|
|
1 | /* |
|
|
2 | * libeio implementation |
|
|
3 | * |
|
|
4 | * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> |
|
|
5 | * All rights reserved. |
|
|
6 | * |
|
|
7 | * Redistribution and use in source and binary forms, with or without modifica- |
|
|
8 | * tion, are permitted provided that the following conditions are met: |
|
|
9 | * |
|
|
10 | * 1. Redistributions of source code must retain the above copyright notice, |
|
|
11 | * this list of conditions and the following disclaimer. |
|
|
12 | * |
|
|
13 | * 2. Redistributions in binary form must reproduce the above copyright |
|
|
14 | * notice, this list of conditions and the following disclaimer in the |
|
|
15 | * documentation and/or other materials provided with the distribution. |
|
|
16 | * |
|
|
17 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED |
|
|
18 | * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- |
|
|
19 | * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO |
|
|
20 | * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- |
|
|
21 | * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
|
|
22 | * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; |
|
|
23 | * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, |
|
|
24 | * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH- |
|
|
25 | * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED |
|
|
26 | * OF THE POSSIBILITY OF SUCH DAMAGE. |
|
|
27 | * |
|
|
28 | * Alternatively, the contents of this file may be used under the terms of |
|
|
29 | * the GNU General Public License ("GPL") version 2 or any later version, |
|
|
30 | * in which case the provisions of the GPL are applicable instead of |
|
|
31 | * the above. If you wish to allow the use of your version of this file |
|
|
32 | * only under the terms of the GPL and not to allow others to use your |
|
|
33 | * version of this file under the BSD license, indicate your decision |
|
|
34 | * by deleting the provisions above and replace them with the notice |
|
|
35 | * and other provisions required by the GPL. If you do not delete the |
|
|
36 | * provisions above, a recipient may use your version of this file under |
|
|
37 | * either the BSD or the GPL. |
|
|
38 | */ |
|
|
39 | |
1 | #include "eio.h" |
40 | #include "eio.h" |
2 | #include "xthread.h" |
41 | #include "xthread.h" |
3 | |
42 | |
4 | #include <errno.h> |
43 | #include <errno.h> |
5 | #include <stddef.h> |
44 | #include <stddef.h> |
… | |
… | |
70 | /* buffer size for various temporary buffers */ |
109 | /* buffer size for various temporary buffers */ |
71 | #define EIO_BUFSIZE 65536 |
110 | #define EIO_BUFSIZE 65536 |
72 | |
111 | |
73 | #define dBUF \ |
112 | #define dBUF \ |
74 | char *eio_buf; \ |
113 | char *eio_buf; \ |
75 | X_LOCK (wrklock); \ |
114 | ETP_WORKER_LOCK (self); \ |
76 | self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ |
115 | self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ |
77 | X_UNLOCK (wrklock); \ |
116 | ETP_WORKER_UNLOCK (self); \ |
78 | errno = ENOMEM; \ |
117 | errno = ENOMEM; \ |
79 | if (!eio_buf) \ |
118 | if (!eio_buf) \ |
80 | return -1; |
119 | return -1; |
81 | |
120 | |
82 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
121 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
83 | |
122 | |
|
|
123 | /*****************************************************************************/ |
|
|
124 | |
|
|
125 | #if __GNUC__ >= 3 |
|
|
126 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
127 | #else |
|
|
128 | # define expect(expr,value) (expr) |
|
|
129 | #endif |
|
|
130 | |
|
|
131 | #define expect_false(expr) expect ((expr) != 0, 0) |
|
|
132 | #define expect_true(expr) expect ((expr) != 0, 1) |
|
|
133 | |
|
|
134 | /*****************************************************************************/ |
|
|
135 | |
|
|
136 | #define ETP_PRI_MIN EIO_PRI_MIN |
|
|
137 | #define ETP_PRI_MAX EIO_PRI_MAX |
|
|
138 | |
|
|
139 | struct etp_worker; |
|
|
140 | |
|
|
141 | #define ETP_REQ eio_req |
|
|
142 | #define ETP_DESTROY(req) eio_destroy (req) |
|
|
143 | static int eio_finish (eio_req *req); |
|
|
144 | #define ETP_FINISH(req) eio_finish (req) |
|
|
145 | static void eio_execute (struct etp_worker *self, eio_req *req); |
|
|
146 | #define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) |
|
|
147 | |
|
|
148 | #define ETP_WORKER_CLEAR(req) \ |
|
|
149 | if (wrk->dbuf) \ |
|
|
150 | { \ |
|
|
151 | free (wrk->dbuf); \ |
|
|
152 | wrk->dbuf = 0; \ |
|
|
153 | } \ |
|
|
154 | \ |
|
|
155 | if (wrk->dirp) \ |
|
|
156 | { \ |
|
|
157 | closedir (wrk->dirp); \ |
|
|
158 | wrk->dirp = 0; \ |
|
|
159 | } |
|
|
160 | #define ETP_WORKER_COMMON \ |
|
|
161 | void *dbuf; \ |
|
|
162 | DIR *dirp; |
|
|
163 | |
|
|
164 | /*****************************************************************************/ |
|
|
165 | |
|
|
166 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
|
|
167 | |
|
|
168 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
|
|
169 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
|
|
170 | { |
|
|
171 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
|
|
172 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
|
|
173 | } |
|
|
174 | |
|
|
175 | static unsigned int started, idle, wanted = 4; |
|
|
176 | |
84 | static void (*want_poll_cb) (void); |
177 | static void (*want_poll_cb) (void); |
85 | static void (*done_poll_cb) (void); |
178 | static void (*done_poll_cb) (void); |
86 | |
179 | |
87 | static unsigned int max_poll_time = 0; |
180 | static unsigned int max_poll_time; /* reslock */ |
88 | static unsigned int max_poll_reqs = 0; |
181 | static unsigned int max_poll_reqs; /* reslock */ |
89 | |
182 | |
90 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
183 | static volatile unsigned int nreqs; /* reqlock */ |
91 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
184 | static volatile unsigned int nready; /* reqlock */ |
92 | { |
185 | static volatile unsigned int npending; /* reqlock */ |
93 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
186 | static volatile unsigned int max_idle = 4; |
94 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
|
|
95 | } |
|
|
96 | |
187 | |
97 | static unsigned int started, idle, wanted = 4; |
|
|
98 | |
|
|
99 | /* worker threads management */ |
|
|
100 | static mutex_t wrklock = X_MUTEX_INIT; |
188 | static mutex_t wrklock = X_MUTEX_INIT; |
101 | |
|
|
102 | typedef struct worker |
|
|
103 | { |
|
|
104 | /* locked by wrklock */ |
|
|
105 | struct worker *prev, *next; |
|
|
106 | |
|
|
107 | thread_t tid; |
|
|
108 | |
|
|
109 | /* locked by reslock, reqlock or wrklock */ |
|
|
110 | eio_req *req; /* currently processed request */ |
|
|
111 | void *dbuf; |
|
|
112 | DIR *dirp; |
|
|
113 | } worker; |
|
|
114 | |
|
|
115 | static worker wrk_first = { &wrk_first, &wrk_first, 0 }; |
|
|
116 | |
|
|
117 | static void worker_clear (worker *wrk) |
|
|
118 | { |
|
|
119 | if (wrk->dirp) |
|
|
120 | { |
|
|
121 | closedir (wrk->dirp); |
|
|
122 | wrk->dirp = 0; |
|
|
123 | } |
|
|
124 | |
|
|
125 | if (wrk->dbuf) |
|
|
126 | { |
|
|
127 | free (wrk->dbuf); |
|
|
128 | wrk->dbuf = 0; |
|
|
129 | } |
|
|
130 | } |
|
|
131 | |
|
|
132 | static void worker_free (worker *wrk) |
|
|
133 | { |
|
|
134 | wrk->next->prev = wrk->prev; |
|
|
135 | wrk->prev->next = wrk->next; |
|
|
136 | |
|
|
137 | free (wrk); |
|
|
138 | } |
|
|
139 | |
|
|
140 | static volatile unsigned int nreqs, nready, npending; |
|
|
141 | static volatile unsigned int max_idle = 4; |
|
|
142 | |
|
|
143 | static mutex_t reslock = X_MUTEX_INIT; |
189 | static mutex_t reslock = X_MUTEX_INIT; |
144 | static mutex_t reqlock = X_MUTEX_INIT; |
190 | static mutex_t reqlock = X_MUTEX_INIT; |
145 | static cond_t reqwait = X_COND_INIT; |
191 | static cond_t reqwait = X_COND_INIT; |
146 | |
192 | |
|
|
193 | #if !HAVE_PREADWRITE |
|
|
194 | /* |
|
|
195 | * make our pread/pwrite emulation safe against themselves, but not against |
|
|
196 | * normal read/write by using a mutex. slows down execution a lot, |
|
|
197 | * but that's your problem, not mine. |
|
|
198 | */ |
|
|
199 | static mutex_t preadwritelock = X_MUTEX_INIT; |
|
|
200 | #endif |
|
|
201 | |
|
|
202 | typedef struct etp_worker |
|
|
203 | { |
|
|
204 | /* locked by wrklock */ |
|
|
205 | struct etp_worker *prev, *next; |
|
|
206 | |
|
|
207 | thread_t tid; |
|
|
208 | |
|
|
209 | /* locked by reslock, reqlock or wrklock */ |
|
|
210 | ETP_REQ *req; /* currently processed request */ |
|
|
211 | |
|
|
212 | ETP_WORKER_COMMON |
|
|
213 | } etp_worker; |
|
|
214 | |
|
|
215 | static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ |
|
|
216 | |
|
|
217 | #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
|
|
218 | #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
|
|
219 | |
|
|
220 | /* worker threads management */ |
|
|
221 | |
|
|
222 | static void etp_worker_clear (etp_worker *wrk) |
|
|
223 | { |
|
|
224 | ETP_WORKER_CLEAR (wrk); |
|
|
225 | } |
|
|
226 | |
|
|
227 | static void etp_worker_free (etp_worker *wrk) |
|
|
228 | { |
|
|
229 | wrk->next->prev = wrk->prev; |
|
|
230 | wrk->prev->next = wrk->next; |
|
|
231 | |
|
|
232 | free (wrk); |
|
|
233 | } |
|
|
234 | |
147 | unsigned int eio_nreqs (void) |
235 | static unsigned int etp_nreqs (void) |
148 | { |
236 | { |
|
|
237 | int retval; |
|
|
238 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
239 | retval = nreqs; |
|
|
240 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
149 | return nreqs; |
241 | return retval; |
150 | } |
242 | } |
151 | |
243 | |
152 | unsigned int eio_nready (void) |
244 | static unsigned int etp_nready (void) |
153 | { |
245 | { |
154 | unsigned int retval; |
246 | unsigned int retval; |
155 | |
247 | |
156 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
248 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
157 | retval = nready; |
249 | retval = nready; |
158 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
250 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
159 | |
251 | |
160 | return retval; |
252 | return retval; |
161 | } |
253 | } |
162 | |
254 | |
163 | unsigned int eio_npending (void) |
255 | static unsigned int etp_npending (void) |
164 | { |
256 | { |
165 | unsigned int retval; |
257 | unsigned int retval; |
166 | |
258 | |
167 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
259 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
168 | retval = npending; |
260 | retval = npending; |
169 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
261 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
170 | |
262 | |
171 | return retval; |
263 | return retval; |
172 | } |
264 | } |
173 | |
265 | |
174 | unsigned int eio_nthreads (void) |
266 | static unsigned int etp_nthreads (void) |
175 | { |
267 | { |
176 | unsigned int retval; |
268 | unsigned int retval; |
177 | |
269 | |
178 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
270 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
179 | retval = started; |
271 | retval = started; |
… | |
… | |
186 | * a somewhat faster data structure might be nice, but |
278 | * a somewhat faster data structure might be nice, but |
187 | * with 8 priorities this actually needs <20 insns |
279 | * with 8 priorities this actually needs <20 insns |
188 | * per shift, the most expensive operation. |
280 | * per shift, the most expensive operation. |
189 | */ |
281 | */ |
190 | typedef struct { |
282 | typedef struct { |
191 | eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */ |
283 | ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */ |
192 | int size; |
284 | int size; |
193 | } reqq; |
285 | } etp_reqq; |
194 | |
286 | |
195 | static reqq req_queue; |
287 | static etp_reqq req_queue; |
196 | static reqq res_queue; |
288 | static etp_reqq res_queue; |
197 | |
289 | |
198 | static int reqq_push (reqq *q, eio_req *req) |
290 | static int reqq_push (etp_reqq *q, ETP_REQ *req) |
199 | { |
291 | { |
200 | int pri = req->pri; |
292 | int pri = req->pri; |
201 | req->next = 0; |
293 | req->next = 0; |
202 | |
294 | |
203 | if (q->qe[pri]) |
295 | if (q->qe[pri]) |
… | |
… | |
209 | q->qe[pri] = q->qs[pri] = req; |
301 | q->qe[pri] = q->qs[pri] = req; |
210 | |
302 | |
211 | return q->size++; |
303 | return q->size++; |
212 | } |
304 | } |
213 | |
305 | |
214 | static eio_req *reqq_shift (reqq *q) |
306 | static ETP_REQ *reqq_shift (etp_reqq *q) |
215 | { |
307 | { |
216 | int pri; |
308 | int pri; |
217 | |
309 | |
218 | if (!q->size) |
310 | if (!q->size) |
219 | return 0; |
311 | return 0; |
220 | |
312 | |
221 | --q->size; |
313 | --q->size; |
222 | |
314 | |
223 | for (pri = EIO_NUM_PRI; pri--; ) |
315 | for (pri = ETP_NUM_PRI; pri--; ) |
224 | { |
316 | { |
225 | eio_req *req = q->qs[pri]; |
317 | eio_req *req = q->qs[pri]; |
226 | |
318 | |
227 | if (req) |
319 | if (req) |
228 | { |
320 | { |
… | |
… | |
233 | } |
325 | } |
234 | } |
326 | } |
235 | |
327 | |
236 | abort (); |
328 | abort (); |
237 | } |
329 | } |
|
|
330 | |
|
|
331 | static void etp_atfork_prepare (void) |
|
|
332 | { |
|
|
333 | X_LOCK (wrklock); |
|
|
334 | X_LOCK (reqlock); |
|
|
335 | X_LOCK (reslock); |
|
|
336 | #if !HAVE_PREADWRITE |
|
|
337 | X_LOCK (preadwritelock); |
|
|
338 | #endif |
|
|
339 | } |
|
|
340 | |
|
|
341 | static void etp_atfork_parent (void) |
|
|
342 | { |
|
|
343 | #if !HAVE_PREADWRITE |
|
|
344 | X_UNLOCK (preadwritelock); |
|
|
345 | #endif |
|
|
346 | X_UNLOCK (reslock); |
|
|
347 | X_UNLOCK (reqlock); |
|
|
348 | X_UNLOCK (wrklock); |
|
|
349 | } |
|
|
350 | |
|
|
351 | static void etp_atfork_child (void) |
|
|
352 | { |
|
|
353 | ETP_REQ *prv; |
|
|
354 | |
|
|
355 | while ((prv = reqq_shift (&req_queue))) |
|
|
356 | ETP_DESTROY (prv); |
|
|
357 | |
|
|
358 | while ((prv = reqq_shift (&res_queue))) |
|
|
359 | ETP_DESTROY (prv); |
|
|
360 | |
|
|
361 | while (wrk_first.next != &wrk_first) |
|
|
362 | { |
|
|
363 | etp_worker *wrk = wrk_first.next; |
|
|
364 | |
|
|
365 | if (wrk->req) |
|
|
366 | ETP_DESTROY (wrk->req); |
|
|
367 | |
|
|
368 | etp_worker_clear (wrk); |
|
|
369 | etp_worker_free (wrk); |
|
|
370 | } |
|
|
371 | |
|
|
372 | started = 0; |
|
|
373 | idle = 0; |
|
|
374 | nreqs = 0; |
|
|
375 | nready = 0; |
|
|
376 | npending = 0; |
|
|
377 | |
|
|
378 | etp_atfork_parent (); |
|
|
379 | } |
|
|
380 | |
|
|
381 | static void |
|
|
382 | etp_once_init (void) |
|
|
383 | { |
|
|
384 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
|
|
385 | } |
|
|
386 | |
|
|
387 | static int |
|
|
388 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
|
|
389 | { |
|
|
390 | static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
|
|
391 | |
|
|
392 | pthread_once (&doinit, etp_once_init); |
|
|
393 | |
|
|
394 | want_poll_cb = want_poll; |
|
|
395 | done_poll_cb = done_poll; |
|
|
396 | |
|
|
397 | return 0; |
|
|
398 | } |
|
|
399 | |
|
|
400 | X_THREAD_PROC (etp_proc); |
|
|
401 | |
|
|
402 | static void etp_start_thread (void) |
|
|
403 | { |
|
|
404 | etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
|
|
405 | |
|
|
406 | /*TODO*/ |
|
|
407 | assert (("unable to allocate worker thread data", wrk)); |
|
|
408 | |
|
|
409 | X_LOCK (wrklock); |
|
|
410 | |
|
|
411 | if (thread_create (&wrk->tid, etp_proc, (void *)wrk)) |
|
|
412 | { |
|
|
413 | wrk->prev = &wrk_first; |
|
|
414 | wrk->next = wrk_first.next; |
|
|
415 | wrk_first.next->prev = wrk; |
|
|
416 | wrk_first.next = wrk; |
|
|
417 | ++started; |
|
|
418 | } |
|
|
419 | else |
|
|
420 | free (wrk); |
|
|
421 | |
|
|
422 | X_UNLOCK (wrklock); |
|
|
423 | } |
|
|
424 | |
|
|
425 | static void etp_maybe_start_thread (void) |
|
|
426 | { |
|
|
427 | if (expect_true (etp_nthreads () >= wanted)) |
|
|
428 | return; |
|
|
429 | |
|
|
430 | /* todo: maybe use idle here, but might be less exact */ |
|
|
431 | if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
|
|
432 | return; |
|
|
433 | |
|
|
434 | etp_start_thread (); |
|
|
435 | } |
|
|
436 | |
|
|
437 | static void etp_end_thread (void) |
|
|
438 | { |
|
|
439 | eio_req *req = calloc (1, sizeof (eio_req)); |
|
|
440 | |
|
|
441 | req->type = -1; |
|
|
442 | req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
|
|
443 | |
|
|
444 | X_LOCK (reqlock); |
|
|
445 | reqq_push (&req_queue, req); |
|
|
446 | X_COND_SIGNAL (reqwait); |
|
|
447 | X_UNLOCK (reqlock); |
|
|
448 | |
|
|
449 | X_LOCK (wrklock); |
|
|
450 | --started; |
|
|
451 | X_UNLOCK (wrklock); |
|
|
452 | } |
|
|
453 | |
|
|
454 | static int etp_poll (void) |
|
|
455 | { |
|
|
456 | unsigned int maxreqs; |
|
|
457 | unsigned int maxtime; |
|
|
458 | struct timeval tv_start, tv_now; |
|
|
459 | |
|
|
460 | X_LOCK (reslock); |
|
|
461 | maxreqs = max_poll_reqs; |
|
|
462 | maxtime = max_poll_time; |
|
|
463 | X_UNLOCK (reslock); |
|
|
464 | |
|
|
465 | if (maxtime) |
|
|
466 | gettimeofday (&tv_start, 0); |
|
|
467 | |
|
|
468 | for (;;) |
|
|
469 | { |
|
|
470 | ETP_REQ *req; |
|
|
471 | |
|
|
472 | etp_maybe_start_thread (); |
|
|
473 | |
|
|
474 | X_LOCK (reslock); |
|
|
475 | req = reqq_shift (&res_queue); |
|
|
476 | |
|
|
477 | if (req) |
|
|
478 | { |
|
|
479 | --npending; |
|
|
480 | |
|
|
481 | if (!res_queue.size && done_poll_cb) |
|
|
482 | done_poll_cb (); |
|
|
483 | } |
|
|
484 | |
|
|
485 | X_UNLOCK (reslock); |
|
|
486 | |
|
|
487 | if (!req) |
|
|
488 | return 0; |
|
|
489 | |
|
|
490 | X_LOCK (reqlock); |
|
|
491 | --nreqs; |
|
|
492 | X_UNLOCK (reqlock); |
|
|
493 | |
|
|
494 | if (expect_false (req->type == EIO_GROUP && req->size)) |
|
|
495 | { |
|
|
496 | req->int1 = 1; /* mark request as delayed */ |
|
|
497 | continue; |
|
|
498 | } |
|
|
499 | else |
|
|
500 | { |
|
|
501 | int res = ETP_FINISH (req); |
|
|
502 | if (expect_false (res)) |
|
|
503 | return res; |
|
|
504 | } |
|
|
505 | |
|
|
506 | if (expect_false (maxreqs && !--maxreqs)) |
|
|
507 | break; |
|
|
508 | |
|
|
509 | if (maxtime) |
|
|
510 | { |
|
|
511 | gettimeofday (&tv_now, 0); |
|
|
512 | |
|
|
513 | if (tvdiff (&tv_start, &tv_now) >= maxtime) |
|
|
514 | break; |
|
|
515 | } |
|
|
516 | } |
|
|
517 | |
|
|
518 | errno = EAGAIN; |
|
|
519 | return -1; |
|
|
520 | } |
|
|
521 | |
|
|
522 | static void etp_cancel (ETP_REQ *req) |
|
|
523 | { |
|
|
524 | X_LOCK (wrklock); |
|
|
525 | req->flags |= EIO_FLAG_CANCELLED; |
|
|
526 | X_UNLOCK (wrklock); |
|
|
527 | |
|
|
528 | eio_grp_cancel (req); |
|
|
529 | } |
|
|
530 | |
|
|
531 | static void etp_submit (ETP_REQ *req) |
|
|
532 | { |
|
|
533 | req->pri -= ETP_PRI_MIN; |
|
|
534 | |
|
|
535 | if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
|
|
536 | if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
|
|
537 | |
|
|
538 | if (expect_false (req->type == EIO_GROUP)) |
|
|
539 | { |
|
|
540 | /* I hope this is worth it :/ */ |
|
|
541 | X_LOCK (reqlock); |
|
|
542 | ++nreqs; |
|
|
543 | X_UNLOCK (reqlock); |
|
|
544 | |
|
|
545 | X_LOCK (reslock); |
|
|
546 | |
|
|
547 | ++npending; |
|
|
548 | |
|
|
549 | if (!reqq_push (&res_queue, req) && want_poll_cb) |
|
|
550 | want_poll_cb (); |
|
|
551 | |
|
|
552 | X_UNLOCK (reslock); |
|
|
553 | } |
|
|
554 | else |
|
|
555 | { |
|
|
556 | X_LOCK (reqlock); |
|
|
557 | ++nreqs; |
|
|
558 | ++nready; |
|
|
559 | reqq_push (&req_queue, req); |
|
|
560 | X_COND_SIGNAL (reqwait); |
|
|
561 | X_UNLOCK (reqlock); |
|
|
562 | |
|
|
563 | etp_maybe_start_thread (); |
|
|
564 | } |
|
|
565 | } |
|
|
566 | |
|
|
567 | static void etp_set_max_poll_time (double nseconds) |
|
|
568 | { |
|
|
569 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
|
|
570 | max_poll_time = nseconds; |
|
|
571 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
|
|
572 | } |
|
|
573 | |
|
|
574 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
|
|
575 | { |
|
|
576 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
|
|
577 | max_poll_reqs = maxreqs; |
|
|
578 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
|
|
579 | } |
|
|
580 | |
|
|
581 | static void etp_set_max_idle (unsigned int nthreads) |
|
|
582 | { |
|
|
583 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
584 | max_idle = nthreads <= 0 ? 1 : nthreads; |
|
|
585 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
586 | } |
|
|
587 | |
|
|
588 | static void etp_set_min_parallel (unsigned int nthreads) |
|
|
589 | { |
|
|
590 | if (wanted < nthreads) |
|
|
591 | wanted = nthreads; |
|
|
592 | } |
|
|
593 | |
|
|
594 | static void etp_set_max_parallel (unsigned int nthreads) |
|
|
595 | { |
|
|
596 | if (wanted > nthreads) |
|
|
597 | wanted = nthreads; |
|
|
598 | |
|
|
599 | while (started > wanted) |
|
|
600 | etp_end_thread (); |
|
|
601 | } |
|
|
602 | |
|
|
603 | /*****************************************************************************/ |
238 | |
604 | |
239 | static void grp_try_feed (eio_req *grp) |
605 | static void grp_try_feed (eio_req *grp) |
240 | { |
606 | { |
241 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
607 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
242 | { |
608 | { |
… | |
… | |
251 | break; |
617 | break; |
252 | } |
618 | } |
253 | } |
619 | } |
254 | } |
620 | } |
255 | |
621 | |
256 | static int eio_finish (eio_req *req); |
|
|
257 | |
|
|
258 | static int grp_dec (eio_req *grp) |
622 | static int grp_dec (eio_req *grp) |
259 | { |
623 | { |
260 | --grp->size; |
624 | --grp->size; |
261 | |
625 | |
262 | /* call feeder, if applicable */ |
626 | /* call feeder, if applicable */ |
… | |
… | |
310 | eio_cancel (grp); |
674 | eio_cancel (grp); |
311 | } |
675 | } |
312 | |
676 | |
313 | void eio_cancel (eio_req *req) |
677 | void eio_cancel (eio_req *req) |
314 | { |
678 | { |
315 | X_LOCK (wrklock); |
|
|
316 | req->flags |= EIO_FLAG_CANCELLED; |
|
|
317 | X_UNLOCK (wrklock); |
|
|
318 | |
|
|
319 | eio_grp_cancel (req); |
679 | etp_cancel (req); |
320 | } |
|
|
321 | |
|
|
322 | X_THREAD_PROC (eio_proc); |
|
|
323 | |
|
|
324 | static void start_thread (void) |
|
|
325 | { |
|
|
326 | worker *wrk = calloc (1, sizeof (worker)); |
|
|
327 | |
|
|
328 | /*TODO*/ |
|
|
329 | assert (("unable to allocate worker thread data", wrk)); |
|
|
330 | |
|
|
331 | X_LOCK (wrklock); |
|
|
332 | |
|
|
333 | if (thread_create (&wrk->tid, eio_proc, (void *)wrk)) |
|
|
334 | { |
|
|
335 | wrk->prev = &wrk_first; |
|
|
336 | wrk->next = wrk_first.next; |
|
|
337 | wrk_first.next->prev = wrk; |
|
|
338 | wrk_first.next = wrk; |
|
|
339 | ++started; |
|
|
340 | } |
|
|
341 | else |
|
|
342 | free (wrk); |
|
|
343 | |
|
|
344 | X_UNLOCK (wrklock); |
|
|
345 | } |
|
|
346 | |
|
|
347 | static void maybe_start_thread (void) |
|
|
348 | { |
|
|
349 | if (eio_nthreads () >= wanted) |
|
|
350 | return; |
|
|
351 | |
|
|
352 | /* todo: maybe use idle here, but might be less exact */ |
|
|
353 | if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ()) |
|
|
354 | return; |
|
|
355 | |
|
|
356 | start_thread (); |
|
|
357 | } |
680 | } |
358 | |
681 | |
359 | void eio_submit (eio_req *req) |
682 | void eio_submit (eio_req *req) |
360 | { |
683 | { |
361 | req->pri += EIO_PRI_BIAS; |
684 | etp_submit (req); |
362 | |
|
|
363 | if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS; |
|
|
364 | if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; |
|
|
365 | |
|
|
366 | ++nreqs; |
|
|
367 | |
|
|
368 | X_LOCK (reqlock); |
|
|
369 | ++nready; |
|
|
370 | reqq_push (&req_queue, req); |
|
|
371 | X_COND_SIGNAL (reqwait); |
|
|
372 | X_UNLOCK (reqlock); |
|
|
373 | |
|
|
374 | maybe_start_thread (); |
|
|
375 | } |
685 | } |
376 | |
686 | |
377 | static void end_thread (void) |
687 | unsigned int eio_nreqs (void) |
378 | { |
688 | { |
379 | eio_req *req = calloc (1, sizeof (eio_req)); |
689 | return etp_nreqs (); |
|
|
690 | } |
380 | |
691 | |
381 | req->type = EIO_QUIT; |
692 | unsigned int eio_nready (void) |
382 | req->pri = EIO_PRI_MAX + EIO_PRI_BIAS; |
693 | { |
|
|
694 | return etp_nready (); |
|
|
695 | } |
383 | |
696 | |
384 | X_LOCK (reqlock); |
697 | unsigned int eio_npending (void) |
385 | reqq_push (&req_queue, req); |
698 | { |
386 | X_COND_SIGNAL (reqwait); |
699 | return etp_npending (); |
387 | X_UNLOCK (reqlock); |
700 | } |
388 | |
701 | |
389 | X_LOCK (wrklock); |
702 | unsigned int eio_nthreads (void) |
390 | --started; |
703 | { |
391 | X_UNLOCK (wrklock); |
704 | return etp_nthreads (); |
392 | } |
705 | } |
393 | |
706 | |
394 | void eio_set_max_poll_time (double nseconds) |
707 | void eio_set_max_poll_time (double nseconds) |
395 | { |
708 | { |
396 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
397 | max_poll_time = nseconds; |
709 | etp_set_max_poll_time (nseconds); |
398 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
399 | } |
710 | } |
400 | |
711 | |
401 | void eio_set_max_poll_reqs (unsigned int maxreqs) |
712 | void eio_set_max_poll_reqs (unsigned int maxreqs) |
402 | { |
713 | { |
403 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
404 | max_poll_reqs = maxreqs; |
714 | etp_set_max_poll_reqs (maxreqs); |
405 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
406 | } |
715 | } |
407 | |
716 | |
408 | void eio_set_max_idle (unsigned int nthreads) |
717 | void eio_set_max_idle (unsigned int nthreads) |
409 | { |
718 | { |
410 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
719 | etp_set_max_idle (nthreads); |
411 | max_idle = nthreads <= 0 ? 1 : nthreads; |
|
|
412 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
413 | } |
720 | } |
414 | |
721 | |
415 | void eio_set_min_parallel (unsigned int nthreads) |
722 | void eio_set_min_parallel (unsigned int nthreads) |
416 | { |
723 | { |
417 | if (wanted < nthreads) |
724 | etp_set_min_parallel (nthreads); |
418 | wanted = nthreads; |
|
|
419 | } |
725 | } |
420 | |
726 | |
421 | void eio_set_max_parallel (unsigned int nthreads) |
727 | void eio_set_max_parallel (unsigned int nthreads) |
422 | { |
728 | { |
423 | if (wanted > nthreads) |
729 | etp_set_max_parallel (nthreads); |
424 | wanted = nthreads; |
|
|
425 | |
|
|
426 | while (started > wanted) |
|
|
427 | end_thread (); |
|
|
428 | } |
730 | } |
429 | |
731 | |
430 | int eio_poll (void) |
732 | int eio_poll (void) |
431 | { |
733 | { |
432 | int maxreqs = max_poll_reqs; |
734 | return etp_poll (); |
433 | struct timeval tv_start, tv_now; |
|
|
434 | eio_req *req; |
|
|
435 | |
|
|
436 | if (max_poll_time) |
|
|
437 | gettimeofday (&tv_start, 0); |
|
|
438 | |
|
|
439 | for (;;) |
|
|
440 | { |
|
|
441 | maybe_start_thread (); |
|
|
442 | |
|
|
443 | X_LOCK (reslock); |
|
|
444 | req = reqq_shift (&res_queue); |
|
|
445 | |
|
|
446 | if (req) |
|
|
447 | { |
|
|
448 | --npending; |
|
|
449 | |
|
|
450 | if (!res_queue.size && done_poll_cb) |
|
|
451 | done_poll_cb (); |
|
|
452 | } |
|
|
453 | |
|
|
454 | X_UNLOCK (reslock); |
|
|
455 | |
|
|
456 | if (!req) |
|
|
457 | return 0; |
|
|
458 | |
|
|
459 | --nreqs; |
|
|
460 | |
|
|
461 | if (req->type == EIO_GROUP && req->size) |
|
|
462 | { |
|
|
463 | req->int1 = 1; /* mark request as delayed */ |
|
|
464 | continue; |
|
|
465 | } |
|
|
466 | else |
|
|
467 | { |
|
|
468 | int res = eio_finish (req); |
|
|
469 | if (res) |
|
|
470 | return res; |
|
|
471 | } |
|
|
472 | |
|
|
473 | if (maxreqs && !--maxreqs) |
|
|
474 | break; |
|
|
475 | |
|
|
476 | if (max_poll_time) |
|
|
477 | { |
|
|
478 | gettimeofday (&tv_now, 0); |
|
|
479 | |
|
|
480 | if (tvdiff (&tv_start, &tv_now) >= max_poll_time) |
|
|
481 | break; |
|
|
482 | } |
|
|
483 | } |
|
|
484 | |
|
|
485 | errno = EAGAIN; |
|
|
486 | return -1; |
|
|
487 | } |
735 | } |
488 | |
736 | |
489 | /*****************************************************************************/ |
737 | /*****************************************************************************/ |
490 | /* work around various missing functions */ |
738 | /* work around various missing functions */ |
491 | |
739 | |
492 | #if !HAVE_PREADWRITE |
740 | #if !HAVE_PREADWRITE |
493 | # define pread eio__pread |
741 | # define pread eio__pread |
494 | # define pwrite eio__pwrite |
742 | # define pwrite eio__pwrite |
495 | |
|
|
496 | /* |
|
|
497 | * make our pread/pwrite safe against themselves, but not against |
|
|
498 | * normal read/write by using a mutex. slows down execution a lot, |
|
|
499 | * but that's your problem, not mine. |
|
|
500 | */ |
|
|
501 | static mutex_t preadwritelock = X_MUTEX_INIT; |
|
|
502 | |
743 | |
503 | static ssize_t |
744 | static ssize_t |
504 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
745 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
505 | { |
746 | { |
506 | ssize_t res; |
747 | ssize_t res; |
… | |
… | |
568 | |
809 | |
569 | #if !HAVE_READAHEAD |
810 | #if !HAVE_READAHEAD |
570 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
811 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
571 | |
812 | |
572 | static ssize_t |
813 | static ssize_t |
573 | eio__readahead (int fd, off_t offset, size_t count, worker *self) |
814 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
574 | { |
815 | { |
575 | size_t todo = count; |
816 | size_t todo = count; |
576 | dBUF; |
817 | dBUF; |
577 | |
818 | |
578 | while (todo > 0) |
819 | while (todo > 0) |
… | |
… | |
588 | return count; |
829 | return count; |
589 | } |
830 | } |
590 | |
831 | |
591 | #endif |
832 | #endif |
592 | |
833 | |
593 | #if !HAVE_READDIR_R |
|
|
594 | # define readdir_r eio__readdir_r |
|
|
595 | |
|
|
596 | static mutex_t readdirlock = X_MUTEX_INIT; |
|
|
597 | |
|
|
598 | static int |
|
|
599 | eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res) |
|
|
600 | { |
|
|
601 | EIO_STRUCT_DIRENT *e; |
|
|
602 | int errorno; |
|
|
603 | |
|
|
604 | X_LOCK (readdirlock); |
|
|
605 | |
|
|
606 | e = readdir (dirp); |
|
|
607 | errorno = errno; |
|
|
608 | |
|
|
609 | if (e) |
|
|
610 | { |
|
|
611 | *res = ent; |
|
|
612 | strcpy (ent->d_name, e->d_name); |
|
|
613 | } |
|
|
614 | else |
|
|
615 | *res = 0; |
|
|
616 | |
|
|
617 | X_UNLOCK (readdirlock); |
|
|
618 | |
|
|
619 | errno = errorno; |
|
|
620 | return e ? 0 : -1; |
|
|
621 | } |
|
|
622 | #endif |
|
|
623 | |
|
|
624 | /* sendfile always needs emulation */ |
834 | /* sendfile always needs emulation */ |
625 | static ssize_t |
835 | static ssize_t |
626 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) |
836 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
627 | { |
837 | { |
628 | ssize_t res; |
838 | ssize_t res; |
629 | |
839 | |
630 | if (!count) |
840 | if (!count) |
631 | return 0; |
841 | return 0; |
… | |
… | |
715 | return res; |
925 | return res; |
716 | } |
926 | } |
717 | |
927 | |
718 | /* read a full directory */ |
928 | /* read a full directory */ |
719 | static void |
929 | static void |
720 | eio__scandir (eio_req *req, worker *self) |
930 | eio__scandir (eio_req *req, etp_worker *self) |
721 | { |
931 | { |
722 | DIR *dirp; |
932 | DIR *dirp; |
723 | union |
|
|
724 | { |
|
|
725 | EIO_STRUCT_DIRENT d; |
|
|
726 | char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1]; |
|
|
727 | } *u; |
|
|
728 | EIO_STRUCT_DIRENT *entp; |
933 | EIO_STRUCT_DIRENT *entp; |
729 | char *name, *names; |
934 | char *name, *names; |
730 | int memlen = 4096; |
935 | int memlen = 4096; |
731 | int memofs = 0; |
936 | int memofs = 0; |
732 | int res = 0; |
937 | int res = 0; |
733 | |
938 | |
734 | X_LOCK (wrklock); |
939 | X_LOCK (wrklock); |
|
|
940 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
735 | self->dirp = dirp = opendir (req->ptr1); |
941 | self->dirp = dirp = opendir (req->ptr1); |
736 | self->dbuf = u = malloc (sizeof (*u)); |
|
|
737 | req->flags |= EIO_FLAG_PTR2_FREE; |
942 | req->flags |= EIO_FLAG_PTR2_FREE; |
738 | req->ptr2 = names = malloc (memlen); |
943 | req->ptr2 = names = malloc (memlen); |
739 | X_UNLOCK (wrklock); |
944 | X_UNLOCK (wrklock); |
740 | |
945 | |
741 | if (dirp && u && names) |
946 | if (dirp && names) |
742 | for (;;) |
947 | for (;;) |
743 | { |
948 | { |
744 | errno = 0; |
949 | errno = 0; |
745 | readdir_r (dirp, &u->d, &entp); |
950 | entp = readdir (dirp); |
746 | |
951 | |
747 | if (!entp) |
952 | if (!entp) |
748 | break; |
953 | break; |
749 | |
954 | |
750 | name = entp->d_name; |
955 | name = entp->d_name; |
… | |
… | |
773 | |
978 | |
774 | if (errno) |
979 | if (errno) |
775 | res = -1; |
980 | res = -1; |
776 | |
981 | |
777 | req->result = res; |
982 | req->result = res; |
|
|
983 | } |
|
|
984 | |
|
|
985 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
|
|
986 | # define msync(a,b,c) ENOSYS |
|
|
987 | #endif |
|
|
988 | |
|
|
989 | int |
|
|
990 | eio__mtouch (void *mem, size_t len, int flags) |
|
|
991 | { |
|
|
992 | intptr_t addr = (intptr_t)mem; |
|
|
993 | intptr_t end = addr + len; |
|
|
994 | #ifdef PAGESIZE |
|
|
995 | const intptr_t page = PAGESIZE; |
|
|
996 | #else |
|
|
997 | static intptr_t page; |
|
|
998 | |
|
|
999 | if (!page) |
|
|
1000 | page = sysconf (_SC_PAGESIZE); |
|
|
1001 | #endif |
|
|
1002 | |
|
|
1003 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1004 | |
|
|
1005 | if (addr < end) |
|
|
1006 | if (flags) /* modify */ |
|
|
1007 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
|
|
1008 | else |
|
|
1009 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1010 | |
|
|
1011 | return 0; |
778 | } |
1012 | } |
779 | |
1013 | |
780 | /*****************************************************************************/ |
1014 | /*****************************************************************************/ |
781 | |
1015 | |
782 | #define ALLOC(len) \ |
1016 | #define ALLOC(len) \ |
… | |
… | |
792 | req->result = -1; \ |
1026 | req->result = -1; \ |
793 | break; \ |
1027 | break; \ |
794 | } \ |
1028 | } \ |
795 | } |
1029 | } |
796 | |
1030 | |
797 | X_THREAD_PROC (eio_proc) |
1031 | X_THREAD_PROC (etp_proc) |
798 | { |
1032 | { |
799 | eio_req *req; |
1033 | ETP_REQ *req; |
800 | struct timespec ts; |
1034 | struct timespec ts; |
801 | worker *self = (worker *)thr_arg; |
1035 | etp_worker *self = (etp_worker *)thr_arg; |
802 | |
1036 | |
803 | /* try to distribute timeouts somewhat randomly */ |
1037 | /* try to distribute timeouts somewhat randomly */ |
804 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1038 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
805 | |
1039 | |
806 | for (;;) |
1040 | for (;;) |
807 | { |
1041 | { |
808 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
|
|
809 | |
|
|
810 | X_LOCK (reqlock); |
1042 | X_LOCK (reqlock); |
811 | |
1043 | |
812 | for (;;) |
1044 | for (;;) |
813 | { |
1045 | { |
814 | self->req = req = reqq_shift (&req_queue); |
1046 | self->req = req = reqq_shift (&req_queue); |
… | |
… | |
816 | if (req) |
1048 | if (req) |
817 | break; |
1049 | break; |
818 | |
1050 | |
819 | ++idle; |
1051 | ++idle; |
820 | |
1052 | |
|
|
1053 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
821 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1054 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
822 | { |
1055 | { |
823 | if (idle > max_idle) |
1056 | if (idle > max_idle) |
824 | { |
1057 | { |
825 | --idle; |
1058 | --idle; |
… | |
… | |
830 | goto quit; |
1063 | goto quit; |
831 | } |
1064 | } |
832 | |
1065 | |
833 | /* we are allowed to idle, so do so without any timeout */ |
1066 | /* we are allowed to idle, so do so without any timeout */ |
834 | X_COND_WAIT (reqwait, reqlock); |
1067 | X_COND_WAIT (reqwait, reqlock); |
835 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
|
|
836 | } |
1068 | } |
837 | |
1069 | |
838 | --idle; |
1070 | --idle; |
839 | } |
1071 | } |
840 | |
1072 | |
841 | --nready; |
1073 | --nready; |
842 | |
1074 | |
843 | X_UNLOCK (reqlock); |
1075 | X_UNLOCK (reqlock); |
844 | |
1076 | |
845 | errno = 0; /* strictly unnecessary */ |
1077 | if (req->type < 0) |
|
|
1078 | goto quit; |
846 | |
1079 | |
847 | if (!EIO_CANCELLED (req)) |
1080 | if (!EIO_CANCELLED (req)) |
848 | switch (req->type) |
1081 | ETP_EXECUTE (self, req); |
849 | { |
|
|
850 | case EIO_READ: ALLOC (req->size); |
|
|
851 | req->result = req->offs >= 0 |
|
|
852 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
|
|
853 | : read (req->int1, req->ptr2, req->size); break; |
|
|
854 | case EIO_WRITE: req->result = req->offs >= 0 |
|
|
855 | ? pwrite (req->int1, req->ptr2, req->size, req->offs) |
|
|
856 | : write (req->int1, req->ptr2, req->size); break; |
|
|
857 | |
|
|
858 | case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; |
|
|
859 | case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; |
|
|
860 | |
|
|
861 | case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
862 | req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
863 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
864 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
865 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
866 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
867 | |
|
|
868 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
|
|
869 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
|
|
870 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
|
|
871 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
|
|
872 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
|
|
873 | case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; |
|
|
874 | |
|
|
875 | case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; |
|
|
876 | case EIO_CLOSE: req->result = close (req->int1); break; |
|
|
877 | case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; |
|
|
878 | case EIO_UNLINK: req->result = unlink (req->ptr1); break; |
|
|
879 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
|
|
880 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
|
|
881 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
|
|
882 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
|
|
883 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
|
|
884 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
|
|
885 | |
|
|
886 | case EIO_READLINK: ALLOC (NAME_MAX); |
|
|
887 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
|
|
888 | |
|
|
889 | case EIO_SYNC: req->result = 0; sync (); break; |
|
|
890 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
|
|
891 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
|
|
892 | |
|
|
893 | case EIO_READDIR: eio__scandir (req, self); break; |
|
|
894 | |
|
|
895 | case EIO_BUSY: |
|
|
896 | #ifdef _WIN32 |
|
|
897 | Sleep (req->nv1 * 1000.); |
|
|
898 | #else |
|
|
899 | { |
|
|
900 | struct timeval tv; |
|
|
901 | |
|
|
902 | tv.tv_sec = req->nv1; |
|
|
903 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
|
|
904 | |
|
|
905 | req->result = select (0, 0, 0, 0, &tv); |
|
|
906 | } |
|
|
907 | #endif |
|
|
908 | break; |
|
|
909 | |
|
|
910 | case EIO_UTIME: |
|
|
911 | case EIO_FUTIME: |
|
|
912 | { |
|
|
913 | struct timeval tv[2]; |
|
|
914 | struct timeval *times; |
|
|
915 | |
|
|
916 | if (req->nv1 != -1. || req->nv2 != -1.) |
|
|
917 | { |
|
|
918 | tv[0].tv_sec = req->nv1; |
|
|
919 | tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; |
|
|
920 | tv[1].tv_sec = req->nv2; |
|
|
921 | tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; |
|
|
922 | |
|
|
923 | times = tv; |
|
|
924 | } |
|
|
925 | else |
|
|
926 | times = 0; |
|
|
927 | |
|
|
928 | |
|
|
929 | req->result = req->type == EIO_FUTIME |
|
|
930 | ? futimes (req->int1, times) |
|
|
931 | : utimes (req->ptr1, times); |
|
|
932 | } |
|
|
933 | |
|
|
934 | case EIO_GROUP: |
|
|
935 | case EIO_NOP: |
|
|
936 | req->result = 0; |
|
|
937 | break; |
|
|
938 | |
|
|
939 | case EIO_QUIT: |
|
|
940 | goto quit; |
|
|
941 | |
|
|
942 | default: |
|
|
943 | req->result = -1; |
|
|
944 | break; |
|
|
945 | } |
|
|
946 | |
|
|
947 | req->errorno = errno; |
|
|
948 | |
1082 | |
949 | X_LOCK (reslock); |
1083 | X_LOCK (reslock); |
950 | |
1084 | |
951 | ++npending; |
1085 | ++npending; |
952 | |
1086 | |
953 | if (!reqq_push (&res_queue, req) && want_poll_cb) |
1087 | if (!reqq_push (&res_queue, req) && want_poll_cb) |
954 | want_poll_cb (); |
1088 | want_poll_cb (); |
955 | |
1089 | |
956 | self->req = 0; |
1090 | self->req = 0; |
957 | worker_clear (self); |
1091 | etp_worker_clear (self); |
958 | |
1092 | |
959 | X_UNLOCK (reslock); |
1093 | X_UNLOCK (reslock); |
960 | } |
1094 | } |
961 | |
1095 | |
962 | quit: |
1096 | quit: |
963 | X_LOCK (wrklock); |
1097 | X_LOCK (wrklock); |
964 | worker_free (self); |
1098 | etp_worker_free (self); |
965 | X_UNLOCK (wrklock); |
1099 | X_UNLOCK (wrklock); |
966 | |
1100 | |
967 | return 0; |
1101 | return 0; |
968 | } |
1102 | } |
969 | |
1103 | |
970 | /*****************************************************************************/ |
1104 | /*****************************************************************************/ |
971 | |
1105 | |
972 | static void eio_atfork_prepare (void) |
|
|
973 | { |
|
|
974 | X_LOCK (wrklock); |
|
|
975 | X_LOCK (reqlock); |
|
|
976 | X_LOCK (reslock); |
|
|
977 | #if !HAVE_PREADWRITE |
|
|
978 | X_LOCK (preadwritelock); |
|
|
979 | #endif |
|
|
980 | #if !HAVE_READDIR_R |
|
|
981 | X_LOCK (readdirlock); |
|
|
982 | #endif |
|
|
983 | } |
|
|
984 | |
|
|
985 | static void eio_atfork_parent (void) |
|
|
986 | { |
|
|
987 | #if !HAVE_READDIR_R |
|
|
988 | X_UNLOCK (readdirlock); |
|
|
989 | #endif |
|
|
990 | #if !HAVE_PREADWRITE |
|
|
991 | X_UNLOCK (preadwritelock); |
|
|
992 | #endif |
|
|
993 | X_UNLOCK (reslock); |
|
|
994 | X_UNLOCK (reqlock); |
|
|
995 | X_UNLOCK (wrklock); |
|
|
996 | } |
|
|
997 | |
|
|
998 | static void eio_atfork_child (void) |
|
|
999 | { |
|
|
1000 | eio_req *prv; |
|
|
1001 | |
|
|
1002 | while (prv = reqq_shift (&req_queue)) |
|
|
1003 | eio_destroy (prv); |
|
|
1004 | |
|
|
1005 | while (prv = reqq_shift (&res_queue)) |
|
|
1006 | eio_destroy (prv); |
|
|
1007 | |
|
|
1008 | while (wrk_first.next != &wrk_first) |
|
|
1009 | { |
|
|
1010 | worker *wrk = wrk_first.next; |
|
|
1011 | |
|
|
1012 | if (wrk->req) |
|
|
1013 | eio_destroy (wrk->req); |
|
|
1014 | |
|
|
1015 | worker_clear (wrk); |
|
|
1016 | worker_free (wrk); |
|
|
1017 | } |
|
|
1018 | |
|
|
1019 | started = 0; |
|
|
1020 | idle = 0; |
|
|
1021 | nreqs = 0; |
|
|
1022 | nready = 0; |
|
|
1023 | npending = 0; |
|
|
1024 | |
|
|
1025 | eio_atfork_parent (); |
|
|
1026 | } |
|
|
1027 | |
|
|
1028 | int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1106 | int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1029 | { |
1107 | { |
1030 | want_poll_cb = want_poll; |
1108 | return etp_init (want_poll, done_poll); |
1031 | done_poll_cb = done_poll; |
|
|
1032 | |
|
|
1033 | #ifdef _WIN32 |
|
|
1034 | X_MUTEX_CHECK (wrklock); |
|
|
1035 | X_MUTEX_CHECK (reslock); |
|
|
1036 | X_MUTEX_CHECK (reqlock); |
|
|
1037 | X_MUTEX_CHECK (reqwait); |
|
|
1038 | X_MUTEX_CHECK (preadwritelock); |
|
|
1039 | X_MUTEX_CHECK (readdirlock); |
|
|
1040 | |
|
|
1041 | X_COND_CHECK (reqwait); |
|
|
1042 | #endif |
|
|
1043 | |
|
|
1044 | X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child); |
|
|
1045 | } |
1109 | } |
1046 | |
1110 | |
1047 | static void eio_api_destroy (eio_req *req) |
1111 | static void eio_api_destroy (eio_req *req) |
1048 | { |
1112 | { |
1049 | free (req); |
1113 | free (req); |
… | |
… | |
1071 | { \ |
1135 | { \ |
1072 | eio_api_destroy (req); \ |
1136 | eio_api_destroy (req); \ |
1073 | return 0; \ |
1137 | return 0; \ |
1074 | } |
1138 | } |
1075 | |
1139 | |
|
|
1140 | static void eio_execute (etp_worker *self, eio_req *req) |
|
|
1141 | { |
|
|
1142 | errno = 0; |
|
|
1143 | |
|
|
1144 | switch (req->type) |
|
|
1145 | { |
|
|
1146 | case EIO_READ: ALLOC (req->size); |
|
|
1147 | req->result = req->offs >= 0 |
|
|
1148 | ? pread (req->int1, req->ptr2, req->size, req->offs) |
|
|
1149 | : read (req->int1, req->ptr2, req->size); break; |
|
|
1150 | case EIO_WRITE: req->result = req->offs >= 0 |
|
|
1151 | ? pwrite (req->int1, req->ptr2, req->size, req->offs) |
|
|
1152 | : write (req->int1, req->ptr2, req->size); break; |
|
|
1153 | |
|
|
1154 | case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; |
|
|
1155 | case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; |
|
|
1156 | |
|
|
1157 | case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
1158 | req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
1159 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
1160 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
1161 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
|
|
1162 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
|
|
1163 | |
|
|
1164 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
|
|
1165 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
|
|
1166 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
|
|
1167 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
|
|
1168 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
|
|
1169 | case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; |
|
|
1170 | |
|
|
1171 | case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break; |
|
|
1172 | case EIO_CLOSE: req->result = close (req->int1); break; |
|
|
1173 | case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; |
|
|
1174 | case EIO_UNLINK: req->result = unlink (req->ptr1); break; |
|
|
1175 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
|
|
1176 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
|
|
1177 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
|
|
1178 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
|
|
1179 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
|
|
1180 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
|
|
1181 | |
|
|
1182 | case EIO_READLINK: ALLOC (NAME_MAX); |
|
|
1183 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
|
|
1184 | |
|
|
1185 | case EIO_SYNC: req->result = 0; sync (); break; |
|
|
1186 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
|
|
1187 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
|
|
1188 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
|
|
1189 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
|
|
1190 | |
|
|
1191 | case EIO_READDIR: eio__scandir (req, self); break; |
|
|
1192 | |
|
|
1193 | case EIO_BUSY: |
|
|
1194 | #ifdef _WIN32 |
|
|
1195 | Sleep (req->nv1 * 1000.); |
|
|
1196 | #else |
|
|
1197 | { |
|
|
1198 | struct timeval tv; |
|
|
1199 | |
|
|
1200 | tv.tv_sec = req->nv1; |
|
|
1201 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
|
|
1202 | |
|
|
1203 | req->result = select (0, 0, 0, 0, &tv); |
|
|
1204 | } |
|
|
1205 | #endif |
|
|
1206 | break; |
|
|
1207 | |
|
|
1208 | case EIO_UTIME: |
|
|
1209 | case EIO_FUTIME: |
|
|
1210 | { |
|
|
1211 | struct timeval tv[2]; |
|
|
1212 | struct timeval *times; |
|
|
1213 | |
|
|
1214 | if (req->nv1 != -1. || req->nv2 != -1.) |
|
|
1215 | { |
|
|
1216 | tv[0].tv_sec = req->nv1; |
|
|
1217 | tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.; |
|
|
1218 | tv[1].tv_sec = req->nv2; |
|
|
1219 | tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.; |
|
|
1220 | |
|
|
1221 | times = tv; |
|
|
1222 | } |
|
|
1223 | else |
|
|
1224 | times = 0; |
|
|
1225 | |
|
|
1226 | |
|
|
1227 | req->result = req->type == EIO_FUTIME |
|
|
1228 | ? futimes (req->int1, times) |
|
|
1229 | : utimes (req->ptr1, times); |
|
|
1230 | } |
|
|
1231 | |
|
|
1232 | case EIO_GROUP: |
|
|
1233 | abort (); /* handled in eio_request */ |
|
|
1234 | |
|
|
1235 | case EIO_NOP: |
|
|
1236 | req->result = 0; |
|
|
1237 | break; |
|
|
1238 | |
|
|
1239 | case EIO_CUSTOM: |
|
|
1240 | ((void (*)(eio_req *))req->feed) (req); |
|
|
1241 | break; |
|
|
1242 | |
|
|
1243 | default: |
|
|
1244 | req->result = -1; |
|
|
1245 | break; |
|
|
1246 | } |
|
|
1247 | |
|
|
1248 | req->errorno = errno; |
|
|
1249 | } |
|
|
1250 | |
1076 | #ifndef EIO_NO_WRAPPERS |
1251 | #ifndef EIO_NO_WRAPPERS |
1077 | |
1252 | |
1078 | eio_req *eio_nop (int pri, eio_cb cb, void *data) |
1253 | eio_req *eio_nop (int pri, eio_cb cb, void *data) |
1079 | { |
1254 | { |
1080 | REQ (EIO_NOP); SEND; |
1255 | REQ (EIO_NOP); SEND; |
… | |
… | |
1091 | } |
1266 | } |
1092 | |
1267 | |
1093 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1268 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1094 | { |
1269 | { |
1095 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1270 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
|
|
1271 | } |
|
|
1272 | |
|
|
1273 | eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1274 | { |
|
|
1275 | REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
1276 | } |
|
|
1277 | |
|
|
1278 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
1279 | { |
|
|
1280 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1096 | } |
1281 | } |
1097 | |
1282 | |
1098 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1283 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1099 | { |
1284 | { |
1100 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1285 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
… | |
… | |
1221 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
1406 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
1222 | } |
1407 | } |
1223 | |
1408 | |
1224 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1409 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1225 | { |
1410 | { |
1226 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; |
1411 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
1227 | } |
1412 | } |
1228 | |
1413 | |
1229 | static eio_req * |
1414 | static eio_req * |
1230 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1415 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1231 | { |
1416 | { |
… | |
… | |
1255 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1440 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1256 | { |
1441 | { |
1257 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1442 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1258 | } |
1443 | } |
1259 | |
1444 | |
|
|
1445 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
|
|
1446 | { |
|
|
1447 | REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
|
|
1448 | } |
|
|
1449 | |
1260 | #endif |
1450 | #endif |
1261 | |
1451 | |
1262 | eio_req *eio_grp (eio_cb cb, void *data) |
1452 | eio_req *eio_grp (eio_cb cb, void *data) |
1263 | { |
1453 | { |
1264 | const int pri = EIO_PRI_MAX; |
1454 | const int pri = EIO_PRI_MAX; |
… | |
… | |
1307 | /*****************************************************************************/ |
1497 | /*****************************************************************************/ |
1308 | /* misc garbage */ |
1498 | /* misc garbage */ |
1309 | |
1499 | |
1310 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1500 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1311 | { |
1501 | { |
1312 | worker wrk; |
1502 | etp_worker wrk; |
1313 | |
1503 | |
1314 | wrk.dbuf = 0; |
1504 | wrk.dbuf = 0; |
1315 | |
1505 | |
1316 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
1506 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
1317 | |
1507 | |