Root/fs/select.c

1/*
2 * This file contains the procedures for the handling of select and poll
3 *
4 * Created for Linux based loosely upon Mathius Lattner's minix
5 * patches by Peter MacDonald. Heavily edited by Linus.
6 *
7 * 4 February 1994
8 * COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS
9 * flag set in its personality we do *not* modify the given timeout
10 * parameter to reflect time remaining.
11 *
12 * 24 January 2000
13 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation
14 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
15 */
16
17#include <linux/kernel.h>
18#include <linux/syscalls.h>
19#include <linux/module.h>
20#include <linux/slab.h>
21#include <linux/poll.h>
22#include <linux/personality.h> /* for STICKY_TIMEOUTS */
23#include <linux/file.h>
24#include <linux/fdtable.h>
25#include <linux/fs.h>
26#include <linux/rcupdate.h>
27#include <linux/hrtimer.h>
28
29#include <asm/uaccess.h>
30
31
32/*
33 * Estimate expected accuracy in ns from a timeval.
34 *
35 * After quite a bit of churning around, we've settled on
36 * a simple thing of taking 0.1% of the timeout as the
37 * slack, with a cap of 100 msec.
38 * "nice" tasks get a 0.5% slack instead.
39 *
40 * Consider this comment an open invitation to come up with even
41 * better solutions..
42 */
43
44static long __estimate_accuracy(struct timespec *tv)
45{
46    long slack;
47    int divfactor = 1000;
48
49    if (task_nice(current) > 0)
50        divfactor = divfactor / 5;
51
52    slack = tv->tv_nsec / divfactor;
53    slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
54
55    if (slack > 100 * NSEC_PER_MSEC)
56        slack = 100 * NSEC_PER_MSEC;
57
58    if (slack < 0)
59        slack = 0;
60    return slack;
61}
62
63static long estimate_accuracy(struct timespec *tv)
64{
65    unsigned long ret;
66    struct timespec now;
67
68    /*
69     * Realtime tasks get a slack of 0 for obvious reasons.
70     */
71
72    if (rt_task(current))
73        return 0;
74
75    ktime_get_ts(&now);
76    now = timespec_sub(*tv, now);
77    ret = __estimate_accuracy(&now);
78    if (ret < current->timer_slack_ns)
79        return current->timer_slack_ns;
80    return ret;
81}
82
83
84
85struct poll_table_page {
86    struct poll_table_page * next;
87    struct poll_table_entry * entry;
88    struct poll_table_entry entries[0];
89};
90
91#define POLL_TABLE_FULL(table) \
92    ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
93
94/*
95 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
96 * I have rewritten this, taking some shortcuts: This code may not be easy to
97 * follow, but it should be free of race-conditions, and it's practical. If you
98 * understand what I'm doing here, then you understand how the linux
99 * sleep/wakeup mechanism works.
100 *
101 * Two very simple procedures, poll_wait() and poll_freewait() make all the
102 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
103 * as all select/poll functions have to call it to add an entry to the
104 * poll table.
105 */
106static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
107               poll_table *p);
108
109void poll_initwait(struct poll_wqueues *pwq)
110{
111    init_poll_funcptr(&pwq->pt, __pollwait);
112    pwq->polling_task = current;
113    pwq->triggered = 0;
114    pwq->error = 0;
115    pwq->table = NULL;
116    pwq->inline_index = 0;
117}
118EXPORT_SYMBOL(poll_initwait);
119
120static void free_poll_entry(struct poll_table_entry *entry)
121{
122    remove_wait_queue(entry->wait_address, &entry->wait);
123    fput(entry->filp);
124}
125
126void poll_freewait(struct poll_wqueues *pwq)
127{
128    struct poll_table_page * p = pwq->table;
129    int i;
130    for (i = 0; i < pwq->inline_index; i++)
131        free_poll_entry(pwq->inline_entries + i);
132    while (p) {
133        struct poll_table_entry * entry;
134        struct poll_table_page *old;
135
136        entry = p->entry;
137        do {
138            entry--;
139            free_poll_entry(entry);
140        } while (entry > p->entries);
141        old = p;
142        p = p->next;
143        free_page((unsigned long) old);
144    }
145}
146EXPORT_SYMBOL(poll_freewait);
147
148static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
149{
150    struct poll_table_page *table = p->table;
151
152    if (p->inline_index < N_INLINE_POLL_ENTRIES)
153        return p->inline_entries + p->inline_index++;
154
155    if (!table || POLL_TABLE_FULL(table)) {
156        struct poll_table_page *new_table;
157
158        new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
159        if (!new_table) {
160            p->error = -ENOMEM;
161            return NULL;
162        }
163        new_table->entry = new_table->entries;
164        new_table->next = table;
165        p->table = new_table;
166        table = new_table;
167    }
168
169    return table->entry++;
170}
171
172static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
173{
174    struct poll_wqueues *pwq = wait->private;
175    DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
176
177    /*
178     * Although this function is called under waitqueue lock, LOCK
179     * doesn't imply write barrier and the users expect write
180     * barrier semantics on wakeup functions. The following
181     * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
182     * and is paired with set_mb() in poll_schedule_timeout.
183     */
184    smp_wmb();
185    pwq->triggered = 1;
186
187    /*
188     * Perform the default wake up operation using a dummy
189     * waitqueue.
190     *
191     * TODO: This is hacky but there currently is no interface to
192     * pass in @sync. @sync is scheduled to be removed and once
193     * that happens, wake_up_process() can be used directly.
194     */
195    return default_wake_function(&dummy_wait, mode, sync, key);
196}
197
198static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
199{
200    struct poll_table_entry *entry;
201
202    entry = container_of(wait, struct poll_table_entry, wait);
203    if (key && !((unsigned long)key & entry->key))
204        return 0;
205    return __pollwake(wait, mode, sync, key);
206}
207
208/* Add a new entry */
209static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
210                poll_table *p)
211{
212    struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
213    struct poll_table_entry *entry = poll_get_entry(pwq);
214    if (!entry)
215        return;
216    get_file(filp);
217    entry->filp = filp;
218    entry->wait_address = wait_address;
219    entry->key = p->key;
220    init_waitqueue_func_entry(&entry->wait, pollwake);
221    entry->wait.private = pwq;
222    add_wait_queue(wait_address, &entry->wait);
223}
224
225int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
226              ktime_t *expires, unsigned long slack)
227{
228    int rc = -EINTR;
229
230    set_current_state(state);
231    if (!pwq->triggered)
232        rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
233    __set_current_state(TASK_RUNNING);
234
235    /*
236     * Prepare for the next iteration.
237     *
238     * The following set_mb() serves two purposes. First, it's
239     * the counterpart rmb of the wmb in pollwake() such that data
240     * written before wake up is always visible after wake up.
241     * Second, the full barrier guarantees that triggered clearing
242     * doesn't pass event check of the next iteration. Note that
243     * this problem doesn't exist for the first iteration as
244     * add_wait_queue() has full barrier semantics.
245     */
246    set_mb(pwq->triggered, 0);
247
248    return rc;
249}
250EXPORT_SYMBOL(poll_schedule_timeout);
251
252/**
253 * poll_select_set_timeout - helper function to setup the timeout value
254 * @to: pointer to timespec variable for the final timeout
255 * @sec: seconds (from user space)
256 * @nsec: nanoseconds (from user space)
257 *
258 * Note, we do not use a timespec for the user space value here, That
259 * way we can use the function for timeval and compat interfaces as well.
260 *
261 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
262 */
263int poll_select_set_timeout(struct timespec *to, long sec, long nsec)
264{
265    struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec};
266
267    if (!timespec_valid(&ts))
268        return -EINVAL;
269
270    /* Optimize for the zero timeout value here */
271    if (!sec && !nsec) {
272        to->tv_sec = to->tv_nsec = 0;
273    } else {
274        ktime_get_ts(to);
275        *to = timespec_add_safe(*to, ts);
276    }
277    return 0;
278}
279
280static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
281                      int timeval, int ret)
282{
283    struct timespec rts;
284    struct timeval rtv;
285
286    if (!p)
287        return ret;
288
289    if (current->personality & STICKY_TIMEOUTS)
290        goto sticky;
291
292    /* No update for zero timeout */
293    if (!end_time->tv_sec && !end_time->tv_nsec)
294        return ret;
295
296    ktime_get_ts(&rts);
297    rts = timespec_sub(*end_time, rts);
298    if (rts.tv_sec < 0)
299        rts.tv_sec = rts.tv_nsec = 0;
300
301    if (timeval) {
302        rtv.tv_sec = rts.tv_sec;
303        rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
304
305        if (!copy_to_user(p, &rtv, sizeof(rtv)))
306            return ret;
307
308    } else if (!copy_to_user(p, &rts, sizeof(rts)))
309        return ret;
310
311    /*
312     * If an application puts its timeval in read-only memory, we
313     * don't want the Linux-specific update to the timeval to
314     * cause a fault after the select has completed
315     * successfully. However, because we're not updating the
316     * timeval, we can't restart the system call.
317     */
318
319sticky:
320    if (ret == -ERESTARTNOHAND)
321        ret = -EINTR;
322    return ret;
323}
324
325#define FDS_IN(fds, n) (fds->in + n)
326#define FDS_OUT(fds, n) (fds->out + n)
327#define FDS_EX(fds, n) (fds->ex + n)
328
329#define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
330
331static int max_select_fd(unsigned long n, fd_set_bits *fds)
332{
333    unsigned long *open_fds;
334    unsigned long set;
335    int max;
336    struct fdtable *fdt;
337
338    /* handle last in-complete long-word first */
339    set = ~(~0UL << (n & (__NFDBITS-1)));
340    n /= __NFDBITS;
341    fdt = files_fdtable(current->files);
342    open_fds = fdt->open_fds->fds_bits+n;
343    max = 0;
344    if (set) {
345        set &= BITS(fds, n);
346        if (set) {
347            if (!(set & ~*open_fds))
348                goto get_max;
349            return -EBADF;
350        }
351    }
352    while (n) {
353        open_fds--;
354        n--;
355        set = BITS(fds, n);
356        if (!set)
357            continue;
358        if (set & ~*open_fds)
359            return -EBADF;
360        if (max)
361            continue;
362get_max:
363        do {
364            max++;
365            set >>= 1;
366        } while (set);
367        max += n * __NFDBITS;
368    }
369
370    return max;
371}
372
373#define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
374#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
375#define POLLEX_SET (POLLPRI)
376
377static inline void wait_key_set(poll_table *wait, unsigned long in,
378                unsigned long out, unsigned long bit)
379{
380    if (wait) {
381        wait->key = POLLEX_SET;
382        if (in & bit)
383            wait->key |= POLLIN_SET;
384        if (out & bit)
385            wait->key |= POLLOUT_SET;
386    }
387}
388
389int do_select(int n, fd_set_bits *fds, struct timespec *end_time)
390{
391    ktime_t expire, *to = NULL;
392    struct poll_wqueues table;
393    poll_table *wait;
394    int retval, i, timed_out = 0;
395    unsigned long slack = 0;
396
397    rcu_read_lock();
398    retval = max_select_fd(n, fds);
399    rcu_read_unlock();
400
401    if (retval < 0)
402        return retval;
403    n = retval;
404
405    poll_initwait(&table);
406    wait = &table.pt;
407    if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
408        wait = NULL;
409        timed_out = 1;
410    }
411
412    if (end_time && !timed_out)
413        slack = estimate_accuracy(end_time);
414
415    retval = 0;
416    for (;;) {
417        unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
418
419        inp = fds->in; outp = fds->out; exp = fds->ex;
420        rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
421
422        for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
423            unsigned long in, out, ex, all_bits, bit = 1, mask, j;
424            unsigned long res_in = 0, res_out = 0, res_ex = 0;
425            const struct file_operations *f_op = NULL;
426            struct file *file = NULL;
427
428            in = *inp++; out = *outp++; ex = *exp++;
429            all_bits = in | out | ex;
430            if (all_bits == 0) {
431                i += __NFDBITS;
432                continue;
433            }
434
435            for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
436                int fput_needed;
437                if (i >= n)
438                    break;
439                if (!(bit & all_bits))
440                    continue;
441                file = fget_light(i, &fput_needed);
442                if (file) {
443                    f_op = file->f_op;
444                    mask = DEFAULT_POLLMASK;
445                    if (f_op && f_op->poll) {
446                        wait_key_set(wait, in, out, bit);
447                        mask = (*f_op->poll)(file, wait);
448                    }
449                    fput_light(file, fput_needed);
450                    if ((mask & POLLIN_SET) && (in & bit)) {
451                        res_in |= bit;
452                        retval++;
453                        wait = NULL;
454                    }
455                    if ((mask & POLLOUT_SET) && (out & bit)) {
456                        res_out |= bit;
457                        retval++;
458                        wait = NULL;
459                    }
460                    if ((mask & POLLEX_SET) && (ex & bit)) {
461                        res_ex |= bit;
462                        retval++;
463                        wait = NULL;
464                    }
465                }
466            }
467            if (res_in)
468                *rinp = res_in;
469            if (res_out)
470                *routp = res_out;
471            if (res_ex)
472                *rexp = res_ex;
473            cond_resched();
474        }
475        wait = NULL;
476        if (retval || timed_out || signal_pending(current))
477            break;
478        if (table.error) {
479            retval = table.error;
480            break;
481        }
482
483        /*
484         * If this is the first loop and we have a timeout
485         * given, then we convert to ktime_t and set the to
486         * pointer to the expiry value.
487         */
488        if (end_time && !to) {
489            expire = timespec_to_ktime(*end_time);
490            to = &expire;
491        }
492
493        if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
494                       to, slack))
495            timed_out = 1;
496    }
497
498    poll_freewait(&table);
499
500    return retval;
501}
502
503/*
504 * We can actually return ERESTARTSYS instead of EINTR, but I'd
505 * like to be certain this leads to no problems. So I return
506 * EINTR just for safety.
507 *
508 * Update: ERESTARTSYS breaks at least the xview clock binary, so
509 * I'm trying ERESTARTNOHAND which restart only when you want to.
510 */
511#define MAX_SELECT_SECONDS \
512    ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
513
514int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
515               fd_set __user *exp, struct timespec *end_time)
516{
517    fd_set_bits fds;
518    void *bits;
519    int ret, max_fds;
520    unsigned int size;
521    struct fdtable *fdt;
522    /* Allocate small arguments on the stack to save memory and be faster */
523    long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
524
525    ret = -EINVAL;
526    if (n < 0)
527        goto out_nofds;
528
529    /* max_fds can increase, so grab it once to avoid race */
530    rcu_read_lock();
531    fdt = files_fdtable(current->files);
532    max_fds = fdt->max_fds;
533    rcu_read_unlock();
534    if (n > max_fds)
535        n = max_fds;
536
537    /*
538     * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
539     * since we used fdset we need to allocate memory in units of
540     * long-words.
541     */
542    size = FDS_BYTES(n);
543    bits = stack_fds;
544    if (size > sizeof(stack_fds) / 6) {
545        /* Not enough space in on-stack array; must use kmalloc */
546        ret = -ENOMEM;
547        bits = kmalloc(6 * size, GFP_KERNEL);
548        if (!bits)
549            goto out_nofds;
550    }
551    fds.in = bits;
552    fds.out = bits + size;
553    fds.ex = bits + 2*size;
554    fds.res_in = bits + 3*size;
555    fds.res_out = bits + 4*size;
556    fds.res_ex = bits + 5*size;
557
558    if ((ret = get_fd_set(n, inp, fds.in)) ||
559        (ret = get_fd_set(n, outp, fds.out)) ||
560        (ret = get_fd_set(n, exp, fds.ex)))
561        goto out;
562    zero_fd_set(n, fds.res_in);
563    zero_fd_set(n, fds.res_out);
564    zero_fd_set(n, fds.res_ex);
565
566    ret = do_select(n, &fds, end_time);
567
568    if (ret < 0)
569        goto out;
570    if (!ret) {
571        ret = -ERESTARTNOHAND;
572        if (signal_pending(current))
573            goto out;
574        ret = 0;
575    }
576
577    if (set_fd_set(n, inp, fds.res_in) ||
578        set_fd_set(n, outp, fds.res_out) ||
579        set_fd_set(n, exp, fds.res_ex))
580        ret = -EFAULT;
581
582out:
583    if (bits != stack_fds)
584        kfree(bits);
585out_nofds:
586    return ret;
587}
588
589SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
590        fd_set __user *, exp, struct timeval __user *, tvp)
591{
592    struct timespec end_time, *to = NULL;
593    struct timeval tv;
594    int ret;
595
596    if (tvp) {
597        if (copy_from_user(&tv, tvp, sizeof(tv)))
598            return -EFAULT;
599
600        to = &end_time;
601        if (poll_select_set_timeout(to,
602                tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
603                (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
604            return -EINVAL;
605    }
606
607    ret = core_sys_select(n, inp, outp, exp, to);
608    ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
609
610    return ret;
611}
612
613#ifdef HAVE_SET_RESTORE_SIGMASK
614static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
615               fd_set __user *exp, struct timespec __user *tsp,
616               const sigset_t __user *sigmask, size_t sigsetsize)
617{
618    sigset_t ksigmask, sigsaved;
619    struct timespec ts, end_time, *to = NULL;
620    int ret;
621
622    if (tsp) {
623        if (copy_from_user(&ts, tsp, sizeof(ts)))
624            return -EFAULT;
625
626        to = &end_time;
627        if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
628            return -EINVAL;
629    }
630
631    if (sigmask) {
632        /* XXX: Don't preclude handling different sized sigset_t's. */
633        if (sigsetsize != sizeof(sigset_t))
634            return -EINVAL;
635        if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
636            return -EFAULT;
637
638        sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
639        sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
640    }
641
642    ret = core_sys_select(n, inp, outp, exp, to);
643    ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
644
645    if (ret == -ERESTARTNOHAND) {
646        /*
647         * Don't restore the signal mask yet. Let do_signal() deliver
648         * the signal on the way back to userspace, before the signal
649         * mask is restored.
650         */
651        if (sigmask) {
652            memcpy(&current->saved_sigmask, &sigsaved,
653                    sizeof(sigsaved));
654            set_restore_sigmask();
655        }
656    } else if (sigmask)
657        sigprocmask(SIG_SETMASK, &sigsaved, NULL);
658
659    return ret;
660}
661
662/*
663 * Most architectures can't handle 7-argument syscalls. So we provide a
664 * 6-argument version where the sixth argument is a pointer to a structure
665 * which has a pointer to the sigset_t itself followed by a size_t containing
666 * the sigset size.
667 */
668SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
669        fd_set __user *, exp, struct timespec __user *, tsp,
670        void __user *, sig)
671{
672    size_t sigsetsize = 0;
673    sigset_t __user *up = NULL;
674
675    if (sig) {
676        if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
677            || __get_user(up, (sigset_t __user * __user *)sig)
678            || __get_user(sigsetsize,
679                (size_t __user *)(sig+sizeof(void *))))
680            return -EFAULT;
681    }
682
683    return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
684}
685#endif /* HAVE_SET_RESTORE_SIGMASK */
686
687struct poll_list {
688    struct poll_list *next;
689    int len;
690    struct pollfd entries[0];
691};
692
693#define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
694
695/*
696 * Fish for pollable events on the pollfd->fd file descriptor. We're only
697 * interested in events matching the pollfd->events mask, and the result
698 * matching that mask is both recorded in pollfd->revents and returned. The
699 * pwait poll_table will be used by the fd-provided poll handler for waiting,
700 * if non-NULL.
701 */
702static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
703{
704    unsigned int mask;
705    int fd;
706
707    mask = 0;
708    fd = pollfd->fd;
709    if (fd >= 0) {
710        int fput_needed;
711        struct file * file;
712
713        file = fget_light(fd, &fput_needed);
714        mask = POLLNVAL;
715        if (file != NULL) {
716            mask = DEFAULT_POLLMASK;
717            if (file->f_op && file->f_op->poll) {
718                if (pwait)
719                    pwait->key = pollfd->events |
720                            POLLERR | POLLHUP;
721                mask = file->f_op->poll(file, pwait);
722            }
723            /* Mask out unneeded events. */
724            mask &= pollfd->events | POLLERR | POLLHUP;
725            fput_light(file, fput_needed);
726        }
727    }
728    pollfd->revents = mask;
729
730    return mask;
731}
732
733static int do_poll(unsigned int nfds, struct poll_list *list,
734           struct poll_wqueues *wait, struct timespec *end_time)
735{
736    poll_table* pt = &wait->pt;
737    ktime_t expire, *to = NULL;
738    int timed_out = 0, count = 0;
739    unsigned long slack = 0;
740
741    /* Optimise the no-wait case */
742    if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
743        pt = NULL;
744        timed_out = 1;
745    }
746
747    if (end_time && !timed_out)
748        slack = estimate_accuracy(end_time);
749
750    for (;;) {
751        struct poll_list *walk;
752
753        for (walk = list; walk != NULL; walk = walk->next) {
754            struct pollfd * pfd, * pfd_end;
755
756            pfd = walk->entries;
757            pfd_end = pfd + walk->len;
758            for (; pfd != pfd_end; pfd++) {
759                /*
760                 * Fish for events. If we found one, record it
761                 * and kill the poll_table, so we don't
762                 * needlessly register any other waiters after
763                 * this. They'll get immediately deregistered
764                 * when we break out and return.
765                 */
766                if (do_pollfd(pfd, pt)) {
767                    count++;
768                    pt = NULL;
769                }
770            }
771        }
772        /*
773         * All waiters have already been registered, so don't provide
774         * a poll_table to them on the next loop iteration.
775         */
776        pt = NULL;
777        if (!count) {
778            count = wait->error;
779            if (signal_pending(current))
780                count = -EINTR;
781        }
782        if (count || timed_out)
783            break;
784
785        /*
786         * If this is the first loop and we have a timeout
787         * given, then we convert to ktime_t and set the to
788         * pointer to the expiry value.
789         */
790        if (end_time && !to) {
791            expire = timespec_to_ktime(*end_time);
792            to = &expire;
793        }
794
795        if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
796            timed_out = 1;
797    }
798    return count;
799}
800
801#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
802            sizeof(struct pollfd))
803
804int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
805        struct timespec *end_time)
806{
807    struct poll_wqueues table;
808     int err = -EFAULT, fdcount, len, size;
809    /* Allocate small arguments on the stack to save memory and be
810       faster - use long to make sure the buffer is aligned properly
811       on 64 bit archs to avoid unaligned access */
812    long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
813    struct poll_list *const head = (struct poll_list *)stack_pps;
814     struct poll_list *walk = head;
815     unsigned long todo = nfds;
816
817    if (nfds > current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
818        return -EINVAL;
819
820    len = min_t(unsigned int, nfds, N_STACK_PPS);
821    for (;;) {
822        walk->next = NULL;
823        walk->len = len;
824        if (!len)
825            break;
826
827        if (copy_from_user(walk->entries, ufds + nfds-todo,
828                    sizeof(struct pollfd) * walk->len))
829            goto out_fds;
830
831        todo -= walk->len;
832        if (!todo)
833            break;
834
835        len = min(todo, POLLFD_PER_PAGE);
836        size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
837        walk = walk->next = kmalloc(size, GFP_KERNEL);
838        if (!walk) {
839            err = -ENOMEM;
840            goto out_fds;
841        }
842    }
843
844    poll_initwait(&table);
845    fdcount = do_poll(nfds, head, &table, end_time);
846    poll_freewait(&table);
847
848    for (walk = head; walk; walk = walk->next) {
849        struct pollfd *fds = walk->entries;
850        int j;
851
852        for (j = 0; j < walk->len; j++, ufds++)
853            if (__put_user(fds[j].revents, &ufds->revents))
854                goto out_fds;
855      }
856
857    err = fdcount;
858out_fds:
859    walk = head->next;
860    while (walk) {
861        struct poll_list *pos = walk;
862        walk = walk->next;
863        kfree(pos);
864    }
865
866    return err;
867}
868
869static long do_restart_poll(struct restart_block *restart_block)
870{
871    struct pollfd __user *ufds = restart_block->poll.ufds;
872    int nfds = restart_block->poll.nfds;
873    struct timespec *to = NULL, end_time;
874    int ret;
875
876    if (restart_block->poll.has_timeout) {
877        end_time.tv_sec = restart_block->poll.tv_sec;
878        end_time.tv_nsec = restart_block->poll.tv_nsec;
879        to = &end_time;
880    }
881
882    ret = do_sys_poll(ufds, nfds, to);
883
884    if (ret == -EINTR) {
885        restart_block->fn = do_restart_poll;
886        ret = -ERESTART_RESTARTBLOCK;
887    }
888    return ret;
889}
890
891SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
892        long, timeout_msecs)
893{
894    struct timespec end_time, *to = NULL;
895    int ret;
896
897    if (timeout_msecs >= 0) {
898        to = &end_time;
899        poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
900            NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
901    }
902
903    ret = do_sys_poll(ufds, nfds, to);
904
905    if (ret == -EINTR) {
906        struct restart_block *restart_block;
907
908        restart_block = &current_thread_info()->restart_block;
909        restart_block->fn = do_restart_poll;
910        restart_block->poll.ufds = ufds;
911        restart_block->poll.nfds = nfds;
912
913        if (timeout_msecs >= 0) {
914            restart_block->poll.tv_sec = end_time.tv_sec;
915            restart_block->poll.tv_nsec = end_time.tv_nsec;
916            restart_block->poll.has_timeout = 1;
917        } else
918            restart_block->poll.has_timeout = 0;
919
920        ret = -ERESTART_RESTARTBLOCK;
921    }
922    return ret;
923}
924
925#ifdef HAVE_SET_RESTORE_SIGMASK
926SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
927        struct timespec __user *, tsp, const sigset_t __user *, sigmask,
928        size_t, sigsetsize)
929{
930    sigset_t ksigmask, sigsaved;
931    struct timespec ts, end_time, *to = NULL;
932    int ret;
933
934    if (tsp) {
935        if (copy_from_user(&ts, tsp, sizeof(ts)))
936            return -EFAULT;
937
938        to = &end_time;
939        if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
940            return -EINVAL;
941    }
942
943    if (sigmask) {
944        /* XXX: Don't preclude handling different sized sigset_t's. */
945        if (sigsetsize != sizeof(sigset_t))
946            return -EINVAL;
947        if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
948            return -EFAULT;
949
950        sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
951        sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
952    }
953
954    ret = do_sys_poll(ufds, nfds, to);
955
956    /* We can restart this syscall, usually */
957    if (ret == -EINTR) {
958        /*
959         * Don't restore the signal mask yet. Let do_signal() deliver
960         * the signal on the way back to userspace, before the signal
961         * mask is restored.
962         */
963        if (sigmask) {
964            memcpy(&current->saved_sigmask, &sigsaved,
965                    sizeof(sigsaved));
966            set_restore_sigmask();
967        }
968        ret = -ERESTARTNOHAND;
969    } else if (sigmask)
970        sigprocmask(SIG_SETMASK, &sigsaved, NULL);
971
972    ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
973
974    return ret;
975}
976#endif /* HAVE_SET_RESTORE_SIGMASK */
977

Archive Download this file



interactive