Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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/syscalls.h> |
| 18 | #include <linux/module.h> |
| 19 | #include <linux/slab.h> |
| 20 | #include <linux/smp_lock.h> |
| 21 | #include <linux/poll.h> |
| 22 | #include <linux/personality.h> /* for STICKY_TIMEOUTS */ |
| 23 | #include <linux/file.h> |
| 24 | #include <linux/fs.h> |
| 25 | |
| 26 | #include <asm/uaccess.h> |
| 27 | |
| 28 | #define ROUND_UP(x,y) (((x)+(y)-1)/(y)) |
| 29 | #define DEFAULT_POLLMASK (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM) |
| 30 | |
| 31 | struct poll_table_entry { |
| 32 | struct file * filp; |
| 33 | wait_queue_t wait; |
| 34 | wait_queue_head_t * wait_address; |
| 35 | }; |
| 36 | |
| 37 | struct poll_table_page { |
| 38 | struct poll_table_page * next; |
| 39 | struct poll_table_entry * entry; |
| 40 | struct poll_table_entry entries[0]; |
| 41 | }; |
| 42 | |
| 43 | #define POLL_TABLE_FULL(table) \ |
| 44 | ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) |
| 45 | |
| 46 | /* |
| 47 | * Ok, Peter made a complicated, but straightforward multiple_wait() function. |
| 48 | * I have rewritten this, taking some shortcuts: This code may not be easy to |
| 49 | * follow, but it should be free of race-conditions, and it's practical. If you |
| 50 | * understand what I'm doing here, then you understand how the linux |
| 51 | * sleep/wakeup mechanism works. |
| 52 | * |
| 53 | * Two very simple procedures, poll_wait() and poll_freewait() make all the |
| 54 | * work. poll_wait() is an inline-function defined in <linux/poll.h>, |
| 55 | * as all select/poll functions have to call it to add an entry to the |
| 56 | * poll table. |
| 57 | */ |
Adrian Bunk | 75c96f8 | 2005-05-05 16:16:09 -0700 | [diff] [blame] | 58 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
| 59 | poll_table *p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 60 | |
| 61 | void poll_initwait(struct poll_wqueues *pwq) |
| 62 | { |
| 63 | init_poll_funcptr(&pwq->pt, __pollwait); |
| 64 | pwq->error = 0; |
| 65 | pwq->table = NULL; |
| 66 | } |
| 67 | |
| 68 | EXPORT_SYMBOL(poll_initwait); |
| 69 | |
| 70 | void poll_freewait(struct poll_wqueues *pwq) |
| 71 | { |
| 72 | struct poll_table_page * p = pwq->table; |
| 73 | while (p) { |
| 74 | struct poll_table_entry * entry; |
| 75 | struct poll_table_page *old; |
| 76 | |
| 77 | entry = p->entry; |
| 78 | do { |
| 79 | entry--; |
| 80 | remove_wait_queue(entry->wait_address,&entry->wait); |
| 81 | fput(entry->filp); |
| 82 | } while (entry > p->entries); |
| 83 | old = p; |
| 84 | p = p->next; |
| 85 | free_page((unsigned long) old); |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | EXPORT_SYMBOL(poll_freewait); |
| 90 | |
Adrian Bunk | 75c96f8 | 2005-05-05 16:16:09 -0700 | [diff] [blame] | 91 | static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, |
| 92 | poll_table *_p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 93 | { |
| 94 | struct poll_wqueues *p = container_of(_p, struct poll_wqueues, pt); |
| 95 | struct poll_table_page *table = p->table; |
| 96 | |
| 97 | if (!table || POLL_TABLE_FULL(table)) { |
| 98 | struct poll_table_page *new_table; |
| 99 | |
| 100 | new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); |
| 101 | if (!new_table) { |
| 102 | p->error = -ENOMEM; |
| 103 | __set_current_state(TASK_RUNNING); |
| 104 | return; |
| 105 | } |
| 106 | new_table->entry = new_table->entries; |
| 107 | new_table->next = table; |
| 108 | p->table = new_table; |
| 109 | table = new_table; |
| 110 | } |
| 111 | |
| 112 | /* Add a new entry */ |
| 113 | { |
| 114 | struct poll_table_entry * entry = table->entry; |
| 115 | table->entry = entry+1; |
| 116 | get_file(filp); |
| 117 | entry->filp = filp; |
| 118 | entry->wait_address = wait_address; |
| 119 | init_waitqueue_entry(&entry->wait, current); |
| 120 | add_wait_queue(wait_address,&entry->wait); |
| 121 | } |
| 122 | } |
| 123 | |
| 124 | #define FDS_IN(fds, n) (fds->in + n) |
| 125 | #define FDS_OUT(fds, n) (fds->out + n) |
| 126 | #define FDS_EX(fds, n) (fds->ex + n) |
| 127 | |
| 128 | #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n)) |
| 129 | |
| 130 | static int max_select_fd(unsigned long n, fd_set_bits *fds) |
| 131 | { |
| 132 | unsigned long *open_fds; |
| 133 | unsigned long set; |
| 134 | int max; |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame^] | 135 | struct fdtable *fdt; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 136 | |
| 137 | /* handle last in-complete long-word first */ |
| 138 | set = ~(~0UL << (n & (__NFDBITS-1))); |
| 139 | n /= __NFDBITS; |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame^] | 140 | fdt = files_fdtable(current->files); |
| 141 | open_fds = fdt->open_fds->fds_bits+n; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | max = 0; |
| 143 | if (set) { |
| 144 | set &= BITS(fds, n); |
| 145 | if (set) { |
| 146 | if (!(set & ~*open_fds)) |
| 147 | goto get_max; |
| 148 | return -EBADF; |
| 149 | } |
| 150 | } |
| 151 | while (n) { |
| 152 | open_fds--; |
| 153 | n--; |
| 154 | set = BITS(fds, n); |
| 155 | if (!set) |
| 156 | continue; |
| 157 | if (set & ~*open_fds) |
| 158 | return -EBADF; |
| 159 | if (max) |
| 160 | continue; |
| 161 | get_max: |
| 162 | do { |
| 163 | max++; |
| 164 | set >>= 1; |
| 165 | } while (set); |
| 166 | max += n * __NFDBITS; |
| 167 | } |
| 168 | |
| 169 | return max; |
| 170 | } |
| 171 | |
| 172 | #define BIT(i) (1UL << ((i)&(__NFDBITS-1))) |
| 173 | #define MEM(i,m) ((m)+(unsigned)(i)/__NFDBITS) |
| 174 | #define ISSET(i,m) (((i)&*(m)) != 0) |
| 175 | #define SET(i,m) (*(m) |= (i)) |
| 176 | |
| 177 | #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR) |
| 178 | #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR) |
| 179 | #define POLLEX_SET (POLLPRI) |
| 180 | |
| 181 | int do_select(int n, fd_set_bits *fds, long *timeout) |
| 182 | { |
| 183 | struct poll_wqueues table; |
| 184 | poll_table *wait; |
| 185 | int retval, i; |
| 186 | long __timeout = *timeout; |
| 187 | |
| 188 | spin_lock(¤t->files->file_lock); |
| 189 | retval = max_select_fd(n, fds); |
| 190 | spin_unlock(¤t->files->file_lock); |
| 191 | |
| 192 | if (retval < 0) |
| 193 | return retval; |
| 194 | n = retval; |
| 195 | |
| 196 | poll_initwait(&table); |
| 197 | wait = &table.pt; |
| 198 | if (!__timeout) |
| 199 | wait = NULL; |
| 200 | retval = 0; |
| 201 | for (;;) { |
| 202 | unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; |
| 203 | |
| 204 | set_current_state(TASK_INTERRUPTIBLE); |
| 205 | |
| 206 | inp = fds->in; outp = fds->out; exp = fds->ex; |
| 207 | rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; |
| 208 | |
| 209 | for (i = 0; i < n; ++rinp, ++routp, ++rexp) { |
| 210 | unsigned long in, out, ex, all_bits, bit = 1, mask, j; |
| 211 | unsigned long res_in = 0, res_out = 0, res_ex = 0; |
| 212 | struct file_operations *f_op = NULL; |
| 213 | struct file *file = NULL; |
| 214 | |
| 215 | in = *inp++; out = *outp++; ex = *exp++; |
| 216 | all_bits = in | out | ex; |
| 217 | if (all_bits == 0) { |
| 218 | i += __NFDBITS; |
| 219 | continue; |
| 220 | } |
| 221 | |
| 222 | for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) { |
| 223 | if (i >= n) |
| 224 | break; |
| 225 | if (!(bit & all_bits)) |
| 226 | continue; |
| 227 | file = fget(i); |
| 228 | if (file) { |
| 229 | f_op = file->f_op; |
| 230 | mask = DEFAULT_POLLMASK; |
| 231 | if (f_op && f_op->poll) |
| 232 | mask = (*f_op->poll)(file, retval ? NULL : wait); |
| 233 | fput(file); |
| 234 | if ((mask & POLLIN_SET) && (in & bit)) { |
| 235 | res_in |= bit; |
| 236 | retval++; |
| 237 | } |
| 238 | if ((mask & POLLOUT_SET) && (out & bit)) { |
| 239 | res_out |= bit; |
| 240 | retval++; |
| 241 | } |
| 242 | if ((mask & POLLEX_SET) && (ex & bit)) { |
| 243 | res_ex |= bit; |
| 244 | retval++; |
| 245 | } |
| 246 | } |
| 247 | cond_resched(); |
| 248 | } |
| 249 | if (res_in) |
| 250 | *rinp = res_in; |
| 251 | if (res_out) |
| 252 | *routp = res_out; |
| 253 | if (res_ex) |
| 254 | *rexp = res_ex; |
| 255 | } |
| 256 | wait = NULL; |
| 257 | if (retval || !__timeout || signal_pending(current)) |
| 258 | break; |
| 259 | if(table.error) { |
| 260 | retval = table.error; |
| 261 | break; |
| 262 | } |
| 263 | __timeout = schedule_timeout(__timeout); |
| 264 | } |
| 265 | __set_current_state(TASK_RUNNING); |
| 266 | |
| 267 | poll_freewait(&table); |
| 268 | |
| 269 | /* |
| 270 | * Up-to-date the caller timeout. |
| 271 | */ |
| 272 | *timeout = __timeout; |
| 273 | return retval; |
| 274 | } |
| 275 | |
| 276 | static void *select_bits_alloc(int size) |
| 277 | { |
| 278 | return kmalloc(6 * size, GFP_KERNEL); |
| 279 | } |
| 280 | |
| 281 | static void select_bits_free(void *bits, int size) |
| 282 | { |
| 283 | kfree(bits); |
| 284 | } |
| 285 | |
| 286 | /* |
| 287 | * We can actually return ERESTARTSYS instead of EINTR, but I'd |
| 288 | * like to be certain this leads to no problems. So I return |
| 289 | * EINTR just for safety. |
| 290 | * |
| 291 | * Update: ERESTARTSYS breaks at least the xview clock binary, so |
| 292 | * I'm trying ERESTARTNOHAND which restart only when you want to. |
| 293 | */ |
| 294 | #define MAX_SELECT_SECONDS \ |
| 295 | ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1) |
| 296 | |
| 297 | asmlinkage long |
| 298 | sys_select(int n, fd_set __user *inp, fd_set __user *outp, fd_set __user *exp, struct timeval __user *tvp) |
| 299 | { |
| 300 | fd_set_bits fds; |
| 301 | char *bits; |
| 302 | long timeout; |
| 303 | int ret, size, max_fdset; |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame^] | 304 | struct fdtable *fdt; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | |
| 306 | timeout = MAX_SCHEDULE_TIMEOUT; |
| 307 | if (tvp) { |
| 308 | time_t sec, usec; |
| 309 | |
| 310 | if (!access_ok(VERIFY_READ, tvp, sizeof(*tvp)) |
| 311 | || __get_user(sec, &tvp->tv_sec) |
| 312 | || __get_user(usec, &tvp->tv_usec)) { |
| 313 | ret = -EFAULT; |
| 314 | goto out_nofds; |
| 315 | } |
| 316 | |
| 317 | ret = -EINVAL; |
| 318 | if (sec < 0 || usec < 0) |
| 319 | goto out_nofds; |
| 320 | |
| 321 | if ((unsigned long) sec < MAX_SELECT_SECONDS) { |
| 322 | timeout = ROUND_UP(usec, 1000000/HZ); |
| 323 | timeout += sec * (unsigned long) HZ; |
| 324 | } |
| 325 | } |
| 326 | |
| 327 | ret = -EINVAL; |
| 328 | if (n < 0) |
| 329 | goto out_nofds; |
| 330 | |
| 331 | /* max_fdset can increase, so grab it once to avoid race */ |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame^] | 332 | fdt = files_fdtable(current->files); |
| 333 | max_fdset = fdt->max_fdset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | if (n > max_fdset) |
| 335 | n = max_fdset; |
| 336 | |
| 337 | /* |
| 338 | * We need 6 bitmaps (in/out/ex for both incoming and outgoing), |
| 339 | * since we used fdset we need to allocate memory in units of |
| 340 | * long-words. |
| 341 | */ |
| 342 | ret = -ENOMEM; |
| 343 | size = FDS_BYTES(n); |
| 344 | bits = select_bits_alloc(size); |
| 345 | if (!bits) |
| 346 | goto out_nofds; |
| 347 | fds.in = (unsigned long *) bits; |
| 348 | fds.out = (unsigned long *) (bits + size); |
| 349 | fds.ex = (unsigned long *) (bits + 2*size); |
| 350 | fds.res_in = (unsigned long *) (bits + 3*size); |
| 351 | fds.res_out = (unsigned long *) (bits + 4*size); |
| 352 | fds.res_ex = (unsigned long *) (bits + 5*size); |
| 353 | |
| 354 | if ((ret = get_fd_set(n, inp, fds.in)) || |
| 355 | (ret = get_fd_set(n, outp, fds.out)) || |
| 356 | (ret = get_fd_set(n, exp, fds.ex))) |
| 357 | goto out; |
| 358 | zero_fd_set(n, fds.res_in); |
| 359 | zero_fd_set(n, fds.res_out); |
| 360 | zero_fd_set(n, fds.res_ex); |
| 361 | |
| 362 | ret = do_select(n, &fds, &timeout); |
| 363 | |
| 364 | if (tvp && !(current->personality & STICKY_TIMEOUTS)) { |
| 365 | time_t sec = 0, usec = 0; |
| 366 | if (timeout) { |
| 367 | sec = timeout / HZ; |
| 368 | usec = timeout % HZ; |
| 369 | usec *= (1000000/HZ); |
| 370 | } |
| 371 | put_user(sec, &tvp->tv_sec); |
| 372 | put_user(usec, &tvp->tv_usec); |
| 373 | } |
| 374 | |
| 375 | if (ret < 0) |
| 376 | goto out; |
| 377 | if (!ret) { |
| 378 | ret = -ERESTARTNOHAND; |
| 379 | if (signal_pending(current)) |
| 380 | goto out; |
| 381 | ret = 0; |
| 382 | } |
| 383 | |
| 384 | if (set_fd_set(n, inp, fds.res_in) || |
| 385 | set_fd_set(n, outp, fds.res_out) || |
| 386 | set_fd_set(n, exp, fds.res_ex)) |
| 387 | ret = -EFAULT; |
| 388 | |
| 389 | out: |
| 390 | select_bits_free(bits, size); |
| 391 | out_nofds: |
| 392 | return ret; |
| 393 | } |
| 394 | |
| 395 | struct poll_list { |
| 396 | struct poll_list *next; |
| 397 | int len; |
| 398 | struct pollfd entries[0]; |
| 399 | }; |
| 400 | |
| 401 | #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) |
| 402 | |
| 403 | static void do_pollfd(unsigned int num, struct pollfd * fdpage, |
| 404 | poll_table ** pwait, int *count) |
| 405 | { |
| 406 | int i; |
| 407 | |
| 408 | for (i = 0; i < num; i++) { |
| 409 | int fd; |
| 410 | unsigned int mask; |
| 411 | struct pollfd *fdp; |
| 412 | |
| 413 | mask = 0; |
| 414 | fdp = fdpage+i; |
| 415 | fd = fdp->fd; |
| 416 | if (fd >= 0) { |
| 417 | struct file * file = fget(fd); |
| 418 | mask = POLLNVAL; |
| 419 | if (file != NULL) { |
| 420 | mask = DEFAULT_POLLMASK; |
| 421 | if (file->f_op && file->f_op->poll) |
| 422 | mask = file->f_op->poll(file, *pwait); |
| 423 | mask &= fdp->events | POLLERR | POLLHUP; |
| 424 | fput(file); |
| 425 | } |
| 426 | if (mask) { |
| 427 | *pwait = NULL; |
| 428 | (*count)++; |
| 429 | } |
| 430 | } |
| 431 | fdp->revents = mask; |
| 432 | } |
| 433 | } |
| 434 | |
| 435 | static int do_poll(unsigned int nfds, struct poll_list *list, |
| 436 | struct poll_wqueues *wait, long timeout) |
| 437 | { |
| 438 | int count = 0; |
| 439 | poll_table* pt = &wait->pt; |
| 440 | |
| 441 | if (!timeout) |
| 442 | pt = NULL; |
| 443 | |
| 444 | for (;;) { |
| 445 | struct poll_list *walk; |
| 446 | set_current_state(TASK_INTERRUPTIBLE); |
| 447 | walk = list; |
| 448 | while(walk != NULL) { |
| 449 | do_pollfd( walk->len, walk->entries, &pt, &count); |
| 450 | walk = walk->next; |
| 451 | } |
| 452 | pt = NULL; |
| 453 | if (count || !timeout || signal_pending(current)) |
| 454 | break; |
| 455 | count = wait->error; |
| 456 | if (count) |
| 457 | break; |
| 458 | timeout = schedule_timeout(timeout); |
| 459 | } |
| 460 | __set_current_state(TASK_RUNNING); |
| 461 | return count; |
| 462 | } |
| 463 | |
| 464 | asmlinkage long sys_poll(struct pollfd __user * ufds, unsigned int nfds, long timeout) |
| 465 | { |
| 466 | struct poll_wqueues table; |
| 467 | int fdcount, err; |
| 468 | unsigned int i; |
| 469 | struct poll_list *head; |
| 470 | struct poll_list *walk; |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame^] | 471 | struct fdtable *fdt; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 472 | |
| 473 | /* Do a sanity check on nfds ... */ |
Dipankar Sarma | badf166 | 2005-09-09 13:04:10 -0700 | [diff] [blame^] | 474 | fdt = files_fdtable(current->files); |
| 475 | if (nfds > fdt->max_fdset && nfds > OPEN_MAX) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | return -EINVAL; |
| 477 | |
| 478 | if (timeout) { |
| 479 | /* Careful about overflow in the intermediate values */ |
| 480 | if ((unsigned long) timeout < MAX_SCHEDULE_TIMEOUT / HZ) |
| 481 | timeout = (unsigned long)(timeout*HZ+999)/1000+1; |
| 482 | else /* Negative or overflow */ |
| 483 | timeout = MAX_SCHEDULE_TIMEOUT; |
| 484 | } |
| 485 | |
| 486 | poll_initwait(&table); |
| 487 | |
| 488 | head = NULL; |
| 489 | walk = NULL; |
| 490 | i = nfds; |
| 491 | err = -ENOMEM; |
| 492 | while(i!=0) { |
| 493 | struct poll_list *pp; |
| 494 | pp = kmalloc(sizeof(struct poll_list)+ |
| 495 | sizeof(struct pollfd)* |
| 496 | (i>POLLFD_PER_PAGE?POLLFD_PER_PAGE:i), |
| 497 | GFP_KERNEL); |
| 498 | if(pp==NULL) |
| 499 | goto out_fds; |
| 500 | pp->next=NULL; |
| 501 | pp->len = (i>POLLFD_PER_PAGE?POLLFD_PER_PAGE:i); |
| 502 | if (head == NULL) |
| 503 | head = pp; |
| 504 | else |
| 505 | walk->next = pp; |
| 506 | |
| 507 | walk = pp; |
| 508 | if (copy_from_user(pp->entries, ufds + nfds-i, |
| 509 | sizeof(struct pollfd)*pp->len)) { |
| 510 | err = -EFAULT; |
| 511 | goto out_fds; |
| 512 | } |
| 513 | i -= pp->len; |
| 514 | } |
| 515 | fdcount = do_poll(nfds, head, &table, timeout); |
| 516 | |
| 517 | /* OK, now copy the revents fields back to user space. */ |
| 518 | walk = head; |
| 519 | err = -EFAULT; |
| 520 | while(walk != NULL) { |
| 521 | struct pollfd *fds = walk->entries; |
| 522 | int j; |
| 523 | |
| 524 | for (j=0; j < walk->len; j++, ufds++) { |
| 525 | if(__put_user(fds[j].revents, &ufds->revents)) |
| 526 | goto out_fds; |
| 527 | } |
| 528 | walk = walk->next; |
| 529 | } |
| 530 | err = fdcount; |
| 531 | if (!fdcount && signal_pending(current)) |
| 532 | err = -EINTR; |
| 533 | out_fds: |
| 534 | walk = head; |
| 535 | while(walk!=NULL) { |
| 536 | struct poll_list *pp = walk->next; |
| 537 | kfree(walk); |
| 538 | walk = pp; |
| 539 | } |
| 540 | poll_freewait(&table); |
| 541 | return err; |
| 542 | } |