1 | /* |
1 | /* |
2 | * libeio implementation |
2 | * libeio implementation |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008,2009,2010 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" |
41 | |
46 | |
42 | #ifdef EIO_STACKSIZE |
47 | #ifdef EIO_STACKSIZE |
43 | # define XTHREAD_STACKSIZE EIO_STACKSIZE |
48 | # define XTHREAD_STACKSIZE EIO_STACKSIZE |
44 | #endif |
49 | #endif |
45 | #include "xthread.h" |
50 | #include "xthread.h" |
… | |
… | |
49 | #include <stdlib.h> |
54 | #include <stdlib.h> |
50 | #include <string.h> |
55 | #include <string.h> |
51 | #include <errno.h> |
56 | #include <errno.h> |
52 | #include <sys/types.h> |
57 | #include <sys/types.h> |
53 | #include <sys/stat.h> |
58 | #include <sys/stat.h> |
54 | #include <sys/statvfs.h> |
|
|
55 | #include <limits.h> |
59 | #include <limits.h> |
56 | #include <fcntl.h> |
60 | #include <fcntl.h> |
57 | #include <assert.h> |
61 | #include <assert.h> |
58 | |
62 | |
|
|
63 | /* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ |
|
|
64 | /* intptr_t only comes from stdint.h, says idiot openbsd coder */ |
|
|
65 | #if HAVE_STDINT_H |
|
|
66 | # include <stdint.h> |
|
|
67 | #endif |
|
|
68 | |
|
|
69 | #ifndef ECANCELED |
|
|
70 | # define ECANCELED EDOM |
|
|
71 | #endif |
|
|
72 | #ifndef ELOOP |
|
|
73 | # define ELOOP EDOM |
|
|
74 | #endif |
|
|
75 | |
|
|
76 | static void eio_destroy (eio_req *req); |
|
|
77 | |
59 | #ifndef EIO_FINISH |
78 | #ifndef EIO_FINISH |
60 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
79 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
61 | #endif |
80 | #endif |
62 | |
81 | |
63 | #ifndef EIO_DESTROY |
82 | #ifndef EIO_DESTROY |
… | |
… | |
66 | |
85 | |
67 | #ifndef EIO_FEED |
86 | #ifndef EIO_FEED |
68 | # define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
87 | # define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
69 | #endif |
88 | #endif |
70 | |
89 | |
|
|
90 | #ifndef EIO_FD_TO_WIN32_HANDLE |
|
|
91 | # define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd) |
|
|
92 | #endif |
|
|
93 | #ifndef EIO_WIN32_HANDLE_TO_FD |
|
|
94 | # define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0) |
|
|
95 | #endif |
|
|
96 | |
|
|
97 | #define EIO_ERRNO(errval,retval) ((errno = errval), retval) |
|
|
98 | |
|
|
99 | #define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1) |
|
|
100 | |
71 | #ifdef _WIN32 |
101 | #ifdef _WIN32 |
72 | |
102 | |
73 | /*doh*/ |
103 | #undef PAGESIZE |
|
|
104 | #define PAGESIZE 4096 /* GetSystemInfo? */ |
|
|
105 | |
|
|
106 | #ifdef EIO_STRUCT_STATI64 |
|
|
107 | #define stat(path,buf) _stati64 (path,buf) |
|
|
108 | #define fstat(fd,buf) _fstati64 (path,buf) |
|
|
109 | #endif |
|
|
110 | #define lstat(path,buf) stat (path,buf) |
|
|
111 | #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) |
|
|
112 | #define mkdir(path,mode) _mkdir (path) |
|
|
113 | #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) |
|
|
114 | |
|
|
115 | #define chmod(path,mode) _chmod (path, mode) |
|
|
116 | #define dup(fd) _dup (fd) |
|
|
117 | #define dup2(fd1,fd2) _dup2 (fd1, fd2) |
|
|
118 | |
|
|
119 | #define fchmod(fd,mode) EIO_ENOSYS () |
|
|
120 | #define chown(path,uid,gid) EIO_ENOSYS () |
|
|
121 | #define fchown(fd,uid,gid) EIO_ENOSYS () |
|
|
122 | #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ |
|
|
123 | #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */ |
|
|
124 | #define mknod(path,mode,dev) EIO_ENOSYS () |
|
|
125 | #define sync() EIO_ENOSYS () |
|
|
126 | #define readlink(path,buf,s) EIO_ENOSYS () |
|
|
127 | #define statvfs(path,buf) EIO_ENOSYS () |
|
|
128 | #define fstatvfs(fd,buf) EIO_ENOSYS () |
|
|
129 | |
|
|
130 | /* we could even stat and see if it exists */ |
|
|
131 | static int |
|
|
132 | symlink (const char *old, const char *neu) |
|
|
133 | { |
|
|
134 | #if WINVER >= 0x0600 |
|
|
135 | if (CreateSymbolicLink (neu, old, 1)) |
|
|
136 | return 0; |
|
|
137 | |
|
|
138 | if (CreateSymbolicLink (neu, old, 0)) |
|
|
139 | return 0; |
|
|
140 | #endif |
|
|
141 | |
|
|
142 | return EIO_ERRNO (ENOENT, -1); |
|
|
143 | } |
|
|
144 | |
|
|
145 | /* POSIX API only */ |
|
|
146 | #define CreateHardLink(neu,old,flags) 0 |
|
|
147 | #define CreateSymbolicLink(neu,old,flags) 0 |
|
|
148 | |
|
|
149 | struct statvfs |
|
|
150 | { |
|
|
151 | int dummy; |
|
|
152 | }; |
|
|
153 | |
|
|
154 | #define DT_DIR EIO_DT_DIR |
|
|
155 | #define DT_REG EIO_DT_REG |
|
|
156 | #define D_NAME(entp) entp.cFileName |
|
|
157 | #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG) |
|
|
158 | |
74 | #else |
159 | #else |
75 | |
160 | |
76 | # include "config.h" |
|
|
77 | # include <sys/time.h> |
161 | #include <sys/time.h> |
78 | # include <sys/select.h> |
162 | #include <sys/select.h> |
79 | # include <sys/mman.h> |
163 | #include <sys/statvfs.h> |
80 | # include <unistd.h> |
164 | #include <unistd.h> |
81 | # include <utime.h> |
165 | #include <utime.h> |
82 | # include <signal.h> |
166 | #include <signal.h> |
83 | # include <dirent.h> |
167 | #include <dirent.h> |
84 | |
168 | |
|
|
169 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
170 | #include <sys/mman.h> |
|
|
171 | #endif |
|
|
172 | |
|
|
173 | #define D_NAME(entp) entp->d_name |
|
|
174 | |
85 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
175 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
86 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
176 | #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
87 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
177 | #define _DIRENT_HAVE_D_TYPE /* sigh */ |
88 | # define D_INO(de) (de)->d_fileno |
178 | #define D_INO(de) (de)->d_fileno |
89 | # define D_NAMLEN(de) (de)->d_namlen |
179 | #define D_NAMLEN(de) (de)->d_namlen |
90 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
180 | #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
91 | # define D_INO(de) (de)->d_ino |
181 | #define D_INO(de) (de)->d_ino |
92 | # endif |
182 | #endif |
93 | |
183 | |
94 | #ifdef _D_EXACT_NAMLEN |
184 | #ifdef _D_EXACT_NAMLEN |
95 | # undef D_NAMLEN |
185 | #undef D_NAMLEN |
96 | # define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
186 | #define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
97 | #endif |
187 | #endif |
98 | |
188 | |
99 | # ifdef _DIRENT_HAVE_D_TYPE |
189 | #ifdef _DIRENT_HAVE_D_TYPE |
100 | # define D_TYPE(de) (de)->d_type |
190 | #define D_TYPE(de) (de)->d_type |
101 | # endif |
191 | #endif |
102 | |
192 | |
103 | # ifndef EIO_STRUCT_DIRENT |
193 | #ifndef EIO_STRUCT_DIRENT |
104 | # define EIO_STRUCT_DIRENT struct dirent |
194 | #define EIO_STRUCT_DIRENT struct dirent |
105 | # endif |
195 | #endif |
106 | |
196 | |
107 | #endif |
197 | #endif |
108 | |
198 | |
109 | #if HAVE_SENDFILE |
199 | #if HAVE_SENDFILE |
110 | # if __linux |
200 | # if __linux |
… | |
… | |
126 | #endif |
216 | #endif |
127 | #ifndef D_INO |
217 | #ifndef D_INO |
128 | # define D_INO(de) 0 |
218 | # define D_INO(de) 0 |
129 | #endif |
219 | #endif |
130 | #ifndef D_NAMLEN |
220 | #ifndef D_NAMLEN |
131 | # define D_NAMLEN(de) strlen ((de)->d_name) |
221 | # define D_NAMLEN(entp) strlen (D_NAME (entp)) |
132 | #endif |
222 | #endif |
133 | |
|
|
134 | /* number of seconds after which an idle threads exit */ |
|
|
135 | #define IDLE_TIMEOUT 10 |
|
|
136 | |
223 | |
137 | /* used for struct dirent, AIX doesn't provide it */ |
224 | /* used for struct dirent, AIX doesn't provide it */ |
138 | #ifndef NAME_MAX |
225 | #ifndef NAME_MAX |
139 | # define NAME_MAX 4096 |
226 | # define NAME_MAX 4096 |
|
|
227 | #endif |
|
|
228 | |
|
|
229 | /* used for readlink etc. */ |
|
|
230 | #ifndef PATH_MAX |
|
|
231 | # define PATH_MAX 4096 |
140 | #endif |
232 | #endif |
141 | |
233 | |
142 | /* buffer size for various temporary buffers */ |
234 | /* buffer size for various temporary buffers */ |
143 | #define EIO_BUFSIZE 65536 |
235 | #define EIO_BUFSIZE 65536 |
144 | |
236 | |
… | |
… | |
151 | if (!eio_buf) \ |
243 | if (!eio_buf) \ |
152 | return -1; |
244 | return -1; |
153 | |
245 | |
154 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
246 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
155 | |
247 | |
156 | /*****************************************************************************/ |
|
|
157 | |
|
|
158 | #if __GNUC__ >= 3 |
|
|
159 | # define expect(expr,value) __builtin_expect ((expr),(value)) |
|
|
160 | #else |
|
|
161 | # define expect(expr,value) (expr) |
|
|
162 | #endif |
|
|
163 | |
|
|
164 | #define expect_false(expr) expect ((expr) != 0, 0) |
|
|
165 | #define expect_true(expr) expect ((expr) != 0, 1) |
|
|
166 | |
|
|
167 | /*****************************************************************************/ |
|
|
168 | |
|
|
169 | #define ETP_PRI_MIN EIO_PRI_MIN |
248 | #define ETP_PRI_MIN EIO_PRI_MIN |
170 | #define ETP_PRI_MAX EIO_PRI_MAX |
249 | #define ETP_PRI_MAX EIO_PRI_MAX |
171 | |
250 | |
172 | struct etp_worker; |
251 | struct etp_worker; |
173 | |
252 | |
… | |
… | |
181 | #define ETP_WORKER_CLEAR(req) \ |
260 | #define ETP_WORKER_CLEAR(req) \ |
182 | if (wrk->dbuf) \ |
261 | if (wrk->dbuf) \ |
183 | { \ |
262 | { \ |
184 | free (wrk->dbuf); \ |
263 | free (wrk->dbuf); \ |
185 | wrk->dbuf = 0; \ |
264 | wrk->dbuf = 0; \ |
186 | } \ |
|
|
187 | \ |
|
|
188 | if (wrk->dirp) \ |
|
|
189 | { \ |
|
|
190 | closedir (wrk->dirp); \ |
|
|
191 | wrk->dirp = 0; \ |
|
|
192 | } |
265 | } |
193 | |
266 | |
194 | #define ETP_WORKER_COMMON \ |
267 | #define ETP_WORKER_COMMON \ |
195 | void *dbuf; \ |
268 | void *dbuf; |
196 | DIR *dirp; |
|
|
197 | |
269 | |
198 | /*****************************************************************************/ |
270 | /*****************************************************************************/ |
199 | |
271 | |
200 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
272 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
201 | |
273 | |
202 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
274 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
|
|
275 | ecb_inline int |
203 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
276 | tvdiff (struct timeval *tv1, struct timeval *tv2) |
204 | { |
277 | { |
205 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
278 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
206 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
279 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
207 | } |
280 | } |
208 | |
281 | |
… | |
… | |
215 | static unsigned int max_poll_reqs; /* reslock */ |
288 | static unsigned int max_poll_reqs; /* reslock */ |
216 | |
289 | |
217 | static volatile unsigned int nreqs; /* reqlock */ |
290 | static volatile unsigned int nreqs; /* reqlock */ |
218 | static volatile unsigned int nready; /* reqlock */ |
291 | static volatile unsigned int nready; /* reqlock */ |
219 | static volatile unsigned int npending; /* reqlock */ |
292 | static volatile unsigned int npending; /* reqlock */ |
220 | static volatile unsigned int max_idle = 4; |
293 | static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
|
|
294 | static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
221 | |
295 | |
222 | static mutex_t wrklock = X_MUTEX_INIT; |
296 | static xmutex_t wrklock; |
223 | static mutex_t reslock = X_MUTEX_INIT; |
297 | static xmutex_t reslock; |
224 | static mutex_t reqlock = X_MUTEX_INIT; |
298 | static xmutex_t reqlock; |
225 | static cond_t reqwait = X_COND_INIT; |
299 | static xcond_t reqwait; |
226 | |
300 | |
227 | #if !HAVE_PREADWRITE |
301 | #if !HAVE_PREADWRITE |
228 | /* |
302 | /* |
229 | * make our pread/pwrite emulation safe against themselves, but not against |
303 | * make our pread/pwrite emulation safe against themselves, but not against |
230 | * normal read/write by using a mutex. slows down execution a lot, |
304 | * normal read/write by using a mutex. slows down execution a lot, |
231 | * but that's your problem, not mine. |
305 | * but that's your problem, not mine. |
232 | */ |
306 | */ |
233 | static mutex_t preadwritelock = X_MUTEX_INIT; |
307 | static xmutex_t preadwritelock; |
234 | #endif |
308 | #endif |
235 | |
309 | |
236 | typedef struct etp_worker |
310 | typedef struct etp_worker |
237 | { |
311 | { |
238 | /* locked by wrklock */ |
312 | /* locked by wrklock */ |
239 | struct etp_worker *prev, *next; |
313 | struct etp_worker *prev, *next; |
240 | |
314 | |
241 | thread_t tid; |
315 | xthread_t tid; |
242 | |
316 | |
243 | /* locked by reslock, reqlock or wrklock */ |
317 | /* locked by reslock, reqlock or wrklock */ |
244 | ETP_REQ *req; /* currently processed request */ |
318 | ETP_REQ *req; /* currently processed request */ |
245 | |
319 | |
246 | ETP_WORKER_COMMON |
320 | ETP_WORKER_COMMON |
247 | } etp_worker; |
321 | } etp_worker; |
248 | |
322 | |
249 | static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ |
323 | static etp_worker wrk_first; /* NOT etp */ |
250 | |
324 | |
251 | #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
325 | #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
252 | #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
326 | #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
253 | |
327 | |
254 | /* worker threads management */ |
328 | /* worker threads management */ |
255 | |
329 | |
|
|
330 | static void ecb_cold |
256 | static void etp_worker_clear (etp_worker *wrk) |
331 | etp_worker_clear (etp_worker *wrk) |
257 | { |
332 | { |
258 | ETP_WORKER_CLEAR (wrk); |
333 | ETP_WORKER_CLEAR (wrk); |
259 | } |
334 | } |
260 | |
335 | |
|
|
336 | static void ecb_cold |
261 | static void etp_worker_free (etp_worker *wrk) |
337 | etp_worker_free (etp_worker *wrk) |
262 | { |
338 | { |
263 | wrk->next->prev = wrk->prev; |
339 | wrk->next->prev = wrk->prev; |
264 | wrk->prev->next = wrk->next; |
340 | wrk->prev->next = wrk->next; |
265 | |
341 | |
266 | free (wrk); |
342 | free (wrk); |
267 | } |
343 | } |
268 | |
344 | |
269 | static unsigned int etp_nreqs (void) |
345 | static unsigned int |
|
|
346 | etp_nreqs (void) |
270 | { |
347 | { |
271 | int retval; |
348 | int retval; |
272 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
349 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
273 | retval = nreqs; |
350 | retval = nreqs; |
274 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
351 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
275 | return retval; |
352 | return retval; |
276 | } |
353 | } |
277 | |
354 | |
278 | static unsigned int etp_nready (void) |
355 | static unsigned int |
|
|
356 | etp_nready (void) |
279 | { |
357 | { |
280 | unsigned int retval; |
358 | unsigned int retval; |
281 | |
359 | |
282 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
360 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
283 | retval = nready; |
361 | retval = nready; |
284 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
362 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
285 | |
363 | |
286 | return retval; |
364 | return retval; |
287 | } |
365 | } |
288 | |
366 | |
289 | static unsigned int etp_npending (void) |
367 | static unsigned int |
|
|
368 | etp_npending (void) |
290 | { |
369 | { |
291 | unsigned int retval; |
370 | unsigned int retval; |
292 | |
371 | |
293 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
372 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
294 | retval = npending; |
373 | retval = npending; |
295 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
374 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
296 | |
375 | |
297 | return retval; |
376 | return retval; |
298 | } |
377 | } |
299 | |
378 | |
300 | static unsigned int etp_nthreads (void) |
379 | static unsigned int |
|
|
380 | etp_nthreads (void) |
301 | { |
381 | { |
302 | unsigned int retval; |
382 | unsigned int retval; |
303 | |
383 | |
304 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
384 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
305 | retval = started; |
385 | retval = started; |
… | |
… | |
319 | } etp_reqq; |
399 | } etp_reqq; |
320 | |
400 | |
321 | static etp_reqq req_queue; |
401 | static etp_reqq req_queue; |
322 | static etp_reqq res_queue; |
402 | static etp_reqq res_queue; |
323 | |
403 | |
|
|
404 | static void ecb_noinline ecb_cold |
|
|
405 | reqq_init (etp_reqq *q) |
|
|
406 | { |
|
|
407 | int pri; |
|
|
408 | |
|
|
409 | for (pri = 0; pri < ETP_NUM_PRI; ++pri) |
|
|
410 | q->qs[pri] = q->qe[pri] = 0; |
|
|
411 | |
|
|
412 | q->size = 0; |
|
|
413 | } |
|
|
414 | |
|
|
415 | static int ecb_noinline |
324 | static int reqq_push (etp_reqq *q, ETP_REQ *req) |
416 | reqq_push (etp_reqq *q, ETP_REQ *req) |
325 | { |
417 | { |
326 | int pri = req->pri; |
418 | int pri = req->pri; |
327 | req->next = 0; |
419 | req->next = 0; |
328 | |
420 | |
329 | if (q->qe[pri]) |
421 | if (q->qe[pri]) |
… | |
… | |
335 | q->qe[pri] = q->qs[pri] = req; |
427 | q->qe[pri] = q->qs[pri] = req; |
336 | |
428 | |
337 | return q->size++; |
429 | return q->size++; |
338 | } |
430 | } |
339 | |
431 | |
|
|
432 | static ETP_REQ * ecb_noinline |
340 | static ETP_REQ *reqq_shift (etp_reqq *q) |
433 | reqq_shift (etp_reqq *q) |
341 | { |
434 | { |
342 | int pri; |
435 | int pri; |
343 | |
436 | |
344 | if (!q->size) |
437 | if (!q->size) |
345 | return 0; |
438 | return 0; |
… | |
… | |
360 | } |
453 | } |
361 | |
454 | |
362 | abort (); |
455 | abort (); |
363 | } |
456 | } |
364 | |
457 | |
365 | static void etp_atfork_prepare (void) |
458 | static int ecb_cold |
|
|
459 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
366 | { |
460 | { |
367 | X_LOCK (wrklock); |
461 | X_MUTEX_CREATE (wrklock); |
368 | X_LOCK (reqlock); |
462 | X_MUTEX_CREATE (reslock); |
369 | X_LOCK (reslock); |
463 | X_MUTEX_CREATE (reqlock); |
370 | #if !HAVE_PREADWRITE |
464 | X_COND_CREATE (reqwait); |
371 | X_LOCK (preadwritelock); |
|
|
372 | #endif |
|
|
373 | } |
|
|
374 | |
465 | |
375 | static void etp_atfork_parent (void) |
466 | reqq_init (&req_queue); |
376 | { |
467 | reqq_init (&res_queue); |
377 | #if !HAVE_PREADWRITE |
|
|
378 | X_UNLOCK (preadwritelock); |
|
|
379 | #endif |
|
|
380 | X_UNLOCK (reslock); |
|
|
381 | X_UNLOCK (reqlock); |
|
|
382 | X_UNLOCK (wrklock); |
|
|
383 | } |
|
|
384 | |
468 | |
385 | static void etp_atfork_child (void) |
469 | wrk_first.next = |
386 | { |
470 | wrk_first.prev = &wrk_first; |
387 | ETP_REQ *prv; |
|
|
388 | |
|
|
389 | while ((prv = reqq_shift (&req_queue))) |
|
|
390 | ETP_DESTROY (prv); |
|
|
391 | |
|
|
392 | while ((prv = reqq_shift (&res_queue))) |
|
|
393 | ETP_DESTROY (prv); |
|
|
394 | |
|
|
395 | while (wrk_first.next != &wrk_first) |
|
|
396 | { |
|
|
397 | etp_worker *wrk = wrk_first.next; |
|
|
398 | |
|
|
399 | if (wrk->req) |
|
|
400 | ETP_DESTROY (wrk->req); |
|
|
401 | |
|
|
402 | etp_worker_clear (wrk); |
|
|
403 | etp_worker_free (wrk); |
|
|
404 | } |
|
|
405 | |
471 | |
406 | started = 0; |
472 | started = 0; |
407 | idle = 0; |
473 | idle = 0; |
408 | nreqs = 0; |
474 | nreqs = 0; |
409 | nready = 0; |
475 | nready = 0; |
410 | npending = 0; |
476 | npending = 0; |
411 | |
477 | |
412 | etp_atfork_parent (); |
|
|
413 | } |
|
|
414 | |
|
|
415 | static void |
|
|
416 | etp_once_init (void) |
|
|
417 | { |
|
|
418 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
|
|
419 | } |
|
|
420 | |
|
|
421 | static int |
|
|
422 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
|
|
423 | { |
|
|
424 | static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
|
|
425 | |
|
|
426 | pthread_once (&doinit, etp_once_init); |
|
|
427 | |
|
|
428 | want_poll_cb = want_poll; |
478 | want_poll_cb = want_poll; |
429 | done_poll_cb = done_poll; |
479 | done_poll_cb = done_poll; |
430 | |
480 | |
431 | return 0; |
481 | return 0; |
432 | } |
482 | } |
433 | |
483 | |
434 | X_THREAD_PROC (etp_proc); |
484 | X_THREAD_PROC (etp_proc); |
435 | |
485 | |
|
|
486 | static void ecb_cold |
436 | static void etp_start_thread (void) |
487 | etp_start_thread (void) |
437 | { |
488 | { |
438 | etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
489 | etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
439 | |
490 | |
440 | /*TODO*/ |
491 | /*TODO*/ |
441 | assert (("unable to allocate worker thread data", wrk)); |
492 | assert (("unable to allocate worker thread data", wrk)); |
… | |
… | |
454 | free (wrk); |
505 | free (wrk); |
455 | |
506 | |
456 | X_UNLOCK (wrklock); |
507 | X_UNLOCK (wrklock); |
457 | } |
508 | } |
458 | |
509 | |
|
|
510 | static void |
459 | static void etp_maybe_start_thread (void) |
511 | etp_maybe_start_thread (void) |
460 | { |
512 | { |
461 | if (expect_true (etp_nthreads () >= wanted)) |
513 | if (ecb_expect_true (etp_nthreads () >= wanted)) |
462 | return; |
514 | return; |
463 | |
515 | |
464 | /* todo: maybe use idle here, but might be less exact */ |
516 | /* todo: maybe use idle here, but might be less exact */ |
465 | if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
517 | if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
466 | return; |
518 | return; |
467 | |
519 | |
468 | etp_start_thread (); |
520 | etp_start_thread (); |
469 | } |
521 | } |
470 | |
522 | |
|
|
523 | static void ecb_cold |
471 | static void etp_end_thread (void) |
524 | etp_end_thread (void) |
472 | { |
525 | { |
473 | eio_req *req = calloc (1, sizeof (eio_req)); |
526 | eio_req *req = calloc (1, sizeof (eio_req)); |
474 | |
527 | |
475 | req->type = -1; |
528 | req->type = -1; |
476 | req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
529 | req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
… | |
… | |
483 | X_LOCK (wrklock); |
536 | X_LOCK (wrklock); |
484 | --started; |
537 | --started; |
485 | X_UNLOCK (wrklock); |
538 | X_UNLOCK (wrklock); |
486 | } |
539 | } |
487 | |
540 | |
488 | static int etp_poll (void) |
541 | static int |
|
|
542 | etp_poll (void) |
489 | { |
543 | { |
490 | unsigned int maxreqs; |
544 | unsigned int maxreqs; |
491 | unsigned int maxtime; |
545 | unsigned int maxtime; |
492 | struct timeval tv_start, tv_now; |
546 | struct timeval tv_start, tv_now; |
493 | |
547 | |
… | |
… | |
523 | |
577 | |
524 | X_LOCK (reqlock); |
578 | X_LOCK (reqlock); |
525 | --nreqs; |
579 | --nreqs; |
526 | X_UNLOCK (reqlock); |
580 | X_UNLOCK (reqlock); |
527 | |
581 | |
528 | if (expect_false (req->type == EIO_GROUP && req->size)) |
582 | if (ecb_expect_false (req->type == EIO_GROUP && req->size)) |
529 | { |
583 | { |
530 | req->int1 = 1; /* mark request as delayed */ |
584 | req->int1 = 1; /* mark request as delayed */ |
531 | continue; |
585 | continue; |
532 | } |
586 | } |
533 | else |
587 | else |
534 | { |
588 | { |
535 | int res = ETP_FINISH (req); |
589 | int res = ETP_FINISH (req); |
536 | if (expect_false (res)) |
590 | if (ecb_expect_false (res)) |
537 | return res; |
591 | return res; |
538 | } |
592 | } |
539 | |
593 | |
540 | if (expect_false (maxreqs && !--maxreqs)) |
594 | if (ecb_expect_false (maxreqs && !--maxreqs)) |
541 | break; |
595 | break; |
542 | |
596 | |
543 | if (maxtime) |
597 | if (maxtime) |
544 | { |
598 | { |
545 | gettimeofday (&tv_now, 0); |
599 | gettimeofday (&tv_now, 0); |
… | |
… | |
551 | |
605 | |
552 | errno = EAGAIN; |
606 | errno = EAGAIN; |
553 | return -1; |
607 | return -1; |
554 | } |
608 | } |
555 | |
609 | |
|
|
610 | static void |
556 | static void etp_cancel (ETP_REQ *req) |
611 | etp_cancel (ETP_REQ *req) |
557 | { |
612 | { |
558 | X_LOCK (wrklock); |
613 | req->cancelled = 1; |
559 | req->flags |= EIO_FLAG_CANCELLED; |
|
|
560 | X_UNLOCK (wrklock); |
|
|
561 | |
614 | |
562 | eio_grp_cancel (req); |
615 | eio_grp_cancel (req); |
563 | } |
616 | } |
564 | |
617 | |
|
|
618 | static void |
565 | static void etp_submit (ETP_REQ *req) |
619 | etp_submit (ETP_REQ *req) |
566 | { |
620 | { |
567 | req->pri -= ETP_PRI_MIN; |
621 | req->pri -= ETP_PRI_MIN; |
568 | |
622 | |
569 | if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
623 | if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
570 | if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
624 | if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
571 | |
625 | |
572 | if (expect_false (req->type == EIO_GROUP)) |
626 | if (ecb_expect_false (req->type == EIO_GROUP)) |
573 | { |
627 | { |
574 | /* I hope this is worth it :/ */ |
628 | /* I hope this is worth it :/ */ |
575 | X_LOCK (reqlock); |
629 | X_LOCK (reqlock); |
576 | ++nreqs; |
630 | ++nreqs; |
577 | X_UNLOCK (reqlock); |
631 | X_UNLOCK (reqlock); |
… | |
… | |
596 | |
650 | |
597 | etp_maybe_start_thread (); |
651 | etp_maybe_start_thread (); |
598 | } |
652 | } |
599 | } |
653 | } |
600 | |
654 | |
|
|
655 | static void ecb_cold |
601 | static void etp_set_max_poll_time (double nseconds) |
656 | etp_set_max_poll_time (double nseconds) |
602 | { |
657 | { |
603 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
658 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
604 | max_poll_time = nseconds * EIO_TICKS; |
659 | max_poll_time = nseconds * EIO_TICKS; |
605 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
660 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
606 | } |
661 | } |
607 | |
662 | |
|
|
663 | static void ecb_cold |
608 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
664 | etp_set_max_poll_reqs (unsigned int maxreqs) |
609 | { |
665 | { |
610 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
666 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
611 | max_poll_reqs = maxreqs; |
667 | max_poll_reqs = maxreqs; |
612 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
668 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
613 | } |
669 | } |
614 | |
670 | |
|
|
671 | static void ecb_cold |
615 | static void etp_set_max_idle (unsigned int nthreads) |
672 | etp_set_max_idle (unsigned int nthreads) |
616 | { |
673 | { |
617 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
674 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
618 | max_idle = nthreads <= 0 ? 1 : nthreads; |
675 | max_idle = nthreads; |
619 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
676 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
620 | } |
677 | } |
621 | |
678 | |
|
|
679 | static void ecb_cold |
|
|
680 | etp_set_idle_timeout (unsigned int seconds) |
|
|
681 | { |
|
|
682 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
683 | idle_timeout = seconds; |
|
|
684 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
685 | } |
|
|
686 | |
|
|
687 | static void ecb_cold |
622 | static void etp_set_min_parallel (unsigned int nthreads) |
688 | etp_set_min_parallel (unsigned int nthreads) |
623 | { |
689 | { |
624 | if (wanted < nthreads) |
690 | if (wanted < nthreads) |
625 | wanted = nthreads; |
691 | wanted = nthreads; |
626 | } |
692 | } |
627 | |
693 | |
|
|
694 | static void ecb_cold |
628 | static void etp_set_max_parallel (unsigned int nthreads) |
695 | etp_set_max_parallel (unsigned int nthreads) |
629 | { |
696 | { |
630 | if (wanted > nthreads) |
697 | if (wanted > nthreads) |
631 | wanted = nthreads; |
698 | wanted = nthreads; |
632 | |
699 | |
633 | while (started > wanted) |
700 | while (started > wanted) |
634 | etp_end_thread (); |
701 | etp_end_thread (); |
635 | } |
702 | } |
636 | |
703 | |
637 | /*****************************************************************************/ |
704 | /*****************************************************************************/ |
638 | |
705 | |
|
|
706 | static void |
639 | static void grp_try_feed (eio_req *grp) |
707 | grp_try_feed (eio_req *grp) |
640 | { |
708 | { |
641 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
709 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
642 | { |
710 | { |
643 | grp->flags &= ~EIO_FLAG_GROUPADD; |
711 | grp->flags &= ~EIO_FLAG_GROUPADD; |
644 | |
712 | |
… | |
… | |
651 | break; |
719 | break; |
652 | } |
720 | } |
653 | } |
721 | } |
654 | } |
722 | } |
655 | |
723 | |
|
|
724 | static int |
656 | static int grp_dec (eio_req *grp) |
725 | grp_dec (eio_req *grp) |
657 | { |
726 | { |
658 | --grp->size; |
727 | --grp->size; |
659 | |
728 | |
660 | /* call feeder, if applicable */ |
729 | /* call feeder, if applicable */ |
661 | grp_try_feed (grp); |
730 | grp_try_feed (grp); |
… | |
… | |
665 | return eio_finish (grp); |
734 | return eio_finish (grp); |
666 | else |
735 | else |
667 | return 0; |
736 | return 0; |
668 | } |
737 | } |
669 | |
738 | |
|
|
739 | static void |
670 | void eio_destroy (eio_req *req) |
740 | eio_destroy (eio_req *req) |
671 | { |
741 | { |
672 | if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
742 | if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
673 | if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
743 | if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
674 | |
744 | |
675 | EIO_DESTROY (req); |
745 | EIO_DESTROY (req); |
676 | } |
746 | } |
677 | |
747 | |
|
|
748 | static int |
678 | static int eio_finish (eio_req *req) |
749 | eio_finish (eio_req *req) |
679 | { |
750 | { |
680 | int res = EIO_FINISH (req); |
751 | int res = EIO_FINISH (req); |
681 | |
752 | |
682 | if (req->grp) |
753 | if (req->grp) |
683 | { |
754 | { |
… | |
… | |
691 | if (grp->grp_first == req) |
762 | if (grp->grp_first == req) |
692 | grp->grp_first = req->grp_next; |
763 | grp->grp_first = req->grp_next; |
693 | |
764 | |
694 | res2 = grp_dec (grp); |
765 | res2 = grp_dec (grp); |
695 | |
766 | |
696 | if (!res && res2) |
767 | if (!res) |
697 | res = res2; |
768 | res = res2; |
698 | } |
769 | } |
699 | |
770 | |
700 | eio_destroy (req); |
771 | eio_destroy (req); |
701 | |
772 | |
702 | return res; |
773 | return res; |
703 | } |
774 | } |
704 | |
775 | |
|
|
776 | void |
705 | void eio_grp_cancel (eio_req *grp) |
777 | eio_grp_cancel (eio_req *grp) |
706 | { |
778 | { |
707 | for (grp = grp->grp_first; grp; grp = grp->grp_next) |
779 | for (grp = grp->grp_first; grp; grp = grp->grp_next) |
708 | eio_cancel (grp); |
780 | eio_cancel (grp); |
709 | } |
781 | } |
710 | |
782 | |
|
|
783 | void |
711 | void eio_cancel (eio_req *req) |
784 | eio_cancel (eio_req *req) |
712 | { |
785 | { |
713 | etp_cancel (req); |
786 | etp_cancel (req); |
714 | } |
787 | } |
715 | |
788 | |
|
|
789 | void |
716 | void eio_submit (eio_req *req) |
790 | eio_submit (eio_req *req) |
717 | { |
791 | { |
718 | etp_submit (req); |
792 | etp_submit (req); |
719 | } |
793 | } |
720 | |
794 | |
721 | unsigned int eio_nreqs (void) |
795 | unsigned int |
|
|
796 | eio_nreqs (void) |
722 | { |
797 | { |
723 | return etp_nreqs (); |
798 | return etp_nreqs (); |
724 | } |
799 | } |
725 | |
800 | |
726 | unsigned int eio_nready (void) |
801 | unsigned int |
|
|
802 | eio_nready (void) |
727 | { |
803 | { |
728 | return etp_nready (); |
804 | return etp_nready (); |
729 | } |
805 | } |
730 | |
806 | |
731 | unsigned int eio_npending (void) |
807 | unsigned int |
|
|
808 | eio_npending (void) |
732 | { |
809 | { |
733 | return etp_npending (); |
810 | return etp_npending (); |
734 | } |
811 | } |
735 | |
812 | |
736 | unsigned int eio_nthreads (void) |
813 | unsigned int ecb_cold |
|
|
814 | eio_nthreads (void) |
737 | { |
815 | { |
738 | return etp_nthreads (); |
816 | return etp_nthreads (); |
739 | } |
817 | } |
740 | |
818 | |
|
|
819 | void ecb_cold |
741 | void eio_set_max_poll_time (double nseconds) |
820 | eio_set_max_poll_time (double nseconds) |
742 | { |
821 | { |
743 | etp_set_max_poll_time (nseconds); |
822 | etp_set_max_poll_time (nseconds); |
744 | } |
823 | } |
745 | |
824 | |
|
|
825 | void ecb_cold |
746 | void eio_set_max_poll_reqs (unsigned int maxreqs) |
826 | eio_set_max_poll_reqs (unsigned int maxreqs) |
747 | { |
827 | { |
748 | etp_set_max_poll_reqs (maxreqs); |
828 | etp_set_max_poll_reqs (maxreqs); |
749 | } |
829 | } |
750 | |
830 | |
|
|
831 | void ecb_cold |
751 | void eio_set_max_idle (unsigned int nthreads) |
832 | eio_set_max_idle (unsigned int nthreads) |
752 | { |
833 | { |
753 | etp_set_max_idle (nthreads); |
834 | etp_set_max_idle (nthreads); |
754 | } |
835 | } |
755 | |
836 | |
|
|
837 | void ecb_cold |
|
|
838 | eio_set_idle_timeout (unsigned int seconds) |
|
|
839 | { |
|
|
840 | etp_set_idle_timeout (seconds); |
|
|
841 | } |
|
|
842 | |
|
|
843 | void ecb_cold |
756 | void eio_set_min_parallel (unsigned int nthreads) |
844 | eio_set_min_parallel (unsigned int nthreads) |
757 | { |
845 | { |
758 | etp_set_min_parallel (nthreads); |
846 | etp_set_min_parallel (nthreads); |
759 | } |
847 | } |
760 | |
848 | |
|
|
849 | void ecb_cold |
761 | void eio_set_max_parallel (unsigned int nthreads) |
850 | eio_set_max_parallel (unsigned int nthreads) |
762 | { |
851 | { |
763 | etp_set_max_parallel (nthreads); |
852 | etp_set_max_parallel (nthreads); |
764 | } |
853 | } |
765 | |
854 | |
766 | int eio_poll (void) |
855 | int eio_poll (void) |
… | |
… | |
775 | # undef pread |
864 | # undef pread |
776 | # undef pwrite |
865 | # undef pwrite |
777 | # define pread eio__pread |
866 | # define pread eio__pread |
778 | # define pwrite eio__pwrite |
867 | # define pwrite eio__pwrite |
779 | |
868 | |
780 | static ssize_t |
869 | static eio_ssize_t |
781 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
870 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
782 | { |
871 | { |
783 | ssize_t res; |
872 | eio_ssize_t res; |
784 | off_t ooffset; |
873 | off_t ooffset; |
785 | |
874 | |
786 | X_LOCK (preadwritelock); |
875 | X_LOCK (preadwritelock); |
787 | ooffset = lseek (fd, 0, SEEK_CUR); |
876 | ooffset = lseek (fd, 0, SEEK_CUR); |
788 | lseek (fd, offset, SEEK_SET); |
877 | lseek (fd, offset, SEEK_SET); |
… | |
… | |
791 | X_UNLOCK (preadwritelock); |
880 | X_UNLOCK (preadwritelock); |
792 | |
881 | |
793 | return res; |
882 | return res; |
794 | } |
883 | } |
795 | |
884 | |
796 | static ssize_t |
885 | static eio_ssize_t |
797 | eio__pwrite (int fd, void *buf, size_t count, off_t offset) |
886 | eio__pwrite (int fd, void *buf, size_t count, off_t offset) |
798 | { |
887 | { |
799 | ssize_t res; |
888 | eio_ssize_t res; |
800 | off_t ooffset; |
889 | off_t ooffset; |
801 | |
890 | |
802 | X_LOCK (preadwritelock); |
891 | X_LOCK (preadwritelock); |
803 | ooffset = lseek (fd, 0, SEEK_CUR); |
892 | ooffset = lseek (fd, 0, SEEK_CUR); |
804 | lseek (fd, offset, SEEK_SET); |
893 | lseek (fd, offset, SEEK_SET); |
… | |
… | |
808 | |
897 | |
809 | return res; |
898 | return res; |
810 | } |
899 | } |
811 | #endif |
900 | #endif |
812 | |
901 | |
813 | #ifndef HAVE_FUTIMES |
902 | #ifndef HAVE_UTIMES |
814 | |
903 | |
815 | # undef utimes |
904 | # undef utimes |
816 | # undef futimes |
|
|
817 | # define utimes(path,times) eio__utimes (path, times) |
905 | # define utimes(path,times) eio__utimes (path, times) |
818 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
819 | |
906 | |
820 | static int |
907 | static int |
821 | eio__utimes (const char *filename, const struct timeval times[2]) |
908 | eio__utimes (const char *filename, const struct timeval times[2]) |
822 | { |
909 | { |
823 | if (times) |
910 | if (times) |
… | |
… | |
831 | } |
918 | } |
832 | else |
919 | else |
833 | return utime (filename, 0); |
920 | return utime (filename, 0); |
834 | } |
921 | } |
835 | |
922 | |
|
|
923 | #endif |
|
|
924 | |
|
|
925 | #ifndef HAVE_FUTIMES |
|
|
926 | |
|
|
927 | # undef futimes |
|
|
928 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
929 | |
|
|
930 | static int |
836 | static int eio__futimes (int fd, const struct timeval tv[2]) |
931 | eio__futimes (int fd, const struct timeval tv[2]) |
837 | { |
932 | { |
838 | errno = ENOSYS; |
933 | errno = ENOSYS; |
839 | return -1; |
934 | return -1; |
840 | } |
935 | } |
841 | |
936 | |
… | |
… | |
845 | # undef fdatasync |
940 | # undef fdatasync |
846 | # define fdatasync(fd) fsync (fd) |
941 | # define fdatasync(fd) fsync (fd) |
847 | #endif |
942 | #endif |
848 | |
943 | |
849 | /* sync_file_range always needs emulation */ |
944 | /* sync_file_range always needs emulation */ |
850 | int |
945 | static int |
851 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
946 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
852 | { |
947 | { |
853 | #if HAVE_SYNC_FILE_RANGE |
948 | #if HAVE_SYNC_FILE_RANGE |
854 | int res; |
949 | int res; |
855 | |
950 | |
… | |
… | |
872 | /* even though we could play tricks with the flags, it's better to always |
967 | /* even though we could play tricks with the flags, it's better to always |
873 | * call fdatasync, as that matches the expectation of its users best */ |
968 | * call fdatasync, as that matches the expectation of its users best */ |
874 | return fdatasync (fd); |
969 | return fdatasync (fd); |
875 | } |
970 | } |
876 | |
971 | |
|
|
972 | static int |
|
|
973 | eio__fallocate (int fd, int mode, off_t offset, size_t len) |
|
|
974 | { |
|
|
975 | #if HAVE_FALLOCATE |
|
|
976 | return fallocate (fd, mode, offset, len); |
|
|
977 | #else |
|
|
978 | errno = ENOSYS; |
|
|
979 | return -1; |
|
|
980 | #endif |
|
|
981 | } |
|
|
982 | |
877 | #if !HAVE_READAHEAD |
983 | #if !HAVE_READAHEAD |
878 | # undef readahead |
984 | # undef readahead |
879 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
985 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
880 | |
986 | |
881 | static ssize_t |
987 | static eio_ssize_t |
882 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
988 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
883 | { |
989 | { |
884 | size_t todo = count; |
990 | size_t todo = count; |
885 | dBUF; |
991 | dBUF; |
886 | |
992 | |
… | |
… | |
898 | } |
1004 | } |
899 | |
1005 | |
900 | #endif |
1006 | #endif |
901 | |
1007 | |
902 | /* sendfile always needs emulation */ |
1008 | /* sendfile always needs emulation */ |
903 | static ssize_t |
1009 | static eio_ssize_t |
904 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
1010 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
905 | { |
1011 | { |
|
|
1012 | eio_ssize_t written = 0; |
906 | ssize_t res; |
1013 | eio_ssize_t res; |
907 | |
1014 | |
908 | if (!count) |
1015 | if (!count) |
909 | return 0; |
1016 | return 0; |
910 | |
1017 | |
|
|
1018 | for (;;) |
|
|
1019 | { |
|
|
1020 | #ifdef __APPLE__ |
|
|
1021 | # undef HAVE_SENDFILE /* broken, as everything on os x */ |
|
|
1022 | #endif |
911 | #if HAVE_SENDFILE |
1023 | #if HAVE_SENDFILE |
912 | # if __linux |
1024 | # if __linux |
|
|
1025 | off_t soffset = offset; |
913 | res = sendfile (ofd, ifd, &offset, count); |
1026 | res = sendfile (ofd, ifd, &soffset, count); |
914 | |
1027 | |
915 | # elif __FreeBSD__ |
1028 | # elif __FreeBSD__ |
916 | /* |
1029 | /* |
917 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
1030 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
918 | * wasted on making it similar to other I/O functions. |
1031 | * wasted on making it similar to other I/O functions. |
919 | */ |
1032 | */ |
920 | { |
|
|
921 | off_t sbytes; |
1033 | off_t sbytes; |
922 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
1034 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
923 | |
1035 | |
|
|
1036 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
924 | /* freebsd' sendfile will return 0 on success */ |
1037 | /* freebsd' sendfile will return 0 on success */ |
925 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
1038 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
926 | /* not on e.g. EIO or EPIPE - sounds broken */ |
1039 | /* not on e.g. EIO or EPIPE - sounds broken */ |
927 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
1040 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
928 | res = sbytes; |
1041 | res = sbytes; |
929 | } |
1042 | #endif |
|
|
1043 | |
|
|
1044 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
1045 | if (sbytes) |
|
|
1046 | res = sbytes; |
930 | |
1047 | |
931 | # elif defined (__APPLE__) |
1048 | # elif defined (__APPLE__) |
932 | |
|
|
933 | { |
|
|
934 | off_t sbytes = count; |
1049 | off_t sbytes = count; |
935 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
1050 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
936 | |
1051 | |
937 | /* according to the manpage, sbytes is always valid */ |
1052 | /* according to the manpage, sbytes is always valid */ |
938 | if (sbytes) |
1053 | if (sbytes) |
939 | res = sbytes; |
1054 | res = sbytes; |
940 | } |
|
|
941 | |
1055 | |
942 | # elif __hpux |
1056 | # elif __hpux |
943 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
1057 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
944 | |
1058 | |
945 | # elif __solaris |
1059 | # elif __solaris |
946 | { |
|
|
947 | struct sendfilevec vec; |
1060 | struct sendfilevec vec; |
948 | size_t sbytes; |
1061 | size_t sbytes; |
949 | |
1062 | |
950 | vec.sfv_fd = ifd; |
1063 | vec.sfv_fd = ifd; |
951 | vec.sfv_flag = 0; |
1064 | vec.sfv_flag = 0; |
952 | vec.sfv_off = offset; |
1065 | vec.sfv_off = offset; |
953 | vec.sfv_len = count; |
1066 | vec.sfv_len = count; |
954 | |
1067 | |
955 | res = sendfilev (ofd, &vec, 1, &sbytes); |
1068 | res = sendfilev (ofd, &vec, 1, &sbytes); |
956 | |
1069 | |
957 | if (res < 0 && sbytes) |
1070 | if (res < 0 && sbytes) |
958 | res = sbytes; |
1071 | res = sbytes; |
959 | } |
|
|
960 | |
1072 | |
961 | # endif |
1073 | # endif |
962 | |
1074 | |
963 | #elif defined (_WIN32) |
1075 | #elif defined (_WIN32) && 0 |
964 | |
|
|
965 | /* does not work, just for documentation of what would need to be done */ |
1076 | /* does not work, just for documentation of what would need to be done */ |
966 | { |
1077 | /* actually, cannot be done like this, as TransmitFile changes the file offset, */ |
|
|
1078 | /* libeio guarantees that the file offset does not change, and windows */ |
|
|
1079 | /* has no way to get an independent handle to the same file description */ |
967 | HANDLE h = TO_SOCKET (ifd); |
1080 | HANDLE h = TO_SOCKET (ifd); |
968 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
1081 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
969 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
1082 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
970 | } |
|
|
971 | |
1083 | |
972 | #else |
1084 | #else |
973 | res = -1; |
1085 | res = -1; |
974 | errno = ENOSYS; |
1086 | errno = ENOSYS; |
975 | #endif |
1087 | #endif |
976 | |
1088 | |
|
|
1089 | /* we assume sendfile can copy at least 128mb in one go */ |
|
|
1090 | if (res <= 128 * 1024 * 1024) |
|
|
1091 | { |
|
|
1092 | if (res > 0) |
|
|
1093 | written += res; |
|
|
1094 | |
|
|
1095 | if (written) |
|
|
1096 | return written; |
|
|
1097 | |
|
|
1098 | break; |
|
|
1099 | } |
|
|
1100 | else |
|
|
1101 | { |
|
|
1102 | /* if we requested more, then probably the kernel was lazy */ |
|
|
1103 | written += res; |
|
|
1104 | offset += res; |
|
|
1105 | count -= res; |
|
|
1106 | |
|
|
1107 | if (!count) |
|
|
1108 | return written; |
|
|
1109 | } |
|
|
1110 | } |
|
|
1111 | |
977 | if (res < 0 |
1112 | if (res < 0 |
978 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1113 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
979 | /* BSDs */ |
1114 | /* BSDs */ |
980 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
1115 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
981 | || errno == ENOTSUP |
1116 | || errno == ENOTSUP |
982 | #endif |
1117 | #endif |
|
|
1118 | #ifdef EOPNOTSUPP /* windows */ |
983 | || errno == EOPNOTSUPP /* BSDs */ |
1119 | || errno == EOPNOTSUPP /* BSDs */ |
|
|
1120 | #endif |
984 | #if __solaris |
1121 | #if __solaris |
985 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1122 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
986 | #endif |
1123 | #endif |
987 | ) |
1124 | ) |
988 | ) |
1125 | ) |
… | |
… | |
992 | |
1129 | |
993 | res = 0; |
1130 | res = 0; |
994 | |
1131 | |
995 | while (count) |
1132 | while (count) |
996 | { |
1133 | { |
997 | ssize_t cnt; |
1134 | eio_ssize_t cnt; |
998 | |
1135 | |
999 | cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); |
1136 | cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); |
1000 | |
1137 | |
1001 | if (cnt <= 0) |
1138 | if (cnt <= 0) |
1002 | { |
1139 | { |
… | |
… | |
1019 | } |
1156 | } |
1020 | |
1157 | |
1021 | return res; |
1158 | return res; |
1022 | } |
1159 | } |
1023 | |
1160 | |
|
|
1161 | #ifdef PAGESIZE |
|
|
1162 | # define eio_pagesize() PAGESIZE |
|
|
1163 | #else |
|
|
1164 | static intptr_t |
|
|
1165 | eio_pagesize (void) |
|
|
1166 | { |
|
|
1167 | static intptr_t page; |
|
|
1168 | |
|
|
1169 | if (!page) |
|
|
1170 | page = sysconf (_SC_PAGESIZE); |
|
|
1171 | |
|
|
1172 | return page; |
|
|
1173 | } |
|
|
1174 | #endif |
|
|
1175 | |
|
|
1176 | static void |
|
|
1177 | eio_page_align (void **addr, size_t *length) |
|
|
1178 | { |
|
|
1179 | intptr_t mask = eio_pagesize () - 1; |
|
|
1180 | |
|
|
1181 | /* round down addr */ |
|
|
1182 | intptr_t adj = mask & (intptr_t)*addr; |
|
|
1183 | |
|
|
1184 | *addr = (void *)((intptr_t)*addr - adj); |
|
|
1185 | *length += adj; |
|
|
1186 | |
|
|
1187 | /* round up length */ |
|
|
1188 | *length = (*length + mask) & ~mask; |
|
|
1189 | } |
|
|
1190 | |
|
|
1191 | #if !_POSIX_MEMLOCK |
|
|
1192 | # define eio__mlockall(a) EIO_ENOSYS () |
|
|
1193 | #else |
|
|
1194 | |
|
|
1195 | static int |
|
|
1196 | eio__mlockall (int flags) |
|
|
1197 | { |
|
|
1198 | #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
|
|
1199 | extern int mallopt (int, int); |
|
|
1200 | mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
|
|
1201 | #endif |
|
|
1202 | |
|
|
1203 | if (EIO_MCL_CURRENT != MCL_CURRENT |
|
|
1204 | || EIO_MCL_FUTURE != MCL_FUTURE) |
|
|
1205 | { |
|
|
1206 | flags = 0 |
|
|
1207 | | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
|
|
1208 | | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
|
|
1209 | } |
|
|
1210 | |
|
|
1211 | return mlockall (flags); |
|
|
1212 | } |
|
|
1213 | #endif |
|
|
1214 | |
|
|
1215 | #if !_POSIX_MEMLOCK_RANGE |
|
|
1216 | # define eio__mlock(a,b) EIO_ENOSYS () |
|
|
1217 | #else |
|
|
1218 | |
|
|
1219 | static int |
|
|
1220 | eio__mlock (void *addr, size_t length) |
|
|
1221 | { |
|
|
1222 | eio_page_align (&addr, &length); |
|
|
1223 | |
|
|
1224 | return mlock (addr, length); |
|
|
1225 | } |
|
|
1226 | |
|
|
1227 | #endif |
|
|
1228 | |
|
|
1229 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
|
|
1230 | # define eio__msync(a,b,c) EIO_ENOSYS () |
|
|
1231 | #else |
|
|
1232 | |
|
|
1233 | static int |
|
|
1234 | eio__msync (void *mem, size_t len, int flags) |
|
|
1235 | { |
|
|
1236 | eio_page_align (&mem, &len); |
|
|
1237 | |
|
|
1238 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1239 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1240 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1241 | { |
|
|
1242 | flags = 0 |
|
|
1243 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1244 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1245 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1246 | } |
|
|
1247 | |
|
|
1248 | return msync (mem, len, flags); |
|
|
1249 | } |
|
|
1250 | |
|
|
1251 | #endif |
|
|
1252 | |
|
|
1253 | static int |
|
|
1254 | eio__mtouch (eio_req *req) |
|
|
1255 | { |
|
|
1256 | void *mem = req->ptr2; |
|
|
1257 | size_t len = req->size; |
|
|
1258 | int flags = req->int1; |
|
|
1259 | |
|
|
1260 | eio_page_align (&mem, &len); |
|
|
1261 | |
|
|
1262 | { |
|
|
1263 | intptr_t addr = (intptr_t)mem; |
|
|
1264 | intptr_t end = addr + len; |
|
|
1265 | intptr_t page = eio_pagesize (); |
|
|
1266 | |
|
|
1267 | if (addr < end) |
|
|
1268 | if (flags & EIO_MT_MODIFY) /* modify */ |
|
|
1269 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
|
|
1270 | else |
|
|
1271 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
|
|
1272 | } |
|
|
1273 | |
|
|
1274 | return 0; |
|
|
1275 | } |
|
|
1276 | |
|
|
1277 | /*****************************************************************************/ |
|
|
1278 | /* requests implemented outside eio_execute, because they are so large */ |
|
|
1279 | |
|
|
1280 | static void |
|
|
1281 | eio__realpath (eio_req *req, etp_worker *self) |
|
|
1282 | { |
|
|
1283 | char *rel = req->ptr1; |
|
|
1284 | char *res; |
|
|
1285 | char *tmp1, *tmp2; |
|
|
1286 | #if SYMLOOP_MAX > 32 |
|
|
1287 | int symlinks = SYMLOOP_MAX; |
|
|
1288 | #else |
|
|
1289 | int symlinks = 32; |
|
|
1290 | #endif |
|
|
1291 | |
|
|
1292 | req->result = -1; |
|
|
1293 | |
|
|
1294 | errno = EINVAL; |
|
|
1295 | if (!rel) |
|
|
1296 | return; |
|
|
1297 | |
|
|
1298 | errno = ENOENT; |
|
|
1299 | if (!*rel) |
|
|
1300 | return; |
|
|
1301 | |
|
|
1302 | if (!req->ptr2) |
|
|
1303 | { |
|
|
1304 | X_LOCK (wrklock); |
|
|
1305 | req->flags |= EIO_FLAG_PTR2_FREE; |
|
|
1306 | X_UNLOCK (wrklock); |
|
|
1307 | req->ptr2 = malloc (PATH_MAX * 3); |
|
|
1308 | |
|
|
1309 | errno = ENOMEM; |
|
|
1310 | if (!req->ptr2) |
|
|
1311 | return; |
|
|
1312 | } |
|
|
1313 | |
|
|
1314 | res = req->ptr2; |
|
|
1315 | tmp1 = res + PATH_MAX; |
|
|
1316 | tmp2 = tmp1 + PATH_MAX; |
|
|
1317 | |
|
|
1318 | #if 0 /* disabled, the musl way to do things is just too racy */ |
|
|
1319 | #if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) |
|
|
1320 | /* on linux we may be able to ask the kernel */ |
|
|
1321 | { |
|
|
1322 | int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME); |
|
|
1323 | |
|
|
1324 | if (fd >= 0) |
|
|
1325 | { |
|
|
1326 | sprintf (tmp1, "/proc/self/fd/%d", fd); |
|
|
1327 | req->result = readlink (tmp1, res, PATH_MAX); |
|
|
1328 | close (fd); |
|
|
1329 | |
|
|
1330 | /* here we should probably stat the open file and the disk file, to make sure they still match */ |
|
|
1331 | |
|
|
1332 | if (req->result > 0) |
|
|
1333 | goto done; |
|
|
1334 | } |
|
|
1335 | else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) |
|
|
1336 | return; |
|
|
1337 | } |
|
|
1338 | #endif |
|
|
1339 | #endif |
|
|
1340 | |
|
|
1341 | if (*rel != '/') |
|
|
1342 | { |
|
|
1343 | if (!getcwd (res, PATH_MAX)) |
|
|
1344 | return; |
|
|
1345 | |
|
|
1346 | if (res [1]) /* only use if not / */ |
|
|
1347 | res += strlen (res); |
|
|
1348 | } |
|
|
1349 | |
|
|
1350 | while (*rel) |
|
|
1351 | { |
|
|
1352 | eio_ssize_t len, linklen; |
|
|
1353 | char *beg = rel; |
|
|
1354 | |
|
|
1355 | while (*rel && *rel != '/') |
|
|
1356 | ++rel; |
|
|
1357 | |
|
|
1358 | len = rel - beg; |
|
|
1359 | |
|
|
1360 | if (!len) /* skip slashes */ |
|
|
1361 | { |
|
|
1362 | ++rel; |
|
|
1363 | continue; |
|
|
1364 | } |
|
|
1365 | |
|
|
1366 | if (beg [0] == '.') |
|
|
1367 | { |
|
|
1368 | if (len == 1) |
|
|
1369 | continue; /* . - nop */ |
|
|
1370 | |
|
|
1371 | if (beg [1] == '.' && len == 2) |
|
|
1372 | { |
|
|
1373 | /* .. - back up one component, if possible */ |
|
|
1374 | |
|
|
1375 | while (res != req->ptr2) |
|
|
1376 | if (*--res == '/') |
|
|
1377 | break; |
|
|
1378 | |
|
|
1379 | continue; |
|
|
1380 | } |
|
|
1381 | } |
|
|
1382 | |
|
|
1383 | errno = ENAMETOOLONG; |
|
|
1384 | if (res + 1 + len + 1 >= tmp1) |
|
|
1385 | return; |
|
|
1386 | |
|
|
1387 | /* copy one component */ |
|
|
1388 | *res = '/'; |
|
|
1389 | memcpy (res + 1, beg, len); |
|
|
1390 | |
|
|
1391 | /* zero-terminate, for readlink */ |
|
|
1392 | res [len + 1] = 0; |
|
|
1393 | |
|
|
1394 | /* now check if it's a symlink */ |
|
|
1395 | linklen = readlink (req->ptr2, tmp1, PATH_MAX); |
|
|
1396 | |
|
|
1397 | if (linklen < 0) |
|
|
1398 | { |
|
|
1399 | if (errno != EINVAL) |
|
|
1400 | return; |
|
|
1401 | |
|
|
1402 | /* it's a normal directory. hopefully */ |
|
|
1403 | res += len + 1; |
|
|
1404 | } |
|
|
1405 | else |
|
|
1406 | { |
|
|
1407 | /* yay, it was a symlink - build new path in tmp2 */ |
|
|
1408 | int rellen = strlen (rel); |
|
|
1409 | |
|
|
1410 | errno = ENAMETOOLONG; |
|
|
1411 | if (linklen + 1 + rellen >= PATH_MAX) |
|
|
1412 | return; |
|
|
1413 | |
|
|
1414 | errno = ELOOP; |
|
|
1415 | if (!--symlinks) |
|
|
1416 | return; |
|
|
1417 | |
|
|
1418 | if (*tmp1 == '/') |
|
|
1419 | res = req->ptr2; /* symlink resolves to an absolute path */ |
|
|
1420 | |
|
|
1421 | /* we need to be careful, as rel might point into tmp2 already */ |
|
|
1422 | memmove (tmp2 + linklen + 1, rel, rellen + 1); |
|
|
1423 | tmp2 [linklen] = '/'; |
|
|
1424 | memcpy (tmp2, tmp1, linklen); |
|
|
1425 | |
|
|
1426 | rel = tmp2; |
|
|
1427 | } |
|
|
1428 | } |
|
|
1429 | |
|
|
1430 | /* special case for the lone root path */ |
|
|
1431 | if (res == req->ptr2) |
|
|
1432 | *res++ = '/'; |
|
|
1433 | |
|
|
1434 | req->result = res - (char *)req->ptr2; |
|
|
1435 | |
|
|
1436 | done: |
|
|
1437 | req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */ |
|
|
1438 | } |
|
|
1439 | |
1024 | static signed char |
1440 | static signed char |
1025 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1441 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
1026 | { |
1442 | { |
1027 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
1443 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
1028 | : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; |
1444 | : a->inode < b->inode ? -1 |
|
|
1445 | : a->inode > b->inode ? 1 |
|
|
1446 | : 0; |
1029 | } |
1447 | } |
1030 | |
1448 | |
1031 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1449 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
1032 | |
1450 | |
1033 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1451 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1034 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
1452 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
1035 | |
1453 | |
1036 | static void |
1454 | static void |
1037 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1455 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits) |
1038 | { |
1456 | { |
1039 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
1457 | unsigned char bits [9 + sizeof (eio_ino_t) * 8]; |
1040 | unsigned char *bit = bits; |
1458 | unsigned char *bit = bits; |
1041 | |
1459 | |
1042 | assert (CHAR_BIT == 8); |
1460 | assert (CHAR_BIT == 8); |
1043 | assert (sizeof (eio_dirent) * 8 < 256); |
1461 | assert (sizeof (eio_dirent) * 8 < 256); |
1044 | assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ |
1462 | assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
1045 | assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ |
1463 | assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ |
1046 | |
1464 | |
1047 | if (size <= EIO_SORT_FAST) |
1465 | if (size <= EIO_SORT_FAST) |
1048 | return; |
1466 | return; |
1049 | |
1467 | |
1050 | /* first prepare an array of bits to test in our radix sort */ |
1468 | /* first prepare an array of bits to test in our radix sort */ |
1051 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
1469 | /* try to take endianness into account, as well as differences in eio_ino_t sizes */ |
1052 | /* inode_bits must contain all inodes ORed together */ |
1470 | /* inode_bits must contain all inodes ORed together */ |
1053 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
1471 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
1054 | { |
1472 | { |
1055 | ino_t endianness; |
1473 | eio_ino_t endianness; |
1056 | int i, j; |
1474 | int i, j; |
1057 | |
1475 | |
1058 | /* we store the byte offset of byte n into byte n of "endianness" */ |
1476 | /* we store the byte offset of byte n into byte n of "endianness" */ |
1059 | for (i = 0; i < sizeof (ino_t); ++i) |
1477 | for (i = 0; i < sizeof (eio_ino_t); ++i) |
1060 | ((unsigned char *)&endianness)[i] = i; |
1478 | ((unsigned char *)&endianness)[i] = i; |
1061 | |
1479 | |
1062 | *bit++ = 0; |
1480 | *bit++ = 0; |
1063 | |
1481 | |
1064 | for (i = 0; i < sizeof (ino_t); ++i) |
1482 | for (i = 0; i < sizeof (eio_ino_t); ++i) |
1065 | { |
1483 | { |
1066 | /* shifting off the byte offsets out of "endianness" */ |
1484 | /* shifting off the byte offsets out of "endianness" */ |
1067 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
1485 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
1068 | endianness >>= 8; |
1486 | endianness >>= 8; |
1069 | |
1487 | |
1070 | for (j = 0; j < 8; ++j) |
1488 | for (j = 0; j < 8; ++j) |
1071 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
1489 | if (inode_bits & (((eio_ino_t)1) << (i * 8 + j))) |
1072 | *bit++ = offs + j; |
1490 | *bit++ = offs + j; |
1073 | } |
1491 | } |
1074 | |
1492 | |
1075 | for (j = 0; j < 8; ++j) |
1493 | for (j = 0; j < 8; ++j) |
1076 | if (score_bits & (1 << j)) |
1494 | if (score_bits & (1 << j)) |
1077 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
1495 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
1078 | } |
1496 | } |
1079 | |
1497 | |
1080 | /* now actually do the sorting (a variant of MSD radix sort) */ |
1498 | /* now actually do the sorting (a variant of MSD radix sort) */ |
1081 | { |
1499 | { |
1082 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
1500 | eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base; |
1083 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
1501 | eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end; |
1084 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
1502 | unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8]; |
1085 | int stk_idx = 0; |
1503 | int stk_idx = 0; |
1086 | |
1504 | |
1087 | base_stk [stk_idx] = dents; |
1505 | base_stk [stk_idx] = dents; |
1088 | end_stk [stk_idx] = dents + size; |
1506 | end_stk [stk_idx] = dents + size; |
1089 | bit_stk [stk_idx] = bit - 1; |
1507 | bit_stk [stk_idx] = bit - 1; |
… | |
… | |
1168 | } |
1586 | } |
1169 | } |
1587 | } |
1170 | } |
1588 | } |
1171 | |
1589 | |
1172 | static void |
1590 | static void |
1173 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1591 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits) |
1174 | { |
1592 | { |
1175 | if (size <= 1) |
1593 | if (size <= 1) |
1176 | return; /* our insertion sort relies on size > 0 */ |
1594 | return; /* our insertion sort relies on size > 0 */ |
1177 | |
1595 | |
1178 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
1596 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
… | |
… | |
1186 | |
1604 | |
1187 | /* read a full directory */ |
1605 | /* read a full directory */ |
1188 | static void |
1606 | static void |
1189 | eio__scandir (eio_req *req, etp_worker *self) |
1607 | eio__scandir (eio_req *req, etp_worker *self) |
1190 | { |
1608 | { |
1191 | DIR *dirp; |
|
|
1192 | EIO_STRUCT_DIRENT *entp; |
|
|
1193 | char *name, *names; |
1609 | char *name, *names; |
1194 | int namesalloc = 4096; |
1610 | int namesalloc = 4096 - sizeof (void *) * 4; |
1195 | int namesoffs = 0; |
1611 | int namesoffs = 0; |
1196 | int flags = req->int1; |
1612 | int flags = req->int1; |
1197 | eio_dirent *dents = 0; |
1613 | eio_dirent *dents = 0; |
1198 | int dentalloc = 128; |
1614 | int dentalloc = 128; |
1199 | int dentoffs = 0; |
1615 | int dentoffs = 0; |
1200 | ino_t inode_bits = 0; |
1616 | eio_ino_t inode_bits = 0; |
|
|
1617 | #ifdef _WIN32 |
|
|
1618 | HANDLE dirp; |
|
|
1619 | WIN32_FIND_DATA entp; |
|
|
1620 | #else |
|
|
1621 | DIR *dirp; |
|
|
1622 | EIO_STRUCT_DIRENT *entp; |
|
|
1623 | #endif |
1201 | |
1624 | |
1202 | req->result = -1; |
1625 | req->result = -1; |
1203 | |
1626 | |
1204 | if (!(flags & EIO_READDIR_DENTS)) |
1627 | if (!(flags & EIO_READDIR_DENTS)) |
1205 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1628 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1206 | |
1629 | |
1207 | X_LOCK (wrklock); |
1630 | #ifdef _WIN32 |
1208 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1631 | { |
|
|
1632 | int len = strlen ((const char *)req->ptr1); |
|
|
1633 | char *path = malloc (MAX_PATH); |
|
|
1634 | const char *fmt; |
|
|
1635 | |
|
|
1636 | if (!len) |
|
|
1637 | fmt = "./*"; |
|
|
1638 | else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\') |
|
|
1639 | fmt = "%s*"; |
|
|
1640 | else |
|
|
1641 | fmt = "%s/*"; |
|
|
1642 | |
|
|
1643 | _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1); |
|
|
1644 | dirp = FindFirstFile (path, &entp); |
|
|
1645 | free (path); |
|
|
1646 | |
|
|
1647 | if (dirp == INVALID_HANDLE_VALUE) |
|
|
1648 | { |
|
|
1649 | dirp = 0; |
|
|
1650 | |
|
|
1651 | switch (GetLastError ()) |
|
|
1652 | { |
|
|
1653 | case ERROR_FILE_NOT_FOUND: |
|
|
1654 | req->result = 0; |
|
|
1655 | break; |
|
|
1656 | |
|
|
1657 | case ERROR_INVALID_NAME: |
|
|
1658 | case ERROR_PATH_NOT_FOUND: |
|
|
1659 | case ERROR_NO_MORE_FILES: |
|
|
1660 | errno = ENOENT; |
|
|
1661 | break; |
|
|
1662 | |
|
|
1663 | case ERROR_NOT_ENOUGH_MEMORY: |
|
|
1664 | errno = ENOMEM; |
|
|
1665 | break; |
|
|
1666 | |
|
|
1667 | default: |
|
|
1668 | errno = EINVAL; |
|
|
1669 | break; |
|
|
1670 | } |
|
|
1671 | } |
|
|
1672 | } |
|
|
1673 | #else |
1209 | self->dirp = dirp = opendir (req->ptr1); |
1674 | dirp = opendir (req->ptr1); |
|
|
1675 | #endif |
|
|
1676 | |
|
|
1677 | if (req->flags & EIO_FLAG_PTR1_FREE) |
|
|
1678 | free (req->ptr1); |
|
|
1679 | |
1210 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1680 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1211 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1681 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1212 | req->ptr2 = names = malloc (namesalloc); |
1682 | req->ptr2 = names = malloc (namesalloc); |
1213 | X_UNLOCK (wrklock); |
|
|
1214 | |
1683 | |
1215 | if (dirp && names && (!flags || dents)) |
1684 | if (dirp && names && (!flags || dents)) |
1216 | for (;;) |
1685 | for (;;) |
1217 | { |
1686 | { |
|
|
1687 | int done; |
|
|
1688 | |
|
|
1689 | #ifdef _WIN32 |
|
|
1690 | done = !dirp; |
|
|
1691 | #else |
1218 | errno = 0; |
1692 | errno = 0; |
1219 | entp = readdir (dirp); |
1693 | entp = readdir (dirp); |
|
|
1694 | done = !entp; |
|
|
1695 | #endif |
1220 | |
1696 | |
1221 | if (!entp) |
1697 | if (done) |
1222 | { |
1698 | { |
|
|
1699 | #ifndef _WIN32 |
|
|
1700 | int old_errno = errno; |
|
|
1701 | closedir (dirp); |
|
|
1702 | errno = old_errno; |
|
|
1703 | |
1223 | if (errno) |
1704 | if (errno) |
1224 | break; |
1705 | break; |
|
|
1706 | #endif |
1225 | |
1707 | |
1226 | /* sort etc. */ |
1708 | /* sort etc. */ |
1227 | req->int1 = flags; |
1709 | req->int1 = flags; |
1228 | req->result = dentoffs; |
1710 | req->result = dentoffs; |
1229 | |
1711 | |
1230 | if (flags & EIO_READDIR_STAT_ORDER) |
1712 | if (flags & EIO_READDIR_STAT_ORDER) |
1231 | eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ |
1713 | eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); |
1232 | else if (flags & EIO_READDIR_DIRS_FIRST) |
1714 | else if (flags & EIO_READDIR_DIRS_FIRST) |
1233 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
1715 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
1234 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
1716 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
1235 | else |
1717 | else |
1236 | { |
1718 | { |
… | |
… | |
1238 | eio_dirent *oth = dents + dentoffs; |
1720 | eio_dirent *oth = dents + dentoffs; |
1239 | eio_dirent *dir = dents; |
1721 | eio_dirent *dir = dents; |
1240 | |
1722 | |
1241 | /* now partition dirs to the front, and non-dirs to the back */ |
1723 | /* now partition dirs to the front, and non-dirs to the back */ |
1242 | /* by walking from both sides and swapping if necessary */ |
1724 | /* by walking from both sides and swapping if necessary */ |
1243 | /* also clear score, so it doesn't influence sorting */ |
|
|
1244 | while (oth > dir) |
1725 | while (oth > dir) |
1245 | { |
1726 | { |
1246 | if (dir->type == EIO_DT_DIR) |
1727 | if (dir->type == EIO_DT_DIR) |
1247 | ++dir; |
1728 | ++dir; |
1248 | else if ((--oth)->type == EIO_DT_DIR) |
1729 | else if ((--oth)->type == EIO_DT_DIR) |
… | |
… | |
1251 | |
1732 | |
1252 | ++dir; |
1733 | ++dir; |
1253 | } |
1734 | } |
1254 | } |
1735 | } |
1255 | |
1736 | |
1256 | /* now sort the dirs only */ |
1737 | /* now sort the dirs only (dirs all have the same score) */ |
1257 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1738 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
1258 | } |
1739 | } |
1259 | |
1740 | |
1260 | break; |
1741 | break; |
1261 | } |
1742 | } |
1262 | |
1743 | |
1263 | /* now add the entry to our list(s) */ |
1744 | /* now add the entry to our list(s) */ |
1264 | name = entp->d_name; |
1745 | name = D_NAME (entp); |
1265 | |
1746 | |
1266 | /* skip . and .. entries */ |
1747 | /* skip . and .. entries */ |
1267 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1748 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1268 | { |
1749 | { |
1269 | int len = D_NAMLEN (entp) + 1; |
1750 | int len = D_NAMLEN (entp) + 1; |
1270 | |
1751 | |
1271 | while (expect_false (namesoffs + len > namesalloc)) |
1752 | while (ecb_expect_false (namesoffs + len > namesalloc)) |
1272 | { |
1753 | { |
1273 | namesalloc *= 2; |
1754 | namesalloc *= 2; |
1274 | X_LOCK (wrklock); |
|
|
1275 | req->ptr2 = names = realloc (names, namesalloc); |
1755 | req->ptr2 = names = realloc (names, namesalloc); |
1276 | X_UNLOCK (wrklock); |
|
|
1277 | |
1756 | |
1278 | if (!names) |
1757 | if (!names) |
1279 | break; |
1758 | break; |
1280 | } |
1759 | } |
1281 | |
1760 | |
… | |
… | |
1283 | |
1762 | |
1284 | if (dents) |
1763 | if (dents) |
1285 | { |
1764 | { |
1286 | struct eio_dirent *ent; |
1765 | struct eio_dirent *ent; |
1287 | |
1766 | |
1288 | if (expect_false (dentoffs == dentalloc)) |
1767 | if (ecb_expect_false (dentoffs == dentalloc)) |
1289 | { |
1768 | { |
1290 | dentalloc *= 2; |
1769 | dentalloc *= 2; |
1291 | X_LOCK (wrklock); |
|
|
1292 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1770 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1293 | X_UNLOCK (wrklock); |
|
|
1294 | |
1771 | |
1295 | if (!dents) |
1772 | if (!dents) |
1296 | break; |
1773 | break; |
1297 | } |
1774 | } |
1298 | |
1775 | |
… | |
… | |
1378 | if (EIO_CANCELLED (req)) |
1855 | if (EIO_CANCELLED (req)) |
1379 | { |
1856 | { |
1380 | errno = ECANCELED; |
1857 | errno = ECANCELED; |
1381 | break; |
1858 | break; |
1382 | } |
1859 | } |
|
|
1860 | |
|
|
1861 | #ifdef _WIN32 |
|
|
1862 | if (!FindNextFile (dirp, &entp)) |
|
|
1863 | { |
|
|
1864 | FindClose (dirp); |
|
|
1865 | dirp = 0; |
|
|
1866 | } |
|
|
1867 | #endif |
1383 | } |
1868 | } |
1384 | } |
|
|
1385 | |
|
|
1386 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
|
|
1387 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
|
|
1388 | #else |
|
|
1389 | |
|
|
1390 | int |
|
|
1391 | eio__msync (void *mem, size_t len, int flags) |
|
|
1392 | { |
|
|
1393 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1394 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1395 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1396 | { |
|
|
1397 | flags = 0 |
|
|
1398 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1399 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1400 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1401 | } |
|
|
1402 | |
|
|
1403 | return msync (mem, len, flags); |
|
|
1404 | } |
|
|
1405 | |
|
|
1406 | #endif |
|
|
1407 | |
|
|
1408 | int |
|
|
1409 | eio__mtouch (void *mem, size_t len, int flags) |
|
|
1410 | { |
|
|
1411 | intptr_t addr = (intptr_t)mem; |
|
|
1412 | intptr_t end = addr + len; |
|
|
1413 | #ifdef PAGESIZE |
|
|
1414 | const intptr_t page = PAGESIZE; |
|
|
1415 | #else |
|
|
1416 | static intptr_t page; |
|
|
1417 | |
|
|
1418 | if (!page) |
|
|
1419 | page = sysconf (_SC_PAGESIZE); |
|
|
1420 | #endif |
|
|
1421 | |
|
|
1422 | /* round down to start of page, although this is probably useless */ |
|
|
1423 | addr &= ~(page - 1); /* assume page size is always a power of two */ |
|
|
1424 | |
|
|
1425 | if (addr < end) |
|
|
1426 | if (flags & EIO_MT_MODIFY) /* modify */ |
|
|
1427 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len); |
|
|
1428 | else |
|
|
1429 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len); |
|
|
1430 | |
|
|
1431 | return 0; |
|
|
1432 | } |
1869 | } |
1433 | |
1870 | |
1434 | /*****************************************************************************/ |
1871 | /*****************************************************************************/ |
1435 | |
1872 | |
1436 | #define ALLOC(len) \ |
1873 | #define ALLOC(len) \ |
… | |
… | |
1452 | { |
1889 | { |
1453 | ETP_REQ *req; |
1890 | ETP_REQ *req; |
1454 | struct timespec ts; |
1891 | struct timespec ts; |
1455 | etp_worker *self = (etp_worker *)thr_arg; |
1892 | etp_worker *self = (etp_worker *)thr_arg; |
1456 | |
1893 | |
1457 | /* try to distribute timeouts somewhat randomly */ |
1894 | /* try to distribute timeouts somewhat evenly */ |
1458 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1895 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1459 | |
1896 | |
1460 | for (;;) |
1897 | for (;;) |
1461 | { |
1898 | { |
|
|
1899 | ts.tv_sec = 0; |
|
|
1900 | |
1462 | X_LOCK (reqlock); |
1901 | X_LOCK (reqlock); |
1463 | |
1902 | |
1464 | for (;;) |
1903 | for (;;) |
1465 | { |
1904 | { |
1466 | self->req = req = reqq_shift (&req_queue); |
1905 | self->req = req = reqq_shift (&req_queue); |
1467 | |
1906 | |
1468 | if (req) |
1907 | if (req) |
1469 | break; |
1908 | break; |
1470 | |
1909 | |
|
|
1910 | if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ |
|
|
1911 | { |
|
|
1912 | X_UNLOCK (reqlock); |
|
|
1913 | X_LOCK (wrklock); |
|
|
1914 | --started; |
|
|
1915 | X_UNLOCK (wrklock); |
|
|
1916 | goto quit; |
|
|
1917 | } |
|
|
1918 | |
1471 | ++idle; |
1919 | ++idle; |
1472 | |
1920 | |
1473 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
1921 | if (idle <= max_idle) |
1474 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1922 | /* we are allowed to idle, so do so without any timeout */ |
|
|
1923 | X_COND_WAIT (reqwait, reqlock); |
|
|
1924 | else |
1475 | { |
1925 | { |
1476 | if (idle > max_idle) |
1926 | /* initialise timeout once */ |
1477 | { |
1927 | if (!ts.tv_sec) |
1478 | --idle; |
1928 | ts.tv_sec = time (0) + idle_timeout; |
1479 | X_UNLOCK (reqlock); |
|
|
1480 | X_LOCK (wrklock); |
|
|
1481 | --started; |
|
|
1482 | X_UNLOCK (wrklock); |
|
|
1483 | goto quit; |
|
|
1484 | } |
|
|
1485 | |
1929 | |
1486 | /* we are allowed to idle, so do so without any timeout */ |
|
|
1487 | X_COND_WAIT (reqwait, reqlock); |
1930 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
|
|
1931 | ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */ |
1488 | } |
1932 | } |
1489 | |
1933 | |
1490 | --idle; |
1934 | --idle; |
1491 | } |
1935 | } |
1492 | |
1936 | |
… | |
… | |
1495 | X_UNLOCK (reqlock); |
1939 | X_UNLOCK (reqlock); |
1496 | |
1940 | |
1497 | if (req->type < 0) |
1941 | if (req->type < 0) |
1498 | goto quit; |
1942 | goto quit; |
1499 | |
1943 | |
1500 | if (!EIO_CANCELLED (req)) |
|
|
1501 | ETP_EXECUTE (self, req); |
1944 | ETP_EXECUTE (self, req); |
1502 | |
1945 | |
1503 | X_LOCK (reslock); |
1946 | X_LOCK (reslock); |
1504 | |
1947 | |
1505 | ++npending; |
1948 | ++npending; |
1506 | |
1949 | |
… | |
… | |
1521 | return 0; |
1964 | return 0; |
1522 | } |
1965 | } |
1523 | |
1966 | |
1524 | /*****************************************************************************/ |
1967 | /*****************************************************************************/ |
1525 | |
1968 | |
|
|
1969 | int ecb_cold |
1526 | int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1970 | eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1527 | { |
1971 | { |
|
|
1972 | #if !HAVE_PREADWRITE |
|
|
1973 | X_MUTEX_CREATE (preadwritelock); |
|
|
1974 | #endif |
|
|
1975 | |
1528 | return etp_init (want_poll, done_poll); |
1976 | return etp_init (want_poll, done_poll); |
1529 | } |
1977 | } |
1530 | |
1978 | |
|
|
1979 | ecb_inline void |
1531 | static void eio_api_destroy (eio_req *req) |
1980 | eio_api_destroy (eio_req *req) |
1532 | { |
1981 | { |
1533 | free (req); |
1982 | free (req); |
1534 | } |
1983 | } |
1535 | |
1984 | |
1536 | #define REQ(rtype) \ |
1985 | #define REQ(rtype) \ |
… | |
… | |
1555 | { \ |
2004 | { \ |
1556 | eio_api_destroy (req); \ |
2005 | eio_api_destroy (req); \ |
1557 | return 0; \ |
2006 | return 0; \ |
1558 | } |
2007 | } |
1559 | |
2008 | |
|
|
2009 | static void |
1560 | static void eio_execute (etp_worker *self, eio_req *req) |
2010 | eio_execute (etp_worker *self, eio_req *req) |
1561 | { |
2011 | { |
1562 | errno = 0; |
2012 | if (ecb_expect_false (EIO_CANCELLED (req))) |
|
|
2013 | { |
|
|
2014 | req->result = -1; |
|
|
2015 | req->errorno = ECANCELED; |
|
|
2016 | return; |
|
|
2017 | } |
1563 | |
2018 | |
1564 | switch (req->type) |
2019 | switch (req->type) |
1565 | { |
2020 | { |
1566 | case EIO_READ: ALLOC (req->size); |
2021 | case EIO_READ: ALLOC (req->size); |
1567 | req->result = req->offs >= 0 |
2022 | req->result = req->offs >= 0 |
… | |
… | |
1600 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
2055 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1601 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
2056 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1602 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
2057 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1603 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
2058 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1604 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
2059 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1605 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
2060 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
1606 | |
2061 | |
|
|
2062 | case EIO_REALPATH: eio__realpath (req, self); break; |
|
|
2063 | |
1607 | case EIO_READLINK: ALLOC (NAME_MAX); |
2064 | case EIO_READLINK: ALLOC (PATH_MAX); |
1608 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
2065 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1609 | |
2066 | |
1610 | case EIO_SYNC: req->result = 0; sync (); break; |
2067 | case EIO_SYNC: req->result = 0; sync (); break; |
1611 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
2068 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1612 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
2069 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
1613 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
2070 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
|
|
2071 | case EIO_MTOUCH: req->result = eio__mtouch (req); break; |
1614 | case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; |
2072 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
|
|
2073 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
1615 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
2074 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
|
|
2075 | case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; |
1616 | |
2076 | |
1617 | case EIO_READDIR: eio__scandir (req, self); break; |
2077 | case EIO_READDIR: eio__scandir (req, self); break; |
1618 | |
2078 | |
1619 | case EIO_BUSY: |
2079 | case EIO_BUSY: |
1620 | #ifdef _WIN32 |
2080 | #ifdef _WIN32 |
1621 | Sleep (req->nv1 * 1000.); |
2081 | Sleep (req->nv1 * 1e3); |
1622 | #else |
2082 | #else |
1623 | { |
2083 | { |
1624 | struct timeval tv; |
2084 | struct timeval tv; |
1625 | |
2085 | |
1626 | tv.tv_sec = req->nv1; |
2086 | tv.tv_sec = req->nv1; |
1627 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
2087 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1628 | |
2088 | |
1629 | req->result = select (0, 0, 0, 0, &tv); |
2089 | req->result = select (0, 0, 0, 0, &tv); |
1630 | } |
2090 | } |
1631 | #endif |
2091 | #endif |
1632 | break; |
2092 | break; |
… | |
… | |
1647 | times = tv; |
2107 | times = tv; |
1648 | } |
2108 | } |
1649 | else |
2109 | else |
1650 | times = 0; |
2110 | times = 0; |
1651 | |
2111 | |
1652 | |
|
|
1653 | req->result = req->type == EIO_FUTIME |
2112 | req->result = req->type == EIO_FUTIME |
1654 | ? futimes (req->int1, times) |
2113 | ? futimes (req->int1, times) |
1655 | : utimes (req->ptr1, times); |
2114 | : utimes (req->ptr1, times); |
1656 | } |
2115 | } |
1657 | break; |
2116 | break; |
… | |
… | |
1662 | case EIO_NOP: |
2121 | case EIO_NOP: |
1663 | req->result = 0; |
2122 | req->result = 0; |
1664 | break; |
2123 | break; |
1665 | |
2124 | |
1666 | case EIO_CUSTOM: |
2125 | case EIO_CUSTOM: |
1667 | ((void (*)(eio_req *))req->feed) (req); |
2126 | req->feed (req); |
1668 | break; |
2127 | break; |
1669 | |
2128 | |
1670 | default: |
2129 | default: |
|
|
2130 | errno = ENOSYS; |
1671 | req->result = -1; |
2131 | req->result = -1; |
1672 | break; |
2132 | break; |
1673 | } |
2133 | } |
1674 | |
2134 | |
1675 | req->errorno = errno; |
2135 | req->errorno = errno; |
… | |
… | |
1705 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
2165 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
1706 | { |
2166 | { |
1707 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
2167 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
1708 | } |
2168 | } |
1709 | |
2169 | |
|
|
2170 | eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
|
|
2171 | { |
|
|
2172 | REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
|
|
2173 | } |
|
|
2174 | |
|
|
2175 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
|
|
2176 | { |
|
|
2177 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
|
|
2178 | } |
|
|
2179 | |
1710 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
2180 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
1711 | { |
2181 | { |
1712 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
2182 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
1713 | } |
2183 | } |
1714 | |
2184 | |
|
|
2185 | eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) |
|
|
2186 | { |
|
|
2187 | REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; |
|
|
2188 | } |
|
|
2189 | |
1715 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
2190 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1716 | { |
2191 | { |
1717 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
2192 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1718 | } |
2193 | } |
1719 | |
2194 | |
… | |
… | |
1760 | eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) |
2235 | eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) |
1761 | { |
2236 | { |
1762 | REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; |
2237 | REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; |
1763 | } |
2238 | } |
1764 | |
2239 | |
1765 | eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) |
2240 | eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data) |
1766 | { |
2241 | { |
1767 | REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
2242 | REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
1768 | } |
2243 | } |
1769 | |
2244 | |
1770 | eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) |
2245 | eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) |
… | |
… | |
1790 | eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) |
2265 | eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) |
1791 | { |
2266 | { |
1792 | REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; |
2267 | REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; |
1793 | } |
2268 | } |
1794 | |
2269 | |
1795 | eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) |
2270 | eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data) |
1796 | { |
2271 | { |
1797 | REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
2272 | REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
1798 | } |
2273 | } |
1799 | |
2274 | |
1800 | eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
2275 | eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
… | |
… | |
1816 | eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
2291 | eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
1817 | { |
2292 | { |
1818 | return eio__1path (EIO_READLINK, path, pri, cb, data); |
2293 | return eio__1path (EIO_READLINK, path, pri, cb, data); |
1819 | } |
2294 | } |
1820 | |
2295 | |
|
|
2296 | eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data) |
|
|
2297 | { |
|
|
2298 | return eio__1path (EIO_REALPATH, path, pri, cb, data); |
|
|
2299 | } |
|
|
2300 | |
1821 | eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
2301 | eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
1822 | { |
2302 | { |
1823 | return eio__1path (EIO_STAT, path, pri, cb, data); |
2303 | return eio__1path (EIO_STAT, path, pri, cb, data); |
1824 | } |
2304 | } |
1825 | |
2305 | |
… | |
… | |
1848 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
2328 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1849 | } |
2329 | } |
1850 | |
2330 | |
1851 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
2331 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1852 | { |
2332 | { |
1853 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
2333 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; |
1854 | } |
2334 | } |
1855 | |
2335 | |
1856 | static eio_req * |
2336 | static eio_req * |
1857 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2337 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1858 | { |
2338 | { |
… | |
… | |
1882 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2362 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1883 | { |
2363 | { |
1884 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
2364 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1885 | } |
2365 | } |
1886 | |
2366 | |
1887 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
2367 | eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) |
1888 | { |
2368 | { |
1889 | REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; |
2369 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
1890 | } |
2370 | } |
1891 | |
2371 | |
1892 | #endif |
2372 | #endif |
1893 | |
2373 | |
1894 | eio_req *eio_grp (eio_cb cb, void *data) |
2374 | eio_req *eio_grp (eio_cb cb, void *data) |
… | |
… | |
1903 | #undef SEND |
2383 | #undef SEND |
1904 | |
2384 | |
1905 | /*****************************************************************************/ |
2385 | /*****************************************************************************/ |
1906 | /* grp functions */ |
2386 | /* grp functions */ |
1907 | |
2387 | |
|
|
2388 | void |
1908 | void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
2389 | eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
1909 | { |
2390 | { |
1910 | grp->int2 = limit; |
2391 | grp->int2 = limit; |
1911 | grp->feed = feed; |
2392 | grp->feed = feed; |
1912 | |
2393 | |
1913 | grp_try_feed (grp); |
2394 | grp_try_feed (grp); |
1914 | } |
2395 | } |
1915 | |
2396 | |
|
|
2397 | void |
1916 | void eio_grp_limit (eio_req *grp, int limit) |
2398 | eio_grp_limit (eio_req *grp, int limit) |
1917 | { |
2399 | { |
1918 | grp->int2 = limit; |
2400 | grp->int2 = limit; |
1919 | |
2401 | |
1920 | grp_try_feed (grp); |
2402 | grp_try_feed (grp); |
1921 | } |
2403 | } |
1922 | |
2404 | |
|
|
2405 | void |
1923 | void eio_grp_add (eio_req *grp, eio_req *req) |
2406 | eio_grp_add (eio_req *grp, eio_req *req) |
1924 | { |
2407 | { |
1925 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
2408 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
1926 | |
2409 | |
1927 | grp->flags |= EIO_FLAG_GROUPADD; |
2410 | grp->flags |= EIO_FLAG_GROUPADD; |
1928 | |
2411 | |
… | |
… | |
1939 | } |
2422 | } |
1940 | |
2423 | |
1941 | /*****************************************************************************/ |
2424 | /*****************************************************************************/ |
1942 | /* misc garbage */ |
2425 | /* misc garbage */ |
1943 | |
2426 | |
|
|
2427 | eio_ssize_t |
1944 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2428 | eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1945 | { |
2429 | { |
1946 | etp_worker wrk; |
2430 | etp_worker wrk; |
1947 | ssize_t ret; |
2431 | eio_ssize_t ret; |
1948 | |
2432 | |
1949 | wrk.dbuf = 0; |
2433 | wrk.dbuf = 0; |
1950 | |
2434 | |
1951 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
2435 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1952 | |
2436 | |