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> |
… | |
… | |
49 | #include <sys/stat.h> |
58 | #include <sys/stat.h> |
50 | #include <limits.h> |
59 | #include <limits.h> |
51 | #include <fcntl.h> |
60 | #include <fcntl.h> |
52 | #include <assert.h> |
61 | #include <assert.h> |
53 | |
62 | |
|
|
63 | #include <sys/statvfs.h> |
|
|
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 |
… | |
… | |
61 | |
83 | |
62 | #ifndef EIO_FEED |
84 | #ifndef EIO_FEED |
63 | # define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
85 | # define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
64 | #endif |
86 | #endif |
65 | |
87 | |
|
|
88 | #ifndef EIO_FD_TO_WIN32_HANDLE |
|
|
89 | # define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd) |
|
|
90 | #endif |
|
|
91 | #ifndef EIO_WIN32_HANDLE_TO_FD |
|
|
92 | # define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0) |
|
|
93 | #endif |
|
|
94 | |
|
|
95 | #define EIO_ERRNO(errval,retval) ((errno = errval), retval) |
|
|
96 | |
|
|
97 | #define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1) |
|
|
98 | |
66 | #ifdef _WIN32 |
99 | #ifdef _WIN32 |
67 | |
100 | |
68 | /*doh*/ |
101 | #define PAGESIZE 4096 /* GetSystemInfo? */ |
|
|
102 | |
|
|
103 | #define stat(path,buf) _stati64 (path,buf) |
|
|
104 | #define lstat(path,buf) stat (path,buf) |
|
|
105 | #define fstat(fd,buf) _fstati64 (path,buf) |
|
|
106 | #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) |
|
|
107 | #define mkdir(path,mode) _mkdir (path) |
|
|
108 | #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) |
|
|
109 | |
|
|
110 | #define chown(path,uid,gid) EIO_ENOSYS () |
|
|
111 | #define fchown(fd,uid,gid) EIO_ENOSYS () |
|
|
112 | #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ |
|
|
113 | #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */ |
|
|
114 | #define mknod(path,mode,dev) EIO_ENOSYS () |
|
|
115 | #define sync() EIO_ENOSYS () |
|
|
116 | |
|
|
117 | /* we could even stat and see if it exists */ |
|
|
118 | static int |
|
|
119 | symlink (const char *old, const char *neu) |
|
|
120 | { |
|
|
121 | if (CreateSymbolicLink (neu, old, 1)) |
|
|
122 | return 0; |
|
|
123 | |
|
|
124 | if (CreateSymbolicLink (neu, old, 0)) |
|
|
125 | return 0; |
|
|
126 | |
|
|
127 | return EIO_ERRNO (ENOENT, -1); |
|
|
128 | } |
69 | |
129 | |
70 | #else |
130 | #else |
71 | |
131 | |
72 | # include "config.h" |
|
|
73 | # include <sys/time.h> |
132 | #include <sys/time.h> |
74 | # include <sys/select.h> |
133 | #include <sys/select.h> |
|
|
134 | #include <sys/statvfs.h> |
75 | # include <unistd.h> |
135 | #include <unistd.h> |
76 | # include <utime.h> |
136 | #include <utime.h> |
77 | # include <signal.h> |
137 | #include <signal.h> |
78 | # include <dirent.h> |
138 | #include <dirent.h> |
79 | |
139 | |
|
|
140 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
141 | #include <sys/mman.h> |
|
|
142 | #endif |
|
|
143 | |
|
|
144 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
|
|
145 | #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
|
|
146 | #define _DIRENT_HAVE_D_TYPE /* sigh */ |
|
|
147 | #define D_INO(de) (de)->d_fileno |
|
|
148 | #define D_NAMLEN(de) (de)->d_namlen |
|
|
149 | #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
|
|
150 | #define D_INO(de) (de)->d_ino |
|
|
151 | #endif |
|
|
152 | |
|
|
153 | #ifdef _D_EXACT_NAMLEN |
|
|
154 | #undef D_NAMLEN |
|
|
155 | #define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
|
|
156 | #endif |
|
|
157 | |
|
|
158 | #ifdef _DIRENT_HAVE_D_TYPE |
|
|
159 | #define D_TYPE(de) (de)->d_type |
|
|
160 | #endif |
|
|
161 | |
80 | # ifndef EIO_STRUCT_DIRENT |
162 | #ifndef EIO_STRUCT_DIRENT |
81 | # define EIO_STRUCT_DIRENT struct dirent |
163 | #define EIO_STRUCT_DIRENT struct dirent |
82 | # endif |
164 | #endif |
83 | |
165 | |
84 | #endif |
166 | #endif |
85 | |
167 | |
86 | #if HAVE_SENDFILE |
168 | #if HAVE_SENDFILE |
87 | # if __linux |
169 | # if __linux |
88 | # include <sys/sendfile.h> |
170 | # include <sys/sendfile.h> |
89 | # elif __freebsd |
171 | # elif __FreeBSD__ || defined __APPLE__ |
90 | # include <sys/socket.h> |
172 | # include <sys/socket.h> |
91 | # include <sys/uio.h> |
173 | # include <sys/uio.h> |
92 | # elif __hpux |
174 | # elif __hpux |
93 | # include <sys/socket.h> |
175 | # include <sys/socket.h> |
94 | # elif __solaris /* not yet */ |
176 | # elif __solaris |
95 | # include <sys/sendfile.h> |
177 | # include <sys/sendfile.h> |
96 | # else |
178 | # else |
97 | # error sendfile support requested but not available |
179 | # error sendfile support requested but not available |
98 | # endif |
180 | # endif |
99 | #endif |
181 | #endif |
100 | |
182 | |
101 | /* number of seconds after which an idle threads exit */ |
183 | #ifndef D_TYPE |
102 | #define IDLE_TIMEOUT 10 |
184 | # define D_TYPE(de) 0 |
|
|
185 | #endif |
|
|
186 | #ifndef D_INO |
|
|
187 | # define D_INO(de) 0 |
|
|
188 | #endif |
|
|
189 | #ifndef D_NAMLEN |
|
|
190 | # define D_NAMLEN(de) strlen ((de)->d_name) |
|
|
191 | #endif |
103 | |
192 | |
104 | /* used for struct dirent, AIX doesn't provide it */ |
193 | /* used for struct dirent, AIX doesn't provide it */ |
105 | #ifndef NAME_MAX |
194 | #ifndef NAME_MAX |
106 | # define NAME_MAX 4096 |
195 | # define NAME_MAX 4096 |
|
|
196 | #endif |
|
|
197 | |
|
|
198 | /* used for readlink etc. */ |
|
|
199 | #ifndef PATH_MAX |
|
|
200 | # define PATH_MAX 4096 |
107 | #endif |
201 | #endif |
108 | |
202 | |
109 | /* buffer size for various temporary buffers */ |
203 | /* buffer size for various temporary buffers */ |
110 | #define EIO_BUFSIZE 65536 |
204 | #define EIO_BUFSIZE 65536 |
111 | |
205 | |
… | |
… | |
117 | errno = ENOMEM; \ |
211 | errno = ENOMEM; \ |
118 | if (!eio_buf) \ |
212 | if (!eio_buf) \ |
119 | return -1; |
213 | return -1; |
120 | |
214 | |
121 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
215 | #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 | |
216 | |
136 | #define ETP_PRI_MIN EIO_PRI_MIN |
217 | #define ETP_PRI_MIN EIO_PRI_MIN |
137 | #define ETP_PRI_MAX EIO_PRI_MAX |
218 | #define ETP_PRI_MAX EIO_PRI_MAX |
138 | |
219 | |
139 | struct etp_worker; |
220 | struct etp_worker; |
… | |
… | |
155 | if (wrk->dirp) \ |
236 | if (wrk->dirp) \ |
156 | { \ |
237 | { \ |
157 | closedir (wrk->dirp); \ |
238 | closedir (wrk->dirp); \ |
158 | wrk->dirp = 0; \ |
239 | wrk->dirp = 0; \ |
159 | } |
240 | } |
|
|
241 | |
160 | #define ETP_WORKER_COMMON \ |
242 | #define ETP_WORKER_COMMON \ |
161 | void *dbuf; \ |
243 | void *dbuf; \ |
162 | DIR *dirp; |
244 | DIR *dirp; |
163 | |
245 | |
164 | /*****************************************************************************/ |
246 | /*****************************************************************************/ |
165 | |
247 | |
166 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
248 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
167 | |
249 | |
168 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
250 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
|
|
251 | ecb_inline int |
169 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
252 | tvdiff (struct timeval *tv1, struct timeval *tv2) |
170 | { |
253 | { |
171 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
254 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
172 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
255 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
173 | } |
256 | } |
174 | |
257 | |
… | |
… | |
181 | static unsigned int max_poll_reqs; /* reslock */ |
264 | static unsigned int max_poll_reqs; /* reslock */ |
182 | |
265 | |
183 | static volatile unsigned int nreqs; /* reqlock */ |
266 | static volatile unsigned int nreqs; /* reqlock */ |
184 | static volatile unsigned int nready; /* reqlock */ |
267 | static volatile unsigned int nready; /* reqlock */ |
185 | static volatile unsigned int npending; /* reqlock */ |
268 | static volatile unsigned int npending; /* reqlock */ |
186 | static volatile unsigned int max_idle = 4; |
269 | static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
|
|
270 | static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
187 | |
271 | |
188 | static mutex_t wrklock = X_MUTEX_INIT; |
272 | static xmutex_t wrklock; |
189 | static mutex_t reslock = X_MUTEX_INIT; |
273 | static xmutex_t reslock; |
190 | static mutex_t reqlock = X_MUTEX_INIT; |
274 | static xmutex_t reqlock; |
191 | static cond_t reqwait = X_COND_INIT; |
275 | static xcond_t reqwait; |
192 | |
276 | |
193 | #if !HAVE_PREADWRITE |
277 | #if !HAVE_PREADWRITE |
194 | /* |
278 | /* |
195 | * make our pread/pwrite emulation safe against themselves, but not against |
279 | * make our pread/pwrite emulation safe against themselves, but not against |
196 | * normal read/write by using a mutex. slows down execution a lot, |
280 | * normal read/write by using a mutex. slows down execution a lot, |
197 | * but that's your problem, not mine. |
281 | * but that's your problem, not mine. |
198 | */ |
282 | */ |
199 | static mutex_t preadwritelock = X_MUTEX_INIT; |
283 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
200 | #endif |
284 | #endif |
201 | |
285 | |
202 | typedef struct etp_worker |
286 | typedef struct etp_worker |
203 | { |
287 | { |
204 | /* locked by wrklock */ |
288 | /* locked by wrklock */ |
205 | struct etp_worker *prev, *next; |
289 | struct etp_worker *prev, *next; |
206 | |
290 | |
207 | thread_t tid; |
291 | xthread_t tid; |
208 | |
292 | |
209 | /* locked by reslock, reqlock or wrklock */ |
293 | /* locked by reslock, reqlock or wrklock */ |
210 | ETP_REQ *req; /* currently processed request */ |
294 | ETP_REQ *req; /* currently processed request */ |
211 | |
295 | |
212 | ETP_WORKER_COMMON |
296 | ETP_WORKER_COMMON |
… | |
… | |
217 | #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
301 | #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
218 | #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
302 | #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
219 | |
303 | |
220 | /* worker threads management */ |
304 | /* worker threads management */ |
221 | |
305 | |
|
|
306 | static void ecb_cold |
222 | static void etp_worker_clear (etp_worker *wrk) |
307 | etp_worker_clear (etp_worker *wrk) |
223 | { |
308 | { |
224 | ETP_WORKER_CLEAR (wrk); |
309 | ETP_WORKER_CLEAR (wrk); |
225 | } |
310 | } |
226 | |
311 | |
|
|
312 | static void ecb_cold |
227 | static void etp_worker_free (etp_worker *wrk) |
313 | etp_worker_free (etp_worker *wrk) |
228 | { |
314 | { |
229 | wrk->next->prev = wrk->prev; |
315 | wrk->next->prev = wrk->prev; |
230 | wrk->prev->next = wrk->next; |
316 | wrk->prev->next = wrk->next; |
231 | |
317 | |
232 | free (wrk); |
318 | free (wrk); |
233 | } |
319 | } |
234 | |
320 | |
235 | static unsigned int etp_nreqs (void) |
321 | static unsigned int |
|
|
322 | etp_nreqs (void) |
236 | { |
323 | { |
237 | int retval; |
324 | int retval; |
238 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
325 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
239 | retval = nreqs; |
326 | retval = nreqs; |
240 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
327 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
241 | return retval; |
328 | return retval; |
242 | } |
329 | } |
243 | |
330 | |
244 | static unsigned int etp_nready (void) |
331 | static unsigned int |
|
|
332 | etp_nready (void) |
245 | { |
333 | { |
246 | unsigned int retval; |
334 | unsigned int retval; |
247 | |
335 | |
248 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
336 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
249 | retval = nready; |
337 | retval = nready; |
250 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
338 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
251 | |
339 | |
252 | return retval; |
340 | return retval; |
253 | } |
341 | } |
254 | |
342 | |
255 | static unsigned int etp_npending (void) |
343 | static unsigned int |
|
|
344 | etp_npending (void) |
256 | { |
345 | { |
257 | unsigned int retval; |
346 | unsigned int retval; |
258 | |
347 | |
259 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
348 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
260 | retval = npending; |
349 | retval = npending; |
261 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
350 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
262 | |
351 | |
263 | return retval; |
352 | return retval; |
264 | } |
353 | } |
265 | |
354 | |
266 | static unsigned int etp_nthreads (void) |
355 | static unsigned int |
|
|
356 | etp_nthreads (void) |
267 | { |
357 | { |
268 | unsigned int retval; |
358 | unsigned int retval; |
269 | |
359 | |
270 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
360 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
271 | retval = started; |
361 | retval = started; |
… | |
… | |
285 | } etp_reqq; |
375 | } etp_reqq; |
286 | |
376 | |
287 | static etp_reqq req_queue; |
377 | static etp_reqq req_queue; |
288 | static etp_reqq res_queue; |
378 | static etp_reqq res_queue; |
289 | |
379 | |
|
|
380 | static int ecb_noinline |
290 | static int reqq_push (etp_reqq *q, ETP_REQ *req) |
381 | reqq_push (etp_reqq *q, ETP_REQ *req) |
291 | { |
382 | { |
292 | int pri = req->pri; |
383 | int pri = req->pri; |
293 | req->next = 0; |
384 | req->next = 0; |
294 | |
385 | |
295 | if (q->qe[pri]) |
386 | if (q->qe[pri]) |
… | |
… | |
301 | q->qe[pri] = q->qs[pri] = req; |
392 | q->qe[pri] = q->qs[pri] = req; |
302 | |
393 | |
303 | return q->size++; |
394 | return q->size++; |
304 | } |
395 | } |
305 | |
396 | |
|
|
397 | static ETP_REQ * ecb_noinline |
306 | static ETP_REQ *reqq_shift (etp_reqq *q) |
398 | reqq_shift (etp_reqq *q) |
307 | { |
399 | { |
308 | int pri; |
400 | int pri; |
309 | |
401 | |
310 | if (!q->size) |
402 | if (!q->size) |
311 | return 0; |
403 | return 0; |
… | |
… | |
326 | } |
418 | } |
327 | |
419 | |
328 | abort (); |
420 | abort (); |
329 | } |
421 | } |
330 | |
422 | |
331 | static void etp_atfork_prepare (void) |
423 | static void ecb_cold |
|
|
424 | etp_thread_init (void) |
332 | { |
425 | { |
333 | X_LOCK (wrklock); |
|
|
334 | X_LOCK (reqlock); |
|
|
335 | X_LOCK (reslock); |
|
|
336 | #if !HAVE_PREADWRITE |
426 | #if !HAVE_PREADWRITE |
337 | X_LOCK (preadwritelock); |
427 | X_MUTEX_CREATE (preadwritelock); |
338 | #endif |
428 | #endif |
|
|
429 | X_MUTEX_CREATE (wrklock); |
|
|
430 | X_MUTEX_CREATE (reslock); |
|
|
431 | X_MUTEX_CREATE (reqlock); |
|
|
432 | X_COND_CREATE (reqwait); |
339 | } |
433 | } |
340 | |
434 | |
|
|
435 | static void ecb_cold |
|
|
436 | etp_atfork_prepare (void) |
|
|
437 | { |
|
|
438 | } |
|
|
439 | |
|
|
440 | static void ecb_cold |
341 | static void etp_atfork_parent (void) |
441 | etp_atfork_parent (void) |
342 | { |
442 | { |
343 | #if !HAVE_PREADWRITE |
|
|
344 | X_UNLOCK (preadwritelock); |
|
|
345 | #endif |
|
|
346 | X_UNLOCK (reslock); |
|
|
347 | X_UNLOCK (reqlock); |
|
|
348 | X_UNLOCK (wrklock); |
|
|
349 | } |
443 | } |
350 | |
444 | |
|
|
445 | static void ecb_cold |
351 | static void etp_atfork_child (void) |
446 | etp_atfork_child (void) |
352 | { |
447 | { |
353 | ETP_REQ *prv; |
448 | ETP_REQ *prv; |
354 | |
449 | |
355 | while ((prv = reqq_shift (&req_queue))) |
450 | while ((prv = reqq_shift (&req_queue))) |
356 | ETP_DESTROY (prv); |
451 | ETP_DESTROY (prv); |
… | |
… | |
373 | idle = 0; |
468 | idle = 0; |
374 | nreqs = 0; |
469 | nreqs = 0; |
375 | nready = 0; |
470 | nready = 0; |
376 | npending = 0; |
471 | npending = 0; |
377 | |
472 | |
378 | etp_atfork_parent (); |
473 | etp_thread_init (); |
379 | } |
474 | } |
380 | |
475 | |
381 | static void |
476 | static void ecb_cold |
382 | etp_once_init (void) |
477 | etp_once_init (void) |
383 | { |
478 | { |
|
|
479 | etp_thread_init (); |
384 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
480 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
385 | } |
481 | } |
386 | |
482 | |
387 | static int |
483 | static int ecb_cold |
388 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
484 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
389 | { |
485 | { |
390 | static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
486 | static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
391 | |
487 | |
392 | pthread_once (&doinit, etp_once_init); |
488 | pthread_once (&doinit, etp_once_init); |
… | |
… | |
397 | return 0; |
493 | return 0; |
398 | } |
494 | } |
399 | |
495 | |
400 | X_THREAD_PROC (etp_proc); |
496 | X_THREAD_PROC (etp_proc); |
401 | |
497 | |
|
|
498 | static void ecb_cold |
402 | static void etp_start_thread (void) |
499 | etp_start_thread (void) |
403 | { |
500 | { |
404 | etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
501 | etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
405 | |
502 | |
406 | /*TODO*/ |
503 | /*TODO*/ |
407 | assert (("unable to allocate worker thread data", wrk)); |
504 | assert (("unable to allocate worker thread data", wrk)); |
… | |
… | |
420 | free (wrk); |
517 | free (wrk); |
421 | |
518 | |
422 | X_UNLOCK (wrklock); |
519 | X_UNLOCK (wrklock); |
423 | } |
520 | } |
424 | |
521 | |
|
|
522 | static void |
425 | static void etp_maybe_start_thread (void) |
523 | etp_maybe_start_thread (void) |
426 | { |
524 | { |
427 | if (expect_true (etp_nthreads () >= wanted)) |
525 | if (ecb_expect_true (etp_nthreads () >= wanted)) |
428 | return; |
526 | return; |
429 | |
527 | |
430 | /* todo: maybe use idle here, but might be less exact */ |
528 | /* todo: maybe use idle here, but might be less exact */ |
431 | if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
529 | if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
432 | return; |
530 | return; |
433 | |
531 | |
434 | etp_start_thread (); |
532 | etp_start_thread (); |
435 | } |
533 | } |
436 | |
534 | |
|
|
535 | static void ecb_cold |
437 | static void etp_end_thread (void) |
536 | etp_end_thread (void) |
438 | { |
537 | { |
439 | eio_req *req = calloc (1, sizeof (eio_req)); |
538 | eio_req *req = calloc (1, sizeof (eio_req)); |
440 | |
539 | |
441 | req->type = -1; |
540 | req->type = -1; |
442 | req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
541 | req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
… | |
… | |
449 | X_LOCK (wrklock); |
548 | X_LOCK (wrklock); |
450 | --started; |
549 | --started; |
451 | X_UNLOCK (wrklock); |
550 | X_UNLOCK (wrklock); |
452 | } |
551 | } |
453 | |
552 | |
454 | static int etp_poll (void) |
553 | static int |
|
|
554 | etp_poll (void) |
455 | { |
555 | { |
456 | unsigned int maxreqs; |
556 | unsigned int maxreqs; |
457 | unsigned int maxtime; |
557 | unsigned int maxtime; |
458 | struct timeval tv_start, tv_now; |
558 | struct timeval tv_start, tv_now; |
459 | |
559 | |
… | |
… | |
489 | |
589 | |
490 | X_LOCK (reqlock); |
590 | X_LOCK (reqlock); |
491 | --nreqs; |
591 | --nreqs; |
492 | X_UNLOCK (reqlock); |
592 | X_UNLOCK (reqlock); |
493 | |
593 | |
494 | if (expect_false (req->type == EIO_GROUP && req->size)) |
594 | if (ecb_expect_false (req->type == EIO_GROUP && req->size)) |
495 | { |
595 | { |
496 | req->int1 = 1; /* mark request as delayed */ |
596 | req->int1 = 1; /* mark request as delayed */ |
497 | continue; |
597 | continue; |
498 | } |
598 | } |
499 | else |
599 | else |
500 | { |
600 | { |
501 | int res = ETP_FINISH (req); |
601 | int res = ETP_FINISH (req); |
502 | if (expect_false (res)) |
602 | if (ecb_expect_false (res)) |
503 | return res; |
603 | return res; |
504 | } |
604 | } |
505 | |
605 | |
506 | if (expect_false (maxreqs && !--maxreqs)) |
606 | if (ecb_expect_false (maxreqs && !--maxreqs)) |
507 | break; |
607 | break; |
508 | |
608 | |
509 | if (maxtime) |
609 | if (maxtime) |
510 | { |
610 | { |
511 | gettimeofday (&tv_now, 0); |
611 | gettimeofday (&tv_now, 0); |
… | |
… | |
517 | |
617 | |
518 | errno = EAGAIN; |
618 | errno = EAGAIN; |
519 | return -1; |
619 | return -1; |
520 | } |
620 | } |
521 | |
621 | |
|
|
622 | static void |
522 | static void etp_cancel (ETP_REQ *req) |
623 | etp_cancel (ETP_REQ *req) |
523 | { |
624 | { |
524 | X_LOCK (wrklock); |
625 | req->cancelled = 1; |
525 | req->flags |= EIO_FLAG_CANCELLED; |
|
|
526 | X_UNLOCK (wrklock); |
|
|
527 | |
626 | |
528 | eio_grp_cancel (req); |
627 | eio_grp_cancel (req); |
529 | } |
628 | } |
530 | |
629 | |
|
|
630 | static void |
531 | static void etp_submit (ETP_REQ *req) |
631 | etp_submit (ETP_REQ *req) |
532 | { |
632 | { |
533 | req->pri -= ETP_PRI_MIN; |
633 | req->pri -= ETP_PRI_MIN; |
534 | |
634 | |
535 | if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
635 | 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; |
636 | if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
537 | |
637 | |
538 | if (expect_false (req->type == EIO_GROUP)) |
638 | if (ecb_expect_false (req->type == EIO_GROUP)) |
539 | { |
639 | { |
540 | /* I hope this is worth it :/ */ |
640 | /* I hope this is worth it :/ */ |
541 | X_LOCK (reqlock); |
641 | X_LOCK (reqlock); |
542 | ++nreqs; |
642 | ++nreqs; |
543 | X_UNLOCK (reqlock); |
643 | X_UNLOCK (reqlock); |
… | |
… | |
562 | |
662 | |
563 | etp_maybe_start_thread (); |
663 | etp_maybe_start_thread (); |
564 | } |
664 | } |
565 | } |
665 | } |
566 | |
666 | |
|
|
667 | static void ecb_cold |
567 | static void etp_set_max_poll_time (double nseconds) |
668 | etp_set_max_poll_time (double nseconds) |
568 | { |
669 | { |
569 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
670 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
570 | max_poll_time = nseconds; |
671 | max_poll_time = nseconds * EIO_TICKS; |
571 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
672 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
572 | } |
673 | } |
573 | |
674 | |
|
|
675 | static void ecb_cold |
574 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
676 | etp_set_max_poll_reqs (unsigned int maxreqs) |
575 | { |
677 | { |
576 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
678 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
577 | max_poll_reqs = maxreqs; |
679 | max_poll_reqs = maxreqs; |
578 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
680 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
579 | } |
681 | } |
580 | |
682 | |
|
|
683 | static void ecb_cold |
581 | static void etp_set_max_idle (unsigned int nthreads) |
684 | etp_set_max_idle (unsigned int nthreads) |
582 | { |
685 | { |
583 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
686 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
584 | max_idle = nthreads <= 0 ? 1 : nthreads; |
687 | max_idle = nthreads; |
585 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
688 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
586 | } |
689 | } |
587 | |
690 | |
|
|
691 | static void ecb_cold |
|
|
692 | etp_set_idle_timeout (unsigned int seconds) |
|
|
693 | { |
|
|
694 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
695 | idle_timeout = seconds; |
|
|
696 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
697 | } |
|
|
698 | |
|
|
699 | static void ecb_cold |
588 | static void etp_set_min_parallel (unsigned int nthreads) |
700 | etp_set_min_parallel (unsigned int nthreads) |
589 | { |
701 | { |
590 | if (wanted < nthreads) |
702 | if (wanted < nthreads) |
591 | wanted = nthreads; |
703 | wanted = nthreads; |
592 | } |
704 | } |
593 | |
705 | |
|
|
706 | static void ecb_cold |
594 | static void etp_set_max_parallel (unsigned int nthreads) |
707 | etp_set_max_parallel (unsigned int nthreads) |
595 | { |
708 | { |
596 | if (wanted > nthreads) |
709 | if (wanted > nthreads) |
597 | wanted = nthreads; |
710 | wanted = nthreads; |
598 | |
711 | |
599 | while (started > wanted) |
712 | while (started > wanted) |
600 | etp_end_thread (); |
713 | etp_end_thread (); |
601 | } |
714 | } |
602 | |
715 | |
603 | /*****************************************************************************/ |
716 | /*****************************************************************************/ |
604 | |
717 | |
|
|
718 | static void |
605 | static void grp_try_feed (eio_req *grp) |
719 | grp_try_feed (eio_req *grp) |
606 | { |
720 | { |
607 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
721 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
608 | { |
722 | { |
609 | grp->flags &= ~EIO_FLAG_GROUPADD; |
723 | grp->flags &= ~EIO_FLAG_GROUPADD; |
610 | |
724 | |
… | |
… | |
617 | break; |
731 | break; |
618 | } |
732 | } |
619 | } |
733 | } |
620 | } |
734 | } |
621 | |
735 | |
|
|
736 | static int |
622 | static int grp_dec (eio_req *grp) |
737 | grp_dec (eio_req *grp) |
623 | { |
738 | { |
624 | --grp->size; |
739 | --grp->size; |
625 | |
740 | |
626 | /* call feeder, if applicable */ |
741 | /* call feeder, if applicable */ |
627 | grp_try_feed (grp); |
742 | grp_try_feed (grp); |
… | |
… | |
631 | return eio_finish (grp); |
746 | return eio_finish (grp); |
632 | else |
747 | else |
633 | return 0; |
748 | return 0; |
634 | } |
749 | } |
635 | |
750 | |
|
|
751 | static void |
636 | void eio_destroy (eio_req *req) |
752 | eio_destroy (eio_req *req) |
637 | { |
753 | { |
638 | if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
754 | if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
639 | if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
755 | if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
640 | |
756 | |
641 | EIO_DESTROY (req); |
757 | EIO_DESTROY (req); |
642 | } |
758 | } |
643 | |
759 | |
|
|
760 | static int |
644 | static int eio_finish (eio_req *req) |
761 | eio_finish (eio_req *req) |
645 | { |
762 | { |
646 | int res = EIO_FINISH (req); |
763 | int res = EIO_FINISH (req); |
647 | |
764 | |
648 | if (req->grp) |
765 | if (req->grp) |
649 | { |
766 | { |
… | |
… | |
657 | if (grp->grp_first == req) |
774 | if (grp->grp_first == req) |
658 | grp->grp_first = req->grp_next; |
775 | grp->grp_first = req->grp_next; |
659 | |
776 | |
660 | res2 = grp_dec (grp); |
777 | res2 = grp_dec (grp); |
661 | |
778 | |
662 | if (!res && res2) |
779 | if (!res) |
663 | res = res2; |
780 | res = res2; |
664 | } |
781 | } |
665 | |
782 | |
666 | eio_destroy (req); |
783 | eio_destroy (req); |
667 | |
784 | |
668 | return res; |
785 | return res; |
669 | } |
786 | } |
670 | |
787 | |
|
|
788 | void |
671 | void eio_grp_cancel (eio_req *grp) |
789 | eio_grp_cancel (eio_req *grp) |
672 | { |
790 | { |
673 | for (grp = grp->grp_first; grp; grp = grp->grp_next) |
791 | for (grp = grp->grp_first; grp; grp = grp->grp_next) |
674 | eio_cancel (grp); |
792 | eio_cancel (grp); |
675 | } |
793 | } |
676 | |
794 | |
|
|
795 | void |
677 | void eio_cancel (eio_req *req) |
796 | eio_cancel (eio_req *req) |
678 | { |
797 | { |
679 | etp_cancel (req); |
798 | etp_cancel (req); |
680 | } |
799 | } |
681 | |
800 | |
|
|
801 | void |
682 | void eio_submit (eio_req *req) |
802 | eio_submit (eio_req *req) |
683 | { |
803 | { |
684 | etp_submit (req); |
804 | etp_submit (req); |
685 | } |
805 | } |
686 | |
806 | |
687 | unsigned int eio_nreqs (void) |
807 | unsigned int |
|
|
808 | eio_nreqs (void) |
688 | { |
809 | { |
689 | return etp_nreqs (); |
810 | return etp_nreqs (); |
690 | } |
811 | } |
691 | |
812 | |
692 | unsigned int eio_nready (void) |
813 | unsigned int |
|
|
814 | eio_nready (void) |
693 | { |
815 | { |
694 | return etp_nready (); |
816 | return etp_nready (); |
695 | } |
817 | } |
696 | |
818 | |
697 | unsigned int eio_npending (void) |
819 | unsigned int |
|
|
820 | eio_npending (void) |
698 | { |
821 | { |
699 | return etp_npending (); |
822 | return etp_npending (); |
700 | } |
823 | } |
701 | |
824 | |
702 | unsigned int eio_nthreads (void) |
825 | unsigned int ecb_cold |
|
|
826 | eio_nthreads (void) |
703 | { |
827 | { |
704 | return etp_nthreads (); |
828 | return etp_nthreads (); |
705 | } |
829 | } |
706 | |
830 | |
|
|
831 | void ecb_cold |
707 | void eio_set_max_poll_time (double nseconds) |
832 | eio_set_max_poll_time (double nseconds) |
708 | { |
833 | { |
709 | etp_set_max_poll_time (nseconds); |
834 | etp_set_max_poll_time (nseconds); |
710 | } |
835 | } |
711 | |
836 | |
|
|
837 | void ecb_cold |
712 | void eio_set_max_poll_reqs (unsigned int maxreqs) |
838 | eio_set_max_poll_reqs (unsigned int maxreqs) |
713 | { |
839 | { |
714 | etp_set_max_poll_reqs (maxreqs); |
840 | etp_set_max_poll_reqs (maxreqs); |
715 | } |
841 | } |
716 | |
842 | |
|
|
843 | void ecb_cold |
717 | void eio_set_max_idle (unsigned int nthreads) |
844 | eio_set_max_idle (unsigned int nthreads) |
718 | { |
845 | { |
719 | etp_set_max_idle (nthreads); |
846 | etp_set_max_idle (nthreads); |
720 | } |
847 | } |
721 | |
848 | |
|
|
849 | void ecb_cold |
|
|
850 | eio_set_idle_timeout (unsigned int seconds) |
|
|
851 | { |
|
|
852 | etp_set_idle_timeout (seconds); |
|
|
853 | } |
|
|
854 | |
|
|
855 | void ecb_cold |
722 | void eio_set_min_parallel (unsigned int nthreads) |
856 | eio_set_min_parallel (unsigned int nthreads) |
723 | { |
857 | { |
724 | etp_set_min_parallel (nthreads); |
858 | etp_set_min_parallel (nthreads); |
725 | } |
859 | } |
726 | |
860 | |
|
|
861 | void ecb_cold |
727 | void eio_set_max_parallel (unsigned int nthreads) |
862 | eio_set_max_parallel (unsigned int nthreads) |
728 | { |
863 | { |
729 | etp_set_max_parallel (nthreads); |
864 | etp_set_max_parallel (nthreads); |
730 | } |
865 | } |
731 | |
866 | |
732 | int eio_poll (void) |
867 | int eio_poll (void) |
… | |
… | |
767 | |
902 | |
768 | X_LOCK (preadwritelock); |
903 | X_LOCK (preadwritelock); |
769 | ooffset = lseek (fd, 0, SEEK_CUR); |
904 | ooffset = lseek (fd, 0, SEEK_CUR); |
770 | lseek (fd, offset, SEEK_SET); |
905 | lseek (fd, offset, SEEK_SET); |
771 | res = write (fd, buf, count); |
906 | res = write (fd, buf, count); |
772 | lseek (fd, offset, SEEK_SET); |
907 | lseek (fd, ooffset, SEEK_SET); |
773 | X_UNLOCK (preadwritelock); |
908 | X_UNLOCK (preadwritelock); |
774 | |
909 | |
775 | return res; |
910 | return res; |
776 | } |
911 | } |
777 | #endif |
912 | #endif |
778 | |
913 | |
779 | #ifndef HAVE_FUTIMES |
914 | #ifndef HAVE_UTIMES |
780 | |
915 | |
781 | # undef utimes |
916 | # undef utimes |
782 | # undef futimes |
|
|
783 | # define utimes(path,times) eio__utimes (path, times) |
917 | # define utimes(path,times) eio__utimes (path, times) |
784 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
785 | |
918 | |
786 | static int |
919 | static int |
787 | eio__utimes (const char *filename, const struct timeval times[2]) |
920 | eio__utimes (const char *filename, const struct timeval times[2]) |
788 | { |
921 | { |
789 | if (times) |
922 | if (times) |
… | |
… | |
797 | } |
930 | } |
798 | else |
931 | else |
799 | return utime (filename, 0); |
932 | return utime (filename, 0); |
800 | } |
933 | } |
801 | |
934 | |
|
|
935 | #endif |
|
|
936 | |
|
|
937 | #ifndef HAVE_FUTIMES |
|
|
938 | |
|
|
939 | # undef futimes |
|
|
940 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
941 | |
|
|
942 | static int |
802 | static int eio__futimes (int fd, const struct timeval tv[2]) |
943 | eio__futimes (int fd, const struct timeval tv[2]) |
803 | { |
944 | { |
804 | errno = ENOSYS; |
945 | errno = ENOSYS; |
805 | return -1; |
946 | return -1; |
806 | } |
947 | } |
807 | |
948 | |
… | |
… | |
811 | # undef fdatasync |
952 | # undef fdatasync |
812 | # define fdatasync(fd) fsync (fd) |
953 | # define fdatasync(fd) fsync (fd) |
813 | #endif |
954 | #endif |
814 | |
955 | |
815 | /* sync_file_range always needs emulation */ |
956 | /* sync_file_range always needs emulation */ |
816 | int |
957 | static int |
817 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
958 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
818 | { |
959 | { |
819 | #if HAVE_SYNC_FILE_RANGE |
960 | #if HAVE_SYNC_FILE_RANGE |
820 | int res; |
961 | int res; |
821 | |
962 | |
… | |
… | |
829 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
970 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
830 | } |
971 | } |
831 | |
972 | |
832 | res = sync_file_range (fd, offset, nbytes, flags); |
973 | res = sync_file_range (fd, offset, nbytes, flags); |
833 | |
974 | |
834 | if (res != ENOSYS) |
975 | if (!res || errno != ENOSYS) |
835 | return res; |
976 | return res; |
836 | #endif |
977 | #endif |
837 | |
978 | |
838 | /* even though we could play tricks with the flags, it's better to always |
979 | /* even though we could play tricks with the flags, it's better to always |
839 | * call fdatasync, as thta matches the expectation of it's users best */ |
980 | * call fdatasync, as that matches the expectation of its users best */ |
840 | return fdatasync (fd); |
981 | return fdatasync (fd); |
|
|
982 | } |
|
|
983 | |
|
|
984 | static int |
|
|
985 | eio__fallocate (int fd, int mode, off_t offset, size_t len) |
|
|
986 | { |
|
|
987 | #if HAVE_FALLOCATE |
|
|
988 | return fallocate (fd, mode, offset, len); |
|
|
989 | #else |
|
|
990 | errno = ENOSYS; |
|
|
991 | return -1; |
|
|
992 | #endif |
841 | } |
993 | } |
842 | |
994 | |
843 | #if !HAVE_READAHEAD |
995 | #if !HAVE_READAHEAD |
844 | # undef readahead |
996 | # undef readahead |
845 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
997 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
… | |
… | |
867 | |
1019 | |
868 | /* sendfile always needs emulation */ |
1020 | /* sendfile always needs emulation */ |
869 | static ssize_t |
1021 | static ssize_t |
870 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
1022 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
871 | { |
1023 | { |
|
|
1024 | ssize_t written = 0; |
872 | ssize_t res; |
1025 | ssize_t res; |
873 | |
1026 | |
874 | if (!count) |
1027 | if (!count) |
875 | return 0; |
1028 | return 0; |
876 | |
1029 | |
|
|
1030 | for (;;) |
|
|
1031 | { |
|
|
1032 | #ifdef __APPLE__ |
|
|
1033 | # undef HAVE_SENDFILE /* broken, as everything on os x */ |
|
|
1034 | #endif |
877 | #if HAVE_SENDFILE |
1035 | #if HAVE_SENDFILE |
878 | # if __linux |
1036 | # if __linux |
|
|
1037 | off_t soffset = offset; |
879 | res = sendfile (ofd, ifd, &offset, count); |
1038 | res = sendfile (ofd, ifd, &soffset, count); |
880 | |
1039 | |
881 | # elif __freebsd |
1040 | # elif __FreeBSD__ |
882 | /* |
1041 | /* |
883 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
1042 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
884 | * wasted on making it similar to other I/O functions. |
1043 | * wasted on making it similar to other I/O functions. |
885 | */ |
1044 | */ |
886 | { |
|
|
887 | off_t sbytes; |
1045 | off_t sbytes; |
888 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
1046 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
889 | |
1047 | |
890 | if (res < 0 && sbytes) |
1048 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
891 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
1049 | /* freebsd' sendfile will return 0 on success */ |
|
|
1050 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
|
|
1051 | /* not on e.g. EIO or EPIPE - sounds broken */ |
|
|
1052 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
892 | res = sbytes; |
1053 | res = sbytes; |
893 | } |
1054 | #endif |
|
|
1055 | |
|
|
1056 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
1057 | if (sbytes) |
|
|
1058 | res = sbytes; |
|
|
1059 | |
|
|
1060 | # elif defined (__APPLE__) |
|
|
1061 | off_t sbytes = count; |
|
|
1062 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
1063 | |
|
|
1064 | /* according to the manpage, sbytes is always valid */ |
|
|
1065 | if (sbytes) |
|
|
1066 | res = sbytes; |
894 | |
1067 | |
895 | # elif __hpux |
1068 | # elif __hpux |
896 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
1069 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
897 | |
1070 | |
898 | # elif __solaris |
1071 | # elif __solaris |
899 | { |
|
|
900 | struct sendfilevec vec; |
1072 | struct sendfilevec vec; |
901 | size_t sbytes; |
1073 | size_t sbytes; |
902 | |
1074 | |
903 | vec.sfv_fd = ifd; |
1075 | vec.sfv_fd = ifd; |
904 | vec.sfv_flag = 0; |
1076 | vec.sfv_flag = 0; |
905 | vec.sfv_off = offset; |
1077 | vec.sfv_off = offset; |
906 | vec.sfv_len = count; |
1078 | vec.sfv_len = count; |
907 | |
1079 | |
908 | res = sendfilev (ofd, &vec, 1, &sbytes); |
1080 | res = sendfilev (ofd, &vec, 1, &sbytes); |
909 | |
1081 | |
910 | if (res < 0 && sbytes) |
1082 | if (res < 0 && sbytes) |
911 | res = sbytes; |
1083 | res = sbytes; |
912 | } |
|
|
913 | |
1084 | |
914 | # endif |
1085 | # endif |
|
|
1086 | |
|
|
1087 | #elif defined (_WIN32) && 0 |
|
|
1088 | /* does not work, just for documentation of what would need to be done */ |
|
|
1089 | /* actually, cannot be done like this, as TransmitFile changes the file offset, */ |
|
|
1090 | /* libeio guarantees that the file offset does not change, and windows */ |
|
|
1091 | /* has no way to get an independent handle to the same file description */ |
|
|
1092 | HANDLE h = TO_SOCKET (ifd); |
|
|
1093 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
1094 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
1095 | |
915 | #else |
1096 | #else |
916 | res = -1; |
1097 | res = -1; |
917 | errno = ENOSYS; |
1098 | errno = ENOSYS; |
918 | #endif |
1099 | #endif |
919 | |
1100 | |
|
|
1101 | /* we assume sendfile can copy at least 128mb in one go */ |
|
|
1102 | if (res <= 128 * 1024 * 1024) |
|
|
1103 | { |
|
|
1104 | if (res > 0) |
|
|
1105 | written += res; |
|
|
1106 | |
|
|
1107 | if (written) |
|
|
1108 | return written; |
|
|
1109 | |
|
|
1110 | break; |
|
|
1111 | } |
|
|
1112 | else |
|
|
1113 | { |
|
|
1114 | /* if we requested more, then probably the kernel was lazy */ |
|
|
1115 | written += res; |
|
|
1116 | offset += res; |
|
|
1117 | count -= res; |
|
|
1118 | |
|
|
1119 | if (!count) |
|
|
1120 | return written; |
|
|
1121 | } |
|
|
1122 | } |
|
|
1123 | |
920 | if (res < 0 |
1124 | if (res < 0 |
921 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1125 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
1126 | /* BSDs */ |
|
|
1127 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
1128 | || errno == ENOTSUP |
|
|
1129 | #endif |
|
|
1130 | || errno == EOPNOTSUPP /* BSDs */ |
922 | #if __solaris |
1131 | #if __solaris |
923 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1132 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
924 | #endif |
1133 | #endif |
925 | ) |
1134 | ) |
926 | ) |
1135 | ) |
… | |
… | |
957 | } |
1166 | } |
958 | |
1167 | |
959 | return res; |
1168 | return res; |
960 | } |
1169 | } |
961 | |
1170 | |
|
|
1171 | #ifdef PAGESIZE |
|
|
1172 | # define eio_pagesize() PAGESIZE |
|
|
1173 | #else |
|
|
1174 | static intptr_t |
|
|
1175 | eio_pagesize (void) |
|
|
1176 | { |
|
|
1177 | static intptr_t page; |
|
|
1178 | |
|
|
1179 | if (!page) |
|
|
1180 | page = sysconf (_SC_PAGESIZE); |
|
|
1181 | |
|
|
1182 | return page; |
|
|
1183 | } |
|
|
1184 | #endif |
|
|
1185 | |
|
|
1186 | static void |
|
|
1187 | eio_page_align (void **addr, size_t *length) |
|
|
1188 | { |
|
|
1189 | intptr_t mask = eio_pagesize () - 1; |
|
|
1190 | |
|
|
1191 | /* round down addr */ |
|
|
1192 | intptr_t adj = mask & (intptr_t)*addr; |
|
|
1193 | |
|
|
1194 | *addr = (void *)((intptr_t)*addr - adj); |
|
|
1195 | *length += adj; |
|
|
1196 | |
|
|
1197 | /* round up length */ |
|
|
1198 | *length = (*length + mask) & ~mask; |
|
|
1199 | } |
|
|
1200 | |
|
|
1201 | #if !_POSIX_MEMLOCK |
|
|
1202 | # define eio__mlockall(a) eio_nosyscall() |
|
|
1203 | #else |
|
|
1204 | |
|
|
1205 | static int |
|
|
1206 | eio__mlockall (int flags) |
|
|
1207 | { |
|
|
1208 | #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
|
|
1209 | extern int mallopt (int, int); |
|
|
1210 | mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
|
|
1211 | #endif |
|
|
1212 | |
|
|
1213 | if (EIO_MCL_CURRENT != MCL_CURRENT |
|
|
1214 | || EIO_MCL_FUTURE != MCL_FUTURE) |
|
|
1215 | { |
|
|
1216 | flags = 0 |
|
|
1217 | | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
|
|
1218 | | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
|
|
1219 | } |
|
|
1220 | |
|
|
1221 | return mlockall (flags); |
|
|
1222 | } |
|
|
1223 | #endif |
|
|
1224 | |
|
|
1225 | #if !_POSIX_MEMLOCK_RANGE |
|
|
1226 | # define eio__mlock(a,b) EIO_ENOSYS () |
|
|
1227 | #else |
|
|
1228 | |
|
|
1229 | static int |
|
|
1230 | eio__mlock (void *addr, size_t length) |
|
|
1231 | { |
|
|
1232 | eio_page_align (&addr, &length); |
|
|
1233 | |
|
|
1234 | return mlock (addr, length); |
|
|
1235 | } |
|
|
1236 | |
|
|
1237 | #endif |
|
|
1238 | |
|
|
1239 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
|
|
1240 | # define eio__msync(a,b,c) EIO_ENOSYS () |
|
|
1241 | #else |
|
|
1242 | |
|
|
1243 | static int |
|
|
1244 | eio__msync (void *mem, size_t len, int flags) |
|
|
1245 | { |
|
|
1246 | eio_page_align (&mem, &len); |
|
|
1247 | |
|
|
1248 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1249 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1250 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1251 | { |
|
|
1252 | flags = 0 |
|
|
1253 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1254 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1255 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1256 | } |
|
|
1257 | |
|
|
1258 | return msync (mem, len, flags); |
|
|
1259 | } |
|
|
1260 | |
|
|
1261 | #endif |
|
|
1262 | |
|
|
1263 | static int |
|
|
1264 | eio__mtouch (eio_req *req) |
|
|
1265 | { |
|
|
1266 | void *mem = req->ptr2; |
|
|
1267 | size_t len = req->size; |
|
|
1268 | int flags = req->int1; |
|
|
1269 | |
|
|
1270 | eio_page_align (&mem, &len); |
|
|
1271 | |
|
|
1272 | { |
|
|
1273 | intptr_t addr = (intptr_t)mem; |
|
|
1274 | intptr_t end = addr + len; |
|
|
1275 | intptr_t page = eio_pagesize (); |
|
|
1276 | |
|
|
1277 | if (addr < end) |
|
|
1278 | if (flags & EIO_MT_MODIFY) /* modify */ |
|
|
1279 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
|
|
1280 | else |
|
|
1281 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
|
|
1282 | } |
|
|
1283 | |
|
|
1284 | return 0; |
|
|
1285 | } |
|
|
1286 | |
|
|
1287 | /*****************************************************************************/ |
|
|
1288 | /* requests implemented outside eio_execute, because they are so large */ |
|
|
1289 | |
|
|
1290 | static void |
|
|
1291 | eio__realpath (eio_req *req, etp_worker *self) |
|
|
1292 | { |
|
|
1293 | char *rel = req->ptr1; |
|
|
1294 | char *res; |
|
|
1295 | char *tmp1, *tmp2; |
|
|
1296 | #if SYMLOOP_MAX > 32 |
|
|
1297 | int symlinks = SYMLOOP_MAX; |
|
|
1298 | #else |
|
|
1299 | int symlinks = 32; |
|
|
1300 | #endif |
|
|
1301 | |
|
|
1302 | req->result = -1; |
|
|
1303 | |
|
|
1304 | errno = EINVAL; |
|
|
1305 | if (!rel) |
|
|
1306 | return; |
|
|
1307 | |
|
|
1308 | errno = ENOENT; |
|
|
1309 | if (!*rel) |
|
|
1310 | return; |
|
|
1311 | |
|
|
1312 | if (!req->ptr2) |
|
|
1313 | { |
|
|
1314 | X_LOCK (wrklock); |
|
|
1315 | req->flags |= EIO_FLAG_PTR2_FREE; |
|
|
1316 | X_UNLOCK (wrklock); |
|
|
1317 | req->ptr2 = malloc (PATH_MAX * 3); |
|
|
1318 | |
|
|
1319 | errno = ENOMEM; |
|
|
1320 | if (!req->ptr2) |
|
|
1321 | return; |
|
|
1322 | } |
|
|
1323 | |
|
|
1324 | res = req->ptr2; |
|
|
1325 | tmp1 = res + PATH_MAX; |
|
|
1326 | tmp2 = tmp1 + PATH_MAX; |
|
|
1327 | |
|
|
1328 | #if 0 /* disabled, the musl way to do things is just too racy */ |
|
|
1329 | #if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) |
|
|
1330 | /* on linux we may be able to ask the kernel */ |
|
|
1331 | { |
|
|
1332 | int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME); |
|
|
1333 | |
|
|
1334 | if (fd >= 0) |
|
|
1335 | { |
|
|
1336 | sprintf (tmp1, "/proc/self/fd/%d", fd); |
|
|
1337 | req->result = readlink (tmp1, res, PATH_MAX); |
|
|
1338 | close (fd); |
|
|
1339 | |
|
|
1340 | /* here we should probably stat the open file and the disk file, to make sure they still match */ |
|
|
1341 | |
|
|
1342 | if (req->result > 0) |
|
|
1343 | goto done; |
|
|
1344 | } |
|
|
1345 | else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) |
|
|
1346 | return; |
|
|
1347 | } |
|
|
1348 | #endif |
|
|
1349 | #endif |
|
|
1350 | |
|
|
1351 | if (*rel != '/') |
|
|
1352 | { |
|
|
1353 | if (!getcwd (res, PATH_MAX)) |
|
|
1354 | return; |
|
|
1355 | |
|
|
1356 | if (res [1]) /* only use if not / */ |
|
|
1357 | res += strlen (res); |
|
|
1358 | } |
|
|
1359 | |
|
|
1360 | while (*rel) |
|
|
1361 | { |
|
|
1362 | ssize_t len, linklen; |
|
|
1363 | char *beg = rel; |
|
|
1364 | |
|
|
1365 | while (*rel && *rel != '/') |
|
|
1366 | ++rel; |
|
|
1367 | |
|
|
1368 | len = rel - beg; |
|
|
1369 | |
|
|
1370 | if (!len) /* skip slashes */ |
|
|
1371 | { |
|
|
1372 | ++rel; |
|
|
1373 | continue; |
|
|
1374 | } |
|
|
1375 | |
|
|
1376 | if (beg [0] == '.') |
|
|
1377 | { |
|
|
1378 | if (len == 1) |
|
|
1379 | continue; /* . - nop */ |
|
|
1380 | |
|
|
1381 | if (beg [1] == '.' && len == 2) |
|
|
1382 | { |
|
|
1383 | /* .. - back up one component, if possible */ |
|
|
1384 | |
|
|
1385 | while (res != req->ptr2) |
|
|
1386 | if (*--res == '/') |
|
|
1387 | break; |
|
|
1388 | |
|
|
1389 | continue; |
|
|
1390 | } |
|
|
1391 | } |
|
|
1392 | |
|
|
1393 | errno = ENAMETOOLONG; |
|
|
1394 | if (res + 1 + len + 1 >= tmp1) |
|
|
1395 | return; |
|
|
1396 | |
|
|
1397 | /* copy one component */ |
|
|
1398 | *res = '/'; |
|
|
1399 | memcpy (res + 1, beg, len); |
|
|
1400 | |
|
|
1401 | /* zero-terminate, for readlink */ |
|
|
1402 | res [len + 1] = 0; |
|
|
1403 | |
|
|
1404 | /* now check if it's a symlink */ |
|
|
1405 | linklen = readlink (req->ptr2, tmp1, PATH_MAX); |
|
|
1406 | |
|
|
1407 | if (linklen < 0) |
|
|
1408 | { |
|
|
1409 | if (errno != EINVAL) |
|
|
1410 | return; |
|
|
1411 | |
|
|
1412 | /* it's a normal directory. hopefully */ |
|
|
1413 | res += len + 1; |
|
|
1414 | } |
|
|
1415 | else |
|
|
1416 | { |
|
|
1417 | /* yay, it was a symlink - build new path in tmp2 */ |
|
|
1418 | int rellen = strlen (rel); |
|
|
1419 | |
|
|
1420 | errno = ENAMETOOLONG; |
|
|
1421 | if (linklen + 1 + rellen >= PATH_MAX) |
|
|
1422 | return; |
|
|
1423 | |
|
|
1424 | errno = ELOOP; |
|
|
1425 | if (!--symlinks) |
|
|
1426 | return; |
|
|
1427 | |
|
|
1428 | if (*tmp1 == '/') |
|
|
1429 | res = req->ptr2; /* symlink resolves to an absolute path */ |
|
|
1430 | |
|
|
1431 | /* we need to be careful, as rel might point into tmp2 already */ |
|
|
1432 | memmove (tmp2 + linklen + 1, rel, rellen + 1); |
|
|
1433 | tmp2 [linklen] = '/'; |
|
|
1434 | memcpy (tmp2, tmp1, linklen); |
|
|
1435 | |
|
|
1436 | rel = tmp2; |
|
|
1437 | } |
|
|
1438 | } |
|
|
1439 | |
|
|
1440 | /* special case for the lone root path */ |
|
|
1441 | if (res == req->ptr2) |
|
|
1442 | *res++ = '/'; |
|
|
1443 | |
|
|
1444 | req->result = res - (char *)req->ptr2; |
|
|
1445 | |
|
|
1446 | done: |
|
|
1447 | req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */ |
|
|
1448 | } |
|
|
1449 | |
|
|
1450 | static signed char |
|
|
1451 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
|
|
1452 | { |
|
|
1453 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
|
|
1454 | : a->inode < b->inode ? -1 |
|
|
1455 | : a->inode > b->inode ? 1 |
|
|
1456 | : 0; |
|
|
1457 | } |
|
|
1458 | |
|
|
1459 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1460 | |
|
|
1461 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1462 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
|
|
1463 | |
|
|
1464 | static void |
|
|
1465 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1466 | { |
|
|
1467 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1468 | unsigned char *bit = bits; |
|
|
1469 | |
|
|
1470 | assert (CHAR_BIT == 8); |
|
|
1471 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1472 | assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
|
|
1473 | assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ |
|
|
1474 | |
|
|
1475 | if (size <= EIO_SORT_FAST) |
|
|
1476 | return; |
|
|
1477 | |
|
|
1478 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1479 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1480 | /* inode_bits must contain all inodes ORed together */ |
|
|
1481 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1482 | { |
|
|
1483 | ino_t endianness; |
|
|
1484 | int i, j; |
|
|
1485 | |
|
|
1486 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1487 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1488 | ((unsigned char *)&endianness)[i] = i; |
|
|
1489 | |
|
|
1490 | *bit++ = 0; |
|
|
1491 | |
|
|
1492 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1493 | { |
|
|
1494 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1495 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1496 | endianness >>= 8; |
|
|
1497 | |
|
|
1498 | for (j = 0; j < 8; ++j) |
|
|
1499 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1500 | *bit++ = offs + j; |
|
|
1501 | } |
|
|
1502 | |
|
|
1503 | for (j = 0; j < 8; ++j) |
|
|
1504 | if (score_bits & (1 << j)) |
|
|
1505 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1506 | } |
|
|
1507 | |
|
|
1508 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1509 | { |
|
|
1510 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1511 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1512 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1513 | int stk_idx = 0; |
|
|
1514 | |
|
|
1515 | base_stk [stk_idx] = dents; |
|
|
1516 | end_stk [stk_idx] = dents + size; |
|
|
1517 | bit_stk [stk_idx] = bit - 1; |
|
|
1518 | |
|
|
1519 | do |
|
|
1520 | { |
|
|
1521 | base = base_stk [stk_idx]; |
|
|
1522 | end = end_stk [stk_idx]; |
|
|
1523 | bit = bit_stk [stk_idx]; |
|
|
1524 | |
|
|
1525 | for (;;) |
|
|
1526 | { |
|
|
1527 | unsigned char O = *bit >> 3; |
|
|
1528 | unsigned char M = 1 << (*bit & 7); |
|
|
1529 | |
|
|
1530 | eio_dirent *a = base; |
|
|
1531 | eio_dirent *b = end; |
|
|
1532 | |
|
|
1533 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1534 | break; |
|
|
1535 | |
|
|
1536 | /* now bit-partition the array on the bit */ |
|
|
1537 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1538 | /* partition algos found in the literature */ |
|
|
1539 | do |
|
|
1540 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1541 | ++a; |
|
|
1542 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1543 | { |
|
|
1544 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1545 | ++a; |
|
|
1546 | } |
|
|
1547 | while (b > a); |
|
|
1548 | |
|
|
1549 | /* next bit, or stop, if no bits left in this path */ |
|
|
1550 | if (!*--bit) |
|
|
1551 | break; |
|
|
1552 | |
|
|
1553 | base_stk [stk_idx] = a; |
|
|
1554 | end_stk [stk_idx] = end; |
|
|
1555 | bit_stk [stk_idx] = bit; |
|
|
1556 | ++stk_idx; |
|
|
1557 | |
|
|
1558 | end = a; |
|
|
1559 | } |
|
|
1560 | } |
|
|
1561 | while (stk_idx--); |
|
|
1562 | } |
|
|
1563 | } |
|
|
1564 | |
|
|
1565 | static void |
|
|
1566 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1567 | { |
|
|
1568 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1569 | { |
|
|
1570 | int i; |
|
|
1571 | eio_dirent *min = dents; |
|
|
1572 | |
|
|
1573 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1574 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1575 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1576 | min = &dents [i]; |
|
|
1577 | |
|
|
1578 | /* swap elements 0 and j (minimum) */ |
|
|
1579 | { |
|
|
1580 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1581 | } |
|
|
1582 | } |
|
|
1583 | |
|
|
1584 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1585 | { |
|
|
1586 | eio_dirent *i, *j; |
|
|
1587 | |
|
|
1588 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1589 | { |
|
|
1590 | eio_dirent value = *i; |
|
|
1591 | |
|
|
1592 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1593 | j [1] = j [0]; |
|
|
1594 | |
|
|
1595 | j [1] = value; |
|
|
1596 | } |
|
|
1597 | } |
|
|
1598 | } |
|
|
1599 | |
|
|
1600 | static void |
|
|
1601 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1602 | { |
|
|
1603 | if (size <= 1) |
|
|
1604 | return; /* our insertion sort relies on size > 0 */ |
|
|
1605 | |
|
|
1606 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
|
|
1607 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1608 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
|
|
1609 | |
|
|
1610 | /* use an insertion sort at the end, or for small arrays, */ |
|
|
1611 | /* as insertion sort is more efficient for small partitions */ |
|
|
1612 | eio_dent_insertion_sort (dents, size); |
|
|
1613 | } |
|
|
1614 | |
962 | /* read a full directory */ |
1615 | /* read a full directory */ |
963 | static void |
1616 | static void |
964 | eio__scandir (eio_req *req, etp_worker *self) |
1617 | eio__scandir (eio_req *req, etp_worker *self) |
965 | { |
1618 | { |
966 | DIR *dirp; |
1619 | DIR *dirp; |
967 | EIO_STRUCT_DIRENT *entp; |
1620 | EIO_STRUCT_DIRENT *entp; |
968 | char *name, *names; |
1621 | char *name, *names; |
969 | int memlen = 4096; |
1622 | int namesalloc = 4096; |
970 | int memofs = 0; |
1623 | int namesoffs = 0; |
|
|
1624 | int flags = req->int1; |
|
|
1625 | eio_dirent *dents = 0; |
|
|
1626 | int dentalloc = 128; |
971 | int res = 0; |
1627 | int dentoffs = 0; |
|
|
1628 | ino_t inode_bits = 0; |
|
|
1629 | |
|
|
1630 | req->result = -1; |
|
|
1631 | |
|
|
1632 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1633 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
972 | |
1634 | |
973 | X_LOCK (wrklock); |
1635 | X_LOCK (wrklock); |
974 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1636 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
975 | self->dirp = dirp = opendir (req->ptr1); |
1637 | self->dirp = dirp = opendir (req->ptr1); |
|
|
1638 | |
976 | req->flags |= EIO_FLAG_PTR2_FREE; |
1639 | if (req->flags & EIO_FLAG_PTR1_FREE) |
|
|
1640 | free (req->ptr1); |
|
|
1641 | |
|
|
1642 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
|
|
1643 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
977 | req->ptr2 = names = malloc (memlen); |
1644 | req->ptr2 = names = malloc (namesalloc); |
978 | X_UNLOCK (wrklock); |
1645 | X_UNLOCK (wrklock); |
979 | |
1646 | |
980 | if (dirp && names) |
1647 | if (dirp && names && (!flags || dents)) |
981 | for (;;) |
1648 | for (;;) |
982 | { |
1649 | { |
983 | errno = 0; |
1650 | errno = 0; |
984 | entp = readdir (dirp); |
1651 | entp = readdir (dirp); |
985 | |
1652 | |
986 | if (!entp) |
1653 | if (!entp) |
|
|
1654 | { |
|
|
1655 | if (errno) |
|
|
1656 | break; |
|
|
1657 | |
|
|
1658 | /* sort etc. */ |
|
|
1659 | req->int1 = flags; |
|
|
1660 | req->result = dentoffs; |
|
|
1661 | |
|
|
1662 | if (flags & EIO_READDIR_STAT_ORDER) |
|
|
1663 | eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); |
|
|
1664 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1665 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1666 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1667 | else |
|
|
1668 | { |
|
|
1669 | /* in this case, all is known, and we just put dirs first and sort them */ |
|
|
1670 | eio_dirent *oth = dents + dentoffs; |
|
|
1671 | eio_dirent *dir = dents; |
|
|
1672 | |
|
|
1673 | /* now partition dirs to the front, and non-dirs to the back */ |
|
|
1674 | /* by walking from both sides and swapping if necessary */ |
|
|
1675 | while (oth > dir) |
|
|
1676 | { |
|
|
1677 | if (dir->type == EIO_DT_DIR) |
|
|
1678 | ++dir; |
|
|
1679 | else if ((--oth)->type == EIO_DT_DIR) |
|
|
1680 | { |
|
|
1681 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
|
|
1682 | |
|
|
1683 | ++dir; |
|
|
1684 | } |
|
|
1685 | } |
|
|
1686 | |
|
|
1687 | /* now sort the dirs only (dirs all have the same score) */ |
|
|
1688 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
|
|
1689 | } |
|
|
1690 | |
987 | break; |
1691 | break; |
|
|
1692 | } |
988 | |
1693 | |
|
|
1694 | /* now add the entry to our list(s) */ |
989 | name = entp->d_name; |
1695 | name = entp->d_name; |
990 | |
1696 | |
|
|
1697 | /* skip . and .. entries */ |
991 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1698 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
992 | { |
1699 | { |
993 | int len = strlen (name) + 1; |
1700 | int len = D_NAMLEN (entp) + 1; |
994 | |
1701 | |
995 | res++; |
1702 | while (ecb_expect_false (namesoffs + len > namesalloc)) |
996 | |
|
|
997 | while (memofs + len > memlen) |
|
|
998 | { |
1703 | { |
999 | memlen *= 2; |
1704 | namesalloc *= 2; |
1000 | X_LOCK (wrklock); |
1705 | X_LOCK (wrklock); |
1001 | req->ptr2 = names = realloc (names, memlen); |
1706 | req->ptr2 = names = realloc (names, namesalloc); |
1002 | X_UNLOCK (wrklock); |
1707 | X_UNLOCK (wrklock); |
1003 | |
1708 | |
1004 | if (!names) |
1709 | if (!names) |
1005 | break; |
1710 | break; |
1006 | } |
1711 | } |
1007 | |
1712 | |
1008 | memcpy (names + memofs, name, len); |
1713 | memcpy (names + namesoffs, name, len); |
|
|
1714 | |
|
|
1715 | if (dents) |
|
|
1716 | { |
|
|
1717 | struct eio_dirent *ent; |
|
|
1718 | |
|
|
1719 | if (ecb_expect_false (dentoffs == dentalloc)) |
|
|
1720 | { |
|
|
1721 | dentalloc *= 2; |
|
|
1722 | X_LOCK (wrklock); |
|
|
1723 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
|
|
1724 | X_UNLOCK (wrklock); |
|
|
1725 | |
|
|
1726 | if (!dents) |
|
|
1727 | break; |
|
|
1728 | } |
|
|
1729 | |
|
|
1730 | ent = dents + dentoffs; |
|
|
1731 | |
|
|
1732 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
|
|
1733 | ent->namelen = len - 1; |
|
|
1734 | ent->inode = D_INO (entp); |
|
|
1735 | |
|
|
1736 | inode_bits |= ent->inode; |
|
|
1737 | |
|
|
1738 | switch (D_TYPE (entp)) |
|
|
1739 | { |
|
|
1740 | default: |
|
|
1741 | ent->type = EIO_DT_UNKNOWN; |
|
|
1742 | flags |= EIO_READDIR_FOUND_UNKNOWN; |
|
|
1743 | break; |
|
|
1744 | |
|
|
1745 | #ifdef DT_FIFO |
|
|
1746 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
|
|
1747 | #endif |
|
|
1748 | #ifdef DT_CHR |
|
|
1749 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
|
|
1750 | #endif |
|
|
1751 | #ifdef DT_MPC |
|
|
1752 | case DT_MPC: ent->type = EIO_DT_MPC; break; |
|
|
1753 | #endif |
|
|
1754 | #ifdef DT_DIR |
|
|
1755 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
|
|
1756 | #endif |
|
|
1757 | #ifdef DT_NAM |
|
|
1758 | case DT_NAM: ent->type = EIO_DT_NAM; break; |
|
|
1759 | #endif |
|
|
1760 | #ifdef DT_BLK |
|
|
1761 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
|
|
1762 | #endif |
|
|
1763 | #ifdef DT_MPB |
|
|
1764 | case DT_MPB: ent->type = EIO_DT_MPB; break; |
|
|
1765 | #endif |
|
|
1766 | #ifdef DT_REG |
|
|
1767 | case DT_REG: ent->type = EIO_DT_REG; break; |
|
|
1768 | #endif |
|
|
1769 | #ifdef DT_NWK |
|
|
1770 | case DT_NWK: ent->type = EIO_DT_NWK; break; |
|
|
1771 | #endif |
|
|
1772 | #ifdef DT_CMP |
|
|
1773 | case DT_CMP: ent->type = EIO_DT_CMP; break; |
|
|
1774 | #endif |
|
|
1775 | #ifdef DT_LNK |
|
|
1776 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
|
|
1777 | #endif |
|
|
1778 | #ifdef DT_SOCK |
|
|
1779 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
|
|
1780 | #endif |
|
|
1781 | #ifdef DT_DOOR |
|
|
1782 | case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
|
|
1783 | #endif |
|
|
1784 | #ifdef DT_WHT |
|
|
1785 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
|
|
1786 | #endif |
|
|
1787 | } |
|
|
1788 | |
|
|
1789 | ent->score = 7; |
|
|
1790 | |
|
|
1791 | if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1792 | { |
|
|
1793 | if (ent->type == EIO_DT_UNKNOWN) |
|
|
1794 | { |
|
|
1795 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
|
|
1796 | ent->score = 1; |
|
|
1797 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
|
|
1798 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
|
|
1799 | } |
|
|
1800 | else if (ent->type == EIO_DT_DIR) |
|
|
1801 | ent->score = 0; |
|
|
1802 | } |
|
|
1803 | } |
|
|
1804 | |
1009 | memofs += len; |
1805 | namesoffs += len; |
|
|
1806 | ++dentoffs; |
|
|
1807 | } |
|
|
1808 | |
|
|
1809 | if (EIO_CANCELLED (req)) |
|
|
1810 | { |
|
|
1811 | errno = ECANCELED; |
|
|
1812 | break; |
1010 | } |
1813 | } |
1011 | } |
1814 | } |
1012 | |
|
|
1013 | if (errno) |
|
|
1014 | res = -1; |
|
|
1015 | |
|
|
1016 | req->result = res; |
|
|
1017 | } |
|
|
1018 | |
|
|
1019 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
|
|
1020 | # undef msync |
|
|
1021 | # define msync(a,b,c) ENOSYS |
|
|
1022 | #endif |
|
|
1023 | |
|
|
1024 | int |
|
|
1025 | eio__mtouch (void *mem, size_t len, int flags) |
|
|
1026 | { |
|
|
1027 | intptr_t addr = (intptr_t)mem; |
|
|
1028 | intptr_t end = addr + len; |
|
|
1029 | #ifdef PAGESIZE |
|
|
1030 | const intptr_t page = PAGESIZE; |
|
|
1031 | #else |
|
|
1032 | static intptr_t page; |
|
|
1033 | |
|
|
1034 | if (!page) |
|
|
1035 | page = sysconf (_SC_PAGESIZE); |
|
|
1036 | #endif |
|
|
1037 | |
|
|
1038 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1039 | |
|
|
1040 | if (addr < end) |
|
|
1041 | if (flags) /* modify */ |
|
|
1042 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
|
|
1043 | else |
|
|
1044 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1045 | |
|
|
1046 | return 0; |
|
|
1047 | } |
1815 | } |
1048 | |
1816 | |
1049 | /*****************************************************************************/ |
1817 | /*****************************************************************************/ |
1050 | |
1818 | |
1051 | #define ALLOC(len) \ |
1819 | #define ALLOC(len) \ |
… | |
… | |
1066 | X_THREAD_PROC (etp_proc) |
1834 | X_THREAD_PROC (etp_proc) |
1067 | { |
1835 | { |
1068 | ETP_REQ *req; |
1836 | ETP_REQ *req; |
1069 | struct timespec ts; |
1837 | struct timespec ts; |
1070 | etp_worker *self = (etp_worker *)thr_arg; |
1838 | etp_worker *self = (etp_worker *)thr_arg; |
|
|
1839 | int timeout; |
1071 | |
1840 | |
1072 | /* try to distribute timeouts somewhat randomly */ |
1841 | /* try to distribute timeouts somewhat evenly */ |
1073 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1842 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1074 | |
1843 | |
1075 | for (;;) |
1844 | for (;;) |
1076 | { |
1845 | { |
|
|
1846 | ts.tv_sec = 0; |
|
|
1847 | |
1077 | X_LOCK (reqlock); |
1848 | X_LOCK (reqlock); |
1078 | |
1849 | |
1079 | for (;;) |
1850 | for (;;) |
1080 | { |
1851 | { |
1081 | self->req = req = reqq_shift (&req_queue); |
1852 | self->req = req = reqq_shift (&req_queue); |
1082 | |
1853 | |
1083 | if (req) |
1854 | if (req) |
1084 | break; |
1855 | break; |
1085 | |
1856 | |
|
|
1857 | if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ |
|
|
1858 | { |
|
|
1859 | X_UNLOCK (reqlock); |
|
|
1860 | X_LOCK (wrklock); |
|
|
1861 | --started; |
|
|
1862 | X_UNLOCK (wrklock); |
|
|
1863 | goto quit; |
|
|
1864 | } |
|
|
1865 | |
1086 | ++idle; |
1866 | ++idle; |
1087 | |
1867 | |
1088 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
1868 | if (idle <= max_idle) |
1089 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1869 | /* we are allowed to idle, so do so without any timeout */ |
|
|
1870 | X_COND_WAIT (reqwait, reqlock); |
|
|
1871 | else |
1090 | { |
1872 | { |
1091 | if (idle > max_idle) |
1873 | /* initialise timeout once */ |
1092 | { |
1874 | if (!ts.tv_sec) |
1093 | --idle; |
1875 | ts.tv_sec = time (0) + idle_timeout; |
1094 | X_UNLOCK (reqlock); |
|
|
1095 | X_LOCK (wrklock); |
|
|
1096 | --started; |
|
|
1097 | X_UNLOCK (wrklock); |
|
|
1098 | goto quit; |
|
|
1099 | } |
|
|
1100 | |
1876 | |
1101 | /* we are allowed to idle, so do so without any timeout */ |
|
|
1102 | X_COND_WAIT (reqwait, reqlock); |
1877 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
|
|
1878 | ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */ |
1103 | } |
1879 | } |
1104 | |
1880 | |
1105 | --idle; |
1881 | --idle; |
1106 | } |
1882 | } |
1107 | |
1883 | |
… | |
… | |
1110 | X_UNLOCK (reqlock); |
1886 | X_UNLOCK (reqlock); |
1111 | |
1887 | |
1112 | if (req->type < 0) |
1888 | if (req->type < 0) |
1113 | goto quit; |
1889 | goto quit; |
1114 | |
1890 | |
1115 | if (!EIO_CANCELLED (req)) |
|
|
1116 | ETP_EXECUTE (self, req); |
1891 | ETP_EXECUTE (self, req); |
1117 | |
1892 | |
1118 | X_LOCK (reslock); |
1893 | X_LOCK (reslock); |
1119 | |
1894 | |
1120 | ++npending; |
1895 | ++npending; |
1121 | |
1896 | |
… | |
… | |
1136 | return 0; |
1911 | return 0; |
1137 | } |
1912 | } |
1138 | |
1913 | |
1139 | /*****************************************************************************/ |
1914 | /*****************************************************************************/ |
1140 | |
1915 | |
|
|
1916 | int ecb_cold |
1141 | int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1917 | eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1142 | { |
1918 | { |
1143 | return etp_init (want_poll, done_poll); |
1919 | return etp_init (want_poll, done_poll); |
1144 | } |
1920 | } |
1145 | |
1921 | |
|
|
1922 | ecb_inline void |
1146 | static void eio_api_destroy (eio_req *req) |
1923 | eio_api_destroy (eio_req *req) |
1147 | { |
1924 | { |
1148 | free (req); |
1925 | free (req); |
1149 | } |
1926 | } |
1150 | |
1927 | |
1151 | #define REQ(rtype) \ |
1928 | #define REQ(rtype) \ |
… | |
… | |
1170 | { \ |
1947 | { \ |
1171 | eio_api_destroy (req); \ |
1948 | eio_api_destroy (req); \ |
1172 | return 0; \ |
1949 | return 0; \ |
1173 | } |
1950 | } |
1174 | |
1951 | |
|
|
1952 | static void |
1175 | static void eio_execute (etp_worker *self, eio_req *req) |
1953 | eio_execute (etp_worker *self, eio_req *req) |
1176 | { |
1954 | { |
1177 | errno = 0; |
1955 | if (ecb_expect_false (EIO_CANCELLED (req))) |
|
|
1956 | { |
|
|
1957 | req->result = -1; |
|
|
1958 | req->errorno = ECANCELED; |
|
|
1959 | return; |
|
|
1960 | } |
1178 | |
1961 | |
1179 | switch (req->type) |
1962 | switch (req->type) |
1180 | { |
1963 | { |
1181 | case EIO_READ: ALLOC (req->size); |
1964 | case EIO_READ: ALLOC (req->size); |
1182 | req->result = req->offs >= 0 |
1965 | req->result = req->offs >= 0 |
… | |
… | |
1194 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1977 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1195 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1978 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1196 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1979 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1197 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1980 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1198 | |
1981 | |
|
|
1982 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1983 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1984 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1985 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1986 | |
1199 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1987 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1200 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1988 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1201 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1989 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1202 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1990 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1203 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1991 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1210 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1998 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1211 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1999 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1212 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
2000 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1213 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
2001 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1214 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
2002 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1215 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
2003 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
1216 | |
2004 | |
|
|
2005 | case EIO_REALPATH: eio__realpath (req, self); break; |
|
|
2006 | |
1217 | case EIO_READLINK: ALLOC (NAME_MAX); |
2007 | case EIO_READLINK: ALLOC (PATH_MAX); |
1218 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
2008 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1219 | |
2009 | |
1220 | case EIO_SYNC: req->result = 0; sync (); break; |
2010 | case EIO_SYNC: req->result = 0; sync (); break; |
1221 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
2011 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1222 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
2012 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1223 | case EIO_MSYNC: req->result = msync (req->ptr2, req->size, req->int1); break; |
2013 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
|
|
2014 | case EIO_MTOUCH: req->result = eio__mtouch (req); break; |
1224 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
2015 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
|
|
2016 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
1225 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
2017 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
|
|
2018 | case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; |
1226 | |
2019 | |
1227 | case EIO_READDIR: eio__scandir (req, self); break; |
2020 | case EIO_READDIR: eio__scandir (req, self); break; |
1228 | |
2021 | |
1229 | case EIO_BUSY: |
2022 | case EIO_BUSY: |
1230 | #ifdef _WIN32 |
2023 | #ifdef _WIN32 |
1231 | Sleep (req->nv1 * 1000.); |
2024 | Sleep (req->nv1 * 1e3); |
1232 | #else |
2025 | #else |
1233 | { |
2026 | { |
1234 | struct timeval tv; |
2027 | struct timeval tv; |
1235 | |
2028 | |
1236 | tv.tv_sec = req->nv1; |
2029 | tv.tv_sec = req->nv1; |
1237 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
2030 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1238 | |
2031 | |
1239 | req->result = select (0, 0, 0, 0, &tv); |
2032 | req->result = select (0, 0, 0, 0, &tv); |
1240 | } |
2033 | } |
1241 | #endif |
2034 | #endif |
1242 | break; |
2035 | break; |
… | |
… | |
1257 | times = tv; |
2050 | times = tv; |
1258 | } |
2051 | } |
1259 | else |
2052 | else |
1260 | times = 0; |
2053 | times = 0; |
1261 | |
2054 | |
1262 | |
|
|
1263 | req->result = req->type == EIO_FUTIME |
2055 | req->result = req->type == EIO_FUTIME |
1264 | ? futimes (req->int1, times) |
2056 | ? futimes (req->int1, times) |
1265 | : utimes (req->ptr1, times); |
2057 | : utimes (req->ptr1, times); |
1266 | } |
2058 | } |
1267 | break; |
2059 | break; |
… | |
… | |
1272 | case EIO_NOP: |
2064 | case EIO_NOP: |
1273 | req->result = 0; |
2065 | req->result = 0; |
1274 | break; |
2066 | break; |
1275 | |
2067 | |
1276 | case EIO_CUSTOM: |
2068 | case EIO_CUSTOM: |
1277 | ((void (*)(eio_req *))req->feed) (req); |
2069 | req->feed (req); |
1278 | break; |
2070 | break; |
1279 | |
2071 | |
1280 | default: |
2072 | default: |
|
|
2073 | errno = ENOSYS; |
1281 | req->result = -1; |
2074 | req->result = -1; |
1282 | break; |
2075 | break; |
1283 | } |
2076 | } |
1284 | |
2077 | |
1285 | req->errorno = errno; |
2078 | req->errorno = errno; |
… | |
… | |
1315 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
2108 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1316 | { |
2109 | { |
1317 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
2110 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1318 | } |
2111 | } |
1319 | |
2112 | |
|
|
2113 | eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
|
|
2114 | { |
|
|
2115 | REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
|
|
2116 | } |
|
|
2117 | |
|
|
2118 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
|
|
2119 | { |
|
|
2120 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
|
|
2121 | } |
|
|
2122 | |
1320 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
2123 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1321 | { |
2124 | { |
1322 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
2125 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1323 | } |
2126 | } |
1324 | |
2127 | |
|
|
2128 | eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) |
|
|
2129 | { |
|
|
2130 | REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; |
|
|
2131 | } |
|
|
2132 | |
1325 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
2133 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1326 | { |
2134 | { |
1327 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
2135 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1328 | } |
2136 | } |
1329 | |
2137 | |
… | |
… | |
1348 | } |
2156 | } |
1349 | |
2157 | |
1350 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
2158 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1351 | { |
2159 | { |
1352 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
2160 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
|
|
2161 | } |
|
|
2162 | |
|
|
2163 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
2164 | { |
|
|
2165 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
1353 | } |
2166 | } |
1354 | |
2167 | |
1355 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
2168 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1356 | { |
2169 | { |
1357 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
2170 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
… | |
… | |
1421 | eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
2234 | eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
1422 | { |
2235 | { |
1423 | return eio__1path (EIO_READLINK, path, pri, cb, data); |
2236 | return eio__1path (EIO_READLINK, path, pri, cb, data); |
1424 | } |
2237 | } |
1425 | |
2238 | |
|
|
2239 | eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data) |
|
|
2240 | { |
|
|
2241 | return eio__1path (EIO_REALPATH, path, pri, cb, data); |
|
|
2242 | } |
|
|
2243 | |
1426 | eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
2244 | eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
1427 | { |
2245 | { |
1428 | return eio__1path (EIO_STAT, path, pri, cb, data); |
2246 | return eio__1path (EIO_STAT, path, pri, cb, data); |
1429 | } |
2247 | } |
1430 | |
2248 | |
1431 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
2249 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1432 | { |
2250 | { |
1433 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
2251 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1434 | } |
2252 | } |
1435 | |
2253 | |
|
|
2254 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
2255 | { |
|
|
2256 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
|
|
2257 | } |
|
|
2258 | |
1436 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
2259 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1437 | { |
2260 | { |
1438 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
2261 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1439 | } |
2262 | } |
1440 | |
2263 | |
1441 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
2264 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1442 | { |
2265 | { |
1443 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
2266 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1444 | } |
2267 | } |
1445 | |
2268 | |
1446 | eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) |
2269 | eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
1447 | { |
2270 | { |
1448 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
2271 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1449 | } |
2272 | } |
1450 | |
2273 | |
1451 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
2274 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1452 | { |
2275 | { |
1453 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
2276 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; |
1454 | } |
2277 | } |
1455 | |
2278 | |
1456 | static eio_req * |
2279 | static eio_req * |
1457 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2280 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1458 | { |
2281 | { |
… | |
… | |
1482 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2305 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1483 | { |
2306 | { |
1484 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
2307 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1485 | } |
2308 | } |
1486 | |
2309 | |
1487 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
2310 | eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) |
1488 | { |
2311 | { |
1489 | REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
2312 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
1490 | } |
2313 | } |
1491 | |
2314 | |
1492 | #endif |
2315 | #endif |
1493 | |
2316 | |
1494 | eio_req *eio_grp (eio_cb cb, void *data) |
2317 | eio_req *eio_grp (eio_cb cb, void *data) |
… | |
… | |
1503 | #undef SEND |
2326 | #undef SEND |
1504 | |
2327 | |
1505 | /*****************************************************************************/ |
2328 | /*****************************************************************************/ |
1506 | /* grp functions */ |
2329 | /* grp functions */ |
1507 | |
2330 | |
|
|
2331 | void |
1508 | void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
2332 | eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
1509 | { |
2333 | { |
1510 | grp->int2 = limit; |
2334 | grp->int2 = limit; |
1511 | grp->feed = feed; |
2335 | grp->feed = feed; |
1512 | |
2336 | |
1513 | grp_try_feed (grp); |
2337 | grp_try_feed (grp); |
1514 | } |
2338 | } |
1515 | |
2339 | |
|
|
2340 | void |
1516 | void eio_grp_limit (eio_req *grp, int limit) |
2341 | eio_grp_limit (eio_req *grp, int limit) |
1517 | { |
2342 | { |
1518 | grp->int2 = limit; |
2343 | grp->int2 = limit; |
1519 | |
2344 | |
1520 | grp_try_feed (grp); |
2345 | grp_try_feed (grp); |
1521 | } |
2346 | } |
1522 | |
2347 | |
|
|
2348 | void |
1523 | void eio_grp_add (eio_req *grp, eio_req *req) |
2349 | eio_grp_add (eio_req *grp, eio_req *req) |
1524 | { |
2350 | { |
1525 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
2351 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1526 | |
2352 | |
1527 | grp->flags |= EIO_FLAG_GROUPADD; |
2353 | grp->flags |= EIO_FLAG_GROUPADD; |
1528 | |
2354 | |
… | |
… | |
1539 | } |
2365 | } |
1540 | |
2366 | |
1541 | /*****************************************************************************/ |
2367 | /*****************************************************************************/ |
1542 | /* misc garbage */ |
2368 | /* misc garbage */ |
1543 | |
2369 | |
|
|
2370 | ssize_t |
1544 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2371 | eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1545 | { |
2372 | { |
1546 | etp_worker wrk; |
2373 | etp_worker wrk; |
|
|
2374 | ssize_t ret; |
1547 | |
2375 | |
1548 | wrk.dbuf = 0; |
2376 | wrk.dbuf = 0; |
1549 | |
2377 | |
1550 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
2378 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1551 | |
2379 | |
1552 | if (wrk.dbuf) |
2380 | if (wrk.dbuf) |
1553 | free (wrk.dbuf); |
2381 | free (wrk.dbuf); |
1554 | } |
|
|
1555 | |
2382 | |
|
|
2383 | return ret; |
|
|
2384 | } |
|
|
2385 | |