ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/cvsroot/libeio/eio.c
(Generate patch)

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.61 by root, Fri Feb 11 00:53:24 2011 UTC vs.
Revision 1.116 by root, Tue Mar 27 19:27:35 2012 UTC

35 * and other provisions required by the GPL. If you do not delete the 35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under 36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL. 37 * either the BSD or the GPL.
38 */ 38 */
39 39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
40#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
41 46
42#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
43# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
44#endif 49#endif
45#include "xthread.h" 50#include "xthread.h"
46 51
47#include <errno.h> 52#include <errno.h>
48#include <stddef.h> 53#include <stddef.h>
49#include <stdlib.h> 54#include <stdlib.h>
50#include <string.h> 55#include <string.h>
51#include <errno.h> 56#include <errno.h>
52#include <sys/types.h> 57#include <sys/types.h>
53#include <sys/stat.h> 58#include <sys/stat.h>
54#include <sys/statvfs.h>
55#include <limits.h> 59#include <limits.h>
56#include <fcntl.h> 60#include <fcntl.h>
57#include <assert.h> 61#include <assert.h>
58 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H
66# include <stdint.h>
67#endif
68
69#ifndef ECANCELED
70# define ECANCELED EDOM
71#endif
72#ifndef ELOOP
73# define ELOOP EDOM
74#endif
75
76#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
77# define ENOTSOCK WSAENOTSOCK
78#endif
79
80static void eio_destroy (eio_req *req);
81
59#ifndef EIO_FINISH 82#ifndef EIO_FINISH
60# 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
61#endif 84#endif
62 85
63#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
66 89
67#ifndef EIO_FEED 90#ifndef EIO_FEED
68# 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)
69#endif 92#endif
70 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
71#ifdef _WIN32 105#ifdef _WIN32
72 106
73 /*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
74#else 197#else
75 198
76# include "config.h"
77# include <sys/time.h> 199 #include <sys/time.h>
78# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
79# 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
80# include <utime.h> 237# include <utime.h>
81# include <signal.h>
82# include <dirent.h>
83
84#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
85# include <sys/mman.h>
86#endif
87
88/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
89# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
90# define _DIRENT_HAVE_D_TYPE /* sigh */
91# define D_INO(de) (de)->d_fileno
92# define D_NAMLEN(de) (de)->d_namlen
93# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
94# define D_INO(de) (de)->d_ino
95# endif 238#endif
96 239
97#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
98# undef D_NAMLEN 241# include <sys/syscall.h>
99# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
100#endif
101
102# ifdef _DIRENT_HAVE_D_TYPE
103# define D_TYPE(de) (de)->d_type
104# endif 242#endif
105 243
106# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
107# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
108# endif
109
110#endif 246#endif
111 247
112#if HAVE_SENDFILE 248#if HAVE_SENDFILE
113# if __linux 249# if __linux
114# include <sys/sendfile.h> 250# include <sys/sendfile.h>
129#endif 265#endif
130#ifndef D_INO 266#ifndef D_INO
131# define D_INO(de) 0 267# define D_INO(de) 0
132#endif 268#endif
133#ifndef D_NAMLEN 269#ifndef D_NAMLEN
134# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
135#endif 271#endif
136
137/* number of seconds after which an idle threads exit */
138#define IDLE_TIMEOUT 10
139 272
140/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
141#ifndef NAME_MAX 274#ifndef NAME_MAX
142# define NAME_MAX 4096 275# define NAME_MAX 4096
143#endif 276#endif
149 282
150/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
151#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
152 285
153#define dBUF \ 286#define dBUF \
154 char *eio_buf; \
155 ETP_WORKER_LOCK (self); \
156 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
157 ETP_WORKER_UNLOCK (self); \
158 errno = ENOMEM; \ 288 errno = ENOMEM; \
159 if (!eio_buf) \ 289 if (!eio_buf) \
160 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
161 294
162#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
163 296
164/*****************************************************************************/ 297/*****************************************************************************/
165 298
166#if __GNUC__ >= 3 299struct tmpbuf
167# define expect(expr,value) __builtin_expect ((expr),(value)) 300{
301 void *ptr;
302 int len;
303};
304
305static void *
306tmpbuf_get (struct tmpbuf *buf, int len)
307{
308 if (buf->len < len)
309 {
310 free (buf->ptr);
311 buf->ptr = malloc (buf->len = len);
312 }
313
314 return buf->ptr;
315}
316
317struct tmpbuf;
318
319#if _POSIX_VERSION >= 200809L
320 #define HAVE_AT 1
321 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
322 #ifndef O_SEARCH
323 #define O_SEARCH O_RDONLY
324 #endif
168#else 325#else
169# define expect(expr,value) (expr) 326 #define HAVE_AT 0
327 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
170#endif 328#endif
171 329
172#define expect_false(expr) expect ((expr) != 0, 0) 330struct eio_pwd
173#define expect_true(expr) expect ((expr) != 0, 1) 331{
332#if HAVE_AT
333 int fd;
334#endif
335 int len;
336 char str[1]; /* actually, a 0-terminated canonical path */
337};
174 338
175/*****************************************************************************/ 339/*****************************************************************************/
176 340
177#define ETP_PRI_MIN EIO_PRI_MIN 341#define ETP_PRI_MIN EIO_PRI_MIN
178#define ETP_PRI_MAX EIO_PRI_MAX 342#define ETP_PRI_MAX EIO_PRI_MAX
184static int eio_finish (eio_req *req); 348static int eio_finish (eio_req *req);
185#define ETP_FINISH(req) eio_finish (req) 349#define ETP_FINISH(req) eio_finish (req)
186static void eio_execute (struct etp_worker *self, eio_req *req); 350static void eio_execute (struct etp_worker *self, eio_req *req);
187#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
188 352
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/*****************************************************************************/ 353/*****************************************************************************/
207 354
208#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
209 356
210/* calculate time difference in ~1/EIO_TICKS of a second */ 357/* calculate time difference in ~1/EIO_TICKS of a second */
358ecb_inline int
211static int tvdiff (struct timeval *tv1, struct timeval *tv2) 359tvdiff (struct timeval *tv1, struct timeval *tv2)
212{ 360{
213 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 361 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
214 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 362 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
215} 363}
216 364
220static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
221 369
222static unsigned int max_poll_time; /* reslock */ 370static unsigned int max_poll_time; /* reslock */
223static unsigned int max_poll_reqs; /* reslock */ 371static unsigned int max_poll_reqs; /* reslock */
224 372
225static volatile unsigned int nreqs; /* reqlock */ 373static unsigned int nreqs; /* reqlock */
226static volatile unsigned int nready; /* reqlock */ 374static unsigned int nready; /* reqlock */
227static volatile unsigned int npending; /* reqlock */ 375static unsigned int npending; /* reqlock */
228static volatile unsigned int max_idle = 4; 376static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
377static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
229 378
230static xmutex_t wrklock = X_MUTEX_INIT; 379static xmutex_t wrklock;
231static xmutex_t reslock = X_MUTEX_INIT; 380static xmutex_t reslock;
232static xmutex_t reqlock = X_MUTEX_INIT; 381static xmutex_t reqlock;
233static xcond_t reqwait = X_COND_INIT; 382static xcond_t reqwait;
234 383
235#if !HAVE_PREADWRITE 384#if !HAVE_PREADWRITE
236/* 385/*
237 * make our pread/pwrite emulation safe against themselves, but not against 386 * make our pread/pwrite emulation safe against themselves, but not against
238 * normal read/write by using a mutex. slows down execution a lot, 387 * normal read/write by using a mutex. slows down execution a lot,
239 * but that's your problem, not mine. 388 * but that's your problem, not mine.
240 */ 389 */
241static xmutex_t preadwritelock = X_MUTEX_INIT; 390static xmutex_t preadwritelock;
242#endif 391#endif
243 392
244typedef struct etp_worker 393typedef struct etp_worker
245{ 394{
395 struct tmpbuf tmpbuf;
396
246 /* locked by wrklock */ 397 /* locked by wrklock */
247 struct etp_worker *prev, *next; 398 struct etp_worker *prev, *next;
248 399
249 xthread_t tid; 400 xthread_t tid;
250 401
251 /* locked by reslock, reqlock or wrklock */ 402#ifdef ETP_WORKER_COMMON
252 ETP_REQ *req; /* currently processed request */
253
254 ETP_WORKER_COMMON 403 ETP_WORKER_COMMON
404#endif
255} etp_worker; 405} etp_worker;
256 406
257static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 407static etp_worker wrk_first; /* NOT etp */
258 408
259#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
260#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
261 411
262/* worker threads management */ 412/* worker threads management */
263 413
414static void
264static void etp_worker_clear (etp_worker *wrk) 415etp_worker_clear (etp_worker *wrk)
265{ 416{
266 ETP_WORKER_CLEAR (wrk);
267} 417}
268 418
419static void ecb_cold
269static void etp_worker_free (etp_worker *wrk) 420etp_worker_free (etp_worker *wrk)
270{ 421{
422 free (wrk->tmpbuf.ptr);
423
271 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
272 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
273 426
274 free (wrk); 427 free (wrk);
275} 428}
276 429
277static unsigned int etp_nreqs (void) 430static unsigned int
431etp_nreqs (void)
278{ 432{
279 int retval; 433 int retval;
280 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 434 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
281 retval = nreqs; 435 retval = nreqs;
282 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 436 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
283 return retval; 437 return retval;
284} 438}
285 439
286static unsigned int etp_nready (void) 440static unsigned int
441etp_nready (void)
287{ 442{
288 unsigned int retval; 443 unsigned int retval;
289 444
290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 445 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
291 retval = nready; 446 retval = nready;
292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
293 448
294 return retval; 449 return retval;
295} 450}
296 451
297static unsigned int etp_npending (void) 452static unsigned int
453etp_npending (void)
298{ 454{
299 unsigned int retval; 455 unsigned int retval;
300 456
301 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 457 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
302 retval = npending; 458 retval = npending;
303 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 459 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
304 460
305 return retval; 461 return retval;
306} 462}
307 463
308static unsigned int etp_nthreads (void) 464static unsigned int
465etp_nthreads (void)
309{ 466{
310 unsigned int retval; 467 unsigned int retval;
311 468
312 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 469 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
313 retval = started; 470 retval = started;
327} etp_reqq; 484} etp_reqq;
328 485
329static etp_reqq req_queue; 486static etp_reqq req_queue;
330static etp_reqq res_queue; 487static etp_reqq res_queue;
331 488
489static void ecb_noinline ecb_cold
490reqq_init (etp_reqq *q)
491{
492 int pri;
493
494 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
495 q->qs[pri] = q->qe[pri] = 0;
496
497 q->size = 0;
498}
499
500static int ecb_noinline
332static int reqq_push (etp_reqq *q, ETP_REQ *req) 501reqq_push (etp_reqq *q, ETP_REQ *req)
333{ 502{
334 int pri = req->pri; 503 int pri = req->pri;
335 req->next = 0; 504 req->next = 0;
336 505
337 if (q->qe[pri]) 506 if (q->qe[pri])
343 q->qe[pri] = q->qs[pri] = req; 512 q->qe[pri] = q->qs[pri] = req;
344 513
345 return q->size++; 514 return q->size++;
346} 515}
347 516
517static ETP_REQ * ecb_noinline
348static ETP_REQ *reqq_shift (etp_reqq *q) 518reqq_shift (etp_reqq *q)
349{ 519{
350 int pri; 520 int pri;
351 521
352 if (!q->size) 522 if (!q->size)
353 return 0; 523 return 0;
368 } 538 }
369 539
370 abort (); 540 abort ();
371} 541}
372 542
373static void etp_atfork_prepare (void) 543static int ecb_cold
544etp_init (void (*want_poll)(void), void (*done_poll)(void))
374{ 545{
375 X_LOCK (wrklock); 546 X_MUTEX_CREATE (wrklock);
376 X_LOCK (reqlock); 547 X_MUTEX_CREATE (reslock);
377 X_LOCK (reslock); 548 X_MUTEX_CREATE (reqlock);
378#if !HAVE_PREADWRITE 549 X_COND_CREATE (reqwait);
379 X_LOCK (preadwritelock);
380#endif
381}
382 550
383static void etp_atfork_parent (void) 551 reqq_init (&req_queue);
384{ 552 reqq_init (&res_queue);
385#if !HAVE_PREADWRITE
386 X_UNLOCK (preadwritelock);
387#endif
388 X_UNLOCK (reslock);
389 X_UNLOCK (reqlock);
390 X_UNLOCK (wrklock);
391}
392 553
393static void etp_atfork_child (void) 554 wrk_first.next =
394{ 555 wrk_first.prev = &wrk_first;
395 ETP_REQ *prv;
396
397 while ((prv = reqq_shift (&req_queue)))
398 ETP_DESTROY (prv);
399
400 while ((prv = reqq_shift (&res_queue)))
401 ETP_DESTROY (prv);
402
403 while (wrk_first.next != &wrk_first)
404 {
405 etp_worker *wrk = wrk_first.next;
406
407 if (wrk->req)
408 ETP_DESTROY (wrk->req);
409
410 etp_worker_clear (wrk);
411 etp_worker_free (wrk);
412 }
413 556
414 started = 0; 557 started = 0;
415 idle = 0; 558 idle = 0;
416 nreqs = 0; 559 nreqs = 0;
417 nready = 0; 560 nready = 0;
418 npending = 0; 561 npending = 0;
419 562
420 etp_atfork_parent ();
421}
422
423static void
424etp_once_init (void)
425{
426 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
427}
428
429static int
430etp_init (void (*want_poll)(void), void (*done_poll)(void))
431{
432 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
433
434 pthread_once (&doinit, etp_once_init);
435
436 want_poll_cb = want_poll; 563 want_poll_cb = want_poll;
437 done_poll_cb = done_poll; 564 done_poll_cb = done_poll;
438 565
439 return 0; 566 return 0;
440} 567}
441 568
442X_THREAD_PROC (etp_proc); 569X_THREAD_PROC (etp_proc);
443 570
571static void ecb_cold
444static void etp_start_thread (void) 572etp_start_thread (void)
445{ 573{
446 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 574 etp_worker *wrk = calloc (1, sizeof (etp_worker));
447 575
448 /*TODO*/ 576 /*TODO*/
449 assert (("unable to allocate worker thread data", wrk)); 577 assert (("unable to allocate worker thread data", wrk));
462 free (wrk); 590 free (wrk);
463 591
464 X_UNLOCK (wrklock); 592 X_UNLOCK (wrklock);
465} 593}
466 594
595static void
467static void etp_maybe_start_thread (void) 596etp_maybe_start_thread (void)
468{ 597{
469 if (expect_true (etp_nthreads () >= wanted)) 598 if (ecb_expect_true (etp_nthreads () >= wanted))
470 return; 599 return;
471 600
472 /* todo: maybe use idle here, but might be less exact */ 601 /* todo: maybe use idle here, but might be less exact */
473 if (expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ())) 602 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
474 return; 603 return;
475 604
476 etp_start_thread (); 605 etp_start_thread ();
477} 606}
478 607
608static void ecb_cold
479static void etp_end_thread (void) 609etp_end_thread (void)
480{ 610{
481 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
482 612
483 req->type = -1; 613 req->type = -1;
484 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
485 615
486 X_LOCK (reqlock); 616 X_LOCK (reqlock);
491 X_LOCK (wrklock); 621 X_LOCK (wrklock);
492 --started; 622 --started;
493 X_UNLOCK (wrklock); 623 X_UNLOCK (wrklock);
494} 624}
495 625
496static int etp_poll (void) 626static int
627etp_poll (void)
497{ 628{
498 unsigned int maxreqs; 629 unsigned int maxreqs;
499 unsigned int maxtime; 630 unsigned int maxtime;
500 struct timeval tv_start, tv_now; 631 struct timeval tv_start, tv_now;
501 632
531 662
532 X_LOCK (reqlock); 663 X_LOCK (reqlock);
533 --nreqs; 664 --nreqs;
534 X_UNLOCK (reqlock); 665 X_UNLOCK (reqlock);
535 666
536 if (expect_false (req->type == EIO_GROUP && req->size)) 667 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
537 { 668 {
538 req->int1 = 1; /* mark request as delayed */ 669 req->int1 = 1; /* mark request as delayed */
539 continue; 670 continue;
540 } 671 }
541 else 672 else
542 { 673 {
543 int res = ETP_FINISH (req); 674 int res = ETP_FINISH (req);
544 if (expect_false (res)) 675 if (ecb_expect_false (res))
545 return res; 676 return res;
546 } 677 }
547 678
548 if (expect_false (maxreqs && !--maxreqs)) 679 if (ecb_expect_false (maxreqs && !--maxreqs))
549 break; 680 break;
550 681
551 if (maxtime) 682 if (maxtime)
552 { 683 {
553 gettimeofday (&tv_now, 0); 684 gettimeofday (&tv_now, 0);
559 690
560 errno = EAGAIN; 691 errno = EAGAIN;
561 return -1; 692 return -1;
562} 693}
563 694
695static void
564static void etp_cancel (ETP_REQ *req) 696etp_cancel (ETP_REQ *req)
565{ 697{
566 X_LOCK (wrklock); 698 req->cancelled = 1;
567 req->flags |= EIO_FLAG_CANCELLED;
568 X_UNLOCK (wrklock);
569 699
570 eio_grp_cancel (req); 700 eio_grp_cancel (req);
571} 701}
572 702
703static void
573static void etp_submit (ETP_REQ *req) 704etp_submit (ETP_REQ *req)
574{ 705{
575 req->pri -= ETP_PRI_MIN; 706 req->pri -= ETP_PRI_MIN;
576 707
577 if (expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN; 708 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
578 if (expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 709 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
579 710
580 if (expect_false (req->type == EIO_GROUP)) 711 if (ecb_expect_false (req->type == EIO_GROUP))
581 { 712 {
582 /* I hope this is worth it :/ */ 713 /* I hope this is worth it :/ */
583 X_LOCK (reqlock); 714 X_LOCK (reqlock);
584 ++nreqs; 715 ++nreqs;
585 X_UNLOCK (reqlock); 716 X_UNLOCK (reqlock);
604 735
605 etp_maybe_start_thread (); 736 etp_maybe_start_thread ();
606 } 737 }
607} 738}
608 739
740static void ecb_cold
609static void etp_set_max_poll_time (double nseconds) 741etp_set_max_poll_time (double nseconds)
610{ 742{
611 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 743 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
612 max_poll_time = nseconds * EIO_TICKS; 744 max_poll_time = nseconds * EIO_TICKS;
613 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 745 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
614} 746}
615 747
748static void ecb_cold
616static void etp_set_max_poll_reqs (unsigned int maxreqs) 749etp_set_max_poll_reqs (unsigned int maxreqs)
617{ 750{
618 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 751 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
619 max_poll_reqs = maxreqs; 752 max_poll_reqs = maxreqs;
620 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 753 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
621} 754}
622 755
756static void ecb_cold
623static void etp_set_max_idle (unsigned int nthreads) 757etp_set_max_idle (unsigned int nthreads)
624{ 758{
625 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 759 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
626 max_idle = nthreads <= 0 ? 1 : nthreads; 760 max_idle = nthreads;
627 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 761 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
628} 762}
629 763
764static void ecb_cold
765etp_set_idle_timeout (unsigned int seconds)
766{
767 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
768 idle_timeout = seconds;
769 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
770}
771
772static void ecb_cold
630static void etp_set_min_parallel (unsigned int nthreads) 773etp_set_min_parallel (unsigned int nthreads)
631{ 774{
632 if (wanted < nthreads) 775 if (wanted < nthreads)
633 wanted = nthreads; 776 wanted = nthreads;
634} 777}
635 778
779static void ecb_cold
636static void etp_set_max_parallel (unsigned int nthreads) 780etp_set_max_parallel (unsigned int nthreads)
637{ 781{
638 if (wanted > nthreads) 782 if (wanted > nthreads)
639 wanted = nthreads; 783 wanted = nthreads;
640 784
641 while (started > wanted) 785 while (started > wanted)
642 etp_end_thread (); 786 etp_end_thread ();
643} 787}
644 788
645/*****************************************************************************/ 789/*****************************************************************************/
646 790
791static void
647static void grp_try_feed (eio_req *grp) 792grp_try_feed (eio_req *grp)
648{ 793{
649 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 794 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
650 { 795 {
651 grp->flags &= ~EIO_FLAG_GROUPADD; 796 grp->flags &= ~EIO_FLAG_GROUPADD;
652 797
659 break; 804 break;
660 } 805 }
661 } 806 }
662} 807}
663 808
809static int
664static int grp_dec (eio_req *grp) 810grp_dec (eio_req *grp)
665{ 811{
666 --grp->size; 812 --grp->size;
667 813
668 /* call feeder, if applicable */ 814 /* call feeder, if applicable */
669 grp_try_feed (grp); 815 grp_try_feed (grp);
673 return eio_finish (grp); 819 return eio_finish (grp);
674 else 820 else
675 return 0; 821 return 0;
676} 822}
677 823
824static void
678void eio_destroy (eio_req *req) 825eio_destroy (eio_req *req)
679{ 826{
680 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 827 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
681 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 828 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
682 829
683 EIO_DESTROY (req); 830 EIO_DESTROY (req);
684} 831}
685 832
833static int
686static int eio_finish (eio_req *req) 834eio_finish (eio_req *req)
687{ 835{
688 int res = EIO_FINISH (req); 836 int res = EIO_FINISH (req);
689 837
690 if (req->grp) 838 if (req->grp)
691 { 839 {
699 if (grp->grp_first == req) 847 if (grp->grp_first == req)
700 grp->grp_first = req->grp_next; 848 grp->grp_first = req->grp_next;
701 849
702 res2 = grp_dec (grp); 850 res2 = grp_dec (grp);
703 851
704 if (!res && res2) 852 if (!res)
705 res = res2; 853 res = res2;
706 } 854 }
707 855
708 eio_destroy (req); 856 eio_destroy (req);
709 857
710 return res; 858 return res;
711} 859}
712 860
861void
713void eio_grp_cancel (eio_req *grp) 862eio_grp_cancel (eio_req *grp)
714{ 863{
715 for (grp = grp->grp_first; grp; grp = grp->grp_next) 864 for (grp = grp->grp_first; grp; grp = grp->grp_next)
716 eio_cancel (grp); 865 eio_cancel (grp);
717} 866}
718 867
868void
719void eio_cancel (eio_req *req) 869eio_cancel (eio_req *req)
720{ 870{
721 etp_cancel (req); 871 etp_cancel (req);
722} 872}
723 873
874void
724void eio_submit (eio_req *req) 875eio_submit (eio_req *req)
725{ 876{
726 etp_submit (req); 877 etp_submit (req);
727} 878}
728 879
729unsigned int eio_nreqs (void) 880unsigned int
881eio_nreqs (void)
730{ 882{
731 return etp_nreqs (); 883 return etp_nreqs ();
732} 884}
733 885
734unsigned int eio_nready (void) 886unsigned int
887eio_nready (void)
735{ 888{
736 return etp_nready (); 889 return etp_nready ();
737} 890}
738 891
739unsigned int eio_npending (void) 892unsigned int
893eio_npending (void)
740{ 894{
741 return etp_npending (); 895 return etp_npending ();
742} 896}
743 897
744unsigned int eio_nthreads (void) 898unsigned int ecb_cold
899eio_nthreads (void)
745{ 900{
746 return etp_nthreads (); 901 return etp_nthreads ();
747} 902}
748 903
904void ecb_cold
749void eio_set_max_poll_time (double nseconds) 905eio_set_max_poll_time (double nseconds)
750{ 906{
751 etp_set_max_poll_time (nseconds); 907 etp_set_max_poll_time (nseconds);
752} 908}
753 909
910void ecb_cold
754void eio_set_max_poll_reqs (unsigned int maxreqs) 911eio_set_max_poll_reqs (unsigned int maxreqs)
755{ 912{
756 etp_set_max_poll_reqs (maxreqs); 913 etp_set_max_poll_reqs (maxreqs);
757} 914}
758 915
916void ecb_cold
759void eio_set_max_idle (unsigned int nthreads) 917eio_set_max_idle (unsigned int nthreads)
760{ 918{
761 etp_set_max_idle (nthreads); 919 etp_set_max_idle (nthreads);
762} 920}
763 921
922void ecb_cold
923eio_set_idle_timeout (unsigned int seconds)
924{
925 etp_set_idle_timeout (seconds);
926}
927
928void ecb_cold
764void eio_set_min_parallel (unsigned int nthreads) 929eio_set_min_parallel (unsigned int nthreads)
765{ 930{
766 etp_set_min_parallel (nthreads); 931 etp_set_min_parallel (nthreads);
767} 932}
768 933
934void ecb_cold
769void eio_set_max_parallel (unsigned int nthreads) 935eio_set_max_parallel (unsigned int nthreads)
770{ 936{
771 etp_set_max_parallel (nthreads); 937 etp_set_max_parallel (nthreads);
772} 938}
773 939
774int eio_poll (void) 940int eio_poll (void)
783# undef pread 949# undef pread
784# undef pwrite 950# undef pwrite
785# define pread eio__pread 951# define pread eio__pread
786# define pwrite eio__pwrite 952# define pwrite eio__pwrite
787 953
788static ssize_t 954static eio_ssize_t
789eio__pread (int fd, void *buf, size_t count, off_t offset) 955eio__pread (int fd, void *buf, size_t count, off_t offset)
790{ 956{
791 ssize_t res; 957 eio_ssize_t res;
792 off_t ooffset; 958 off_t ooffset;
793 959
794 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
795 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
796 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
799 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
800 966
801 return res; 967 return res;
802} 968}
803 969
804static ssize_t 970static eio_ssize_t
805eio__pwrite (int fd, void *buf, size_t count, off_t offset) 971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
806{ 972{
807 ssize_t res; 973 eio_ssize_t res;
808 off_t ooffset; 974 off_t ooffset;
809 975
810 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
811 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
812 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
844#ifndef HAVE_FUTIMES 1010#ifndef HAVE_FUTIMES
845 1011
846# undef futimes 1012# undef futimes
847# define futimes(fd,times) eio__futimes (fd, times) 1013# define futimes(fd,times) eio__futimes (fd, times)
848 1014
1015static int
849static int eio__futimes (int fd, const struct timeval tv[2]) 1016eio__futimes (int fd, const struct timeval tv[2])
850{ 1017{
851 errno = ENOSYS; 1018 errno = ENOSYS;
852 return -1; 1019 return -1;
853} 1020}
854 1021
857#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
858# undef fdatasync 1025# undef fdatasync
859# define fdatasync(fd) fsync (fd) 1026# define fdatasync(fd) fsync (fd)
860#endif 1027#endif
861 1028
1029static int
1030eio__syncfs (int fd)
1031{
1032 int res;
1033
1034#if HAVE_SYS_SYNCFS
1035 res = (int)syscall (__NR_syncfs, (int)(fd));
1036#else
1037 res = -1;
1038 errno = ENOSYS;
1039#endif
1040
1041 if (res < 0 && errno == ENOSYS && fd >= 0)
1042 sync ();
1043
1044 return res;
1045}
1046
862/* sync_file_range always needs emulation */ 1047/* sync_file_range always needs emulation */
863int 1048static int
864eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1049eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
865{ 1050{
866#if HAVE_SYNC_FILE_RANGE 1051#if HAVE_SYNC_FILE_RANGE
867 int res; 1052 int res;
868 1053
885 /* even though we could play tricks with the flags, it's better to always 1070 /* even though we could play tricks with the flags, it's better to always
886 * call fdatasync, as that matches the expectation of its users best */ 1071 * call fdatasync, as that matches the expectation of its users best */
887 return fdatasync (fd); 1072 return fdatasync (fd);
888} 1073}
889 1074
1075static int
1076eio__fallocate (int fd, int mode, off_t offset, size_t len)
1077{
1078#if HAVE_FALLOCATE
1079 return fallocate (fd, mode, offset, len);
1080#else
1081 errno = ENOSYS;
1082 return -1;
1083#endif
1084}
1085
890#if !HAVE_READAHEAD 1086#if !HAVE_READAHEAD
891# undef readahead 1087# undef readahead
892# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1088# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
893 1089
894static ssize_t 1090static eio_ssize_t
895eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1091eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
896{ 1092{
897 size_t todo = count; 1093 size_t todo = count;
898 dBUF; 1094 dBUF;
899 1095
904 pread (fd, eio_buf, len, offset); 1100 pread (fd, eio_buf, len, offset);
905 offset += len; 1101 offset += len;
906 todo -= len; 1102 todo -= len;
907 } 1103 }
908 1104
909 errno = 0; 1105 FUBd;
1106
1107 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1108 /* but not for e.g. EIO or eof, so we also never fail */
910 return count; 1109 return 0;
911} 1110}
912 1111
913#endif 1112#endif
914 1113
915/* sendfile always needs emulation */ 1114/* sendfile always needs emulation */
916static ssize_t 1115static eio_ssize_t
917eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1116eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
918{ 1117{
1118 eio_ssize_t written = 0;
919 ssize_t res; 1119 eio_ssize_t res;
920 1120
921 if (!count) 1121 if (!count)
922 return 0; 1122 return 0;
923 1123
1124 for (;;)
1125 {
1126#ifdef __APPLE__
1127# undef HAVE_SENDFILE /* broken, as everything on os x */
1128#endif
924#if HAVE_SENDFILE 1129#if HAVE_SENDFILE
925# if __linux 1130# if __linux
1131 off_t soffset = offset;
926 res = sendfile (ofd, ifd, &offset, count); 1132 res = sendfile (ofd, ifd, &soffset, count);
927 1133
928# elif __FreeBSD__ 1134# elif __FreeBSD__
929 /* 1135 /*
930 * Of course, the freebsd sendfile is a dire hack with no thoughts 1136 * Of course, the freebsd sendfile is a dire hack with no thoughts
931 * wasted on making it similar to other I/O functions. 1137 * wasted on making it similar to other I/O functions.
932 */ 1138 */
933 {
934 off_t sbytes; 1139 off_t sbytes;
935 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1140 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
936 1141
937 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1142 #if 0 /* according to the manpage, this is correct, but broken behaviour */
938 /* freebsd' sendfile will return 0 on success */ 1143 /* freebsd' sendfile will return 0 on success */
939 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1144 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
940 /* not on e.g. EIO or EPIPE - sounds broken */ 1145 /* not on e.g. EIO or EPIPE - sounds broken */
941 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1146 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
942 res = sbytes; 1147 res = sbytes;
943 #endif 1148 #endif
944 1149
945 /* according to source inspection, this is correct, and useful behaviour */ 1150 /* according to source inspection, this is correct, and useful behaviour */
946 if (sbytes) 1151 if (sbytes)
947 res = sbytes; 1152 res = sbytes;
948 }
949 1153
950# elif defined (__APPLE__) 1154# elif defined (__APPLE__)
951
952 {
953 off_t sbytes = count; 1155 off_t sbytes = count;
954 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1156 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
955 1157
956 /* according to the manpage, sbytes is always valid */ 1158 /* according to the manpage, sbytes is always valid */
957 if (sbytes) 1159 if (sbytes)
958 res = sbytes; 1160 res = sbytes;
959 }
960 1161
961# elif __hpux 1162# elif __hpux
962 res = sendfile (ofd, ifd, offset, count, 0, 0); 1163 res = sendfile (ofd, ifd, offset, count, 0, 0);
963 1164
964# elif __solaris 1165# elif __solaris
965 {
966 struct sendfilevec vec; 1166 struct sendfilevec vec;
967 size_t sbytes; 1167 size_t sbytes;
968 1168
969 vec.sfv_fd = ifd; 1169 vec.sfv_fd = ifd;
970 vec.sfv_flag = 0; 1170 vec.sfv_flag = 0;
971 vec.sfv_off = offset; 1171 vec.sfv_off = offset;
972 vec.sfv_len = count; 1172 vec.sfv_len = count;
973 1173
974 res = sendfilev (ofd, &vec, 1, &sbytes); 1174 res = sendfilev (ofd, &vec, 1, &sbytes);
975 1175
976 if (res < 0 && sbytes) 1176 if (res < 0 && sbytes)
977 res = sbytes; 1177 res = sbytes;
978 }
979 1178
980# endif 1179# endif
981 1180
982#elif defined (_WIN32) 1181#elif defined (_WIN32) && 0
983
984 /* does not work, just for documentation of what would need to be done */ 1182 /* does not work, just for documentation of what would need to be done */
985 { 1183 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1184 /* libeio guarantees that the file offset does not change, and windows */
1185 /* has no way to get an independent handle to the same file description */
986 HANDLE h = TO_SOCKET (ifd); 1186 HANDLE h = TO_SOCKET (ifd);
987 SetFilePointer (h, offset, 0, FILE_BEGIN); 1187 SetFilePointer (h, offset, 0, FILE_BEGIN);
988 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1188 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
989 }
990 1189
991#else 1190#else
992 res = -1; 1191 res = -1;
993 errno = ENOSYS; 1192 errno = ENOSYS;
994#endif 1193#endif
995 1194
1195 /* we assume sendfile can copy at least 128mb in one go */
1196 if (res <= 128 * 1024 * 1024)
1197 {
1198 if (res > 0)
1199 written += res;
1200
1201 if (written)
1202 return written;
1203
1204 break;
1205 }
1206 else
1207 {
1208 /* if we requested more, then probably the kernel was lazy */
1209 written += res;
1210 offset += res;
1211 count -= res;
1212
1213 if (!count)
1214 return written;
1215 }
1216 }
1217
996 if (res < 0 1218 if (res < 0
997 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1219 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
998 /* BSDs */ 1220 /* BSDs */
999#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1221#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1000 || errno == ENOTSUP 1222 || errno == ENOTSUP
1001#endif 1223#endif
1224#ifdef EOPNOTSUPP /* windows */
1002 || errno == EOPNOTSUPP /* BSDs */ 1225 || errno == EOPNOTSUPP /* BSDs */
1226#endif
1003#if __solaris 1227#if __solaris
1004 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1228 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1005#endif 1229#endif
1006 ) 1230 )
1007 ) 1231 )
1011 1235
1012 res = 0; 1236 res = 0;
1013 1237
1014 while (count) 1238 while (count)
1015 { 1239 {
1016 ssize_t cnt; 1240 eio_ssize_t cnt;
1017 1241
1018 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1242 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1019 1243
1020 if (cnt <= 0) 1244 if (cnt <= 0)
1021 { 1245 {
1033 1257
1034 offset += cnt; 1258 offset += cnt;
1035 res += cnt; 1259 res += cnt;
1036 count -= cnt; 1260 count -= cnt;
1037 } 1261 }
1262
1263 FUBd;
1038 } 1264 }
1039 1265
1040 return res; 1266 return res;
1267}
1268
1269#ifdef PAGESIZE
1270# define eio_pagesize() PAGESIZE
1271#else
1272static intptr_t
1273eio_pagesize (void)
1274{
1275 static intptr_t page;
1276
1277 if (!page)
1278 page = sysconf (_SC_PAGESIZE);
1279
1280 return page;
1281}
1282#endif
1283
1284static void
1285eio_page_align (void **addr, size_t *length)
1286{
1287 intptr_t mask = eio_pagesize () - 1;
1288
1289 /* round down addr */
1290 intptr_t adj = mask & (intptr_t)*addr;
1291
1292 *addr = (void *)((intptr_t)*addr - adj);
1293 *length += adj;
1294
1295 /* round up length */
1296 *length = (*length + mask) & ~mask;
1297}
1298
1299#if !_POSIX_MEMLOCK
1300# define eio__mlockall(a) EIO_ENOSYS ()
1301#else
1302
1303static int
1304eio__mlockall (int flags)
1305{
1306 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1307 extern int mallopt (int, int);
1308 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1309 #endif
1310
1311 if (EIO_MCL_CURRENT != MCL_CURRENT
1312 || EIO_MCL_FUTURE != MCL_FUTURE)
1313 {
1314 flags = 0
1315 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1316 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1317 }
1318
1319 return mlockall (flags);
1320}
1321#endif
1322
1323#if !_POSIX_MEMLOCK_RANGE
1324# define eio__mlock(a,b) EIO_ENOSYS ()
1325#else
1326
1327static int
1328eio__mlock (void *addr, size_t length)
1329{
1330 eio_page_align (&addr, &length);
1331
1332 return mlock (addr, length);
1333}
1334
1335#endif
1336
1337#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1338# define eio__msync(a,b,c) EIO_ENOSYS ()
1339#else
1340
1341static int
1342eio__msync (void *mem, size_t len, int flags)
1343{
1344 eio_page_align (&mem, &len);
1345
1346 if (EIO_MS_ASYNC != MS_SYNC
1347 || EIO_MS_INVALIDATE != MS_INVALIDATE
1348 || EIO_MS_SYNC != MS_SYNC)
1349 {
1350 flags = 0
1351 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1352 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1353 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1354 }
1355
1356 return msync (mem, len, flags);
1357}
1358
1359#endif
1360
1361static int
1362eio__mtouch (eio_req *req)
1363{
1364 void *mem = req->ptr2;
1365 size_t len = req->size;
1366 int flags = req->int1;
1367
1368 eio_page_align (&mem, &len);
1369
1370 {
1371 intptr_t addr = (intptr_t)mem;
1372 intptr_t end = addr + len;
1373 intptr_t page = eio_pagesize ();
1374
1375 if (addr < end)
1376 if (flags & EIO_MT_MODIFY) /* modify */
1377 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1378 else
1379 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1380 }
1381
1382 return 0;
1383}
1384
1385/*****************************************************************************/
1386/* requests implemented outside eio_execute, because they are so large */
1387
1388/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1389static int
1390eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1391{
1392 const char *rel = path;
1393 char *res;
1394 char *tmp1, *tmp2;
1395#if SYMLOOP_MAX > 32
1396 int symlinks = SYMLOOP_MAX;
1397#else
1398 int symlinks = 32;
1399#endif
1400
1401 errno = EINVAL;
1402 if (!rel)
1403 return -1;
1404
1405 errno = ENOENT;
1406 if (!*rel)
1407 return -1;
1408
1409 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1410 tmp1 = res + PATH_MAX;
1411 tmp2 = tmp1 + PATH_MAX;
1412
1413#if 0 /* disabled, the musl way to do things is just too racy */
1414#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1415 /* on linux we may be able to ask the kernel */
1416 {
1417 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1418
1419 if (fd >= 0)
1420 {
1421 sprintf (tmp1, "/proc/self/fd/%d", fd);
1422 req->result = readlink (tmp1, res, PATH_MAX);
1423 close (fd);
1424
1425 /* here we should probably stat the open file and the disk file, to make sure they still match */
1426
1427 if (req->result > 0)
1428 goto done;
1429 }
1430 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1431 return;
1432 }
1433#endif
1434#endif
1435
1436 if (*rel != '/')
1437 {
1438 int len;
1439
1440 errno = ENOENT;
1441 if (wd == EIO_INVALID_WD)
1442 return -1;
1443
1444 if (wd == EIO_CWD)
1445 {
1446 if (!getcwd (res, PATH_MAX))
1447 return -1;
1448
1449 len = strlen (res);
1450 }
1451 else
1452 memcpy (res, wd->str, len = wd->len);
1453
1454 if (res [1]) /* only use if not / */
1455 res += len;
1456 }
1457
1458 while (*rel)
1459 {
1460 eio_ssize_t len, linklen;
1461 const char *beg = rel;
1462
1463 while (*rel && *rel != '/')
1464 ++rel;
1465
1466 len = rel - beg;
1467
1468 if (!len) /* skip slashes */
1469 {
1470 ++rel;
1471 continue;
1472 }
1473
1474 if (beg [0] == '.')
1475 {
1476 if (len == 1)
1477 continue; /* . - nop */
1478
1479 if (beg [1] == '.' && len == 2)
1480 {
1481 /* .. - back up one component, if possible */
1482
1483 while (res != tmpbuf->ptr)
1484 if (*--res == '/')
1485 break;
1486
1487 continue;
1488 }
1489 }
1490
1491 errno = ENAMETOOLONG;
1492 if (res + 1 + len + 1 >= tmp1)
1493 return -1;
1494
1495 /* copy one component */
1496 *res = '/';
1497 memcpy (res + 1, beg, len);
1498
1499 /* zero-terminate, for readlink */
1500 res [len + 1] = 0;
1501
1502 /* now check if it's a symlink */
1503 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1504
1505 if (linklen < 0)
1506 {
1507 if (errno != EINVAL)
1508 return -1;
1509
1510 /* it's a normal directory. hopefully */
1511 res += len + 1;
1512 }
1513 else
1514 {
1515 /* yay, it was a symlink - build new path in tmp2 */
1516 int rellen = strlen (rel);
1517
1518 errno = ENAMETOOLONG;
1519 if (linklen + 1 + rellen >= PATH_MAX)
1520 return -1;
1521
1522 errno = ELOOP;
1523 if (!--symlinks)
1524 return -1;
1525
1526 if (*tmp1 == '/')
1527 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1528
1529 /* we need to be careful, as rel might point into tmp2 already */
1530 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1531 tmp2 [linklen] = '/';
1532 memcpy (tmp2, tmp1, linklen);
1533
1534 rel = tmp2;
1535 }
1536 }
1537
1538 /* special case for the lone root path */
1539 if (res == tmpbuf->ptr)
1540 *res++ = '/';
1541
1542 return res - (char *)tmpbuf->ptr;
1041} 1543}
1042 1544
1043static signed char 1545static signed char
1044eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1546eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1045{ 1547{
1046 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */ 1548 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1047 : a->inode < b->inode ? -1 : a->inode > b->inode ? 1 : 0; 1549 : a->inode < b->inode ? -1
1550 : a->inode > b->inode ? 1
1551 : 0;
1048} 1552}
1049 1553
1050#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0 1554#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1051 1555
1052#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1556#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1053#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1557#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1054 1558
1055static void 1559static void
1056eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1560eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1057{ 1561{
1058 unsigned char bits [9 + sizeof (ino_t) * 8]; 1562 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1059 unsigned char *bit = bits; 1563 unsigned char *bit = bits;
1060 1564
1061 assert (CHAR_BIT == 8); 1565 assert (CHAR_BIT == 8);
1062 assert (sizeof (eio_dirent) * 8 < 256); 1566 assert (sizeof (eio_dirent) * 8 < 256);
1063 assert (offsetof (eio_dirent, inode)); /* we use 0 as sentinel */ 1567 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1064 assert (offsetof (eio_dirent, score)); /* we use 0 as sentinel */ 1568 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1065 1569
1066 if (size <= EIO_SORT_FAST) 1570 if (size <= EIO_SORT_FAST)
1067 return; 1571 return;
1068 1572
1069 /* first prepare an array of bits to test in our radix sort */ 1573 /* first prepare an array of bits to test in our radix sort */
1070 /* try to take endianness into account, as well as differences in ino_t sizes */ 1574 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1071 /* inode_bits must contain all inodes ORed together */ 1575 /* inode_bits must contain all inodes ORed together */
1072 /* which is used to skip bits that are 0 everywhere, which is very common */ 1576 /* which is used to skip bits that are 0 everywhere, which is very common */
1073 { 1577 {
1074 ino_t endianness; 1578 eio_ino_t endianness;
1075 int i, j; 1579 int i, j;
1076 1580
1077 /* we store the byte offset of byte n into byte n of "endianness" */ 1581 /* we store the byte offset of byte n into byte n of "endianness" */
1078 for (i = 0; i < sizeof (ino_t); ++i) 1582 for (i = 0; i < sizeof (eio_ino_t); ++i)
1079 ((unsigned char *)&endianness)[i] = i; 1583 ((unsigned char *)&endianness)[i] = i;
1080 1584
1081 *bit++ = 0; 1585 *bit++ = 0;
1082 1586
1083 for (i = 0; i < sizeof (ino_t); ++i) 1587 for (i = 0; i < sizeof (eio_ino_t); ++i)
1084 { 1588 {
1085 /* shifting off the byte offsets out of "endianness" */ 1589 /* shifting off the byte offsets out of "endianness" */
1086 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1590 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1087 endianness >>= 8; 1591 endianness >>= 8;
1088 1592
1089 for (j = 0; j < 8; ++j) 1593 for (j = 0; j < 8; ++j)
1090 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1594 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1091 *bit++ = offs + j; 1595 *bit++ = offs + j;
1092 } 1596 }
1093 1597
1094 for (j = 0; j < 8; ++j) 1598 for (j = 0; j < 8; ++j)
1095 if (score_bits & (1 << j)) 1599 if (score_bits & (1 << j))
1096 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1600 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1097 } 1601 }
1098 1602
1099 /* now actually do the sorting (a variant of MSD radix sort) */ 1603 /* now actually do the sorting (a variant of MSD radix sort) */
1100 { 1604 {
1101 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1605 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1102 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1606 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1103 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1607 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1104 int stk_idx = 0; 1608 int stk_idx = 0;
1105 1609
1106 base_stk [stk_idx] = dents; 1610 base_stk [stk_idx] = dents;
1107 end_stk [stk_idx] = dents + size; 1611 end_stk [stk_idx] = dents + size;
1108 bit_stk [stk_idx] = bit - 1; 1612 bit_stk [stk_idx] = bit - 1;
1187 } 1691 }
1188 } 1692 }
1189} 1693}
1190 1694
1191static void 1695static void
1192eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1696eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1193{ 1697{
1194 if (size <= 1) 1698 if (size <= 1)
1195 return; /* our insertion sort relies on size > 0 */ 1699 return; /* our insertion sort relies on size > 0 */
1196 1700
1197 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1701 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1205 1709
1206/* read a full directory */ 1710/* read a full directory */
1207static void 1711static void
1208eio__scandir (eio_req *req, etp_worker *self) 1712eio__scandir (eio_req *req, etp_worker *self)
1209{ 1713{
1210 DIR *dirp;
1211 EIO_STRUCT_DIRENT *entp;
1212 char *name, *names; 1714 char *name, *names;
1213 int namesalloc = 4096; 1715 int namesalloc = 4096 - sizeof (void *) * 4;
1214 int namesoffs = 0; 1716 int namesoffs = 0;
1215 int flags = req->int1; 1717 int flags = req->int1;
1216 eio_dirent *dents = 0; 1718 eio_dirent *dents = 0;
1217 int dentalloc = 128; 1719 int dentalloc = 128;
1218 int dentoffs = 0; 1720 int dentoffs = 0;
1219 ino_t inode_bits = 0; 1721 eio_ino_t inode_bits = 0;
1722#ifdef _WIN32
1723 HANDLE dirp;
1724 WIN32_FIND_DATA entp;
1725#else
1726 DIR *dirp;
1727 EIO_STRUCT_DIRENT *entp;
1728#endif
1220 1729
1221 req->result = -1; 1730 req->result = -1;
1222 1731
1223 if (!(flags & EIO_READDIR_DENTS)) 1732 if (!(flags & EIO_READDIR_DENTS))
1224 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1733 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1225 1734
1226 X_LOCK (wrklock); 1735#ifdef _WIN32
1227 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1736 {
1737 int len = strlen ((const char *)req->ptr1);
1738 char *path = malloc (MAX_PATH);
1739 const char *fmt;
1740 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1741
1742 if (!len)
1743 fmt = "./*";
1744 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1745 fmt = "%s*";
1746 else
1747 fmt = "%s/*";
1748
1749 _snprintf (path, MAX_PATH, fmt, reqpath);
1750 dirp = FindFirstFile (path, &entp);
1751 free (path);
1752
1753 if (dirp == INVALID_HANDLE_VALUE)
1754 {
1755 /* should steal _dosmaperr */
1756 switch (GetLastError ())
1757 {
1758 case ERROR_FILE_NOT_FOUND:
1759 req->result = 0;
1760 break;
1761
1762 case ERROR_INVALID_NAME:
1763 case ERROR_PATH_NOT_FOUND:
1764 case ERROR_NO_MORE_FILES:
1765 errno = ENOENT;
1766 break;
1767
1768 case ERROR_NOT_ENOUGH_MEMORY:
1769 errno = ENOMEM;
1770 break;
1771
1772 default:
1773 errno = EINVAL;
1774 break;
1775 }
1776
1777 return;
1778 }
1779 }
1780#else
1781 #if HAVE_AT
1782 if (req->wd)
1783 {
1784 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1785
1786 if (fd < 0)
1787 return;
1788
1789 dirp = fdopendir (fd);
1790
1791 if (!dirp)
1792 close (fd);
1793 }
1794 else
1228 self->dirp = dirp = opendir (req->ptr1); 1795 dirp = opendir (req->ptr1);
1796 #else
1797 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1798 #endif
1799
1800 if (!dirp)
1801 return;
1802#endif
1803
1804 if (req->flags & EIO_FLAG_PTR1_FREE)
1805 free (req->ptr1);
1806
1229 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1807 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1230 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1808 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1231 req->ptr2 = names = malloc (namesalloc); 1809 req->ptr2 = names = malloc (namesalloc);
1232 X_UNLOCK (wrklock);
1233 1810
1234 if (dirp && names && (!flags || dents)) 1811 if (!names || (flags && !dents))
1812 return;
1813
1235 for (;;) 1814 for (;;)
1236 { 1815 {
1816 int done;
1817
1818#ifdef _WIN32
1819 done = !dirp;
1820#else
1237 errno = 0; 1821 errno = 0;
1238 entp = readdir (dirp); 1822 entp = readdir (dirp);
1823 done = !entp;
1824#endif
1239 1825
1240 if (!entp) 1826 if (done)
1241 { 1827 {
1828#ifndef _WIN32
1829 int old_errno = errno;
1830 closedir (dirp);
1831 errno = old_errno;
1832
1242 if (errno) 1833 if (errno)
1243 break; 1834 break;
1835#endif
1244 1836
1245 /* sort etc. */ 1837 /* sort etc. */
1246 req->int1 = flags; 1838 req->int1 = flags;
1247 req->result = dentoffs; 1839 req->result = dentoffs;
1248 1840
1249 if (flags & EIO_READDIR_STAT_ORDER) 1841 if (flags & EIO_READDIR_STAT_ORDER)
1250 eio_dent_sort (dents, dentoffs, 0, inode_bits); /* sort by inode exclusively */ 1842 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1251 else if (flags & EIO_READDIR_DIRS_FIRST) 1843 else if (flags & EIO_READDIR_DIRS_FIRST)
1252 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1844 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1253 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1845 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1254 else 1846 else
1847 {
1848 /* in this case, all is known, and we just put dirs first and sort them */
1849 eio_dirent *oth = dents + dentoffs;
1850 eio_dirent *dir = dents;
1851
1852 /* now partition dirs to the front, and non-dirs to the back */
1853 /* by walking from both sides and swapping if necessary */
1854 while (oth > dir)
1855 {
1856 if (dir->type == EIO_DT_DIR)
1857 ++dir;
1858 else if ((--oth)->type == EIO_DT_DIR)
1859 {
1860 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1861
1862 ++dir;
1863 }
1864 }
1865
1866 /* now sort the dirs only (dirs all have the same score) */
1867 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1868 }
1869
1870 break;
1871 }
1872
1873 /* now add the entry to our list(s) */
1874 name = D_NAME (entp);
1875
1876 /* skip . and .. entries */
1877 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1878 {
1879 int len = D_NAMLEN (entp) + 1;
1880
1881 while (ecb_expect_false (namesoffs + len > namesalloc))
1882 {
1883 namesalloc *= 2;
1884 req->ptr2 = names = realloc (names, namesalloc);
1885
1886 if (!names)
1887 break;
1888 }
1889
1890 memcpy (names + namesoffs, name, len);
1891
1892 if (dents)
1893 {
1894 struct eio_dirent *ent;
1895
1896 if (ecb_expect_false (dentoffs == dentalloc))
1255 { 1897 {
1256 /* in this case, all is known, and we just put dirs first and sort them */ 1898 dentalloc *= 2;
1899 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1900
1901 if (!dents)
1902 break;
1903 }
1904
1257 eio_dirent *oth = dents + dentoffs; 1905 ent = dents + dentoffs;
1258 eio_dirent *dir = dents;
1259 1906
1260 /* now partition dirs to the front, and non-dirs to the back */ 1907 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1261 /* by walking from both sides and swapping if necessary */ 1908 ent->namelen = len - 1;
1262 /* also clear score, so it doesn't influence sorting */ 1909 ent->inode = D_INO (entp);
1263 while (oth > dir) 1910
1911 inode_bits |= ent->inode;
1912
1913 switch (D_TYPE (entp))
1914 {
1915 default:
1916 ent->type = EIO_DT_UNKNOWN;
1917 flags |= EIO_READDIR_FOUND_UNKNOWN;
1918 break;
1919
1920 #ifdef DT_FIFO
1921 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1922 #endif
1923 #ifdef DT_CHR
1924 case DT_CHR: ent->type = EIO_DT_CHR; break;
1925 #endif
1926 #ifdef DT_MPC
1927 case DT_MPC: ent->type = EIO_DT_MPC; break;
1928 #endif
1929 #ifdef DT_DIR
1930 case DT_DIR: ent->type = EIO_DT_DIR; break;
1931 #endif
1932 #ifdef DT_NAM
1933 case DT_NAM: ent->type = EIO_DT_NAM; break;
1934 #endif
1935 #ifdef DT_BLK
1936 case DT_BLK: ent->type = EIO_DT_BLK; break;
1937 #endif
1938 #ifdef DT_MPB
1939 case DT_MPB: ent->type = EIO_DT_MPB; break;
1940 #endif
1941 #ifdef DT_REG
1942 case DT_REG: ent->type = EIO_DT_REG; break;
1943 #endif
1944 #ifdef DT_NWK
1945 case DT_NWK: ent->type = EIO_DT_NWK; break;
1946 #endif
1947 #ifdef DT_CMP
1948 case DT_CMP: ent->type = EIO_DT_CMP; break;
1949 #endif
1950 #ifdef DT_LNK
1951 case DT_LNK: ent->type = EIO_DT_LNK; break;
1952 #endif
1953 #ifdef DT_SOCK
1954 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1955 #endif
1956 #ifdef DT_DOOR
1957 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1958 #endif
1959 #ifdef DT_WHT
1960 case DT_WHT: ent->type = EIO_DT_WHT; break;
1961 #endif
1962 }
1963
1964 ent->score = 7;
1965
1966 if (flags & EIO_READDIR_DIRS_FIRST)
1967 {
1968 if (ent->type == EIO_DT_UNKNOWN)
1264 { 1969 {
1265 if (dir->type == EIO_DT_DIR) 1970 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1266 ++dir; 1971 ent->score = 1;
1267 else if ((--oth)->type == EIO_DT_DIR) 1972 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1268 { 1973 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1269 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1270
1271 ++dir;
1272 }
1273 } 1974 }
1274 1975 else if (ent->type == EIO_DT_DIR)
1275 /* now sort the dirs only */ 1976 ent->score = 0;
1276 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1277 } 1977 }
1278
1279 break;
1280 }
1281
1282 /* now add the entry to our list(s) */
1283 name = entp->d_name;
1284
1285 /* skip . and .. entries */
1286 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1287 {
1288 int len = D_NAMLEN (entp) + 1;
1289
1290 while (expect_false (namesoffs + len > namesalloc))
1291 {
1292 namesalloc *= 2;
1293 X_LOCK (wrklock);
1294 req->ptr2 = names = realloc (names, namesalloc);
1295 X_UNLOCK (wrklock);
1296
1297 if (!names)
1298 break;
1299 } 1978 }
1300 1979
1301 memcpy (names + namesoffs, name, len);
1302
1303 if (dents)
1304 {
1305 struct eio_dirent *ent;
1306
1307 if (expect_false (dentoffs == dentalloc))
1308 {
1309 dentalloc *= 2;
1310 X_LOCK (wrklock);
1311 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1312 X_UNLOCK (wrklock);
1313
1314 if (!dents)
1315 break;
1316 }
1317
1318 ent = dents + dentoffs;
1319
1320 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1321 ent->namelen = len - 1;
1322 ent->inode = D_INO (entp);
1323
1324 inode_bits |= ent->inode;
1325
1326 switch (D_TYPE (entp))
1327 {
1328 default:
1329 ent->type = EIO_DT_UNKNOWN;
1330 flags |= EIO_READDIR_FOUND_UNKNOWN;
1331 break;
1332
1333 #ifdef DT_FIFO
1334 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1335 #endif
1336 #ifdef DT_CHR
1337 case DT_CHR: ent->type = EIO_DT_CHR; break;
1338 #endif
1339 #ifdef DT_MPC
1340 case DT_MPC: ent->type = EIO_DT_MPC; break;
1341 #endif
1342 #ifdef DT_DIR
1343 case DT_DIR: ent->type = EIO_DT_DIR; break;
1344 #endif
1345 #ifdef DT_NAM
1346 case DT_NAM: ent->type = EIO_DT_NAM; break;
1347 #endif
1348 #ifdef DT_BLK
1349 case DT_BLK: ent->type = EIO_DT_BLK; break;
1350 #endif
1351 #ifdef DT_MPB
1352 case DT_MPB: ent->type = EIO_DT_MPB; break;
1353 #endif
1354 #ifdef DT_REG
1355 case DT_REG: ent->type = EIO_DT_REG; break;
1356 #endif
1357 #ifdef DT_NWK
1358 case DT_NWK: ent->type = EIO_DT_NWK; break;
1359 #endif
1360 #ifdef DT_CMP
1361 case DT_CMP: ent->type = EIO_DT_CMP; break;
1362 #endif
1363 #ifdef DT_LNK
1364 case DT_LNK: ent->type = EIO_DT_LNK; break;
1365 #endif
1366 #ifdef DT_SOCK
1367 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1368 #endif
1369 #ifdef DT_DOOR
1370 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1371 #endif
1372 #ifdef DT_WHT
1373 case DT_WHT: ent->type = EIO_DT_WHT; break;
1374 #endif
1375 }
1376
1377 ent->score = 7;
1378
1379 if (flags & EIO_READDIR_DIRS_FIRST)
1380 {
1381 if (ent->type == EIO_DT_UNKNOWN)
1382 {
1383 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1384 ent->score = 1;
1385 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1386 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1387 }
1388 else if (ent->type == EIO_DT_DIR)
1389 ent->score = 0;
1390 }
1391 }
1392
1393 namesoffs += len; 1980 namesoffs += len;
1394 ++dentoffs; 1981 ++dentoffs;
1395 } 1982 }
1396 1983
1397 if (EIO_CANCELLED (req)) 1984 if (EIO_CANCELLED (req))
1398 { 1985 {
1399 errno = ECANCELED; 1986 errno = ECANCELED;
1400 break; 1987 break;
1401 } 1988 }
1989
1990#ifdef _WIN32
1991 if (!FindNextFile (dirp, &entp))
1992 {
1993 FindClose (dirp);
1994 dirp = 0;
1402 } 1995 }
1996#endif
1997 }
1403} 1998}
1404 1999
1405#ifdef PAGESIZE 2000/*****************************************************************************/
1406# define eio_pagesize() PAGESIZE 2001/* working directory stuff */
1407#else 2002/* various deficiencies in the posix 2008 api force us to */
1408static intptr_t 2003/* keep the absolute path in string form at all times */
1409eio_pagesize (void) 2004/* fuck yeah. */
1410{
1411 static intptr_t page;
1412 2005
1413 if (!page) 2006#if !HAVE_AT
1414 page = sysconf (_SC_PAGESIZE);
1415 2007
2008/* a bit like realpath, but usually faster because it doesn'T have to return */
2009/* an absolute or canonical path */
2010static const char *
2011wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2012{
2013 if (!wd || *path == '/')
2014 return path;
2015
2016 if (path [0] == '.' && !path [1])
2017 return wd->str;
2018
2019 {
2020 int l1 = wd->len;
2021 int l2 = strlen (path);
2022
2023 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2024
2025 memcpy (res, wd->str, l1);
2026 res [l1] = '/';
2027 memcpy (res + l1 + 1, path, l2 + 1);
2028
2029 return res;
2030 }
2031}
2032
2033#endif
2034
2035static eio_wd
2036eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2037{
2038 int fd;
2039 eio_wd res;
2040 int len = eio__realpath (tmpbuf, wd, path);
2041
2042 if (len < 0)
2043 return EIO_INVALID_WD;
2044
2045#if HAVE_AT
2046 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2047
2048 if (fd < 0)
2049 return EIO_INVALID_WD;
2050#endif
2051
2052 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2053
2054#if HAVE_AT
2055 res->fd = fd;
2056#endif
2057
2058 res->len = len;
2059 memcpy (res->str, tmpbuf->ptr, len);
2060 res->str [len] = 0;
2061
1416 return page; 2062 return res;
1417} 2063}
1418#endif
1419 2064
1420static void 2065eio_wd
1421eio_page_align (void **addr, size_t *length) 2066eio_wd_open_sync (eio_wd wd, const char *path)
1422{ 2067{
1423 intptr_t mask = eio_pagesize () - 1; 2068 struct tmpbuf tmpbuf = { 0 };
2069 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2070 free (tmpbuf.ptr);
1424 2071
1425 /* round down addr */ 2072 return wd;
1426 intptr_t adj = mask & (intptr_t)*addr;
1427
1428 *addr = (void *)((intptr_t)*addr - adj);
1429 *length += adj;
1430
1431 /* round up length */
1432 *length = (*length + mask) & ~mask;
1433} 2073}
1434 2074
1435#if !_POSIX_MEMLOCK 2075void
1436# define eio__mlockall(a) ((errno = ENOSYS), -1) 2076eio_wd_close_sync (eio_wd wd)
1437#else 2077{
2078 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2079 {
2080 #if HAVE_AT
2081 close (wd->fd);
2082 #endif
2083 free (wd);
2084 }
2085}
2086
2087#if HAVE_AT
2088
2089/* they forgot these */
1438 2090
1439static int 2091static int
1440eio__mlockall (int flags) 2092eio__truncateat (int dirfd, const char *path, off_t length)
1441{ 2093{
1442 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 2094 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1443 extern int mallopt (int, int); 2095 int res;
1444 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1445 #endif
1446 2096
1447 if (EIO_MCL_CURRENT != MCL_CURRENT 2097 if (fd < 0)
1448 || EIO_MCL_FUTURE != MCL_FUTURE) 2098 return fd;
1449 {
1450 flags = 0
1451 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1452 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1453 }
1454 2099
1455 return mlockall (flags); 2100 res = ftruncate (fd, length);
2101 close (fd);
2102 return res;
1456} 2103}
1457#endif
1458
1459#if !_POSIX_MEMLOCK_RANGE
1460# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1461#else
1462 2104
1463static int 2105static int
1464eio__mlock (void *addr, size_t length) 2106eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1465{ 2107{
1466 eio_page_align (&addr, &length); 2108 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2109 int res;
1467 2110
1468 return mlock (addr, length); 2111 if (fd < 0)
1469} 2112 return fd;
1470 2113
1471#endif 2114 res = fstatvfs (fd, buf);
1472 2115 close (fd);
1473#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1474# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1475#else
1476
1477int
1478eio__msync (void *mem, size_t len, int flags)
1479{
1480 eio_page_align (&mem, &len);
1481
1482 if (EIO_MS_ASYNC != MS_SYNC
1483 || EIO_MS_INVALIDATE != MS_INVALIDATE
1484 || EIO_MS_SYNC != MS_SYNC)
1485 {
1486 flags = 0
1487 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1488 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1489 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1490 }
1491
1492 return msync (mem, len, flags);
1493}
1494
1495#endif
1496
1497int
1498eio__mtouch (void *mem, size_t len, int flags)
1499{
1500 eio_page_align (&mem, &len);
1501
1502 {
1503 intptr_t addr = (intptr_t)mem;
1504 intptr_t end = addr + len;
1505 intptr_t page = eio_pagesize ();
1506
1507 if (addr < end)
1508 if (flags & EIO_MT_MODIFY) /* modify */
1509 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1510 else
1511 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1512 }
1513
1514 return 0; 2116 return res;
2117
1515} 2118}
2119
2120#endif
1516 2121
1517/*****************************************************************************/ 2122/*****************************************************************************/
1518 2123
1519#define ALLOC(len) \ 2124#define ALLOC(len) \
1520 if (!req->ptr2) \ 2125 if (!req->ptr2) \
1529 req->result = -1; \ 2134 req->result = -1; \
1530 break; \ 2135 break; \
1531 } \ 2136 } \
1532 } 2137 }
1533 2138
2139static void ecb_noinline ecb_cold
2140etp_proc_init (void)
2141{
2142#if HAVE_PRCTL_SET_NAME
2143 /* provide a more sensible "thread name" */
2144 char name[16 + 1];
2145 const int namelen = sizeof (name) - 1;
2146 int len;
2147
2148 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2149 name [namelen] = 0;
2150 len = strlen (name);
2151 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2152 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2153#endif
2154}
2155
1534X_THREAD_PROC (etp_proc) 2156X_THREAD_PROC (etp_proc)
1535{ 2157{
1536 ETP_REQ *req; 2158 ETP_REQ *req;
1537 struct timespec ts; 2159 struct timespec ts;
1538 etp_worker *self = (etp_worker *)thr_arg; 2160 etp_worker *self = (etp_worker *)thr_arg;
1539 2161
2162 etp_proc_init ();
2163
1540 /* try to distribute timeouts somewhat randomly */ 2164 /* try to distribute timeouts somewhat evenly */
1541 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2165 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1542 2166
1543 for (;;) 2167 for (;;)
1544 { 2168 {
2169 ts.tv_sec = 0;
2170
1545 X_LOCK (reqlock); 2171 X_LOCK (reqlock);
1546 2172
1547 for (;;) 2173 for (;;)
1548 { 2174 {
1549 self->req = req = reqq_shift (&req_queue); 2175 req = reqq_shift (&req_queue);
1550 2176
1551 if (req) 2177 if (req)
1552 break; 2178 break;
1553 2179
2180 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2181 {
2182 X_UNLOCK (reqlock);
2183 X_LOCK (wrklock);
2184 --started;
2185 X_UNLOCK (wrklock);
2186 goto quit;
2187 }
2188
1554 ++idle; 2189 ++idle;
1555 2190
1556 ts.tv_sec = time (0) + IDLE_TIMEOUT; 2191 if (idle <= max_idle)
1557 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2192 /* we are allowed to idle, so do so without any timeout */
2193 X_COND_WAIT (reqwait, reqlock);
2194 else
1558 { 2195 {
1559 if (idle > max_idle) 2196 /* initialise timeout once */
1560 { 2197 if (!ts.tv_sec)
1561 --idle; 2198 ts.tv_sec = time (0) + idle_timeout;
1562 X_UNLOCK (reqlock);
1563 X_LOCK (wrklock);
1564 --started;
1565 X_UNLOCK (wrklock);
1566 goto quit;
1567 }
1568 2199
1569 /* we are allowed to idle, so do so without any timeout */
1570 X_COND_WAIT (reqwait, reqlock); 2200 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2201 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1571 } 2202 }
1572 2203
1573 --idle; 2204 --idle;
1574 } 2205 }
1575 2206
1578 X_UNLOCK (reqlock); 2209 X_UNLOCK (reqlock);
1579 2210
1580 if (req->type < 0) 2211 if (req->type < 0)
1581 goto quit; 2212 goto quit;
1582 2213
1583 if (!EIO_CANCELLED (req))
1584 ETP_EXECUTE (self, req); 2214 ETP_EXECUTE (self, req);
1585 2215
1586 X_LOCK (reslock); 2216 X_LOCK (reslock);
1587 2217
1588 ++npending; 2218 ++npending;
1589 2219
1590 if (!reqq_push (&res_queue, req) && want_poll_cb) 2220 if (!reqq_push (&res_queue, req) && want_poll_cb)
1591 want_poll_cb (); 2221 want_poll_cb ();
1592 2222
1593 self->req = 0;
1594 etp_worker_clear (self); 2223 etp_worker_clear (self);
1595 2224
1596 X_UNLOCK (reslock); 2225 X_UNLOCK (reslock);
1597 } 2226 }
1598 2227
1599quit: 2228quit:
2229 free (req);
2230
1600 X_LOCK (wrklock); 2231 X_LOCK (wrklock);
1601 etp_worker_free (self); 2232 etp_worker_free (self);
1602 X_UNLOCK (wrklock); 2233 X_UNLOCK (wrklock);
1603 2234
1604 return 0; 2235 return 0;
1605} 2236}
1606 2237
1607/*****************************************************************************/ 2238/*****************************************************************************/
1608 2239
2240int ecb_cold
1609int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2241eio_init (void (*want_poll)(void), void (*done_poll)(void))
1610{ 2242{
2243#if !HAVE_PREADWRITE
2244 X_MUTEX_CREATE (preadwritelock);
2245#endif
2246
1611 return etp_init (want_poll, done_poll); 2247 return etp_init (want_poll, done_poll);
1612} 2248}
1613 2249
2250ecb_inline void
1614static void eio_api_destroy (eio_req *req) 2251eio_api_destroy (eio_req *req)
1615{ 2252{
1616 free (req); 2253 free (req);
1617} 2254}
1618 2255
1619#define REQ(rtype) \ 2256#define REQ(rtype) \
1638 { \ 2275 { \
1639 eio_api_destroy (req); \ 2276 eio_api_destroy (req); \
1640 return 0; \ 2277 return 0; \
1641 } 2278 }
1642 2279
2280static void
1643static void eio_execute (etp_worker *self, eio_req *req) 2281eio_execute (etp_worker *self, eio_req *req)
1644{ 2282{
2283#if HAVE_AT
2284 int dirfd;
2285#else
2286 const char *path;
2287#endif
2288
2289 if (ecb_expect_false (EIO_CANCELLED (req)))
2290 {
2291 req->result = -1;
2292 req->errorno = ECANCELED;
2293 return;
2294 }
2295
2296 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2297 {
2298 req->result = -1;
2299 req->errorno = ENOENT;
2300 return;
2301 }
2302
2303 if (req->type >= EIO_OPEN)
2304 {
2305 #if HAVE_AT
2306 dirfd = WD2FD (req->wd);
2307 #else
2308 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2309 #endif
2310 }
2311
1645 switch (req->type) 2312 switch (req->type)
1646 { 2313 {
2314 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2315 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2316 break;
2317 case EIO_WD_CLOSE: req->result = 0;
2318 eio_wd_close_sync (req->wd); break;
2319
1647 case EIO_READ: ALLOC (req->size); 2320 case EIO_READ: ALLOC (req->size);
1648 req->result = req->offs >= 0 2321 req->result = req->offs >= 0
1649 ? pread (req->int1, req->ptr2, req->size, req->offs) 2322 ? pread (req->int1, req->ptr2, req->size, req->offs)
1650 : read (req->int1, req->ptr2, req->size); break; 2323 : read (req->int1, req->ptr2, req->size); break;
1651 case EIO_WRITE: req->result = req->offs >= 0 2324 case EIO_WRITE: req->result = req->offs >= 0
1652 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2325 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1653 : write (req->int1, req->ptr2, req->size); break; 2326 : write (req->int1, req->ptr2, req->size); break;
1654 2327
1655 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2328 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1656 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2329 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2330
2331#if HAVE_AT
1657 2332
1658 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2333 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1659 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2334 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1660 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2335 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2336 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2337 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2338 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2339 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2340 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2341
2342 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2343 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2344 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2345 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2346 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2347 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2348 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2349 case EIO_READLINK: ALLOC (PATH_MAX);
2350 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2351 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2352 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2353 case EIO_UTIME:
2354 case EIO_FUTIME:
2355 {
2356 struct timespec ts[2];
2357 struct timespec *times;
2358
2359 if (req->nv1 != -1. || req->nv2 != -1.)
2360 {
2361 ts[0].tv_sec = req->nv1;
2362 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2363 ts[1].tv_sec = req->nv2;
2364 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2365
2366 times = ts;
2367 }
2368 else
2369 times = 0;
2370
2371 req->result = req->type == EIO_FUTIME
2372 ? futimens (req->int1, times)
2373 : utimensat (dirfd, req->ptr1, times, 0);
2374 }
2375 break;
2376
2377#else
2378
2379 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2380 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2381 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1661 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2382 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2383 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2384 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2385 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2386 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2387
2388 case EIO_UNLINK: req->result = unlink (path ); break;
2389 case EIO_RMDIR: req->result = rmdir (path ); break;
2390 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2391 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2392 case EIO_LINK: req->result = link (path , req->ptr2); break;
2393 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2394 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2395 case EIO_READLINK: ALLOC (PATH_MAX);
2396 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2397 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2398 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2399
2400 case EIO_UTIME:
2401 case EIO_FUTIME:
2402 {
2403 struct timeval tv[2];
2404 struct timeval *times;
2405
2406 if (req->nv1 != -1. || req->nv2 != -1.)
2407 {
2408 tv[0].tv_sec = req->nv1;
2409 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2410 tv[1].tv_sec = req->nv2;
2411 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2412
2413 times = tv;
2414 }
2415 else
2416 times = 0;
2417
2418 req->result = req->type == EIO_FUTIME
2419 ? futimes (req->int1, times)
2420 : utimes (req->ptr1, times);
2421 }
2422 break;
2423
2424#endif
2425
2426 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2427 {
2428 ALLOC (req->result);
2429 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2430 }
2431 break;
2432
1662 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2433 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1663 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2434 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1664 2435
1665 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1666 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1667 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2436 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1668 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2437 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1669 2438
1670 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1671 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2439 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1672 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1673 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2440 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1674 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1675 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2441 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1676 2442
1677 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1678 case EIO_CLOSE: req->result = close (req->int1); break; 2443 case EIO_CLOSE: req->result = close (req->int1); break;
1679 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2444 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1680 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1681 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1682 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1683 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1684 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1685 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1686 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1687
1688 case EIO_READLINK: ALLOC (PATH_MAX);
1689 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1690
1691 case EIO_SYNC: req->result = 0; sync (); break; 2445 case EIO_SYNC: req->result = 0; sync (); break;
1692 case EIO_FSYNC: req->result = fsync (req->int1); break; 2446 case EIO_FSYNC: req->result = fsync (req->int1); break;
1693 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2447 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2448 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2449 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1694 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2450 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1695 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2451 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1696 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2452 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1697 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2453 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1698 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2454 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1699 2455
1700 case EIO_READDIR: eio__scandir (req, self); break; 2456 case EIO_READDIR: eio__scandir (req, self); break;
1701 2457
1702 case EIO_BUSY: 2458 case EIO_BUSY:
1703#ifdef _WIN32 2459#ifdef _WIN32
1712 req->result = select (0, 0, 0, 0, &tv); 2468 req->result = select (0, 0, 0, 0, &tv);
1713 } 2469 }
1714#endif 2470#endif
1715 break; 2471 break;
1716 2472
1717 case EIO_UTIME:
1718 case EIO_FUTIME:
1719 {
1720 struct timeval tv[2];
1721 struct timeval *times;
1722
1723 if (req->nv1 != -1. || req->nv2 != -1.)
1724 {
1725 tv[0].tv_sec = req->nv1;
1726 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1727 tv[1].tv_sec = req->nv2;
1728 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1729
1730 times = tv;
1731 }
1732 else
1733 times = 0;
1734
1735 req->result = req->type == EIO_FUTIME
1736 ? futimes (req->int1, times)
1737 : utimes (req->ptr1, times);
1738 }
1739 break;
1740
1741 case EIO_GROUP: 2473 case EIO_GROUP:
1742 abort (); /* handled in eio_request */ 2474 abort (); /* handled in eio_request */
1743 2475
1744 case EIO_NOP: 2476 case EIO_NOP:
1745 req->result = 0; 2477 req->result = 0;
1746 break; 2478 break;
1747 2479
1748 case EIO_CUSTOM: 2480 case EIO_CUSTOM:
1749 ((void (*)(eio_req *))req->feed) (req); 2481 req->feed (req);
1750 break; 2482 break;
1751 2483
1752 default: 2484 default:
1753 errno = ENOSYS; 2485 errno = ENOSYS;
1754 req->result = -1; 2486 req->result = -1;
1758 req->errorno = errno; 2490 req->errorno = errno;
1759} 2491}
1760 2492
1761#ifndef EIO_NO_WRAPPERS 2493#ifndef EIO_NO_WRAPPERS
1762 2494
2495eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2496{
2497 REQ (EIO_WD_OPEN); PATH; SEND;
2498}
2499
2500eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2501{
2502 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2503}
2504
1763eio_req *eio_nop (int pri, eio_cb cb, void *data) 2505eio_req *eio_nop (int pri, eio_cb cb, void *data)
1764{ 2506{
1765 REQ (EIO_NOP); SEND; 2507 REQ (EIO_NOP); SEND;
1766} 2508}
1767 2509
1783eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2525eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1784{ 2526{
1785 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2527 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1786} 2528}
1787 2529
2530eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2531{
2532 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2533}
2534
2535eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2536{
2537 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2538}
2539
2540eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2541{
2542 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2543}
2544
1788eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2545eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1789{ 2546{
1790 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2547 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1791} 2548}
1792 2549
1798eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2555eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1799{ 2556{
1800 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2557 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1801} 2558}
1802 2559
1803eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2560eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1804{ 2561{
1805 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2562 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1806}
1807
1808eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1809{
1810 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1811} 2563}
1812 2564
1813eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2565eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1814{ 2566{
1815 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2567 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1853eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2605eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1854{ 2606{
1855 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2607 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1856} 2608}
1857 2609
1858eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2610eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1859{ 2611{
1860 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2612 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1861} 2613}
1862 2614
1863eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2615eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1883eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2635eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1884{ 2636{
1885 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2637 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1886} 2638}
1887 2639
1888eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2640eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1889{ 2641{
1890 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2642 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891} 2643}
1892 2644
1893eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2645eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1907} 2659}
1908 2660
1909eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2661eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1910{ 2662{
1911 return eio__1path (EIO_READLINK, path, pri, cb, data); 2663 return eio__1path (EIO_READLINK, path, pri, cb, data);
2664}
2665
2666eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2667{
2668 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1912} 2669}
1913 2670
1914eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2671eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1915{ 2672{
1916 return eio__1path (EIO_STAT, path, pri, cb, data); 2673 return eio__1path (EIO_STAT, path, pri, cb, data);
1975eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2732eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1976{ 2733{
1977 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2734 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1978} 2735}
1979 2736
1980eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2737eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
1981{ 2738{
1982 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2739 REQ (EIO_CUSTOM); req->feed = execute; SEND;
1983} 2740}
1984 2741
1985#endif 2742#endif
1986 2743
1987eio_req *eio_grp (eio_cb cb, void *data) 2744eio_req *eio_grp (eio_cb cb, void *data)
1996#undef SEND 2753#undef SEND
1997 2754
1998/*****************************************************************************/ 2755/*****************************************************************************/
1999/* grp functions */ 2756/* grp functions */
2000 2757
2758void
2001void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2759eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2002{ 2760{
2003 grp->int2 = limit; 2761 grp->int2 = limit;
2004 grp->feed = feed; 2762 grp->feed = feed;
2005 2763
2006 grp_try_feed (grp); 2764 grp_try_feed (grp);
2007} 2765}
2008 2766
2767void
2009void eio_grp_limit (eio_req *grp, int limit) 2768eio_grp_limit (eio_req *grp, int limit)
2010{ 2769{
2011 grp->int2 = limit; 2770 grp->int2 = limit;
2012 2771
2013 grp_try_feed (grp); 2772 grp_try_feed (grp);
2014} 2773}
2015 2774
2775void
2016void eio_grp_add (eio_req *grp, eio_req *req) 2776eio_grp_add (eio_req *grp, eio_req *req)
2017{ 2777{
2018 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2778 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2019 2779
2020 grp->flags |= EIO_FLAG_GROUPADD; 2780 grp->flags |= EIO_FLAG_GROUPADD;
2021 2781
2032} 2792}
2033 2793
2034/*****************************************************************************/ 2794/*****************************************************************************/
2035/* misc garbage */ 2795/* misc garbage */
2036 2796
2797eio_ssize_t
2037ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2798eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2038{ 2799{
2039 etp_worker wrk;
2040 ssize_t ret;
2041
2042 wrk.dbuf = 0;
2043
2044 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2800 return eio__sendfile (ofd, ifd, offset, count);
2045
2046 if (wrk.dbuf)
2047 free (wrk.dbuf);
2048
2049 return ret;
2050} 2801}
2051 2802

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines