproc = NULL;
 551					}
 552					/* got something, stop busy polling */
 553					if (retval) {
 554						can_busy_loop = false;
 555						busy_flag = 0;
 556
 557					/*
 558					 * only remember a returned
 559					 * POLL_BUSY_LOOP if we asked for it
 560					 */
 561					} else if (busy_flag & mask)
 562						can_busy_loop = true;
 563
 564				}
 565			}
 566			if (res_in)
 567				*rinp = res_in;
 568			if (res_out)
 569				*routp = res_out;
 570			if (res_ex)
 571				*rexp = res_ex;
 572			cond_resched();
 573		}
 574		wait->_qproc = NULL;
 575		if (retval || timed_out || signal_pending(current))
 576			break;
 577		if (table.error) {
 578			retval = table.error;
 579			break;
 580		}
 581
 582		/* only if found POLL_BUSY_LOOP sockets && not out of time */
 583		if (can_busy_loop && !need_resched()) {
 584			if (!busy_start) {
 585				busy_start = busy_loop_current_time();
 586				continue;
 587			}
 588			if (!busy_loop_timeout(busy_start))
 589				continue;
 590		}
 591		busy_flag = 0;
 592
 593		/*
 594		 * If this is the first loop and we have a timeout
 595		 * given, then we convert to ktime_t and set the to
 596		 * pointer to the expiry value.
 597		 */
 598		if (end_time && !to) {
 599			expire = timespec64_to_ktime(*end_time);
 600			to = &expire;
 601		}
 602
 603		if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
 604					   to, slack))
 605			timed_out = 1;
 606	}
 607
 608	poll_freewait(&table);
 609
 610	return retval;
 611}
 612
 613/*
 614 * We can actually return ERESTARTSYS instead of EINTR, but I'd
 615 * like to be certain this leads to no problems. So I return
 616 * EINTR just for safety.
 617 *
 618 * Update: ERESTARTSYS breaks at least the xview clock binary, so
 619 * I'm trying ERESTARTNOHAND which restart only when you want to.
 620 */
 621int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
 622			   fd_set __user *exp, struct timespec64 *end_time)
 623{
 624	fd_set_bits fds;
 625	void *bits;
 626	int ret, max_fds;
 627	size_t size, alloc_size;
 628	struct fdtable *fdt;
 629	/* Allocate small arguments on the stack to save memory and be faster */
 630	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
 631
 632	ret = -EINVAL;
 633	if (n < 0)
 634		goto out_nofds;
 635
 636	/* max_fds can increase, so grab it once to avoid race */
 637	rcu_read_lock();
 638	fdt = files_fdtable(current->files);
 639	max_fds = fdt->max_fds;
 640	rcu_read_unlock();
 641	if (n > max_fds)
 642		n = max_fds;
 643
 644	/*
 645	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
 646	 * since we used fdset we need to allocate memory in units of
 647	 * long-words. 
 648	 */
 649	size = FDS_BYTES(n);
 650	bits = stack_fds;
 651	if (size > sizeof(stack_fds) / 6) {
 652		/* Not enough space in on-stack array; must use kmalloc */
 653		ret = -ENOMEM;
 654		if (size > (SIZE_MAX / 6))
 655			goto out_nofds;
 656
 657		alloc_size = 6 * size;
 658		bits = kvmalloc(alloc_size, GFP_KERNEL);
 659		if (!bits)
 660			goto out_nofds;
 661	}
 662	fds.in      = bits;
 663	fds.out     = bits +   size;
 664	fds.ex      = bits + 2*size;
 665	fds.res_in  = bits + 3*size;
 666	fds.res_out = bits + 4*size;
 667	fds.res_ex  = bits + 5*size;
 668
 669	if ((ret = get_fd_set(n, inp, fds.in)) ||
 670	    (ret = get_fd_set(n, outp, fds.out)) ||
 671	    (ret = get_fd_set(n, exp, fds.ex)))
 672		goto out;
 673	zero_fd_set(n, fds.res_in);
 674	zero_fd_set(n, fds.res_out);
 675	zero_fd_set(n, fds.res_ex);
 676
 677	ret = do_select(n, &fds, end_time);
 678
 679	if (ret < 0)
 680		goto out;
 681	if (!ret) {
 682		ret = -ERESTARTNOHAND;
 683		if (signal_pending(current))
 684			goto out;
 685		ret = 0;
 686	}
 687
 688	if (set_fd_set(n, inp, fds.res_in) ||
 689	    set_fd_set(n, outp, fds.res_out) ||
 690	    set_fd_set(n, exp, fds.res_ex))
 691		ret = -EFAULT;
 692
 693out:
 694	if (bits != stack_fds)
 695		kvfree(bits);
 696out_nofds:
 697	return ret;
 698}
 699
 700static int kern_select(int n, fd_set __user *inp, fd_set __user *outp,
 701		       fd_set __user *exp, struct __kernel_old_timeval __user *tvp)
 702{
 703	struct timespec64 end_time, *to = NULL;
 704	struct __kernel_old_timeval tv;
 705	int ret;
 706
 707	if (tvp) {
 708		if (copy_from_user(&tv, tvp, sizeof(tv)))
 709			return -EFAULT;
 710
 711		to = &end_time;
 712		if (poll_select_set_timeout(to,
 713				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
 714				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
 715			return -EINVAL;
 716	}
 717
 718	ret = core_sys_select(n, inp, outp, exp, to);
 719	return poll_select_finish(&end_time, tvp, PT_TIMEVAL, ret);
 720}
 721
 722SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
 723		fd_set __user *, exp, struct __kernel_old_timeval __user *, tvp)
 724{
 725	return kern_select(n, inp, outp, exp, tvp);
 726}
 727
 728static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
 729		       fd_set __user *exp, void __user *tsp,
 730		       const sigset_t __user *sigmask, size_t sigsetsize,
 731		       enum poll_time_type type)
 732{
 733	struct timespec64 ts, end_time, *to = NULL;
 734	int ret;
 735
 736	if (tsp) {
 737		switch (type) {
 738		case PT_TIMESPEC:
 739			if (get_timespec64(&ts, tsp))
 740				return -EFAULT;
 741			break;
 742		case PT_OLD_TIMESPEC:
 743			if (get_old_timespec32(&ts, tsp))
 744				return -EFAULT;
 745			break;
 746		default:
 747			BUG();
 748		}
 749
 750		to = &end_time;
 751		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
 752			return -EINVAL;
 753	}
 754
 755	ret = set_user_sigmask(sigmask, sigsetsize);
 756	if (ret)
 757		return ret;
 758
 759	ret = core_sys_select(n, inp, outp, exp, to);
 760	return poll_select_finish(&end_time, tsp, type, ret);
 761}
 762
 763/*
 764 * Most architectures can't handle 7-argument syscalls. So we provide a
 765 * 6-argument version where the sixth argument is a pointer to a structure
 766 * which has a pointer to the sigset_t itself followed by a size_t containing
 767 * the sigset size.
 768 */
 769struct sigset_argpack {
 770	sigset_t __user *p;
 771	size_t size;
 772};
 773
 774static inline int get_sigset_argpack(struct sigset_argpack *to,
 775				     struct sigset_argpack __user *from)
 776{
 777	// the path is hot enough for overhead of copy_from_user() to matter
 778	if (from) {
 779		if (!user_read_access_begin(from, sizeof(*from)))
 
 
 780			return -EFAULT;
 781		unsafe_get_user(to->p, &from->p, Efault);
 782		unsafe_get_user(to->size, &from->size, Efault);
 783		user_read_access_end();
 784	}
 785	return 0;
 786Efault:
 787	user_access_end();
 788	return -EFAULT;
 789}
 790
 791SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
 792		fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
 793		void __user *, sig)
 794{
 795	struct sigset_argpack x = {NULL, 0};
 796
 797	if (get_sigset_argpack(&x, sig))
 798		return -EFAULT;
 799
 800	return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_TIMESPEC);
 801}
 802
 803#if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
 804
 805SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
 806		fd_set __user *, exp, struct old_timespec32 __user *, tsp,
 807		void __user *, sig)
 808{
 809	struct sigset_argpack x = {NULL, 0};
 810
 811	if (get_sigset_argpack(&x, sig))
 812		return -EFAULT;
 813
 814	return do_pselect(n, inp, outp, exp, tsp, x.p, x.size, PT_OLD_TIMESPEC);
 815}
 816
 817#endif
 818
 819#ifdef __ARCH_WANT_SYS_OLD_SELECT
 820struct sel_arg_struct {
 821	unsigned long n;
 822	fd_set __user *inp, *outp, *exp;
 823	struct __kernel_old_timeval __user *tvp;
 824};
 825
 826SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
 827{
 828	struct sel_arg_struct a;
 829
 830	if (copy_from_user(&a, arg, sizeof(a)))
 831		return -EFAULT;
 832	return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
 833}
 834#endif
 835
 836struct poll_list {
 837	struct poll_list *next;
 838	int len;
 839	struct pollfd entries[0];
 840};
 841
 842#define POLLFD_PER_PAGE  ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
 843
 844/*
 845 * Fish for pollable events on the pollfd->fd file descriptor. We're only
 846 * interested in events matching the pollfd->events mask, and the result
 847 * matching that mask is both recorded in pollfd->revents and returned. The
 848 * pwait poll_table will be used by the fd-provided poll handler for waiting,
 849 * if pwait->_qproc is non-NULL.
 850 */
 851static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
 852				     bool *can_busy_poll,
 853				     __poll_t busy_flag)
 854{
 855	int fd = pollfd->fd;
 856	__poll_t mask = 0, filter;
 857	struct fd f;
 858
 859	if (fd < 0)
 860		goto out;
 861	mask = EPOLLNVAL;
 862	f = fdget(fd);
 863	if (!f.file)
 864		goto out;
 865
 866	/* userland u16 ->events contains POLL... bitmap */
 867	filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
 868	pwait->_key = filter | busy_flag;
 869	mask = vfs_poll(f.file, pwait);
 870	if (mask & busy_flag)
 871		*can_busy_poll = true;
 872	mask &= filter;		/* Mask out unneeded events. */
 873	fdput(f);
 874
 875out:
 876	/* ... and so does ->revents */
 877	pollfd->revents = mangle_poll(mask);
 878	return mask;
 879}
 880
 881static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
 882		   struct timespec64 *end_time)
 883{
 884	poll_table* pt = &wait->pt;
 885	ktime_t expire, *to = NULL;
 886	int timed_out = 0, count = 0;
 887	u64 slack = 0;
 888	__poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
 889	unsigned long busy_start = 0;
 890
 891	/* Optimise the no-wait case */
 892	if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
 893		pt->_qproc = NULL;
 894		timed_out = 1;
 895	}
 896
 897	if (end_time && !timed_out)
 898		slack = select_estimate_accuracy(end_time);
 899
 900	for (;;) {
 901		struct poll_list *walk;
 902		bool can_busy_loop = false;
 903
 904		for (walk = list; walk != NULL; walk = walk->next) {
 905			struct pollfd * pfd, * pfd_end;
 906
 907			pfd = walk->entries;
 908			pfd_end = pfd + walk->len;
 909			for (; pfd != pfd_end; pfd++) {
 
 910				/*
 911				 * Fish for events. If we found one, record it
 912				 * and kill poll_table->_qproc, so we don't
 913				 * needlessly register any other waiters after
 914				 * this. They'll get immediately deregistered
 915				 * when we break out and return.
 916				 */
 917				if (do_pollfd(pfd, pt, &can_busy_loop,
 918					      busy_flag)) {
 
 919					count++;
 920					pt->_qproc = NULL;
 921					/* found something, stop busy polling */
 922					busy_flag = 0;
 923					can_busy_loop = false;
 924				}
 925			}
 926		}
 927		/*
 928		 * All waiters have already been registered, so don't provide
 929		 * a poll_table->_qproc to them on the next loop iteration.
 930		 */
 931		pt->_qproc = NULL;
 932		if (!count) {
 933			count = wait->error;
 934			if (signal_pending(current))
 935				count = -ERESTARTNOHAND;
 936		}
 937		if (count || timed_out)
 938			break;
 939
 940		/* only if found POLL_BUSY_LOOP sockets && not out of time */
 941		if (can_busy_loop && !need_resched()) {
 942			if (!busy_start) {
 943				busy_start = busy_loop_current_time();
 944				continue;
 945			}
 946			if (!busy_loop_timeout(busy_start))
 947				continue;
 948		}
 949		busy_flag = 0;
 950
 951		/*
 952		 * If this is the first loop and we have a timeout
 953		 * given, then we convert to ktime_t and set the to
 954		 * pointer to the expiry value.
 955		 */
 956		if (end_time && !to) {
 957			expire = timespec64_to_ktime(*end_time);
 958			to = &expire;
 959		}
 960
 961		if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
 962			timed_out = 1;
 963	}
 964	return count;
 965}
 966
 967#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list))  / \
 968			sizeof(struct pollfd))
 969
 970static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
 971		struct timespec64 *end_time)
 972{
 973	struct poll_wqueues table;
 974	int err = -EFAULT, fdcount, len;
 975	/* Allocate small arguments on the stack to save memory and be
 976	   faster - use long to make sure the buffer is aligned properly
 977	   on 64 bit archs to avoid unaligned access */
 978	long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
 979	struct poll_list *const head = (struct poll_list *)stack_pps;
 980 	struct poll_list *walk = head;
 981 	unsigned long todo = nfds;
 
 982
 983	if (nfds > rlimit(RLIMIT_NOFILE))
 984		return -EINVAL;
 985
 986	len = min_t(unsigned int, nfds, N_STACK_PPS);
 987	for (;;) {
 988		walk->next = NULL;
 989		walk->len = len;
 990		if (!len)
 991			break;
 992
 993		if (copy_from_user(walk->entries, ufds + nfds-todo,
 994					sizeof(struct pollfd) * walk->len))
 995			goto out_fds;
 996
 
 
 997		todo -= walk->len;
 998		if (!todo)
 999			break;
1000
1001		len = min(todo, POLLFD_PER_PAGE);
1002		walk = walk->next = kmalloc(struct_size(walk, entries, len),
1003					    GFP_KERNEL);
1004		if (!walk) {
1005			err = -ENOMEM;
1006			goto out_fds;
1007		}
1008	}
1009
1010	poll_initwait(&table);
1011	fdcount = do_poll(head, &table, end_time);
1012	poll_freewait(&table);
1013
 
 
 
1014	for (walk = head; walk; walk = walk->next) {
1015		struct pollfd *fds = walk->entries;
1016		int j;
1017
1018		for (j = 0; j < walk->len; j++, ufds++)
1019			if (__put_user(fds[j].revents, &ufds->revents))
1020				goto out_fds;
1021  	}
 
1022
1023	err = fdcount;
1024out_fds:
1025	walk = head->next;
1026	while (walk) {
1027		struct poll_list *pos = walk;
1028		walk = walk->next;
1029		kfree(pos);
1030	}
1031
1032	return err;
 
 
 
 
 
1033}
1034
1035static long do_restart_poll(struct restart_block *restart_block)
1036{
1037	struct pollfd __user *ufds = restart_block->poll.ufds;
1038	int nfds = restart_block->poll.nfds;
1039	struct timespec64 *to = NULL, end_time;
1040	int ret;
1041
1042	if (restart_block->poll.has_timeout) {
1043		end_time.tv_sec = restart_block->poll.tv_sec;
1044		end_time.tv_nsec = restart_block->poll.tv_nsec;
1045		to = &end_time;
1046	}
1047
1048	ret = do_sys_poll(ufds, nfds, to);
1049
1050	if (ret == -ERESTARTNOHAND) {
1051		restart_block->fn = do_restart_poll;
1052		ret = -ERESTART_RESTARTBLOCK;
1053	}
1054	return ret;
1055}
1056
1057SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1058		int, timeout_msecs)
1059{
1060	struct timespec64 end_time, *to = NULL;
1061	int ret;
1062
1063	if (timeout_msecs >= 0) {
1064		to = &end_time;
1065		poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1066			NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1067	}
1068
1069	ret = do_sys_poll(ufds, nfds, to);
1070
1071	if (ret == -ERESTARTNOHAND) {
1072		struct restart_block *restart_block;
1073
1074		restart_block = &current->restart_block;
1075		restart_block->fn = do_restart_poll;
1076		restart_block->poll.ufds = ufds;
1077		restart_block->poll.nfds = nfds;
1078
1079		if (timeout_msecs >= 0) {
1080			restart_block->poll.tv_sec = end_time.tv_sec;
1081			restart_block->poll.tv_nsec = end_time.tv_nsec;
1082			restart_block->poll.has_timeout = 1;
1083		} else
1084			restart_block->poll.has_timeout = 0;
1085
1086		ret = -ERESTART_RESTARTBLOCK;
1087	}
1088	return ret;
1089}
1090
1091SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1092		struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
1093		size_t, sigsetsize)
1094{
1095	struct timespec64 ts, end_time, *to = NULL;
1096	int ret;
1097
1098	if (tsp) {
1099		if (get_timespec64(&ts, tsp))
1100			return -EFAULT;
1101
1102		to = &end_time;
1103		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1104			return -EINVAL;
1105	}
1106
1107	ret = set_user_sigmask(sigmask, sigsetsize);
1108	if (ret)
1109		return ret;
1110
1111	ret = do_sys_poll(ufds, nfds, to);
1112	return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1113}
1114
1115#if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1116
1117SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1118		struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1119		size_t, sigsetsize)
1120{
1121	struct timespec64 ts, end_time, *to = NULL;
1122	int ret;
1123
1124	if (tsp) {
1125		if (get_old_timespec32(&ts, tsp))
1126			return -EFAULT;
1127
1128		to = &end_time;
1129		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1130			return -EINVAL;
1131	}
1132
1133	ret = set_user_sigmask(sigmask, sigsetsize);
1134	if (ret)
1135		return ret;
1136
1137	ret = do_sys_poll(ufds, nfds, to);
1138	return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1139}
1140#endif
1141
1142#ifdef CONFIG_COMPAT
1143#define __COMPAT_NFDBITS       (8 * sizeof(compat_ulong_t))
1144
1145/*
1146 * Ooo, nasty.  We need here to frob 32-bit unsigned longs to
1147 * 64-bit unsigned longs.
1148 */
1149static
1150int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1151			unsigned long *fdset)
1152{
1153	if (ufdset) {
1154		return compat_get_bitmap(fdset, ufdset, nr);
1155	} else {
1156		zero_fd_set(nr, fdset);
1157		return 0;
1158	}
1159}
1160
1161static
1162int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1163		      unsigned long *fdset)
1164{
1165	if (!ufdset)
1166		return 0;
1167	return compat_put_bitmap(ufdset, fdset, nr);
1168}
1169
1170
1171/*
1172 * This is a virtual copy of sys_select from fs/select.c and probably
1173 * should be compared to it from time to time
1174 */
1175
1176/*
1177 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1178 * like to be certain this leads to no problems. So I return
1179 * EINTR just for safety.
1180 *
1181 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1182 * I'm trying ERESTARTNOHAND which restart only when you want to.
1183 */
1184static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1185	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1186	struct timespec64 *end_time)
1187{
1188	fd_set_bits fds;
1189	void *bits;
1190	int size, max_fds, ret = -EINVAL;
1191	struct fdtable *fdt;
1192	long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1193
1194	if (n < 0)
1195		goto out_nofds;
1196
1197	/* max_fds can increase, so grab it once to avoid race */
1198	rcu_read_lock();
1199	fdt = files_fdtable(current->files);
1200	max_fds = fdt->max_fds;
1201	rcu_read_unlock();
1202	if (n > max_fds)
1203		n = max_fds;
1204
1205	/*
1206	 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1207	 * since we used fdset we need to allocate memory in units of
1208	 * long-words.
1209	 */
1210	size = FDS_BYTES(n);
1211	bits = stack_fds;
1212	if (size > sizeof(stack_fds) / 6) {
1213		bits = kmalloc_array(6, size, GFP_KERNEL);
1214		ret = -ENOMEM;
1215		if (!bits)
1216			goto out_nofds;
1217	}
1218	fds.in      = (unsigned long *)  bits;
1219	fds.out     = (unsigned long *) (bits +   size);
1220	fds.ex      = (unsigned long *) (bits + 2*size);
1221	fds.res_in  = (unsigned long *) (bits + 3*size);
1222	fds.res_out = (unsigned long *) (bits + 4*size);
1223	fds.res_ex  = (unsigned long *) (bits + 5*size);
1224
1225	if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1226	    (ret = compat_get_fd_set(n, outp, fds.out)) ||
1227	    (ret = compat_get_fd_set(n, exp, fds.ex)))
1228		goto out;
1229	zero_fd_set(n, fds.res_in);
1230	zero_fd_set(n, fds.res_out);
1231	zero_fd_set(n, fds.res_ex);
1232
1233	ret = do_select(n, &fds, end_time);
1234
1235	if (ret < 0)
1236		goto out;
1237	if (!ret) {
1238		ret = -ERESTARTNOHAND;
1239		if (signal_pending(current))
1240			goto out;
1241		ret = 0;
1242	}
1243
1244	if (compat_set_fd_set(n, inp, fds.res_in) ||
1245	    compat_set_fd_set(n, outp, fds.res_out) ||
1246	    compat_set_fd_set(n, exp, fds.res_ex))
1247		ret = -EFAULT;
1248out:
1249	if (bits != stack_fds)
1250		kfree(bits);
1251out_nofds:
1252	return ret;
1253}
1254
1255static int do_compat_select(int n, compat_ulong_t __user *inp,
1256	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1257	struct old_timeval32 __user *tvp)
1258{
1259	struct timespec64 end_time, *to = NULL;
1260	struct old_timeval32 tv;
1261	int ret;
1262
1263	if (tvp) {
1264		if (copy_from_user(&tv, tvp, sizeof(tv)))
1265			return -EFAULT;
1266
1267		to = &end_time;
1268		if (poll_select_set_timeout(to,
1269				tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1270				(tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1271			return -EINVAL;
1272	}
1273
1274	ret = compat_core_sys_select(n, inp, outp, exp, to);
1275	return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret);
1276}
1277
1278COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1279	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1280	struct old_timeval32 __user *, tvp)
1281{
1282	return do_compat_select(n, inp, outp, exp, tvp);
1283}
1284
1285struct compat_sel_arg_struct {
1286	compat_ulong_t n;
1287	compat_uptr_t inp;
1288	compat_uptr_t outp;
1289	compat_uptr_t exp;
1290	compat_uptr_t tvp;
1291};
1292
1293COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1294{
1295	struct compat_sel_arg_struct a;
1296
1297	if (copy_from_user(&a, arg, sizeof(a)))
1298		return -EFAULT;
1299	return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1300				compat_ptr(a.exp), compat_ptr(a.tvp));
1301}
1302
1303static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1304	compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1305	void __user *tsp, compat_sigset_t __user *sigmask,
1306	compat_size_t sigsetsize, enum poll_time_type type)
1307{
1308	struct timespec64 ts, end_time, *to = NULL;
1309	int ret;
1310
1311	if (tsp) {
1312		switch (type) {
1313		case PT_OLD_TIMESPEC:
1314			if (get_old_timespec32(&ts, tsp))
1315				return -EFAULT;
1316			break;
1317		case PT_TIMESPEC:
1318			if (get_timespec64(&ts, tsp))
1319				return -EFAULT;
1320			break;
1321		default:
1322			BUG();
1323		}
1324
1325		to = &end_time;
1326		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1327			return -EINVAL;
1328	}
1329
1330	ret = set_compat_user_sigmask(sigmask, sigsetsize);
1331	if (ret)
1332		return ret;
1333
1334	ret = compat_core_sys_select(n, inp, outp, exp, to);
1335	return poll_select_finish(&end_time, tsp, type, ret);
1336}
1337
1338struct compat_sigset_argpack {
1339	compat_uptr_t p;
1340	compat_size_t size;
1341};
1342static inline int get_compat_sigset_argpack(struct compat_sigset_argpack *to,
1343					    struct compat_sigset_argpack __user *from)
1344{
1345	if (from) {
1346		if (!user_read_access_begin(from, sizeof(*from)))
1347			return -EFAULT;
1348		unsafe_get_user(to->p, &from->p, Efault);
1349		unsafe_get_user(to->size, &from->size, Efault);
1350		user_read_access_end();
1351	}
1352	return 0;
1353Efault:
1354	user_access_end();
1355	return -EFAULT;
1356}
1357
1358COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1359	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1360	struct __kernel_timespec __user *, tsp, void __user *, sig)
1361{
1362	struct compat_sigset_argpack x = {0, 0};
1363
1364	if (get_compat_sigset_argpack(&x, sig))
1365		return -EFAULT;
1366
1367	return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1368				 x.size, PT_TIMESPEC);
1369}
1370
1371#if defined(CONFIG_COMPAT_32BIT_TIME)
1372
1373COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp,
1374	compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1375	struct old_timespec32 __user *, tsp, void __user *, sig)
1376{
1377	struct compat_sigset_argpack x = {0, 0};
1378
1379	if (get_compat_sigset_argpack(&x, sig))
1380		return -EFAULT;
1381
1382	return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(x.p),
1383				 x.size, PT_OLD_TIMESPEC);
1384}
1385
1386#endif
1387
1388#if defined(CONFIG_COMPAT_32BIT_TIME)
1389COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds,
1390	unsigned int,  nfds, struct old_timespec32 __user *, tsp,
1391	const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1392{
1393	struct timespec64 ts, end_time, *to = NULL;
1394	int ret;
1395
1396	if (tsp) {
1397		if (get_old_timespec32(&ts, tsp))
1398			return -EFAULT;
1399
1400		to = &end_time;
1401		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1402			return -EINVAL;
1403	}
1404
1405	ret = set_compat_user_sigmask(sigmask, sigsetsize);
1406	if (ret)
1407		return ret;
1408
1409	ret = do_sys_poll(ufds, nfds, to);
1410	return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1411}
1412#endif
1413
1414/* New compat syscall for 64 bit time_t*/
1415COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1416	unsigned int,  nfds, struct __kernel_timespec __user *, tsp,
1417	const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1418{
1419	struct timespec64 ts, end_time, *to = NULL;
1420	int ret;
1421
1422	if (tsp) {
1423		if (get_timespec64(&ts, tsp))
1424			return -EFAULT;
1425
1426		to = &end_time;
1427		if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1428			return -EINVAL;
1429	}
1430
1431	ret = set_compat_user_sigmask(sigmask, sigsetsize);
1432	if (ret)
1433		return ret;
1434
1435	ret = do_sys_poll(ufds, nfds, to);
1436	return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1437}
1438
1439#endif