~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/select.c

Version: ~ [ linux-5.17-rc1 ] ~ [ linux-5.16.2 ] ~ [ linux-5.15.16 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.93 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.173 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.225 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.262 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.297 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.299 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 
 52 static 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 
 75 u64 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 
 97 struct 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  */
118 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
119                        poll_table *p);
120 
121 void 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 }
130 EXPORT_SYMBOL(poll_initwait);
131 
132 static 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 
138 void 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 }
158 EXPORT_SYMBOL(poll_freewait);
159 
160 static 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 
184 static 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 
210 static 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 */
221 static 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 
236 static 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  */
273 int 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 
290 enum poll_time_type {
291         PT_TIMEVAL = 0,
292         PT_OLD_TIMEVAL = 1,
293         PT_TIMESPEC = 2,
294         PT_OLD_TIMESPEC = 3,
295 };
296 
297 static int poll_select_copy_remaining(struct timespec64 *end_time,
298                                       void __user *p,
299                                       enum poll_time_type pt_type, int ret)
300 {
301         struct timespec64 rts;
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 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 
361 sticky:
362         if (ret == -ERESTARTNOHAND)
363                 ret = -EINTR;
364         return ret;
365 }
366 
367 /*
368  * Scalable version of the fd_set.
369  */
370 
371 typedef 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  */
386 static inline
387 int 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 
397 static inline unsigned long __must_check
398 set_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 
405 static inline
406 void 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 
417 static 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;
448 get_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 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR)
461 #define POLLEX_SET (EPOLLPRI)
462 
463 static inline void wait_key_set(poll_table *wait, unsigned long in,
464                                 unsigned long out, unsigned long bit,
465                                 __poll_t ll_flag)
466 {
467         wait->_key = POLLEX_SET | ll_flag;
468         if (in & bit)
469                 wait->_key |= POLLIN_SET;
470         if (out & bit)
471                 wait->_key |= POLLOUT_SET;
472 }
473 
474 static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
475 {
476         ktime_t expire, *to = NULL;
477         struct poll_wqueues table;
478         poll_table *wait;
479         int retval, i, timed_out = 0;
480         u64 slack = 0;
481         __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
482         unsigned long busy_start = 0;
483 
484         rcu_read_lock();
485         retval = max_select_fd(n, fds);
486         rcu_read_unlock();
487 
488         if (retval < 0)
489                 return retval;
490         n = retval;
491 
492         poll_initwait(&table);
493         wait = &table.pt;
494         if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
495                 wait->_qproc = NULL;
496                 timed_out = 1;
497         }
498 
499         if (end_time && !timed_out)
500                 slack = select_estimate_accuracy(end_time);
501 
502         retval = 0;
503         for (;;) {
504                 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
505                 bool can_busy_loop = false;
506 
507                 inp = fds->in; outp = fds->out; exp = fds->ex;
508                 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
509 
510                 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
511                         unsigned long in, out, ex, all_bits, bit = 1, j;
512                         unsigned long res_in = 0, res_out = 0, res_ex = 0;
513                         __poll_t mask;
514 
515                         in = *inp++; out = *outp++; ex = *exp++;
516                         all_bits = in | out | ex;
517                         if (all_bits == 0) {
518                                 i += BITS_PER_LONG;
519                                 continue;
520                         }
521 
522                         for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
523                                 struct fd f;
524                                 if (i >= n)
525                                         break;
526                                 if (!(bit & all_bits))
527                                         continue;
528                                 f = fdget(i);
529                                 if (f.file) {
530                                         wait_key_set(wait, in, out, bit,
531                                                      busy_flag);
532                                         mask = vfs_poll(f.file, wait);
533 
534                                         fdput(f);
535                                         if ((mask & POLLIN_SET) && (in & bit)) {
536                                                 res_in |= bit;
537                                                 retval++;
538                                                 wait->_qproc = NULL;
539                                         }
540                                         if ((mask & POLLOUT_SET) && (out & bit)) {
541                                                 res_out |= bit;
542                                                 retval++;
543                                                 wait->_qproc = NULL;
544                                         }
545                                         if ((mask & POLLEX_SET) && (ex & bit)) {
546                                                 res_ex |= bit;
547                                                 retval++;
548                                                 wait->_qproc = NULL;
549                                         }
550                                         /* got something, stop busy polling */
551                                         if (retval) {
552                                                 can_busy_loop = false;
553                                                 busy_flag = 0;
554 
555                                         /*
556                                          * only remember a returned
557                                          * POLL_BUSY_LOOP if we asked for it
558                                          */
559                                         } else if (busy_flag & mask)
560                                                 can_busy_loop = true;
561 
562                                 }
563                         }
564                         if (res_in)
565                                 *rinp = res_in;
566                         if (res_out)
567                                 *routp = res_out;
568                         if (res_ex)
569                                 *rexp = res_ex;
570                         cond_resched();
571                 }
572                 wait->_qproc = NULL;
573                 if (retval || timed_out || signal_pending(current))
574                         break;
575                 if (table.error) {
576                         retval = table.error;
577                         break;
578                 }
579 
580                 /* only if found POLL_BUSY_LOOP sockets && not out of time */
581                 if (can_busy_loop && !need_resched()) {
582                         if (!busy_start) {
583                                 busy_start = busy_loop_current_time();
584                                 continue;
585                         }
586                         if (!busy_loop_timeout(busy_start))
587                                 continue;
588                 }
589                 busy_flag = 0;
590 
591                 /*
592                  * If this is the first loop and we have a timeout
593                  * given, then we convert to ktime_t and set the to
594                  * pointer to the expiry value.
595                  */
596                 if (end_time && !to) {
597                         expire = timespec64_to_ktime(*end_time);
598                         to = &expire;
599                 }
600 
601                 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
602                                            to, slack))
603                         timed_out = 1;
604         }
605 
606         poll_freewait(&table);
607 
608         return retval;
609 }
610 
611 /*
612  * We can actually return ERESTARTSYS instead of EINTR, but I'd
613  * like to be certain this leads to no problems. So I return
614  * EINTR just for safety.
615  *
616  * Update: ERESTARTSYS breaks at least the xview clock binary, so
617  * I'm trying ERESTARTNOHAND which restart only when you want to.
618  */
619 int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
620                            fd_set __user *exp, struct timespec64 *end_time)
621 {
622         fd_set_bits fds;
623         void *bits;
624         int ret, max_fds;
625         size_t size, alloc_size;
626         struct fdtable *fdt;
627         /* Allocate small arguments on the stack to save memory and be faster */
628         long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
629 
630         ret = -EINVAL;
631         if (n < 0)
632                 goto out_nofds;
633 
634         /* max_fds can increase, so grab it once to avoid race */
635         rcu_read_lock();
636         fdt = files_fdtable(current->files);
637         max_fds = fdt->max_fds;
638         rcu_read_unlock();
639         if (n > max_fds)
640                 n = max_fds;
641 
642         /*
643          * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
644          * since we used fdset we need to allocate memory in units of
645          * long-words. 
646          */
647         size = FDS_BYTES(n);
648         bits = stack_fds;
649         if (size > sizeof(stack_fds) / 6) {
650                 /* Not enough space in on-stack array; must use kmalloc */
651                 ret = -ENOMEM;
652                 if (size > (SIZE_MAX / 6))
653                         goto out_nofds;
654 
655                 alloc_size = 6 * size;
656                 bits = kvmalloc(alloc_size, GFP_KERNEL);
657                 if (!bits)
658                         goto out_nofds;
659         }
660         fds.in      = bits;
661         fds.out     = bits +   size;
662         fds.ex      = bits + 2*size;
663         fds.res_in  = bits + 3*size;
664         fds.res_out = bits + 4*size;
665         fds.res_ex  = bits + 5*size;
666 
667         if ((ret = get_fd_set(n, inp, fds.in)) ||
668             (ret = get_fd_set(n, outp, fds.out)) ||
669             (ret = get_fd_set(n, exp, fds.ex)))
670                 goto out;
671         zero_fd_set(n, fds.res_in);
672         zero_fd_set(n, fds.res_out);
673         zero_fd_set(n, fds.res_ex);
674 
675         ret = do_select(n, &fds, end_time);
676 
677         if (ret < 0)
678                 goto out;
679         if (!ret) {
680                 ret = -ERESTARTNOHAND;
681                 if (signal_pending(current))
682                         goto out;
683                 ret = 0;
684         }
685 
686         if (set_fd_set(n, inp, fds.res_in) ||
687             set_fd_set(n, outp, fds.res_out) ||
688             set_fd_set(n, exp, fds.res_ex))
689                 ret = -EFAULT;
690 
691 out:
692         if (bits != stack_fds)
693                 kvfree(bits);
694 out_nofds:
695         return ret;
696 }
697 
698 static int kern_select(int n, fd_set __user *inp, fd_set __user *outp,
699                        fd_set __user *exp, struct timeval __user *tvp)
700 {
701         struct timespec64 end_time, *to = NULL;
702         struct timeval tv;
703         int ret;
704 
705         if (tvp) {
706                 if (copy_from_user(&tv, tvp, sizeof(tv)))
707                         return -EFAULT;
708 
709                 to = &end_time;
710                 if (poll_select_set_timeout(to,
711                                 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
712                                 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
713                         return -EINVAL;
714         }
715 
716         ret = core_sys_select(n, inp, outp, exp, to);
717         ret = poll_select_copy_remaining(&end_time, tvp, PT_TIMEVAL, ret);
718 
719         return ret;
720 }
721 
722 SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
723                 fd_set __user *, exp, struct timeval __user *, tvp)
724 {
725         return kern_select(n, inp, outp, exp, tvp);
726 }
727 
728 static 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         sigset_t ksigmask, sigsaved;
734         struct timespec64 ts, end_time, *to = NULL;
735         int ret;
736 
737         if (tsp) {
738                 switch (type) {
739                 case PT_TIMESPEC:
740                         if (get_timespec64(&ts, tsp))
741                                 return -EFAULT;
742                         break;
743                 case PT_OLD_TIMESPEC:
744                         if (get_old_timespec32(&ts, tsp))
745                                 return -EFAULT;
746                         break;
747                 default:
748                         BUG();
749                 }
750 
751                 to = &end_time;
752                 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
753                         return -EINVAL;
754         }
755 
756         ret = set_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize);
757         if (ret)
758                 return ret;
759 
760         ret = core_sys_select(n, inp, outp, exp, to);
761         restore_user_sigmask(sigmask, &sigsaved, ret == -ERESTARTNOHAND);
762         ret = poll_select_copy_remaining(&end_time, tsp, type, ret);
763 
764         return ret;
765 }
766 
767 /*
768  * Most architectures can't handle 7-argument syscalls. So we provide a
769  * 6-argument version where the sixth argument is a pointer to a structure
770  * which has a pointer to the sigset_t itself followed by a size_t containing
771  * the sigset size.
772  */
773 SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
774                 fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
775                 void __user *, sig)
776 {
777         size_t sigsetsize = 0;
778         sigset_t __user *up = NULL;
779 
780         if (sig) {
781                 if (!access_ok(sig, sizeof(void *)+sizeof(size_t))
782                     || __get_user(up, (sigset_t __user * __user *)sig)
783                     || __get_user(sigsetsize,
784                                 (size_t __user *)(sig+sizeof(void *))))
785                         return -EFAULT;
786         }
787 
788         return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize, PT_TIMESPEC);
789 }
790 
791 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
792 
793 SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
794                 fd_set __user *, exp, struct old_timespec32 __user *, tsp,
795                 void __user *, sig)
796 {
797         size_t sigsetsize = 0;
798         sigset_t __user *up = NULL;
799 
800         if (sig) {
801                 if (!access_ok(sig, sizeof(void *)+sizeof(size_t))
802                     || __get_user(up, (sigset_t __user * __user *)sig)
803                     || __get_user(sigsetsize,
804                                 (size_t __user *)(sig+sizeof(void *))))
805                         return -EFAULT;
806         }
807 
808         return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize, PT_OLD_TIMESPEC);
809 }
810 
811 #endif
812 
813 #ifdef __ARCH_WANT_SYS_OLD_SELECT
814 struct sel_arg_struct {
815         unsigned long n;
816         fd_set __user *inp, *outp, *exp;
817         struct timeval __user *tvp;
818 };
819 
820 SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
821 {
822         struct sel_arg_struct a;
823 
824         if (copy_from_user(&a, arg, sizeof(a)))
825                 return -EFAULT;
826         return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
827 }
828 #endif
829 
830 struct poll_list {
831         struct poll_list *next;
832         int len;
833         struct pollfd entries[0];
834 };
835 
836 #define POLLFD_PER_PAGE  ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
837 
838 /*
839  * Fish for pollable events on the pollfd->fd file descriptor. We're only
840  * interested in events matching the pollfd->events mask, and the result
841  * matching that mask is both recorded in pollfd->revents and returned. The
842  * pwait poll_table will be used by the fd-provided poll handler for waiting,
843  * if pwait->_qproc is non-NULL.
844  */
845 static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
846                                      bool *can_busy_poll,
847                                      __poll_t busy_flag)
848 {
849         int fd = pollfd->fd;
850         __poll_t mask = 0, filter;
851         struct fd f;
852 
853         if (fd < 0)
854                 goto out;
855         mask = EPOLLNVAL;
856         f = fdget(fd);
857         if (!f.file)
858                 goto out;
859 
860         /* userland u16 ->events contains POLL... bitmap */
861         filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
862         pwait->_key = filter | busy_flag;
863         mask = vfs_poll(f.file, pwait);
864         if (mask & busy_flag)
865                 *can_busy_poll = true;
866         mask &= filter;         /* Mask out unneeded events. */
867         fdput(f);
868 
869 out:
870         /* ... and so does ->revents */
871         pollfd->revents = mangle_poll(mask);
872         return mask;
873 }
874 
875 static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
876                    struct timespec64 *end_time)
877 {
878         poll_table* pt = &wait->pt;
879         ktime_t expire, *to = NULL;
880         int timed_out = 0, count = 0;
881         u64 slack = 0;
882         __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
883         unsigned long busy_start = 0;
884 
885         /* Optimise the no-wait case */
886         if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
887                 pt->_qproc = NULL;
888                 timed_out = 1;
889         }
890 
891         if (end_time && !timed_out)
892                 slack = select_estimate_accuracy(end_time);
893 
894         for (;;) {
895                 struct poll_list *walk;
896                 bool can_busy_loop = false;
897 
898                 for (walk = list; walk != NULL; walk = walk->next) {
899                         struct pollfd * pfd, * pfd_end;
900 
901                         pfd = walk->entries;
902                         pfd_end = pfd + walk->len;
903                         for (; pfd != pfd_end; pfd++) {
904                                 /*
905                                  * Fish for events. If we found one, record it
906                                  * and kill poll_table->_qproc, so we don't
907                                  * needlessly register any other waiters after
908                                  * this. They'll get immediately deregistered
909                                  * when we break out and return.
910                                  */
911                                 if (do_pollfd(pfd, pt, &can_busy_loop,
912                                               busy_flag)) {
913                                         count++;
914                                         pt->_qproc = NULL;
915                                         /* found something, stop busy polling */
916                                         busy_flag = 0;
917                                         can_busy_loop = false;
918                                 }
919                         }
920                 }
921                 /*
922                  * All waiters have already been registered, so don't provide
923                  * a poll_table->_qproc to them on the next loop iteration.
924                  */
925                 pt->_qproc = NULL;
926                 if (!count) {
927                         count = wait->error;
928                         if (signal_pending(current))
929                                 count = -EINTR;
930                 }
931                 if (count || timed_out)
932                         break;
933 
934                 /* only if found POLL_BUSY_LOOP sockets && not out of time */
935                 if (can_busy_loop && !need_resched()) {
936                         if (!busy_start) {
937                                 busy_start = busy_loop_current_time();
938                                 continue;
939                         }
940                         if (!busy_loop_timeout(busy_start))
941                                 continue;
942                 }
943                 busy_flag = 0;
944 
945                 /*
946                  * If this is the first loop and we have a timeout
947                  * given, then we convert to ktime_t and set the to
948                  * pointer to the expiry value.
949                  */
950                 if (end_time && !to) {
951                         expire = timespec64_to_ktime(*end_time);
952                         to = &expire;
953                 }
954 
955                 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
956                         timed_out = 1;
957         }
958         return count;
959 }
960 
961 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list))  / \
962                         sizeof(struct pollfd))
963 
964 static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
965                 struct timespec64 *end_time)
966 {
967         struct poll_wqueues table;
968         int err = -EFAULT, fdcount, len, size;
969         /* Allocate small arguments on the stack to save memory and be
970            faster - use long to make sure the buffer is aligned properly
971            on 64 bit archs to avoid unaligned access */
972         long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
973         struct poll_list *const head = (struct poll_list *)stack_pps;
974         struct poll_list *walk = head;
975         unsigned long todo = nfds;
976 
977         if (nfds > rlimit(RLIMIT_NOFILE))
978                 return -EINVAL;
979 
980         len = min_t(unsigned int, nfds, N_STACK_PPS);
981         for (;;) {
982                 walk->next = NULL;
983                 walk->len = len;
984                 if (!len)
985                         break;
986 
987                 if (copy_from_user(walk->entries, ufds + nfds-todo,
988                                         sizeof(struct pollfd) * walk->len))
989                         goto out_fds;
990 
991                 todo -= walk->len;
992                 if (!todo)
993                         break;
994 
995                 len = min(todo, POLLFD_PER_PAGE);
996                 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
997                 walk = walk->next = kmalloc(size, GFP_KERNEL);
998                 if (!walk) {
999                         err = -ENOMEM;
1000                         goto out_fds;
1001                 }
1002         }
1003 
1004         poll_initwait(&table);
1005         fdcount = do_poll(head, &table, end_time);
1006         poll_freewait(&table);
1007 
1008         for (walk = head; walk; walk = walk->next) {
1009                 struct pollfd *fds = walk->entries;
1010                 int j;
1011 
1012                 for (j = 0; j < walk->len; j++, ufds++)
1013                         if (__put_user(fds[j].revents, &ufds->revents))
1014                                 goto out_fds;
1015         }
1016 
1017         err = fdcount;
1018 out_fds:
1019         walk = head->next;
1020         while (walk) {
1021                 struct poll_list *pos = walk;
1022                 walk = walk->next;
1023                 kfree(pos);
1024         }
1025 
1026         return err;
1027 }
1028 
1029 static long do_restart_poll(struct restart_block *restart_block)
1030 {
1031         struct pollfd __user *ufds = restart_block->poll.ufds;
1032         int nfds = restart_block->poll.nfds;
1033         struct timespec64 *to = NULL, end_time;
1034         int ret;
1035 
1036         if (restart_block->poll.has_timeout) {
1037                 end_time.tv_sec = restart_block->poll.tv_sec;
1038                 end_time.tv_nsec = restart_block->poll.tv_nsec;
1039                 to = &end_time;
1040         }
1041 
1042         ret = do_sys_poll(ufds, nfds, to);
1043 
1044         if (ret == -EINTR) {
1045                 restart_block->fn = do_restart_poll;
1046                 ret = -ERESTART_RESTARTBLOCK;
1047         }
1048         return ret;
1049 }
1050 
1051 SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1052                 int, timeout_msecs)
1053 {
1054         struct timespec64 end_time, *to = NULL;
1055         int ret;
1056 
1057         if (timeout_msecs >= 0) {
1058                 to = &end_time;
1059                 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1060                         NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1061         }
1062 
1063         ret = do_sys_poll(ufds, nfds, to);
1064 
1065         if (ret == -EINTR) {
1066                 struct restart_block *restart_block;
1067 
1068                 restart_block = &current->restart_block;
1069                 restart_block->fn = do_restart_poll;
1070                 restart_block->poll.ufds = ufds;
1071                 restart_block->poll.nfds = nfds;
1072 
1073                 if (timeout_msecs >= 0) {
1074                         restart_block->poll.tv_sec = end_time.tv_sec;
1075                         restart_block->poll.tv_nsec = end_time.tv_nsec;
1076                         restart_block->poll.has_timeout = 1;
1077                 } else
1078                         restart_block->poll.has_timeout = 0;
1079 
1080                 ret = -ERESTART_RESTARTBLOCK;
1081         }
1082         return ret;
1083 }
1084 
1085 SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1086                 struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
1087                 size_t, sigsetsize)
1088 {
1089         sigset_t ksigmask, sigsaved;
1090         struct timespec64 ts, end_time, *to = NULL;
1091         int ret;
1092 
1093         if (tsp) {
1094                 if (get_timespec64(&ts, tsp))
1095                         return -EFAULT;
1096 
1097                 to = &end_time;
1098                 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1099                         return -EINVAL;
1100         }
1101 
1102         ret = set_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize);
1103         if (ret)
1104                 return ret;
1105 
1106         ret = do_sys_poll(ufds, nfds, to);
1107 
1108         restore_user_sigmask(sigmask, &sigsaved, ret == -EINTR);
1109         /* We can restart this syscall, usually */
1110         if (ret == -EINTR)
1111                 ret = -ERESTARTNOHAND;
1112 
1113         ret = poll_select_copy_remaining(&end_time, tsp, PT_TIMESPEC, ret);
1114 
1115         return ret;
1116 }
1117 
1118 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1119 
1120 SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1121                 struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1122                 size_t, sigsetsize)
1123 {
1124         sigset_t ksigmask, sigsaved;
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, &ksigmask, &sigsaved, sigsetsize);
1138         if (ret)
1139                 return ret;
1140 
1141         ret = do_sys_poll(ufds, nfds, to);
1142 
1143         restore_user_sigmask(sigmask, &sigsaved, ret == -EINTR);
1144         /* We can restart this syscall, usually */
1145         if (ret == -EINTR)
1146                 ret = -ERESTARTNOHAND;
1147 
1148         ret = poll_select_copy_remaining(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1149 
1150         return ret;
1151 }
1152 #endif
1153 
1154 #ifdef CONFIG_COMPAT
1155 #define __COMPAT_NFDBITS       (8 * sizeof(compat_ulong_t))
1156 
1157 /*
1158  * Ooo, nasty.  We need here to frob 32-bit unsigned longs to
1159  * 64-bit unsigned longs.
1160  */
1161 static
1162 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1163                         unsigned long *fdset)
1164 {
1165         if (ufdset) {
1166                 return compat_get_bitmap(fdset, ufdset, nr);
1167         } else {
1168                 zero_fd_set(nr, fdset);
1169                 return 0;
1170         }
1171 }
1172 
1173 static
1174 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1175                       unsigned long *fdset)
1176 {
1177         if (!ufdset)
1178                 return 0;
1179         return compat_put_bitmap(ufdset, fdset, nr);
1180 }
1181 
1182 
1183 /*
1184  * This is a virtual copy of sys_select from fs/select.c and probably
1185  * should be compared to it from time to time
1186  */
1187 
1188 /*
1189  * We can actually return ERESTARTSYS instead of EINTR, but I'd
1190  * like to be certain this leads to no problems. So I return
1191  * EINTR just for safety.
1192  *
1193  * Update: ERESTARTSYS breaks at least the xview clock binary, so
1194  * I'm trying ERESTARTNOHAND which restart only when you want to.
1195  */
1196 static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1197         compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1198         struct timespec64 *end_time)
1199 {
1200         fd_set_bits fds;
1201         void *bits;
1202         int size, max_fds, ret = -EINVAL;
1203         struct fdtable *fdt;
1204         long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1205 
1206         if (n < 0)
1207                 goto out_nofds;
1208 
1209         /* max_fds can increase, so grab it once to avoid race */
1210         rcu_read_lock();
1211         fdt = files_fdtable(current->files);
1212         max_fds = fdt->max_fds;
1213         rcu_read_unlock();
1214         if (n > max_fds)
1215                 n = max_fds;
1216 
1217         /*
1218          * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1219          * since we used fdset we need to allocate memory in units of
1220          * long-words.
1221          */
1222         size = FDS_BYTES(n);
1223         bits = stack_fds;
1224         if (size > sizeof(stack_fds) / 6) {
1225                 bits = kmalloc_array(6, size, GFP_KERNEL);
1226                 ret = -ENOMEM;
1227                 if (!bits)
1228                         goto out_nofds;
1229         }
1230         fds.in      = (unsigned long *)  bits;
1231         fds.out     = (unsigned long *) (bits +   size);
1232         fds.ex      = (unsigned long *) (bits + 2*size);
1233         fds.res_in  = (unsigned long *) (bits + 3*size);
1234         fds.res_out = (unsigned long *) (bits + 4*size);
1235         fds.res_ex  = (unsigned long *) (bits + 5*size);
1236 
1237         if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1238             (ret = compat_get_fd_set(n, outp, fds.out)) ||
1239             (ret = compat_get_fd_set(n, exp, fds.ex)))
1240                 goto out;
1241         zero_fd_set(n, fds.res_in);
1242         zero_fd_set(n, fds.res_out);
1243         zero_fd_set(n, fds.res_ex);
1244 
1245         ret = do_select(n, &fds, end_time);
1246 
1247         if (ret < 0)
1248                 goto out;
1249         if (!ret) {
1250                 ret = -ERESTARTNOHAND;
1251                 if (signal_pending(current))
1252                         goto out;
1253                 ret = 0;
1254         }
1255 
1256         if (compat_set_fd_set(n, inp, fds.res_in) ||
1257             compat_set_fd_set(n, outp, fds.res_out) ||
1258             compat_set_fd_set(n, exp, fds.res_ex))
1259                 ret = -EFAULT;
1260 out:
1261         if (bits != stack_fds)
1262                 kfree(bits);
1263 out_nofds:
1264         return ret;
1265 }
1266 
1267 static int do_compat_select(int n, compat_ulong_t __user *inp,
1268         compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1269         struct old_timeval32 __user *tvp)
1270 {
1271         struct timespec64 end_time, *to = NULL;
1272         struct old_timeval32 tv;
1273         int ret;
1274 
1275         if (tvp) {
1276                 if (copy_from_user(&tv, tvp, sizeof(tv)))
1277                         return -EFAULT;
1278 
1279                 to = &end_time;
1280                 if (poll_select_set_timeout(to,
1281                                 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1282                                 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1283                         return -EINVAL;
1284         }
1285 
1286         ret = compat_core_sys_select(n, inp, outp, exp, to);
1287         ret = poll_select_copy_remaining(&end_time, tvp, PT_OLD_TIMEVAL, ret);
1288 
1289         return ret;
1290 }
1291 
1292 COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1293         compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1294         struct old_timeval32 __user *, tvp)
1295 {
1296         return do_compat_select(n, inp, outp, exp, tvp);
1297 }
1298 
1299 struct compat_sel_arg_struct {
1300         compat_ulong_t n;
1301         compat_uptr_t inp;
1302         compat_uptr_t outp;
1303         compat_uptr_t exp;
1304         compat_uptr_t tvp;
1305 };
1306 
1307 COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1308 {
1309         struct compat_sel_arg_struct a;
1310 
1311         if (copy_from_user(&a, arg, sizeof(a)))
1312                 return -EFAULT;
1313         return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1314                                 compat_ptr(a.exp), compat_ptr(a.tvp));
1315 }
1316 
1317 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1318         compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1319         void __user *tsp, compat_sigset_t __user *sigmask,
1320         compat_size_t sigsetsize, enum poll_time_type type)
1321 {
1322         sigset_t ksigmask, sigsaved;
1323         struct timespec64 ts, end_time, *to = NULL;
1324         int ret;
1325 
1326         if (tsp) {
1327                 switch (type) {
1328                 case PT_OLD_TIMESPEC:
1329                         if (get_old_timespec32(&ts, tsp))
1330                                 return -EFAULT;
1331                         break;
1332                 case PT_TIMESPEC:
1333                         if (get_timespec64(&ts, tsp))
1334                                 return -EFAULT;
1335                         break;
1336                 default:
1337                         BUG();
1338                 }
1339 
1340                 to = &end_time;
1341                 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1342                         return -EINVAL;
1343         }
1344 
1345         ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize);
1346         if (ret)
1347                 return ret;
1348 
1349         ret = compat_core_sys_select(n, inp, outp, exp, to);
1350         restore_user_sigmask(sigmask, &sigsaved, ret == -ERESTARTNOHAND);
1351         ret = poll_select_copy_remaining(&end_time, tsp, type, ret);
1352 
1353         return ret;
1354 }
1355 
1356 COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1357         compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1358         struct __kernel_timespec __user *, tsp, void __user *, sig)
1359 {
1360         compat_size_t sigsetsize = 0;
1361         compat_uptr_t up = 0;
1362 
1363         if (sig) {
1364                 if (!access_ok(sig,
1365                                 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1366                                 __get_user(up, (compat_uptr_t __user *)sig) ||
1367                                 __get_user(sigsetsize,
1368                                 (compat_size_t __user *)(sig+sizeof(up))))
1369                         return -EFAULT;
1370         }
1371 
1372         return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1373                                  sigsetsize, PT_TIMESPEC);
1374 }
1375 
1376 #if defined(CONFIG_COMPAT_32BIT_TIME)
1377 
1378 COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp,
1379         compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1380         struct old_timespec32 __user *, tsp, void __user *, sig)
1381 {
1382         compat_size_t sigsetsize = 0;
1383         compat_uptr_t up = 0;
1384 
1385         if (sig) {
1386                 if (!access_ok(sig,
1387                                 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1388                         __get_user(up, (compat_uptr_t __user *)sig) ||
1389                         __get_user(sigsetsize,
1390                                 (compat_size_t __user *)(sig+sizeof(up))))
1391                         return -EFAULT;
1392         }
1393 
1394         return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1395                                  sigsetsize, PT_OLD_TIMESPEC);
1396 }
1397 
1398 #endif
1399 
1400 #if defined(CONFIG_COMPAT_32BIT_TIME)
1401 COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds,
1402         unsigned int,  nfds, struct old_timespec32 __user *, tsp,
1403         const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1404 {
1405         sigset_t ksigmask, sigsaved;
1406         struct timespec64 ts, end_time, *to = NULL;
1407         int ret;
1408 
1409         if (tsp) {
1410                 if (get_old_timespec32(&ts, tsp))
1411                         return -EFAULT;
1412 
1413                 to = &end_time;
1414                 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1415                         return -EINVAL;
1416         }
1417 
1418         ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize);
1419         if (ret)
1420                 return ret;
1421 
1422         ret = do_sys_poll(ufds, nfds, to);
1423 
1424         restore_user_sigmask(sigmask, &sigsaved, ret == -EINTR);
1425         /* We can restart this syscall, usually */
1426         if (ret == -EINTR)
1427                 ret = -ERESTARTNOHAND;
1428 
1429         ret = poll_select_copy_remaining(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1430 
1431         return ret;
1432 }
1433 #endif
1434 
1435 /* New compat syscall for 64 bit time_t*/
1436 COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1437         unsigned int,  nfds, struct __kernel_timespec __user *, tsp,
1438         const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1439 {
1440         sigset_t ksigmask, sigsaved;
1441         struct timespec64 ts, end_time, *to = NULL;
1442         int ret;
1443 
1444         if (tsp) {
1445                 if (get_timespec64(&ts, tsp))
1446                         return -EFAULT;
1447 
1448                 to = &end_time;
1449                 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1450                         return -EINVAL;
1451         }
1452 
1453         ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize);
1454         if (ret)
1455                 return ret;
1456 
1457         ret = do_sys_poll(ufds, nfds, to);
1458 
1459         restore_user_sigmask(sigmask, &sigsaved, ret == -EINTR);
1460         /* We can restart this syscall, usually */
1461         if (ret == -EINTR)
1462                 ret = -ERESTARTNOHAND;
1463 
1464         ret = poll_select_copy_remaining(&end_time, tsp, PT_TIMESPEC, ret);
1465 
1466         return ret;
1467 }
1468 
1469 #endif
1470 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp