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 1.66 by root, Sun Jun 5 19:58:37 2011 UTC vs.
Revision 1.115 by root, Tue Mar 27 18:54:45 2012 UTC

40#ifndef _WIN32 40#ifndef _WIN32
41# include "config.h" 41# include "config.h"
42#endif 42#endif
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h"
45 46
46#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
47# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
48#endif 49#endif
49#include "xthread.h" 50#include "xthread.h"
50 51
51#include <errno.h> 52#include <errno.h>
52#include <stddef.h> 53#include <stddef.h>
53#include <stdlib.h> 54#include <stdlib.h>
54#include <string.h> 55#include <string.h>
55#include <errno.h> 56#include <errno.h>
56#include <sys/types.h> 57#include <sys/types.h>
57#include <sys/stat.h> 58#include <sys/stat.h>
58#include <sys/statvfs.h>
59#include <limits.h> 59#include <limits.h>
60#include <fcntl.h> 60#include <fcntl.h>
61#include <assert.h> 61#include <assert.h>
62 62
63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
64/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
65#if HAVE_STDINT_H 65#if HAVE_STDINT_H
66# include <stdint.h> 66# include <stdint.h>
67#endif 67#endif
68 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
69#ifndef EIO_FINISH 82#ifndef EIO_FINISH
70# 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
71#endif 84#endif
72 85
73#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
76 89
77#ifndef EIO_FEED 90#ifndef EIO_FEED
78# 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)
79#endif 92#endif
80 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
81#ifdef _WIN32 105#ifdef _WIN32
82 106
83 /*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
84#else 197#else
85 198
86# include <sys/time.h> 199 #include <sys/time.h>
87# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
88# 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
89# include <utime.h> 237# include <utime.h>
90# include <signal.h>
91# include <dirent.h>
92
93#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
94# include <sys/mman.h>
95#endif
96
97/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
98# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
99# define _DIRENT_HAVE_D_TYPE /* sigh */
100# define D_INO(de) (de)->d_fileno
101# define D_NAMLEN(de) (de)->d_namlen
102# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
103# define D_INO(de) (de)->d_ino
104# endif 238#endif
105 239
106#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
107# undef D_NAMLEN 241# include <sys/syscall.h>
108# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
109#endif
110
111# ifdef _DIRENT_HAVE_D_TYPE
112# define D_TYPE(de) (de)->d_type
113# endif 242#endif
114 243
115# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
116# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
117# endif
118
119#endif 246#endif
120 247
121#if HAVE_SENDFILE 248#if HAVE_SENDFILE
122# if __linux 249# if __linux
123# include <sys/sendfile.h> 250# include <sys/sendfile.h>
138#endif 265#endif
139#ifndef D_INO 266#ifndef D_INO
140# define D_INO(de) 0 267# define D_INO(de) 0
141#endif 268#endif
142#ifndef D_NAMLEN 269#ifndef D_NAMLEN
143# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
144#endif 271#endif
145 272
146/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
147#ifndef NAME_MAX 274#ifndef NAME_MAX
148# define NAME_MAX 4096 275# define NAME_MAX 4096
155 282
156/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
157#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
158 285
159#define dBUF \ 286#define dBUF \
160 char *eio_buf; \
161 ETP_WORKER_LOCK (self); \
162 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
163 ETP_WORKER_UNLOCK (self); \
164 errno = ENOMEM; \ 288 errno = ENOMEM; \
165 if (!eio_buf) \ 289 if (!eio_buf) \
166 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
167 294
168#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
169 296
170/*****************************************************************************/ 297/*****************************************************************************/
171 298
172#if __GNUC__ >= 3 299struct tmpbuf
173# 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
174#else 325#else
175# 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);
176#endif 328#endif
177 329
178#define expect_false(expr) expect ((expr) != 0, 0) 330struct eio_pwd
179#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};
180 338
181/*****************************************************************************/ 339/*****************************************************************************/
182 340
183#define ETP_PRI_MIN EIO_PRI_MIN 341#define ETP_PRI_MIN EIO_PRI_MIN
184#define ETP_PRI_MAX EIO_PRI_MAX 342#define ETP_PRI_MAX EIO_PRI_MAX
190static int eio_finish (eio_req *req); 348static int eio_finish (eio_req *req);
191#define ETP_FINISH(req) eio_finish (req) 349#define ETP_FINISH(req) eio_finish (req)
192static void eio_execute (struct etp_worker *self, eio_req *req); 350static void eio_execute (struct etp_worker *self, eio_req *req);
193#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
194 352
195#define ETP_WORKER_CLEAR(req) \
196 if (wrk->dbuf) \
197 { \
198 free (wrk->dbuf); \
199 wrk->dbuf = 0; \
200 } \
201 \
202 if (wrk->dirp) \
203 { \
204 closedir (wrk->dirp); \
205 wrk->dirp = 0; \
206 }
207
208#define ETP_WORKER_COMMON \
209 void *dbuf; \
210 DIR *dirp;
211
212/*****************************************************************************/ 353/*****************************************************************************/
213 354
214#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
215 356
216/* calculate time difference in ~1/EIO_TICKS of a second */ 357/* calculate time difference in ~1/EIO_TICKS of a second */
358ecb_inline int
217static int tvdiff (struct timeval *tv1, struct timeval *tv2) 359tvdiff (struct timeval *tv1, struct timeval *tv2)
218{ 360{
219 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 361 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
220 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 362 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
221} 363}
222 364
226static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
227 369
228static unsigned int max_poll_time; /* reslock */ 370static unsigned int max_poll_time; /* reslock */
229static unsigned int max_poll_reqs; /* reslock */ 371static unsigned int max_poll_reqs; /* reslock */
230 372
231static volatile unsigned int nreqs; /* reqlock */ 373static unsigned int nreqs; /* reqlock */
232static volatile unsigned int nready; /* reqlock */ 374static unsigned int nready; /* reqlock */
233static volatile unsigned int npending; /* reqlock */ 375static unsigned int npending; /* reqlock */
234static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 376static unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
235static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */ 377static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
236 378
237static xmutex_t wrklock; 379static xmutex_t wrklock;
238static xmutex_t reslock; 380static xmutex_t reslock;
239static xmutex_t reqlock; 381static xmutex_t reqlock;
240static xcond_t reqwait; 382static xcond_t reqwait;
243/* 385/*
244 * make our pread/pwrite emulation safe against themselves, but not against 386 * make our pread/pwrite emulation safe against themselves, but not against
245 * 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,
246 * but that's your problem, not mine. 388 * but that's your problem, not mine.
247 */ 389 */
248static xmutex_t preadwritelock = X_MUTEX_INIT; 390static xmutex_t preadwritelock;
249#endif 391#endif
250 392
251typedef struct etp_worker 393typedef struct etp_worker
252{ 394{
395 struct tmpbuf tmpbuf;
396
253 /* locked by wrklock */ 397 /* locked by wrklock */
254 struct etp_worker *prev, *next; 398 struct etp_worker *prev, *next;
255 399
256 xthread_t tid; 400 xthread_t tid;
257 401
258 /* locked by reslock, reqlock or wrklock */ 402#ifdef ETP_WORKER_COMMON
259 ETP_REQ *req; /* currently processed request */
260
261 ETP_WORKER_COMMON 403 ETP_WORKER_COMMON
404#endif
262} etp_worker; 405} etp_worker;
263 406
264static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 407static etp_worker wrk_first; /* NOT etp */
265 408
266#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
267#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
268 411
269/* worker threads management */ 412/* worker threads management */
270 413
414static void
271static void etp_worker_clear (etp_worker *wrk) 415etp_worker_clear (etp_worker *wrk)
272{ 416{
273 ETP_WORKER_CLEAR (wrk);
274} 417}
275 418
419static void ecb_cold
276static void etp_worker_free (etp_worker *wrk) 420etp_worker_free (etp_worker *wrk)
277{ 421{
422 free (wrk->tmpbuf.ptr);
423
278 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
279 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
280 426
281 free (wrk); 427 free (wrk);
282} 428}
283 429
284static unsigned int etp_nreqs (void) 430static unsigned int
431etp_nreqs (void)
285{ 432{
286 int retval; 433 int retval;
287 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 434 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
288 retval = nreqs; 435 retval = nreqs;
289 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 436 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
290 return retval; 437 return retval;
291} 438}
292 439
293static unsigned int etp_nready (void) 440static unsigned int
441etp_nready (void)
294{ 442{
295 unsigned int retval; 443 unsigned int retval;
296 444
297 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 445 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
298 retval = nready; 446 retval = nready;
299 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 447 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
300 448
301 return retval; 449 return retval;
302} 450}
303 451
304static unsigned int etp_npending (void) 452static unsigned int
453etp_npending (void)
305{ 454{
306 unsigned int retval; 455 unsigned int retval;
307 456
308 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 457 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
309 retval = npending; 458 retval = npending;
310 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 459 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
311 460
312 return retval; 461 return retval;
313} 462}
314 463
315static unsigned int etp_nthreads (void) 464static unsigned int
465etp_nthreads (void)
316{ 466{
317 unsigned int retval; 467 unsigned int retval;
318 468
319 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 469 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
320 retval = started; 470 retval = started;
334} etp_reqq; 484} etp_reqq;
335 485
336static etp_reqq req_queue; 486static etp_reqq req_queue;
337static etp_reqq res_queue; 487static etp_reqq res_queue;
338 488
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
339static int reqq_push (etp_reqq *q, ETP_REQ *req) 501reqq_push (etp_reqq *q, ETP_REQ *req)
340{ 502{
341 int pri = req->pri; 503 int pri = req->pri;
342 req->next = 0; 504 req->next = 0;
343 505
344 if (q->qe[pri]) 506 if (q->qe[pri])
350 q->qe[pri] = q->qs[pri] = req; 512 q->qe[pri] = q->qs[pri] = req;
351 513
352 return q->size++; 514 return q->size++;
353} 515}
354 516
517static ETP_REQ * ecb_noinline
355static ETP_REQ *reqq_shift (etp_reqq *q) 518reqq_shift (etp_reqq *q)
356{ 519{
357 int pri; 520 int pri;
358 521
359 if (!q->size) 522 if (!q->size)
360 return 0; 523 return 0;
375 } 538 }
376 539
377 abort (); 540 abort ();
378} 541}
379 542
380static void etp_thread_init (void) 543static int ecb_cold
544etp_init (void (*want_poll)(void), void (*done_poll)(void))
381{ 545{
382 X_MUTEX_CREATE (wrklock); 546 X_MUTEX_CREATE (wrklock);
383 X_MUTEX_CREATE (reslock); 547 X_MUTEX_CREATE (reslock);
384 X_MUTEX_CREATE (reqlock); 548 X_MUTEX_CREATE (reqlock);
385 X_COND_CREATE (reqwait); 549 X_COND_CREATE (reqwait);
386}
387 550
388static void etp_atfork_prepare (void) 551 reqq_init (&req_queue);
389{ 552 reqq_init (&res_queue);
390 X_LOCK (wrklock);
391 X_LOCK (reqlock);
392 X_LOCK (reslock);
393#if !HAVE_PREADWRITE
394 X_LOCK (preadwritelock);
395#endif
396}
397 553
398static void etp_atfork_parent (void) 554 wrk_first.next =
399{ 555 wrk_first.prev = &wrk_first;
400#if !HAVE_PREADWRITE
401 X_UNLOCK (preadwritelock);
402#endif
403 X_UNLOCK (reslock);
404 X_UNLOCK (reqlock);
405 X_UNLOCK (wrklock);
406}
407
408static void etp_atfork_child (void)
409{
410 ETP_REQ *prv;
411
412 while ((prv = reqq_shift (&req_queue)))
413 ETP_DESTROY (prv);
414
415 while ((prv = reqq_shift (&res_queue)))
416 ETP_DESTROY (prv);
417
418 while (wrk_first.next != &wrk_first)
419 {
420 etp_worker *wrk = wrk_first.next;
421
422 if (wrk->req)
423 ETP_DESTROY (wrk->req);
424
425 etp_worker_clear (wrk);
426 etp_worker_free (wrk);
427 }
428 556
429 started = 0; 557 started = 0;
430 idle = 0; 558 idle = 0;
431 nreqs = 0; 559 nreqs = 0;
432 nready = 0; 560 nready = 0;
433 npending = 0; 561 npending = 0;
434 562
435 etp_thread_init ();
436}
437
438static void
439etp_once_init (void)
440{
441 etp_thread_init ();
442 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
443}
444
445static int
446etp_init (void (*want_poll)(void), void (*done_poll)(void))
447{
448 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
449
450 pthread_once (&doinit, etp_once_init);
451
452 want_poll_cb = want_poll; 563 want_poll_cb = want_poll;
453 done_poll_cb = done_poll; 564 done_poll_cb = done_poll;
454 565
455 return 0; 566 return 0;
456} 567}
457 568
458X_THREAD_PROC (etp_proc); 569X_THREAD_PROC (etp_proc);
459 570
571static void ecb_cold
460static void etp_start_thread (void) 572etp_start_thread (void)
461{ 573{
462 etp_worker *wrk = calloc (1, sizeof (etp_worker)); 574 etp_worker *wrk = calloc (1, sizeof (etp_worker));
463 575
464 /*TODO*/ 576 /*TODO*/
465 assert (("unable to allocate worker thread data", wrk)); 577 assert (("unable to allocate worker thread data", wrk));
478 free (wrk); 590 free (wrk);
479 591
480 X_UNLOCK (wrklock); 592 X_UNLOCK (wrklock);
481} 593}
482 594
595static void
483static void etp_maybe_start_thread (void) 596etp_maybe_start_thread (void)
484{ 597{
485 if (expect_true (etp_nthreads () >= wanted)) 598 if (ecb_expect_true (etp_nthreads () >= wanted))
486 return; 599 return;
487 600
488 /* todo: maybe use idle here, but might be less exact */ 601 /* todo: maybe use idle here, but might be less exact */
489 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 ()))
490 return; 603 return;
491 604
492 etp_start_thread (); 605 etp_start_thread ();
493} 606}
494 607
608static void ecb_cold
495static void etp_end_thread (void) 609etp_end_thread (void)
496{ 610{
497 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
498 612
499 req->type = -1; 613 req->type = -1;
500 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
501 615
502 X_LOCK (reqlock); 616 X_LOCK (reqlock);
507 X_LOCK (wrklock); 621 X_LOCK (wrklock);
508 --started; 622 --started;
509 X_UNLOCK (wrklock); 623 X_UNLOCK (wrklock);
510} 624}
511 625
512static int etp_poll (void) 626static int
627etp_poll (void)
513{ 628{
514 unsigned int maxreqs; 629 unsigned int maxreqs;
515 unsigned int maxtime; 630 unsigned int maxtime;
516 struct timeval tv_start, tv_now; 631 struct timeval tv_start, tv_now;
517 632
547 662
548 X_LOCK (reqlock); 663 X_LOCK (reqlock);
549 --nreqs; 664 --nreqs;
550 X_UNLOCK (reqlock); 665 X_UNLOCK (reqlock);
551 666
552 if (expect_false (req->type == EIO_GROUP && req->size)) 667 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
553 { 668 {
554 req->int1 = 1; /* mark request as delayed */ 669 req->int1 = 1; /* mark request as delayed */
555 continue; 670 continue;
556 } 671 }
557 else 672 else
558 { 673 {
559 int res = ETP_FINISH (req); 674 int res = ETP_FINISH (req);
560 if (expect_false (res)) 675 if (ecb_expect_false (res))
561 return res; 676 return res;
562 } 677 }
563 678
564 if (expect_false (maxreqs && !--maxreqs)) 679 if (ecb_expect_false (maxreqs && !--maxreqs))
565 break; 680 break;
566 681
567 if (maxtime) 682 if (maxtime)
568 { 683 {
569 gettimeofday (&tv_now, 0); 684 gettimeofday (&tv_now, 0);
575 690
576 errno = EAGAIN; 691 errno = EAGAIN;
577 return -1; 692 return -1;
578} 693}
579 694
695static void
580static void etp_cancel (ETP_REQ *req) 696etp_cancel (ETP_REQ *req)
581{ 697{
582 X_LOCK (wrklock); 698 req->cancelled = 1;
583 req->flags |= EIO_FLAG_CANCELLED;
584 X_UNLOCK (wrklock);
585 699
586 eio_grp_cancel (req); 700 eio_grp_cancel (req);
587} 701}
588 702
703static void
589static void etp_submit (ETP_REQ *req) 704etp_submit (ETP_REQ *req)
590{ 705{
591 req->pri -= ETP_PRI_MIN; 706 req->pri -= ETP_PRI_MIN;
592 707
593 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;
594 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;
595 710
596 if (expect_false (req->type == EIO_GROUP)) 711 if (ecb_expect_false (req->type == EIO_GROUP))
597 { 712 {
598 /* I hope this is worth it :/ */ 713 /* I hope this is worth it :/ */
599 X_LOCK (reqlock); 714 X_LOCK (reqlock);
600 ++nreqs; 715 ++nreqs;
601 X_UNLOCK (reqlock); 716 X_UNLOCK (reqlock);
620 735
621 etp_maybe_start_thread (); 736 etp_maybe_start_thread ();
622 } 737 }
623} 738}
624 739
740static void ecb_cold
625static void etp_set_max_poll_time (double nseconds) 741etp_set_max_poll_time (double nseconds)
626{ 742{
627 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 743 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
628 max_poll_time = nseconds * EIO_TICKS; 744 max_poll_time = nseconds * EIO_TICKS;
629 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 745 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
630} 746}
631 747
748static void ecb_cold
632static void etp_set_max_poll_reqs (unsigned int maxreqs) 749etp_set_max_poll_reqs (unsigned int maxreqs)
633{ 750{
634 if (WORDACCESS_UNSAFE) X_LOCK (reslock); 751 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
635 max_poll_reqs = maxreqs; 752 max_poll_reqs = maxreqs;
636 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock); 753 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
637} 754}
638 755
756static void ecb_cold
639static void etp_set_max_idle (unsigned int nthreads) 757etp_set_max_idle (unsigned int nthreads)
640{ 758{
641 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 759 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
642 max_idle = nthreads; 760 max_idle = nthreads;
643 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 761 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
644} 762}
645 763
764static void ecb_cold
646static void etp_set_idle_timeout (unsigned int seconds) 765etp_set_idle_timeout (unsigned int seconds)
647{ 766{
648 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 767 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
649 idle_timeout = seconds; 768 idle_timeout = seconds;
650 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock); 769 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
651} 770}
652 771
772static void ecb_cold
653static void etp_set_min_parallel (unsigned int nthreads) 773etp_set_min_parallel (unsigned int nthreads)
654{ 774{
655 if (wanted < nthreads) 775 if (wanted < nthreads)
656 wanted = nthreads; 776 wanted = nthreads;
657} 777}
658 778
779static void ecb_cold
659static void etp_set_max_parallel (unsigned int nthreads) 780etp_set_max_parallel (unsigned int nthreads)
660{ 781{
661 if (wanted > nthreads) 782 if (wanted > nthreads)
662 wanted = nthreads; 783 wanted = nthreads;
663 784
664 while (started > wanted) 785 while (started > wanted)
665 etp_end_thread (); 786 etp_end_thread ();
666} 787}
667 788
668/*****************************************************************************/ 789/*****************************************************************************/
669 790
791static void
670static void grp_try_feed (eio_req *grp) 792grp_try_feed (eio_req *grp)
671{ 793{
672 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 794 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
673 { 795 {
674 grp->flags &= ~EIO_FLAG_GROUPADD; 796 grp->flags &= ~EIO_FLAG_GROUPADD;
675 797
682 break; 804 break;
683 } 805 }
684 } 806 }
685} 807}
686 808
809static int
687static int grp_dec (eio_req *grp) 810grp_dec (eio_req *grp)
688{ 811{
689 --grp->size; 812 --grp->size;
690 813
691 /* call feeder, if applicable */ 814 /* call feeder, if applicable */
692 grp_try_feed (grp); 815 grp_try_feed (grp);
696 return eio_finish (grp); 819 return eio_finish (grp);
697 else 820 else
698 return 0; 821 return 0;
699} 822}
700 823
824static void
701void eio_destroy (eio_req *req) 825eio_destroy (eio_req *req)
702{ 826{
703 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 827 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
704 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 828 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
705 829
706 EIO_DESTROY (req); 830 EIO_DESTROY (req);
707} 831}
708 832
833static int
709static int eio_finish (eio_req *req) 834eio_finish (eio_req *req)
710{ 835{
711 int res = EIO_FINISH (req); 836 int res = EIO_FINISH (req);
712 837
713 if (req->grp) 838 if (req->grp)
714 { 839 {
722 if (grp->grp_first == req) 847 if (grp->grp_first == req)
723 grp->grp_first = req->grp_next; 848 grp->grp_first = req->grp_next;
724 849
725 res2 = grp_dec (grp); 850 res2 = grp_dec (grp);
726 851
727 if (!res && res2) 852 if (!res)
728 res = res2; 853 res = res2;
729 } 854 }
730 855
731 eio_destroy (req); 856 eio_destroy (req);
732 857
733 return res; 858 return res;
734} 859}
735 860
861void
736void eio_grp_cancel (eio_req *grp) 862eio_grp_cancel (eio_req *grp)
737{ 863{
738 for (grp = grp->grp_first; grp; grp = grp->grp_next) 864 for (grp = grp->grp_first; grp; grp = grp->grp_next)
739 eio_cancel (grp); 865 eio_cancel (grp);
740} 866}
741 867
868void
742void eio_cancel (eio_req *req) 869eio_cancel (eio_req *req)
743{ 870{
744 etp_cancel (req); 871 etp_cancel (req);
745} 872}
746 873
874void
747void eio_submit (eio_req *req) 875eio_submit (eio_req *req)
748{ 876{
749 etp_submit (req); 877 etp_submit (req);
750} 878}
751 879
752unsigned int eio_nreqs (void) 880unsigned int
881eio_nreqs (void)
753{ 882{
754 return etp_nreqs (); 883 return etp_nreqs ();
755} 884}
756 885
757unsigned int eio_nready (void) 886unsigned int
887eio_nready (void)
758{ 888{
759 return etp_nready (); 889 return etp_nready ();
760} 890}
761 891
762unsigned int eio_npending (void) 892unsigned int
893eio_npending (void)
763{ 894{
764 return etp_npending (); 895 return etp_npending ();
765} 896}
766 897
767unsigned int eio_nthreads (void) 898unsigned int ecb_cold
899eio_nthreads (void)
768{ 900{
769 return etp_nthreads (); 901 return etp_nthreads ();
770} 902}
771 903
904void ecb_cold
772void eio_set_max_poll_time (double nseconds) 905eio_set_max_poll_time (double nseconds)
773{ 906{
774 etp_set_max_poll_time (nseconds); 907 etp_set_max_poll_time (nseconds);
775} 908}
776 909
910void ecb_cold
777void eio_set_max_poll_reqs (unsigned int maxreqs) 911eio_set_max_poll_reqs (unsigned int maxreqs)
778{ 912{
779 etp_set_max_poll_reqs (maxreqs); 913 etp_set_max_poll_reqs (maxreqs);
780} 914}
781 915
916void ecb_cold
782void eio_set_max_idle (unsigned int nthreads) 917eio_set_max_idle (unsigned int nthreads)
783{ 918{
784 etp_set_max_idle (nthreads); 919 etp_set_max_idle (nthreads);
785} 920}
786 921
922void ecb_cold
787void eio_set_idle_timeout (unsigned int seconds) 923eio_set_idle_timeout (unsigned int seconds)
788{ 924{
789 etp_set_idle_timeout (seconds); 925 etp_set_idle_timeout (seconds);
790} 926}
791 927
928void ecb_cold
792void eio_set_min_parallel (unsigned int nthreads) 929eio_set_min_parallel (unsigned int nthreads)
793{ 930{
794 etp_set_min_parallel (nthreads); 931 etp_set_min_parallel (nthreads);
795} 932}
796 933
934void ecb_cold
797void eio_set_max_parallel (unsigned int nthreads) 935eio_set_max_parallel (unsigned int nthreads)
798{ 936{
799 etp_set_max_parallel (nthreads); 937 etp_set_max_parallel (nthreads);
800} 938}
801 939
802int eio_poll (void) 940int eio_poll (void)
811# undef pread 949# undef pread
812# undef pwrite 950# undef pwrite
813# define pread eio__pread 951# define pread eio__pread
814# define pwrite eio__pwrite 952# define pwrite eio__pwrite
815 953
816static ssize_t 954static eio_ssize_t
817eio__pread (int fd, void *buf, size_t count, off_t offset) 955eio__pread (int fd, void *buf, size_t count, off_t offset)
818{ 956{
819 ssize_t res; 957 eio_ssize_t res;
820 off_t ooffset; 958 off_t ooffset;
821 959
822 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
823 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
824 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
827 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
828 966
829 return res; 967 return res;
830} 968}
831 969
832static ssize_t 970static eio_ssize_t
833eio__pwrite (int fd, void *buf, size_t count, off_t offset) 971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
834{ 972{
835 ssize_t res; 973 eio_ssize_t res;
836 off_t ooffset; 974 off_t ooffset;
837 975
838 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
839 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
840 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
872#ifndef HAVE_FUTIMES 1010#ifndef HAVE_FUTIMES
873 1011
874# undef futimes 1012# undef futimes
875# define futimes(fd,times) eio__futimes (fd, times) 1013# define futimes(fd,times) eio__futimes (fd, times)
876 1014
1015static int
877static int eio__futimes (int fd, const struct timeval tv[2]) 1016eio__futimes (int fd, const struct timeval tv[2])
878{ 1017{
879 errno = ENOSYS; 1018 errno = ENOSYS;
880 return -1; 1019 return -1;
881} 1020}
882 1021
885#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
886# undef fdatasync 1025# undef fdatasync
887# define fdatasync(fd) fsync (fd) 1026# define fdatasync(fd) fsync (fd)
888#endif 1027#endif
889 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
890/* sync_file_range always needs emulation */ 1047/* sync_file_range always needs emulation */
891int 1048static int
892eio__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)
893{ 1050{
894#if HAVE_SYNC_FILE_RANGE 1051#if HAVE_SYNC_FILE_RANGE
895 int res; 1052 int res;
896 1053
913 /* 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
914 * call fdatasync, as that matches the expectation of its users best */ 1071 * call fdatasync, as that matches the expectation of its users best */
915 return fdatasync (fd); 1072 return fdatasync (fd);
916} 1073}
917 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
918#if !HAVE_READAHEAD 1086#if !HAVE_READAHEAD
919# undef readahead 1087# undef readahead
920# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1088# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
921 1089
922static ssize_t 1090static eio_ssize_t
923eio__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)
924{ 1092{
925 size_t todo = count; 1093 size_t todo = count;
926 dBUF; 1094 dBUF;
927 1095
928 while (todo > 0) 1096 while (todo > 0)
929 { 1097 {
930 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE; 1098 size_t len = todo < EIO_BUFSIZE ? todo : EIO_BUFSIZE;
931
932 pread (fd, eio_buf, len, offset);
933 offset += len; 1099 offset += len;
934 todo -= len; 1100 todo -= len;
935 } 1101 }
936 1102
937 errno = 0; 1103 FUBd;
1104
1105 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
1106 /* but not for e.g. EIO or eof, so we also never fail */
938 return count; 1107 return 0;
939} 1108}
940 1109
941#endif 1110#endif
942 1111
943/* sendfile always needs emulation */ 1112/* sendfile always needs emulation */
944static ssize_t 1113static eio_ssize_t
945eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1114eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
946{ 1115{
1116 eio_ssize_t written = 0;
947 ssize_t res; 1117 eio_ssize_t res;
948 1118
949 if (!count) 1119 if (!count)
950 return 0; 1120 return 0;
951 1121
1122 for (;;)
1123 {
1124#ifdef __APPLE__
1125# undef HAVE_SENDFILE /* broken, as everything on os x */
1126#endif
952#if HAVE_SENDFILE 1127#if HAVE_SENDFILE
953# if __linux 1128# if __linux
1129 off_t soffset = offset;
954 res = sendfile (ofd, ifd, &offset, count); 1130 res = sendfile (ofd, ifd, &soffset, count);
955 1131
956# elif __FreeBSD__ 1132# elif __FreeBSD__
957 /* 1133 /*
958 * Of course, the freebsd sendfile is a dire hack with no thoughts 1134 * Of course, the freebsd sendfile is a dire hack with no thoughts
959 * wasted on making it similar to other I/O functions. 1135 * wasted on making it similar to other I/O functions.
960 */ 1136 */
961 {
962 off_t sbytes; 1137 off_t sbytes;
963 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 1138 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
964 1139
965 #if 0 /* according to the manpage, this is correct, but broken behaviour */ 1140 #if 0 /* according to the manpage, this is correct, but broken behaviour */
966 /* freebsd' sendfile will return 0 on success */ 1141 /* freebsd' sendfile will return 0 on success */
967 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */ 1142 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
968 /* not on e.g. EIO or EPIPE - sounds broken */ 1143 /* not on e.g. EIO or EPIPE - sounds broken */
969 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0) 1144 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
970 res = sbytes; 1145 res = sbytes;
971 #endif 1146 #endif
972 1147
973 /* according to source inspection, this is correct, and useful behaviour */ 1148 /* according to source inspection, this is correct, and useful behaviour */
974 if (sbytes) 1149 if (sbytes)
975 res = sbytes; 1150 res = sbytes;
976 }
977 1151
978# elif defined (__APPLE__) 1152# elif defined (__APPLE__)
979
980 {
981 off_t sbytes = count; 1153 off_t sbytes = count;
982 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1154 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
983 1155
984 /* according to the manpage, sbytes is always valid */ 1156 /* according to the manpage, sbytes is always valid */
985 if (sbytes) 1157 if (sbytes)
986 res = sbytes; 1158 res = sbytes;
987 }
988 1159
989# elif __hpux 1160# elif __hpux
990 res = sendfile (ofd, ifd, offset, count, 0, 0); 1161 res = sendfile (ofd, ifd, offset, count, 0, 0);
991 1162
992# elif __solaris 1163# elif __solaris
993 {
994 struct sendfilevec vec; 1164 struct sendfilevec vec;
995 size_t sbytes; 1165 size_t sbytes;
996 1166
997 vec.sfv_fd = ifd; 1167 vec.sfv_fd = ifd;
998 vec.sfv_flag = 0; 1168 vec.sfv_flag = 0;
999 vec.sfv_off = offset; 1169 vec.sfv_off = offset;
1000 vec.sfv_len = count; 1170 vec.sfv_len = count;
1001 1171
1002 res = sendfilev (ofd, &vec, 1, &sbytes); 1172 res = sendfilev (ofd, &vec, 1, &sbytes);
1003 1173
1004 if (res < 0 && sbytes) 1174 if (res < 0 && sbytes)
1005 res = sbytes; 1175 res = sbytes;
1006 }
1007 1176
1008# endif 1177# endif
1009 1178
1010#elif defined (_WIN32) 1179#elif defined (_WIN32) && 0
1011
1012 /* does not work, just for documentation of what would need to be done */ 1180 /* does not work, just for documentation of what would need to be done */
1013 { 1181 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1182 /* libeio guarantees that the file offset does not change, and windows */
1183 /* has no way to get an independent handle to the same file description */
1014 HANDLE h = TO_SOCKET (ifd); 1184 HANDLE h = TO_SOCKET (ifd);
1015 SetFilePointer (h, offset, 0, FILE_BEGIN); 1185 SetFilePointer (h, offset, 0, FILE_BEGIN);
1016 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 1186 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1017 }
1018 1187
1019#else 1188#else
1020 res = -1; 1189 res = -1;
1021 errno = ENOSYS; 1190 errno = ENOSYS;
1022#endif 1191#endif
1023 1192
1193 /* we assume sendfile can copy at least 128mb in one go */
1194 if (res <= 128 * 1024 * 1024)
1195 {
1196 if (res > 0)
1197 written += res;
1198
1199 if (written)
1200 return written;
1201
1202 break;
1203 }
1204 else
1205 {
1206 /* if we requested more, then probably the kernel was lazy */
1207 written += res;
1208 offset += res;
1209 count -= res;
1210
1211 if (!count)
1212 return written;
1213 }
1214 }
1215
1024 if (res < 0 1216 if (res < 0
1025 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1217 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1026 /* BSDs */ 1218 /* BSDs */
1027#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1219#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1028 || errno == ENOTSUP 1220 || errno == ENOTSUP
1029#endif 1221#endif
1222#ifdef EOPNOTSUPP /* windows */
1030 || errno == EOPNOTSUPP /* BSDs */ 1223 || errno == EOPNOTSUPP /* BSDs */
1224#endif
1031#if __solaris 1225#if __solaris
1032 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1226 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1033#endif 1227#endif
1034 ) 1228 )
1035 ) 1229 )
1039 1233
1040 res = 0; 1234 res = 0;
1041 1235
1042 while (count) 1236 while (count)
1043 { 1237 {
1044 ssize_t cnt; 1238 eio_ssize_t cnt;
1045 1239
1046 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1240 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1047 1241
1048 if (cnt <= 0) 1242 if (cnt <= 0)
1049 { 1243 {
1061 1255
1062 offset += cnt; 1256 offset += cnt;
1063 res += cnt; 1257 res += cnt;
1064 count -= cnt; 1258 count -= cnt;
1065 } 1259 }
1260
1261 FUBd;
1066 } 1262 }
1067 1263
1068 return res; 1264 return res;
1265}
1266
1267#ifdef PAGESIZE
1268# define eio_pagesize() PAGESIZE
1269#else
1270static intptr_t
1271eio_pagesize (void)
1272{
1273 static intptr_t page;
1274
1275 if (!page)
1276 page = sysconf (_SC_PAGESIZE);
1277
1278 return page;
1279}
1280#endif
1281
1282static void
1283eio_page_align (void **addr, size_t *length)
1284{
1285 intptr_t mask = eio_pagesize () - 1;
1286
1287 /* round down addr */
1288 intptr_t adj = mask & (intptr_t)*addr;
1289
1290 *addr = (void *)((intptr_t)*addr - adj);
1291 *length += adj;
1292
1293 /* round up length */
1294 *length = (*length + mask) & ~mask;
1295}
1296
1297#if !_POSIX_MEMLOCK
1298# define eio__mlockall(a) EIO_ENOSYS ()
1299#else
1300
1301static int
1302eio__mlockall (int flags)
1303{
1304 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1305 extern int mallopt (int, int);
1306 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1307 #endif
1308
1309 if (EIO_MCL_CURRENT != MCL_CURRENT
1310 || EIO_MCL_FUTURE != MCL_FUTURE)
1311 {
1312 flags = 0
1313 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1314 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1315 }
1316
1317 return mlockall (flags);
1318}
1319#endif
1320
1321#if !_POSIX_MEMLOCK_RANGE
1322# define eio__mlock(a,b) EIO_ENOSYS ()
1323#else
1324
1325static int
1326eio__mlock (void *addr, size_t length)
1327{
1328 eio_page_align (&addr, &length);
1329
1330 return mlock (addr, length);
1331}
1332
1333#endif
1334
1335#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1336# define eio__msync(a,b,c) EIO_ENOSYS ()
1337#else
1338
1339static int
1340eio__msync (void *mem, size_t len, int flags)
1341{
1342 eio_page_align (&mem, &len);
1343
1344 if (EIO_MS_ASYNC != MS_SYNC
1345 || EIO_MS_INVALIDATE != MS_INVALIDATE
1346 || EIO_MS_SYNC != MS_SYNC)
1347 {
1348 flags = 0
1349 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1350 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1351 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1352 }
1353
1354 return msync (mem, len, flags);
1355}
1356
1357#endif
1358
1359static int
1360eio__mtouch (eio_req *req)
1361{
1362 void *mem = req->ptr2;
1363 size_t len = req->size;
1364 int flags = req->int1;
1365
1366 eio_page_align (&mem, &len);
1367
1368 {
1369 intptr_t addr = (intptr_t)mem;
1370 intptr_t end = addr + len;
1371 intptr_t page = eio_pagesize ();
1372
1373 if (addr < end)
1374 if (flags & EIO_MT_MODIFY) /* modify */
1375 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1376 else
1377 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1378 }
1379
1380 return 0;
1381}
1382
1383/*****************************************************************************/
1384/* requests implemented outside eio_execute, because they are so large */
1385
1386/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1387static int
1388eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1389{
1390 const char *rel = path;
1391 char *res;
1392 char *tmp1, *tmp2;
1393#if SYMLOOP_MAX > 32
1394 int symlinks = SYMLOOP_MAX;
1395#else
1396 int symlinks = 32;
1397#endif
1398
1399 errno = EINVAL;
1400 if (!rel)
1401 return -1;
1402
1403 errno = ENOENT;
1404 if (!*rel)
1405 return -1;
1406
1407 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1408 tmp1 = res + PATH_MAX;
1409 tmp2 = tmp1 + PATH_MAX;
1410
1411#if 0 /* disabled, the musl way to do things is just too racy */
1412#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1413 /* on linux we may be able to ask the kernel */
1414 {
1415 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1416
1417 if (fd >= 0)
1418 {
1419 sprintf (tmp1, "/proc/self/fd/%d", fd);
1420 req->result = readlink (tmp1, res, PATH_MAX);
1421 close (fd);
1422
1423 /* here we should probably stat the open file and the disk file, to make sure they still match */
1424
1425 if (req->result > 0)
1426 goto done;
1427 }
1428 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1429 return;
1430 }
1431#endif
1432#endif
1433
1434 if (*rel != '/')
1435 {
1436 int len;
1437
1438 errno = ENOENT;
1439 if (wd == EIO_INVALID_WD)
1440 return -1;
1441
1442 if (wd == EIO_CWD)
1443 {
1444 if (!getcwd (res, PATH_MAX))
1445 return -1;
1446
1447 len = strlen (res);
1448 }
1449 else
1450 memcpy (res, wd->str, len = wd->len);
1451
1452 if (res [1]) /* only use if not / */
1453 res += len;
1454 }
1455
1456 while (*rel)
1457 {
1458 eio_ssize_t len, linklen;
1459 const char *beg = rel;
1460
1461 while (*rel && *rel != '/')
1462 ++rel;
1463
1464 len = rel - beg;
1465
1466 if (!len) /* skip slashes */
1467 {
1468 ++rel;
1469 continue;
1470 }
1471
1472 if (beg [0] == '.')
1473 {
1474 if (len == 1)
1475 continue; /* . - nop */
1476
1477 if (beg [1] == '.' && len == 2)
1478 {
1479 /* .. - back up one component, if possible */
1480
1481 while (res != tmpbuf->ptr)
1482 if (*--res == '/')
1483 break;
1484
1485 continue;
1486 }
1487 }
1488
1489 errno = ENAMETOOLONG;
1490 if (res + 1 + len + 1 >= tmp1)
1491 return -1;
1492
1493 /* copy one component */
1494 *res = '/';
1495 memcpy (res + 1, beg, len);
1496
1497 /* zero-terminate, for readlink */
1498 res [len + 1] = 0;
1499
1500 /* now check if it's a symlink */
1501 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1502
1503 if (linklen < 0)
1504 {
1505 if (errno != EINVAL)
1506 return -1;
1507
1508 /* it's a normal directory. hopefully */
1509 res += len + 1;
1510 }
1511 else
1512 {
1513 /* yay, it was a symlink - build new path in tmp2 */
1514 int rellen = strlen (rel);
1515
1516 errno = ENAMETOOLONG;
1517 if (linklen + 1 + rellen >= PATH_MAX)
1518 return -1;
1519
1520 errno = ELOOP;
1521 if (!--symlinks)
1522 return -1;
1523
1524 if (*tmp1 == '/')
1525 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1526
1527 /* we need to be careful, as rel might point into tmp2 already */
1528 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1529 tmp2 [linklen] = '/';
1530 memcpy (tmp2, tmp1, linklen);
1531
1532 rel = tmp2;
1533 }
1534 }
1535
1536 /* special case for the lone root path */
1537 if (res == tmpbuf->ptr)
1538 *res++ = '/';
1539
1540 return res - (char *)tmpbuf->ptr;
1069} 1541}
1070 1542
1071static signed char 1543static signed char
1072eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1544eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1073{ 1545{
1081 1553
1082#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1554#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1083#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1555#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1084 1556
1085static void 1557static void
1086eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1558eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1087{ 1559{
1088 unsigned char bits [9 + sizeof (ino_t) * 8]; 1560 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1089 unsigned char *bit = bits; 1561 unsigned char *bit = bits;
1090 1562
1091 assert (CHAR_BIT == 8); 1563 assert (CHAR_BIT == 8);
1092 assert (sizeof (eio_dirent) * 8 < 256); 1564 assert (sizeof (eio_dirent) * 8 < 256);
1093 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1565 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1095 1567
1096 if (size <= EIO_SORT_FAST) 1568 if (size <= EIO_SORT_FAST)
1097 return; 1569 return;
1098 1570
1099 /* first prepare an array of bits to test in our radix sort */ 1571 /* first prepare an array of bits to test in our radix sort */
1100 /* try to take endianness into account, as well as differences in ino_t sizes */ 1572 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1101 /* inode_bits must contain all inodes ORed together */ 1573 /* inode_bits must contain all inodes ORed together */
1102 /* which is used to skip bits that are 0 everywhere, which is very common */ 1574 /* which is used to skip bits that are 0 everywhere, which is very common */
1103 { 1575 {
1104 ino_t endianness; 1576 eio_ino_t endianness;
1105 int i, j; 1577 int i, j;
1106 1578
1107 /* we store the byte offset of byte n into byte n of "endianness" */ 1579 /* we store the byte offset of byte n into byte n of "endianness" */
1108 for (i = 0; i < sizeof (ino_t); ++i) 1580 for (i = 0; i < sizeof (eio_ino_t); ++i)
1109 ((unsigned char *)&endianness)[i] = i; 1581 ((unsigned char *)&endianness)[i] = i;
1110 1582
1111 *bit++ = 0; 1583 *bit++ = 0;
1112 1584
1113 for (i = 0; i < sizeof (ino_t); ++i) 1585 for (i = 0; i < sizeof (eio_ino_t); ++i)
1114 { 1586 {
1115 /* shifting off the byte offsets out of "endianness" */ 1587 /* shifting off the byte offsets out of "endianness" */
1116 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1588 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1117 endianness >>= 8; 1589 endianness >>= 8;
1118 1590
1119 for (j = 0; j < 8; ++j) 1591 for (j = 0; j < 8; ++j)
1120 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1592 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1121 *bit++ = offs + j; 1593 *bit++ = offs + j;
1122 } 1594 }
1123 1595
1124 for (j = 0; j < 8; ++j) 1596 for (j = 0; j < 8; ++j)
1125 if (score_bits & (1 << j)) 1597 if (score_bits & (1 << j))
1126 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1598 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1127 } 1599 }
1128 1600
1129 /* now actually do the sorting (a variant of MSD radix sort) */ 1601 /* now actually do the sorting (a variant of MSD radix sort) */
1130 { 1602 {
1131 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1603 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1132 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1604 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1133 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1605 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1134 int stk_idx = 0; 1606 int stk_idx = 0;
1135 1607
1136 base_stk [stk_idx] = dents; 1608 base_stk [stk_idx] = dents;
1137 end_stk [stk_idx] = dents + size; 1609 end_stk [stk_idx] = dents + size;
1138 bit_stk [stk_idx] = bit - 1; 1610 bit_stk [stk_idx] = bit - 1;
1217 } 1689 }
1218 } 1690 }
1219} 1691}
1220 1692
1221static void 1693static void
1222eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1694eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1223{ 1695{
1224 if (size <= 1) 1696 if (size <= 1)
1225 return; /* our insertion sort relies on size > 0 */ 1697 return; /* our insertion sort relies on size > 0 */
1226 1698
1227 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1699 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1235 1707
1236/* read a full directory */ 1708/* read a full directory */
1237static void 1709static void
1238eio__scandir (eio_req *req, etp_worker *self) 1710eio__scandir (eio_req *req, etp_worker *self)
1239{ 1711{
1240 DIR *dirp;
1241 EIO_STRUCT_DIRENT *entp;
1242 char *name, *names; 1712 char *name, *names;
1243 int namesalloc = 4096; 1713 int namesalloc = 4096 - sizeof (void *) * 4;
1244 int namesoffs = 0; 1714 int namesoffs = 0;
1245 int flags = req->int1; 1715 int flags = req->int1;
1246 eio_dirent *dents = 0; 1716 eio_dirent *dents = 0;
1247 int dentalloc = 128; 1717 int dentalloc = 128;
1248 int dentoffs = 0; 1718 int dentoffs = 0;
1249 ino_t inode_bits = 0; 1719 eio_ino_t inode_bits = 0;
1720#ifdef _WIN32
1721 HANDLE dirp;
1722 WIN32_FIND_DATA entp;
1723#else
1724 DIR *dirp;
1725 EIO_STRUCT_DIRENT *entp;
1726#endif
1250 1727
1251 req->result = -1; 1728 req->result = -1;
1252 1729
1253 if (!(flags & EIO_READDIR_DENTS)) 1730 if (!(flags & EIO_READDIR_DENTS))
1254 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1731 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1255 1732
1256 X_LOCK (wrklock); 1733#ifdef _WIN32
1257 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1734 {
1735 int len = strlen ((const char *)req->ptr1);
1736 char *path = malloc (MAX_PATH);
1737 const char *fmt;
1738 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1739
1740 if (!len)
1741 fmt = "./*";
1742 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1743 fmt = "%s*";
1744 else
1745 fmt = "%s/*";
1746
1747 _snprintf (path, MAX_PATH, fmt, reqpath);
1748 dirp = FindFirstFile (path, &entp);
1749 free (path);
1750
1751 if (dirp == INVALID_HANDLE_VALUE)
1752 {
1753 /* should steal _dosmaperr */
1754 switch (GetLastError ())
1755 {
1756 case ERROR_FILE_NOT_FOUND:
1757 req->result = 0;
1758 break;
1759
1760 case ERROR_INVALID_NAME:
1761 case ERROR_PATH_NOT_FOUND:
1762 case ERROR_NO_MORE_FILES:
1763 errno = ENOENT;
1764 break;
1765
1766 case ERROR_NOT_ENOUGH_MEMORY:
1767 errno = ENOMEM;
1768 break;
1769
1770 default:
1771 errno = EINVAL;
1772 break;
1773 }
1774
1775 return;
1776 }
1777 }
1778#else
1779 #if HAVE_AT
1780 if (req->wd)
1781 {
1782 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1783
1784 if (fd < 0)
1785 return;
1786
1787 dirp = fdopendir (fd);
1788
1789 if (!dirp)
1790 close (fd);
1791 }
1792 else
1258 self->dirp = dirp = opendir (req->ptr1); 1793 dirp = opendir (req->ptr1);
1794 #else
1795 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1796 #endif
1797
1798 if (!dirp)
1799 return;
1800#endif
1801
1802 if (req->flags & EIO_FLAG_PTR1_FREE)
1803 free (req->ptr1);
1259 1804
1260 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1805 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1261 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1806 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1262 req->ptr2 = names = malloc (namesalloc); 1807 req->ptr2 = names = malloc (namesalloc);
1263 X_UNLOCK (wrklock);
1264 1808
1265 if (dirp && names && (!flags || dents)) 1809 if (!names || (flags && !dents))
1810 return;
1811
1266 for (;;) 1812 for (;;)
1267 { 1813 {
1814 int done;
1815
1816#ifdef _WIN32
1817 done = !dirp;
1818#else
1268 errno = 0; 1819 errno = 0;
1269 entp = readdir (dirp); 1820 entp = readdir (dirp);
1821 done = !entp;
1822#endif
1270 1823
1271 if (!entp) 1824 if (done)
1272 { 1825 {
1826#ifndef _WIN32
1827 int old_errno = errno;
1828 closedir (dirp);
1829 errno = old_errno;
1830
1273 if (errno) 1831 if (errno)
1274 break; 1832 break;
1833#endif
1275 1834
1276 /* sort etc. */ 1835 /* sort etc. */
1277 req->int1 = flags; 1836 req->int1 = flags;
1278 req->result = dentoffs; 1837 req->result = dentoffs;
1279 1838
1280 if (flags & EIO_READDIR_STAT_ORDER) 1839 if (flags & EIO_READDIR_STAT_ORDER)
1281 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1840 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1282 else if (flags & EIO_READDIR_DIRS_FIRST) 1841 else if (flags & EIO_READDIR_DIRS_FIRST)
1283 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1842 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1284 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1843 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1285 else 1844 else
1845 {
1846 /* in this case, all is known, and we just put dirs first and sort them */
1847 eio_dirent *oth = dents + dentoffs;
1848 eio_dirent *dir = dents;
1849
1850 /* now partition dirs to the front, and non-dirs to the back */
1851 /* by walking from both sides and swapping if necessary */
1852 while (oth > dir)
1853 {
1854 if (dir->type == EIO_DT_DIR)
1855 ++dir;
1856 else if ((--oth)->type == EIO_DT_DIR)
1857 {
1858 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1859
1860 ++dir;
1861 }
1862 }
1863
1864 /* now sort the dirs only (dirs all have the same score) */
1865 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1866 }
1867
1868 break;
1869 }
1870
1871 /* now add the entry to our list(s) */
1872 name = D_NAME (entp);
1873
1874 /* skip . and .. entries */
1875 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1876 {
1877 int len = D_NAMLEN (entp) + 1;
1878
1879 while (ecb_expect_false (namesoffs + len > namesalloc))
1880 {
1881 namesalloc *= 2;
1882 req->ptr2 = names = realloc (names, namesalloc);
1883
1884 if (!names)
1885 break;
1886 }
1887
1888 memcpy (names + namesoffs, name, len);
1889
1890 if (dents)
1891 {
1892 struct eio_dirent *ent;
1893
1894 if (ecb_expect_false (dentoffs == dentalloc))
1286 { 1895 {
1287 /* in this case, all is known, and we just put dirs first and sort them */ 1896 dentalloc *= 2;
1897 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1898
1899 if (!dents)
1900 break;
1901 }
1902
1288 eio_dirent *oth = dents + dentoffs; 1903 ent = dents + dentoffs;
1289 eio_dirent *dir = dents;
1290 1904
1291 /* now partition dirs to the front, and non-dirs to the back */ 1905 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1292 /* by walking from both sides and swapping if necessary */ 1906 ent->namelen = len - 1;
1293 while (oth > dir) 1907 ent->inode = D_INO (entp);
1908
1909 inode_bits |= ent->inode;
1910
1911 switch (D_TYPE (entp))
1912 {
1913 default:
1914 ent->type = EIO_DT_UNKNOWN;
1915 flags |= EIO_READDIR_FOUND_UNKNOWN;
1916 break;
1917
1918 #ifdef DT_FIFO
1919 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1920 #endif
1921 #ifdef DT_CHR
1922 case DT_CHR: ent->type = EIO_DT_CHR; break;
1923 #endif
1924 #ifdef DT_MPC
1925 case DT_MPC: ent->type = EIO_DT_MPC; break;
1926 #endif
1927 #ifdef DT_DIR
1928 case DT_DIR: ent->type = EIO_DT_DIR; break;
1929 #endif
1930 #ifdef DT_NAM
1931 case DT_NAM: ent->type = EIO_DT_NAM; break;
1932 #endif
1933 #ifdef DT_BLK
1934 case DT_BLK: ent->type = EIO_DT_BLK; break;
1935 #endif
1936 #ifdef DT_MPB
1937 case DT_MPB: ent->type = EIO_DT_MPB; break;
1938 #endif
1939 #ifdef DT_REG
1940 case DT_REG: ent->type = EIO_DT_REG; break;
1941 #endif
1942 #ifdef DT_NWK
1943 case DT_NWK: ent->type = EIO_DT_NWK; break;
1944 #endif
1945 #ifdef DT_CMP
1946 case DT_CMP: ent->type = EIO_DT_CMP; break;
1947 #endif
1948 #ifdef DT_LNK
1949 case DT_LNK: ent->type = EIO_DT_LNK; break;
1950 #endif
1951 #ifdef DT_SOCK
1952 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1953 #endif
1954 #ifdef DT_DOOR
1955 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1956 #endif
1957 #ifdef DT_WHT
1958 case DT_WHT: ent->type = EIO_DT_WHT; break;
1959 #endif
1960 }
1961
1962 ent->score = 7;
1963
1964 if (flags & EIO_READDIR_DIRS_FIRST)
1965 {
1966 if (ent->type == EIO_DT_UNKNOWN)
1294 { 1967 {
1295 if (dir->type == EIO_DT_DIR) 1968 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1296 ++dir; 1969 ent->score = 1;
1297 else if ((--oth)->type == EIO_DT_DIR) 1970 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1298 { 1971 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1299 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1300
1301 ++dir;
1302 }
1303 } 1972 }
1304 1973 else if (ent->type == EIO_DT_DIR)
1305 /* now sort the dirs only (dirs all have the same score) */ 1974 ent->score = 0;
1306 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1307 } 1975 }
1308
1309 break;
1310 }
1311
1312 /* now add the entry to our list(s) */
1313 name = entp->d_name;
1314
1315 /* skip . and .. entries */
1316 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1317 {
1318 int len = D_NAMLEN (entp) + 1;
1319
1320 while (expect_false (namesoffs + len > namesalloc))
1321 {
1322 namesalloc *= 2;
1323 X_LOCK (wrklock);
1324 req->ptr2 = names = realloc (names, namesalloc);
1325 X_UNLOCK (wrklock);
1326
1327 if (!names)
1328 break;
1329 } 1976 }
1330 1977
1331 memcpy (names + namesoffs, name, len);
1332
1333 if (dents)
1334 {
1335 struct eio_dirent *ent;
1336
1337 if (expect_false (dentoffs == dentalloc))
1338 {
1339 dentalloc *= 2;
1340 X_LOCK (wrklock);
1341 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1342 X_UNLOCK (wrklock);
1343
1344 if (!dents)
1345 break;
1346 }
1347
1348 ent = dents + dentoffs;
1349
1350 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1351 ent->namelen = len - 1;
1352 ent->inode = D_INO (entp);
1353
1354 inode_bits |= ent->inode;
1355
1356 switch (D_TYPE (entp))
1357 {
1358 default:
1359 ent->type = EIO_DT_UNKNOWN;
1360 flags |= EIO_READDIR_FOUND_UNKNOWN;
1361 break;
1362
1363 #ifdef DT_FIFO
1364 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1365 #endif
1366 #ifdef DT_CHR
1367 case DT_CHR: ent->type = EIO_DT_CHR; break;
1368 #endif
1369 #ifdef DT_MPC
1370 case DT_MPC: ent->type = EIO_DT_MPC; break;
1371 #endif
1372 #ifdef DT_DIR
1373 case DT_DIR: ent->type = EIO_DT_DIR; break;
1374 #endif
1375 #ifdef DT_NAM
1376 case DT_NAM: ent->type = EIO_DT_NAM; break;
1377 #endif
1378 #ifdef DT_BLK
1379 case DT_BLK: ent->type = EIO_DT_BLK; break;
1380 #endif
1381 #ifdef DT_MPB
1382 case DT_MPB: ent->type = EIO_DT_MPB; break;
1383 #endif
1384 #ifdef DT_REG
1385 case DT_REG: ent->type = EIO_DT_REG; break;
1386 #endif
1387 #ifdef DT_NWK
1388 case DT_NWK: ent->type = EIO_DT_NWK; break;
1389 #endif
1390 #ifdef DT_CMP
1391 case DT_CMP: ent->type = EIO_DT_CMP; break;
1392 #endif
1393 #ifdef DT_LNK
1394 case DT_LNK: ent->type = EIO_DT_LNK; break;
1395 #endif
1396 #ifdef DT_SOCK
1397 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1398 #endif
1399 #ifdef DT_DOOR
1400 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1401 #endif
1402 #ifdef DT_WHT
1403 case DT_WHT: ent->type = EIO_DT_WHT; break;
1404 #endif
1405 }
1406
1407 ent->score = 7;
1408
1409 if (flags & EIO_READDIR_DIRS_FIRST)
1410 {
1411 if (ent->type == EIO_DT_UNKNOWN)
1412 {
1413 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1414 ent->score = 1;
1415 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1416 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1417 }
1418 else if (ent->type == EIO_DT_DIR)
1419 ent->score = 0;
1420 }
1421 }
1422
1423 namesoffs += len; 1978 namesoffs += len;
1424 ++dentoffs; 1979 ++dentoffs;
1425 } 1980 }
1426 1981
1427 if (EIO_CANCELLED (req)) 1982 if (EIO_CANCELLED (req))
1428 { 1983 {
1429 errno = ECANCELED; 1984 errno = ECANCELED;
1430 break; 1985 break;
1431 } 1986 }
1987
1988#ifdef _WIN32
1989 if (!FindNextFile (dirp, &entp))
1990 {
1991 FindClose (dirp);
1992 dirp = 0;
1432 } 1993 }
1994#endif
1995 }
1433} 1996}
1434 1997
1435#ifdef PAGESIZE 1998/*****************************************************************************/
1436# define eio_pagesize() PAGESIZE 1999/* working directory stuff */
1437#else 2000/* various deficiencies in the posix 2008 api force us to */
1438static intptr_t 2001/* keep the absolute path in string form at all times */
1439eio_pagesize (void) 2002/* fuck yeah. */
1440{
1441 static intptr_t page;
1442 2003
1443 if (!page) 2004#if !HAVE_AT
1444 page = sysconf (_SC_PAGESIZE);
1445 2005
2006/* a bit like realpath, but usually faster because it doesn'T have to return */
2007/* an absolute or canonical path */
2008static const char *
2009wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2010{
2011 if (!wd || *path == '/')
2012 return path;
2013
2014 if (path [0] == '.' && !path [1])
2015 return wd->str;
2016
2017 {
2018 int l1 = wd->len;
2019 int l2 = strlen (path);
2020
2021 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2022
2023 memcpy (res, wd->str, l1);
2024 res [l1] = '/';
2025 memcpy (res + l1 + 1, path, l2 + 1);
2026
2027 return res;
2028 }
2029}
2030
2031#endif
2032
2033static eio_wd
2034eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2035{
2036 int fd;
2037 eio_wd res;
2038 int len = eio__realpath (tmpbuf, wd, path);
2039
2040 if (len < 0)
2041 return EIO_INVALID_WD;
2042
2043#if HAVE_AT
2044 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2045
2046 if (fd < 0)
2047 return EIO_INVALID_WD;
2048#endif
2049
2050 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2051
2052#if HAVE_AT
2053 res->fd = fd;
2054#endif
2055
2056 res->len = len;
2057 memcpy (res->str, tmpbuf->ptr, len);
2058 res->str [len] = 0;
2059
1446 return page; 2060 return res;
1447} 2061}
1448#endif
1449 2062
1450static void 2063eio_wd
1451eio_page_align (void **addr, size_t *length) 2064eio_wd_open_sync (eio_wd wd, const char *path)
1452{ 2065{
1453 intptr_t mask = eio_pagesize () - 1; 2066 struct tmpbuf tmpbuf = { 0 };
2067 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2068 free (tmpbuf.ptr);
1454 2069
1455 /* round down addr */ 2070 return wd;
1456 intptr_t adj = mask & (intptr_t)*addr;
1457
1458 *addr = (void *)((intptr_t)*addr - adj);
1459 *length += adj;
1460
1461 /* round up length */
1462 *length = (*length + mask) & ~mask;
1463} 2071}
1464 2072
1465#if !_POSIX_MEMLOCK 2073void
1466# define eio__mlockall(a) ((errno = ENOSYS), -1) 2074eio_wd_close_sync (eio_wd wd)
1467#else 2075{
2076 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2077 {
2078 #if HAVE_AT
2079 close (wd->fd);
2080 #endif
2081 free (wd);
2082 }
2083}
2084
2085#if HAVE_AT
2086
2087/* they forgot these */
1468 2088
1469static int 2089static int
1470eio__mlockall (int flags) 2090eio__truncateat (int dirfd, const char *path, off_t length)
1471{ 2091{
1472 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 2092 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1473 extern int mallopt (int, int); 2093 int res;
1474 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1475 #endif
1476 2094
1477 if (EIO_MCL_CURRENT != MCL_CURRENT 2095 if (fd < 0)
1478 || EIO_MCL_FUTURE != MCL_FUTURE) 2096 return fd;
1479 {
1480 flags = 0
1481 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1482 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1483 }
1484 2097
1485 return mlockall (flags); 2098 res = ftruncate (fd, length);
2099 close (fd);
2100 return res;
1486} 2101}
1487#endif
1488
1489#if !_POSIX_MEMLOCK_RANGE
1490# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1491#else
1492 2102
1493static int 2103static int
1494eio__mlock (void *addr, size_t length) 2104eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1495{ 2105{
1496 eio_page_align (&addr, &length); 2106 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2107 int res;
1497 2108
1498 return mlock (addr, length); 2109 if (fd < 0)
1499} 2110 return fd;
1500 2111
1501#endif 2112 res = fstatvfs (fd, buf);
1502 2113 close (fd);
1503#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1504# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1505#else
1506
1507int
1508eio__msync (void *mem, size_t len, int flags)
1509{
1510 eio_page_align (&mem, &len);
1511
1512 if (EIO_MS_ASYNC != MS_SYNC
1513 || EIO_MS_INVALIDATE != MS_INVALIDATE
1514 || EIO_MS_SYNC != MS_SYNC)
1515 {
1516 flags = 0
1517 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1518 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1519 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1520 }
1521
1522 return msync (mem, len, flags);
1523}
1524
1525#endif
1526
1527int
1528eio__mtouch (void *mem, size_t len, int flags)
1529{
1530 eio_page_align (&mem, &len);
1531
1532 {
1533 intptr_t addr = (intptr_t)mem;
1534 intptr_t end = addr + len;
1535 intptr_t page = eio_pagesize ();
1536
1537 if (addr < end)
1538 if (flags & EIO_MT_MODIFY) /* modify */
1539 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len);
1540 else
1541 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len);
1542 }
1543
1544 return 0; 2114 return res;
2115
1545} 2116}
2117
2118#endif
1546 2119
1547/*****************************************************************************/ 2120/*****************************************************************************/
1548 2121
1549#define ALLOC(len) \ 2122#define ALLOC(len) \
1550 if (!req->ptr2) \ 2123 if (!req->ptr2) \
1559 req->result = -1; \ 2132 req->result = -1; \
1560 break; \ 2133 break; \
1561 } \ 2134 } \
1562 } 2135 }
1563 2136
2137static void ecb_noinline ecb_cold
2138etp_proc_init (void)
2139{
2140#if HAVE_PRCTL_SET_NAME
2141 /* provide a more sensible "thread name" */
2142 char name[16 + 1];
2143 const int namelen = sizeof (name) - 1;
2144 int len;
2145
2146 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2147 name [namelen] = 0;
2148 len = strlen (name);
2149 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2150 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2151#endif
2152}
2153
1564X_THREAD_PROC (etp_proc) 2154X_THREAD_PROC (etp_proc)
1565{ 2155{
1566 ETP_REQ *req; 2156 ETP_REQ *req;
1567 struct timespec ts; 2157 struct timespec ts;
1568 etp_worker *self = (etp_worker *)thr_arg; 2158 etp_worker *self = (etp_worker *)thr_arg;
1569 2159
2160 etp_proc_init ();
2161
1570 /* try to distribute timeouts somewhat randomly */ 2162 /* try to distribute timeouts somewhat evenly */
1571 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2163 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1572 2164
1573 for (;;) 2165 for (;;)
1574 { 2166 {
2167 ts.tv_sec = 0;
2168
1575 X_LOCK (reqlock); 2169 X_LOCK (reqlock);
1576 2170
1577 for (;;) 2171 for (;;)
1578 { 2172 {
1579 self->req = req = reqq_shift (&req_queue); 2173 req = reqq_shift (&req_queue);
1580 2174
1581 if (req) 2175 if (req)
1582 break; 2176 break;
1583 2177
2178 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2179 {
2180 X_UNLOCK (reqlock);
2181 X_LOCK (wrklock);
2182 --started;
2183 X_UNLOCK (wrklock);
2184 goto quit;
2185 }
2186
1584 ++idle; 2187 ++idle;
1585 2188
1586 ts.tv_sec = time (0) + idle_timeout; 2189 if (idle <= max_idle)
1587 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2190 /* we are allowed to idle, so do so without any timeout */
2191 X_COND_WAIT (reqwait, reqlock);
2192 else
1588 { 2193 {
1589 if (idle > max_idle) 2194 /* initialise timeout once */
1590 { 2195 if (!ts.tv_sec)
1591 --idle; 2196 ts.tv_sec = time (0) + idle_timeout;
1592 X_UNLOCK (reqlock);
1593 X_LOCK (wrklock);
1594 --started;
1595 X_UNLOCK (wrklock);
1596 goto quit;
1597 }
1598 2197
1599 /* we are allowed to idle, so do so without any timeout */
1600 X_COND_WAIT (reqwait, reqlock); 2198 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2199 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1601 } 2200 }
1602 2201
1603 --idle; 2202 --idle;
1604 } 2203 }
1605 2204
1608 X_UNLOCK (reqlock); 2207 X_UNLOCK (reqlock);
1609 2208
1610 if (req->type < 0) 2209 if (req->type < 0)
1611 goto quit; 2210 goto quit;
1612 2211
1613 if (!EIO_CANCELLED (req))
1614 ETP_EXECUTE (self, req); 2212 ETP_EXECUTE (self, req);
1615 2213
1616 X_LOCK (reslock); 2214 X_LOCK (reslock);
1617 2215
1618 ++npending; 2216 ++npending;
1619 2217
1620 if (!reqq_push (&res_queue, req) && want_poll_cb) 2218 if (!reqq_push (&res_queue, req) && want_poll_cb)
1621 want_poll_cb (); 2219 want_poll_cb ();
1622 2220
1623 self->req = 0;
1624 etp_worker_clear (self); 2221 etp_worker_clear (self);
1625 2222
1626 X_UNLOCK (reslock); 2223 X_UNLOCK (reslock);
1627 } 2224 }
1628 2225
1629quit: 2226quit:
2227 free (req);
2228
1630 X_LOCK (wrklock); 2229 X_LOCK (wrklock);
1631 etp_worker_free (self); 2230 etp_worker_free (self);
1632 X_UNLOCK (wrklock); 2231 X_UNLOCK (wrklock);
1633 2232
1634 return 0; 2233 return 0;
1635} 2234}
1636 2235
1637/*****************************************************************************/ 2236/*****************************************************************************/
1638 2237
2238int ecb_cold
1639int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2239eio_init (void (*want_poll)(void), void (*done_poll)(void))
1640{ 2240{
2241#if !HAVE_PREADWRITE
2242 X_MUTEX_CREATE (preadwritelock);
2243#endif
2244
1641 return etp_init (want_poll, done_poll); 2245 return etp_init (want_poll, done_poll);
1642} 2246}
1643 2247
2248ecb_inline void
1644static void eio_api_destroy (eio_req *req) 2249eio_api_destroy (eio_req *req)
1645{ 2250{
1646 free (req); 2251 free (req);
1647} 2252}
1648 2253
1649#define REQ(rtype) \ 2254#define REQ(rtype) \
1668 { \ 2273 { \
1669 eio_api_destroy (req); \ 2274 eio_api_destroy (req); \
1670 return 0; \ 2275 return 0; \
1671 } 2276 }
1672 2277
2278static void
1673static void eio_execute (etp_worker *self, eio_req *req) 2279eio_execute (etp_worker *self, eio_req *req)
1674{ 2280{
2281#if HAVE_AT
2282 int dirfd;
2283#else
2284 const char *path;
2285#endif
2286
2287 if (ecb_expect_false (EIO_CANCELLED (req)))
2288 {
2289 req->result = -1;
2290 req->errorno = ECANCELED;
2291 return;
2292 }
2293
2294 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2295 {
2296 req->result = -1;
2297 req->errorno = ENOENT;
2298 return;
2299 }
2300
2301 if (req->type >= EIO_OPEN)
2302 {
2303 #if HAVE_AT
2304 dirfd = WD2FD (req->wd);
2305 #else
2306 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2307 #endif
2308 }
2309
1675 switch (req->type) 2310 switch (req->type)
1676 { 2311 {
2312 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2313 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2314 break;
2315 case EIO_WD_CLOSE: req->result = 0;
2316 eio_wd_close_sync (req->wd); break;
2317
1677 case EIO_READ: ALLOC (req->size); 2318 case EIO_READ: ALLOC (req->size);
1678 req->result = req->offs >= 0 2319 req->result = req->offs >= 0
1679 ? pread (req->int1, req->ptr2, req->size, req->offs) 2320 ? pread (req->int1, req->ptr2, req->size, req->offs)
1680 : read (req->int1, req->ptr2, req->size); break; 2321 : read (req->int1, req->ptr2, req->size); break;
1681 case EIO_WRITE: req->result = req->offs >= 0 2322 case EIO_WRITE: req->result = req->offs >= 0
1682 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2323 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1683 : write (req->int1, req->ptr2, req->size); break; 2324 : write (req->int1, req->ptr2, req->size); break;
1684 2325
1685 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2326 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1686 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2327 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2328
2329#if HAVE_AT
1687 2330
1688 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2331 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1689 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2332 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1690 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2333 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2334 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2335 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2336 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2337 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2338 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2339
2340 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2341 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2342 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2343 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2344 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2345 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2346 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2347 case EIO_READLINK: ALLOC (PATH_MAX);
2348 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2349 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2350 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2351 case EIO_UTIME:
2352 case EIO_FUTIME:
2353 {
2354 struct timespec ts[2];
2355 struct timespec *times;
2356
2357 if (req->nv1 != -1. || req->nv2 != -1.)
2358 {
2359 ts[0].tv_sec = req->nv1;
2360 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2361 ts[1].tv_sec = req->nv2;
2362 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2363
2364 times = ts;
2365 }
2366 else
2367 times = 0;
2368
2369 req->result = req->type == EIO_FUTIME
2370 ? futimens (req->int1, times)
2371 : utimensat (dirfd, req->ptr1, times, 0);
2372 }
2373 break;
2374
2375#else
2376
2377 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2378 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2379 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1691 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2380 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2381 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2382 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2383 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2384 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2385
2386 case EIO_UNLINK: req->result = unlink (path ); break;
2387 case EIO_RMDIR: req->result = rmdir (path ); break;
2388 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2389 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2390 case EIO_LINK: req->result = link (path , req->ptr2); break;
2391 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2392 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2393 case EIO_READLINK: ALLOC (PATH_MAX);
2394 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2395 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2396 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2397
2398 case EIO_UTIME:
2399 case EIO_FUTIME:
2400 {
2401 struct timeval tv[2];
2402 struct timeval *times;
2403
2404 if (req->nv1 != -1. || req->nv2 != -1.)
2405 {
2406 tv[0].tv_sec = req->nv1;
2407 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2408 tv[1].tv_sec = req->nv2;
2409 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2410
2411 times = tv;
2412 }
2413 else
2414 times = 0;
2415
2416 req->result = req->type == EIO_FUTIME
2417 ? futimes (req->int1, times)
2418 : utimes (req->ptr1, times);
2419 }
2420 break;
2421
2422#endif
2423
2424 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2425 {
2426 ALLOC (req->result);
2427 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2428 }
2429 break;
2430
1692 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2431 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1693 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2432 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1694 2433
1695 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1696 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1697 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2434 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1698 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2435 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1699 2436
1700 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1701 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2437 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1702 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1703 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2438 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1704 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1705 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2439 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1706 2440
1707 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1708 case EIO_CLOSE: req->result = close (req->int1); break; 2441 case EIO_CLOSE: req->result = close (req->int1); break;
1709 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2442 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1710 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1711 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1712 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1713 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1714 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1715 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1716 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1717
1718 case EIO_READLINK: ALLOC (PATH_MAX);
1719 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1720
1721 case EIO_SYNC: req->result = 0; sync (); break; 2443 case EIO_SYNC: req->result = 0; sync (); break;
1722 case EIO_FSYNC: req->result = fsync (req->int1); break; 2444 case EIO_FSYNC: req->result = fsync (req->int1); break;
1723 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2445 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2446 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2447 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1724 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2448 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1725 case EIO_MTOUCH: req->result = eio__mtouch (req->ptr2, req->size, req->int1); break; 2449 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1726 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2450 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1727 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2451 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1728 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2452 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1729 2453
1730 case EIO_READDIR: eio__scandir (req, self); break; 2454 case EIO_READDIR: eio__scandir (req, self); break;
1731 2455
1732 case EIO_BUSY: 2456 case EIO_BUSY:
1733#ifdef _WIN32 2457#ifdef _WIN32
1742 req->result = select (0, 0, 0, 0, &tv); 2466 req->result = select (0, 0, 0, 0, &tv);
1743 } 2467 }
1744#endif 2468#endif
1745 break; 2469 break;
1746 2470
1747 case EIO_UTIME:
1748 case EIO_FUTIME:
1749 {
1750 struct timeval tv[2];
1751 struct timeval *times;
1752
1753 if (req->nv1 != -1. || req->nv2 != -1.)
1754 {
1755 tv[0].tv_sec = req->nv1;
1756 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1757 tv[1].tv_sec = req->nv2;
1758 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1759
1760 times = tv;
1761 }
1762 else
1763 times = 0;
1764
1765 req->result = req->type == EIO_FUTIME
1766 ? futimes (req->int1, times)
1767 : utimes (req->ptr1, times);
1768 }
1769 break;
1770
1771 case EIO_GROUP: 2471 case EIO_GROUP:
1772 abort (); /* handled in eio_request */ 2472 abort (); /* handled in eio_request */
1773 2473
1774 case EIO_NOP: 2474 case EIO_NOP:
1775 req->result = 0; 2475 req->result = 0;
1776 break; 2476 break;
1777 2477
1778 case EIO_CUSTOM: 2478 case EIO_CUSTOM:
1779 ((void (*)(eio_req *))req->feed) (req); 2479 req->feed (req);
1780 break; 2480 break;
1781 2481
1782 default: 2482 default:
1783 errno = ENOSYS; 2483 errno = ENOSYS;
1784 req->result = -1; 2484 req->result = -1;
1788 req->errorno = errno; 2488 req->errorno = errno;
1789} 2489}
1790 2490
1791#ifndef EIO_NO_WRAPPERS 2491#ifndef EIO_NO_WRAPPERS
1792 2492
2493eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2494{
2495 REQ (EIO_WD_OPEN); PATH; SEND;
2496}
2497
2498eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2499{
2500 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2501}
2502
1793eio_req *eio_nop (int pri, eio_cb cb, void *data) 2503eio_req *eio_nop (int pri, eio_cb cb, void *data)
1794{ 2504{
1795 REQ (EIO_NOP); SEND; 2505 REQ (EIO_NOP); SEND;
1796} 2506}
1797 2507
1813eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2523eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1814{ 2524{
1815 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2525 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1816} 2526}
1817 2527
2528eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2529{
2530 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2531}
2532
2533eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2534{
2535 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2536}
2537
2538eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2539{
2540 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2541}
2542
1818eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2543eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1819{ 2544{
1820 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2545 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1821} 2546}
1822 2547
1828eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2553eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1829{ 2554{
1830 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2555 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1831} 2556}
1832 2557
1833eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2558eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1834{ 2559{
1835 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2560 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1836}
1837
1838eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1839{
1840 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1841} 2561}
1842 2562
1843eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2563eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1844{ 2564{
1845 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2565 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1883eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2603eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1884{ 2604{
1885 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2605 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1886} 2606}
1887 2607
1888eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2608eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1889{ 2609{
1890 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2610 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1891} 2611}
1892 2612
1893eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2613eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1913eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2633eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1914{ 2634{
1915 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2635 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1916} 2636}
1917 2637
1918eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2638eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1919{ 2639{
1920 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2640 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1921} 2641}
1922 2642
1923eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2643eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1937} 2657}
1938 2658
1939eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2659eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1940{ 2660{
1941 return eio__1path (EIO_READLINK, path, pri, cb, data); 2661 return eio__1path (EIO_READLINK, path, pri, cb, data);
2662}
2663
2664eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2665{
2666 return eio__1path (EIO_REALPATH, path, pri, cb, data);
1942} 2667}
1943 2668
1944eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2669eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1945{ 2670{
1946 return eio__1path (EIO_STAT, path, pri, cb, data); 2671 return eio__1path (EIO_STAT, path, pri, cb, data);
2005eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2730eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2006{ 2731{
2007 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2732 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2008} 2733}
2009 2734
2010eio_req *eio_custom (eio_cb execute, int pri, eio_cb cb, void *data) 2735eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2011{ 2736{
2012 REQ (EIO_CUSTOM); req->feed = (void (*)(eio_req *))execute; SEND; 2737 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2013} 2738}
2014 2739
2015#endif 2740#endif
2016 2741
2017eio_req *eio_grp (eio_cb cb, void *data) 2742eio_req *eio_grp (eio_cb cb, void *data)
2026#undef SEND 2751#undef SEND
2027 2752
2028/*****************************************************************************/ 2753/*****************************************************************************/
2029/* grp functions */ 2754/* grp functions */
2030 2755
2756void
2031void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2757eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
2032{ 2758{
2033 grp->int2 = limit; 2759 grp->int2 = limit;
2034 grp->feed = feed; 2760 grp->feed = feed;
2035 2761
2036 grp_try_feed (grp); 2762 grp_try_feed (grp);
2037} 2763}
2038 2764
2765void
2039void eio_grp_limit (eio_req *grp, int limit) 2766eio_grp_limit (eio_req *grp, int limit)
2040{ 2767{
2041 grp->int2 = limit; 2768 grp->int2 = limit;
2042 2769
2043 grp_try_feed (grp); 2770 grp_try_feed (grp);
2044} 2771}
2045 2772
2773void
2046void eio_grp_add (eio_req *grp, eio_req *req) 2774eio_grp_add (eio_req *grp, eio_req *req)
2047{ 2775{
2048 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2776 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2049 2777
2050 grp->flags |= EIO_FLAG_GROUPADD; 2778 grp->flags |= EIO_FLAG_GROUPADD;
2051 2779
2062} 2790}
2063 2791
2064/*****************************************************************************/ 2792/*****************************************************************************/
2065/* misc garbage */ 2793/* misc garbage */
2066 2794
2795eio_ssize_t
2067ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2796eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2068{ 2797{
2069 etp_worker wrk;
2070 ssize_t ret;
2071
2072 wrk.dbuf = 0;
2073
2074 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2798 return eio__sendfile (ofd, ifd, offset, count);
2075
2076 if (wrk.dbuf)
2077 free (wrk.dbuf);
2078
2079 return ret;
2080} 2799}
2081 2800

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines