1 | /* |
1 | /* |
2 | * libeio implementation |
2 | * libeio implementation |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> |
4 | * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> |
5 | * All rights reserved. |
5 | * All rights reserved. |
6 | * |
6 | * |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * tion, are permitted provided that the following conditions are met: |
8 | * tion, are permitted provided that the following conditions are met: |
9 | * |
9 | * |
… | |
… | |
35 | * and other provisions required by the GPL. If you do not delete the |
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 |
36 | * provisions above, a recipient may use your version of this file under |
37 | * either the BSD or the GPL. |
37 | * either the BSD or the GPL. |
38 | */ |
38 | */ |
39 | |
39 | |
|
|
40 | #ifndef _WIN32 |
|
|
41 | # include "config.h" |
|
|
42 | #endif |
|
|
43 | |
40 | #include "eio.h" |
44 | #include "eio.h" |
|
|
45 | #include "ecb.h" |
|
|
46 | |
|
|
47 | #ifdef EIO_STACKSIZE |
|
|
48 | # define XTHREAD_STACKSIZE EIO_STACKSIZE |
|
|
49 | #endif |
41 | #include "xthread.h" |
50 | #include "xthread.h" |
42 | |
51 | |
43 | #include <errno.h> |
52 | #include <errno.h> |
44 | #include <stddef.h> |
53 | #include <stddef.h> |
45 | #include <stdlib.h> |
54 | #include <stdlib.h> |
46 | #include <string.h> |
55 | #include <string.h> |
47 | #include <errno.h> |
56 | #include <errno.h> |
48 | #include <sys/types.h> |
57 | #include <sys/types.h> |
49 | #include <sys/stat.h> |
58 | #include <sys/stat.h> |
|
|
59 | #include <sys/statvfs.h> |
50 | #include <limits.h> |
60 | #include <limits.h> |
51 | #include <fcntl.h> |
61 | #include <fcntl.h> |
52 | #include <assert.h> |
62 | #include <assert.h> |
53 | |
63 | |
|
|
64 | /* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ |
|
|
65 | /* intptr_t only comes from stdint.h, says idiot openbsd coder */ |
|
|
66 | #if HAVE_STDINT_H |
|
|
67 | # include <stdint.h> |
|
|
68 | #endif |
|
|
69 | |
|
|
70 | #ifndef ECANCELED |
|
|
71 | # define ECANCELED EDOM |
|
|
72 | #endif |
|
|
73 | |
|
|
74 | static void eio_destroy (eio_req *req); |
|
|
75 | |
54 | #ifndef EIO_FINISH |
76 | #ifndef EIO_FINISH |
55 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
77 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
56 | #endif |
78 | #endif |
57 | |
79 | |
58 | #ifndef EIO_DESTROY |
80 | #ifndef EIO_DESTROY |
… | |
… | |
64 | #endif |
86 | #endif |
65 | |
87 | |
66 | #ifdef _WIN32 |
88 | #ifdef _WIN32 |
67 | |
89 | |
68 | /*doh*/ |
90 | /*doh*/ |
69 | |
|
|
70 | #else |
91 | #else |
71 | |
92 | |
72 | # include "config.h" |
|
|
73 | # include <sys/time.h> |
93 | # include <sys/time.h> |
74 | # include <sys/select.h> |
94 | # include <sys/select.h> |
75 | # include <unistd.h> |
95 | # include <unistd.h> |
76 | # include <utime.h> |
96 | # include <utime.h> |
77 | # include <signal.h> |
97 | # include <signal.h> |
78 | # include <dirent.h> |
98 | # include <dirent.h> |
79 | |
99 | |
|
|
100 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
101 | # include <sys/mman.h> |
|
|
102 | #endif |
|
|
103 | |
|
|
104 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
|
|
105 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
|
|
106 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
|
|
107 | # define D_INO(de) (de)->d_fileno |
|
|
108 | # define D_NAMLEN(de) (de)->d_namlen |
|
|
109 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
|
|
110 | # define D_INO(de) (de)->d_ino |
|
|
111 | # endif |
|
|
112 | |
|
|
113 | #ifdef _D_EXACT_NAMLEN |
|
|
114 | # undef D_NAMLEN |
|
|
115 | # define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
|
|
116 | #endif |
|
|
117 | |
|
|
118 | # ifdef _DIRENT_HAVE_D_TYPE |
|
|
119 | # define D_TYPE(de) (de)->d_type |
|
|
120 | # endif |
|
|
121 | |
80 | # ifndef EIO_STRUCT_DIRENT |
122 | # ifndef EIO_STRUCT_DIRENT |
81 | # define EIO_STRUCT_DIRENT struct dirent |
123 | # define EIO_STRUCT_DIRENT struct dirent |
82 | # endif |
124 | # endif |
83 | |
125 | |
84 | #endif |
126 | #endif |
85 | |
127 | |
86 | #if HAVE_SENDFILE |
128 | #if HAVE_SENDFILE |
87 | # if __linux |
129 | # if __linux |
88 | # include <sys/sendfile.h> |
130 | # include <sys/sendfile.h> |
89 | # elif __freebsd |
131 | # elif __FreeBSD__ || defined __APPLE__ |
90 | # include <sys/socket.h> |
132 | # include <sys/socket.h> |
91 | # include <sys/uio.h> |
133 | # include <sys/uio.h> |
92 | # elif __hpux |
134 | # elif __hpux |
93 | # include <sys/socket.h> |
135 | # include <sys/socket.h> |
94 | # elif __solaris /* not yet */ |
136 | # elif __solaris |
95 | # include <sys/sendfile.h> |
137 | # include <sys/sendfile.h> |
96 | # else |
138 | # else |
97 | # error sendfile support requested but not available |
139 | # error sendfile support requested but not available |
98 | # endif |
140 | # endif |
99 | #endif |
141 | #endif |
100 | |
142 | |
101 | /* number of seconds after which an idle threads exit */ |
143 | #ifndef D_TYPE |
102 | #define IDLE_TIMEOUT 10 |
144 | # define D_TYPE(de) 0 |
|
|
145 | #endif |
|
|
146 | #ifndef D_INO |
|
|
147 | # define D_INO(de) 0 |
|
|
148 | #endif |
|
|
149 | #ifndef D_NAMLEN |
|
|
150 | # define D_NAMLEN(de) strlen ((de)->d_name) |
|
|
151 | #endif |
103 | |
152 | |
104 | /* used for struct dirent, AIX doesn't provide it */ |
153 | /* used for struct dirent, AIX doesn't provide it */ |
105 | #ifndef NAME_MAX |
154 | #ifndef NAME_MAX |
106 | # define NAME_MAX 4096 |
155 | # define NAME_MAX 4096 |
|
|
156 | #endif |
|
|
157 | |
|
|
158 | /* used for readlink etc. */ |
|
|
159 | #ifndef PATH_MAX |
|
|
160 | # define PATH_MAX 4096 |
107 | #endif |
161 | #endif |
108 | |
162 | |
109 | /* buffer size for various temporary buffers */ |
163 | /* buffer size for various temporary buffers */ |
110 | #define EIO_BUFSIZE 65536 |
164 | #define EIO_BUFSIZE 65536 |
111 | |
165 | |
… | |
… | |
117 | errno = ENOMEM; \ |
171 | errno = ENOMEM; \ |
118 | if (!eio_buf) \ |
172 | if (!eio_buf) \ |
119 | return -1; |
173 | return -1; |
120 | |
174 | |
121 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
175 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
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 | |
176 | |
136 | #define ETP_PRI_MIN EIO_PRI_MIN |
177 | #define ETP_PRI_MIN EIO_PRI_MIN |
137 | #define ETP_PRI_MAX EIO_PRI_MAX |
178 | #define ETP_PRI_MAX EIO_PRI_MAX |
138 | |
179 | |
139 | struct etp_worker; |
180 | struct etp_worker; |
… | |
… | |
155 | if (wrk->dirp) \ |
196 | if (wrk->dirp) \ |
156 | { \ |
197 | { \ |
157 | closedir (wrk->dirp); \ |
198 | closedir (wrk->dirp); \ |
158 | wrk->dirp = 0; \ |
199 | wrk->dirp = 0; \ |
159 | } |
200 | } |
|
|
201 | |
160 | #define ETP_WORKER_COMMON \ |
202 | #define ETP_WORKER_COMMON \ |
161 | void *dbuf; \ |
203 | void *dbuf; \ |
162 | DIR *dirp; |
204 | DIR *dirp; |
163 | |
205 | |
164 | /*****************************************************************************/ |
206 | /*****************************************************************************/ |
165 | |
207 | |
166 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
208 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
167 | |
209 | |
168 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
210 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
|
|
211 | ecb_inline int |
169 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
212 | tvdiff (struct timeval *tv1, struct timeval *tv2) |
170 | { |
213 | { |
171 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
214 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
172 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
215 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
173 | } |
216 | } |
174 | |
217 | |
… | |
… | |
181 | static unsigned int max_poll_reqs; /* reslock */ |
224 | static unsigned int max_poll_reqs; /* reslock */ |
182 | |
225 | |
183 | static volatile unsigned int nreqs; /* reqlock */ |
226 | static volatile unsigned int nreqs; /* reqlock */ |
184 | static volatile unsigned int nready; /* reqlock */ |
227 | static volatile unsigned int nready; /* reqlock */ |
185 | static volatile unsigned int npending; /* reqlock */ |
228 | static volatile unsigned int npending; /* reqlock */ |
186 | static volatile unsigned int max_idle = 4; |
229 | static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
|
|
230 | static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
187 | |
231 | |
188 | static mutex_t wrklock = X_MUTEX_INIT; |
232 | static xmutex_t wrklock; |
189 | static mutex_t reslock = X_MUTEX_INIT; |
233 | static xmutex_t reslock; |
190 | static mutex_t reqlock = X_MUTEX_INIT; |
234 | static xmutex_t reqlock; |
191 | static cond_t reqwait = X_COND_INIT; |
235 | static xcond_t reqwait; |
192 | |
236 | |
193 | #if !HAVE_PREADWRITE |
237 | #if !HAVE_PREADWRITE |
194 | /* |
238 | /* |
195 | * make our pread/pwrite emulation safe against themselves, but not against |
239 | * make our pread/pwrite emulation safe against themselves, but not against |
196 | * normal read/write by using a mutex. slows down execution a lot, |
240 | * normal read/write by using a mutex. slows down execution a lot, |
197 | * but that's your problem, not mine. |
241 | * but that's your problem, not mine. |
198 | */ |
242 | */ |
199 | static mutex_t preadwritelock = X_MUTEX_INIT; |
243 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
200 | #endif |
244 | #endif |
201 | |
245 | |
202 | typedef struct etp_worker |
246 | typedef struct etp_worker |
203 | { |
247 | { |
204 | /* locked by wrklock */ |
248 | /* locked by wrklock */ |
205 | struct etp_worker *prev, *next; |
249 | struct etp_worker *prev, *next; |
206 | |
250 | |
207 | thread_t tid; |
251 | xthread_t tid; |
208 | |
252 | |
209 | /* locked by reslock, reqlock or wrklock */ |
253 | /* locked by reslock, reqlock or wrklock */ |
210 | ETP_REQ *req; /* currently processed request */ |
254 | ETP_REQ *req; /* currently processed request */ |
211 | |
255 | |
212 | ETP_WORKER_COMMON |
256 | ETP_WORKER_COMMON |
… | |
… | |
217 | #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
261 | #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
218 | #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
262 | #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
219 | |
263 | |
220 | /* worker threads management */ |
264 | /* worker threads management */ |
221 | |
265 | |
|
|
266 | static void ecb_cold |
222 | static void etp_worker_clear (etp_worker *wrk) |
267 | etp_worker_clear (etp_worker *wrk) |
223 | { |
268 | { |
224 | ETP_WORKER_CLEAR (wrk); |
269 | ETP_WORKER_CLEAR (wrk); |
225 | } |
270 | } |
226 | |
271 | |
|
|
272 | static void ecb_cold |
227 | static void etp_worker_free (etp_worker *wrk) |
273 | etp_worker_free (etp_worker *wrk) |
228 | { |
274 | { |
229 | wrk->next->prev = wrk->prev; |
275 | wrk->next->prev = wrk->prev; |
230 | wrk->prev->next = wrk->next; |
276 | wrk->prev->next = wrk->next; |
231 | |
277 | |
232 | free (wrk); |
278 | free (wrk); |
233 | } |
279 | } |
234 | |
280 | |
235 | static unsigned int etp_nreqs (void) |
281 | static unsigned int |
|
|
282 | etp_nreqs (void) |
236 | { |
283 | { |
237 | int retval; |
284 | int retval; |
238 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
285 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
239 | retval = nreqs; |
286 | retval = nreqs; |
240 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
287 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
241 | return retval; |
288 | return retval; |
242 | } |
289 | } |
243 | |
290 | |
244 | static unsigned int etp_nready (void) |
291 | static unsigned int |
|
|
292 | etp_nready (void) |
245 | { |
293 | { |
246 | unsigned int retval; |
294 | unsigned int retval; |
247 | |
295 | |
248 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
296 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
249 | retval = nready; |
297 | retval = nready; |
250 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
298 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
251 | |
299 | |
252 | return retval; |
300 | return retval; |
253 | } |
301 | } |
254 | |
302 | |
255 | static unsigned int etp_npending (void) |
303 | static unsigned int |
|
|
304 | etp_npending (void) |
256 | { |
305 | { |
257 | unsigned int retval; |
306 | unsigned int retval; |
258 | |
307 | |
259 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
308 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
260 | retval = npending; |
309 | retval = npending; |
261 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
310 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
262 | |
311 | |
263 | return retval; |
312 | return retval; |
264 | } |
313 | } |
265 | |
314 | |
266 | static unsigned int etp_nthreads (void) |
315 | static unsigned int |
|
|
316 | etp_nthreads (void) |
267 | { |
317 | { |
268 | unsigned int retval; |
318 | unsigned int retval; |
269 | |
319 | |
270 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
320 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
271 | retval = started; |
321 | retval = started; |
… | |
… | |
285 | } etp_reqq; |
335 | } etp_reqq; |
286 | |
336 | |
287 | static etp_reqq req_queue; |
337 | static etp_reqq req_queue; |
288 | static etp_reqq res_queue; |
338 | static etp_reqq res_queue; |
289 | |
339 | |
|
|
340 | static int ecb_noinline |
290 | static int reqq_push (etp_reqq *q, ETP_REQ *req) |
341 | reqq_push (etp_reqq *q, ETP_REQ *req) |
291 | { |
342 | { |
292 | int pri = req->pri; |
343 | int pri = req->pri; |
293 | req->next = 0; |
344 | req->next = 0; |
294 | |
345 | |
295 | if (q->qe[pri]) |
346 | if (q->qe[pri]) |
… | |
… | |
301 | q->qe[pri] = q->qs[pri] = req; |
352 | q->qe[pri] = q->qs[pri] = req; |
302 | |
353 | |
303 | return q->size++; |
354 | return q->size++; |
304 | } |
355 | } |
305 | |
356 | |
|
|
357 | static ETP_REQ * ecb_noinline |
306 | static ETP_REQ *reqq_shift (etp_reqq *q) |
358 | reqq_shift (etp_reqq *q) |
307 | { |
359 | { |
308 | int pri; |
360 | int pri; |
309 | |
361 | |
310 | if (!q->size) |
362 | if (!q->size) |
311 | return 0; |
363 | return 0; |
… | |
… | |
326 | } |
378 | } |
327 | |
379 | |
328 | abort (); |
380 | abort (); |
329 | } |
381 | } |
330 | |
382 | |
331 | static void etp_atfork_prepare (void) |
383 | static void ecb_cold |
|
|
384 | etp_thread_init (void) |
332 | { |
385 | { |
333 | X_LOCK (wrklock); |
|
|
334 | X_LOCK (reqlock); |
|
|
335 | X_LOCK (reslock); |
|
|
336 | #if !HAVE_PREADWRITE |
386 | #if !HAVE_PREADWRITE |
337 | X_LOCK (preadwritelock); |
387 | X_MUTEX_CREATE (preadwritelock); |
338 | #endif |
388 | #endif |
|
|
389 | X_MUTEX_CREATE (wrklock); |
|
|
390 | X_MUTEX_CREATE (reslock); |
|
|
391 | X_MUTEX_CREATE (reqlock); |
|
|
392 | X_COND_CREATE (reqwait); |
339 | } |
393 | } |
340 | |
394 | |
|
|
395 | static void ecb_cold |
|
|
396 | etp_atfork_prepare (void) |
|
|
397 | { |
|
|
398 | } |
|
|
399 | |
|
|
400 | static void ecb_cold |
341 | static void etp_atfork_parent (void) |
401 | etp_atfork_parent (void) |
342 | { |
402 | { |
343 | #if !HAVE_PREADWRITE |
|
|
344 | X_UNLOCK (preadwritelock); |
|
|
345 | #endif |
|
|
346 | X_UNLOCK (reslock); |
|
|
347 | X_UNLOCK (reqlock); |
|
|
348 | X_UNLOCK (wrklock); |
|
|
349 | } |
403 | } |
350 | |
404 | |
|
|
405 | static void ecb_cold |
351 | static void etp_atfork_child (void) |
406 | etp_atfork_child (void) |
352 | { |
407 | { |
353 | ETP_REQ *prv; |
408 | ETP_REQ *prv; |
354 | |
409 | |
355 | while ((prv = reqq_shift (&req_queue))) |
410 | while ((prv = reqq_shift (&req_queue))) |
356 | ETP_DESTROY (prv); |
411 | ETP_DESTROY (prv); |
… | |
… | |
373 | idle = 0; |
428 | idle = 0; |
374 | nreqs = 0; |
429 | nreqs = 0; |
375 | nready = 0; |
430 | nready = 0; |
376 | npending = 0; |
431 | npending = 0; |
377 | |
432 | |
378 | etp_atfork_parent (); |
433 | etp_thread_init (); |
379 | } |
434 | } |
380 | |
435 | |
381 | static void |
436 | static void ecb_cold |
382 | etp_once_init (void) |
437 | etp_once_init (void) |
383 | { |
438 | { |
|
|
439 | etp_thread_init (); |
384 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
440 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
385 | } |
441 | } |
386 | |
442 | |
387 | static int |
443 | static int ecb_cold |
388 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
444 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
389 | { |
445 | { |
390 | static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
446 | static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
391 | |
447 | |
392 | pthread_once (&doinit, etp_once_init); |
448 | pthread_once (&doinit, etp_once_init); |
… | |
… | |
397 | return 0; |
453 | return 0; |
398 | } |
454 | } |
399 | |
455 | |
400 | X_THREAD_PROC (etp_proc); |
456 | X_THREAD_PROC (etp_proc); |
401 | |
457 | |
|
|
458 | static void ecb_cold |
402 | static void etp_start_thread (void) |
459 | etp_start_thread (void) |
403 | { |
460 | { |
404 | etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
461 | etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
405 | |
462 | |
406 | /*TODO*/ |
463 | /*TODO*/ |
407 | assert (("unable to allocate worker thread data", wrk)); |
464 | assert (("unable to allocate worker thread data", wrk)); |
… | |
… | |
420 | free (wrk); |
477 | free (wrk); |
421 | |
478 | |
422 | X_UNLOCK (wrklock); |
479 | X_UNLOCK (wrklock); |
423 | } |
480 | } |
424 | |
481 | |
|
|
482 | static void |
425 | static void etp_maybe_start_thread (void) |
483 | etp_maybe_start_thread (void) |
426 | { |
484 | { |
427 | if (expect_true (etp_nthreads () >= wanted)) |
485 | if (ecb_expect_true (etp_nthreads () >= wanted)) |
428 | return; |
486 | return; |
429 | |
487 | |
430 | /* todo: maybe use idle here, but might be less exact */ |
488 | /* todo: maybe use idle here, but might be less exact */ |
431 | if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
489 | if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
432 | return; |
490 | return; |
433 | |
491 | |
434 | etp_start_thread (); |
492 | etp_start_thread (); |
435 | } |
493 | } |
436 | |
494 | |
|
|
495 | static void ecb_cold |
437 | static void etp_end_thread (void) |
496 | etp_end_thread (void) |
438 | { |
497 | { |
439 | eio_req *req = calloc (1, sizeof (eio_req)); |
498 | eio_req *req = calloc (1, sizeof (eio_req)); |
440 | |
499 | |
441 | req->type = -1; |
500 | req->type = -1; |
442 | req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
501 | req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
… | |
… | |
449 | X_LOCK (wrklock); |
508 | X_LOCK (wrklock); |
450 | --started; |
509 | --started; |
451 | X_UNLOCK (wrklock); |
510 | X_UNLOCK (wrklock); |
452 | } |
511 | } |
453 | |
512 | |
454 | static int etp_poll (void) |
513 | static int |
|
|
514 | etp_poll (void) |
455 | { |
515 | { |
456 | unsigned int maxreqs; |
516 | unsigned int maxreqs; |
457 | unsigned int maxtime; |
517 | unsigned int maxtime; |
458 | struct timeval tv_start, tv_now; |
518 | struct timeval tv_start, tv_now; |
459 | |
519 | |
… | |
… | |
489 | |
549 | |
490 | X_LOCK (reqlock); |
550 | X_LOCK (reqlock); |
491 | --nreqs; |
551 | --nreqs; |
492 | X_UNLOCK (reqlock); |
552 | X_UNLOCK (reqlock); |
493 | |
553 | |
494 | if (expect_false (req->type == EIO_GROUP && req->size)) |
554 | if (ecb_expect_false (req->type == EIO_GROUP && req->size)) |
495 | { |
555 | { |
496 | req->int1 = 1; /* mark request as delayed */ |
556 | req->int1 = 1; /* mark request as delayed */ |
497 | continue; |
557 | continue; |
498 | } |
558 | } |
499 | else |
559 | else |
500 | { |
560 | { |
501 | int res = ETP_FINISH (req); |
561 | int res = ETP_FINISH (req); |
502 | if (expect_false (res)) |
562 | if (ecb_expect_false (res)) |
503 | return res; |
563 | return res; |
504 | } |
564 | } |
505 | |
565 | |
506 | if (expect_false (maxreqs && !--maxreqs)) |
566 | if (ecb_expect_false (maxreqs && !--maxreqs)) |
507 | break; |
567 | break; |
508 | |
568 | |
509 | if (maxtime) |
569 | if (maxtime) |
510 | { |
570 | { |
511 | gettimeofday (&tv_now, 0); |
571 | gettimeofday (&tv_now, 0); |
… | |
… | |
517 | |
577 | |
518 | errno = EAGAIN; |
578 | errno = EAGAIN; |
519 | return -1; |
579 | return -1; |
520 | } |
580 | } |
521 | |
581 | |
|
|
582 | static void |
522 | static void etp_cancel (ETP_REQ *req) |
583 | etp_cancel (ETP_REQ *req) |
523 | { |
584 | { |
524 | X_LOCK (wrklock); |
585 | req->cancelled = 1; |
525 | req->flags |= EIO_FLAG_CANCELLED; |
|
|
526 | X_UNLOCK (wrklock); |
|
|
527 | |
586 | |
528 | eio_grp_cancel (req); |
587 | eio_grp_cancel (req); |
529 | } |
588 | } |
530 | |
589 | |
|
|
590 | static void |
531 | static void etp_submit (ETP_REQ *req) |
591 | etp_submit (ETP_REQ *req) |
532 | { |
592 | { |
533 | req->pri -= ETP_PRI_MIN; |
593 | req->pri -= ETP_PRI_MIN; |
534 | |
594 | |
535 | if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
595 | if (ecb_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; |
596 | if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
537 | |
597 | |
538 | if (expect_false (req->type == EIO_GROUP)) |
598 | if (ecb_expect_false (req->type == EIO_GROUP)) |
539 | { |
599 | { |
540 | /* I hope this is worth it :/ */ |
600 | /* I hope this is worth it :/ */ |
541 | X_LOCK (reqlock); |
601 | X_LOCK (reqlock); |
542 | ++nreqs; |
602 | ++nreqs; |
543 | X_UNLOCK (reqlock); |
603 | X_UNLOCK (reqlock); |
… | |
… | |
562 | |
622 | |
563 | etp_maybe_start_thread (); |
623 | etp_maybe_start_thread (); |
564 | } |
624 | } |
565 | } |
625 | } |
566 | |
626 | |
|
|
627 | static void ecb_cold |
567 | static void etp_set_max_poll_time (double nseconds) |
628 | etp_set_max_poll_time (double nseconds) |
568 | { |
629 | { |
569 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
630 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
570 | max_poll_time = nseconds; |
631 | max_poll_time = nseconds * EIO_TICKS; |
571 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
632 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
572 | } |
633 | } |
573 | |
634 | |
|
|
635 | static void ecb_cold |
574 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
636 | etp_set_max_poll_reqs (unsigned int maxreqs) |
575 | { |
637 | { |
576 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
638 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
577 | max_poll_reqs = maxreqs; |
639 | max_poll_reqs = maxreqs; |
578 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
640 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
579 | } |
641 | } |
580 | |
642 | |
|
|
643 | static void ecb_cold |
581 | static void etp_set_max_idle (unsigned int nthreads) |
644 | etp_set_max_idle (unsigned int nthreads) |
582 | { |
645 | { |
583 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
646 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
584 | max_idle = nthreads <= 0 ? 1 : nthreads; |
647 | max_idle = nthreads; |
585 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
648 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
586 | } |
649 | } |
587 | |
650 | |
|
|
651 | static void ecb_cold |
|
|
652 | etp_set_idle_timeout (unsigned int seconds) |
|
|
653 | { |
|
|
654 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
655 | idle_timeout = seconds; |
|
|
656 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
657 | } |
|
|
658 | |
|
|
659 | static void ecb_cold |
588 | static void etp_set_min_parallel (unsigned int nthreads) |
660 | etp_set_min_parallel (unsigned int nthreads) |
589 | { |
661 | { |
590 | if (wanted < nthreads) |
662 | if (wanted < nthreads) |
591 | wanted = nthreads; |
663 | wanted = nthreads; |
592 | } |
664 | } |
593 | |
665 | |
|
|
666 | static void ecb_cold |
594 | static void etp_set_max_parallel (unsigned int nthreads) |
667 | etp_set_max_parallel (unsigned int nthreads) |
595 | { |
668 | { |
596 | if (wanted > nthreads) |
669 | if (wanted > nthreads) |
597 | wanted = nthreads; |
670 | wanted = nthreads; |
598 | |
671 | |
599 | while (started > wanted) |
672 | while (started > wanted) |
600 | etp_end_thread (); |
673 | etp_end_thread (); |
601 | } |
674 | } |
602 | |
675 | |
603 | /*****************************************************************************/ |
676 | /*****************************************************************************/ |
604 | |
677 | |
|
|
678 | static void |
605 | static void grp_try_feed (eio_req *grp) |
679 | grp_try_feed (eio_req *grp) |
606 | { |
680 | { |
607 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
681 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
608 | { |
682 | { |
609 | int old_len = grp->size; |
683 | grp->flags &= ~EIO_FLAG_GROUPADD; |
610 | |
684 | |
611 | EIO_FEED (grp); |
685 | EIO_FEED (grp); |
612 | |
686 | |
613 | /* stop if no progress has been made */ |
687 | /* stop if no progress has been made */ |
614 | if (old_len == grp->size) |
688 | if (!(grp->flags & EIO_FLAG_GROUPADD)) |
615 | { |
689 | { |
616 | grp->feed = 0; |
690 | grp->feed = 0; |
617 | break; |
691 | break; |
618 | } |
692 | } |
619 | } |
693 | } |
620 | } |
694 | } |
621 | |
695 | |
|
|
696 | static int |
622 | static int grp_dec (eio_req *grp) |
697 | grp_dec (eio_req *grp) |
623 | { |
698 | { |
624 | --grp->size; |
699 | --grp->size; |
625 | |
700 | |
626 | /* call feeder, if applicable */ |
701 | /* call feeder, if applicable */ |
627 | grp_try_feed (grp); |
702 | grp_try_feed (grp); |
… | |
… | |
631 | return eio_finish (grp); |
706 | return eio_finish (grp); |
632 | else |
707 | else |
633 | return 0; |
708 | return 0; |
634 | } |
709 | } |
635 | |
710 | |
|
|
711 | static void |
636 | void eio_destroy (eio_req *req) |
712 | eio_destroy (eio_req *req) |
637 | { |
713 | { |
638 | if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
714 | if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
639 | if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
715 | if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
640 | |
716 | |
641 | EIO_DESTROY (req); |
717 | EIO_DESTROY (req); |
642 | } |
718 | } |
643 | |
719 | |
|
|
720 | static int |
644 | static int eio_finish (eio_req *req) |
721 | eio_finish (eio_req *req) |
645 | { |
722 | { |
646 | int res = EIO_FINISH (req); |
723 | int res = EIO_FINISH (req); |
647 | |
724 | |
648 | if (req->grp) |
725 | if (req->grp) |
649 | { |
726 | { |
… | |
… | |
657 | if (grp->grp_first == req) |
734 | if (grp->grp_first == req) |
658 | grp->grp_first = req->grp_next; |
735 | grp->grp_first = req->grp_next; |
659 | |
736 | |
660 | res2 = grp_dec (grp); |
737 | res2 = grp_dec (grp); |
661 | |
738 | |
662 | if (!res && res2) |
739 | if (!res) |
663 | res = res2; |
740 | res = res2; |
664 | } |
741 | } |
665 | |
742 | |
666 | eio_destroy (req); |
743 | eio_destroy (req); |
667 | |
744 | |
668 | return res; |
745 | return res; |
669 | } |
746 | } |
670 | |
747 | |
|
|
748 | void |
671 | void eio_grp_cancel (eio_req *grp) |
749 | eio_grp_cancel (eio_req *grp) |
672 | { |
750 | { |
673 | for (grp = grp->grp_first; grp; grp = grp->grp_next) |
751 | for (grp = grp->grp_first; grp; grp = grp->grp_next) |
674 | eio_cancel (grp); |
752 | eio_cancel (grp); |
675 | } |
753 | } |
676 | |
754 | |
|
|
755 | void |
677 | void eio_cancel (eio_req *req) |
756 | eio_cancel (eio_req *req) |
678 | { |
757 | { |
679 | etp_cancel (req); |
758 | etp_cancel (req); |
680 | } |
759 | } |
681 | |
760 | |
|
|
761 | void |
682 | void eio_submit (eio_req *req) |
762 | eio_submit (eio_req *req) |
683 | { |
763 | { |
684 | etp_submit (req); |
764 | etp_submit (req); |
685 | } |
765 | } |
686 | |
766 | |
687 | unsigned int eio_nreqs (void) |
767 | unsigned int |
|
|
768 | eio_nreqs (void) |
688 | { |
769 | { |
689 | return etp_nreqs (); |
770 | return etp_nreqs (); |
690 | } |
771 | } |
691 | |
772 | |
692 | unsigned int eio_nready (void) |
773 | unsigned int |
|
|
774 | eio_nready (void) |
693 | { |
775 | { |
694 | return etp_nready (); |
776 | return etp_nready (); |
695 | } |
777 | } |
696 | |
778 | |
697 | unsigned int eio_npending (void) |
779 | unsigned int |
|
|
780 | eio_npending (void) |
698 | { |
781 | { |
699 | return etp_npending (); |
782 | return etp_npending (); |
700 | } |
783 | } |
701 | |
784 | |
702 | unsigned int eio_nthreads (void) |
785 | unsigned int ecb_cold |
|
|
786 | eio_nthreads (void) |
703 | { |
787 | { |
704 | return etp_nthreads (); |
788 | return etp_nthreads (); |
705 | } |
789 | } |
706 | |
790 | |
|
|
791 | void ecb_cold |
707 | void eio_set_max_poll_time (double nseconds) |
792 | eio_set_max_poll_time (double nseconds) |
708 | { |
793 | { |
709 | etp_set_max_poll_time (nseconds); |
794 | etp_set_max_poll_time (nseconds); |
710 | } |
795 | } |
711 | |
796 | |
|
|
797 | void ecb_cold |
712 | void eio_set_max_poll_reqs (unsigned int maxreqs) |
798 | eio_set_max_poll_reqs (unsigned int maxreqs) |
713 | { |
799 | { |
714 | etp_set_max_poll_reqs (maxreqs); |
800 | etp_set_max_poll_reqs (maxreqs); |
715 | } |
801 | } |
716 | |
802 | |
|
|
803 | void ecb_cold |
717 | void eio_set_max_idle (unsigned int nthreads) |
804 | eio_set_max_idle (unsigned int nthreads) |
718 | { |
805 | { |
719 | etp_set_max_idle (nthreads); |
806 | etp_set_max_idle (nthreads); |
720 | } |
807 | } |
721 | |
808 | |
|
|
809 | void ecb_cold |
|
|
810 | eio_set_idle_timeout (unsigned int seconds) |
|
|
811 | { |
|
|
812 | etp_set_idle_timeout (seconds); |
|
|
813 | } |
|
|
814 | |
|
|
815 | void ecb_cold |
722 | void eio_set_min_parallel (unsigned int nthreads) |
816 | eio_set_min_parallel (unsigned int nthreads) |
723 | { |
817 | { |
724 | etp_set_min_parallel (nthreads); |
818 | etp_set_min_parallel (nthreads); |
725 | } |
819 | } |
726 | |
820 | |
|
|
821 | void ecb_cold |
727 | void eio_set_max_parallel (unsigned int nthreads) |
822 | eio_set_max_parallel (unsigned int nthreads) |
728 | { |
823 | { |
729 | etp_set_max_parallel (nthreads); |
824 | etp_set_max_parallel (nthreads); |
730 | } |
825 | } |
731 | |
826 | |
732 | int eio_poll (void) |
827 | int eio_poll (void) |
… | |
… | |
736 | |
831 | |
737 | /*****************************************************************************/ |
832 | /*****************************************************************************/ |
738 | /* work around various missing functions */ |
833 | /* work around various missing functions */ |
739 | |
834 | |
740 | #if !HAVE_PREADWRITE |
835 | #if !HAVE_PREADWRITE |
|
|
836 | # undef pread |
|
|
837 | # undef pwrite |
741 | # define pread eio__pread |
838 | # define pread eio__pread |
742 | # define pwrite eio__pwrite |
839 | # define pwrite eio__pwrite |
743 | |
840 | |
744 | static ssize_t |
841 | static ssize_t |
745 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
842 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
… | |
… | |
765 | |
862 | |
766 | X_LOCK (preadwritelock); |
863 | X_LOCK (preadwritelock); |
767 | ooffset = lseek (fd, 0, SEEK_CUR); |
864 | ooffset = lseek (fd, 0, SEEK_CUR); |
768 | lseek (fd, offset, SEEK_SET); |
865 | lseek (fd, offset, SEEK_SET); |
769 | res = write (fd, buf, count); |
866 | res = write (fd, buf, count); |
770 | lseek (fd, offset, SEEK_SET); |
867 | lseek (fd, ooffset, SEEK_SET); |
771 | X_UNLOCK (preadwritelock); |
868 | X_UNLOCK (preadwritelock); |
772 | |
869 | |
773 | return res; |
870 | return res; |
774 | } |
871 | } |
775 | #endif |
872 | #endif |
776 | |
873 | |
777 | #ifndef HAVE_FUTIMES |
874 | #ifndef HAVE_UTIMES |
778 | |
875 | |
|
|
876 | # undef utimes |
779 | # define utimes(path,times) eio__utimes (path, times) |
877 | # define utimes(path,times) eio__utimes (path, times) |
780 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
781 | |
878 | |
782 | static int |
879 | static int |
783 | eio__utimes (const char *filename, const struct timeval times[2]) |
880 | eio__utimes (const char *filename, const struct timeval times[2]) |
784 | { |
881 | { |
785 | if (times) |
882 | if (times) |
… | |
… | |
793 | } |
890 | } |
794 | else |
891 | else |
795 | return utime (filename, 0); |
892 | return utime (filename, 0); |
796 | } |
893 | } |
797 | |
894 | |
|
|
895 | #endif |
|
|
896 | |
|
|
897 | #ifndef HAVE_FUTIMES |
|
|
898 | |
|
|
899 | # undef futimes |
|
|
900 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
901 | |
|
|
902 | static int |
798 | static int eio__futimes (int fd, const struct timeval tv[2]) |
903 | eio__futimes (int fd, const struct timeval tv[2]) |
799 | { |
904 | { |
800 | errno = ENOSYS; |
905 | errno = ENOSYS; |
801 | return -1; |
906 | return -1; |
802 | } |
907 | } |
803 | |
908 | |
804 | #endif |
909 | #endif |
805 | |
910 | |
806 | #if !HAVE_FDATASYNC |
911 | #if !HAVE_FDATASYNC |
|
|
912 | # undef fdatasync |
807 | # define fdatasync fsync |
913 | # define fdatasync(fd) fsync (fd) |
808 | #endif |
914 | #endif |
|
|
915 | |
|
|
916 | /* sync_file_range always needs emulation */ |
|
|
917 | static int |
|
|
918 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
|
|
919 | { |
|
|
920 | #if HAVE_SYNC_FILE_RANGE |
|
|
921 | int res; |
|
|
922 | |
|
|
923 | if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE |
|
|
924 | || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE |
|
|
925 | || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER) |
|
|
926 | { |
|
|
927 | flags = 0 |
|
|
928 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0) |
|
|
929 | | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0) |
|
|
930 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
|
|
931 | } |
|
|
932 | |
|
|
933 | res = sync_file_range (fd, offset, nbytes, flags); |
|
|
934 | |
|
|
935 | if (!res || errno != ENOSYS) |
|
|
936 | return res; |
|
|
937 | #endif |
|
|
938 | |
|
|
939 | /* even though we could play tricks with the flags, it's better to always |
|
|
940 | * call fdatasync, as that matches the expectation of its users best */ |
|
|
941 | return fdatasync (fd); |
|
|
942 | } |
|
|
943 | |
|
|
944 | static int |
|
|
945 | eio__fallocate (int fd, int mode, off_t offset, size_t len) |
|
|
946 | { |
|
|
947 | #if HAVE_FALLOCATE |
|
|
948 | return fallocate (fd, mode, offset, len); |
|
|
949 | #else |
|
|
950 | errno = ENOSYS; |
|
|
951 | return -1; |
|
|
952 | #endif |
|
|
953 | } |
809 | |
954 | |
810 | #if !HAVE_READAHEAD |
955 | #if !HAVE_READAHEAD |
|
|
956 | # undef readahead |
811 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
957 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
812 | |
958 | |
813 | static ssize_t |
959 | static ssize_t |
814 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
960 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
815 | { |
961 | { |
… | |
… | |
833 | |
979 | |
834 | /* sendfile always needs emulation */ |
980 | /* sendfile always needs emulation */ |
835 | static ssize_t |
981 | static ssize_t |
836 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
982 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
837 | { |
983 | { |
|
|
984 | ssize_t written = 0; |
838 | ssize_t res; |
985 | ssize_t res; |
839 | |
986 | |
840 | if (!count) |
987 | if (!count) |
841 | return 0; |
988 | return 0; |
842 | |
989 | |
|
|
990 | for (;;) |
|
|
991 | { |
|
|
992 | #ifdef __APPLE__ |
|
|
993 | # undef HAVE_SENDFILE /* broken, as everything on os x */ |
|
|
994 | #endif |
843 | #if HAVE_SENDFILE |
995 | #if HAVE_SENDFILE |
844 | # if __linux |
996 | # if __linux |
|
|
997 | off_t soffset = offset; |
845 | res = sendfile (ofd, ifd, &offset, count); |
998 | res = sendfile (ofd, ifd, &soffset, count); |
846 | |
999 | |
847 | # elif __freebsd |
1000 | # elif __FreeBSD__ |
848 | /* |
1001 | /* |
849 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
1002 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
850 | * wasted on making it similar to other I/O functions. |
1003 | * wasted on making it similar to other I/O functions. |
851 | */ |
1004 | */ |
852 | { |
|
|
853 | off_t sbytes; |
1005 | off_t sbytes; |
854 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
1006 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
855 | |
1007 | |
856 | if (res < 0 && sbytes) |
1008 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
857 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
1009 | /* freebsd' sendfile will return 0 on success */ |
|
|
1010 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
|
|
1011 | /* not on e.g. EIO or EPIPE - sounds broken */ |
|
|
1012 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
858 | res = sbytes; |
1013 | res = sbytes; |
859 | } |
1014 | #endif |
|
|
1015 | |
|
|
1016 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
1017 | if (sbytes) |
|
|
1018 | res = sbytes; |
|
|
1019 | |
|
|
1020 | # elif defined (__APPLE__) |
|
|
1021 | off_t sbytes = count; |
|
|
1022 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
1023 | |
|
|
1024 | /* according to the manpage, sbytes is always valid */ |
|
|
1025 | if (sbytes) |
|
|
1026 | res = sbytes; |
860 | |
1027 | |
861 | # elif __hpux |
1028 | # elif __hpux |
862 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
1029 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
863 | |
1030 | |
864 | # elif __solaris |
1031 | # elif __solaris |
865 | { |
|
|
866 | struct sendfilevec vec; |
1032 | struct sendfilevec vec; |
867 | size_t sbytes; |
1033 | size_t sbytes; |
868 | |
1034 | |
869 | vec.sfv_fd = ifd; |
1035 | vec.sfv_fd = ifd; |
870 | vec.sfv_flag = 0; |
1036 | vec.sfv_flag = 0; |
871 | vec.sfv_off = offset; |
1037 | vec.sfv_off = offset; |
872 | vec.sfv_len = count; |
1038 | vec.sfv_len = count; |
873 | |
1039 | |
874 | res = sendfilev (ofd, &vec, 1, &sbytes); |
1040 | res = sendfilev (ofd, &vec, 1, &sbytes); |
875 | |
1041 | |
876 | if (res < 0 && sbytes) |
1042 | if (res < 0 && sbytes) |
877 | res = sbytes; |
1043 | res = sbytes; |
878 | } |
|
|
879 | |
1044 | |
880 | # endif |
1045 | # endif |
|
|
1046 | |
|
|
1047 | #elif defined (_WIN32) |
|
|
1048 | /* does not work, just for documentation of what would need to be done */ |
|
|
1049 | /* actually, cannot be done like this, as TransmitFile changes the file offset, */ |
|
|
1050 | /* libeio guarantees that the file offset does not change, and windows */ |
|
|
1051 | /* has no way to get an independent handle to the same file description */ |
|
|
1052 | HANDLE h = TO_SOCKET (ifd); |
|
|
1053 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
1054 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
1055 | |
881 | #else |
1056 | #else |
882 | res = -1; |
1057 | res = -1; |
883 | errno = ENOSYS; |
1058 | errno = ENOSYS; |
884 | #endif |
1059 | #endif |
885 | |
1060 | |
|
|
1061 | /* we assume sendfile can copy at least 128mb in one go */ |
|
|
1062 | if (res <= 128 * 1024 * 1024) |
|
|
1063 | { |
|
|
1064 | if (res > 0) |
|
|
1065 | written += res; |
|
|
1066 | |
|
|
1067 | if (written) |
|
|
1068 | return written; |
|
|
1069 | |
|
|
1070 | break; |
|
|
1071 | } |
|
|
1072 | else |
|
|
1073 | { |
|
|
1074 | /* if we requested more, then probably the kernel was lazy */ |
|
|
1075 | written += res; |
|
|
1076 | offset += res; |
|
|
1077 | count -= res; |
|
|
1078 | |
|
|
1079 | if (!count) |
|
|
1080 | return written; |
|
|
1081 | } |
|
|
1082 | } |
|
|
1083 | |
886 | if (res < 0 |
1084 | if (res < 0 |
887 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1085 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
1086 | /* BSDs */ |
|
|
1087 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
1088 | || errno == ENOTSUP |
|
|
1089 | #endif |
|
|
1090 | || errno == EOPNOTSUPP /* BSDs */ |
888 | #if __solaris |
1091 | #if __solaris |
889 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1092 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
890 | #endif |
1093 | #endif |
891 | ) |
1094 | ) |
892 | ) |
1095 | ) |
… | |
… | |
923 | } |
1126 | } |
924 | |
1127 | |
925 | return res; |
1128 | return res; |
926 | } |
1129 | } |
927 | |
1130 | |
|
|
1131 | #ifdef PAGESIZE |
|
|
1132 | # define eio_pagesize() PAGESIZE |
|
|
1133 | #else |
|
|
1134 | static intptr_t |
|
|
1135 | eio_pagesize (void) |
|
|
1136 | { |
|
|
1137 | static intptr_t page; |
|
|
1138 | |
|
|
1139 | if (!page) |
|
|
1140 | page = sysconf (_SC_PAGESIZE); |
|
|
1141 | |
|
|
1142 | return page; |
|
|
1143 | } |
|
|
1144 | #endif |
|
|
1145 | |
|
|
1146 | static void |
|
|
1147 | eio_page_align (void **addr, size_t *length) |
|
|
1148 | { |
|
|
1149 | intptr_t mask = eio_pagesize () - 1; |
|
|
1150 | |
|
|
1151 | /* round down addr */ |
|
|
1152 | intptr_t adj = mask & (intptr_t)*addr; |
|
|
1153 | |
|
|
1154 | *addr = (void *)((intptr_t)*addr - adj); |
|
|
1155 | *length += adj; |
|
|
1156 | |
|
|
1157 | /* round up length */ |
|
|
1158 | *length = (*length + mask) & ~mask; |
|
|
1159 | } |
|
|
1160 | |
|
|
1161 | #if !_POSIX_MEMLOCK |
|
|
1162 | # define eio__mlockall(a) ((errno = ENOSYS), -1) |
|
|
1163 | #else |
|
|
1164 | |
|
|
1165 | static int |
|
|
1166 | eio__mlockall (int flags) |
|
|
1167 | { |
|
|
1168 | #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
|
|
1169 | extern int mallopt (int, int); |
|
|
1170 | mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
|
|
1171 | #endif |
|
|
1172 | |
|
|
1173 | if (EIO_MCL_CURRENT != MCL_CURRENT |
|
|
1174 | || EIO_MCL_FUTURE != MCL_FUTURE) |
|
|
1175 | { |
|
|
1176 | flags = 0 |
|
|
1177 | | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
|
|
1178 | | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
|
|
1179 | } |
|
|
1180 | |
|
|
1181 | return mlockall (flags); |
|
|
1182 | } |
|
|
1183 | #endif |
|
|
1184 | |
|
|
1185 | #if !_POSIX_MEMLOCK_RANGE |
|
|
1186 | # define eio__mlock(a,b) ((errno = ENOSYS), -1) |
|
|
1187 | #else |
|
|
1188 | |
|
|
1189 | static int |
|
|
1190 | eio__mlock (void *addr, size_t length) |
|
|
1191 | { |
|
|
1192 | eio_page_align (&addr, &length); |
|
|
1193 | |
|
|
1194 | return mlock (addr, length); |
|
|
1195 | } |
|
|
1196 | |
|
|
1197 | #endif |
|
|
1198 | |
|
|
1199 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
|
|
1200 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1201 | #else |
|
|
1202 | |
|
|
1203 | static int |
|
|
1204 | eio__msync (void *mem, size_t len, int flags) |
|
|
1205 | { |
|
|
1206 | eio_page_align (&mem, &len); |
|
|
1207 | |
|
|
1208 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1209 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1210 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1211 | { |
|
|
1212 | flags = 0 |
|
|
1213 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1214 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1215 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1216 | } |
|
|
1217 | |
|
|
1218 | return msync (mem, len, flags); |
|
|
1219 | } |
|
|
1220 | |
|
|
1221 | #endif |
|
|
1222 | |
|
|
1223 | static int |
|
|
1224 | eio__mtouch (eio_req *req) |
|
|
1225 | { |
|
|
1226 | void *mem = req->ptr2; |
|
|
1227 | size_t len = req->size; |
|
|
1228 | int flags = req->int1; |
|
|
1229 | |
|
|
1230 | eio_page_align (&mem, &len); |
|
|
1231 | |
|
|
1232 | { |
|
|
1233 | intptr_t addr = (intptr_t)mem; |
|
|
1234 | intptr_t end = addr + len; |
|
|
1235 | intptr_t page = eio_pagesize (); |
|
|
1236 | |
|
|
1237 | if (addr < end) |
|
|
1238 | if (flags & EIO_MT_MODIFY) /* modify */ |
|
|
1239 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
|
|
1240 | else |
|
|
1241 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
|
|
1242 | } |
|
|
1243 | |
|
|
1244 | return 0; |
|
|
1245 | } |
|
|
1246 | |
|
|
1247 | /*****************************************************************************/ |
|
|
1248 | /* requests implemented outside eio_execute, because they are so large */ |
|
|
1249 | |
|
|
1250 | static void |
|
|
1251 | eio__realpath (eio_req *req, etp_worker *self) |
|
|
1252 | { |
|
|
1253 | char *rel = req->ptr1; |
|
|
1254 | char *res; |
|
|
1255 | char *tmp1, *tmp2; |
|
|
1256 | #if SYMLOOP_MAX > 32 |
|
|
1257 | int symlinks = SYMLOOP_MAX; |
|
|
1258 | #else |
|
|
1259 | int symlinks = 32; |
|
|
1260 | #endif |
|
|
1261 | |
|
|
1262 | req->result = -1; |
|
|
1263 | |
|
|
1264 | errno = EINVAL; |
|
|
1265 | if (!rel) |
|
|
1266 | return; |
|
|
1267 | |
|
|
1268 | errno = ENOENT; |
|
|
1269 | if (!*rel) |
|
|
1270 | return; |
|
|
1271 | |
|
|
1272 | if (!req->ptr2) |
|
|
1273 | { |
|
|
1274 | X_LOCK (wrklock); |
|
|
1275 | req->flags |= EIO_FLAG_PTR2_FREE; |
|
|
1276 | X_UNLOCK (wrklock); |
|
|
1277 | req->ptr2 = malloc (PATH_MAX * 3); |
|
|
1278 | |
|
|
1279 | errno = ENOMEM; |
|
|
1280 | if (!req->ptr2) |
|
|
1281 | return; |
|
|
1282 | } |
|
|
1283 | |
|
|
1284 | res = req->ptr2; |
|
|
1285 | tmp1 = res + PATH_MAX; |
|
|
1286 | tmp2 = tmp1 + PATH_MAX; |
|
|
1287 | |
|
|
1288 | #if 0 /* disabled, the musl way to do things is just too racy */ |
|
|
1289 | #if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) |
|
|
1290 | /* on linux we may be able to ask the kernel */ |
|
|
1291 | { |
|
|
1292 | int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME); |
|
|
1293 | |
|
|
1294 | if (fd >= 0) |
|
|
1295 | { |
|
|
1296 | sprintf (tmp1, "/proc/self/fd/%d", fd); |
|
|
1297 | req->result = readlink (tmp1, res, PATH_MAX); |
|
|
1298 | close (fd); |
|
|
1299 | |
|
|
1300 | /* here we should probably stat the open file and the disk file, to make sure they still match */ |
|
|
1301 | |
|
|
1302 | if (req->result > 0) |
|
|
1303 | goto done; |
|
|
1304 | } |
|
|
1305 | else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) |
|
|
1306 | return; |
|
|
1307 | } |
|
|
1308 | #endif |
|
|
1309 | #endif |
|
|
1310 | |
|
|
1311 | if (*rel != '/') |
|
|
1312 | { |
|
|
1313 | if (!getcwd (res, PATH_MAX)) |
|
|
1314 | return; |
|
|
1315 | |
|
|
1316 | if (res [1]) /* only use if not / */ |
|
|
1317 | res += strlen (res); |
|
|
1318 | } |
|
|
1319 | |
|
|
1320 | while (*rel) |
|
|
1321 | { |
|
|
1322 | ssize_t len, linklen; |
|
|
1323 | char *beg = rel; |
|
|
1324 | |
|
|
1325 | while (*rel && *rel != '/') |
|
|
1326 | ++rel; |
|
|
1327 | |
|
|
1328 | len = rel - beg; |
|
|
1329 | |
|
|
1330 | if (!len) /* skip slashes */ |
|
|
1331 | { |
|
|
1332 | ++rel; |
|
|
1333 | continue; |
|
|
1334 | } |
|
|
1335 | |
|
|
1336 | if (beg [0] == '.') |
|
|
1337 | { |
|
|
1338 | if (len == 1) |
|
|
1339 | continue; /* . - nop */ |
|
|
1340 | |
|
|
1341 | if (beg [1] == '.' && len == 2) |
|
|
1342 | { |
|
|
1343 | /* .. - back up one component, if possible */ |
|
|
1344 | |
|
|
1345 | while (res != req->ptr2) |
|
|
1346 | if (*--res == '/') |
|
|
1347 | break; |
|
|
1348 | |
|
|
1349 | continue; |
|
|
1350 | } |
|
|
1351 | } |
|
|
1352 | |
|
|
1353 | errno = ENAMETOOLONG; |
|
|
1354 | if (res + 1 + len + 1 >= tmp1) |
|
|
1355 | return; |
|
|
1356 | |
|
|
1357 | /* copy one component */ |
|
|
1358 | *res = '/'; |
|
|
1359 | memcpy (res + 1, beg, len); |
|
|
1360 | |
|
|
1361 | /* zero-terminate, for readlink */ |
|
|
1362 | res [len + 1] = 0; |
|
|
1363 | |
|
|
1364 | /* now check if it's a symlink */ |
|
|
1365 | linklen = readlink (req->ptr2, tmp1, PATH_MAX); |
|
|
1366 | |
|
|
1367 | if (linklen < 0) |
|
|
1368 | { |
|
|
1369 | if (errno != EINVAL) |
|
|
1370 | return; |
|
|
1371 | |
|
|
1372 | /* it's a normal directory. hopefully */ |
|
|
1373 | res += len + 1; |
|
|
1374 | } |
|
|
1375 | else |
|
|
1376 | { |
|
|
1377 | /* yay, it was a symlink - build new path in tmp2 */ |
|
|
1378 | int rellen = strlen (rel); |
|
|
1379 | |
|
|
1380 | errno = ENAMETOOLONG; |
|
|
1381 | if (linklen + 1 + rellen >= PATH_MAX) |
|
|
1382 | return; |
|
|
1383 | |
|
|
1384 | errno = ELOOP; |
|
|
1385 | if (!--symlinks) |
|
|
1386 | return; |
|
|
1387 | |
|
|
1388 | if (*tmp1 == '/') |
|
|
1389 | res = req->ptr2; /* symlink resolves to an absolute path */ |
|
|
1390 | |
|
|
1391 | /* we need to be careful, as rel might point into tmp2 already */ |
|
|
1392 | memmove (tmp2 + linklen + 1, rel, rellen + 1); |
|
|
1393 | tmp2 [linklen] = '/'; |
|
|
1394 | memcpy (tmp2, tmp1, linklen); |
|
|
1395 | |
|
|
1396 | rel = tmp2; |
|
|
1397 | } |
|
|
1398 | } |
|
|
1399 | |
|
|
1400 | /* special case for the lone root path */ |
|
|
1401 | if (res == req->ptr2) |
|
|
1402 | *res++ = '/'; |
|
|
1403 | |
|
|
1404 | req->result = res - (char *)req->ptr2; |
|
|
1405 | |
|
|
1406 | done: |
|
|
1407 | req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */ |
|
|
1408 | } |
|
|
1409 | |
|
|
1410 | static signed char |
|
|
1411 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
|
|
1412 | { |
|
|
1413 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
|
|
1414 | : a->inode < b->inode ? -1 |
|
|
1415 | : a->inode > b->inode ? 1 |
|
|
1416 | : 0; |
|
|
1417 | } |
|
|
1418 | |
|
|
1419 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1420 | |
|
|
1421 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1422 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
|
|
1423 | |
|
|
1424 | static void |
|
|
1425 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1426 | { |
|
|
1427 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1428 | unsigned char *bit = bits; |
|
|
1429 | |
|
|
1430 | assert (CHAR_BIT == 8); |
|
|
1431 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1432 | assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
|
|
1433 | assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ |
|
|
1434 | |
|
|
1435 | if (size <= EIO_SORT_FAST) |
|
|
1436 | return; |
|
|
1437 | |
|
|
1438 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1439 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1440 | /* inode_bits must contain all inodes ORed together */ |
|
|
1441 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1442 | { |
|
|
1443 | ino_t endianness; |
|
|
1444 | int i, j; |
|
|
1445 | |
|
|
1446 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1447 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1448 | ((unsigned char *)&endianness)[i] = i; |
|
|
1449 | |
|
|
1450 | *bit++ = 0; |
|
|
1451 | |
|
|
1452 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1453 | { |
|
|
1454 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1455 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1456 | endianness >>= 8; |
|
|
1457 | |
|
|
1458 | for (j = 0; j < 8; ++j) |
|
|
1459 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1460 | *bit++ = offs + j; |
|
|
1461 | } |
|
|
1462 | |
|
|
1463 | for (j = 0; j < 8; ++j) |
|
|
1464 | if (score_bits & (1 << j)) |
|
|
1465 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1466 | } |
|
|
1467 | |
|
|
1468 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1469 | { |
|
|
1470 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1471 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1472 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1473 | int stk_idx = 0; |
|
|
1474 | |
|
|
1475 | base_stk [stk_idx] = dents; |
|
|
1476 | end_stk [stk_idx] = dents + size; |
|
|
1477 | bit_stk [stk_idx] = bit - 1; |
|
|
1478 | |
|
|
1479 | do |
|
|
1480 | { |
|
|
1481 | base = base_stk [stk_idx]; |
|
|
1482 | end = end_stk [stk_idx]; |
|
|
1483 | bit = bit_stk [stk_idx]; |
|
|
1484 | |
|
|
1485 | for (;;) |
|
|
1486 | { |
|
|
1487 | unsigned char O = *bit >> 3; |
|
|
1488 | unsigned char M = 1 << (*bit & 7); |
|
|
1489 | |
|
|
1490 | eio_dirent *a = base; |
|
|
1491 | eio_dirent *b = end; |
|
|
1492 | |
|
|
1493 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1494 | break; |
|
|
1495 | |
|
|
1496 | /* now bit-partition the array on the bit */ |
|
|
1497 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1498 | /* partition algos found in the literature */ |
|
|
1499 | do |
|
|
1500 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1501 | ++a; |
|
|
1502 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1503 | { |
|
|
1504 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1505 | ++a; |
|
|
1506 | } |
|
|
1507 | while (b > a); |
|
|
1508 | |
|
|
1509 | /* next bit, or stop, if no bits left in this path */ |
|
|
1510 | if (!*--bit) |
|
|
1511 | break; |
|
|
1512 | |
|
|
1513 | base_stk [stk_idx] = a; |
|
|
1514 | end_stk [stk_idx] = end; |
|
|
1515 | bit_stk [stk_idx] = bit; |
|
|
1516 | ++stk_idx; |
|
|
1517 | |
|
|
1518 | end = a; |
|
|
1519 | } |
|
|
1520 | } |
|
|
1521 | while (stk_idx--); |
|
|
1522 | } |
|
|
1523 | } |
|
|
1524 | |
|
|
1525 | static void |
|
|
1526 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1527 | { |
|
|
1528 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1529 | { |
|
|
1530 | int i; |
|
|
1531 | eio_dirent *min = dents; |
|
|
1532 | |
|
|
1533 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1534 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1535 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1536 | min = &dents [i]; |
|
|
1537 | |
|
|
1538 | /* swap elements 0 and j (minimum) */ |
|
|
1539 | { |
|
|
1540 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1541 | } |
|
|
1542 | } |
|
|
1543 | |
|
|
1544 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1545 | { |
|
|
1546 | eio_dirent *i, *j; |
|
|
1547 | |
|
|
1548 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1549 | { |
|
|
1550 | eio_dirent value = *i; |
|
|
1551 | |
|
|
1552 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1553 | j [1] = j [0]; |
|
|
1554 | |
|
|
1555 | j [1] = value; |
|
|
1556 | } |
|
|
1557 | } |
|
|
1558 | } |
|
|
1559 | |
|
|
1560 | static void |
|
|
1561 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1562 | { |
|
|
1563 | if (size <= 1) |
|
|
1564 | return; /* our insertion sort relies on size > 0 */ |
|
|
1565 | |
|
|
1566 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
|
|
1567 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1568 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
|
|
1569 | |
|
|
1570 | /* use an insertion sort at the end, or for small arrays, */ |
|
|
1571 | /* as insertion sort is more efficient for small partitions */ |
|
|
1572 | eio_dent_insertion_sort (dents, size); |
|
|
1573 | } |
|
|
1574 | |
928 | /* read a full directory */ |
1575 | /* read a full directory */ |
929 | static void |
1576 | static void |
930 | eio__scandir (eio_req *req, etp_worker *self) |
1577 | eio__scandir (eio_req *req, etp_worker *self) |
931 | { |
1578 | { |
932 | DIR *dirp; |
1579 | DIR *dirp; |
933 | EIO_STRUCT_DIRENT *entp; |
1580 | EIO_STRUCT_DIRENT *entp; |
934 | char *name, *names; |
1581 | char *name, *names; |
935 | int memlen = 4096; |
1582 | int namesalloc = 4096; |
936 | int memofs = 0; |
1583 | int namesoffs = 0; |
|
|
1584 | int flags = req->int1; |
|
|
1585 | eio_dirent *dents = 0; |
|
|
1586 | int dentalloc = 128; |
937 | int res = 0; |
1587 | int dentoffs = 0; |
|
|
1588 | ino_t inode_bits = 0; |
|
|
1589 | |
|
|
1590 | req->result = -1; |
|
|
1591 | |
|
|
1592 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1593 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
938 | |
1594 | |
939 | X_LOCK (wrklock); |
1595 | X_LOCK (wrklock); |
940 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1596 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
941 | self->dirp = dirp = opendir (req->ptr1); |
1597 | self->dirp = dirp = opendir (req->ptr1); |
|
|
1598 | |
942 | req->flags |= EIO_FLAG_PTR2_FREE; |
1599 | if (req->flags & EIO_FLAG_PTR1_FREE) |
|
|
1600 | free (req->ptr1); |
|
|
1601 | |
|
|
1602 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
|
|
1603 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
943 | req->ptr2 = names = malloc (memlen); |
1604 | req->ptr2 = names = malloc (namesalloc); |
944 | X_UNLOCK (wrklock); |
1605 | X_UNLOCK (wrklock); |
945 | |
1606 | |
946 | if (dirp && names) |
1607 | if (dirp && names && (!flags || dents)) |
947 | for (;;) |
1608 | for (;;) |
948 | { |
1609 | { |
949 | errno = 0; |
1610 | errno = 0; |
950 | entp = readdir (dirp); |
1611 | entp = readdir (dirp); |
951 | |
1612 | |
952 | if (!entp) |
1613 | if (!entp) |
|
|
1614 | { |
|
|
1615 | if (errno) |
|
|
1616 | break; |
|
|
1617 | |
|
|
1618 | /* sort etc. */ |
|
|
1619 | req->int1 = flags; |
|
|
1620 | req->result = dentoffs; |
|
|
1621 | |
|
|
1622 | if (flags & EIO_READDIR_STAT_ORDER) |
|
|
1623 | eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); |
|
|
1624 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1625 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1626 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1627 | else |
|
|
1628 | { |
|
|
1629 | /* in this case, all is known, and we just put dirs first and sort them */ |
|
|
1630 | eio_dirent *oth = dents + dentoffs; |
|
|
1631 | eio_dirent *dir = dents; |
|
|
1632 | |
|
|
1633 | /* now partition dirs to the front, and non-dirs to the back */ |
|
|
1634 | /* by walking from both sides and swapping if necessary */ |
|
|
1635 | while (oth > dir) |
|
|
1636 | { |
|
|
1637 | if (dir->type == EIO_DT_DIR) |
|
|
1638 | ++dir; |
|
|
1639 | else if ((--oth)->type == EIO_DT_DIR) |
|
|
1640 | { |
|
|
1641 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
|
|
1642 | |
|
|
1643 | ++dir; |
|
|
1644 | } |
|
|
1645 | } |
|
|
1646 | |
|
|
1647 | /* now sort the dirs only (dirs all have the same score) */ |
|
|
1648 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
|
|
1649 | } |
|
|
1650 | |
953 | break; |
1651 | break; |
|
|
1652 | } |
954 | |
1653 | |
|
|
1654 | /* now add the entry to our list(s) */ |
955 | name = entp->d_name; |
1655 | name = entp->d_name; |
956 | |
1656 | |
|
|
1657 | /* skip . and .. entries */ |
957 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1658 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
958 | { |
1659 | { |
959 | int len = strlen (name) + 1; |
1660 | int len = D_NAMLEN (entp) + 1; |
960 | |
1661 | |
961 | res++; |
1662 | while (ecb_expect_false (namesoffs + len > namesalloc)) |
962 | |
|
|
963 | while (memofs + len > memlen) |
|
|
964 | { |
1663 | { |
965 | memlen *= 2; |
1664 | namesalloc *= 2; |
966 | X_LOCK (wrklock); |
1665 | X_LOCK (wrklock); |
967 | req->ptr2 = names = realloc (names, memlen); |
1666 | req->ptr2 = names = realloc (names, namesalloc); |
968 | X_UNLOCK (wrklock); |
1667 | X_UNLOCK (wrklock); |
969 | |
1668 | |
970 | if (!names) |
1669 | if (!names) |
971 | break; |
1670 | break; |
972 | } |
1671 | } |
973 | |
1672 | |
974 | memcpy (names + memofs, name, len); |
1673 | memcpy (names + namesoffs, name, len); |
|
|
1674 | |
|
|
1675 | if (dents) |
|
|
1676 | { |
|
|
1677 | struct eio_dirent *ent; |
|
|
1678 | |
|
|
1679 | if (ecb_expect_false (dentoffs == dentalloc)) |
|
|
1680 | { |
|
|
1681 | dentalloc *= 2; |
|
|
1682 | X_LOCK (wrklock); |
|
|
1683 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
|
|
1684 | X_UNLOCK (wrklock); |
|
|
1685 | |
|
|
1686 | if (!dents) |
|
|
1687 | break; |
|
|
1688 | } |
|
|
1689 | |
|
|
1690 | ent = dents + dentoffs; |
|
|
1691 | |
|
|
1692 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
|
|
1693 | ent->namelen = len - 1; |
|
|
1694 | ent->inode = D_INO (entp); |
|
|
1695 | |
|
|
1696 | inode_bits |= ent->inode; |
|
|
1697 | |
|
|
1698 | switch (D_TYPE (entp)) |
|
|
1699 | { |
|
|
1700 | default: |
|
|
1701 | ent->type = EIO_DT_UNKNOWN; |
|
|
1702 | flags |= EIO_READDIR_FOUND_UNKNOWN; |
|
|
1703 | break; |
|
|
1704 | |
|
|
1705 | #ifdef DT_FIFO |
|
|
1706 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
|
|
1707 | #endif |
|
|
1708 | #ifdef DT_CHR |
|
|
1709 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
|
|
1710 | #endif |
|
|
1711 | #ifdef DT_MPC |
|
|
1712 | case DT_MPC: ent->type = EIO_DT_MPC; break; |
|
|
1713 | #endif |
|
|
1714 | #ifdef DT_DIR |
|
|
1715 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
|
|
1716 | #endif |
|
|
1717 | #ifdef DT_NAM |
|
|
1718 | case DT_NAM: ent->type = EIO_DT_NAM; break; |
|
|
1719 | #endif |
|
|
1720 | #ifdef DT_BLK |
|
|
1721 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
|
|
1722 | #endif |
|
|
1723 | #ifdef DT_MPB |
|
|
1724 | case DT_MPB: ent->type = EIO_DT_MPB; break; |
|
|
1725 | #endif |
|
|
1726 | #ifdef DT_REG |
|
|
1727 | case DT_REG: ent->type = EIO_DT_REG; break; |
|
|
1728 | #endif |
|
|
1729 | #ifdef DT_NWK |
|
|
1730 | case DT_NWK: ent->type = EIO_DT_NWK; break; |
|
|
1731 | #endif |
|
|
1732 | #ifdef DT_CMP |
|
|
1733 | case DT_CMP: ent->type = EIO_DT_CMP; break; |
|
|
1734 | #endif |
|
|
1735 | #ifdef DT_LNK |
|
|
1736 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
|
|
1737 | #endif |
|
|
1738 | #ifdef DT_SOCK |
|
|
1739 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
|
|
1740 | #endif |
|
|
1741 | #ifdef DT_DOOR |
|
|
1742 | case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
|
|
1743 | #endif |
|
|
1744 | #ifdef DT_WHT |
|
|
1745 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
|
|
1746 | #endif |
|
|
1747 | } |
|
|
1748 | |
|
|
1749 | ent->score = 7; |
|
|
1750 | |
|
|
1751 | if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1752 | { |
|
|
1753 | if (ent->type == EIO_DT_UNKNOWN) |
|
|
1754 | { |
|
|
1755 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
|
|
1756 | ent->score = 1; |
|
|
1757 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
|
|
1758 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
|
|
1759 | } |
|
|
1760 | else if (ent->type == EIO_DT_DIR) |
|
|
1761 | ent->score = 0; |
|
|
1762 | } |
|
|
1763 | } |
|
|
1764 | |
975 | memofs += len; |
1765 | namesoffs += len; |
|
|
1766 | ++dentoffs; |
|
|
1767 | } |
|
|
1768 | |
|
|
1769 | if (EIO_CANCELLED (req)) |
|
|
1770 | { |
|
|
1771 | errno = ECANCELED; |
|
|
1772 | break; |
976 | } |
1773 | } |
977 | } |
1774 | } |
978 | |
|
|
979 | if (errno) |
|
|
980 | res = -1; |
|
|
981 | |
|
|
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; |
|
|
1012 | } |
1775 | } |
1013 | |
1776 | |
1014 | /*****************************************************************************/ |
1777 | /*****************************************************************************/ |
1015 | |
1778 | |
1016 | #define ALLOC(len) \ |
1779 | #define ALLOC(len) \ |
… | |
… | |
1031 | X_THREAD_PROC (etp_proc) |
1794 | X_THREAD_PROC (etp_proc) |
1032 | { |
1795 | { |
1033 | ETP_REQ *req; |
1796 | ETP_REQ *req; |
1034 | struct timespec ts; |
1797 | struct timespec ts; |
1035 | etp_worker *self = (etp_worker *)thr_arg; |
1798 | etp_worker *self = (etp_worker *)thr_arg; |
|
|
1799 | int timeout; |
1036 | |
1800 | |
1037 | /* try to distribute timeouts somewhat randomly */ |
1801 | /* try to distribute timeouts somewhat evenly */ |
1038 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1802 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1039 | |
1803 | |
1040 | for (;;) |
1804 | for (;;) |
1041 | { |
1805 | { |
|
|
1806 | ts.tv_sec = 0; |
|
|
1807 | |
1042 | X_LOCK (reqlock); |
1808 | X_LOCK (reqlock); |
1043 | |
1809 | |
1044 | for (;;) |
1810 | for (;;) |
1045 | { |
1811 | { |
1046 | self->req = req = reqq_shift (&req_queue); |
1812 | self->req = req = reqq_shift (&req_queue); |
1047 | |
1813 | |
1048 | if (req) |
1814 | if (req) |
1049 | break; |
1815 | break; |
1050 | |
1816 | |
|
|
1817 | if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ |
|
|
1818 | { |
|
|
1819 | X_UNLOCK (reqlock); |
|
|
1820 | X_LOCK (wrklock); |
|
|
1821 | --started; |
|
|
1822 | X_UNLOCK (wrklock); |
|
|
1823 | goto quit; |
|
|
1824 | } |
|
|
1825 | |
1051 | ++idle; |
1826 | ++idle; |
1052 | |
1827 | |
1053 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
1828 | if (idle <= max_idle) |
1054 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1829 | /* we are allowed to idle, so do so without any timeout */ |
|
|
1830 | X_COND_WAIT (reqwait, reqlock); |
|
|
1831 | else |
1055 | { |
1832 | { |
1056 | if (idle > max_idle) |
1833 | /* initialise timeout once */ |
1057 | { |
1834 | if (!ts.tv_sec) |
1058 | --idle; |
1835 | ts.tv_sec = time (0) + idle_timeout; |
1059 | X_UNLOCK (reqlock); |
|
|
1060 | X_LOCK (wrklock); |
|
|
1061 | --started; |
|
|
1062 | X_UNLOCK (wrklock); |
|
|
1063 | goto quit; |
|
|
1064 | } |
|
|
1065 | |
1836 | |
1066 | /* we are allowed to idle, so do so without any timeout */ |
|
|
1067 | X_COND_WAIT (reqwait, reqlock); |
1837 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
|
|
1838 | ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */ |
1068 | } |
1839 | } |
1069 | |
1840 | |
1070 | --idle; |
1841 | --idle; |
1071 | } |
1842 | } |
1072 | |
1843 | |
… | |
… | |
1075 | X_UNLOCK (reqlock); |
1846 | X_UNLOCK (reqlock); |
1076 | |
1847 | |
1077 | if (req->type < 0) |
1848 | if (req->type < 0) |
1078 | goto quit; |
1849 | goto quit; |
1079 | |
1850 | |
1080 | if (!EIO_CANCELLED (req)) |
|
|
1081 | ETP_EXECUTE (self, req); |
1851 | ETP_EXECUTE (self, req); |
1082 | |
1852 | |
1083 | X_LOCK (reslock); |
1853 | X_LOCK (reslock); |
1084 | |
1854 | |
1085 | ++npending; |
1855 | ++npending; |
1086 | |
1856 | |
… | |
… | |
1101 | return 0; |
1871 | return 0; |
1102 | } |
1872 | } |
1103 | |
1873 | |
1104 | /*****************************************************************************/ |
1874 | /*****************************************************************************/ |
1105 | |
1875 | |
|
|
1876 | int ecb_cold |
1106 | int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1877 | eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1107 | { |
1878 | { |
1108 | return etp_init (want_poll, done_poll); |
1879 | return etp_init (want_poll, done_poll); |
1109 | } |
1880 | } |
1110 | |
1881 | |
|
|
1882 | ecb_inline void |
1111 | static void eio_api_destroy (eio_req *req) |
1883 | eio_api_destroy (eio_req *req) |
1112 | { |
1884 | { |
1113 | free (req); |
1885 | free (req); |
1114 | } |
1886 | } |
1115 | |
1887 | |
1116 | #define REQ(rtype) \ |
1888 | #define REQ(rtype) \ |
… | |
… | |
1135 | { \ |
1907 | { \ |
1136 | eio_api_destroy (req); \ |
1908 | eio_api_destroy (req); \ |
1137 | return 0; \ |
1909 | return 0; \ |
1138 | } |
1910 | } |
1139 | |
1911 | |
|
|
1912 | static void |
1140 | static void eio_execute (etp_worker *self, eio_req *req) |
1913 | eio_execute (etp_worker *self, eio_req *req) |
1141 | { |
1914 | { |
1142 | errno = 0; |
1915 | if (ecb_expect_false (EIO_CANCELLED (req))) |
|
|
1916 | { |
|
|
1917 | req->result = -1; |
|
|
1918 | req->errorno = ECANCELED; |
|
|
1919 | return; |
|
|
1920 | } |
1143 | |
1921 | |
1144 | switch (req->type) |
1922 | switch (req->type) |
1145 | { |
1923 | { |
1146 | case EIO_READ: ALLOC (req->size); |
1924 | case EIO_READ: ALLOC (req->size); |
1147 | req->result = req->offs >= 0 |
1925 | req->result = req->offs >= 0 |
… | |
… | |
1159 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1937 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1160 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1938 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1161 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1939 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1162 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1940 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1163 | |
1941 | |
|
|
1942 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1943 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1944 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1945 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1946 | |
1164 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1947 | 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; |
1948 | 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; |
1949 | 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; |
1950 | 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; |
1951 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1175 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1958 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1176 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1959 | 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; |
1960 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1178 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1961 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1179 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1962 | 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; |
1963 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
1181 | |
1964 | |
|
|
1965 | case EIO_REALPATH: eio__realpath (req, self); break; |
|
|
1966 | |
1182 | case EIO_READLINK: ALLOC (NAME_MAX); |
1967 | case EIO_READLINK: ALLOC (PATH_MAX); |
1183 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
1968 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1184 | |
1969 | |
1185 | case EIO_SYNC: req->result = 0; sync (); break; |
1970 | case EIO_SYNC: req->result = 0; sync (); break; |
1186 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1971 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1187 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1972 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1188 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
1973 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
|
|
1974 | case EIO_MTOUCH: req->result = eio__mtouch (req); break; |
1189 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
1975 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
|
|
1976 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
|
|
1977 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
|
|
1978 | case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; |
1190 | |
1979 | |
1191 | case EIO_READDIR: eio__scandir (req, self); break; |
1980 | case EIO_READDIR: eio__scandir (req, self); break; |
1192 | |
1981 | |
1193 | case EIO_BUSY: |
1982 | case EIO_BUSY: |
1194 | #ifdef _WIN32 |
1983 | #ifdef _WIN32 |
1195 | Sleep (req->nv1 * 1000.); |
1984 | Sleep (req->nv1 * 1e3); |
1196 | #else |
1985 | #else |
1197 | { |
1986 | { |
1198 | struct timeval tv; |
1987 | struct timeval tv; |
1199 | |
1988 | |
1200 | tv.tv_sec = req->nv1; |
1989 | tv.tv_sec = req->nv1; |
1201 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
1990 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1202 | |
1991 | |
1203 | req->result = select (0, 0, 0, 0, &tv); |
1992 | req->result = select (0, 0, 0, 0, &tv); |
1204 | } |
1993 | } |
1205 | #endif |
1994 | #endif |
1206 | break; |
1995 | break; |
… | |
… | |
1221 | times = tv; |
2010 | times = tv; |
1222 | } |
2011 | } |
1223 | else |
2012 | else |
1224 | times = 0; |
2013 | times = 0; |
1225 | |
2014 | |
1226 | |
|
|
1227 | req->result = req->type == EIO_FUTIME |
2015 | req->result = req->type == EIO_FUTIME |
1228 | ? futimes (req->int1, times) |
2016 | ? futimes (req->int1, times) |
1229 | : utimes (req->ptr1, times); |
2017 | : utimes (req->ptr1, times); |
1230 | } |
2018 | } |
1231 | break; |
2019 | break; |
… | |
… | |
1236 | case EIO_NOP: |
2024 | case EIO_NOP: |
1237 | req->result = 0; |
2025 | req->result = 0; |
1238 | break; |
2026 | break; |
1239 | |
2027 | |
1240 | case EIO_CUSTOM: |
2028 | case EIO_CUSTOM: |
1241 | ((void (*)(eio_req *))req->feed) (req); |
2029 | req->feed (req); |
1242 | break; |
2030 | break; |
1243 | |
2031 | |
1244 | default: |
2032 | default: |
|
|
2033 | errno = ENOSYS; |
1245 | req->result = -1; |
2034 | req->result = -1; |
1246 | break; |
2035 | break; |
1247 | } |
2036 | } |
1248 | |
2037 | |
1249 | req->errorno = errno; |
2038 | req->errorno = errno; |
… | |
… | |
1279 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
2068 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1280 | { |
2069 | { |
1281 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
2070 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1282 | } |
2071 | } |
1283 | |
2072 | |
|
|
2073 | eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
|
|
2074 | { |
|
|
2075 | REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
|
|
2076 | } |
|
|
2077 | |
|
|
2078 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
|
|
2079 | { |
|
|
2080 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
|
|
2081 | } |
|
|
2082 | |
|
|
2083 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
|
|
2084 | { |
|
|
2085 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
|
|
2086 | } |
|
|
2087 | |
|
|
2088 | eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) |
|
|
2089 | { |
|
|
2090 | REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; |
|
|
2091 | } |
|
|
2092 | |
1284 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
2093 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1285 | { |
2094 | { |
1286 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
2095 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1287 | } |
2096 | } |
1288 | |
2097 | |
… | |
… | |
1307 | } |
2116 | } |
1308 | |
2117 | |
1309 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
2118 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1310 | { |
2119 | { |
1311 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
2120 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
|
|
2121 | } |
|
|
2122 | |
|
|
2123 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
2124 | { |
|
|
2125 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
1312 | } |
2126 | } |
1313 | |
2127 | |
1314 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
2128 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1315 | { |
2129 | { |
1316 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
2130 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
… | |
… | |
1380 | eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
2194 | eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
1381 | { |
2195 | { |
1382 | return eio__1path (EIO_READLINK, path, pri, cb, data); |
2196 | return eio__1path (EIO_READLINK, path, pri, cb, data); |
1383 | } |
2197 | } |
1384 | |
2198 | |
|
|
2199 | eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data) |
|
|
2200 | { |
|
|
2201 | return eio__1path (EIO_REALPATH, path, pri, cb, data); |
|
|
2202 | } |
|
|
2203 | |
1385 | eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
2204 | eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
1386 | { |
2205 | { |
1387 | return eio__1path (EIO_STAT, path, pri, cb, data); |
2206 | return eio__1path (EIO_STAT, path, pri, cb, data); |
1388 | } |
2207 | } |
1389 | |
2208 | |
1390 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
2209 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1391 | { |
2210 | { |
1392 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
2211 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1393 | } |
2212 | } |
1394 | |
2213 | |
|
|
2214 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
2215 | { |
|
|
2216 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
|
|
2217 | } |
|
|
2218 | |
1395 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
2219 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1396 | { |
2220 | { |
1397 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
2221 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1398 | } |
2222 | } |
1399 | |
2223 | |
1400 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
2224 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1401 | { |
2225 | { |
1402 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
2226 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1403 | } |
2227 | } |
1404 | |
2228 | |
1405 | eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) |
2229 | eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
1406 | { |
2230 | { |
1407 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
2231 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1408 | } |
2232 | } |
1409 | |
2233 | |
1410 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
2234 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1411 | { |
2235 | { |
1412 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
2236 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; |
1413 | } |
2237 | } |
1414 | |
2238 | |
1415 | static eio_req * |
2239 | static eio_req * |
1416 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2240 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1417 | { |
2241 | { |
… | |
… | |
1441 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2265 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1442 | { |
2266 | { |
1443 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
2267 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1444 | } |
2268 | } |
1445 | |
2269 | |
1446 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
2270 | eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) |
1447 | { |
2271 | { |
1448 | REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
2272 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
1449 | } |
2273 | } |
1450 | |
2274 | |
1451 | #endif |
2275 | #endif |
1452 | |
2276 | |
1453 | eio_req *eio_grp (eio_cb cb, void *data) |
2277 | eio_req *eio_grp (eio_cb cb, void *data) |
… | |
… | |
1462 | #undef SEND |
2286 | #undef SEND |
1463 | |
2287 | |
1464 | /*****************************************************************************/ |
2288 | /*****************************************************************************/ |
1465 | /* grp functions */ |
2289 | /* grp functions */ |
1466 | |
2290 | |
|
|
2291 | void |
1467 | void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
2292 | eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
1468 | { |
2293 | { |
1469 | grp->int2 = limit; |
2294 | grp->int2 = limit; |
1470 | grp->feed = feed; |
2295 | grp->feed = feed; |
1471 | |
2296 | |
1472 | grp_try_feed (grp); |
2297 | grp_try_feed (grp); |
1473 | } |
2298 | } |
1474 | |
2299 | |
|
|
2300 | void |
1475 | void eio_grp_limit (eio_req *grp, int limit) |
2301 | eio_grp_limit (eio_req *grp, int limit) |
1476 | { |
2302 | { |
1477 | grp->int2 = limit; |
2303 | grp->int2 = limit; |
1478 | |
2304 | |
1479 | grp_try_feed (grp); |
2305 | grp_try_feed (grp); |
1480 | } |
2306 | } |
1481 | |
2307 | |
|
|
2308 | void |
1482 | void eio_grp_add (eio_req *grp, eio_req *req) |
2309 | eio_grp_add (eio_req *grp, eio_req *req) |
1483 | { |
2310 | { |
1484 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
2311 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
|
|
2312 | |
|
|
2313 | grp->flags |= EIO_FLAG_GROUPADD; |
1485 | |
2314 | |
1486 | ++grp->size; |
2315 | ++grp->size; |
1487 | req->grp = grp; |
2316 | req->grp = grp; |
1488 | |
2317 | |
1489 | req->grp_prev = 0; |
2318 | req->grp_prev = 0; |
… | |
… | |
1496 | } |
2325 | } |
1497 | |
2326 | |
1498 | /*****************************************************************************/ |
2327 | /*****************************************************************************/ |
1499 | /* misc garbage */ |
2328 | /* misc garbage */ |
1500 | |
2329 | |
|
|
2330 | ssize_t |
1501 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2331 | eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1502 | { |
2332 | { |
1503 | etp_worker wrk; |
2333 | etp_worker wrk; |
|
|
2334 | ssize_t ret; |
1504 | |
2335 | |
1505 | wrk.dbuf = 0; |
2336 | wrk.dbuf = 0; |
1506 | |
2337 | |
1507 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
2338 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1508 | |
2339 | |
1509 | if (wrk.dbuf) |
2340 | if (wrk.dbuf) |
1510 | free (wrk.dbuf); |
2341 | free (wrk.dbuf); |
1511 | } |
|
|
1512 | |
2342 | |
|
|
2343 | return ret; |
|
|
2344 | } |
|
|
2345 | |