1 | /* |
1 | /* |
2 | * libeio implementation |
2 | * libeio implementation |
3 | * |
3 | * |
4 | * Copyright (c) 2007,2008 Marc Alexander Lehmann <libeio@schmorp.de> |
4 | * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> |
5 | * All rights reserved. |
5 | * All rights reserved. |
6 | * |
6 | * |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
7 | * Redistribution and use in source and binary forms, with or without modifica- |
8 | * tion, are permitted provided that the following conditions are met: |
8 | * tion, are permitted provided that the following conditions are met: |
9 | * |
9 | * |
… | |
… | |
35 | * and other provisions required by the GPL. If you do not delete the |
35 | * and other provisions required by the GPL. If you do not delete the |
36 | * provisions above, a recipient may use your version of this file under |
36 | * provisions above, a recipient may use your version of this file under |
37 | * either the BSD or the GPL. |
37 | * either the BSD or the GPL. |
38 | */ |
38 | */ |
39 | |
39 | |
|
|
40 | #ifndef _WIN32 |
|
|
41 | # include "config.h" |
|
|
42 | #endif |
|
|
43 | |
40 | #include "eio.h" |
44 | #include "eio.h" |
|
|
45 | #include "ecb.h" |
|
|
46 | |
|
|
47 | #ifdef EIO_STACKSIZE |
|
|
48 | # define XTHREAD_STACKSIZE EIO_STACKSIZE |
|
|
49 | #endif |
41 | #include "xthread.h" |
50 | #include "xthread.h" |
42 | |
51 | |
43 | #include <errno.h> |
52 | #include <errno.h> |
44 | #include <stddef.h> |
53 | #include <stddef.h> |
45 | #include <stdlib.h> |
54 | #include <stdlib.h> |
… | |
… | |
49 | #include <sys/stat.h> |
58 | #include <sys/stat.h> |
50 | #include <limits.h> |
59 | #include <limits.h> |
51 | #include <fcntl.h> |
60 | #include <fcntl.h> |
52 | #include <assert.h> |
61 | #include <assert.h> |
53 | |
62 | |
|
|
63 | #include <sys/statvfs.h> |
|
|
64 | /* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ |
|
|
65 | /* intptr_t only comes from stdint.h, says idiot openbsd coder */ |
|
|
66 | #if HAVE_STDINT_H |
|
|
67 | # include <stdint.h> |
|
|
68 | #endif |
|
|
69 | |
|
|
70 | #ifndef ECANCELED |
|
|
71 | # define ECANCELED EDOM |
|
|
72 | #endif |
|
|
73 | |
|
|
74 | static void eio_destroy (eio_req *req); |
|
|
75 | |
54 | #ifndef EIO_FINISH |
76 | #ifndef EIO_FINISH |
55 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
77 | # define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 |
56 | #endif |
78 | #endif |
57 | |
79 | |
58 | #ifndef EIO_DESTROY |
80 | #ifndef EIO_DESTROY |
… | |
… | |
61 | |
83 | |
62 | #ifndef EIO_FEED |
84 | #ifndef EIO_FEED |
63 | # define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
85 | # define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) |
64 | #endif |
86 | #endif |
65 | |
87 | |
|
|
88 | #ifndef EIO_FD_TO_WIN32_HANDLE |
|
|
89 | # define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd) |
|
|
90 | #endif |
|
|
91 | #ifndef EIO_WIN32_HANDLE_TO_FD |
|
|
92 | # define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0) |
|
|
93 | #endif |
|
|
94 | |
|
|
95 | #define EIO_ERRNO(errval,retval) ((errno = errval), retval) |
|
|
96 | |
|
|
97 | #define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1) |
|
|
98 | |
66 | #ifdef _WIN32 |
99 | #ifdef _WIN32 |
67 | |
100 | |
68 | /*doh*/ |
101 | #define PAGESIZE 4096 /* GetSystemInfo? */ |
|
|
102 | |
|
|
103 | #ifdef EIO_STRUCT_STATI64 |
|
|
104 | #define stat(path,buf) _stati64 (path,buf) |
|
|
105 | #define fstat(fd,buf) _fstati64 (path,buf) |
|
|
106 | #endif |
|
|
107 | #define lstat(path,buf) stat (path,buf) |
|
|
108 | #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) |
|
|
109 | #define mkdir(path,mode) _mkdir (path) |
|
|
110 | #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) |
|
|
111 | |
|
|
112 | #define chown(path,uid,gid) EIO_ENOSYS () |
|
|
113 | #define fchown(fd,uid,gid) EIO_ENOSYS () |
|
|
114 | #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ |
|
|
115 | #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */ |
|
|
116 | #define mknod(path,mode,dev) EIO_ENOSYS () |
|
|
117 | #define sync() EIO_ENOSYS () |
|
|
118 | |
|
|
119 | /* we could even stat and see if it exists */ |
|
|
120 | static int |
|
|
121 | symlink (const char *old, const char *neu) |
|
|
122 | { |
|
|
123 | if (CreateSymbolicLink (neu, old, 1)) |
|
|
124 | return 0; |
|
|
125 | |
|
|
126 | if (CreateSymbolicLink (neu, old, 0)) |
|
|
127 | return 0; |
|
|
128 | |
|
|
129 | return EIO_ERRNO (ENOENT, -1); |
|
|
130 | } |
69 | |
131 | |
70 | #else |
132 | #else |
71 | |
133 | |
72 | # include "config.h" |
|
|
73 | # include <sys/time.h> |
134 | #include <sys/time.h> |
74 | # include <sys/select.h> |
135 | #include <sys/select.h> |
|
|
136 | #include <sys/statvfs.h> |
75 | # include <unistd.h> |
137 | #include <unistd.h> |
76 | # include <utime.h> |
138 | #include <utime.h> |
77 | # include <signal.h> |
139 | #include <signal.h> |
78 | # include <dirent.h> |
140 | #include <dirent.h> |
79 | |
141 | |
|
|
142 | #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES |
|
|
143 | #include <sys/mman.h> |
|
|
144 | #endif |
|
|
145 | |
|
|
146 | /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ |
|
|
147 | #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ |
|
|
148 | #define _DIRENT_HAVE_D_TYPE /* sigh */ |
|
|
149 | #define D_INO(de) (de)->d_fileno |
|
|
150 | #define D_NAMLEN(de) (de)->d_namlen |
|
|
151 | #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 |
|
|
152 | #define D_INO(de) (de)->d_ino |
|
|
153 | #endif |
|
|
154 | |
|
|
155 | #ifdef _D_EXACT_NAMLEN |
|
|
156 | #undef D_NAMLEN |
|
|
157 | #define D_NAMLEN(de) _D_EXACT_NAMLEN (de) |
|
|
158 | #endif |
|
|
159 | |
|
|
160 | #ifdef _DIRENT_HAVE_D_TYPE |
|
|
161 | #define D_TYPE(de) (de)->d_type |
|
|
162 | #endif |
|
|
163 | |
80 | # ifndef EIO_STRUCT_DIRENT |
164 | #ifndef EIO_STRUCT_DIRENT |
81 | # define EIO_STRUCT_DIRENT struct dirent |
165 | #define EIO_STRUCT_DIRENT struct dirent |
82 | # endif |
166 | #endif |
83 | |
167 | |
84 | #endif |
168 | #endif |
85 | |
169 | |
86 | #if HAVE_SENDFILE |
170 | #if HAVE_SENDFILE |
87 | # if __linux |
171 | # if __linux |
88 | # include <sys/sendfile.h> |
172 | # include <sys/sendfile.h> |
89 | # elif __freebsd |
173 | # elif __FreeBSD__ || defined __APPLE__ |
90 | # include <sys/socket.h> |
174 | # include <sys/socket.h> |
91 | # include <sys/uio.h> |
175 | # include <sys/uio.h> |
92 | # elif __hpux |
176 | # elif __hpux |
93 | # include <sys/socket.h> |
177 | # include <sys/socket.h> |
94 | # elif __solaris /* not yet */ |
178 | # elif __solaris |
95 | # include <sys/sendfile.h> |
179 | # include <sys/sendfile.h> |
96 | # else |
180 | # else |
97 | # error sendfile support requested but not available |
181 | # error sendfile support requested but not available |
98 | # endif |
182 | # endif |
99 | #endif |
183 | #endif |
100 | |
184 | |
101 | /* number of seconds after which an idle threads exit */ |
185 | #ifndef D_TYPE |
102 | #define IDLE_TIMEOUT 10 |
186 | # define D_TYPE(de) 0 |
|
|
187 | #endif |
|
|
188 | #ifndef D_INO |
|
|
189 | # define D_INO(de) 0 |
|
|
190 | #endif |
|
|
191 | #ifndef D_NAMLEN |
|
|
192 | # define D_NAMLEN(de) strlen ((de)->d_name) |
|
|
193 | #endif |
103 | |
194 | |
104 | /* used for struct dirent, AIX doesn't provide it */ |
195 | /* used for struct dirent, AIX doesn't provide it */ |
105 | #ifndef NAME_MAX |
196 | #ifndef NAME_MAX |
106 | # define NAME_MAX 4096 |
197 | # define NAME_MAX 4096 |
|
|
198 | #endif |
|
|
199 | |
|
|
200 | /* used for readlink etc. */ |
|
|
201 | #ifndef PATH_MAX |
|
|
202 | # define PATH_MAX 4096 |
107 | #endif |
203 | #endif |
108 | |
204 | |
109 | /* buffer size for various temporary buffers */ |
205 | /* buffer size for various temporary buffers */ |
110 | #define EIO_BUFSIZE 65536 |
206 | #define EIO_BUFSIZE 65536 |
111 | |
207 | |
… | |
… | |
117 | errno = ENOMEM; \ |
213 | errno = ENOMEM; \ |
118 | if (!eio_buf) \ |
214 | if (!eio_buf) \ |
119 | return -1; |
215 | return -1; |
120 | |
216 | |
121 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
217 | #define EIO_TICKS ((1000000 + 1023) >> 10) |
122 | |
|
|
123 | /*****************************************************************************/ |
|
|
124 | |
218 | |
125 | #define ETP_PRI_MIN EIO_PRI_MIN |
219 | #define ETP_PRI_MIN EIO_PRI_MIN |
126 | #define ETP_PRI_MAX EIO_PRI_MAX |
220 | #define ETP_PRI_MAX EIO_PRI_MAX |
127 | |
221 | |
128 | struct etp_worker; |
222 | struct etp_worker; |
… | |
… | |
144 | if (wrk->dirp) \ |
238 | if (wrk->dirp) \ |
145 | { \ |
239 | { \ |
146 | closedir (wrk->dirp); \ |
240 | closedir (wrk->dirp); \ |
147 | wrk->dirp = 0; \ |
241 | wrk->dirp = 0; \ |
148 | } |
242 | } |
|
|
243 | |
149 | #define ETP_WORKER_COMMON \ |
244 | #define ETP_WORKER_COMMON \ |
150 | void *dbuf; \ |
245 | void *dbuf; \ |
151 | DIR *dirp; |
246 | DIR *dirp; |
152 | |
247 | |
153 | /*****************************************************************************/ |
248 | /*****************************************************************************/ |
154 | |
249 | |
155 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
250 | #define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) |
156 | |
251 | |
157 | /* calculcate time difference in ~1/EIO_TICKS of a second */ |
252 | /* calculate time difference in ~1/EIO_TICKS of a second */ |
|
|
253 | ecb_inline int |
158 | static int tvdiff (struct timeval *tv1, struct timeval *tv2) |
254 | tvdiff (struct timeval *tv1, struct timeval *tv2) |
159 | { |
255 | { |
160 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
256 | return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS |
161 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
257 | + ((tv2->tv_usec - tv1->tv_usec) >> 10); |
162 | } |
258 | } |
163 | |
259 | |
… | |
… | |
170 | static unsigned int max_poll_reqs; /* reslock */ |
266 | static unsigned int max_poll_reqs; /* reslock */ |
171 | |
267 | |
172 | static volatile unsigned int nreqs; /* reqlock */ |
268 | static volatile unsigned int nreqs; /* reqlock */ |
173 | static volatile unsigned int nready; /* reqlock */ |
269 | static volatile unsigned int nready; /* reqlock */ |
174 | static volatile unsigned int npending; /* reqlock */ |
270 | static volatile unsigned int npending; /* reqlock */ |
175 | static volatile unsigned int max_idle = 4; |
271 | static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ |
|
|
272 | static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ |
176 | |
273 | |
177 | static mutex_t wrklock = X_MUTEX_INIT; |
274 | static xmutex_t wrklock; |
178 | static mutex_t reslock = X_MUTEX_INIT; |
275 | static xmutex_t reslock; |
179 | static mutex_t reqlock = X_MUTEX_INIT; |
276 | static xmutex_t reqlock; |
180 | static cond_t reqwait = X_COND_INIT; |
277 | static xcond_t reqwait; |
181 | |
278 | |
182 | #if !HAVE_PREADWRITE |
279 | #if !HAVE_PREADWRITE |
183 | /* |
280 | /* |
184 | * make our pread/pwrite emulation safe against themselves, but not against |
281 | * make our pread/pwrite emulation safe against themselves, but not against |
185 | * normal read/write by using a mutex. slows down execution a lot, |
282 | * normal read/write by using a mutex. slows down execution a lot, |
186 | * but that's your problem, not mine. |
283 | * but that's your problem, not mine. |
187 | */ |
284 | */ |
188 | static mutex_t preadwritelock = X_MUTEX_INIT; |
285 | static xmutex_t preadwritelock = X_MUTEX_INIT; |
189 | #endif |
286 | #endif |
190 | |
287 | |
191 | typedef struct etp_worker |
288 | typedef struct etp_worker |
192 | { |
289 | { |
193 | /* locked by wrklock */ |
290 | /* locked by wrklock */ |
194 | struct etp_worker *prev, *next; |
291 | struct etp_worker *prev, *next; |
195 | |
292 | |
196 | thread_t tid; |
293 | xthread_t tid; |
197 | |
294 | |
198 | /* locked by reslock, reqlock or wrklock */ |
295 | /* locked by reslock, reqlock or wrklock */ |
199 | ETP_REQ *req; /* currently processed request */ |
296 | ETP_REQ *req; /* currently processed request */ |
200 | |
297 | |
201 | ETP_WORKER_COMMON |
298 | ETP_WORKER_COMMON |
… | |
… | |
206 | #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
303 | #define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) |
207 | #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
304 | #define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) |
208 | |
305 | |
209 | /* worker threads management */ |
306 | /* worker threads management */ |
210 | |
307 | |
|
|
308 | static void ecb_cold |
211 | static void etp_worker_clear (etp_worker *wrk) |
309 | etp_worker_clear (etp_worker *wrk) |
212 | { |
310 | { |
213 | ETP_WORKER_CLEAR (wrk); |
311 | ETP_WORKER_CLEAR (wrk); |
214 | } |
312 | } |
215 | |
313 | |
|
|
314 | static void ecb_cold |
216 | static void etp_worker_free (etp_worker *wrk) |
315 | etp_worker_free (etp_worker *wrk) |
217 | { |
316 | { |
218 | wrk->next->prev = wrk->prev; |
317 | wrk->next->prev = wrk->prev; |
219 | wrk->prev->next = wrk->next; |
318 | wrk->prev->next = wrk->next; |
220 | |
319 | |
221 | free (wrk); |
320 | free (wrk); |
222 | } |
321 | } |
223 | |
322 | |
224 | static unsigned int etp_nreqs (void) |
323 | static unsigned int |
|
|
324 | etp_nreqs (void) |
225 | { |
325 | { |
226 | int retval; |
326 | int retval; |
227 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
327 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
228 | retval = nreqs; |
328 | retval = nreqs; |
229 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
329 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
230 | return retval; |
330 | return retval; |
231 | } |
331 | } |
232 | |
332 | |
233 | static unsigned int etp_nready (void) |
333 | static unsigned int |
|
|
334 | etp_nready (void) |
234 | { |
335 | { |
235 | unsigned int retval; |
336 | unsigned int retval; |
236 | |
337 | |
237 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
338 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
238 | retval = nready; |
339 | retval = nready; |
239 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
340 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
240 | |
341 | |
241 | return retval; |
342 | return retval; |
242 | } |
343 | } |
243 | |
344 | |
244 | static unsigned int etp_npending (void) |
345 | static unsigned int |
|
|
346 | etp_npending (void) |
245 | { |
347 | { |
246 | unsigned int retval; |
348 | unsigned int retval; |
247 | |
349 | |
248 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
350 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
249 | retval = npending; |
351 | retval = npending; |
250 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
352 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
251 | |
353 | |
252 | return retval; |
354 | return retval; |
253 | } |
355 | } |
254 | |
356 | |
255 | static unsigned int etp_nthreads (void) |
357 | static unsigned int |
|
|
358 | etp_nthreads (void) |
256 | { |
359 | { |
257 | unsigned int retval; |
360 | unsigned int retval; |
258 | |
361 | |
259 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
362 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
260 | retval = started; |
363 | retval = started; |
… | |
… | |
274 | } etp_reqq; |
377 | } etp_reqq; |
275 | |
378 | |
276 | static etp_reqq req_queue; |
379 | static etp_reqq req_queue; |
277 | static etp_reqq res_queue; |
380 | static etp_reqq res_queue; |
278 | |
381 | |
|
|
382 | static int ecb_noinline |
279 | static int reqq_push (etp_reqq *q, ETP_REQ *req) |
383 | reqq_push (etp_reqq *q, ETP_REQ *req) |
280 | { |
384 | { |
281 | int pri = req->pri; |
385 | int pri = req->pri; |
282 | req->next = 0; |
386 | req->next = 0; |
283 | |
387 | |
284 | if (q->qe[pri]) |
388 | if (q->qe[pri]) |
… | |
… | |
290 | q->qe[pri] = q->qs[pri] = req; |
394 | q->qe[pri] = q->qs[pri] = req; |
291 | |
395 | |
292 | return q->size++; |
396 | return q->size++; |
293 | } |
397 | } |
294 | |
398 | |
|
|
399 | static ETP_REQ * ecb_noinline |
295 | static ETP_REQ *reqq_shift (etp_reqq *q) |
400 | reqq_shift (etp_reqq *q) |
296 | { |
401 | { |
297 | int pri; |
402 | int pri; |
298 | |
403 | |
299 | if (!q->size) |
404 | if (!q->size) |
300 | return 0; |
405 | return 0; |
… | |
… | |
315 | } |
420 | } |
316 | |
421 | |
317 | abort (); |
422 | abort (); |
318 | } |
423 | } |
319 | |
424 | |
320 | static void etp_atfork_prepare (void) |
425 | static void ecb_cold |
|
|
426 | etp_thread_init (void) |
321 | { |
427 | { |
322 | X_LOCK (wrklock); |
|
|
323 | X_LOCK (reqlock); |
|
|
324 | X_LOCK (reslock); |
|
|
325 | #if !HAVE_PREADWRITE |
428 | #if !HAVE_PREADWRITE |
326 | X_LOCK (preadwritelock); |
429 | X_MUTEX_CREATE (preadwritelock); |
327 | #endif |
430 | #endif |
|
|
431 | X_MUTEX_CREATE (wrklock); |
|
|
432 | X_MUTEX_CREATE (reslock); |
|
|
433 | X_MUTEX_CREATE (reqlock); |
|
|
434 | X_COND_CREATE (reqwait); |
328 | } |
435 | } |
329 | |
436 | |
|
|
437 | static void ecb_cold |
|
|
438 | etp_atfork_prepare (void) |
|
|
439 | { |
|
|
440 | } |
|
|
441 | |
|
|
442 | static void ecb_cold |
330 | static void etp_atfork_parent (void) |
443 | etp_atfork_parent (void) |
331 | { |
444 | { |
332 | #if !HAVE_PREADWRITE |
|
|
333 | X_UNLOCK (preadwritelock); |
|
|
334 | #endif |
|
|
335 | X_UNLOCK (reslock); |
|
|
336 | X_UNLOCK (reqlock); |
|
|
337 | X_UNLOCK (wrklock); |
|
|
338 | } |
445 | } |
339 | |
446 | |
|
|
447 | static void ecb_cold |
340 | static void etp_atfork_child (void) |
448 | etp_atfork_child (void) |
341 | { |
449 | { |
342 | ETP_REQ *prv; |
450 | ETP_REQ *prv; |
343 | |
451 | |
344 | while ((prv = reqq_shift (&req_queue))) |
452 | while ((prv = reqq_shift (&req_queue))) |
345 | ETP_DESTROY (prv); |
453 | ETP_DESTROY (prv); |
… | |
… | |
362 | idle = 0; |
470 | idle = 0; |
363 | nreqs = 0; |
471 | nreqs = 0; |
364 | nready = 0; |
472 | nready = 0; |
365 | npending = 0; |
473 | npending = 0; |
366 | |
474 | |
367 | etp_atfork_parent (); |
475 | etp_thread_init (); |
368 | } |
476 | } |
369 | |
477 | |
370 | static void |
478 | static void ecb_cold |
371 | etp_once_init (void) |
479 | etp_once_init (void) |
372 | { |
480 | { |
|
|
481 | etp_thread_init (); |
373 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
482 | X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child); |
374 | } |
483 | } |
375 | |
484 | |
376 | static int |
485 | static int ecb_cold |
377 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
486 | etp_init (void (*want_poll)(void), void (*done_poll)(void)) |
378 | { |
487 | { |
379 | static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
488 | static pthread_once_t doinit = PTHREAD_ONCE_INIT; |
380 | |
489 | |
381 | pthread_once (&doinit, etp_once_init); |
490 | pthread_once (&doinit, etp_once_init); |
… | |
… | |
386 | return 0; |
495 | return 0; |
387 | } |
496 | } |
388 | |
497 | |
389 | X_THREAD_PROC (etp_proc); |
498 | X_THREAD_PROC (etp_proc); |
390 | |
499 | |
|
|
500 | static void ecb_cold |
391 | static void etp_start_thread (void) |
501 | etp_start_thread (void) |
392 | { |
502 | { |
393 | etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
503 | etp_worker *wrk = calloc (1, sizeof (etp_worker)); |
394 | |
504 | |
395 | /*TODO*/ |
505 | /*TODO*/ |
396 | assert (("unable to allocate worker thread data", wrk)); |
506 | assert (("unable to allocate worker thread data", wrk)); |
… | |
… | |
409 | free (wrk); |
519 | free (wrk); |
410 | |
520 | |
411 | X_UNLOCK (wrklock); |
521 | X_UNLOCK (wrklock); |
412 | } |
522 | } |
413 | |
523 | |
|
|
524 | static void |
414 | static void etp_maybe_start_thread (void) |
525 | etp_maybe_start_thread (void) |
415 | { |
526 | { |
416 | if (etp_nthreads () >= wanted) |
527 | if (ecb_expect_true (etp_nthreads () >= wanted)) |
417 | return; |
528 | return; |
418 | |
529 | |
419 | /* todo: maybe use idle here, but might be less exact */ |
530 | /* todo: maybe use idle here, but might be less exact */ |
420 | if (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()) |
531 | if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) |
421 | return; |
532 | return; |
422 | |
533 | |
423 | etp_start_thread (); |
534 | etp_start_thread (); |
424 | } |
535 | } |
425 | |
536 | |
|
|
537 | static void ecb_cold |
426 | static void etp_end_thread (void) |
538 | etp_end_thread (void) |
427 | { |
539 | { |
428 | eio_req *req = calloc (1, sizeof (eio_req)); |
540 | eio_req *req = calloc (1, sizeof (eio_req)); |
429 | |
541 | |
430 | req->type = -1; |
542 | req->type = -1; |
431 | req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
543 | req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
… | |
… | |
438 | X_LOCK (wrklock); |
550 | X_LOCK (wrklock); |
439 | --started; |
551 | --started; |
440 | X_UNLOCK (wrklock); |
552 | X_UNLOCK (wrklock); |
441 | } |
553 | } |
442 | |
554 | |
443 | static int etp_poll (void) |
555 | static int |
|
|
556 | etp_poll (void) |
444 | { |
557 | { |
445 | unsigned int maxreqs; |
558 | unsigned int maxreqs; |
446 | unsigned int maxtime; |
559 | unsigned int maxtime; |
447 | struct timeval tv_start, tv_now; |
560 | struct timeval tv_start, tv_now; |
448 | |
561 | |
… | |
… | |
478 | |
591 | |
479 | X_LOCK (reqlock); |
592 | X_LOCK (reqlock); |
480 | --nreqs; |
593 | --nreqs; |
481 | X_UNLOCK (reqlock); |
594 | X_UNLOCK (reqlock); |
482 | |
595 | |
483 | if (req->type == EIO_GROUP && req->size) |
596 | if (ecb_expect_false (req->type == EIO_GROUP && req->size)) |
484 | { |
597 | { |
485 | req->int1 = 1; /* mark request as delayed */ |
598 | req->int1 = 1; /* mark request as delayed */ |
486 | continue; |
599 | continue; |
487 | } |
600 | } |
488 | else |
601 | else |
489 | { |
602 | { |
490 | int res = ETP_FINISH (req); |
603 | int res = ETP_FINISH (req); |
491 | if (res) |
604 | if (ecb_expect_false (res)) |
492 | return res; |
605 | return res; |
493 | } |
606 | } |
494 | |
607 | |
495 | if (maxreqs && !--maxreqs) |
608 | if (ecb_expect_false (maxreqs && !--maxreqs)) |
496 | break; |
609 | break; |
497 | |
610 | |
498 | if (maxtime) |
611 | if (maxtime) |
499 | { |
612 | { |
500 | gettimeofday (&tv_now, 0); |
613 | gettimeofday (&tv_now, 0); |
… | |
… | |
506 | |
619 | |
507 | errno = EAGAIN; |
620 | errno = EAGAIN; |
508 | return -1; |
621 | return -1; |
509 | } |
622 | } |
510 | |
623 | |
|
|
624 | static void |
511 | static void etp_cancel (ETP_REQ *req) |
625 | etp_cancel (ETP_REQ *req) |
512 | { |
626 | { |
513 | X_LOCK (wrklock); |
627 | req->cancelled = 1; |
514 | req->flags |= EIO_FLAG_CANCELLED; |
|
|
515 | X_UNLOCK (wrklock); |
|
|
516 | |
628 | |
517 | eio_grp_cancel (req); |
629 | eio_grp_cancel (req); |
518 | } |
630 | } |
519 | |
631 | |
|
|
632 | static void |
520 | static void etp_submit (ETP_REQ *req) |
633 | etp_submit (ETP_REQ *req) |
521 | { |
634 | { |
522 | req->pri -= ETP_PRI_MIN; |
635 | req->pri -= ETP_PRI_MIN; |
523 | |
636 | |
524 | if (req->pri < ETP_PRI_MIN - ETP_PRI_MIN) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
637 | if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; |
525 | if (req->pri > ETP_PRI_MAX - ETP_PRI_MIN) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
638 | if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; |
526 | |
639 | |
|
|
640 | if (ecb_expect_false (req->type == EIO_GROUP)) |
|
|
641 | { |
|
|
642 | /* I hope this is worth it :/ */ |
527 | X_LOCK (reqlock); |
643 | X_LOCK (reqlock); |
528 | ++nreqs; |
644 | ++nreqs; |
|
|
645 | X_UNLOCK (reqlock); |
|
|
646 | |
|
|
647 | X_LOCK (reslock); |
|
|
648 | |
|
|
649 | ++npending; |
|
|
650 | |
|
|
651 | if (!reqq_push (&res_queue, req) && want_poll_cb) |
|
|
652 | want_poll_cb (); |
|
|
653 | |
|
|
654 | X_UNLOCK (reslock); |
|
|
655 | } |
|
|
656 | else |
|
|
657 | { |
|
|
658 | X_LOCK (reqlock); |
|
|
659 | ++nreqs; |
529 | ++nready; |
660 | ++nready; |
530 | reqq_push (&req_queue, req); |
661 | reqq_push (&req_queue, req); |
531 | X_COND_SIGNAL (reqwait); |
662 | X_COND_SIGNAL (reqwait); |
532 | X_UNLOCK (reqlock); |
663 | X_UNLOCK (reqlock); |
533 | |
664 | |
534 | etp_maybe_start_thread (); |
665 | etp_maybe_start_thread (); |
|
|
666 | } |
535 | } |
667 | } |
536 | |
668 | |
|
|
669 | static void ecb_cold |
537 | static void etp_set_max_poll_time (double nseconds) |
670 | etp_set_max_poll_time (double nseconds) |
538 | { |
671 | { |
539 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
672 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
540 | max_poll_time = nseconds; |
673 | max_poll_time = nseconds * EIO_TICKS; |
541 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
674 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
542 | } |
675 | } |
543 | |
676 | |
|
|
677 | static void ecb_cold |
544 | static void etp_set_max_poll_reqs (unsigned int maxreqs) |
678 | etp_set_max_poll_reqs (unsigned int maxreqs) |
545 | { |
679 | { |
546 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
680 | if (WORDACCESS_UNSAFE) X_LOCK (reslock); |
547 | max_poll_reqs = maxreqs; |
681 | max_poll_reqs = maxreqs; |
548 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
682 | if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); |
549 | } |
683 | } |
550 | |
684 | |
|
|
685 | static void ecb_cold |
551 | static void etp_set_max_idle (unsigned int nthreads) |
686 | etp_set_max_idle (unsigned int nthreads) |
552 | { |
687 | { |
553 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
688 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
554 | max_idle = nthreads <= 0 ? 1 : nthreads; |
689 | max_idle = nthreads; |
555 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
690 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
556 | } |
691 | } |
557 | |
692 | |
|
|
693 | static void ecb_cold |
|
|
694 | etp_set_idle_timeout (unsigned int seconds) |
|
|
695 | { |
|
|
696 | if (WORDACCESS_UNSAFE) X_LOCK (reqlock); |
|
|
697 | idle_timeout = seconds; |
|
|
698 | if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); |
|
|
699 | } |
|
|
700 | |
|
|
701 | static void ecb_cold |
558 | static void etp_set_min_parallel (unsigned int nthreads) |
702 | etp_set_min_parallel (unsigned int nthreads) |
559 | { |
703 | { |
560 | if (wanted < nthreads) |
704 | if (wanted < nthreads) |
561 | wanted = nthreads; |
705 | wanted = nthreads; |
562 | } |
706 | } |
563 | |
707 | |
|
|
708 | static void ecb_cold |
564 | static void etp_set_max_parallel (unsigned int nthreads) |
709 | etp_set_max_parallel (unsigned int nthreads) |
565 | { |
710 | { |
566 | if (wanted > nthreads) |
711 | if (wanted > nthreads) |
567 | wanted = nthreads; |
712 | wanted = nthreads; |
568 | |
713 | |
569 | while (started > wanted) |
714 | while (started > wanted) |
570 | etp_end_thread (); |
715 | etp_end_thread (); |
571 | } |
716 | } |
572 | |
717 | |
573 | /*****************************************************************************/ |
718 | /*****************************************************************************/ |
574 | |
719 | |
|
|
720 | static void |
575 | static void grp_try_feed (eio_req *grp) |
721 | grp_try_feed (eio_req *grp) |
576 | { |
722 | { |
577 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
723 | while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) |
578 | { |
724 | { |
579 | int old_len = grp->size; |
725 | grp->flags &= ~EIO_FLAG_GROUPADD; |
580 | |
726 | |
581 | EIO_FEED (grp); |
727 | EIO_FEED (grp); |
582 | |
728 | |
583 | /* stop if no progress has been made */ |
729 | /* stop if no progress has been made */ |
584 | if (old_len == grp->size) |
730 | if (!(grp->flags & EIO_FLAG_GROUPADD)) |
585 | { |
731 | { |
586 | grp->feed = 0; |
732 | grp->feed = 0; |
587 | break; |
733 | break; |
588 | } |
734 | } |
589 | } |
735 | } |
590 | } |
736 | } |
591 | |
737 | |
|
|
738 | static int |
592 | static int grp_dec (eio_req *grp) |
739 | grp_dec (eio_req *grp) |
593 | { |
740 | { |
594 | --grp->size; |
741 | --grp->size; |
595 | |
742 | |
596 | /* call feeder, if applicable */ |
743 | /* call feeder, if applicable */ |
597 | grp_try_feed (grp); |
744 | grp_try_feed (grp); |
… | |
… | |
601 | return eio_finish (grp); |
748 | return eio_finish (grp); |
602 | else |
749 | else |
603 | return 0; |
750 | return 0; |
604 | } |
751 | } |
605 | |
752 | |
|
|
753 | static void |
606 | void eio_destroy (eio_req *req) |
754 | eio_destroy (eio_req *req) |
607 | { |
755 | { |
608 | if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
756 | if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); |
609 | if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
757 | if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); |
610 | |
758 | |
611 | EIO_DESTROY (req); |
759 | EIO_DESTROY (req); |
612 | } |
760 | } |
613 | |
761 | |
|
|
762 | static int |
614 | static int eio_finish (eio_req *req) |
763 | eio_finish (eio_req *req) |
615 | { |
764 | { |
616 | int res = EIO_FINISH (req); |
765 | int res = EIO_FINISH (req); |
617 | |
766 | |
618 | if (req->grp) |
767 | if (req->grp) |
619 | { |
768 | { |
… | |
… | |
627 | if (grp->grp_first == req) |
776 | if (grp->grp_first == req) |
628 | grp->grp_first = req->grp_next; |
777 | grp->grp_first = req->grp_next; |
629 | |
778 | |
630 | res2 = grp_dec (grp); |
779 | res2 = grp_dec (grp); |
631 | |
780 | |
632 | if (!res && res2) |
781 | if (!res) |
633 | res = res2; |
782 | res = res2; |
634 | } |
783 | } |
635 | |
784 | |
636 | eio_destroy (req); |
785 | eio_destroy (req); |
637 | |
786 | |
638 | return res; |
787 | return res; |
639 | } |
788 | } |
640 | |
789 | |
|
|
790 | void |
641 | void eio_grp_cancel (eio_req *grp) |
791 | eio_grp_cancel (eio_req *grp) |
642 | { |
792 | { |
643 | for (grp = grp->grp_first; grp; grp = grp->grp_next) |
793 | for (grp = grp->grp_first; grp; grp = grp->grp_next) |
644 | eio_cancel (grp); |
794 | eio_cancel (grp); |
645 | } |
795 | } |
646 | |
796 | |
|
|
797 | void |
647 | void eio_cancel (eio_req *req) |
798 | eio_cancel (eio_req *req) |
648 | { |
799 | { |
649 | etp_cancel (req); |
800 | etp_cancel (req); |
650 | } |
801 | } |
651 | |
802 | |
|
|
803 | void |
652 | void eio_submit (eio_req *req) |
804 | eio_submit (eio_req *req) |
653 | { |
805 | { |
654 | etp_submit (req); |
806 | etp_submit (req); |
655 | } |
807 | } |
656 | |
808 | |
657 | unsigned int eio_nreqs (void) |
809 | unsigned int |
|
|
810 | eio_nreqs (void) |
658 | { |
811 | { |
659 | return etp_nreqs (); |
812 | return etp_nreqs (); |
660 | } |
813 | } |
661 | |
814 | |
662 | unsigned int eio_nready (void) |
815 | unsigned int |
|
|
816 | eio_nready (void) |
663 | { |
817 | { |
664 | return etp_nready (); |
818 | return etp_nready (); |
665 | } |
819 | } |
666 | |
820 | |
667 | unsigned int eio_npending (void) |
821 | unsigned int |
|
|
822 | eio_npending (void) |
668 | { |
823 | { |
669 | return etp_npending (); |
824 | return etp_npending (); |
670 | } |
825 | } |
671 | |
826 | |
672 | unsigned int eio_nthreads (void) |
827 | unsigned int ecb_cold |
|
|
828 | eio_nthreads (void) |
673 | { |
829 | { |
674 | return etp_nthreads (); |
830 | return etp_nthreads (); |
675 | } |
831 | } |
676 | |
832 | |
|
|
833 | void ecb_cold |
677 | void eio_set_max_poll_time (double nseconds) |
834 | eio_set_max_poll_time (double nseconds) |
678 | { |
835 | { |
679 | etp_set_max_poll_time (nseconds); |
836 | etp_set_max_poll_time (nseconds); |
680 | } |
837 | } |
681 | |
838 | |
|
|
839 | void ecb_cold |
682 | void eio_set_max_poll_reqs (unsigned int maxreqs) |
840 | eio_set_max_poll_reqs (unsigned int maxreqs) |
683 | { |
841 | { |
684 | etp_set_max_poll_reqs (maxreqs); |
842 | etp_set_max_poll_reqs (maxreqs); |
685 | } |
843 | } |
686 | |
844 | |
|
|
845 | void ecb_cold |
687 | void eio_set_max_idle (unsigned int nthreads) |
846 | eio_set_max_idle (unsigned int nthreads) |
688 | { |
847 | { |
689 | etp_set_max_idle (nthreads); |
848 | etp_set_max_idle (nthreads); |
690 | } |
849 | } |
691 | |
850 | |
|
|
851 | void ecb_cold |
|
|
852 | eio_set_idle_timeout (unsigned int seconds) |
|
|
853 | { |
|
|
854 | etp_set_idle_timeout (seconds); |
|
|
855 | } |
|
|
856 | |
|
|
857 | void ecb_cold |
692 | void eio_set_min_parallel (unsigned int nthreads) |
858 | eio_set_min_parallel (unsigned int nthreads) |
693 | { |
859 | { |
694 | etp_set_min_parallel (nthreads); |
860 | etp_set_min_parallel (nthreads); |
695 | } |
861 | } |
696 | |
862 | |
|
|
863 | void ecb_cold |
697 | void eio_set_max_parallel (unsigned int nthreads) |
864 | eio_set_max_parallel (unsigned int nthreads) |
698 | { |
865 | { |
699 | etp_set_max_parallel (nthreads); |
866 | etp_set_max_parallel (nthreads); |
700 | } |
867 | } |
701 | |
868 | |
702 | int eio_poll (void) |
869 | int eio_poll (void) |
… | |
… | |
706 | |
873 | |
707 | /*****************************************************************************/ |
874 | /*****************************************************************************/ |
708 | /* work around various missing functions */ |
875 | /* work around various missing functions */ |
709 | |
876 | |
710 | #if !HAVE_PREADWRITE |
877 | #if !HAVE_PREADWRITE |
|
|
878 | # undef pread |
|
|
879 | # undef pwrite |
711 | # define pread eio__pread |
880 | # define pread eio__pread |
712 | # define pwrite eio__pwrite |
881 | # define pwrite eio__pwrite |
713 | |
882 | |
714 | static ssize_t |
883 | static ssize_t |
715 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
884 | eio__pread (int fd, void *buf, size_t count, off_t offset) |
… | |
… | |
735 | |
904 | |
736 | X_LOCK (preadwritelock); |
905 | X_LOCK (preadwritelock); |
737 | ooffset = lseek (fd, 0, SEEK_CUR); |
906 | ooffset = lseek (fd, 0, SEEK_CUR); |
738 | lseek (fd, offset, SEEK_SET); |
907 | lseek (fd, offset, SEEK_SET); |
739 | res = write (fd, buf, count); |
908 | res = write (fd, buf, count); |
740 | lseek (fd, offset, SEEK_SET); |
909 | lseek (fd, ooffset, SEEK_SET); |
741 | X_UNLOCK (preadwritelock); |
910 | X_UNLOCK (preadwritelock); |
742 | |
911 | |
743 | return res; |
912 | return res; |
744 | } |
913 | } |
745 | #endif |
914 | #endif |
746 | |
915 | |
747 | #ifndef HAVE_FUTIMES |
916 | #ifndef HAVE_UTIMES |
748 | |
917 | |
|
|
918 | # undef utimes |
749 | # define utimes(path,times) eio__utimes (path, times) |
919 | # define utimes(path,times) eio__utimes (path, times) |
750 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
751 | |
920 | |
752 | static int |
921 | static int |
753 | eio__utimes (const char *filename, const struct timeval times[2]) |
922 | eio__utimes (const char *filename, const struct timeval times[2]) |
754 | { |
923 | { |
755 | if (times) |
924 | if (times) |
… | |
… | |
763 | } |
932 | } |
764 | else |
933 | else |
765 | return utime (filename, 0); |
934 | return utime (filename, 0); |
766 | } |
935 | } |
767 | |
936 | |
|
|
937 | #endif |
|
|
938 | |
|
|
939 | #ifndef HAVE_FUTIMES |
|
|
940 | |
|
|
941 | # undef futimes |
|
|
942 | # define futimes(fd,times) eio__futimes (fd, times) |
|
|
943 | |
|
|
944 | static int |
768 | static int eio__futimes (int fd, const struct timeval tv[2]) |
945 | eio__futimes (int fd, const struct timeval tv[2]) |
769 | { |
946 | { |
770 | errno = ENOSYS; |
947 | errno = ENOSYS; |
771 | return -1; |
948 | return -1; |
772 | } |
949 | } |
773 | |
950 | |
774 | #endif |
951 | #endif |
775 | |
952 | |
776 | #if !HAVE_FDATASYNC |
953 | #if !HAVE_FDATASYNC |
|
|
954 | # undef fdatasync |
777 | # define fdatasync fsync |
955 | # define fdatasync(fd) fsync (fd) |
778 | #endif |
956 | #endif |
|
|
957 | |
|
|
958 | /* sync_file_range always needs emulation */ |
|
|
959 | static int |
|
|
960 | eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) |
|
|
961 | { |
|
|
962 | #if HAVE_SYNC_FILE_RANGE |
|
|
963 | int res; |
|
|
964 | |
|
|
965 | if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE |
|
|
966 | || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE |
|
|
967 | || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER) |
|
|
968 | { |
|
|
969 | flags = 0 |
|
|
970 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0) |
|
|
971 | | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0) |
|
|
972 | | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0); |
|
|
973 | } |
|
|
974 | |
|
|
975 | res = sync_file_range (fd, offset, nbytes, flags); |
|
|
976 | |
|
|
977 | if (!res || errno != ENOSYS) |
|
|
978 | return res; |
|
|
979 | #endif |
|
|
980 | |
|
|
981 | /* even though we could play tricks with the flags, it's better to always |
|
|
982 | * call fdatasync, as that matches the expectation of its users best */ |
|
|
983 | return fdatasync (fd); |
|
|
984 | } |
|
|
985 | |
|
|
986 | static int |
|
|
987 | eio__fallocate (int fd, int mode, off_t offset, size_t len) |
|
|
988 | { |
|
|
989 | #if HAVE_FALLOCATE |
|
|
990 | return fallocate (fd, mode, offset, len); |
|
|
991 | #else |
|
|
992 | errno = ENOSYS; |
|
|
993 | return -1; |
|
|
994 | #endif |
|
|
995 | } |
779 | |
996 | |
780 | #if !HAVE_READAHEAD |
997 | #if !HAVE_READAHEAD |
|
|
998 | # undef readahead |
781 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
999 | # define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) |
782 | |
1000 | |
783 | static ssize_t |
1001 | static ssize_t |
784 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
1002 | eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) |
785 | { |
1003 | { |
… | |
… | |
803 | |
1021 | |
804 | /* sendfile always needs emulation */ |
1022 | /* sendfile always needs emulation */ |
805 | static ssize_t |
1023 | static ssize_t |
806 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
1024 | eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) |
807 | { |
1025 | { |
|
|
1026 | ssize_t written = 0; |
808 | ssize_t res; |
1027 | ssize_t res; |
809 | |
1028 | |
810 | if (!count) |
1029 | if (!count) |
811 | return 0; |
1030 | return 0; |
812 | |
1031 | |
|
|
1032 | for (;;) |
|
|
1033 | { |
|
|
1034 | #ifdef __APPLE__ |
|
|
1035 | # undef HAVE_SENDFILE /* broken, as everything on os x */ |
|
|
1036 | #endif |
813 | #if HAVE_SENDFILE |
1037 | #if HAVE_SENDFILE |
814 | # if __linux |
1038 | # if __linux |
|
|
1039 | off_t soffset = offset; |
815 | res = sendfile (ofd, ifd, &offset, count); |
1040 | res = sendfile (ofd, ifd, &soffset, count); |
816 | |
1041 | |
817 | # elif __freebsd |
1042 | # elif __FreeBSD__ |
818 | /* |
1043 | /* |
819 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
1044 | * Of course, the freebsd sendfile is a dire hack with no thoughts |
820 | * wasted on making it similar to other I/O functions. |
1045 | * wasted on making it similar to other I/O functions. |
821 | */ |
1046 | */ |
822 | { |
|
|
823 | off_t sbytes; |
1047 | off_t sbytes; |
824 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
1048 | res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); |
825 | |
1049 | |
826 | if (res < 0 && sbytes) |
1050 | #if 0 /* according to the manpage, this is correct, but broken behaviour */ |
827 | /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ |
1051 | /* freebsd' sendfile will return 0 on success */ |
|
|
1052 | /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ |
|
|
1053 | /* not on e.g. EIO or EPIPE - sounds broken */ |
|
|
1054 | if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) |
828 | res = sbytes; |
1055 | res = sbytes; |
829 | } |
1056 | #endif |
|
|
1057 | |
|
|
1058 | /* according to source inspection, this is correct, and useful behaviour */ |
|
|
1059 | if (sbytes) |
|
|
1060 | res = sbytes; |
|
|
1061 | |
|
|
1062 | # elif defined (__APPLE__) |
|
|
1063 | off_t sbytes = count; |
|
|
1064 | res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); |
|
|
1065 | |
|
|
1066 | /* according to the manpage, sbytes is always valid */ |
|
|
1067 | if (sbytes) |
|
|
1068 | res = sbytes; |
830 | |
1069 | |
831 | # elif __hpux |
1070 | # elif __hpux |
832 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
1071 | res = sendfile (ofd, ifd, offset, count, 0, 0); |
833 | |
1072 | |
834 | # elif __solaris |
1073 | # elif __solaris |
835 | { |
|
|
836 | struct sendfilevec vec; |
1074 | struct sendfilevec vec; |
837 | size_t sbytes; |
1075 | size_t sbytes; |
838 | |
1076 | |
839 | vec.sfv_fd = ifd; |
1077 | vec.sfv_fd = ifd; |
840 | vec.sfv_flag = 0; |
1078 | vec.sfv_flag = 0; |
841 | vec.sfv_off = offset; |
1079 | vec.sfv_off = offset; |
842 | vec.sfv_len = count; |
1080 | vec.sfv_len = count; |
843 | |
1081 | |
844 | res = sendfilev (ofd, &vec, 1, &sbytes); |
1082 | res = sendfilev (ofd, &vec, 1, &sbytes); |
845 | |
1083 | |
846 | if (res < 0 && sbytes) |
1084 | if (res < 0 && sbytes) |
847 | res = sbytes; |
1085 | res = sbytes; |
848 | } |
|
|
849 | |
1086 | |
850 | # endif |
1087 | # endif |
|
|
1088 | |
|
|
1089 | #elif defined (_WIN32) && 0 |
|
|
1090 | /* does not work, just for documentation of what would need to be done */ |
|
|
1091 | /* actually, cannot be done like this, as TransmitFile changes the file offset, */ |
|
|
1092 | /* libeio guarantees that the file offset does not change, and windows */ |
|
|
1093 | /* has no way to get an independent handle to the same file description */ |
|
|
1094 | HANDLE h = TO_SOCKET (ifd); |
|
|
1095 | SetFilePointer (h, offset, 0, FILE_BEGIN); |
|
|
1096 | res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); |
|
|
1097 | |
851 | #else |
1098 | #else |
852 | res = -1; |
1099 | res = -1; |
853 | errno = ENOSYS; |
1100 | errno = ENOSYS; |
854 | #endif |
1101 | #endif |
855 | |
1102 | |
|
|
1103 | /* we assume sendfile can copy at least 128mb in one go */ |
|
|
1104 | if (res <= 128 * 1024 * 1024) |
|
|
1105 | { |
|
|
1106 | if (res > 0) |
|
|
1107 | written += res; |
|
|
1108 | |
|
|
1109 | if (written) |
|
|
1110 | return written; |
|
|
1111 | |
|
|
1112 | break; |
|
|
1113 | } |
|
|
1114 | else |
|
|
1115 | { |
|
|
1116 | /* if we requested more, then probably the kernel was lazy */ |
|
|
1117 | written += res; |
|
|
1118 | offset += res; |
|
|
1119 | count -= res; |
|
|
1120 | |
|
|
1121 | if (!count) |
|
|
1122 | return written; |
|
|
1123 | } |
|
|
1124 | } |
|
|
1125 | |
856 | if (res < 0 |
1126 | if (res < 0 |
857 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
1127 | && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK |
|
|
1128 | /* BSDs */ |
|
|
1129 | #ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ |
|
|
1130 | || errno == ENOTSUP |
|
|
1131 | #endif |
|
|
1132 | || errno == EOPNOTSUPP /* BSDs */ |
858 | #if __solaris |
1133 | #if __solaris |
859 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
1134 | || errno == EAFNOSUPPORT || errno == EPROTOTYPE |
860 | #endif |
1135 | #endif |
861 | ) |
1136 | ) |
862 | ) |
1137 | ) |
… | |
… | |
893 | } |
1168 | } |
894 | |
1169 | |
895 | return res; |
1170 | return res; |
896 | } |
1171 | } |
897 | |
1172 | |
|
|
1173 | #ifdef PAGESIZE |
|
|
1174 | # define eio_pagesize() PAGESIZE |
|
|
1175 | #else |
|
|
1176 | static intptr_t |
|
|
1177 | eio_pagesize (void) |
|
|
1178 | { |
|
|
1179 | static intptr_t page; |
|
|
1180 | |
|
|
1181 | if (!page) |
|
|
1182 | page = sysconf (_SC_PAGESIZE); |
|
|
1183 | |
|
|
1184 | return page; |
|
|
1185 | } |
|
|
1186 | #endif |
|
|
1187 | |
|
|
1188 | static void |
|
|
1189 | eio_page_align (void **addr, size_t *length) |
|
|
1190 | { |
|
|
1191 | intptr_t mask = eio_pagesize () - 1; |
|
|
1192 | |
|
|
1193 | /* round down addr */ |
|
|
1194 | intptr_t adj = mask & (intptr_t)*addr; |
|
|
1195 | |
|
|
1196 | *addr = (void *)((intptr_t)*addr - adj); |
|
|
1197 | *length += adj; |
|
|
1198 | |
|
|
1199 | /* round up length */ |
|
|
1200 | *length = (*length + mask) & ~mask; |
|
|
1201 | } |
|
|
1202 | |
|
|
1203 | #if !_POSIX_MEMLOCK |
|
|
1204 | # define eio__mlockall(a) eio_nosyscall() |
|
|
1205 | #else |
|
|
1206 | |
|
|
1207 | static int |
|
|
1208 | eio__mlockall (int flags) |
|
|
1209 | { |
|
|
1210 | #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 |
|
|
1211 | extern int mallopt (int, int); |
|
|
1212 | mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */ |
|
|
1213 | #endif |
|
|
1214 | |
|
|
1215 | if (EIO_MCL_CURRENT != MCL_CURRENT |
|
|
1216 | || EIO_MCL_FUTURE != MCL_FUTURE) |
|
|
1217 | { |
|
|
1218 | flags = 0 |
|
|
1219 | | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0) |
|
|
1220 | | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0); |
|
|
1221 | } |
|
|
1222 | |
|
|
1223 | return mlockall (flags); |
|
|
1224 | } |
|
|
1225 | #endif |
|
|
1226 | |
|
|
1227 | #if !_POSIX_MEMLOCK_RANGE |
|
|
1228 | # define eio__mlock(a,b) EIO_ENOSYS () |
|
|
1229 | #else |
|
|
1230 | |
|
|
1231 | static int |
|
|
1232 | eio__mlock (void *addr, size_t length) |
|
|
1233 | { |
|
|
1234 | eio_page_align (&addr, &length); |
|
|
1235 | |
|
|
1236 | return mlock (addr, length); |
|
|
1237 | } |
|
|
1238 | |
|
|
1239 | #endif |
|
|
1240 | |
|
|
1241 | #if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) |
|
|
1242 | # define eio__msync(a,b,c) EIO_ENOSYS () |
|
|
1243 | #else |
|
|
1244 | |
|
|
1245 | static int |
|
|
1246 | eio__msync (void *mem, size_t len, int flags) |
|
|
1247 | { |
|
|
1248 | eio_page_align (&mem, &len); |
|
|
1249 | |
|
|
1250 | if (EIO_MS_ASYNC != MS_SYNC |
|
|
1251 | || EIO_MS_INVALIDATE != MS_INVALIDATE |
|
|
1252 | || EIO_MS_SYNC != MS_SYNC) |
|
|
1253 | { |
|
|
1254 | flags = 0 |
|
|
1255 | | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0) |
|
|
1256 | | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0) |
|
|
1257 | | (flags & EIO_MS_SYNC ? MS_SYNC : 0); |
|
|
1258 | } |
|
|
1259 | |
|
|
1260 | return msync (mem, len, flags); |
|
|
1261 | } |
|
|
1262 | |
|
|
1263 | #endif |
|
|
1264 | |
|
|
1265 | static int |
|
|
1266 | eio__mtouch (eio_req *req) |
|
|
1267 | { |
|
|
1268 | void *mem = req->ptr2; |
|
|
1269 | size_t len = req->size; |
|
|
1270 | int flags = req->int1; |
|
|
1271 | |
|
|
1272 | eio_page_align (&mem, &len); |
|
|
1273 | |
|
|
1274 | { |
|
|
1275 | intptr_t addr = (intptr_t)mem; |
|
|
1276 | intptr_t end = addr + len; |
|
|
1277 | intptr_t page = eio_pagesize (); |
|
|
1278 | |
|
|
1279 | if (addr < end) |
|
|
1280 | if (flags & EIO_MT_MODIFY) /* modify */ |
|
|
1281 | do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
|
|
1282 | else |
|
|
1283 | do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req)); |
|
|
1284 | } |
|
|
1285 | |
|
|
1286 | return 0; |
|
|
1287 | } |
|
|
1288 | |
|
|
1289 | /*****************************************************************************/ |
|
|
1290 | /* requests implemented outside eio_execute, because they are so large */ |
|
|
1291 | |
|
|
1292 | static void |
|
|
1293 | eio__realpath (eio_req *req, etp_worker *self) |
|
|
1294 | { |
|
|
1295 | char *rel = req->ptr1; |
|
|
1296 | char *res; |
|
|
1297 | char *tmp1, *tmp2; |
|
|
1298 | #if SYMLOOP_MAX > 32 |
|
|
1299 | int symlinks = SYMLOOP_MAX; |
|
|
1300 | #else |
|
|
1301 | int symlinks = 32; |
|
|
1302 | #endif |
|
|
1303 | |
|
|
1304 | req->result = -1; |
|
|
1305 | |
|
|
1306 | errno = EINVAL; |
|
|
1307 | if (!rel) |
|
|
1308 | return; |
|
|
1309 | |
|
|
1310 | errno = ENOENT; |
|
|
1311 | if (!*rel) |
|
|
1312 | return; |
|
|
1313 | |
|
|
1314 | if (!req->ptr2) |
|
|
1315 | { |
|
|
1316 | X_LOCK (wrklock); |
|
|
1317 | req->flags |= EIO_FLAG_PTR2_FREE; |
|
|
1318 | X_UNLOCK (wrklock); |
|
|
1319 | req->ptr2 = malloc (PATH_MAX * 3); |
|
|
1320 | |
|
|
1321 | errno = ENOMEM; |
|
|
1322 | if (!req->ptr2) |
|
|
1323 | return; |
|
|
1324 | } |
|
|
1325 | |
|
|
1326 | res = req->ptr2; |
|
|
1327 | tmp1 = res + PATH_MAX; |
|
|
1328 | tmp2 = tmp1 + PATH_MAX; |
|
|
1329 | |
|
|
1330 | #if 0 /* disabled, the musl way to do things is just too racy */ |
|
|
1331 | #if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) |
|
|
1332 | /* on linux we may be able to ask the kernel */ |
|
|
1333 | { |
|
|
1334 | int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME); |
|
|
1335 | |
|
|
1336 | if (fd >= 0) |
|
|
1337 | { |
|
|
1338 | sprintf (tmp1, "/proc/self/fd/%d", fd); |
|
|
1339 | req->result = readlink (tmp1, res, PATH_MAX); |
|
|
1340 | close (fd); |
|
|
1341 | |
|
|
1342 | /* here we should probably stat the open file and the disk file, to make sure they still match */ |
|
|
1343 | |
|
|
1344 | if (req->result > 0) |
|
|
1345 | goto done; |
|
|
1346 | } |
|
|
1347 | else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) |
|
|
1348 | return; |
|
|
1349 | } |
|
|
1350 | #endif |
|
|
1351 | #endif |
|
|
1352 | |
|
|
1353 | if (*rel != '/') |
|
|
1354 | { |
|
|
1355 | if (!getcwd (res, PATH_MAX)) |
|
|
1356 | return; |
|
|
1357 | |
|
|
1358 | if (res [1]) /* only use if not / */ |
|
|
1359 | res += strlen (res); |
|
|
1360 | } |
|
|
1361 | |
|
|
1362 | while (*rel) |
|
|
1363 | { |
|
|
1364 | ssize_t len, linklen; |
|
|
1365 | char *beg = rel; |
|
|
1366 | |
|
|
1367 | while (*rel && *rel != '/') |
|
|
1368 | ++rel; |
|
|
1369 | |
|
|
1370 | len = rel - beg; |
|
|
1371 | |
|
|
1372 | if (!len) /* skip slashes */ |
|
|
1373 | { |
|
|
1374 | ++rel; |
|
|
1375 | continue; |
|
|
1376 | } |
|
|
1377 | |
|
|
1378 | if (beg [0] == '.') |
|
|
1379 | { |
|
|
1380 | if (len == 1) |
|
|
1381 | continue; /* . - nop */ |
|
|
1382 | |
|
|
1383 | if (beg [1] == '.' && len == 2) |
|
|
1384 | { |
|
|
1385 | /* .. - back up one component, if possible */ |
|
|
1386 | |
|
|
1387 | while (res != req->ptr2) |
|
|
1388 | if (*--res == '/') |
|
|
1389 | break; |
|
|
1390 | |
|
|
1391 | continue; |
|
|
1392 | } |
|
|
1393 | } |
|
|
1394 | |
|
|
1395 | errno = ENAMETOOLONG; |
|
|
1396 | if (res + 1 + len + 1 >= tmp1) |
|
|
1397 | return; |
|
|
1398 | |
|
|
1399 | /* copy one component */ |
|
|
1400 | *res = '/'; |
|
|
1401 | memcpy (res + 1, beg, len); |
|
|
1402 | |
|
|
1403 | /* zero-terminate, for readlink */ |
|
|
1404 | res [len + 1] = 0; |
|
|
1405 | |
|
|
1406 | /* now check if it's a symlink */ |
|
|
1407 | linklen = readlink (req->ptr2, tmp1, PATH_MAX); |
|
|
1408 | |
|
|
1409 | if (linklen < 0) |
|
|
1410 | { |
|
|
1411 | if (errno != EINVAL) |
|
|
1412 | return; |
|
|
1413 | |
|
|
1414 | /* it's a normal directory. hopefully */ |
|
|
1415 | res += len + 1; |
|
|
1416 | } |
|
|
1417 | else |
|
|
1418 | { |
|
|
1419 | /* yay, it was a symlink - build new path in tmp2 */ |
|
|
1420 | int rellen = strlen (rel); |
|
|
1421 | |
|
|
1422 | errno = ENAMETOOLONG; |
|
|
1423 | if (linklen + 1 + rellen >= PATH_MAX) |
|
|
1424 | return; |
|
|
1425 | |
|
|
1426 | errno = ELOOP; |
|
|
1427 | if (!--symlinks) |
|
|
1428 | return; |
|
|
1429 | |
|
|
1430 | if (*tmp1 == '/') |
|
|
1431 | res = req->ptr2; /* symlink resolves to an absolute path */ |
|
|
1432 | |
|
|
1433 | /* we need to be careful, as rel might point into tmp2 already */ |
|
|
1434 | memmove (tmp2 + linklen + 1, rel, rellen + 1); |
|
|
1435 | tmp2 [linklen] = '/'; |
|
|
1436 | memcpy (tmp2, tmp1, linklen); |
|
|
1437 | |
|
|
1438 | rel = tmp2; |
|
|
1439 | } |
|
|
1440 | } |
|
|
1441 | |
|
|
1442 | /* special case for the lone root path */ |
|
|
1443 | if (res == req->ptr2) |
|
|
1444 | *res++ = '/'; |
|
|
1445 | |
|
|
1446 | req->result = res - (char *)req->ptr2; |
|
|
1447 | |
|
|
1448 | done: |
|
|
1449 | req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */ |
|
|
1450 | } |
|
|
1451 | |
|
|
1452 | static signed char |
|
|
1453 | eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) |
|
|
1454 | { |
|
|
1455 | return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ |
|
|
1456 | : a->inode < b->inode ? -1 |
|
|
1457 | : a->inode > b->inode ? 1 |
|
|
1458 | : 0; |
|
|
1459 | } |
|
|
1460 | |
|
|
1461 | #define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 |
|
|
1462 | |
|
|
1463 | #define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ |
|
|
1464 | #define EIO_SORT_FAST 60 /* when to only use insertion sort */ |
|
|
1465 | |
|
|
1466 | static void |
|
|
1467 | eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1468 | { |
|
|
1469 | unsigned char bits [9 + sizeof (ino_t) * 8]; |
|
|
1470 | unsigned char *bit = bits; |
|
|
1471 | |
|
|
1472 | assert (CHAR_BIT == 8); |
|
|
1473 | assert (sizeof (eio_dirent) * 8 < 256); |
|
|
1474 | assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ |
|
|
1475 | assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */ |
|
|
1476 | |
|
|
1477 | if (size <= EIO_SORT_FAST) |
|
|
1478 | return; |
|
|
1479 | |
|
|
1480 | /* first prepare an array of bits to test in our radix sort */ |
|
|
1481 | /* try to take endianness into account, as well as differences in ino_t sizes */ |
|
|
1482 | /* inode_bits must contain all inodes ORed together */ |
|
|
1483 | /* which is used to skip bits that are 0 everywhere, which is very common */ |
|
|
1484 | { |
|
|
1485 | ino_t endianness; |
|
|
1486 | int i, j; |
|
|
1487 | |
|
|
1488 | /* we store the byte offset of byte n into byte n of "endianness" */ |
|
|
1489 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1490 | ((unsigned char *)&endianness)[i] = i; |
|
|
1491 | |
|
|
1492 | *bit++ = 0; |
|
|
1493 | |
|
|
1494 | for (i = 0; i < sizeof (ino_t); ++i) |
|
|
1495 | { |
|
|
1496 | /* shifting off the byte offsets out of "endianness" */ |
|
|
1497 | int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; |
|
|
1498 | endianness >>= 8; |
|
|
1499 | |
|
|
1500 | for (j = 0; j < 8; ++j) |
|
|
1501 | if (inode_bits & (((ino_t)1) << (i * 8 + j))) |
|
|
1502 | *bit++ = offs + j; |
|
|
1503 | } |
|
|
1504 | |
|
|
1505 | for (j = 0; j < 8; ++j) |
|
|
1506 | if (score_bits & (1 << j)) |
|
|
1507 | *bit++ = offsetof (eio_dirent, score) * 8 + j; |
|
|
1508 | } |
|
|
1509 | |
|
|
1510 | /* now actually do the sorting (a variant of MSD radix sort) */ |
|
|
1511 | { |
|
|
1512 | eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; |
|
|
1513 | eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; |
|
|
1514 | unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; |
|
|
1515 | int stk_idx = 0; |
|
|
1516 | |
|
|
1517 | base_stk [stk_idx] = dents; |
|
|
1518 | end_stk [stk_idx] = dents + size; |
|
|
1519 | bit_stk [stk_idx] = bit - 1; |
|
|
1520 | |
|
|
1521 | do |
|
|
1522 | { |
|
|
1523 | base = base_stk [stk_idx]; |
|
|
1524 | end = end_stk [stk_idx]; |
|
|
1525 | bit = bit_stk [stk_idx]; |
|
|
1526 | |
|
|
1527 | for (;;) |
|
|
1528 | { |
|
|
1529 | unsigned char O = *bit >> 3; |
|
|
1530 | unsigned char M = 1 << (*bit & 7); |
|
|
1531 | |
|
|
1532 | eio_dirent *a = base; |
|
|
1533 | eio_dirent *b = end; |
|
|
1534 | |
|
|
1535 | if (b - a < EIO_SORT_CUTOFF) |
|
|
1536 | break; |
|
|
1537 | |
|
|
1538 | /* now bit-partition the array on the bit */ |
|
|
1539 | /* this ugly asymmetric loop seems to perform much better than typical */ |
|
|
1540 | /* partition algos found in the literature */ |
|
|
1541 | do |
|
|
1542 | if (!(((unsigned char *)a)[O] & M)) |
|
|
1543 | ++a; |
|
|
1544 | else if (!(((unsigned char *)--b)[O] & M)) |
|
|
1545 | { |
|
|
1546 | eio_dirent tmp = *a; *a = *b; *b = tmp; |
|
|
1547 | ++a; |
|
|
1548 | } |
|
|
1549 | while (b > a); |
|
|
1550 | |
|
|
1551 | /* next bit, or stop, if no bits left in this path */ |
|
|
1552 | if (!*--bit) |
|
|
1553 | break; |
|
|
1554 | |
|
|
1555 | base_stk [stk_idx] = a; |
|
|
1556 | end_stk [stk_idx] = end; |
|
|
1557 | bit_stk [stk_idx] = bit; |
|
|
1558 | ++stk_idx; |
|
|
1559 | |
|
|
1560 | end = a; |
|
|
1561 | } |
|
|
1562 | } |
|
|
1563 | while (stk_idx--); |
|
|
1564 | } |
|
|
1565 | } |
|
|
1566 | |
|
|
1567 | static void |
|
|
1568 | eio_dent_insertion_sort (eio_dirent *dents, int size) |
|
|
1569 | { |
|
|
1570 | /* first move the smallest element to the front, to act as a sentinel */ |
|
|
1571 | { |
|
|
1572 | int i; |
|
|
1573 | eio_dirent *min = dents; |
|
|
1574 | |
|
|
1575 | /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */ |
|
|
1576 | for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; ) |
|
|
1577 | if (EIO_DENT_CMP (dents [i], <, *min)) |
|
|
1578 | min = &dents [i]; |
|
|
1579 | |
|
|
1580 | /* swap elements 0 and j (minimum) */ |
|
|
1581 | { |
|
|
1582 | eio_dirent tmp = *dents; *dents = *min; *min = tmp; |
|
|
1583 | } |
|
|
1584 | } |
|
|
1585 | |
|
|
1586 | /* then do standard insertion sort, assuming that all elements are >= dents [0] */ |
|
|
1587 | { |
|
|
1588 | eio_dirent *i, *j; |
|
|
1589 | |
|
|
1590 | for (i = dents + 1; i < dents + size; ++i) |
|
|
1591 | { |
|
|
1592 | eio_dirent value = *i; |
|
|
1593 | |
|
|
1594 | for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j) |
|
|
1595 | j [1] = j [0]; |
|
|
1596 | |
|
|
1597 | j [1] = value; |
|
|
1598 | } |
|
|
1599 | } |
|
|
1600 | } |
|
|
1601 | |
|
|
1602 | static void |
|
|
1603 | eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) |
|
|
1604 | { |
|
|
1605 | if (size <= 1) |
|
|
1606 | return; /* our insertion sort relies on size > 0 */ |
|
|
1607 | |
|
|
1608 | /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ |
|
|
1609 | /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */ |
|
|
1610 | eio_dent_radix_sort (dents, size, score_bits, inode_bits); |
|
|
1611 | |
|
|
1612 | /* use an insertion sort at the end, or for small arrays, */ |
|
|
1613 | /* as insertion sort is more efficient for small partitions */ |
|
|
1614 | eio_dent_insertion_sort (dents, size); |
|
|
1615 | } |
|
|
1616 | |
898 | /* read a full directory */ |
1617 | /* read a full directory */ |
899 | static void |
1618 | static void |
900 | eio__scandir (eio_req *req, etp_worker *self) |
1619 | eio__scandir (eio_req *req, etp_worker *self) |
901 | { |
1620 | { |
902 | DIR *dirp; |
1621 | DIR *dirp; |
903 | EIO_STRUCT_DIRENT *entp; |
1622 | EIO_STRUCT_DIRENT *entp; |
904 | char *name, *names; |
1623 | char *name, *names; |
905 | int memlen = 4096; |
1624 | int namesalloc = 4096; |
906 | int memofs = 0; |
1625 | int namesoffs = 0; |
|
|
1626 | int flags = req->int1; |
|
|
1627 | eio_dirent *dents = 0; |
|
|
1628 | int dentalloc = 128; |
907 | int res = 0; |
1629 | int dentoffs = 0; |
|
|
1630 | ino_t inode_bits = 0; |
|
|
1631 | |
|
|
1632 | req->result = -1; |
|
|
1633 | |
|
|
1634 | if (!(flags & EIO_READDIR_DENTS)) |
|
|
1635 | flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); |
908 | |
1636 | |
909 | X_LOCK (wrklock); |
1637 | X_LOCK (wrklock); |
910 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
1638 | /* the corresponding closedir is in ETP_WORKER_CLEAR */ |
911 | self->dirp = dirp = opendir (req->ptr1); |
1639 | self->dirp = dirp = opendir (req->ptr1); |
|
|
1640 | |
912 | req->flags |= EIO_FLAG_PTR2_FREE; |
1641 | if (req->flags & EIO_FLAG_PTR1_FREE) |
|
|
1642 | free (req->ptr1); |
|
|
1643 | |
|
|
1644 | req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; |
|
|
1645 | req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; |
913 | req->ptr2 = names = malloc (memlen); |
1646 | req->ptr2 = names = malloc (namesalloc); |
914 | X_UNLOCK (wrklock); |
1647 | X_UNLOCK (wrklock); |
915 | |
1648 | |
916 | if (dirp && names) |
1649 | if (dirp && names && (!flags || dents)) |
917 | for (;;) |
1650 | for (;;) |
918 | { |
1651 | { |
919 | errno = 0; |
1652 | errno = 0; |
920 | entp = readdir (dirp); |
1653 | entp = readdir (dirp); |
921 | |
1654 | |
922 | if (!entp) |
1655 | if (!entp) |
|
|
1656 | { |
|
|
1657 | if (errno) |
|
|
1658 | break; |
|
|
1659 | |
|
|
1660 | /* sort etc. */ |
|
|
1661 | req->int1 = flags; |
|
|
1662 | req->result = dentoffs; |
|
|
1663 | |
|
|
1664 | if (flags & EIO_READDIR_STAT_ORDER) |
|
|
1665 | eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); |
|
|
1666 | else if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1667 | if (flags & EIO_READDIR_FOUND_UNKNOWN) |
|
|
1668 | eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ |
|
|
1669 | else |
|
|
1670 | { |
|
|
1671 | /* in this case, all is known, and we just put dirs first and sort them */ |
|
|
1672 | eio_dirent *oth = dents + dentoffs; |
|
|
1673 | eio_dirent *dir = dents; |
|
|
1674 | |
|
|
1675 | /* now partition dirs to the front, and non-dirs to the back */ |
|
|
1676 | /* by walking from both sides and swapping if necessary */ |
|
|
1677 | while (oth > dir) |
|
|
1678 | { |
|
|
1679 | if (dir->type == EIO_DT_DIR) |
|
|
1680 | ++dir; |
|
|
1681 | else if ((--oth)->type == EIO_DT_DIR) |
|
|
1682 | { |
|
|
1683 | eio_dirent tmp = *dir; *dir = *oth; *oth = tmp; |
|
|
1684 | |
|
|
1685 | ++dir; |
|
|
1686 | } |
|
|
1687 | } |
|
|
1688 | |
|
|
1689 | /* now sort the dirs only (dirs all have the same score) */ |
|
|
1690 | eio_dent_sort (dents, dir - dents, 0, inode_bits); |
|
|
1691 | } |
|
|
1692 | |
923 | break; |
1693 | break; |
|
|
1694 | } |
924 | |
1695 | |
|
|
1696 | /* now add the entry to our list(s) */ |
925 | name = entp->d_name; |
1697 | name = entp->d_name; |
926 | |
1698 | |
|
|
1699 | /* skip . and .. entries */ |
927 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
1700 | if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) |
928 | { |
1701 | { |
929 | int len = strlen (name) + 1; |
1702 | int len = D_NAMLEN (entp) + 1; |
930 | |
1703 | |
931 | res++; |
1704 | while (ecb_expect_false (namesoffs + len > namesalloc)) |
932 | |
|
|
933 | while (memofs + len > memlen) |
|
|
934 | { |
1705 | { |
935 | memlen *= 2; |
1706 | namesalloc *= 2; |
936 | X_LOCK (wrklock); |
1707 | X_LOCK (wrklock); |
937 | req->ptr2 = names = realloc (names, memlen); |
1708 | req->ptr2 = names = realloc (names, namesalloc); |
938 | X_UNLOCK (wrklock); |
1709 | X_UNLOCK (wrklock); |
939 | |
1710 | |
940 | if (!names) |
1711 | if (!names) |
941 | break; |
1712 | break; |
942 | } |
1713 | } |
943 | |
1714 | |
944 | memcpy (names + memofs, name, len); |
1715 | memcpy (names + namesoffs, name, len); |
|
|
1716 | |
|
|
1717 | if (dents) |
|
|
1718 | { |
|
|
1719 | struct eio_dirent *ent; |
|
|
1720 | |
|
|
1721 | if (ecb_expect_false (dentoffs == dentalloc)) |
|
|
1722 | { |
|
|
1723 | dentalloc *= 2; |
|
|
1724 | X_LOCK (wrklock); |
|
|
1725 | req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent)); |
|
|
1726 | X_UNLOCK (wrklock); |
|
|
1727 | |
|
|
1728 | if (!dents) |
|
|
1729 | break; |
|
|
1730 | } |
|
|
1731 | |
|
|
1732 | ent = dents + dentoffs; |
|
|
1733 | |
|
|
1734 | ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */ |
|
|
1735 | ent->namelen = len - 1; |
|
|
1736 | ent->inode = D_INO (entp); |
|
|
1737 | |
|
|
1738 | inode_bits |= ent->inode; |
|
|
1739 | |
|
|
1740 | switch (D_TYPE (entp)) |
|
|
1741 | { |
|
|
1742 | default: |
|
|
1743 | ent->type = EIO_DT_UNKNOWN; |
|
|
1744 | flags |= EIO_READDIR_FOUND_UNKNOWN; |
|
|
1745 | break; |
|
|
1746 | |
|
|
1747 | #ifdef DT_FIFO |
|
|
1748 | case DT_FIFO: ent->type = EIO_DT_FIFO; break; |
|
|
1749 | #endif |
|
|
1750 | #ifdef DT_CHR |
|
|
1751 | case DT_CHR: ent->type = EIO_DT_CHR; break; |
|
|
1752 | #endif |
|
|
1753 | #ifdef DT_MPC |
|
|
1754 | case DT_MPC: ent->type = EIO_DT_MPC; break; |
|
|
1755 | #endif |
|
|
1756 | #ifdef DT_DIR |
|
|
1757 | case DT_DIR: ent->type = EIO_DT_DIR; break; |
|
|
1758 | #endif |
|
|
1759 | #ifdef DT_NAM |
|
|
1760 | case DT_NAM: ent->type = EIO_DT_NAM; break; |
|
|
1761 | #endif |
|
|
1762 | #ifdef DT_BLK |
|
|
1763 | case DT_BLK: ent->type = EIO_DT_BLK; break; |
|
|
1764 | #endif |
|
|
1765 | #ifdef DT_MPB |
|
|
1766 | case DT_MPB: ent->type = EIO_DT_MPB; break; |
|
|
1767 | #endif |
|
|
1768 | #ifdef DT_REG |
|
|
1769 | case DT_REG: ent->type = EIO_DT_REG; break; |
|
|
1770 | #endif |
|
|
1771 | #ifdef DT_NWK |
|
|
1772 | case DT_NWK: ent->type = EIO_DT_NWK; break; |
|
|
1773 | #endif |
|
|
1774 | #ifdef DT_CMP |
|
|
1775 | case DT_CMP: ent->type = EIO_DT_CMP; break; |
|
|
1776 | #endif |
|
|
1777 | #ifdef DT_LNK |
|
|
1778 | case DT_LNK: ent->type = EIO_DT_LNK; break; |
|
|
1779 | #endif |
|
|
1780 | #ifdef DT_SOCK |
|
|
1781 | case DT_SOCK: ent->type = EIO_DT_SOCK; break; |
|
|
1782 | #endif |
|
|
1783 | #ifdef DT_DOOR |
|
|
1784 | case DT_DOOR: ent->type = EIO_DT_DOOR; break; |
|
|
1785 | #endif |
|
|
1786 | #ifdef DT_WHT |
|
|
1787 | case DT_WHT: ent->type = EIO_DT_WHT; break; |
|
|
1788 | #endif |
|
|
1789 | } |
|
|
1790 | |
|
|
1791 | ent->score = 7; |
|
|
1792 | |
|
|
1793 | if (flags & EIO_READDIR_DIRS_FIRST) |
|
|
1794 | { |
|
|
1795 | if (ent->type == EIO_DT_UNKNOWN) |
|
|
1796 | { |
|
|
1797 | if (*name == '.') /* leading dots are likely directories, and, in any case, rare */ |
|
|
1798 | ent->score = 1; |
|
|
1799 | else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */ |
|
|
1800 | ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */ |
|
|
1801 | } |
|
|
1802 | else if (ent->type == EIO_DT_DIR) |
|
|
1803 | ent->score = 0; |
|
|
1804 | } |
|
|
1805 | } |
|
|
1806 | |
945 | memofs += len; |
1807 | namesoffs += len; |
|
|
1808 | ++dentoffs; |
|
|
1809 | } |
|
|
1810 | |
|
|
1811 | if (EIO_CANCELLED (req)) |
|
|
1812 | { |
|
|
1813 | errno = ECANCELED; |
|
|
1814 | break; |
946 | } |
1815 | } |
947 | } |
1816 | } |
948 | |
|
|
949 | if (errno) |
|
|
950 | res = -1; |
|
|
951 | |
|
|
952 | req->result = res; |
|
|
953 | } |
1817 | } |
954 | |
1818 | |
955 | /*****************************************************************************/ |
1819 | /*****************************************************************************/ |
956 | |
1820 | |
957 | #define ALLOC(len) \ |
1821 | #define ALLOC(len) \ |
… | |
… | |
972 | X_THREAD_PROC (etp_proc) |
1836 | X_THREAD_PROC (etp_proc) |
973 | { |
1837 | { |
974 | ETP_REQ *req; |
1838 | ETP_REQ *req; |
975 | struct timespec ts; |
1839 | struct timespec ts; |
976 | etp_worker *self = (etp_worker *)thr_arg; |
1840 | etp_worker *self = (etp_worker *)thr_arg; |
|
|
1841 | int timeout; |
977 | |
1842 | |
978 | /* try to distribute timeouts somewhat randomly */ |
1843 | /* try to distribute timeouts somewhat evenly */ |
979 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
1844 | ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); |
980 | |
1845 | |
981 | for (;;) |
1846 | for (;;) |
982 | { |
1847 | { |
|
|
1848 | ts.tv_sec = 0; |
|
|
1849 | |
983 | X_LOCK (reqlock); |
1850 | X_LOCK (reqlock); |
984 | |
1851 | |
985 | for (;;) |
1852 | for (;;) |
986 | { |
1853 | { |
987 | self->req = req = reqq_shift (&req_queue); |
1854 | self->req = req = reqq_shift (&req_queue); |
988 | |
1855 | |
989 | if (req) |
1856 | if (req) |
990 | break; |
1857 | break; |
991 | |
1858 | |
|
|
1859 | if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */ |
|
|
1860 | { |
|
|
1861 | X_UNLOCK (reqlock); |
|
|
1862 | X_LOCK (wrklock); |
|
|
1863 | --started; |
|
|
1864 | X_UNLOCK (wrklock); |
|
|
1865 | goto quit; |
|
|
1866 | } |
|
|
1867 | |
992 | ++idle; |
1868 | ++idle; |
993 | |
1869 | |
994 | ts.tv_sec = time (0) + IDLE_TIMEOUT; |
1870 | if (idle <= max_idle) |
995 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
1871 | /* we are allowed to idle, so do so without any timeout */ |
|
|
1872 | X_COND_WAIT (reqwait, reqlock); |
|
|
1873 | else |
996 | { |
1874 | { |
997 | if (idle > max_idle) |
1875 | /* initialise timeout once */ |
998 | { |
1876 | if (!ts.tv_sec) |
999 | --idle; |
1877 | ts.tv_sec = time (0) + idle_timeout; |
1000 | X_UNLOCK (reqlock); |
|
|
1001 | X_LOCK (wrklock); |
|
|
1002 | --started; |
|
|
1003 | X_UNLOCK (wrklock); |
|
|
1004 | goto quit; |
|
|
1005 | } |
|
|
1006 | |
1878 | |
1007 | /* we are allowed to idle, so do so without any timeout */ |
|
|
1008 | X_COND_WAIT (reqwait, reqlock); |
1879 | if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) |
|
|
1880 | ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */ |
1009 | } |
1881 | } |
1010 | |
1882 | |
1011 | --idle; |
1883 | --idle; |
1012 | } |
1884 | } |
1013 | |
1885 | |
… | |
… | |
1016 | X_UNLOCK (reqlock); |
1888 | X_UNLOCK (reqlock); |
1017 | |
1889 | |
1018 | if (req->type < 0) |
1890 | if (req->type < 0) |
1019 | goto quit; |
1891 | goto quit; |
1020 | |
1892 | |
1021 | if (!EIO_CANCELLED (req)) |
|
|
1022 | ETP_EXECUTE (self, req); |
1893 | ETP_EXECUTE (self, req); |
1023 | |
1894 | |
1024 | X_LOCK (reslock); |
1895 | X_LOCK (reslock); |
1025 | |
1896 | |
1026 | ++npending; |
1897 | ++npending; |
1027 | |
1898 | |
… | |
… | |
1042 | return 0; |
1913 | return 0; |
1043 | } |
1914 | } |
1044 | |
1915 | |
1045 | /*****************************************************************************/ |
1916 | /*****************************************************************************/ |
1046 | |
1917 | |
|
|
1918 | int ecb_cold |
1047 | int eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1919 | eio_init (void (*want_poll)(void), void (*done_poll)(void)) |
1048 | { |
1920 | { |
1049 | return etp_init (want_poll, done_poll); |
1921 | return etp_init (want_poll, done_poll); |
1050 | } |
1922 | } |
1051 | |
1923 | |
|
|
1924 | ecb_inline void |
1052 | static void eio_api_destroy (eio_req *req) |
1925 | eio_api_destroy (eio_req *req) |
1053 | { |
1926 | { |
1054 | free (req); |
1927 | free (req); |
1055 | } |
1928 | } |
1056 | |
1929 | |
1057 | #define REQ(rtype) \ |
1930 | #define REQ(rtype) \ |
… | |
… | |
1076 | { \ |
1949 | { \ |
1077 | eio_api_destroy (req); \ |
1950 | eio_api_destroy (req); \ |
1078 | return 0; \ |
1951 | return 0; \ |
1079 | } |
1952 | } |
1080 | |
1953 | |
|
|
1954 | static void |
1081 | static void eio_execute (etp_worker *self, eio_req *req) |
1955 | eio_execute (etp_worker *self, eio_req *req) |
1082 | { |
1956 | { |
1083 | errno = 0; |
1957 | if (ecb_expect_false (EIO_CANCELLED (req))) |
|
|
1958 | { |
|
|
1959 | req->result = -1; |
|
|
1960 | req->errorno = ECANCELED; |
|
|
1961 | return; |
|
|
1962 | } |
1084 | |
1963 | |
1085 | switch (req->type) |
1964 | switch (req->type) |
1086 | { |
1965 | { |
1087 | case EIO_READ: ALLOC (req->size); |
1966 | case EIO_READ: ALLOC (req->size); |
1088 | req->result = req->offs >= 0 |
1967 | req->result = req->offs >= 0 |
… | |
… | |
1100 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1979 | case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1101 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1980 | req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1102 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1981 | case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); |
1103 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1982 | req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; |
1104 | |
1983 | |
|
|
1984 | case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1985 | req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1986 | case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); |
|
|
1987 | req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; |
|
|
1988 | |
1105 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1989 | case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break; |
1106 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1990 | case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; |
1107 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1991 | case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break; |
1108 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1992 | case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; |
1109 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
1993 | case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break; |
… | |
… | |
1116 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
2000 | case EIO_RMDIR: req->result = rmdir (req->ptr1); break; |
1117 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
2001 | case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break; |
1118 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
2002 | case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break; |
1119 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
2003 | case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break; |
1120 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
2004 | case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break; |
1121 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->int3); break; |
2005 | case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break; |
1122 | |
2006 | |
|
|
2007 | case EIO_REALPATH: eio__realpath (req, self); break; |
|
|
2008 | |
1123 | case EIO_READLINK: ALLOC (NAME_MAX); |
2009 | case EIO_READLINK: ALLOC (PATH_MAX); |
1124 | req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break; |
2010 | req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break; |
1125 | |
2011 | |
1126 | case EIO_SYNC: req->result = 0; sync (); break; |
2012 | case EIO_SYNC: req->result = 0; sync (); break; |
1127 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
2013 | case EIO_FSYNC: req->result = fsync (req->int1); break; |
1128 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
2014 | case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; |
|
|
2015 | case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; |
|
|
2016 | case EIO_MTOUCH: req->result = eio__mtouch (req); break; |
|
|
2017 | case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; |
|
|
2018 | case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; |
|
|
2019 | case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; |
|
|
2020 | case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; |
1129 | |
2021 | |
1130 | case EIO_READDIR: eio__scandir (req, self); break; |
2022 | case EIO_READDIR: eio__scandir (req, self); break; |
1131 | |
2023 | |
1132 | case EIO_BUSY: |
2024 | case EIO_BUSY: |
1133 | #ifdef _WIN32 |
2025 | #ifdef _WIN32 |
1134 | Sleep (req->nv1 * 1000.); |
2026 | Sleep (req->nv1 * 1e3); |
1135 | #else |
2027 | #else |
1136 | { |
2028 | { |
1137 | struct timeval tv; |
2029 | struct timeval tv; |
1138 | |
2030 | |
1139 | tv.tv_sec = req->nv1; |
2031 | tv.tv_sec = req->nv1; |
1140 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.; |
2032 | tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6; |
1141 | |
2033 | |
1142 | req->result = select (0, 0, 0, 0, &tv); |
2034 | req->result = select (0, 0, 0, 0, &tv); |
1143 | } |
2035 | } |
1144 | #endif |
2036 | #endif |
1145 | break; |
2037 | break; |
… | |
… | |
1160 | times = tv; |
2052 | times = tv; |
1161 | } |
2053 | } |
1162 | else |
2054 | else |
1163 | times = 0; |
2055 | times = 0; |
1164 | |
2056 | |
1165 | |
|
|
1166 | req->result = req->type == EIO_FUTIME |
2057 | req->result = req->type == EIO_FUTIME |
1167 | ? futimes (req->int1, times) |
2058 | ? futimes (req->int1, times) |
1168 | : utimes (req->ptr1, times); |
2059 | : utimes (req->ptr1, times); |
1169 | } |
2060 | } |
|
|
2061 | break; |
1170 | |
2062 | |
1171 | case EIO_GROUP: |
2063 | case EIO_GROUP: |
|
|
2064 | abort (); /* handled in eio_request */ |
|
|
2065 | |
1172 | case EIO_NOP: |
2066 | case EIO_NOP: |
1173 | req->result = 0; |
2067 | req->result = 0; |
1174 | break; |
2068 | break; |
1175 | |
2069 | |
1176 | case EIO_CUSTOM: |
2070 | case EIO_CUSTOM: |
1177 | req->feed (req); |
2071 | req->feed (req); |
1178 | break; |
2072 | break; |
1179 | |
2073 | |
1180 | default: |
2074 | default: |
|
|
2075 | errno = ENOSYS; |
1181 | req->result = -1; |
2076 | req->result = -1; |
1182 | break; |
2077 | break; |
1183 | } |
2078 | } |
1184 | |
2079 | |
1185 | req->errorno = errno; |
2080 | req->errorno = errno; |
… | |
… | |
1205 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
2100 | eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) |
1206 | { |
2101 | { |
1207 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
2102 | REQ (EIO_FSYNC); req->int1 = fd; SEND; |
1208 | } |
2103 | } |
1209 | |
2104 | |
|
|
2105 | eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
2106 | { |
|
|
2107 | REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
2108 | } |
|
|
2109 | |
|
|
2110 | eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) |
|
|
2111 | { |
|
|
2112 | REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; |
|
|
2113 | } |
|
|
2114 | |
|
|
2115 | eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data) |
|
|
2116 | { |
|
|
2117 | REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND; |
|
|
2118 | } |
|
|
2119 | |
|
|
2120 | eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) |
|
|
2121 | { |
|
|
2122 | REQ (EIO_MLOCKALL); req->int1 = flags; SEND; |
|
|
2123 | } |
|
|
2124 | |
|
|
2125 | eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) |
|
|
2126 | { |
|
|
2127 | REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; |
|
|
2128 | } |
|
|
2129 | |
|
|
2130 | eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) |
|
|
2131 | { |
|
|
2132 | REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; |
|
|
2133 | } |
|
|
2134 | |
1210 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
2135 | eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) |
1211 | { |
2136 | { |
1212 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
2137 | REQ (EIO_FDATASYNC); req->int1 = fd; SEND; |
1213 | } |
2138 | } |
1214 | |
2139 | |
… | |
… | |
1233 | } |
2158 | } |
1234 | |
2159 | |
1235 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
2160 | eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) |
1236 | { |
2161 | { |
1237 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
2162 | REQ (EIO_FSTAT); req->int1 = fd; SEND; |
|
|
2163 | } |
|
|
2164 | |
|
|
2165 | eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data) |
|
|
2166 | { |
|
|
2167 | REQ (EIO_FSTATVFS); req->int1 = fd; SEND; |
1238 | } |
2168 | } |
1239 | |
2169 | |
1240 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
2170 | eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) |
1241 | { |
2171 | { |
1242 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
2172 | REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; |
… | |
… | |
1306 | eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
2236 | eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) |
1307 | { |
2237 | { |
1308 | return eio__1path (EIO_READLINK, path, pri, cb, data); |
2238 | return eio__1path (EIO_READLINK, path, pri, cb, data); |
1309 | } |
2239 | } |
1310 | |
2240 | |
|
|
2241 | eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data) |
|
|
2242 | { |
|
|
2243 | return eio__1path (EIO_REALPATH, path, pri, cb, data); |
|
|
2244 | } |
|
|
2245 | |
1311 | eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
2246 | eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) |
1312 | { |
2247 | { |
1313 | return eio__1path (EIO_STAT, path, pri, cb, data); |
2248 | return eio__1path (EIO_STAT, path, pri, cb, data); |
1314 | } |
2249 | } |
1315 | |
2250 | |
1316 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
2251 | eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) |
1317 | { |
2252 | { |
1318 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
2253 | return eio__1path (EIO_LSTAT, path, pri, cb, data); |
1319 | } |
2254 | } |
1320 | |
2255 | |
|
|
2256 | eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data) |
|
|
2257 | { |
|
|
2258 | return eio__1path (EIO_STATVFS, path, pri, cb, data); |
|
|
2259 | } |
|
|
2260 | |
1321 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
2261 | eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) |
1322 | { |
2262 | { |
1323 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
2263 | return eio__1path (EIO_UNLINK, path, pri, cb, data); |
1324 | } |
2264 | } |
1325 | |
2265 | |
1326 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
2266 | eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) |
1327 | { |
2267 | { |
1328 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
2268 | return eio__1path (EIO_RMDIR, path, pri, cb, data); |
1329 | } |
2269 | } |
1330 | |
2270 | |
1331 | eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) |
2271 | eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data) |
1332 | { |
2272 | { |
1333 | return eio__1path (EIO_READDIR, path, pri, cb, data); |
2273 | REQ (EIO_READDIR); PATH; req->int1 = flags; SEND; |
1334 | } |
2274 | } |
1335 | |
2275 | |
1336 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
2276 | eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) |
1337 | { |
2277 | { |
1338 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int3 = (long)dev; SEND; |
2278 | REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND; |
1339 | } |
2279 | } |
1340 | |
2280 | |
1341 | static eio_req * |
2281 | static eio_req * |
1342 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2282 | eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1343 | { |
2283 | { |
… | |
… | |
1367 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
2307 | eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) |
1368 | { |
2308 | { |
1369 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
2309 | return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); |
1370 | } |
2310 | } |
1371 | |
2311 | |
1372 | eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) |
2312 | eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) |
1373 | { |
2313 | { |
1374 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
2314 | REQ (EIO_CUSTOM); req->feed = execute; SEND; |
1375 | } |
2315 | } |
1376 | |
2316 | |
1377 | #endif |
2317 | #endif |
… | |
… | |
1388 | #undef SEND |
2328 | #undef SEND |
1389 | |
2329 | |
1390 | /*****************************************************************************/ |
2330 | /*****************************************************************************/ |
1391 | /* grp functions */ |
2331 | /* grp functions */ |
1392 | |
2332 | |
|
|
2333 | void |
1393 | void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
2334 | eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) |
1394 | { |
2335 | { |
1395 | grp->int2 = limit; |
2336 | grp->int2 = limit; |
1396 | grp->feed = feed; |
2337 | grp->feed = feed; |
1397 | |
2338 | |
1398 | grp_try_feed (grp); |
2339 | grp_try_feed (grp); |
1399 | } |
2340 | } |
1400 | |
2341 | |
|
|
2342 | void |
1401 | void eio_grp_limit (eio_req *grp, int limit) |
2343 | eio_grp_limit (eio_req *grp, int limit) |
1402 | { |
2344 | { |
1403 | grp->int2 = limit; |
2345 | grp->int2 = limit; |
1404 | |
2346 | |
1405 | grp_try_feed (grp); |
2347 | grp_try_feed (grp); |
1406 | } |
2348 | } |
1407 | |
2349 | |
|
|
2350 | void |
1408 | void eio_grp_add (eio_req *grp, eio_req *req) |
2351 | eio_grp_add (eio_req *grp, eio_req *req) |
1409 | { |
2352 | { |
1410 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
2353 | assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); |
|
|
2354 | |
|
|
2355 | grp->flags |= EIO_FLAG_GROUPADD; |
1411 | |
2356 | |
1412 | ++grp->size; |
2357 | ++grp->size; |
1413 | req->grp = grp; |
2358 | req->grp = grp; |
1414 | |
2359 | |
1415 | req->grp_prev = 0; |
2360 | req->grp_prev = 0; |
… | |
… | |
1422 | } |
2367 | } |
1423 | |
2368 | |
1424 | /*****************************************************************************/ |
2369 | /*****************************************************************************/ |
1425 | /* misc garbage */ |
2370 | /* misc garbage */ |
1426 | |
2371 | |
|
|
2372 | ssize_t |
1427 | ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
2373 | eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) |
1428 | { |
2374 | { |
1429 | etp_worker wrk; |
2375 | etp_worker wrk; |
|
|
2376 | ssize_t ret; |
1430 | |
2377 | |
1431 | wrk.dbuf = 0; |
2378 | wrk.dbuf = 0; |
1432 | |
2379 | |
1433 | eio__sendfile (ofd, ifd, offset, count, &wrk); |
2380 | ret = eio__sendfile (ofd, ifd, offset, count, &wrk); |
1434 | |
2381 | |
1435 | if (wrk.dbuf) |
2382 | if (wrk.dbuf) |
1436 | free (wrk.dbuf); |
2383 | free (wrk.dbuf); |
1437 | } |
|
|
1438 | |
2384 | |
|
|
2385 | return ret; |
|
|
2386 | } |
|
|
2387 | |