… | |
… | |
54 | #include <stdlib.h> |
54 | #include <stdlib.h> |
55 | #include <string.h> |
55 | #include <string.h> |
56 | #include <errno.h> |
56 | #include <errno.h> |
57 | #include <sys/types.h> |
57 | #include <sys/types.h> |
58 | #include <sys/stat.h> |
58 | #include <sys/stat.h> |
59 | #include <sys/statvfs.h> |
|
|
60 | #include <limits.h> |
59 | #include <limits.h> |
61 | #include <fcntl.h> |
60 | #include <fcntl.h> |
62 | #include <assert.h> |
61 | #include <assert.h> |
63 | |
62 | |
64 | /* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ |
63 | /* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ |
… | |
… | |
68 | #endif |
67 | #endif |
69 | |
68 | |
70 | #ifndef ECANCELED |
69 | #ifndef ECANCELED |
71 | # define ECANCELED EDOM |
70 | # define ECANCELED EDOM |
72 | #endif |
71 | #endif |
|
|
72 | #ifndef ELOOP |
|
|
73 | # define ELOOP EDOM |
|
|
74 | #endif |
|
|
75 | |
|
|
76 | #if !defined(ENOTSOCK) && defined(WSAENOTSOCK) |
|
|
77 | # define ENOTSOCK WSAENOTSOCK |
|
|
78 | #endif |
73 | |
79 | |
74 | static void eio_destroy (eio_req *req); |
80 | static void eio_destroy (eio_req *req); |
75 | |
81 | |
76 | #ifndef EIO_FINISH |
82 | #ifndef EIO_FINISH |
77 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
83 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
… | |
… | |
83 | |
89 | |
84 | #ifndef EIO_FEED |
90 | #ifndef EIO_FEED |
85 | # define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
91 | # define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
86 | #endif |
92 | #endif |
87 | |
93 | |
|
|
94 | #ifndef EIO_FD_TO_WIN32_HANDLE |
|
|
95 | # define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd) |
|
|
96 | #endif |
|
|
97 | #ifndef EIO_WIN32_HANDLE_TO_FD |
|
|
98 | # define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0) |
|
|
99 | #endif |
|
|
100 | |
|
|
101 | #define EIO_ERRNO(errval,retval) ((errno = errval), retval) |
|
|
102 | |
|
|
103 | #define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1) |
|
|
104 | |
88 | #ifdef _WIN32 |
105 | #ifdef _WIN32 |
89 | |
106 | |
90 | /*doh*/ |
107 | #undef PAGESIZE |
|
|
108 | #define PAGESIZE 4096 /* GetSystemInfo? */ |
|
|
109 | |
|
|
110 | /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */ |
|
|
111 | |
|
|
112 | #ifdef EIO_STRUCT_STATI64 |
|
|
113 | /* look at perl's non-sloppy stat */ |
|
|
114 | #define stat(path,buf) _stati64 (path,buf) |
|
|
115 | #define fstat(fd,buf) _fstati64 (fd,buf) |
|
|
116 | #endif |
|
|
117 | #define lstat(path,buf) stat (path,buf) |
|
|
118 | #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) |
|
|
119 | #define mkdir(path,mode) _mkdir (path) |
|
|
120 | #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) |
|
|
121 | |
|
|
122 | #define chmod(path,mode) _chmod (path, mode) |
|
|
123 | #define dup(fd) _dup (fd) |
|
|
124 | #define dup2(fd1,fd2) _dup2 (fd1, fd2) |
|
|
125 | |
|
|
126 | #define fchmod(fd,mode) EIO_ENOSYS () |
|
|
127 | #define chown(path,uid,gid) EIO_ENOSYS () |
|
|
128 | #define fchown(fd,uid,gid) EIO_ENOSYS () |
|
|
129 | #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ |
|
|
130 | #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */ |
|
|
131 | #define mknod(path,mode,dev) EIO_ENOSYS () |
|
|
132 | #define sync() EIO_ENOSYS () |
|
|
133 | #define readlink(path,buf,s) EIO_ENOSYS () |
|
|
134 | #define statvfs(path,buf) EIO_ENOSYS () |
|
|
135 | #define fstatvfs(fd,buf) EIO_ENOSYS () |
|
|
136 | |
|
|
137 | /* rename() uses MoveFile, which fails to overwrite */ |
|
|
138 | #define rename(old,neu) eio__rename (old, neu) |
|
|
139 | |
|
|
140 | static int |
|
|
141 | eio__rename (const char *old, const char *neu) |
|
|
142 | { |
|
|
143 | if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING)) |
|
|
144 | return 0; |
|
|
145 | |
|
|
146 | /* should steal _dosmaperr */ |
|
|
147 | switch (GetLastError ()) |
|
|
148 | { |
|
|
149 | case ERROR_FILE_NOT_FOUND: |
|
|
150 | case ERROR_PATH_NOT_FOUND: |
|
|
151 | case ERROR_INVALID_DRIVE: |
|
|
152 | case ERROR_NO_MORE_FILES: |
|
|
153 | case ERROR_BAD_NETPATH: |
|
|
154 | case ERROR_BAD_NET_NAME: |
|
|
155 | case ERROR_BAD_PATHNAME: |
|
|
156 | case ERROR_FILENAME_EXCED_RANGE: |
|
|
157 | errno = ENOENT; |
|
|
158 | break; |
|
|
159 | |
|
|
160 | default: |
|
|
161 | errno = EACCES; |
|
|
162 | break; |
|
|
163 | } |
|
|
164 | |
|
|
165 | return -1; |
|
|
166 | } |
|
|
167 | |
|
|
168 | /* we could even stat and see if it exists */ |
|
|
169 | static int |
|
|
170 | symlink (const char *old, const char *neu) |
|
|
171 | { |
|
|
172 | #if WINVER >= 0x0600 |
|
|
173 | if (CreateSymbolicLink (neu, old, 1)) |
|
|
174 | return 0; |
|
|
175 | |
|
|
176 | if (CreateSymbolicLink (neu, old, 0)) |
|
|
177 | return 0; |
|
|
178 | #endif |
|
|
179 | |
|
|
180 | return EIO_ERRNO (ENOENT, -1); |
|
|
181 | } |
|
|
182 | |
|
|
183 | /* POSIX API only */ |
|
|
184 | #define CreateHardLink(neu,old,flags) 0 |
|
|
185 | #define CreateSymbolicLink(neu,old,flags) 0 |
|
|
186 | |
|
|
187 | struct statvfs |
|
|
188 | { |
|
|
189 | int dummy; |
|
|
190 | }; |
|
|
191 | |
|
|
192 | #define DT_DIR EIO_DT_DIR |
|
|
193 | #define DT_REG EIO_DT_REG |
|
|
194 | #define D_NAME(entp) entp.cFileName |
|
|
195 | #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG) |
|
|
196 | |
91 | #else |
197 | #else |
92 | |
198 | |
93 | # include <sys/time.h> |
199 | #include <sys/time.h> |
94 | # include <sys/select.h> |
200 | #include <sys/select.h> |
|
|
201 | #include <sys/statvfs.h> |
95 | # include <unistd.h> |
202 | #include <unistd.h> |
|
|
203 | #include <signal.h> |
|
|
204 | #include <dirent.h> |
|
|
205 | |
|
|
206 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
207 | #include <sys/mman.h> |
|
|
208 | #endif |
|
|
209 | |
|
|
210 | #define D_NAME(entp) entp->d_name |
|
|
211 | |
|
|
212 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
|
|
213 | #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
|
|
214 | #define _DIRENT_HAVE_D_TYPE /* sigh */ |
|
|
215 | #define D_INO(de) (de)->d_fileno |
|
|
216 | #define D_NAMLEN(de) (de)->d_namlen |
|
|
217 | #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
|
|
218 | #define D_INO(de) (de)->d_ino |
|
|
219 | #endif |
|
|
220 | |
|
|
221 | #ifdef _D_EXACT_NAMLEN |
|
|
222 | #undef D_NAMLEN |
|
|
223 | #define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
|
|
224 | #endif |
|
|
225 | |
|
|
226 | #ifdef _DIRENT_HAVE_D_TYPE |
|
|
227 | #define D_TYPE(de) (de)->d_type |
|
|
228 | #endif |
|
|
229 | |
|
|
230 | #ifndef EIO_STRUCT_DIRENT |
|
|
231 | #define EIO_STRUCT_DIRENT struct dirent |
|
|
232 | #endif |
|
|
233 | |
|
|
234 | #endif |
|
|
235 | |
|
|
236 | #if HAVE_UTIMES |
96 | # include <utime.h> |
237 | # include <utime.h> |
97 | # include <signal.h> |
|
|
98 | # include <dirent.h> |
|
|
99 | |
|
|
100 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
101 | # include <sys/mman.h> |
|
|
102 | #endif |
|
|
103 | |
|
|
104 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
|
|
105 | # if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
|
|
106 | # define _DIRENT_HAVE_D_TYPE /* sigh */ |
|
|
107 | # define D_INO(de) (de)->d_fileno |
|
|
108 | # define D_NAMLEN(de) (de)->d_namlen |
|
|
109 | # elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
|
|
110 | # define D_INO(de) (de)->d_ino |
|
|
111 | # endif |
238 | #endif |
112 | |
239 | |
113 | #ifdef _D_EXACT_NAMLEN |
240 | #if HAVE_SYS_SYSCALL_H |
114 | # undef D_NAMLEN |
241 | # include <sys/syscall.h> |
115 | # define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
|
|
116 | #endif |
|
|
117 | |
|
|
118 | # ifdef _DIRENT_HAVE_D_TYPE |
|
|
119 | # define D_TYPE(de) (de)->d_type |
|
|
120 | # endif |
242 | #endif |
121 | |
243 | |
122 | # ifndef EIO_STRUCT_DIRENT |
244 | #if HAVE_SYS_PRCTL_H |
123 | # define EIO_STRUCT_DIRENT struct dirent |
245 | # include <sys/prctl.h> |
124 | # endif |
|
|
125 | |
|
|
126 | #endif |
246 | #endif |
127 | |
247 | |
128 | #if HAVE_SENDFILE |
248 | #if HAVE_SENDFILE |
129 | # if __linux |
249 | # if __linux |
130 | # include <sys/sendfile.h> |
250 | # include <sys/sendfile.h> |
… | |
… | |
145 | #endif |
265 | #endif |
146 | #ifndef D_INO |
266 | #ifndef D_INO |
147 | # define D_INO(de) 0 |
267 | # define D_INO(de) 0 |
148 | #endif |
268 | #endif |
149 | #ifndef D_NAMLEN |
269 | #ifndef D_NAMLEN |
150 | # define D_NAMLEN(de) strlen ((de)->d_name) |
270 | # define D_NAMLEN(entp) strlen (D_NAME (entp)) |
151 | #endif |
271 | #endif |
152 | |
272 | |
153 | /* used for struct dirent, AIX doesn't provide it */ |
273 | /* used for struct dirent, AIX doesn't provide it */ |
154 | #ifndef NAME_MAX |
274 | #ifndef NAME_MAX |
155 | # define NAME_MAX 4096 |
275 | # define NAME_MAX 4096 |
… | |
… | |
162 | |
282 | |
163 | /* buffer size for various temporary buffers */ |
283 | /* buffer size for various temporary buffers */ |
164 | #define EIO_BUFSIZE 65536 |
284 | #define EIO_BUFSIZE 65536 |
165 | |
285 | |
166 | #define dBUF \ |
286 | #define dBUF \ |
167 | char *eio_buf; \ |
|
|
168 | ETP_WORKER_LOCK (self); \ |
|
|
169 | self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ |
287 | char *eio_buf = malloc (EIO_BUFSIZE); \ |
170 | ETP_WORKER_UNLOCK (self); \ |
|
|
171 | errno = ENOMEM; \ |
288 | errno = ENOMEM; \ |
172 | if (!eio_buf) \ |
289 | if (!eio_buf) \ |
173 | return -1; |
290 | return -1 |
|
|
291 | |
|
|
292 | #define FUBd \ |
|
|
293 | free (eio_buf) |
174 | |
294 | |
175 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
295 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
176 | |
296 | |
177 | #define ETP_PRI_MIN EIO_PRI_MIN |
297 | #define ETP_PRI_MIN EIO_PRI_MIN |
178 | #define ETP_PRI_MAX EIO_PRI_MAX |
298 | #define ETP_PRI_MAX EIO_PRI_MAX |
… | |
… | |
184 | static int eio_finish (eio_req *req); |
304 | static int eio_finish (eio_req *req); |
185 | #define ETP_FINISH(req) eio_finish (req) |
305 | #define ETP_FINISH(req) eio_finish (req) |
186 | static void eio_execute (struct etp_worker *self, eio_req *req); |
306 | static void eio_execute (struct etp_worker *self, eio_req *req); |
187 | #define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) |
307 | #define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) |
188 | |
308 | |
189 | #define ETP_WORKER_CLEAR(req) \ |
|
|
190 | if (wrk->dbuf) \ |
|
|
191 | { \ |
|
|
192 | free (wrk->dbuf); \ |
|
|
193 | wrk->dbuf = 0; \ |
|
|
194 | } \ |
|
|
195 | \ |
|
|
196 | if (wrk->dirp) \ |
|
|
197 | { \ |
|
|
198 | closedir (wrk->dirp); \ |
|
|
199 | wrk->dirp = 0; \ |
|
|
200 | } |
|
|
201 | |
|
|
202 | #define ETP_WORKER_COMMON \ |
|
|
203 | void *dbuf; \ |
|
|
204 | DIR *dirp; |
|
|
205 | |
|
|
206 | /*****************************************************************************/ |
309 | /*****************************************************************************/ |
207 | |
310 | |
208 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
311 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
209 | |
312 | |
210 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
313 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
… | |
… | |
221 | static void (*done_poll_cb) (void); |
324 | static void (*done_poll_cb) (void); |
222 | |
325 | |
223 | static unsigned int max_poll_time; /* reslock */ |
326 | static unsigned int max_poll_time; /* reslock */ |
224 | static unsigned int max_poll_reqs; /* reslock */ |
327 | static unsigned int max_poll_reqs; /* reslock */ |
225 | |
328 | |
226 | static volatile unsigned int nreqs; /* reqlock */ |
329 | static unsigned int nreqs; /* reqlock */ |
227 | static volatile unsigned int nready; /* reqlock */ |
330 | static unsigned int nready; /* reqlock */ |
228 | static volatile unsigned int npending; /* reqlock */ |
331 | static unsigned int npending; /* reqlock */ |
229 | static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
332 | static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
230 | static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
333 | static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
231 | |
334 | |
232 | static xmutex_t wrklock; |
335 | static xmutex_t wrklock; |
233 | static xmutex_t reslock; |
336 | static xmutex_t reslock; |
234 | static xmutex_t reqlock; |
337 | static xmutex_t reqlock; |
235 | static xcond_t reqwait; |
338 | static xcond_t reqwait; |
… | |
… | |
238 | /* |
341 | /* |
239 | * make our pread/pwrite emulation safe against themselves, but not against |
342 | * make our pread/pwrite emulation safe against themselves, but not against |
240 | * normal read/write by using a mutex. slows down execution a lot, |
343 | * normal read/write by using a mutex. slows down execution a lot, |
241 | * but that's your problem, not mine. |
344 | * but that's your problem, not mine. |
242 | */ |
345 | */ |
243 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
346 | static xmutex_t preadwritelock; |
244 | #endif |
347 | #endif |
245 | |
348 | |
246 | typedef struct etp_worker |
349 | typedef struct etp_worker |
247 | { |
350 | { |
248 | /* locked by wrklock */ |
351 | /* locked by wrklock */ |
… | |
… | |
251 | xthread_t tid; |
354 | xthread_t tid; |
252 | |
355 | |
253 | /* locked by reslock, reqlock or wrklock */ |
356 | /* locked by reslock, reqlock or wrklock */ |
254 | ETP_REQ *req; /* currently processed request */ |
357 | ETP_REQ *req; /* currently processed request */ |
255 | |
358 | |
|
|
359 | #ifdef ETP_WORKER_COMMON |
256 | ETP_WORKER_COMMON |
360 | ETP_WORKER_COMMON |
|
|
361 | #endif |
257 | } etp_worker; |
362 | } etp_worker; |
258 | |
363 | |
259 | static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ |
364 | static etp_worker wrk_first; /* NOT etp */ |
260 | |
365 | |
261 | #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
366 | #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
262 | #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
367 | #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
263 | |
368 | |
264 | /* worker threads management */ |
369 | /* worker threads management */ |
265 | |
370 | |
266 | static void ecb_cold |
371 | static void ecb_cold |
267 | etp_worker_clear (etp_worker *wrk) |
372 | etp_worker_clear (etp_worker *wrk) |
268 | { |
373 | { |
269 | ETP_WORKER_CLEAR (wrk); |
|
|
270 | } |
374 | } |
271 | |
375 | |
272 | static void ecb_cold |
376 | static void ecb_cold |
273 | etp_worker_free (etp_worker *wrk) |
377 | etp_worker_free (etp_worker *wrk) |
274 | { |
378 | { |
… | |
… | |
335 | } etp_reqq; |
439 | } etp_reqq; |
336 | |
440 | |
337 | static etp_reqq req_queue; |
441 | static etp_reqq req_queue; |
338 | static etp_reqq res_queue; |
442 | static etp_reqq res_queue; |
339 | |
443 | |
|
|
444 | static void ecb_noinline ecb_cold |
|
|
445 | reqq_init (etp_reqq *q) |
|
|
446 | { |
|
|
447 | int pri; |
|
|
448 | |
|
|
449 | for (pri = 0; pri < ETP_NUM_PRI; ++pri) |
|
|
450 | q->qs[pri] = q->qe[pri] = 0; |
|
|
451 | |
|
|
452 | q->size = 0; |
|
|
453 | } |
|
|
454 | |
340 | static int ecb_noinline |
455 | static int ecb_noinline |
341 | reqq_push (etp_reqq *q, ETP_REQ *req) |
456 | reqq_push (etp_reqq *q, ETP_REQ *req) |
342 | { |
457 | { |
343 | int pri = req->pri; |
458 | int pri = req->pri; |
344 | req->next = 0; |
459 | req->next = 0; |
… | |
… | |
378 | } |
493 | } |
379 | |
494 | |
380 | abort (); |
495 | abort (); |
381 | } |
496 | } |
382 | |
497 | |
383 | static void ecb_cold |
498 | static int ecb_cold |
384 | etp_thread_init (void) |
499 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
385 | { |
500 | { |
386 | X_MUTEX_CREATE (wrklock); |
501 | X_MUTEX_CREATE (wrklock); |
387 | X_MUTEX_CREATE (reslock); |
502 | X_MUTEX_CREATE (reslock); |
388 | X_MUTEX_CREATE (reqlock); |
503 | X_MUTEX_CREATE (reqlock); |
389 | X_COND_CREATE (reqwait); |
504 | X_COND_CREATE (reqwait); |
390 | } |
|
|
391 | |
505 | |
392 | static void ecb_cold |
506 | reqq_init (&req_queue); |
393 | etp_atfork_prepare (void) |
507 | reqq_init (&res_queue); |
394 | { |
|
|
395 | X_LOCK (wrklock); |
|
|
396 | X_LOCK (reqlock); |
|
|
397 | X_LOCK (reslock); |
|
|
398 | #if !HAVE_PREADWRITE |
|
|
399 | X_LOCK (preadwritelock); |
|
|
400 | #endif |
|
|
401 | } |
|
|
402 | |
508 | |
403 | static void ecb_cold |
509 | wrk_first.next = |
404 | etp_atfork_parent (void) |
510 | wrk_first.prev = &wrk_first; |
405 | { |
|
|
406 | #if !HAVE_PREADWRITE |
|
|
407 | X_UNLOCK (preadwritelock); |
|
|
408 | #endif |
|
|
409 | X_UNLOCK (reslock); |
|
|
410 | X_UNLOCK (reqlock); |
|
|
411 | X_UNLOCK (wrklock); |
|
|
412 | } |
|
|
413 | |
|
|
414 | static void ecb_cold |
|
|
415 | etp_atfork_child (void) |
|
|
416 | { |
|
|
417 | ETP_REQ *prv; |
|
|
418 | |
|
|
419 | while ((prv = reqq_shift (&req_queue))) |
|
|
420 | ETP_DESTROY (prv); |
|
|
421 | |
|
|
422 | while ((prv = reqq_shift (&res_queue))) |
|
|
423 | ETP_DESTROY (prv); |
|
|
424 | |
|
|
425 | while (wrk_first.next != &wrk_first) |
|
|
426 | { |
|
|
427 | etp_worker *wrk = wrk_first.next; |
|
|
428 | |
|
|
429 | if (wrk->req) |
|
|
430 | ETP_DESTROY (wrk->req); |
|
|
431 | |
|
|
432 | etp_worker_clear (wrk); |
|
|
433 | etp_worker_free (wrk); |
|
|
434 | } |
|
|
435 | |
511 | |
436 | started = 0; |
512 | started = 0; |
437 | idle = 0; |
513 | idle = 0; |
438 | nreqs = 0; |
514 | nreqs = 0; |
439 | nready = 0; |
515 | nready = 0; |
440 | npending = 0; |
516 | npending = 0; |
441 | |
|
|
442 | etp_thread_init (); |
|
|
443 | } |
|
|
444 | |
|
|
445 | static void ecb_cold |
|
|
446 | etp_once_init (void) |
|
|
447 | { |
|
|
448 | etp_thread_init (); |
|
|
449 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
|
|
450 | } |
|
|
451 | |
|
|
452 | static int ecb_cold |
|
|
453 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
|
|
454 | { |
|
|
455 | static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
|
|
456 | |
|
|
457 | pthread_once (&doinit, etp_once_init); |
|
|
458 | |
517 | |
459 | want_poll_cb = want_poll; |
518 | want_poll_cb = want_poll; |
460 | done_poll_cb = done_poll; |
519 | done_poll_cb = done_poll; |
461 | |
520 | |
462 | return 0; |
521 | return 0; |
… | |
… | |
589 | } |
648 | } |
590 | |
649 | |
591 | static void |
650 | static void |
592 | etp_cancel (ETP_REQ *req) |
651 | etp_cancel (ETP_REQ *req) |
593 | { |
652 | { |
594 | X_LOCK (wrklock); |
653 | req->cancelled = 1; |
595 | req->flags |= EIO_FLAG_CANCELLED; |
|
|
596 | X_UNLOCK (wrklock); |
|
|
597 | |
654 | |
598 | eio_grp_cancel (req); |
655 | eio_grp_cancel (req); |
599 | } |
656 | } |
600 | |
657 | |
601 | static void |
658 | static void |
… | |
… | |
847 | # undef pread |
904 | # undef pread |
848 | # undef pwrite |
905 | # undef pwrite |
849 | # define pread eio__pread |
906 | # define pread eio__pread |
850 | # define pwrite eio__pwrite |
907 | # define pwrite eio__pwrite |
851 | |
908 | |
852 | static ssize_t |
909 | static eio_ssize_t |
853 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
910 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
854 | { |
911 | { |
855 | ssize_t res; |
912 | eio_ssize_t res; |
856 | off_t ooffset; |
913 | off_t ooffset; |
857 | |
914 | |
858 | X_LOCK (preadwritelock); |
915 | X_LOCK (preadwritelock); |
859 | ooffset = lseek (fd, 0, SEEK_CUR); |
916 | ooffset = lseek (fd, 0, SEEK_CUR); |
860 | lseek (fd, offset, SEEK_SET); |
917 | lseek (fd, offset, SEEK_SET); |
… | |
… | |
863 | X_UNLOCK (preadwritelock); |
920 | X_UNLOCK (preadwritelock); |
864 | |
921 | |
865 | return res; |
922 | return res; |
866 | } |
923 | } |
867 | |
924 | |
868 | static ssize_t |
925 | static eio_ssize_t |
869 | eio__pwrite (int fd, void *buf, size_t count, off_t offset) |
926 | eio__pwrite (int fd, void *buf, size_t count, off_t offset) |
870 | { |
927 | { |
871 | ssize_t res; |
928 | eio_ssize_t res; |
872 | off_t ooffset; |
929 | off_t ooffset; |
873 | |
930 | |
874 | X_LOCK (preadwritelock); |
931 | X_LOCK (preadwritelock); |
875 | ooffset = lseek (fd, 0, SEEK_CUR); |
932 | ooffset = lseek (fd, 0, SEEK_CUR); |
876 | lseek (fd, offset, SEEK_SET); |
933 | lseek (fd, offset, SEEK_SET); |
… | |
… | |
921 | |
978 | |
922 | #if !HAVE_FDATASYNC |
979 | #if !HAVE_FDATASYNC |
923 | # undef fdatasync |
980 | # undef fdatasync |
924 | # define fdatasync(fd) fsync (fd) |
981 | # define fdatasync(fd) fsync (fd) |
925 | #endif |
982 | #endif |
|
|
983 | |
|
|
984 | static int |
|
|
985 | eio__syncfs (int fd) |
|
|
986 | { |
|
|
987 | int res; |
|
|
988 | |
|
|
989 | #if HAVE_SYS_SYNCFS |
|
|
990 | res = (int)syscall (__NR_syncfs, (int)(fd)); |
|
|
991 | #else |
|
|
992 | res = -1; |
|
|
993 | errno = ENOSYS; |
|
|
994 | #endif |
|
|
995 | |
|
|
996 | if (res < 0 && errno == ENOSYS && fd >= 0) |
|
|
997 | sync (); |
|
|
998 | |
|
|
999 | return res; |
|
|
1000 | } |
926 | |
1001 | |
927 | /* sync_file_range always needs emulation */ |
1002 | /* sync_file_range always needs emulation */ |
928 | static int |
1003 | static int |
929 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
1004 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
930 | { |
1005 | { |
… | |
… | |
950 | /* even though we could play tricks with the flags, it's better to always |
1025 | /* even though we could play tricks with the flags, it's better to always |
951 | * call fdatasync, as that matches the expectation of its users best */ |
1026 | * call fdatasync, as that matches the expectation of its users best */ |
952 | return fdatasync (fd); |
1027 | return fdatasync (fd); |
953 | } |
1028 | } |
954 | |
1029 | |
|
|
1030 | static int |
|
|
1031 | eio__fallocate (int fd, int mode, off_t offset, size_t len) |
|
|
1032 | { |
|
|
1033 | #if HAVE_FALLOCATE |
|
|
1034 | return fallocate (fd, mode, offset, len); |
|
|
1035 | #else |
|
|
1036 | errno = ENOSYS; |
|
|
1037 | return -1; |
|
|
1038 | #endif |
|
|
1039 | } |
|
|
1040 | |
955 | #if !HAVE_READAHEAD |
1041 | #if !HAVE_READAHEAD |
956 | # undef readahead |
1042 | # undef readahead |
957 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
1043 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
958 | |
1044 | |
959 | static ssize_t |
1045 | static eio_ssize_t |
960 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
1046 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
961 | { |
1047 | { |
962 | size_t todo = count; |
1048 | size_t todo = count; |
963 | dBUF; |
1049 | dBUF; |
964 | |
1050 | |
… | |
… | |
969 | pread (fd, eio_buf, len, offset); |
1055 | pread (fd, eio_buf, len, offset); |
970 | offset += len; |
1056 | offset += len; |
971 | todo -= len; |
1057 | todo -= len; |
972 | } |
1058 | } |
973 | |
1059 | |
|
|
1060 | FUBd; |
|
|
1061 | |
974 | errno = 0; |
1062 | errno = 0; |
975 | return count; |
1063 | return count; |
976 | } |
1064 | } |
977 | |
1065 | |
978 | #endif |
1066 | #endif |
979 | |
1067 | |
980 | /* sendfile always needs emulation */ |
1068 | /* sendfile always needs emulation */ |
981 | static ssize_t |
1069 | static eio_ssize_t |
982 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
1070 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count) |
983 | { |
1071 | { |
984 | ssize_t written = 0; |
1072 | eio_ssize_t written = 0; |
985 | ssize_t res; |
1073 | eio_ssize_t res; |
986 | |
1074 | |
987 | if (!count) |
1075 | if (!count) |
988 | return 0; |
1076 | return 0; |
989 | |
1077 | |
990 | for (;;) |
1078 | for (;;) |
991 | { |
1079 | { |
|
|
1080 | #ifdef __APPLE__ |
|
|
1081 | # undef HAVE_SENDFILE /* broken, as everything on os x */ |
|
|
1082 | #endif |
992 | #if HAVE_SENDFILE |
1083 | #if HAVE_SENDFILE |
993 | # if __linux |
1084 | # if __linux |
994 | off_t soffset = offset; |
1085 | off_t soffset = offset; |
995 | res = sendfile (ofd, ifd, &soffset, count); |
1086 | res = sendfile (ofd, ifd, &soffset, count); |
996 | |
1087 | |
… | |
… | |
1012 | |
1103 | |
1013 | /* according to source inspection, this is correct, and useful behaviour */ |
1104 | /* according to source inspection, this is correct, and useful behaviour */ |
1014 | if (sbytes) |
1105 | if (sbytes) |
1015 | res = sbytes; |
1106 | res = sbytes; |
1016 | |
1107 | |
1017 | # elif defined (__APPLE__) && 0 /* broken, as everything on os x */ |
1108 | # elif defined (__APPLE__) |
1018 | off_t sbytes = count; |
1109 | off_t sbytes = count; |
1019 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
1110 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
1020 | |
1111 | |
1021 | /* according to the manpage, sbytes is always valid */ |
1112 | /* according to the manpage, sbytes is always valid */ |
1022 | if (sbytes) |
1113 | if (sbytes) |
… | |
… | |
1039 | if (res < 0 && sbytes) |
1130 | if (res < 0 && sbytes) |
1040 | res = sbytes; |
1131 | res = sbytes; |
1041 | |
1132 | |
1042 | # endif |
1133 | # endif |
1043 | |
1134 | |
1044 | #elif defined (_WIN32) |
1135 | #elif defined (_WIN32) && 0 |
1045 | /* does not work, just for documentation of what would need to be done */ |
1136 | /* does not work, just for documentation of what would need to be done */ |
1046 | /* actually, cannot be done like this, as TransmitFile changes the file offset, */ |
1137 | /* actually, cannot be done like this, as TransmitFile changes the file offset, */ |
1047 | /* libeio guarantees that the file offset does not change, and windows */ |
1138 | /* libeio guarantees that the file offset does not change, and windows */ |
1048 | /* has no way to get an independent handle to the same file description */ |
1139 | /* has no way to get an independent handle to the same file description */ |
1049 | HANDLE h = TO_SOCKET (ifd); |
1140 | HANDLE h = TO_SOCKET (ifd); |
… | |
… | |
1082 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1173 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1083 | /* BSDs */ |
1174 | /* BSDs */ |
1084 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
1175 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
1085 | || errno == ENOTSUP |
1176 | || errno == ENOTSUP |
1086 | #endif |
1177 | #endif |
|
|
1178 | #ifdef EOPNOTSUPP /* windows */ |
1087 | || errno == EOPNOTSUPP /* BSDs */ |
1179 | || errno == EOPNOTSUPP /* BSDs */ |
|
|
1180 | #endif |
1088 | #if __solaris |
1181 | #if __solaris |
1089 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1182 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1090 | #endif |
1183 | #endif |
1091 | ) |
1184 | ) |
1092 | ) |
1185 | ) |
… | |
… | |
1096 | |
1189 | |
1097 | res = 0; |
1190 | res = 0; |
1098 | |
1191 | |
1099 | while (count) |
1192 | while (count) |
1100 | { |
1193 | { |
1101 | ssize_t cnt; |
1194 | eio_ssize_t cnt; |
1102 | |
1195 | |
1103 | cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); |
1196 | cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); |
1104 | |
1197 | |
1105 | if (cnt <= 0) |
1198 | if (cnt <= 0) |
1106 | { |
1199 | { |
… | |
… | |
1118 | |
1211 | |
1119 | offset += cnt; |
1212 | offset += cnt; |
1120 | res += cnt; |
1213 | res += cnt; |
1121 | count -= cnt; |
1214 | count -= cnt; |
1122 | } |
1215 | } |
|
|
1216 | |
|
|
1217 | FUBd; |
1123 | } |
1218 | } |
1124 | |
1219 | |
1125 | return res; |
1220 | return res; |
1126 | } |
1221 | } |
1127 | |
1222 | |
… | |
… | |
1154 | /* round up length */ |
1249 | /* round up length */ |
1155 | *length = (*length + mask) & ~mask; |
1250 | *length = (*length + mask) & ~mask; |
1156 | } |
1251 | } |
1157 | |
1252 | |
1158 | #if !_POSIX_MEMLOCK |
1253 | #if !_POSIX_MEMLOCK |
1159 | # define eio__mlockall(a) ((errno = ENOSYS), -1) |
1254 | # define eio__mlockall(a) EIO_ENOSYS () |
1160 | #else |
1255 | #else |
1161 | |
1256 | |
1162 | static int |
1257 | static int |
1163 | eio__mlockall (int flags) |
1258 | eio__mlockall (int flags) |
1164 | { |
1259 | { |
… | |
… | |
1178 | return mlockall (flags); |
1273 | return mlockall (flags); |
1179 | } |
1274 | } |
1180 | #endif |
1275 | #endif |
1181 | |
1276 | |
1182 | #if !_POSIX_MEMLOCK_RANGE |
1277 | #if !_POSIX_MEMLOCK_RANGE |
1183 | # define eio__mlock(a,b) ((errno = ENOSYS), -1) |
1278 | # define eio__mlock(a,b) EIO_ENOSYS () |
1184 | #else |
1279 | #else |
1185 | |
1280 | |
1186 | static int |
1281 | static int |
1187 | eio__mlock (void *addr, size_t length) |
1282 | eio__mlock (void *addr, size_t length) |
1188 | { |
1283 | { |
… | |
… | |
1192 | } |
1287 | } |
1193 | |
1288 | |
1194 | #endif |
1289 | #endif |
1195 | |
1290 | |
1196 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1291 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
1197 | # define eio__msync(a,b,c) ((errno = ENOSYS), -1) |
1292 | # define eio__msync(a,b,c) EIO_ENOSYS () |
1198 | #else |
1293 | #else |
1199 | |
1294 | |
1200 | static int |
1295 | static int |
1201 | eio__msync (void *mem, size_t len, int flags) |
1296 | eio__msync (void *mem, size_t len, int flags) |
1202 | { |
1297 | { |
… | |
… | |
1314 | res += strlen (res); |
1409 | res += strlen (res); |
1315 | } |
1410 | } |
1316 | |
1411 | |
1317 | while (*rel) |
1412 | while (*rel) |
1318 | { |
1413 | { |
1319 | ssize_t len, linklen; |
1414 | eio_ssize_t len, linklen; |
1320 | char *beg = rel; |
1415 | char *beg = rel; |
1321 | |
1416 | |
1322 | while (*rel && *rel != '/') |
1417 | while (*rel && *rel != '/') |
1323 | ++rel; |
1418 | ++rel; |
1324 | |
1419 | |
… | |
… | |
1417 | |
1512 | |
1418 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1513 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
1419 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
1514 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
1420 | |
1515 | |
1421 | static void |
1516 | static void |
1422 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1517 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits) |
1423 | { |
1518 | { |
1424 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
1519 | unsigned char bits [9 + sizeof (eio_ino_t) * 8]; |
1425 | unsigned char *bit = bits; |
1520 | unsigned char *bit = bits; |
1426 | |
1521 | |
1427 | assert (CHAR_BIT == 8); |
1522 | assert (CHAR_BIT == 8); |
1428 | assert (sizeof (eio_dirent) * 8 < 256); |
1523 | assert (sizeof (eio_dirent) * 8 < 256); |
1429 | assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
1524 | assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
… | |
… | |
1431 | |
1526 | |
1432 | if (size <= EIO_SORT_FAST) |
1527 | if (size <= EIO_SORT_FAST) |
1433 | return; |
1528 | return; |
1434 | |
1529 | |
1435 | /* first prepare an array of bits to test in our radix sort */ |
1530 | /* first prepare an array of bits to test in our radix sort */ |
1436 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
1531 | /* try to take endianness into account, as well as differences in eio_ino_t sizes */ |
1437 | /* inode_bits must contain all inodes ORed together */ |
1532 | /* inode_bits must contain all inodes ORed together */ |
1438 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
1533 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
1439 | { |
1534 | { |
1440 | ino_t endianness; |
1535 | eio_ino_t endianness; |
1441 | int i, j; |
1536 | int i, j; |
1442 | |
1537 | |
1443 | /* we store the byte offset of byte n into byte n of "endianness" */ |
1538 | /* we store the byte offset of byte n into byte n of "endianness" */ |
1444 | for (i = 0; i < sizeof (ino_t); ++i) |
1539 | for (i = 0; i < sizeof (eio_ino_t); ++i) |
1445 | ((unsigned char *)&endianness)[i] = i; |
1540 | ((unsigned char *)&endianness)[i] = i; |
1446 | |
1541 | |
1447 | *bit++ = 0; |
1542 | *bit++ = 0; |
1448 | |
1543 | |
1449 | for (i = 0; i < sizeof (ino_t); ++i) |
1544 | for (i = 0; i < sizeof (eio_ino_t); ++i) |
1450 | { |
1545 | { |
1451 | /* shifting off the byte offsets out of "endianness" */ |
1546 | /* shifting off the byte offsets out of "endianness" */ |
1452 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
1547 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
1453 | endianness >>= 8; |
1548 | endianness >>= 8; |
1454 | |
1549 | |
1455 | for (j = 0; j < 8; ++j) |
1550 | for (j = 0; j < 8; ++j) |
1456 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
1551 | if (inode_bits & (((eio_ino_t)1) << (i * 8 + j))) |
1457 | *bit++ = offs + j; |
1552 | *bit++ = offs + j; |
1458 | } |
1553 | } |
1459 | |
1554 | |
1460 | for (j = 0; j < 8; ++j) |
1555 | for (j = 0; j < 8; ++j) |
1461 | if (score_bits & (1 << j)) |
1556 | if (score_bits & (1 << j)) |
1462 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
1557 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
1463 | } |
1558 | } |
1464 | |
1559 | |
1465 | /* now actually do the sorting (a variant of MSD radix sort) */ |
1560 | /* now actually do the sorting (a variant of MSD radix sort) */ |
1466 | { |
1561 | { |
1467 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
1562 | eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base; |
1468 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
1563 | eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end; |
1469 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
1564 | unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8]; |
1470 | int stk_idx = 0; |
1565 | int stk_idx = 0; |
1471 | |
1566 | |
1472 | base_stk [stk_idx] = dents; |
1567 | base_stk [stk_idx] = dents; |
1473 | end_stk [stk_idx] = dents + size; |
1568 | end_stk [stk_idx] = dents + size; |
1474 | bit_stk [stk_idx] = bit - 1; |
1569 | bit_stk [stk_idx] = bit - 1; |
… | |
… | |
1553 | } |
1648 | } |
1554 | } |
1649 | } |
1555 | } |
1650 | } |
1556 | |
1651 | |
1557 | static void |
1652 | static void |
1558 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
1653 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits) |
1559 | { |
1654 | { |
1560 | if (size <= 1) |
1655 | if (size <= 1) |
1561 | return; /* our insertion sort relies on size > 0 */ |
1656 | return; /* our insertion sort relies on size > 0 */ |
1562 | |
1657 | |
1563 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
1658 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
… | |
… | |
1571 | |
1666 | |
1572 | /* read a full directory */ |
1667 | /* read a full directory */ |
1573 | static void |
1668 | static void |
1574 | eio__scandir (eio_req *req, etp_worker *self) |
1669 | eio__scandir (eio_req *req, etp_worker *self) |
1575 | { |
1670 | { |
1576 | DIR *dirp; |
|
|
1577 | EIO_STRUCT_DIRENT *entp; |
|
|
1578 | char *name, *names; |
1671 | char *name, *names; |
1579 | int namesalloc = 4096; |
1672 | int namesalloc = 4096 - sizeof (void *) * 4; |
1580 | int namesoffs = 0; |
1673 | int namesoffs = 0; |
1581 | int flags = req->int1; |
1674 | int flags = req->int1; |
1582 | eio_dirent *dents = 0; |
1675 | eio_dirent *dents = 0; |
1583 | int dentalloc = 128; |
1676 | int dentalloc = 128; |
1584 | int dentoffs = 0; |
1677 | int dentoffs = 0; |
1585 | ino_t inode_bits = 0; |
1678 | eio_ino_t inode_bits = 0; |
|
|
1679 | #ifdef _WIN32 |
|
|
1680 | HANDLE dirp; |
|
|
1681 | WIN32_FIND_DATA entp; |
|
|
1682 | #else |
|
|
1683 | DIR *dirp; |
|
|
1684 | EIO_STRUCT_DIRENT *entp; |
|
|
1685 | #endif |
1586 | |
1686 | |
1587 | req->result = -1; |
1687 | req->result = -1; |
1588 | |
1688 | |
1589 | if (!(flags & EIO_READDIR_DENTS)) |
1689 | if (!(flags & EIO_READDIR_DENTS)) |
1590 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1690 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
1591 | |
1691 | |
1592 | X_LOCK (wrklock); |
1692 | #ifdef _WIN32 |
1593 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1693 | { |
|
|
1694 | int len = strlen ((const char *)req->ptr1); |
|
|
1695 | char *path = malloc (MAX_PATH); |
|
|
1696 | const char *fmt; |
|
|
1697 | |
|
|
1698 | if (!len) |
|
|
1699 | fmt = "./*"; |
|
|
1700 | else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\') |
|
|
1701 | fmt = "%s*"; |
|
|
1702 | else |
|
|
1703 | fmt = "%s/*"; |
|
|
1704 | |
|
|
1705 | _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1); |
|
|
1706 | dirp = FindFirstFile (path, &entp); |
|
|
1707 | free (path); |
|
|
1708 | |
|
|
1709 | if (dirp == INVALID_HANDLE_VALUE) |
|
|
1710 | { |
|
|
1711 | dirp = 0; |
|
|
1712 | |
|
|
1713 | /* should steal _dosmaperr */ |
|
|
1714 | switch (GetLastError ()) |
|
|
1715 | { |
|
|
1716 | case ERROR_FILE_NOT_FOUND: |
|
|
1717 | req->result = 0; |
|
|
1718 | break; |
|
|
1719 | |
|
|
1720 | case ERROR_INVALID_NAME: |
|
|
1721 | case ERROR_PATH_NOT_FOUND: |
|
|
1722 | case ERROR_NO_MORE_FILES: |
|
|
1723 | errno = ENOENT; |
|
|
1724 | break; |
|
|
1725 | |
|
|
1726 | case ERROR_NOT_ENOUGH_MEMORY: |
|
|
1727 | errno = ENOMEM; |
|
|
1728 | break; |
|
|
1729 | |
|
|
1730 | default: |
|
|
1731 | errno = EINVAL; |
|
|
1732 | break; |
|
|
1733 | } |
|
|
1734 | } |
|
|
1735 | } |
|
|
1736 | #else |
1594 | self->dirp = dirp = opendir (req->ptr1); |
1737 | dirp = opendir (req->ptr1); |
|
|
1738 | #endif |
1595 | |
1739 | |
1596 | if (req->flags & EIO_FLAG_PTR1_FREE) |
1740 | if (req->flags & EIO_FLAG_PTR1_FREE) |
1597 | free (req->ptr1); |
1741 | free (req->ptr1); |
1598 | |
1742 | |
1599 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1743 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
1600 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1744 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
1601 | req->ptr2 = names = malloc (namesalloc); |
1745 | req->ptr2 = names = malloc (namesalloc); |
1602 | X_UNLOCK (wrklock); |
|
|
1603 | |
1746 | |
1604 | if (dirp && names && (!flags || dents)) |
1747 | if (dirp && names && (!flags || dents)) |
1605 | for (;;) |
1748 | for (;;) |
1606 | { |
1749 | { |
|
|
1750 | int done; |
|
|
1751 | |
|
|
1752 | #ifdef _WIN32 |
|
|
1753 | done = !dirp; |
|
|
1754 | #else |
1607 | errno = 0; |
1755 | errno = 0; |
1608 | entp = readdir (dirp); |
1756 | entp = readdir (dirp); |
|
|
1757 | done = !entp; |
|
|
1758 | #endif |
1609 | |
1759 | |
1610 | if (!entp) |
1760 | if (done) |
1611 | { |
1761 | { |
|
|
1762 | #ifndef _WIN32 |
|
|
1763 | int old_errno = errno; |
|
|
1764 | closedir (dirp); |
|
|
1765 | errno = old_errno; |
|
|
1766 | |
1612 | if (errno) |
1767 | if (errno) |
1613 | break; |
1768 | break; |
|
|
1769 | #endif |
1614 | |
1770 | |
1615 | /* sort etc. */ |
1771 | /* sort etc. */ |
1616 | req->int1 = flags; |
1772 | req->int1 = flags; |
1617 | req->result = dentoffs; |
1773 | req->result = dentoffs; |
1618 | |
1774 | |
… | |
… | |
1647 | |
1803 | |
1648 | break; |
1804 | break; |
1649 | } |
1805 | } |
1650 | |
1806 | |
1651 | /* now add the entry to our list(s) */ |
1807 | /* now add the entry to our list(s) */ |
1652 | name = entp->d_name; |
1808 | name = D_NAME (entp); |
1653 | |
1809 | |
1654 | /* skip . and .. entries */ |
1810 | /* skip . and .. entries */ |
1655 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1811 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1656 | { |
1812 | { |
1657 | int len = D_NAMLEN (entp) + 1; |
1813 | int len = D_NAMLEN (entp) + 1; |
1658 | |
1814 | |
1659 | while (ecb_expect_false (namesoffs + len > namesalloc)) |
1815 | while (ecb_expect_false (namesoffs + len > namesalloc)) |
1660 | { |
1816 | { |
1661 | namesalloc *= 2; |
1817 | namesalloc *= 2; |
1662 | X_LOCK (wrklock); |
|
|
1663 | req->ptr2 = names = realloc (names, namesalloc); |
1818 | req->ptr2 = names = realloc (names, namesalloc); |
1664 | X_UNLOCK (wrklock); |
|
|
1665 | |
1819 | |
1666 | if (!names) |
1820 | if (!names) |
1667 | break; |
1821 | break; |
1668 | } |
1822 | } |
1669 | |
1823 | |
… | |
… | |
1674 | struct eio_dirent *ent; |
1828 | struct eio_dirent *ent; |
1675 | |
1829 | |
1676 | if (ecb_expect_false (dentoffs == dentalloc)) |
1830 | if (ecb_expect_false (dentoffs == dentalloc)) |
1677 | { |
1831 | { |
1678 | dentalloc *= 2; |
1832 | dentalloc *= 2; |
1679 | X_LOCK (wrklock); |
|
|
1680 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1833 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
1681 | X_UNLOCK (wrklock); |
|
|
1682 | |
1834 | |
1683 | if (!dents) |
1835 | if (!dents) |
1684 | break; |
1836 | break; |
1685 | } |
1837 | } |
1686 | |
1838 | |
… | |
… | |
1766 | if (EIO_CANCELLED (req)) |
1918 | if (EIO_CANCELLED (req)) |
1767 | { |
1919 | { |
1768 | errno = ECANCELED; |
1920 | errno = ECANCELED; |
1769 | break; |
1921 | break; |
1770 | } |
1922 | } |
|
|
1923 | |
|
|
1924 | #ifdef _WIN32 |
|
|
1925 | if (!FindNextFile (dirp, &entp)) |
|
|
1926 | { |
|
|
1927 | FindClose (dirp); |
|
|
1928 | dirp = 0; |
|
|
1929 | } |
|
|
1930 | #endif |
1771 | } |
1931 | } |
1772 | } |
1932 | } |
1773 | |
1933 | |
1774 | /*****************************************************************************/ |
1934 | /*****************************************************************************/ |
1775 | |
1935 | |
… | |
… | |
1792 | { |
1952 | { |
1793 | ETP_REQ *req; |
1953 | ETP_REQ *req; |
1794 | struct timespec ts; |
1954 | struct timespec ts; |
1795 | etp_worker *self = (etp_worker *)thr_arg; |
1955 | etp_worker *self = (etp_worker *)thr_arg; |
1796 | |
1956 | |
|
|
1957 | #if HAVE_PRCTL_SET_NAME |
|
|
1958 | prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0); |
|
|
1959 | #endif |
|
|
1960 | |
1797 | /* try to distribute timeouts somewhat randomly */ |
1961 | /* try to distribute timeouts somewhat evenly */ |
1798 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1962 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1799 | |
1963 | |
1800 | for (;;) |
1964 | for (;;) |
1801 | { |
1965 | { |
|
|
1966 | ts.tv_sec = 0; |
|
|
1967 | |
1802 | X_LOCK (reqlock); |
1968 | X_LOCK (reqlock); |
1803 | |
1969 | |
1804 | for (;;) |
1970 | for (;;) |
1805 | { |
1971 | { |
1806 | self->req = req = reqq_shift (&req_queue); |
1972 | self->req = req = reqq_shift (&req_queue); |
1807 | |
1973 | |
1808 | if (req) |
1974 | if (req) |
1809 | break; |
1975 | break; |
1810 | |
1976 | |
|
|
1977 | if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ |
|
|
1978 | { |
|
|
1979 | X_UNLOCK (reqlock); |
|
|
1980 | X_LOCK (wrklock); |
|
|
1981 | --started; |
|
|
1982 | X_UNLOCK (wrklock); |
|
|
1983 | goto quit; |
|
|
1984 | } |
|
|
1985 | |
1811 | ++idle; |
1986 | ++idle; |
1812 | |
1987 | |
1813 | ts.tv_sec = time (0) + idle_timeout; |
1988 | if (idle <= max_idle) |
1814 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1989 | /* we are allowed to idle, so do so without any timeout */ |
|
|
1990 | X_COND_WAIT (reqwait, reqlock); |
|
|
1991 | else |
1815 | { |
1992 | { |
1816 | if (idle > max_idle) |
1993 | /* initialise timeout once */ |
1817 | { |
1994 | if (!ts.tv_sec) |
1818 | --idle; |
1995 | ts.tv_sec = time (0) + idle_timeout; |
1819 | X_UNLOCK (reqlock); |
|
|
1820 | X_LOCK (wrklock); |
|
|
1821 | --started; |
|
|
1822 | X_UNLOCK (wrklock); |
|
|
1823 | goto quit; |
|
|
1824 | } |
|
|
1825 | |
1996 | |
1826 | /* we are allowed to idle, so do so without any timeout */ |
|
|
1827 | X_COND_WAIT (reqwait, reqlock); |
1997 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
|
|
1998 | ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */ |
1828 | } |
1999 | } |
1829 | |
2000 | |
1830 | --idle; |
2001 | --idle; |
1831 | } |
2002 | } |
1832 | |
2003 | |
… | |
… | |
1863 | /*****************************************************************************/ |
2034 | /*****************************************************************************/ |
1864 | |
2035 | |
1865 | int ecb_cold |
2036 | int ecb_cold |
1866 | eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
2037 | eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1867 | { |
2038 | { |
|
|
2039 | #if !HAVE_PREADWRITE |
|
|
2040 | X_MUTEX_CREATE (preadwritelock); |
|
|
2041 | #endif |
|
|
2042 | |
1868 | return etp_init (want_poll, done_poll); |
2043 | return etp_init (want_poll, done_poll); |
1869 | } |
2044 | } |
1870 | |
2045 | |
1871 | ecb_inline void |
2046 | ecb_inline void |
1872 | eio_api_destroy (eio_req *req) |
2047 | eio_api_destroy (eio_req *req) |
… | |
… | |
1917 | case EIO_WRITE: req->result = req->offs >= 0 |
2092 | case EIO_WRITE: req->result = req->offs >= 0 |
1918 | ? pwrite (req->int1, req->ptr2, req->size, req->offs) |
2093 | ? pwrite (req->int1, req->ptr2, req->size, req->offs) |
1919 | : write (req->int1, req->ptr2, req->size); break; |
2094 | : write (req->int1, req->ptr2, req->size); break; |
1920 | |
2095 | |
1921 | case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; |
2096 | case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; |
1922 | case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; |
2097 | case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break; |
1923 | |
2098 | |
1924 | case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
2099 | case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1925 | req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
2100 | req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1926 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
2101 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1927 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
2102 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
… | |
… | |
1957 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
2132 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1958 | |
2133 | |
1959 | case EIO_SYNC: req->result = 0; sync (); break; |
2134 | case EIO_SYNC: req->result = 0; sync (); break; |
1960 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
2135 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1961 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
2136 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
|
|
2137 | case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break; |
|
|
2138 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
1962 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
2139 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
1963 | case EIO_MTOUCH: req->result = eio__mtouch (req); break; |
2140 | case EIO_MTOUCH: req->result = eio__mtouch (req); break; |
1964 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
2141 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
1965 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
2142 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
1966 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
2143 | case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; |
1967 | |
2144 | |
1968 | case EIO_READDIR: eio__scandir (req, self); break; |
2145 | case EIO_READDIR: eio__scandir (req, self); break; |
1969 | |
2146 | |
1970 | case EIO_BUSY: |
2147 | case EIO_BUSY: |
1971 | #ifdef _WIN32 |
2148 | #ifdef _WIN32 |
… | |
… | |
2051 | eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
2228 | eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
2052 | { |
2229 | { |
2053 | REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
2230 | REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
2054 | } |
2231 | } |
2055 | |
2232 | |
|
|
2233 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
|
|
2234 | { |
|
|
2235 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
|
|
2236 | } |
|
|
2237 | |
|
|
2238 | eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data) |
|
|
2239 | { |
|
|
2240 | REQ (EIO_SYNCFS); req->int1 = fd; SEND; |
|
|
2241 | } |
|
|
2242 | |
|
|
2243 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
|
|
2244 | { |
|
|
2245 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
|
|
2246 | } |
|
|
2247 | |
2056 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
2248 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
2057 | { |
2249 | { |
2058 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
2250 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
2059 | } |
2251 | } |
2060 | |
2252 | |
… | |
… | |
2066 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
2258 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
2067 | { |
2259 | { |
2068 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
2260 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
2069 | } |
2261 | } |
2070 | |
2262 | |
2071 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
2263 | eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) |
2072 | { |
2264 | { |
2073 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
2265 | REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; |
2074 | } |
|
|
2075 | |
|
|
2076 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
|
|
2077 | { |
|
|
2078 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
|
|
2079 | } |
2266 | } |
2080 | |
2267 | |
2081 | eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) |
2268 | eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) |
2082 | { |
2269 | { |
2083 | REQ (EIO_CLOSE); req->int1 = fd; SEND; |
2270 | REQ (EIO_CLOSE); req->int1 = fd; SEND; |
… | |
… | |
2121 | eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) |
2308 | eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) |
2122 | { |
2309 | { |
2123 | REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; |
2310 | REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; |
2124 | } |
2311 | } |
2125 | |
2312 | |
2126 | eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) |
2313 | eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data) |
2127 | { |
2314 | { |
2128 | REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
2315 | REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
2129 | } |
2316 | } |
2130 | |
2317 | |
2131 | eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) |
2318 | eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) |
… | |
… | |
2151 | eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) |
2338 | eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) |
2152 | { |
2339 | { |
2153 | REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; |
2340 | REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; |
2154 | } |
2341 | } |
2155 | |
2342 | |
2156 | eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) |
2343 | eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data) |
2157 | { |
2344 | { |
2158 | REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
2345 | REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; |
2159 | } |
2346 | } |
2160 | |
2347 | |
2161 | eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
2348 | eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) |
… | |
… | |
2308 | } |
2495 | } |
2309 | |
2496 | |
2310 | /*****************************************************************************/ |
2497 | /*****************************************************************************/ |
2311 | /* misc garbage */ |
2498 | /* misc garbage */ |
2312 | |
2499 | |
2313 | ssize_t |
2500 | eio_ssize_t |
2314 | eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2501 | eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2315 | { |
2502 | { |
2316 | etp_worker wrk; |
|
|
2317 | ssize_t ret; |
|
|
2318 | |
|
|
2319 | wrk.dbuf = 0; |
|
|
2320 | |
|
|
2321 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
2503 | return eio__sendfile (ofd, ifd, offset, count); |
2322 | |
|
|
2323 | if (wrk.dbuf) |
|
|
2324 | free (wrk.dbuf); |
|
|
2325 | |
|
|
2326 | return ret; |
|
|
2327 | } |
2504 | } |
2328 | |
2505 | |