2 * This file contains the procedures for the handling of select and poll
4 * Created for Linux based loosely upon Mathius Lattner's minix
5 * patches by Peter MacDonald. Heavily edited by Linus.
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.
13 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation
14 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
17 #include <linux/slab.h>
18 #include <linux/smp_lock.h>
19 #include <linux/poll.h>
20 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
21 #include <linux/file.h>
24 #include <asm/uaccess.h>
26 #define ROUND_UP(x,y) (((x)+(y)-1)/(y))
27 #define DEFAULT_POLLMASK (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)
29 struct poll_table_entry
{
32 wait_queue_head_t
* wait_address
;
35 struct poll_table_page
{
36 struct poll_table_page
* next
;
37 struct poll_table_entry
* entry
;
38 struct poll_table_entry entries
[0];
41 #define POLL_TABLE_FULL(table) \
42 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
45 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
46 * I have rewritten this, taking some shortcuts: This code may not be easy to
47 * follow, but it should be free of race-conditions, and it's practical. If you
48 * understand what I'm doing here, then you understand how the linux
49 * sleep/wakeup mechanism works.
51 * Two very simple procedures, poll_wait() and poll_freewait() make all the
52 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
53 * as all select/poll functions have to call it to add an entry to the
56 void __pollwait(struct file
*filp
, wait_queue_head_t
*wait_address
, poll_table
*p
);
58 void poll_initwait(struct poll_wqueues
*pwq
)
60 init_poll_funcptr(&pwq
->pt
, __pollwait
);
65 void poll_freewait(struct poll_wqueues
*pwq
)
67 struct poll_table_page
* p
= pwq
->table
;
69 struct poll_table_entry
* entry
;
70 struct poll_table_page
*old
;
75 remove_wait_queue(entry
->wait_address
,&entry
->wait
);
77 } while (entry
> p
->entries
);
80 free_page((unsigned long) old
);
84 void __pollwait(struct file
*filp
, wait_queue_head_t
*wait_address
, poll_table
*_p
)
86 struct poll_wqueues
*p
= container_of(_p
, struct poll_wqueues
, pt
);
87 struct poll_table_page
*table
= p
->table
;
89 if (!table
|| POLL_TABLE_FULL(table
)) {
90 struct poll_table_page
*new_table
;
92 new_table
= (struct poll_table_page
*) __get_free_page(GFP_KERNEL
);
95 __set_current_state(TASK_RUNNING
);
98 new_table
->entry
= new_table
->entries
;
99 new_table
->next
= table
;
100 p
->table
= new_table
;
104 /* Add a new entry */
106 struct poll_table_entry
* entry
= table
->entry
;
107 table
->entry
= entry
+1;
110 entry
->wait_address
= wait_address
;
111 init_waitqueue_entry(&entry
->wait
, current
);
112 add_wait_queue(wait_address
,&entry
->wait
);
117 #define __IN(fds, n) (fds->in + n)
118 #define __OUT(fds, n) (fds->out + n)
119 #define __EX(fds, n) (fds->ex + n)
120 #define __RES_IN(fds, n) (fds->res_in + n)
121 #define __RES_OUT(fds, n) (fds->res_out + n)
122 #define __RES_EX(fds, n) (fds->res_ex + n)
124 #define BITS(fds, n) (*__IN(fds, n)|*__OUT(fds, n)|*__EX(fds, n))
126 static int max_select_fd(unsigned long n
, fd_set_bits
*fds
)
128 unsigned long *open_fds
;
132 /* handle last in-complete long-word first */
