1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 /* 2 /*
3 * This file contains the procedures for the h 3 * This file contains the procedures for the handling of select and poll
4 * 4 *
5 * Created for Linux based loosely upon Mathiu 5 * Created for Linux based loosely upon Mathius Lattner's minix
6 * patches by Peter MacDonald. Heavily edited 6 * patches by Peter MacDonald. Heavily edited by Linus.
7 * 7 *
8 * 4 February 1994 8 * 4 February 1994
9 * COFF/ELF binary emulation. If the proce 9 * COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS
10 * flag set in its personality we do *not* 10 * flag set in its personality we do *not* modify the given timeout
11 * parameter to reflect time remaining. 11 * parameter to reflect time remaining.
12 * 12 *
13 * 24 January 2000 13 * 24 January 2000
14 * Changed sys_poll()/do_poll() to use PAG 14 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation
15 * of fds to overcome nfds < 16390 descrip 15 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
16 */ 16 */
17 17
18 #include <linux/compat.h> <<
19 #include <linux/kernel.h> 18 #include <linux/kernel.h>
20 #include <linux/sched/signal.h> 19 #include <linux/sched/signal.h>
21 #include <linux/sched/rt.h> 20 #include <linux/sched/rt.h>
22 #include <linux/syscalls.h> 21 #include <linux/syscalls.h>
23 #include <linux/export.h> 22 #include <linux/export.h>
24 #include <linux/slab.h> 23 #include <linux/slab.h>
25 #include <linux/poll.h> 24 #include <linux/poll.h>
26 #include <linux/personality.h> /* for STICKY_T 25 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
27 #include <linux/file.h> 26 #include <linux/file.h>
28 #include <linux/fdtable.h> 27 #include <linux/fdtable.h>
29 #include <linux/fs.h> 28 #include <linux/fs.h>
30 #include <linux/rcupdate.h> 29 #include <linux/rcupdate.h>
31 #include <linux/hrtimer.h> 30 #include <linux/hrtimer.h>
32 #include <linux/freezer.h> 31 #include <linux/freezer.h>
33 #include <net/busy_poll.h> 32 #include <net/busy_poll.h>
34 #include <linux/vmalloc.h> 33 #include <linux/vmalloc.h>
35 34
36 #include <linux/uaccess.h> 35 #include <linux/uaccess.h>
37 36
38 37
39 /* 38 /*
40 * Estimate expected accuracy in ns from a tim 39 * Estimate expected accuracy in ns from a timeval.
41 * 40 *
42 * After quite a bit of churning around, we've 41 * After quite a bit of churning around, we've settled on
43 * a simple thing of taking 0.1% of the timeou 42 * a simple thing of taking 0.1% of the timeout as the
44 * slack, with a cap of 100 msec. 43 * slack, with a cap of 100 msec.
45 * "nice" tasks get a 0.5% slack instead. 44 * "nice" tasks get a 0.5% slack instead.
46 * 45 *
47 * Consider this comment an open invitation to 46 * Consider this comment an open invitation to come up with even
48 * better solutions.. 47 * better solutions..
49 */ 48 */
50 49
51 #define MAX_SLACK (100 * NSEC_PER_MSEC) 50 #define MAX_SLACK (100 * NSEC_PER_MSEC)
52 51
53 static long __estimate_accuracy(struct timespe 52 static long __estimate_accuracy(struct timespec64 *tv)
54 { 53 {
55 long slack; 54 long slack;
56 int divfactor = 1000; 55 int divfactor = 1000;
57 56
58 if (tv->tv_sec < 0) 57 if (tv->tv_sec < 0)
59 return 0; 58 return 0;
60 59
61 if (task_nice(current) > 0) 60 if (task_nice(current) > 0)
62 divfactor = divfactor / 5; 61 divfactor = divfactor / 5;
63 62
64 if (tv->tv_sec > MAX_SLACK / (NSEC_PER 63 if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
65 return MAX_SLACK; 64 return MAX_SLACK;
66 65
67 slack = tv->tv_nsec / divfactor; 66 slack = tv->tv_nsec / divfactor;
68 slack += tv->tv_sec * (NSEC_PER_SEC/di 67 slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
69 68
70 if (slack > MAX_SLACK) 69 if (slack > MAX_SLACK)
71 return MAX_SLACK; 70 return MAX_SLACK;
72 71
73 return slack; 72 return slack;
74 } 73 }
75 74
76 u64 select_estimate_accuracy(struct timespec64 75 u64 select_estimate_accuracy(struct timespec64 *tv)
77 { 76 {
78 u64 ret; 77 u64 ret;
79 struct timespec64 now; 78 struct timespec64 now;
80 u64 slack = current->timer_slack_ns; <<
81 79
82 if (slack == 0) !! 80 /*
>> 81 * Realtime tasks get a slack of 0 for obvious reasons.
>> 82 */
>> 83
>> 84 if (rt_task(current))
83 return 0; 85 return 0;
84 86
85 ktime_get_ts64(&now); 87 ktime_get_ts64(&now);
86 now = timespec64_sub(*tv, now); 88 now = timespec64_sub(*tv, now);
87 ret = __estimate_accuracy(&now); 89 ret = __estimate_accuracy(&now);
88 if (ret < slack) !! 90 if (ret < current->timer_slack_ns)
89 return slack; !! 91 return current->timer_slack_ns;
90 return ret; 92 return ret;
91 } 93 }
92 94
93 95
94 96
95 struct poll_table_page { 97 struct poll_table_page {
96 struct poll_table_page * next; 98 struct poll_table_page * next;
97 struct poll_table_entry * entry; 99 struct poll_table_entry * entry;
98 struct poll_table_entry entries[]; !! 100 struct poll_table_entry entries[0];
99 }; 101 };
100 102
101 #define POLL_TABLE_FULL(table) \ 103 #define POLL_TABLE_FULL(table) \
102 ((unsigned long)((table)->entry+1) > P 104 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
103 105
104 /* 106 /*
105 * Ok, Peter made a complicated, but straightf 107 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
106 * I have rewritten this, taking some shortcut 108 * I have rewritten this, taking some shortcuts: This code may not be easy to
107 * follow, but it should be free of race-condi 109 * follow, but it should be free of race-conditions, and it's practical. If you
108 * understand what I'm doing here, then you un 110 * understand what I'm doing here, then you understand how the linux
109 * sleep/wakeup mechanism works. 111 * sleep/wakeup mechanism works.
110 * 112 *
111 * Two very simple procedures, poll_wait() and 113 * Two very simple procedures, poll_wait() and poll_freewait() make all the
112 * work. poll_wait() is an inline-function de 114 * work. poll_wait() is an inline-function defined in <linux/poll.h>,
113 * as all select/poll functions have to call i 115 * as all select/poll functions have to call it to add an entry to the
114 * poll table. 116 * poll table.
115 */ 117 */
116 static void __pollwait(struct file *filp, wait 118 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
117 poll_table *p); 119 poll_table *p);
118 120
119 void poll_initwait(struct poll_wqueues *pwq) 121 void poll_initwait(struct poll_wqueues *pwq)
120 { 122 {
121 init_poll_funcptr(&pwq->pt, __pollwait 123 init_poll_funcptr(&pwq->pt, __pollwait);
122 pwq->polling_task = current; 124 pwq->polling_task = current;
123 pwq->triggered = 0; 125 pwq->triggered = 0;
124 pwq->error = 0; 126 pwq->error = 0;
125 pwq->table = NULL; 127 pwq->table = NULL;
126 pwq->inline_index = 0; 128 pwq->inline_index = 0;
127 } 129 }
128 EXPORT_SYMBOL(poll_initwait); 130 EXPORT_SYMBOL(poll_initwait);
129 131
130 static void free_poll_entry(struct poll_table_ 132 static void free_poll_entry(struct poll_table_entry *entry)
131 { 133 {
132 remove_wait_queue(entry->wait_address, 134 remove_wait_queue(entry->wait_address, &entry->wait);
133 fput(entry->filp); 135 fput(entry->filp);
134 } 136 }
135 137
136 void poll_freewait(struct poll_wqueues *pwq) 138 void poll_freewait(struct poll_wqueues *pwq)
137 { 139 {
138 struct poll_table_page * p = pwq->tabl 140 struct poll_table_page * p = pwq->table;
139 int i; 141 int i;
140 for (i = 0; i < pwq->inline_index; i++ 142 for (i = 0; i < pwq->inline_index; i++)
141 free_poll_entry(pwq->inline_en 143 free_poll_entry(pwq->inline_entries + i);
142 while (p) { 144 while (p) {
143 struct poll_table_entry * entr 145 struct poll_table_entry * entry;
144 struct poll_table_page *old; 146 struct poll_table_page *old;
145 147
146 entry = p->entry; 148 entry = p->entry;
147 do { 149 do {
148 entry--; 150 entry--;
149 free_poll_entry(entry) 151 free_poll_entry(entry);
150 } while (entry > p->entries); 152 } while (entry > p->entries);
151 old = p; 153 old = p;
152 p = p->next; 154 p = p->next;
153 free_page((unsigned long) old) 155 free_page((unsigned long) old);
154 } 156 }
155 } 157 }
156 EXPORT_SYMBOL(poll_freewait); 158 EXPORT_SYMBOL(poll_freewait);
157 159
158 static struct poll_table_entry *poll_get_entry 160 static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
159 { 161 {
160 struct poll_table_page *table = p->tab 162 struct poll_table_page *table = p->table;
161 163
162 if (p->inline_index < N_INLINE_POLL_EN 164 if (p->inline_index < N_INLINE_POLL_ENTRIES)
163 return p->inline_entries + p-> 165 return p->inline_entries + p->inline_index++;
164 166
165 if (!table || POLL_TABLE_FULL(table)) 167 if (!table || POLL_TABLE_FULL(table)) {
166 struct poll_table_page *new_ta 168 struct poll_table_page *new_table;
167 169
168 new_table = (struct poll_table 170 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
169 if (!new_table) { 171 if (!new_table) {
170 p->error = -ENOMEM; 172 p->error = -ENOMEM;
171 return NULL; 173 return NULL;
172 } 174 }
173 new_table->entry = new_table-> 175 new_table->entry = new_table->entries;
174 new_table->next = table; 176 new_table->next = table;
175 p->table = new_table; 177 p->table = new_table;
176 table = new_table; 178 table = new_table;
177 } 179 }
178 180
179 return table->entry++; 181 return table->entry++;
180 } 182 }
181 183
182 static int __pollwake(wait_queue_entry_t *wait 184 static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
183 { 185 {
184 struct poll_wqueues *pwq = wait->priva 186 struct poll_wqueues *pwq = wait->private;
185 DECLARE_WAITQUEUE(dummy_wait, pwq->pol 187 DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
186 188
187 /* 189 /*
188 * Although this function is called un 190 * Although this function is called under waitqueue lock, LOCK
189 * doesn't imply write barrier and the 191 * doesn't imply write barrier and the users expect write
190 * barrier semantics on wakeup functio 192 * barrier semantics on wakeup functions. The following
191 * smp_wmb() is equivalent to smp_wmb( 193 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
192 * and is paired with smp_store_mb() i 194 * and is paired with smp_store_mb() in poll_schedule_timeout.
193 */ 195 */
194 smp_wmb(); 196 smp_wmb();
195 pwq->triggered = 1; 197 pwq->triggered = 1;
196 198
197 /* 199 /*
198 * Perform the default wake up operati 200 * Perform the default wake up operation using a dummy
199 * waitqueue. 201 * waitqueue.
200 * 202 *
201 * TODO: This is hacky but there curre 203 * TODO: This is hacky but there currently is no interface to
202 * pass in @sync. @sync is scheduled 204 * pass in @sync. @sync is scheduled to be removed and once
203 * that happens, wake_up_process() can 205 * that happens, wake_up_process() can be used directly.
204 */ 206 */
205 return default_wake_function(&dummy_wa 207 return default_wake_function(&dummy_wait, mode, sync, key);
206 } 208 }
207 209
208 static int pollwake(wait_queue_entry_t *wait, 210 static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
209 { 211 {
210 struct poll_table_entry *entry; 212 struct poll_table_entry *entry;
211 213
212 entry = container_of(wait, struct poll 214 entry = container_of(wait, struct poll_table_entry, wait);
213 if (key && !(key_to_poll(key) & entry- 215 if (key && !(key_to_poll(key) & entry->key))
214 return 0; 216 return 0;
215 return __pollwake(wait, mode, sync, ke 217 return __pollwake(wait, mode, sync, key);
216 } 218 }
217 219
218 /* Add a new entry */ 220 /* Add a new entry */
219 static void __pollwait(struct file *filp, wait 221 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
220 poll_table *p) 222 poll_table *p)
221 { 223 {
222 struct poll_wqueues *pwq = container_o 224 struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
223 struct poll_table_entry *entry = poll_ 225 struct poll_table_entry *entry = poll_get_entry(pwq);
224 if (!entry) 226 if (!entry)
225 return; 227 return;
226 entry->filp = get_file(filp); 228 entry->filp = get_file(filp);
227 entry->wait_address = wait_address; 229 entry->wait_address = wait_address;
228 entry->key = p->_key; 230 entry->key = p->_key;
229 init_waitqueue_func_entry(&entry->wait 231 init_waitqueue_func_entry(&entry->wait, pollwake);
230 entry->wait.private = pwq; 232 entry->wait.private = pwq;
231 add_wait_queue(wait_address, &entry->w 233 add_wait_queue(wait_address, &entry->wait);
232 } 234 }
233 235
234 static int poll_schedule_timeout(struct poll_w 236 static int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
235 ktime_t *expires, un 237 ktime_t *expires, unsigned long slack)
236 { 238 {
237 int rc = -EINTR; 239 int rc = -EINTR;
238 240
239 set_current_state(state); 241 set_current_state(state);
240 if (!pwq->triggered) 242 if (!pwq->triggered)
241 rc = schedule_hrtimeout_range( 243 rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
242 __set_current_state(TASK_RUNNING); 244 __set_current_state(TASK_RUNNING);
243 245
244 /* 246 /*
245 * Prepare for the next iteration. 247 * Prepare for the next iteration.
246 * 248 *
247 * The following smp_store_mb() serves 249 * The following smp_store_mb() serves two purposes. First, it's
248 * the counterpart rmb of the wmb in p 250 * the counterpart rmb of the wmb in pollwake() such that data
249 * written before wake up is always vi 251 * written before wake up is always visible after wake up.
250 * Second, the full barrier guarantees 252 * Second, the full barrier guarantees that triggered clearing
251 * doesn't pass event check of the nex 253 * doesn't pass event check of the next iteration. Note that
252 * this problem doesn't exist for the 254 * this problem doesn't exist for the first iteration as
253 * add_wait_queue() has full barrier s 255 * add_wait_queue() has full barrier semantics.
254 */ 256 */
255 smp_store_mb(pwq->triggered, 0); 257 smp_store_mb(pwq->triggered, 0);
256 258
257 return rc; 259 return rc;
258 } 260 }
259 261
260 /** 262 /**
261 * poll_select_set_timeout - helper function t 263 * poll_select_set_timeout - helper function to setup the timeout value
262 * @to: pointer to timespec64 variable 264 * @to: pointer to timespec64 variable for the final timeout
263 * @sec: seconds (from user space) 265 * @sec: seconds (from user space)
264 * @nsec: nanoseconds (from user space) 266 * @nsec: nanoseconds (from user space)
265 * 267 *
266 * Note, we do not use a timespec for the user 268 * Note, we do not use a timespec for the user space value here, That
267 * way we can use the function for timeval and 269 * way we can use the function for timeval and compat interfaces as well.
268 * 270 *
269 * Returns -EINVAL if sec/nsec are not normali 271 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
270 */ 272 */
271 int poll_select_set_timeout(struct timespec64 273 int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec)
272 { 274 {
273 struct timespec64 ts = {.tv_sec = sec, 275 struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec};
274 276
275 if (!timespec64_valid(&ts)) 277 if (!timespec64_valid(&ts))
276 return -EINVAL; 278 return -EINVAL;
277 279
278 /* Optimize for the zero timeout value 280 /* Optimize for the zero timeout value here */
279 if (!sec && !nsec) { 281 if (!sec && !nsec) {
280 to->tv_sec = to->tv_nsec = 0; 282 to->tv_sec = to->tv_nsec = 0;
281 } else { 283 } else {
282 ktime_get_ts64(to); 284 ktime_get_ts64(to);
283 *to = timespec64_add_safe(*to, 285 *to = timespec64_add_safe(*to, ts);
284 } 286 }
285 return 0; 287 return 0;
286 } 288 }
287 289
288 enum poll_time_type { 290 enum poll_time_type {
289 PT_TIMEVAL = 0, 291 PT_TIMEVAL = 0,
290 PT_OLD_TIMEVAL = 1, 292 PT_OLD_TIMEVAL = 1,
291 PT_TIMESPEC = 2, 293 PT_TIMESPEC = 2,
292 PT_OLD_TIMESPEC = 3, 294 PT_OLD_TIMESPEC = 3,
293 }; 295 };
294 296
295 static int poll_select_finish(struct timespec6 297 static int poll_select_finish(struct timespec64 *end_time,
296 void __user *p, 298 void __user *p,
297 enum poll_time_t 299 enum poll_time_type pt_type, int ret)
298 { 300 {
299 struct timespec64 rts; 301 struct timespec64 rts;
300 302
301 restore_saved_sigmask_unless(ret == -E 303 restore_saved_sigmask_unless(ret == -ERESTARTNOHAND);
302 304
303 if (!p) 305 if (!p)
304 return ret; 306 return ret;
305 307
306 if (current->personality & STICKY_TIME 308 if (current->personality & STICKY_TIMEOUTS)
307 goto sticky; 309 goto sticky;
308 310
309 /* No update for zero timeout */ 311 /* No update for zero timeout */
310 if (!end_time->tv_sec && !end_time->tv 312 if (!end_time->tv_sec && !end_time->tv_nsec)
311 return ret; 313 return ret;
312 314
313 ktime_get_ts64(&rts); 315 ktime_get_ts64(&rts);
314 rts = timespec64_sub(*end_time, rts); 316 rts = timespec64_sub(*end_time, rts);
315 if (rts.tv_sec < 0) 317 if (rts.tv_sec < 0)
316 rts.tv_sec = rts.tv_nsec = 0; 318 rts.tv_sec = rts.tv_nsec = 0;
317 319
318 320
319 switch (pt_type) { 321 switch (pt_type) {
320 case PT_TIMEVAL: 322 case PT_TIMEVAL:
321 { 323 {
322 struct __kernel_old_ti !! 324 struct timeval rtv;
323 325
324 if (sizeof(rtv) > size 326 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
325 memset(&rtv, 0 327 memset(&rtv, 0, sizeof(rtv));
326 rtv.tv_sec = rts.tv_se 328 rtv.tv_sec = rts.tv_sec;
327 rtv.tv_usec = rts.tv_n 329 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
328 if (!copy_to_user(p, & 330 if (!copy_to_user(p, &rtv, sizeof(rtv)))
329 return ret; 331 return ret;
330 } 332 }
331 break; 333 break;
332 case PT_OLD_TIMEVAL: 334 case PT_OLD_TIMEVAL:
333 { 335 {
334 struct old_timeval32 r 336 struct old_timeval32 rtv;
335 337
336 rtv.tv_sec = rts.tv_se 338 rtv.tv_sec = rts.tv_sec;
337 rtv.tv_usec = rts.tv_n 339 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
338 if (!copy_to_user(p, & 340 if (!copy_to_user(p, &rtv, sizeof(rtv)))
339 return ret; 341 return ret;
340 } 342 }
341 break; 343 break;
342 case PT_TIMESPEC: 344 case PT_TIMESPEC:
343 if (!put_timespec64(&rts, p)) 345 if (!put_timespec64(&rts, p))
344 return ret; 346 return ret;
345 break; 347 break;
346 case PT_OLD_TIMESPEC: 348 case PT_OLD_TIMESPEC:
347 if (!put_old_timespec32(&rts, 349 if (!put_old_timespec32(&rts, p))
348 return ret; 350 return ret;
349 break; 351 break;
350 default: 352 default:
351 BUG(); 353 BUG();
352 } 354 }
353 /* 355 /*
354 * If an application puts its timeval 356 * If an application puts its timeval in read-only memory, we
355 * don't want the Linux-specific updat 357 * don't want the Linux-specific update to the timeval to
356 * cause a fault after the select has 358 * cause a fault after the select has completed
357 * successfully. However, because we'r 359 * successfully. However, because we're not updating the
358 * timeval, we can't restart the syste 360 * timeval, we can't restart the system call.
359 */ 361 */
360 362
361 sticky: 363 sticky:
362 if (ret == -ERESTARTNOHAND) 364 if (ret == -ERESTARTNOHAND)
363 ret = -EINTR; 365 ret = -EINTR;
364 return ret; 366 return ret;
365 } 367 }
366 368
367 /* 369 /*
368 * Scalable version of the fd_set. 370 * Scalable version of the fd_set.
369 */ 371 */
370 372
371 typedef struct { 373 typedef struct {
372 unsigned long *in, *out, *ex; 374 unsigned long *in, *out, *ex;
373 unsigned long *res_in, *res_out, *res_ 375 unsigned long *res_in, *res_out, *res_ex;
374 } fd_set_bits; 376 } fd_set_bits;
375 377
376 /* 378 /*
377 * How many longwords for "nr" bits? 379 * How many longwords for "nr" bits?
378 */ 380 */
379 #define FDS_BITPERLONG (8*sizeof(long)) 381 #define FDS_BITPERLONG (8*sizeof(long))
380 #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG- 382 #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG)
381 #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof( 383 #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long))
382 384
383 /* 385 /*
384 * Use "unsigned long" accesses to let user-mo 386 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned.
385 */ 387 */
386 static inline 388 static inline
387 int get_fd_set(unsigned long nr, void __user * 389 int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
388 { 390 {
389 nr = FDS_BYTES(nr); 391 nr = FDS_BYTES(nr);
390 if (ufdset) 392 if (ufdset)
391 return copy_from_user(fdset, u 393 return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0;
392 394
393 memset(fdset, 0, nr); 395 memset(fdset, 0, nr);
394 return 0; 396 return 0;
395 } 397 }
396 398
397 static inline unsigned long __must_check 399 static inline unsigned long __must_check
398 set_fd_set(unsigned long nr, void __user *ufds 400 set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset)
399 { 401 {
400 if (ufdset) 402 if (ufdset)
401 return __copy_to_user(ufdset, 403 return __copy_to_user(ufdset, fdset, FDS_BYTES(nr));
402 return 0; 404 return 0;
403 } 405 }
404 406
405 static inline 407 static inline
406 void zero_fd_set(unsigned long nr, unsigned lo 408 void zero_fd_set(unsigned long nr, unsigned long *fdset)
407 { 409 {
408 memset(fdset, 0, FDS_BYTES(nr)); 410 memset(fdset, 0, FDS_BYTES(nr));
409 } 411 }
410 412
411 #define FDS_IN(fds, n) (fds->in + n) 413 #define FDS_IN(fds, n) (fds->in + n)
412 #define FDS_OUT(fds, n) (fds->out + n) 414 #define FDS_OUT(fds, n) (fds->out + n)
413 #define FDS_EX(fds, n) (fds->ex + n) 415 #define FDS_EX(fds, n) (fds->ex + n)
414 416
415 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_ 417 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
416 418
417 static int max_select_fd(unsigned long n, fd_s 419 static int max_select_fd(unsigned long n, fd_set_bits *fds)
418 { 420 {
419 unsigned long *open_fds; 421 unsigned long *open_fds;
420 unsigned long set; 422 unsigned long set;
421 int max; 423 int max;
422 struct fdtable *fdt; 424 struct fdtable *fdt;
423 425
424 /* handle last in-complete long-word f 426 /* handle last in-complete long-word first */
425 set = ~(~0UL << (n & (BITS_PER_LONG-1) 427 set = ~(~0UL << (n & (BITS_PER_LONG-1)));
426 n /= BITS_PER_LONG; 428 n /= BITS_PER_LONG;
427 fdt = files_fdtable(current->files); 429 fdt = files_fdtable(current->files);
428 open_fds = fdt->open_fds + n; 430 open_fds = fdt->open_fds + n;
429 max = 0; 431 max = 0;
430 if (set) { 432 if (set) {
431 set &= BITS(fds, n); 433 set &= BITS(fds, n);
432 if (set) { 434 if (set) {
433 if (!(set & ~*open_fds 435 if (!(set & ~*open_fds))
434 goto get_max; 436 goto get_max;
435 return -EBADF; 437 return -EBADF;
436 } 438 }
437 } 439 }
438 while (n) { 440 while (n) {
439 open_fds--; 441 open_fds--;
440 n--; 442 n--;
441 set = BITS(fds, n); 443 set = BITS(fds, n);
442 if (!set) 444 if (!set)
443 continue; 445 continue;
444 if (set & ~*open_fds) 446 if (set & ~*open_fds)
445 return -EBADF; 447 return -EBADF;
446 if (max) 448 if (max)
447 continue; 449 continue;
448 get_max: 450 get_max:
449 do { 451 do {
450 max++; 452 max++;
451 set >>= 1; 453 set >>= 1;
452 } while (set); 454 } while (set);
453 max += n * BITS_PER_LONG; 455 max += n * BITS_PER_LONG;
454 } 456 }
455 457
456 return max; 458 return max;
457 } 459 }
458 460
459 #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND !! 461 #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR)
460 EPOLLNVAL) !! 462 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR)
461 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM !! 463 #define POLLEX_SET (EPOLLPRI)
462 EPOLLNVAL) <<
463 #define POLLEX_SET (EPOLLPRI | EPOLLNVAL) <<
464 464
465 static inline void wait_key_set(poll_table *wa 465 static inline void wait_key_set(poll_table *wait, unsigned long in,
466 unsigned long 466 unsigned long out, unsigned long bit,
467 __poll_t ll_fl 467 __poll_t ll_flag)
468 { 468 {
469 wait->_key = POLLEX_SET | ll_flag; 469 wait->_key = POLLEX_SET | ll_flag;
470 if (in & bit) 470 if (in & bit)
471 wait->_key |= POLLIN_SET; 471 wait->_key |= POLLIN_SET;
472 if (out & bit) 472 if (out & bit)
473 wait->_key |= POLLOUT_SET; 473 wait->_key |= POLLOUT_SET;
474 } 474 }
475 475
476 static noinline_for_stack int do_select(int n, !! 476 static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time)
477 { 477 {
478 ktime_t expire, *to = NULL; 478 ktime_t expire, *to = NULL;
479 struct poll_wqueues table; 479 struct poll_wqueues table;
480 poll_table *wait; 480 poll_table *wait;
481 int retval, i, timed_out = 0; 481 int retval, i, timed_out = 0;
482 u64 slack = 0; 482 u64 slack = 0;
483 __poll_t busy_flag = net_busy_loop_on( 483 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
484 unsigned long busy_start = 0; 484 unsigned long busy_start = 0;
485 485
486 rcu_read_lock(); 486 rcu_read_lock();
487 retval = max_select_fd(n, fds); 487 retval = max_select_fd(n, fds);
488 rcu_read_unlock(); 488 rcu_read_unlock();
489 489
490 if (retval < 0) 490 if (retval < 0)
491 return retval; 491 return retval;
492 n = retval; 492 n = retval;
493 493
494 poll_initwait(&table); 494 poll_initwait(&table);
495 wait = &table.pt; 495 wait = &table.pt;
496 if (end_time && !end_time->tv_sec && ! 496 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
497 wait->_qproc = NULL; 497 wait->_qproc = NULL;
498 timed_out = 1; 498 timed_out = 1;
499 } 499 }
500 500
501 if (end_time && !timed_out) 501 if (end_time && !timed_out)
502 slack = select_estimate_accura 502 slack = select_estimate_accuracy(end_time);
503 503
504 retval = 0; 504 retval = 0;
505 for (;;) { 505 for (;;) {
506 unsigned long *rinp, *routp, * 506 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
507 bool can_busy_loop = false; 507 bool can_busy_loop = false;
508 508
509 inp = fds->in; outp = fds->out 509 inp = fds->in; outp = fds->out; exp = fds->ex;
510 rinp = fds->res_in; routp = fd 510 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
511 511
512 for (i = 0; i < n; ++rinp, ++r 512 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
513 unsigned long in, out, 513 unsigned long in, out, ex, all_bits, bit = 1, j;
514 unsigned long res_in = 514 unsigned long res_in = 0, res_out = 0, res_ex = 0;
515 __poll_t mask; 515 __poll_t mask;
516 516
517 in = *inp++; out = *ou 517 in = *inp++; out = *outp++; ex = *exp++;
518 all_bits = in | out | 518 all_bits = in | out | ex;
519 if (all_bits == 0) { 519 if (all_bits == 0) {
520 i += BITS_PER_ 520 i += BITS_PER_LONG;
521 continue; 521 continue;
522 } 522 }
523 523
524 for (j = 0; j < BITS_P 524 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
525 struct fd f; 525 struct fd f;
526 if (i >= n) 526 if (i >= n)
527 break; 527 break;
528 if (!(bit & al 528 if (!(bit & all_bits))
529 contin 529 continue;
530 mask = EPOLLNV <<
531 f = fdget(i); 530 f = fdget(i);
532 if (fd_file(f) !! 531 if (f.file) {
533 wait_k 532 wait_key_set(wait, in, out, bit,
534 533 busy_flag);
535 mask = !! 534 mask = vfs_poll(f.file, wait);
536 535
537 fdput( 536 fdput(f);
538 } !! 537 if ((mask & POLLIN_SET) && (in & bit)) {
539 if ((mask & PO !! 538 res_in |= bit;
540 res_in !! 539 retval++;
541 retval !! 540 wait->_qproc = NULL;
542 wait-> !! 541 }
543 } !! 542 if ((mask & POLLOUT_SET) && (out & bit)) {
544 if ((mask & PO !! 543 res_out |= bit;
545 res_ou !! 544 retval++;
546 retval !! 545 wait->_qproc = NULL;
547 wait-> !! 546 }
548 } !! 547 if ((mask & POLLEX_SET) && (ex & bit)) {
549 if ((mask & PO !! 548 res_ex |= bit;
550 res_ex !! 549 retval++;
551 retval !! 550 wait->_qproc = NULL;
552 wait-> !! 551 }
553 } !! 552 /* got something, stop busy polling */
554 /* got somethi !! 553 if (retval) {
555 if (retval) { !! 554 can_busy_loop = false;
556 can_bu !! 555 busy_flag = 0;
557 busy_f !! 556
558 !! 557 /*
559 /* !! 558 * only remember a returned
560 * only rememb !! 559 * POLL_BUSY_LOOP if we asked for it
561 * POLL_BUSY_L !! 560 */
562 */ !! 561 } else if (busy_flag & mask)
563 } else if (bus !! 562 can_busy_loop = true;
564 can_bu <<
565 563
>> 564 }
566 } 565 }
567 if (res_in) 566 if (res_in)
568 *rinp = res_in 567 *rinp = res_in;
569 if (res_out) 568 if (res_out)
570 *routp = res_o 569 *routp = res_out;
571 if (res_ex) 570 if (res_ex)
572 *rexp = res_ex 571 *rexp = res_ex;
573 cond_resched(); 572 cond_resched();
574 } 573 }
575 wait->_qproc = NULL; 574 wait->_qproc = NULL;
576 if (retval || timed_out || sig 575 if (retval || timed_out || signal_pending(current))
577 break; 576 break;
578 if (table.error) { 577 if (table.error) {
579 retval = table.error; 578 retval = table.error;
580 break; 579 break;
581 } 580 }
582 581
583 /* only if found POLL_BUSY_LOO 582 /* only if found POLL_BUSY_LOOP sockets && not out of time */
584 if (can_busy_loop && !need_res 583 if (can_busy_loop && !need_resched()) {
585 if (!busy_start) { 584 if (!busy_start) {
586 busy_start = b 585 busy_start = busy_loop_current_time();
587 continue; 586 continue;
588 } 587 }
589 if (!busy_loop_timeout 588 if (!busy_loop_timeout(busy_start))
590 continue; 589 continue;
591 } 590 }
592 busy_flag = 0; 591 busy_flag = 0;
593 592
594 /* 593 /*
595 * If this is the first loop a 594 * If this is the first loop and we have a timeout
596 * given, then we convert to k 595 * given, then we convert to ktime_t and set the to
597 * pointer to the expiry value 596 * pointer to the expiry value.
598 */ 597 */
599 if (end_time && !to) { 598 if (end_time && !to) {
600 expire = timespec64_to 599 expire = timespec64_to_ktime(*end_time);
601 to = &expire; 600 to = &expire;
602 } 601 }
603 602
604 if (!poll_schedule_timeout(&ta 603 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
605 to, 604 to, slack))
606 timed_out = 1; 605 timed_out = 1;
607 } 606 }
608 607
609 poll_freewait(&table); 608 poll_freewait(&table);
610 609
611 return retval; 610 return retval;
612 } 611 }
613 612
614 /* 613 /*
615 * We can actually return ERESTARTSYS instead 614 * We can actually return ERESTARTSYS instead of EINTR, but I'd
616 * like to be certain this leads to no problem 615 * like to be certain this leads to no problems. So I return
617 * EINTR just for safety. 616 * EINTR just for safety.
618 * 617 *
619 * Update: ERESTARTSYS breaks at least the xvi 618 * Update: ERESTARTSYS breaks at least the xview clock binary, so
620 * I'm trying ERESTARTNOHAND which restart onl 619 * I'm trying ERESTARTNOHAND which restart only when you want to.
621 */ 620 */
622 int core_sys_select(int n, fd_set __user *inp, 621 int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
623 fd_set __user *exp, 622 fd_set __user *exp, struct timespec64 *end_time)
624 { 623 {
625 fd_set_bits fds; 624 fd_set_bits fds;
626 void *bits; 625 void *bits;
627 int ret, max_fds; 626 int ret, max_fds;
628 size_t size, alloc_size; 627 size_t size, alloc_size;
629 struct fdtable *fdt; 628 struct fdtable *fdt;
630 /* Allocate small arguments on the sta 629 /* Allocate small arguments on the stack to save memory and be faster */
631 long stack_fds[SELECT_STACK_ALLOC/size 630 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
632 631
633 ret = -EINVAL; 632 ret = -EINVAL;
634 if (n < 0) 633 if (n < 0)
635 goto out_nofds; 634 goto out_nofds;
636 635
637 /* max_fds can increase, so grab it on 636 /* max_fds can increase, so grab it once to avoid race */
638 rcu_read_lock(); 637 rcu_read_lock();
639 fdt = files_fdtable(current->files); 638 fdt = files_fdtable(current->files);
640 max_fds = fdt->max_fds; 639 max_fds = fdt->max_fds;
641 rcu_read_unlock(); 640 rcu_read_unlock();
642 if (n > max_fds) 641 if (n > max_fds)
643 n = max_fds; 642 n = max_fds;
644 643
645 /* 644 /*
646 * We need 6 bitmaps (in/out/ex for bo 645 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
647 * since we used fdset we need to allo 646 * since we used fdset we need to allocate memory in units of
648 * long-words. 647 * long-words.
649 */ 648 */
650 size = FDS_BYTES(n); 649 size = FDS_BYTES(n);
651 bits = stack_fds; 650 bits = stack_fds;
652 if (size > sizeof(stack_fds) / 6) { 651 if (size > sizeof(stack_fds) / 6) {
653 /* Not enough space in on-stac 652 /* Not enough space in on-stack array; must use kmalloc */
654 ret = -ENOMEM; 653 ret = -ENOMEM;
655 if (size > (SIZE_MAX / 6)) 654 if (size > (SIZE_MAX / 6))
656 goto out_nofds; 655 goto out_nofds;
657 656
658 alloc_size = 6 * size; 657 alloc_size = 6 * size;
659 bits = kvmalloc(alloc_size, GF 658 bits = kvmalloc(alloc_size, GFP_KERNEL);
660 if (!bits) 659 if (!bits)
661 goto out_nofds; 660 goto out_nofds;
662 } 661 }
663 fds.in = bits; 662 fds.in = bits;
664 fds.out = bits + size; 663 fds.out = bits + size;
665 fds.ex = bits + 2*size; 664 fds.ex = bits + 2*size;
666 fds.res_in = bits + 3*size; 665 fds.res_in = bits + 3*size;
667 fds.res_out = bits + 4*size; 666 fds.res_out = bits + 4*size;
668 fds.res_ex = bits + 5*size; 667 fds.res_ex = bits + 5*size;
669 668
670 if ((ret = get_fd_set(n, inp, fds.in)) 669 if ((ret = get_fd_set(n, inp, fds.in)) ||
671 (ret = get_fd_set(n, outp, fds.out 670 (ret = get_fd_set(n, outp, fds.out)) ||
672 (ret = get_fd_set(n, exp, fds.ex)) 671 (ret = get_fd_set(n, exp, fds.ex)))
673 goto out; 672 goto out;
674 zero_fd_set(n, fds.res_in); 673 zero_fd_set(n, fds.res_in);
675 zero_fd_set(n, fds.res_out); 674 zero_fd_set(n, fds.res_out);
676 zero_fd_set(n, fds.res_ex); 675 zero_fd_set(n, fds.res_ex);
677 676
678 ret = do_select(n, &fds, end_time); 677 ret = do_select(n, &fds, end_time);
679 678
680 if (ret < 0) 679 if (ret < 0)
681 goto out; 680 goto out;
682 if (!ret) { 681 if (!ret) {
683 ret = -ERESTARTNOHAND; 682 ret = -ERESTARTNOHAND;
684 if (signal_pending(current)) 683 if (signal_pending(current))
685 goto out; 684 goto out;
686 ret = 0; 685 ret = 0;
687 } 686 }
688 687
689 if (set_fd_set(n, inp, fds.res_in) || 688 if (set_fd_set(n, inp, fds.res_in) ||
690 set_fd_set(n, outp, fds.res_out) | 689 set_fd_set(n, outp, fds.res_out) ||
691 set_fd_set(n, exp, fds.res_ex)) 690 set_fd_set(n, exp, fds.res_ex))
692 ret = -EFAULT; 691 ret = -EFAULT;
693 692
694 out: 693 out:
695 if (bits != stack_fds) 694 if (bits != stack_fds)
696 kvfree(bits); 695 kvfree(bits);
697 out_nofds: 696 out_nofds:
698 return ret; 697 return ret;
699 } 698 }
700 699
701 static int kern_select(int n, fd_set __user *i 700 static int kern_select(int n, fd_set __user *inp, fd_set __user *outp,
702 fd_set __user *exp, str !! 701 fd_set __user *exp, struct timeval __user *tvp)
703 { 702 {
704 struct timespec64 end_time, *to = NULL 703 struct timespec64 end_time, *to = NULL;
705 struct __kernel_old_timeval tv; !! 704 struct timeval tv;
706 int ret; 705 int ret;
707 706
708 if (tvp) { 707 if (tvp) {
709 if (copy_from_user(&tv, tvp, s 708 if (copy_from_user(&tv, tvp, sizeof(tv)))
710 return -EFAULT; 709 return -EFAULT;
711 710
712 to = &end_time; 711 to = &end_time;
713 if (poll_select_set_timeout(to 712 if (poll_select_set_timeout(to,
714 tv.tv_sec + (t 713 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
715 (tv.tv_usec % 714 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
716 return -EINVAL; 715 return -EINVAL;
717 } 716 }
718 717
719 ret = core_sys_select(n, inp, outp, ex 718 ret = core_sys_select(n, inp, outp, exp, to);
720 return poll_select_finish(&end_time, t 719 return poll_select_finish(&end_time, tvp, PT_TIMEVAL, ret);
721 } 720 }
722 721
723 SYSCALL_DEFINE5(select, int, n, fd_set __user 722 SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
724 fd_set __user *, exp, struct _ !! 723 fd_set __user *, exp, struct timeval __user *, tvp)
725 { 724 {
726 return kern_select(n, inp, outp, exp, 725 return kern_select(n, inp, outp, exp, tvp);
727 } 726 }
728 727
729 static long do_pselect(int n, fd_set __user *i 728 static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
730 fd_set __user *exp, voi 729 fd_set __user *exp, void __user *tsp,
731 const sigset_t __user * 730 const sigset_t __user *sigmask, size_t sigsetsize,
732 enum poll_time_type typ 731 enum poll_time_type type)
733 { 732 {
734 struct timespec64 ts, end_time, *to = 733 struct timespec64 ts, end_time, *to = NULL;
735 int ret; 734 int ret;
736 735
737 if (tsp) { 736 if (tsp) {
738 switch (type) { 737 switch (type) {
739 case PT_TIMESPEC: 738 case PT_TIMESPEC:
740 if (get_timespec64(&ts 739 if (get_timespec64(&ts, tsp))
741 return -EFAULT 740 return -EFAULT;
742 break; 741 break;
743 case PT_OLD_TIMESPEC: 742 case PT_OLD_TIMESPEC:
744 if (get_old_timespec32 743 if (get_old_timespec32(&ts, tsp))
745 return -EFAULT 744 return -EFAULT;
746 break; 745 break;
747 default: 746 default:
748 BUG(); 747 BUG();
749 } 748 }
750 749
751 to = &end_time; 750 to = &end_time;
752 if (poll_select_set_timeout(to 751 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
753 return -EINVAL; 752 return -EINVAL;
754 } 753 }
755 754
756 ret = set_user_sigmask(sigmask, sigset 755 ret = set_user_sigmask(sigmask, sigsetsize);
757 if (ret) 756 if (ret)
758 return ret; 757 return ret;
759 758
760 ret = core_sys_select(n, inp, outp, ex 759 ret = core_sys_select(n, inp, outp, exp, to);
761 return poll_select_finish(&end_time, t 760 return poll_select_finish(&end_time, tsp, type, ret);
762 } 761 }
763 762
764 /* 763 /*
765 * Most architectures can't handle 7-argument 764 * Most architectures can't handle 7-argument syscalls. So we provide a
766 * 6-argument version where the sixth argument 765 * 6-argument version where the sixth argument is a pointer to a structure
767 * which has a pointer to the sigset_t itself 766 * which has a pointer to the sigset_t itself followed by a size_t containing
768 * the sigset size. 767 * the sigset size.
769 */ 768 */
770 struct sigset_argpack { <<
771 sigset_t __user *p; <<
772 size_t size; <<
773 }; <<
774 <<
775 static inline int get_sigset_argpack(struct si <<
776 struct si <<
777 { <<
778 // the path is hot enough for overhead <<
779 if (from) { <<
780 if (can_do_masked_user_access( <<
781 from = masked_user_acc <<
782 else if (!user_read_access_beg <<
783 return -EFAULT; <<
784 unsafe_get_user(to->p, &from-> <<
785 unsafe_get_user(to->size, &fro <<
786 user_read_access_end(); <<
787 } <<
788 return 0; <<
789 Efault: <<
790 user_access_end(); <<
791 return -EFAULT; <<
792 } <<
793 <<
794 SYSCALL_DEFINE6(pselect6, int, n, fd_set __use 769 SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
795 fd_set __user *, exp, struct _ 770 fd_set __user *, exp, struct __kernel_timespec __user *, tsp,
796 void __user *, sig) 771 void __user *, sig)
797 { 772 {
798 struct sigset_argpack x = {NULL, 0}; !! 773 size_t sigsetsize = 0;
>> 774 sigset_t __user *up = NULL;
799 775
800 if (get_sigset_argpack(&x, sig)) !! 776 if (sig) {
801 return -EFAULT; !! 777 if (!access_ok(sig, sizeof(void *)+sizeof(size_t))
>> 778 || __get_user(up, (sigset_t __user * __user *)sig)
>> 779 || __get_user(sigsetsize,
>> 780 (size_t __user *)(sig+sizeof(void *))))
>> 781 return -EFAULT;
>> 782 }
802 783
803 return do_pselect(n, inp, outp, exp, t !! 784 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize, PT_TIMESPEC);
804 } 785 }
805 786
806 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defi 787 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
807 788
808 SYSCALL_DEFINE6(pselect6_time32, int, n, fd_se 789 SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp,
809 fd_set __user *, exp, struct o 790 fd_set __user *, exp, struct old_timespec32 __user *, tsp,
810 void __user *, sig) 791 void __user *, sig)
811 { 792 {
812 struct sigset_argpack x = {NULL, 0}; !! 793 size_t sigsetsize = 0;
>> 794 sigset_t __user *up = NULL;
813 795
814 if (get_sigset_argpack(&x, sig)) !! 796 if (sig) {
815 return -EFAULT; !! 797 if (!access_ok(sig, sizeof(void *)+sizeof(size_t))
>> 798 || __get_user(up, (sigset_t __user * __user *)sig)
>> 799 || __get_user(sigsetsize,
>> 800 (size_t __user *)(sig+sizeof(void *))))
>> 801 return -EFAULT;
>> 802 }
816 803
817 return do_pselect(n, inp, outp, exp, t !! 804 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize, PT_OLD_TIMESPEC);
818 } 805 }
819 806
820 #endif 807 #endif
821 808
822 #ifdef __ARCH_WANT_SYS_OLD_SELECT 809 #ifdef __ARCH_WANT_SYS_OLD_SELECT
823 struct sel_arg_struct { 810 struct sel_arg_struct {
824 unsigned long n; 811 unsigned long n;
825 fd_set __user *inp, *outp, *exp; 812 fd_set __user *inp, *outp, *exp;
826 struct __kernel_old_timeval __user *tv !! 813 struct timeval __user *tvp;
827 }; 814 };
828 815
829 SYSCALL_DEFINE1(old_select, struct sel_arg_str 816 SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
830 { 817 {
831 struct sel_arg_struct a; 818 struct sel_arg_struct a;
832 819
833 if (copy_from_user(&a, arg, sizeof(a)) 820 if (copy_from_user(&a, arg, sizeof(a)))
834 return -EFAULT; 821 return -EFAULT;
835 return kern_select(a.n, a.inp, a.outp, 822 return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp);
836 } 823 }
837 #endif 824 #endif
838 825
839 struct poll_list { 826 struct poll_list {
840 struct poll_list *next; 827 struct poll_list *next;
841 unsigned int len; !! 828 int len;
842 struct pollfd entries[] __counted_by(l !! 829 struct pollfd entries[0];
843 }; 830 };
844 831
845 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(st 832 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
846 833
847 /* 834 /*
848 * Fish for pollable events on the pollfd->fd 835 * Fish for pollable events on the pollfd->fd file descriptor. We're only
849 * interested in events matching the pollfd->e 836 * interested in events matching the pollfd->events mask, and the result
850 * matching that mask is both recorded in poll 837 * matching that mask is both recorded in pollfd->revents and returned. The
851 * pwait poll_table will be used by the fd-pro 838 * pwait poll_table will be used by the fd-provided poll handler for waiting,
852 * if pwait->_qproc is non-NULL. 839 * if pwait->_qproc is non-NULL.
853 */ 840 */
854 static inline __poll_t do_pollfd(struct pollfd 841 static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait,
855 bool *can 842 bool *can_busy_poll,
856 __poll_t 843 __poll_t busy_flag)
857 { 844 {
858 int fd = pollfd->fd; 845 int fd = pollfd->fd;
859 __poll_t mask = 0, filter; 846 __poll_t mask = 0, filter;
860 struct fd f; 847 struct fd f;
861 848
862 if (fd < 0) 849 if (fd < 0)
863 goto out; 850 goto out;
864 mask = EPOLLNVAL; 851 mask = EPOLLNVAL;
865 f = fdget(fd); 852 f = fdget(fd);
866 if (!fd_file(f)) !! 853 if (!f.file)
867 goto out; 854 goto out;
868 855
869 /* userland u16 ->events contains POLL 856 /* userland u16 ->events contains POLL... bitmap */
870 filter = demangle_poll(pollfd->events) 857 filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP;
871 pwait->_key = filter | busy_flag; 858 pwait->_key = filter | busy_flag;
872 mask = vfs_poll(fd_file(f), pwait); !! 859 mask = vfs_poll(f.file, pwait);
873 if (mask & busy_flag) 860 if (mask & busy_flag)
874 *can_busy_poll = true; 861 *can_busy_poll = true;
875 mask &= filter; /* Mask out un 862 mask &= filter; /* Mask out unneeded events. */
876 fdput(f); 863 fdput(f);
877 864
878 out: 865 out:
879 /* ... and so does ->revents */ 866 /* ... and so does ->revents */
880 pollfd->revents = mangle_poll(mask); 867 pollfd->revents = mangle_poll(mask);
881 return mask; 868 return mask;
882 } 869 }
883 870
884 static int do_poll(struct poll_list *list, str 871 static int do_poll(struct poll_list *list, struct poll_wqueues *wait,
885 struct timespec64 *end_time 872 struct timespec64 *end_time)
886 { 873 {
887 poll_table* pt = &wait->pt; 874 poll_table* pt = &wait->pt;
888 ktime_t expire, *to = NULL; 875 ktime_t expire, *to = NULL;
889 int timed_out = 0, count = 0; 876 int timed_out = 0, count = 0;
890 u64 slack = 0; 877 u64 slack = 0;
891 __poll_t busy_flag = net_busy_loop_on( 878 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0;
892 unsigned long busy_start = 0; 879 unsigned long busy_start = 0;
893 880
894 /* Optimise the no-wait case */ 881 /* Optimise the no-wait case */
895 if (end_time && !end_time->tv_sec && ! 882 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
896 pt->_qproc = NULL; 883 pt->_qproc = NULL;
897 timed_out = 1; 884 timed_out = 1;
898 } 885 }
899 886
900 if (end_time && !timed_out) 887 if (end_time && !timed_out)
901 slack = select_estimate_accura 888 slack = select_estimate_accuracy(end_time);
902 889
903 for (;;) { 890 for (;;) {
904 struct poll_list *walk; 891 struct poll_list *walk;
905 bool can_busy_loop = false; 892 bool can_busy_loop = false;
906 893
907 for (walk = list; walk != NULL 894 for (walk = list; walk != NULL; walk = walk->next) {
908 struct pollfd * pfd, * 895 struct pollfd * pfd, * pfd_end;
909 896
910 pfd = walk->entries; 897 pfd = walk->entries;
911 pfd_end = pfd + walk-> 898 pfd_end = pfd + walk->len;
912 for (; pfd != pfd_end; 899 for (; pfd != pfd_end; pfd++) {
913 /* 900 /*
914 * Fish for ev 901 * Fish for events. If we found one, record it
915 * and kill po 902 * and kill poll_table->_qproc, so we don't
916 * needlessly 903 * needlessly register any other waiters after
917 * this. They' 904 * this. They'll get immediately deregistered
918 * when we bre 905 * when we break out and return.
919 */ 906 */
920 if (do_pollfd( 907 if (do_pollfd(pfd, pt, &can_busy_loop,
921 908 busy_flag)) {
922 count+ 909 count++;
923 pt->_q 910 pt->_qproc = NULL;
924 /* fou 911 /* found something, stop busy polling */
925 busy_f 912 busy_flag = 0;
926 can_bu 913 can_busy_loop = false;
927 } 914 }
928 } 915 }
929 } 916 }
930 /* 917 /*
931 * All waiters have already be 918 * All waiters have already been registered, so don't provide
932 * a poll_table->_qproc to the 919 * a poll_table->_qproc to them on the next loop iteration.
933 */ 920 */
934 pt->_qproc = NULL; 921 pt->_qproc = NULL;
935 if (!count) { 922 if (!count) {
936 count = wait->error; 923 count = wait->error;
937 if (signal_pending(cur 924 if (signal_pending(current))
938 count = -EREST 925 count = -ERESTARTNOHAND;
939 } 926 }
940 if (count || timed_out) 927 if (count || timed_out)
941 break; 928 break;
942 929
943 /* only if found POLL_BUSY_LOO 930 /* only if found POLL_BUSY_LOOP sockets && not out of time */
944 if (can_busy_loop && !need_res 931 if (can_busy_loop && !need_resched()) {
945 if (!busy_start) { 932 if (!busy_start) {
946 busy_start = b 933 busy_start = busy_loop_current_time();
947 continue; 934 continue;
948 } 935 }
949 if (!busy_loop_timeout 936 if (!busy_loop_timeout(busy_start))
950 continue; 937 continue;
951 } 938 }
952 busy_flag = 0; 939 busy_flag = 0;
953 940
954 /* 941 /*
955 * If this is the first loop a 942 * If this is the first loop and we have a timeout
956 * given, then we convert to k 943 * given, then we convert to ktime_t and set the to
957 * pointer to the expiry value 944 * pointer to the expiry value.
958 */ 945 */
959 if (end_time && !to) { 946 if (end_time && !to) {
960 expire = timespec64_to 947 expire = timespec64_to_ktime(*end_time);
961 to = &expire; 948 to = &expire;
962 } 949 }
963 950
964 if (!poll_schedule_timeout(wai 951 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
965 timed_out = 1; 952 timed_out = 1;
966 } 953 }
967 return count; 954 return count;
968 } 955 }
969 956
970 #define N_STACK_PPS ((sizeof(stack_pps) - size 957 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \
971 sizeof(struct pollfd)) 958 sizeof(struct pollfd))
972 959
973 static int do_sys_poll(struct pollfd __user *u 960 static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
974 struct timespec64 *end_time) 961 struct timespec64 *end_time)
975 { 962 {
976 struct poll_wqueues table; 963 struct poll_wqueues table;
977 int err = -EFAULT, fdcount; !! 964 int err = -EFAULT, fdcount, len;
978 /* Allocate small arguments on the sta 965 /* Allocate small arguments on the stack to save memory and be
979 faster - use long to make sure the 966 faster - use long to make sure the buffer is aligned properly
980 on 64 bit archs to avoid unaligned 967 on 64 bit archs to avoid unaligned access */
981 long stack_pps[POLL_STACK_ALLOC/sizeof 968 long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
982 struct poll_list *const head = (struct 969 struct poll_list *const head = (struct poll_list *)stack_pps;
983 struct poll_list *walk = head; 970 struct poll_list *walk = head;
984 unsigned int todo = nfds; !! 971 unsigned long todo = nfds;
985 unsigned int len; <<
986 972
987 if (nfds > rlimit(RLIMIT_NOFILE)) 973 if (nfds > rlimit(RLIMIT_NOFILE))
988 return -EINVAL; 974 return -EINVAL;
989 975
990 len = min_t(unsigned int, nfds, N_STAC 976 len = min_t(unsigned int, nfds, N_STACK_PPS);
991 for (;;) { 977 for (;;) {
992 walk->next = NULL; 978 walk->next = NULL;
993 walk->len = len; 979 walk->len = len;
994 if (!len) 980 if (!len)
995 break; 981 break;
996 982
997 if (copy_from_user(walk->entri 983 if (copy_from_user(walk->entries, ufds + nfds-todo,
998 sizeof 984 sizeof(struct pollfd) * walk->len))
999 goto out_fds; 985 goto out_fds;
1000 986
1001 if (walk->len >= todo) <<
1002 break; <<
1003 todo -= walk->len; 987 todo -= walk->len;
>> 988 if (!todo)
>> 989 break;
1004 990
1005 len = min(todo, POLLFD_PER_PA 991 len = min(todo, POLLFD_PER_PAGE);
1006 walk = walk->next = kmalloc(s 992 walk = walk->next = kmalloc(struct_size(walk, entries, len),
1007 G 993 GFP_KERNEL);
1008 if (!walk) { 994 if (!walk) {
1009 err = -ENOMEM; 995 err = -ENOMEM;
1010 goto out_fds; 996 goto out_fds;
1011 } 997 }
1012 } 998 }
1013 999
1014 poll_initwait(&table); 1000 poll_initwait(&table);
1015 fdcount = do_poll(head, &table, end_t 1001 fdcount = do_poll(head, &table, end_time);
1016 poll_freewait(&table); 1002 poll_freewait(&table);
1017 1003
1018 if (!user_write_access_begin(ufds, nf <<
1019 goto out_fds; <<
1020 <<
1021 for (walk = head; walk; walk = walk-> 1004 for (walk = head; walk; walk = walk->next) {
1022 struct pollfd *fds = walk->en 1005 struct pollfd *fds = walk->entries;
1023 unsigned int j; !! 1006 int j;
1024 1007
1025 for (j = walk->len; j; fds++, !! 1008 for (j = 0; j < walk->len; j++, ufds++)
1026 unsafe_put_user(fds-> !! 1009 if (__put_user(fds[j].revents, &ufds->revents))
>> 1010 goto out_fds;
1027 } 1011 }
1028 user_write_access_end(); <<
1029 1012
1030 err = fdcount; 1013 err = fdcount;
1031 out_fds: 1014 out_fds:
1032 walk = head->next; 1015 walk = head->next;
1033 while (walk) { 1016 while (walk) {
1034 struct poll_list *pos = walk; 1017 struct poll_list *pos = walk;
1035 walk = walk->next; 1018 walk = walk->next;
1036 kfree(pos); 1019 kfree(pos);
1037 } 1020 }
1038 1021
1039 return err; 1022 return err;
1040 <<
1041 Efault: <<
1042 user_write_access_end(); <<
1043 err = -EFAULT; <<
1044 goto out_fds; <<
1045 } 1023 }
1046 1024
1047 static long do_restart_poll(struct restart_bl 1025 static long do_restart_poll(struct restart_block *restart_block)
1048 { 1026 {
1049 struct pollfd __user *ufds = restart_ 1027 struct pollfd __user *ufds = restart_block->poll.ufds;
1050 int nfds = restart_block->poll.nfds; 1028 int nfds = restart_block->poll.nfds;
1051 struct timespec64 *to = NULL, end_tim 1029 struct timespec64 *to = NULL, end_time;
1052 int ret; 1030 int ret;
1053 1031
1054 if (restart_block->poll.has_timeout) 1032 if (restart_block->poll.has_timeout) {
1055 end_time.tv_sec = restart_blo 1033 end_time.tv_sec = restart_block->poll.tv_sec;
1056 end_time.tv_nsec = restart_bl 1034 end_time.tv_nsec = restart_block->poll.tv_nsec;
1057 to = &end_time; 1035 to = &end_time;
1058 } 1036 }
1059 1037
1060 ret = do_sys_poll(ufds, nfds, to); 1038 ret = do_sys_poll(ufds, nfds, to);
1061 1039
1062 if (ret == -ERESTARTNOHAND) !! 1040 if (ret == -ERESTARTNOHAND) {
1063 ret = set_restart_fn(restart_ !! 1041 restart_block->fn = do_restart_poll;
1064 !! 1042 ret = -ERESTART_RESTARTBLOCK;
>> 1043 }
1065 return ret; 1044 return ret;
1066 } 1045 }
1067 1046
1068 SYSCALL_DEFINE3(poll, struct pollfd __user *, 1047 SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
1069 int, timeout_msecs) 1048 int, timeout_msecs)
1070 { 1049 {
1071 struct timespec64 end_time, *to = NUL 1050 struct timespec64 end_time, *to = NULL;
1072 int ret; 1051 int ret;
1073 1052
1074 if (timeout_msecs >= 0) { 1053 if (timeout_msecs >= 0) {
1075 to = &end_time; 1054 to = &end_time;
1076 poll_select_set_timeout(to, t 1055 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
1077 NSEC_PER_MSEC * (time 1056 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
1078 } 1057 }
1079 1058
1080 ret = do_sys_poll(ufds, nfds, to); 1059 ret = do_sys_poll(ufds, nfds, to);
1081 1060
1082 if (ret == -ERESTARTNOHAND) { 1061 if (ret == -ERESTARTNOHAND) {
1083 struct restart_block *restart 1062 struct restart_block *restart_block;
1084 1063
1085 restart_block = ¤t->res 1064 restart_block = ¤t->restart_block;
>> 1065 restart_block->fn = do_restart_poll;
1086 restart_block->poll.ufds = uf 1066 restart_block->poll.ufds = ufds;
1087 restart_block->poll.nfds = nf 1067 restart_block->poll.nfds = nfds;
1088 1068
1089 if (timeout_msecs >= 0) { 1069 if (timeout_msecs >= 0) {
1090 restart_block->poll.t 1070 restart_block->poll.tv_sec = end_time.tv_sec;
1091 restart_block->poll.t 1071 restart_block->poll.tv_nsec = end_time.tv_nsec;
1092 restart_block->poll.h 1072 restart_block->poll.has_timeout = 1;
1093 } else 1073 } else
1094 restart_block->poll.h 1074 restart_block->poll.has_timeout = 0;
1095 1075
1096 ret = set_restart_fn(restart_ !! 1076 ret = -ERESTART_RESTARTBLOCK;
1097 } 1077 }
1098 return ret; 1078 return ret;
1099 } 1079 }
1100 1080
1101 SYSCALL_DEFINE5(ppoll, struct pollfd __user * 1081 SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
1102 struct __kernel_timespec __us 1082 struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask,
1103 size_t, sigsetsize) 1083 size_t, sigsetsize)
1104 { 1084 {
1105 struct timespec64 ts, end_time, *to = 1085 struct timespec64 ts, end_time, *to = NULL;
1106 int ret; 1086 int ret;
1107 1087
1108 if (tsp) { 1088 if (tsp) {
1109 if (get_timespec64(&ts, tsp)) 1089 if (get_timespec64(&ts, tsp))
1110 return -EFAULT; 1090 return -EFAULT;
1111 1091
1112 to = &end_time; 1092 to = &end_time;
1113 if (poll_select_set_timeout(t 1093 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1114 return -EINVAL; 1094 return -EINVAL;
1115 } 1095 }
1116 1096
1117 ret = set_user_sigmask(sigmask, sigse 1097 ret = set_user_sigmask(sigmask, sigsetsize);
1118 if (ret) 1098 if (ret)
1119 return ret; 1099 return ret;
1120 1100
1121 ret = do_sys_poll(ufds, nfds, to); 1101 ret = do_sys_poll(ufds, nfds, to);
1122 return poll_select_finish(&end_time, 1102 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1123 } 1103 }
1124 1104
1125 #if defined(CONFIG_COMPAT_32BIT_TIME) && !def 1105 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT)
1126 1106
1127 SYSCALL_DEFINE5(ppoll_time32, struct pollfd _ 1107 SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds,
1128 struct old_timespec32 __user 1108 struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask,
1129 size_t, sigsetsize) 1109 size_t, sigsetsize)
1130 { 1110 {
1131 struct timespec64 ts, end_time, *to = 1111 struct timespec64 ts, end_time, *to = NULL;
1132 int ret; 1112 int ret;
1133 1113
1134 if (tsp) { 1114 if (tsp) {
1135 if (get_old_timespec32(&ts, t 1115 if (get_old_timespec32(&ts, tsp))
1136 return -EFAULT; 1116 return -EFAULT;
1137 1117
1138 to = &end_time; 1118 to = &end_time;
1139 if (poll_select_set_timeout(t 1119 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1140 return -EINVAL; 1120 return -EINVAL;
1141 } 1121 }
1142 1122
1143 ret = set_user_sigmask(sigmask, sigse 1123 ret = set_user_sigmask(sigmask, sigsetsize);
1144 if (ret) 1124 if (ret)
1145 return ret; 1125 return ret;
1146 1126
1147 ret = do_sys_poll(ufds, nfds, to); 1127 ret = do_sys_poll(ufds, nfds, to);
1148 return poll_select_finish(&end_time, 1128 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1149 } 1129 }
1150 #endif 1130 #endif
1151 1131
1152 #ifdef CONFIG_COMPAT 1132 #ifdef CONFIG_COMPAT
1153 #define __COMPAT_NFDBITS (8 * sizeof(co 1133 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t))
1154 1134
1155 /* 1135 /*
1156 * Ooo, nasty. We need here to frob 32-bit u 1136 * Ooo, nasty. We need here to frob 32-bit unsigned longs to
1157 * 64-bit unsigned longs. 1137 * 64-bit unsigned longs.
1158 */ 1138 */
1159 static 1139 static
1160 int compat_get_fd_set(unsigned long nr, compa 1140 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1161 unsigned long *fdset) 1141 unsigned long *fdset)
1162 { 1142 {
1163 if (ufdset) { 1143 if (ufdset) {
1164 return compat_get_bitmap(fdse 1144 return compat_get_bitmap(fdset, ufdset, nr);
1165 } else { 1145 } else {
1166 zero_fd_set(nr, fdset); 1146 zero_fd_set(nr, fdset);
1167 return 0; 1147 return 0;
1168 } 1148 }
1169 } 1149 }
1170 1150
1171 static 1151 static
1172 int compat_set_fd_set(unsigned long nr, compa 1152 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset,
1173 unsigned long *fdset) 1153 unsigned long *fdset)
1174 { 1154 {
1175 if (!ufdset) 1155 if (!ufdset)
1176 return 0; 1156 return 0;
1177 return compat_put_bitmap(ufdset, fdse 1157 return compat_put_bitmap(ufdset, fdset, nr);
1178 } 1158 }
1179 1159
1180 1160
1181 /* 1161 /*
1182 * This is a virtual copy of sys_select from 1162 * This is a virtual copy of sys_select from fs/select.c and probably
1183 * should be compared to it from time to time 1163 * should be compared to it from time to time
1184 */ 1164 */
1185 1165
1186 /* 1166 /*
1187 * We can actually return ERESTARTSYS instead 1167 * We can actually return ERESTARTSYS instead of EINTR, but I'd
1188 * like to be certain this leads to no proble 1168 * like to be certain this leads to no problems. So I return
1189 * EINTR just for safety. 1169 * EINTR just for safety.
1190 * 1170 *
1191 * Update: ERESTARTSYS breaks at least the xv 1171 * Update: ERESTARTSYS breaks at least the xview clock binary, so
1192 * I'm trying ERESTARTNOHAND which restart on 1172 * I'm trying ERESTARTNOHAND which restart only when you want to.
1193 */ 1173 */
1194 static int compat_core_sys_select(int n, comp 1174 static int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1195 compat_ulong_t __user *outp, compat_u 1175 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1196 struct timespec64 *end_time) 1176 struct timespec64 *end_time)
1197 { 1177 {
1198 fd_set_bits fds; 1178 fd_set_bits fds;
1199 void *bits; 1179 void *bits;
1200 int size, max_fds, ret = -EINVAL; 1180 int size, max_fds, ret = -EINVAL;
1201 struct fdtable *fdt; 1181 struct fdtable *fdt;
1202 long stack_fds[SELECT_STACK_ALLOC/siz 1182 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
1203 1183
1204 if (n < 0) 1184 if (n < 0)
1205 goto out_nofds; 1185 goto out_nofds;
1206 1186
1207 /* max_fds can increase, so grab it o 1187 /* max_fds can increase, so grab it once to avoid race */
1208 rcu_read_lock(); 1188 rcu_read_lock();
1209 fdt = files_fdtable(current->files); 1189 fdt = files_fdtable(current->files);
1210 max_fds = fdt->max_fds; 1190 max_fds = fdt->max_fds;
1211 rcu_read_unlock(); 1191 rcu_read_unlock();
1212 if (n > max_fds) 1192 if (n > max_fds)
1213 n = max_fds; 1193 n = max_fds;
1214 1194
1215 /* 1195 /*
1216 * We need 6 bitmaps (in/out/ex for b 1196 * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
1217 * since we used fdset we need to all 1197 * since we used fdset we need to allocate memory in units of
1218 * long-words. 1198 * long-words.
1219 */ 1199 */
1220 size = FDS_BYTES(n); 1200 size = FDS_BYTES(n);
1221 bits = stack_fds; 1201 bits = stack_fds;
1222 if (size > sizeof(stack_fds) / 6) { 1202 if (size > sizeof(stack_fds) / 6) {
1223 bits = kmalloc_array(6, size, 1203 bits = kmalloc_array(6, size, GFP_KERNEL);
1224 ret = -ENOMEM; 1204 ret = -ENOMEM;
1225 if (!bits) 1205 if (!bits)
1226 goto out_nofds; 1206 goto out_nofds;
1227 } 1207 }
1228 fds.in = (unsigned long *) bits 1208 fds.in = (unsigned long *) bits;
1229 fds.out = (unsigned long *) (bits 1209 fds.out = (unsigned long *) (bits + size);
1230 fds.ex = (unsigned long *) (bits 1210 fds.ex = (unsigned long *) (bits + 2*size);
1231 fds.res_in = (unsigned long *) (bits 1211 fds.res_in = (unsigned long *) (bits + 3*size);
1232 fds.res_out = (unsigned long *) (bits 1212 fds.res_out = (unsigned long *) (bits + 4*size);
1233 fds.res_ex = (unsigned long *) (bits 1213 fds.res_ex = (unsigned long *) (bits + 5*size);
1234 1214
1235 if ((ret = compat_get_fd_set(n, inp, 1215 if ((ret = compat_get_fd_set(n, inp, fds.in)) ||
1236 (ret = compat_get_fd_set(n, outp, 1216 (ret = compat_get_fd_set(n, outp, fds.out)) ||
1237 (ret = compat_get_fd_set(n, exp, 1217 (ret = compat_get_fd_set(n, exp, fds.ex)))
1238 goto out; 1218 goto out;
1239 zero_fd_set(n, fds.res_in); 1219 zero_fd_set(n, fds.res_in);
1240 zero_fd_set(n, fds.res_out); 1220 zero_fd_set(n, fds.res_out);
1241 zero_fd_set(n, fds.res_ex); 1221 zero_fd_set(n, fds.res_ex);
1242 1222
1243 ret = do_select(n, &fds, end_time); 1223 ret = do_select(n, &fds, end_time);
1244 1224
1245 if (ret < 0) 1225 if (ret < 0)
1246 goto out; 1226 goto out;
1247 if (!ret) { 1227 if (!ret) {
1248 ret = -ERESTARTNOHAND; 1228 ret = -ERESTARTNOHAND;
1249 if (signal_pending(current)) 1229 if (signal_pending(current))
1250 goto out; 1230 goto out;
1251 ret = 0; 1231 ret = 0;
1252 } 1232 }
1253 1233
1254 if (compat_set_fd_set(n, inp, fds.res 1234 if (compat_set_fd_set(n, inp, fds.res_in) ||
1255 compat_set_fd_set(n, outp, fds.re 1235 compat_set_fd_set(n, outp, fds.res_out) ||
1256 compat_set_fd_set(n, exp, fds.res 1236 compat_set_fd_set(n, exp, fds.res_ex))
1257 ret = -EFAULT; 1237 ret = -EFAULT;
1258 out: 1238 out:
1259 if (bits != stack_fds) 1239 if (bits != stack_fds)
1260 kfree(bits); 1240 kfree(bits);
1261 out_nofds: 1241 out_nofds:
1262 return ret; 1242 return ret;
1263 } 1243 }
1264 1244
1265 static int do_compat_select(int n, compat_ulo 1245 static int do_compat_select(int n, compat_ulong_t __user *inp,
1266 compat_ulong_t __user *outp, compat_u 1246 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1267 struct old_timeval32 __user *tvp) 1247 struct old_timeval32 __user *tvp)
1268 { 1248 {
1269 struct timespec64 end_time, *to = NUL 1249 struct timespec64 end_time, *to = NULL;
1270 struct old_timeval32 tv; 1250 struct old_timeval32 tv;
1271 int ret; 1251 int ret;
1272 1252
1273 if (tvp) { 1253 if (tvp) {
1274 if (copy_from_user(&tv, tvp, 1254 if (copy_from_user(&tv, tvp, sizeof(tv)))
1275 return -EFAULT; 1255 return -EFAULT;
1276 1256
1277 to = &end_time; 1257 to = &end_time;
1278 if (poll_select_set_timeout(t 1258 if (poll_select_set_timeout(to,
1279 tv.tv_sec + ( 1259 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
1280 (tv.tv_usec % 1260 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
1281 return -EINVAL; 1261 return -EINVAL;
1282 } 1262 }
1283 1263
1284 ret = compat_core_sys_select(n, inp, 1264 ret = compat_core_sys_select(n, inp, outp, exp, to);
1285 return poll_select_finish(&end_time, 1265 return poll_select_finish(&end_time, tvp, PT_OLD_TIMEVAL, ret);
1286 } 1266 }
1287 1267
1288 COMPAT_SYSCALL_DEFINE5(select, int, n, compat 1268 COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp,
1289 compat_ulong_t __user *, outp, compat 1269 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1290 struct old_timeval32 __user *, tvp) 1270 struct old_timeval32 __user *, tvp)
1291 { 1271 {
1292 return do_compat_select(n, inp, outp, 1272 return do_compat_select(n, inp, outp, exp, tvp);
1293 } 1273 }
1294 1274
1295 struct compat_sel_arg_struct { 1275 struct compat_sel_arg_struct {
1296 compat_ulong_t n; 1276 compat_ulong_t n;
1297 compat_uptr_t inp; 1277 compat_uptr_t inp;
1298 compat_uptr_t outp; 1278 compat_uptr_t outp;
1299 compat_uptr_t exp; 1279 compat_uptr_t exp;
1300 compat_uptr_t tvp; 1280 compat_uptr_t tvp;
1301 }; 1281 };
1302 1282
1303 COMPAT_SYSCALL_DEFINE1(old_select, struct com 1283 COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg)
1304 { 1284 {
1305 struct compat_sel_arg_struct a; 1285 struct compat_sel_arg_struct a;
1306 1286
1307 if (copy_from_user(&a, arg, sizeof(a) 1287 if (copy_from_user(&a, arg, sizeof(a)))
1308 return -EFAULT; 1288 return -EFAULT;
1309 return do_compat_select(a.n, compat_p 1289 return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp),
1310 compat_ptr(a. 1290 compat_ptr(a.exp), compat_ptr(a.tvp));
1311 } 1291 }
1312 1292
1313 static long do_compat_pselect(int n, compat_u 1293 static long do_compat_pselect(int n, compat_ulong_t __user *inp,
1314 compat_ulong_t __user *outp, compat_u 1294 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1315 void __user *tsp, compat_sigset_t __u 1295 void __user *tsp, compat_sigset_t __user *sigmask,
1316 compat_size_t sigsetsize, enum poll_t 1296 compat_size_t sigsetsize, enum poll_time_type type)
1317 { 1297 {
1318 struct timespec64 ts, end_time, *to = 1298 struct timespec64 ts, end_time, *to = NULL;
1319 int ret; 1299 int ret;
1320 1300
1321 if (tsp) { 1301 if (tsp) {
1322 switch (type) { 1302 switch (type) {
1323 case PT_OLD_TIMESPEC: 1303 case PT_OLD_TIMESPEC:
1324 if (get_old_timespec3 1304 if (get_old_timespec32(&ts, tsp))
1325 return -EFAUL 1305 return -EFAULT;
1326 break; 1306 break;
1327 case PT_TIMESPEC: 1307 case PT_TIMESPEC:
1328 if (get_timespec64(&t 1308 if (get_timespec64(&ts, tsp))
1329 return -EFAUL 1309 return -EFAULT;
1330 break; 1310 break;
1331 default: 1311 default:
1332 BUG(); 1312 BUG();
1333 } 1313 }
1334 1314
1335 to = &end_time; 1315 to = &end_time;
1336 if (poll_select_set_timeout(t 1316 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1337 return -EINVAL; 1317 return -EINVAL;
1338 } 1318 }
1339 1319
1340 ret = set_compat_user_sigmask(sigmask 1320 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1341 if (ret) 1321 if (ret)
1342 return ret; 1322 return ret;
1343 1323
1344 ret = compat_core_sys_select(n, inp, 1324 ret = compat_core_sys_select(n, inp, outp, exp, to);
1345 return poll_select_finish(&end_time, 1325 return poll_select_finish(&end_time, tsp, type, ret);
1346 } 1326 }
1347 1327
1348 struct compat_sigset_argpack { <<
1349 compat_uptr_t p; <<
1350 compat_size_t size; <<
1351 }; <<
1352 static inline int get_compat_sigset_argpack(s <<
1353 s <<
1354 { <<
1355 if (from) { <<
1356 if (!user_read_access_begin(f <<
1357 return -EFAULT; <<
1358 unsafe_get_user(to->p, &from- <<
1359 unsafe_get_user(to->size, &fr <<
1360 user_read_access_end(); <<
1361 } <<
1362 return 0; <<
1363 Efault: <<
1364 user_access_end(); <<
1365 return -EFAULT; <<
1366 } <<
1367 <<
1368 COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, 1328 COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp,
1369 compat_ulong_t __user *, outp, compat 1329 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1370 struct __kernel_timespec __user *, ts 1330 struct __kernel_timespec __user *, tsp, void __user *, sig)
1371 { 1331 {
1372 struct compat_sigset_argpack x = {0, !! 1332 compat_size_t sigsetsize = 0;
>> 1333 compat_uptr_t up = 0;
1373 1334
1374 if (get_compat_sigset_argpack(&x, sig !! 1335 if (sig) {
1375 return -EFAULT; !! 1336 if (!access_ok(sig,
>> 1337 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
>> 1338 __get_user(up, (compat_uptr_t __user *)sig) ||
>> 1339 __get_user(sigsetsize,
>> 1340 (compat_size_t __user *)(sig+sizeof(up))))
>> 1341 return -EFAULT;
>> 1342 }
1376 1343
1377 return do_compat_pselect(n, inp, outp !! 1344 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1378 x.size, PT_T !! 1345 sigsetsize, PT_TIMESPEC);
1379 } 1346 }
1380 1347
1381 #if defined(CONFIG_COMPAT_32BIT_TIME) 1348 #if defined(CONFIG_COMPAT_32BIT_TIME)
1382 1349
1383 COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, 1350 COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp,
1384 compat_ulong_t __user *, outp, compat 1351 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp,
1385 struct old_timespec32 __user *, tsp, 1352 struct old_timespec32 __user *, tsp, void __user *, sig)
1386 { 1353 {
1387 struct compat_sigset_argpack x = {0, !! 1354 compat_size_t sigsetsize = 0;
>> 1355 compat_uptr_t up = 0;
1388 1356
1389 if (get_compat_sigset_argpack(&x, sig !! 1357 if (sig) {
1390 return -EFAULT; !! 1358 if (!access_ok(sig,
>> 1359 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
>> 1360 __get_user(up, (compat_uptr_t __user *)sig) ||
>> 1361 __get_user(sigsetsize,
>> 1362 (compat_size_t __user *)(sig+sizeof(up))))
>> 1363 return -EFAULT;
>> 1364 }
1391 1365
1392 return do_compat_pselect(n, inp, outp !! 1366 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up),
1393 x.size, PT_O !! 1367 sigsetsize, PT_OLD_TIMESPEC);
1394 } 1368 }
1395 1369
1396 #endif 1370 #endif
1397 1371
1398 #if defined(CONFIG_COMPAT_32BIT_TIME) 1372 #if defined(CONFIG_COMPAT_32BIT_TIME)
1399 COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct p 1373 COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds,
1400 unsigned int, nfds, struct old_times 1374 unsigned int, nfds, struct old_timespec32 __user *, tsp,
1401 const compat_sigset_t __user *, sigma 1375 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1402 { 1376 {
1403 struct timespec64 ts, end_time, *to = 1377 struct timespec64 ts, end_time, *to = NULL;
1404 int ret; 1378 int ret;
1405 1379
1406 if (tsp) { 1380 if (tsp) {
1407 if (get_old_timespec32(&ts, t 1381 if (get_old_timespec32(&ts, tsp))
1408 return -EFAULT; 1382 return -EFAULT;
1409 1383
1410 to = &end_time; 1384 to = &end_time;
1411 if (poll_select_set_timeout(t 1385 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1412 return -EINVAL; 1386 return -EINVAL;
1413 } 1387 }
1414 1388
1415 ret = set_compat_user_sigmask(sigmask 1389 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1416 if (ret) 1390 if (ret)
1417 return ret; 1391 return ret;
1418 1392
1419 ret = do_sys_poll(ufds, nfds, to); 1393 ret = do_sys_poll(ufds, nfds, to);
1420 return poll_select_finish(&end_time, 1394 return poll_select_finish(&end_time, tsp, PT_OLD_TIMESPEC, ret);
1421 } 1395 }
1422 #endif 1396 #endif
1423 1397
1424 /* New compat syscall for 64 bit time_t*/ 1398 /* New compat syscall for 64 bit time_t*/
1425 COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct p 1399 COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds,
1426 unsigned int, nfds, struct __kernel_ 1400 unsigned int, nfds, struct __kernel_timespec __user *, tsp,
1427 const compat_sigset_t __user *, sigma 1401 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize)
1428 { 1402 {
1429 struct timespec64 ts, end_time, *to = 1403 struct timespec64 ts, end_time, *to = NULL;
1430 int ret; 1404 int ret;
1431 1405
1432 if (tsp) { 1406 if (tsp) {
1433 if (get_timespec64(&ts, tsp)) 1407 if (get_timespec64(&ts, tsp))
1434 return -EFAULT; 1408 return -EFAULT;
1435 1409
1436 to = &end_time; 1410 to = &end_time;
1437 if (poll_select_set_timeout(t 1411 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
1438 return -EINVAL; 1412 return -EINVAL;
1439 } 1413 }
1440 1414
1441 ret = set_compat_user_sigmask(sigmask 1415 ret = set_compat_user_sigmask(sigmask, sigsetsize);
1442 if (ret) 1416 if (ret)
1443 return ret; 1417 return ret;
1444 1418
1445 ret = do_sys_poll(ufds, nfds, to); 1419 ret = do_sys_poll(ufds, nfds, to);
1446 return poll_select_finish(&end_time, 1420 return poll_select_finish(&end_time, tsp, PT_TIMESPEC, ret);
1447 } 1421 }
1448 1422
1449 #endif 1423 #endif
1450 1424
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.