133 set
= ~(~0UL << (n
& (__NFDBITS
-1)));
135 open_fds
= current
->files
->open_fds
->fds_bits
+n
;
140 if (!(set
& ~*open_fds
))
151 if (set
& ~*open_fds
)
160 max
+= n
* __NFDBITS
;
166 #define BIT(i) (1UL << ((i)&(__NFDBITS-1)))
167 #define MEM(i,m) ((m)+(unsigned)(i)/__NFDBITS)
168 #define ISSET(i,m) (((i)&*(m)) != 0)
169 #define SET(i,m) (*(m) |= (i))
171 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
172 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
173 #define POLLEX_SET (POLLPRI)
175 int do_select(int n
, fd_set_bits
*fds
, long *timeout
)
177 struct poll_wqueues table
;
180 long __timeout
= *timeout
;
182 spin_lock(¤t
->files
->file_lock
);
183 retval
= max_select_fd(n
, fds
);
184 spin_unlock(¤t
->files
->file_lock
);
190 poll_initwait(&table
);
196 unsigned long *rinp
, *routp
, *rexp
, *inp
, *outp
, *exp
;
198 set_current_state(TASK_INTERRUPTIBLE
);
200 inp
= fds
->in
; outp
= fds
->out
; exp
= fds
->ex
;
201 rinp
= fds
->res_in
; routp
= fds
->res_out
; rexp
= fds
->res_ex
;
203 for (i
= 0; i
< n
; ++rinp
, ++routp
, ++rexp
) {
204 unsigned long in
, out
, ex
, all_bits
, bit
= 1, mask
, j
;
205 unsigned long res_in
= 0, res_out
= 0, res_ex
= 0;
206 struct file_operations
*f_op
= NULL
;
207 struct file
*file
= NULL
;
209 in
= *inp
++; out
= *outp
++; ex
= *exp
++;
210 all_bits
= in
| out
| ex
;
216 for (j
= 0; j
< __NFDBITS
; ++j
, ++i
, bit
<<= 1) {
219 if (!(bit
& all_bits
))
224 mask
= DEFAULT_POLLMASK
;
225 if (f_op
&& f_op
->poll
)
226 mask
= (*f_op
->poll
)(file
, retval
? NULL
: wait
);
228 if ((mask
& POLLIN_SET
) && (in
& bit
)) {
232 if ((mask
& POLLOUT_SET
) && (out
& bit
)) {
236 if ((mask
& POLLEX_SET
) && (ex
& bit
)) {
250 if (retval
|| !__timeout
|| signal_pending(current
))
253 retval
= table
.error
;
256 __timeout
= schedule_timeout(__timeout
);
258 __set_current_state(TASK_RUNNING
);
260 poll_freewait(&table
);
263 * Up-to-date the caller timeout.
265 *timeout
= __timeout
;
269 static void *select_bits_alloc(int size
)
271 return kmalloc(6 * size
, GFP_KERNEL
);
274 static void select_bits_free(void *bits
, int size
)
280 * We can actually return ERESTARTSYS instead of EINTR, but I'd
281 * like to be certain this leads to no problems. So I return
282 * EINTR just for safety.
284 * Update: ERESTARTSYS breaks at least the xview clock binary, so
285 * I'm trying ERESTARTNOHAND which restart only when you want to.
287 #define MAX_SELECT_SECONDS \
288 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
291 sys_select(int n
, fd_set __user
*inp
, fd_set __user
*outp
, fd_set __user
*exp
, struct timeval __user
*tvp
)
296 int ret
, size
, max_fdset
;
298 timeout
= MAX_SCHEDULE_TIMEOUT
;
302 if ((ret
= verify_area(VERIFY_READ
, tvp
, sizeof(*tvp
)))
303 || (ret
= __get_user(sec
, &tvp
->tv_sec
))
304 || (ret
= __get_user(usec
, &tvp
->tv_usec
)))
308 if (sec
< 0 || usec
< 0)
311 if ((unsigned long) sec
< MAX_SELECT_SECONDS
) {
312 timeout
= ROUND_UP(usec
, 1000000/HZ
);
313 timeout
+= sec
* (unsigned long) HZ
;
321 /* max_fdset can increase, so grab it once to avoid race */
322 max_fdset
= current
->files
->max_fdset
;
327 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
328 * since we used fdset we need to allocate memory in units of
333 bits
= select_bits_alloc(size
);
336 fds
.in
= (unsigned long *) bits
;
337 fds
.out
= (unsigned long *) (bits
+ size
);
338 fds
.ex
= (unsigned long *) (bits
+ 2*size
);
339 fds
.res_in
= (unsigned long *) (bits
+ 3*size
);
340 fds
.res_out
= (unsigned long *) (bits
+ 4*size
);
341 fds
.res_ex
= (unsigned long *) (bits
+ 5*size
);
343 if ((ret
= get_fd_set(n
, inp
, fds
.in
)) ||
344 (ret
= get_fd_set(n
, outp
, fds
.out
)) ||
345 (ret
= get_fd_set(n
, exp
, fds
.ex
)))
347 zero_fd_set(n
, fds
.res_in
);
348 zero_fd_set(n
, fds
.res_out
);
349 zero_fd_set(n
, fds
.res_ex
);
351 ret
= do_select(n
, &fds
, &timeout
);
353 if (tvp
&& !(current
->personality
& STICKY_TIMEOUTS
)) {
354 time_t sec
= 0, usec
= 0;
358 usec
*= (1000000/HZ
);
360 put_user(sec
, &tvp
->tv_sec
);
361 put_user(usec
, &tvp
->tv_usec
);
367 ret
= -ERESTARTNOHAND
;
368 if (signal_pending(current
))
373 set_fd_set(n
, inp
, fds
.res_in
);
374 set_fd_set(n
, outp
, fds
.res_out
);
375 set_fd_set(n
, exp
, fds
.res_ex
);
378 select_bits_free(bits
, size
);
384 struct poll_list
*next
;
386 struct pollfd entries
[0];
389 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
391 static void do_pollfd(unsigned int num
, struct pollfd
* fdpage
,
392 poll_table
** pwait
, int *count
)
396 for (i
= 0; i
< num
; i
++) {
405 struct file
* file
= fget(fd
);
408 mask
= DEFAULT_POLLMASK
;
409 if (file
->f_op
&& file
->f_op
->poll
)
410 mask
= file
->f_op
->poll(file
, *pwait
);
411 mask
&= fdp
->events
| POLLERR
| POLLHUP
;
423 static int do_poll(unsigned int nfds
, struct poll_list
*list
,
424 struct poll_wqueues
*wait
, long timeout
)
427 poll_table
* pt
= &wait
->pt
;
433 struct poll_list
*walk
;
434 set_current_state(TASK_INTERRUPTIBLE
);
436 while(walk
!= NULL
) {
437 do_pollfd( walk
->len
, walk
->entries
, &pt
, &count
);
441 if (count
|| !timeout
|| signal_pending(current
))
446 timeout
= schedule_timeout(timeout
);
448 __set_current_state(TASK_RUNNING
);
452 asmlinkage
long sys_poll(struct pollfd __user
* ufds
, unsigned int nfds
, long timeout
)
454 struct poll_wqueues table
;
457 struct poll_list
*head
;
458 struct poll_list
*walk
;
460 /* Do a sanity check on nfds ... */
461 if (nfds
> current
->files
->max_fdset
&& nfds
> OPEN_MAX
)
465 /* Careful about overflow in the intermediate values */
466 if ((unsigned long) timeout
< MAX_SCHEDULE_TIMEOUT
/ HZ
)
467 timeout
= (unsigned long)(timeout
*HZ
+999)/1000+1;
468 else /* Negative or overflow */
469 timeout
= MAX_SCHEDULE_TIMEOUT
;
472 poll_initwait(&table
);
479 struct poll_list
*pp
;
480 pp
= kmalloc(sizeof(struct poll_list
)+
481 sizeof(struct pollfd
)*
482 (i
>POLLFD_PER_PAGE
?POLLFD_PER_PAGE
:i
),
487 pp
->len
= (i
>POLLFD_PER_PAGE
?POLLFD_PER_PAGE
:i
);
494 if (copy_from_user(pp
->entries
, ufds
+ nfds
-i
,
495 sizeof(struct pollfd
)*pp
->len
)) {
501 fdcount
= do_poll(nfds
, head
, &table
, timeout
);
503 /* OK, now copy the revents fields back to user space. */
506 while(walk
!= NULL
) {
507 struct pollfd
*fds
= walk
->entries
;
510 for (j
=0; j
< walk
->len
; j
++, ufds
++) {
511 if(__put_user(fds
[j
].revents
, &ufds
->revents
))
517 if (!fdcount
&& signal_pending(current
))
522 struct poll_list
*pp
= walk
->next
;
526 poll_freewait(&table
);