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

Comparing libeio/eio.c (file contents):
Revision 1.72 by root, Fri Jun 10 12:45:20 2011 UTC vs.
Revision 1.107 by root, Tue Sep 27 00:41:52 2011 UTC

43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h" 45#include "ecb.h"
46 46
47#ifdef EIO_STACKSIZE 47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE 48# define X_STACKSIZE EIO_STACKSIZE
49#endif 49#endif
50#include "xthread.h" 50#include "xthread.h"
51 51
52#include <errno.h> 52#include <errno.h>
53#include <stddef.h> 53#include <stddef.h>
54#include <stdlib.h> 54#include <stdlib.h>
55#include <string.h> 55#include <string.h>
56#include <errno.h> 56#include <errno.h>
57#include <sys/types.h> 57#include <sys/types.h>
58#include <sys/stat.h> 58#include <sys/stat.h>
59#include <sys/statvfs.h>
60#include <limits.h> 59#include <limits.h>
61#include <fcntl.h> 60#include <fcntl.h>
62#include <assert.h> 61#include <assert.h>
63 62
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 63/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 64/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H 65#if HAVE_STDINT_H
67# include <stdint.h> 66# include <stdint.h>
68#endif 67#endif
69 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
70#ifndef EIO_FINISH 82#ifndef EIO_FINISH
71# 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
72#endif 84#endif
73 85
74#ifndef EIO_DESTROY 86#ifndef EIO_DESTROY
77 89
78#ifndef EIO_FEED 90#ifndef EIO_FEED
79# 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)
80#endif 92#endif
81 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
82#ifdef _WIN32 105#ifdef _WIN32
83 106
84 /*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
85#else 197#else
86 198
87# include <sys/time.h> 199 #include <sys/time.h>
88# include <sys/select.h> 200 #include <sys/select.h>
201 #include <sys/statvfs.h>
89# 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
90# include <utime.h> 237# include <utime.h>
91# include <signal.h>
92# include <dirent.h>
93
94#if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
95# include <sys/mman.h>
96#endif
97
98/* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
99# if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
100# define _DIRENT_HAVE_D_TYPE /* sigh */
101# define D_INO(de) (de)->d_fileno
102# define D_NAMLEN(de) (de)->d_namlen
103# elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
104# define D_INO(de) (de)->d_ino
105# endif 238#endif
106 239
107#ifdef _D_EXACT_NAMLEN 240#if HAVE_SYS_SYSCALL_H
108# undef D_NAMLEN 241# include <sys/syscall.h>
109# define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
110#endif
111
112# ifdef _DIRENT_HAVE_D_TYPE
113# define D_TYPE(de) (de)->d_type
114# endif 242#endif
115 243
116# ifndef EIO_STRUCT_DIRENT 244#if HAVE_SYS_PRCTL_H
117# define EIO_STRUCT_DIRENT struct dirent 245# include <sys/prctl.h>
118# endif
119
120#endif 246#endif
121 247
122#if HAVE_SENDFILE 248#if HAVE_SENDFILE
123# if __linux 249# if __linux
124# include <sys/sendfile.h> 250# include <sys/sendfile.h>
139#endif 265#endif
140#ifndef D_INO 266#ifndef D_INO
141# define D_INO(de) 0 267# define D_INO(de) 0
142#endif 268#endif
143#ifndef D_NAMLEN 269#ifndef D_NAMLEN
144# define D_NAMLEN(de) strlen ((de)->d_name) 270# define D_NAMLEN(entp) strlen (D_NAME (entp))
145#endif 271#endif
146 272
147/* used for struct dirent, AIX doesn't provide it */ 273/* used for struct dirent, AIX doesn't provide it */
148#ifndef NAME_MAX 274#ifndef NAME_MAX
149# define NAME_MAX 4096 275# define NAME_MAX 4096
156 282
157/* buffer size for various temporary buffers */ 283/* buffer size for various temporary buffers */
158#define EIO_BUFSIZE 65536 284#define EIO_BUFSIZE 65536
159 285
160#define dBUF \ 286#define dBUF \
161 char *eio_buf; \
162 ETP_WORKER_LOCK (self); \
163 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 287 char *eio_buf = malloc (EIO_BUFSIZE); \
164 ETP_WORKER_UNLOCK (self); \
165 errno = ENOMEM; \ 288 errno = ENOMEM; \
166 if (!eio_buf) \ 289 if (!eio_buf) \
167 return -1; 290 return -1
291
292#define FUBd \
293 free (eio_buf)
168 294
169#define EIO_TICKS ((1000000 + 1023) >> 10) 295#define EIO_TICKS ((1000000 + 1023) >> 10)
296
297/*****************************************************************************/
298
299struct tmpbuf
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
325#else
326 #define HAVE_AT 0
327 static const char *wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path);
328#endif
329
330struct eio_pwd
331{
332#if HAVE_AT
333 int fd;
334#endif
335 int len;
336 char str[1]; /* actually, a 0-terminated canonical path */
337};
338
339/*****************************************************************************/
170 340
171#define ETP_PRI_MIN EIO_PRI_MIN 341#define ETP_PRI_MIN EIO_PRI_MIN
172#define ETP_PRI_MAX EIO_PRI_MAX 342#define ETP_PRI_MAX EIO_PRI_MAX
173 343
174struct etp_worker; 344struct etp_worker;
178static int eio_finish (eio_req *req); 348static int eio_finish (eio_req *req);
179#define ETP_FINISH(req) eio_finish (req) 349#define ETP_FINISH(req) eio_finish (req)
180static void eio_execute (struct etp_worker *self, eio_req *req); 350static void eio_execute (struct etp_worker *self, eio_req *req);
181#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 351#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
182 352
183#define ETP_WORKER_CLEAR(req) \
184 if (wrk->dbuf) \
185 { \
186 free (wrk->dbuf); \
187 wrk->dbuf = 0; \
188 } \
189 \
190 if (wrk->dirp) \
191 { \
192 closedir (wrk->dirp); \
193 wrk->dirp = 0; \
194 }
195
196#define ETP_WORKER_COMMON \
197 void *dbuf; \
198 DIR *dirp;
199
200/*****************************************************************************/ 353/*****************************************************************************/
201 354
202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 355#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203 356
204/* calculate time difference in ~1/EIO_TICKS of a second */ 357/* calculate time difference in ~1/EIO_TICKS of a second */
205ECB_INLINE int 358ecb_inline int
206tvdiff (struct timeval *tv1, struct timeval *tv2) 359tvdiff (struct timeval *tv1, struct timeval *tv2)
207{ 360{
208 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 361 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
209 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 362 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
210} 363}
215static void (*done_poll_cb) (void); 368static void (*done_poll_cb) (void);
216 369
217static unsigned int max_poll_time; /* reslock */ 370static unsigned int max_poll_time; /* reslock */
218static unsigned int max_poll_reqs; /* reslock */ 371static unsigned int max_poll_reqs; /* reslock */
219 372
220static volatile unsigned int nreqs; /* reqlock */ 373static unsigned int nreqs; /* reqlock */
221static volatile unsigned int nready; /* reqlock */ 374static unsigned int nready; /* reqlock */
222static volatile unsigned int npending; /* reqlock */ 375static unsigned int npending; /* reqlock */
223static 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 */
224static 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 */
225 378
226static xmutex_t wrklock; 379static xmutex_t wrklock;
227static xmutex_t reslock; 380static xmutex_t reslock;
228static xmutex_t reqlock; 381static xmutex_t reqlock;
229static xcond_t reqwait; 382static xcond_t reqwait;
232/* 385/*
233 * make our pread/pwrite emulation safe against themselves, but not against 386 * make our pread/pwrite emulation safe against themselves, but not against
234 * 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,
235 * but that's your problem, not mine. 388 * but that's your problem, not mine.
236 */ 389 */
237static xmutex_t preadwritelock = X_MUTEX_INIT; 390static xmutex_t preadwritelock;
238#endif 391#endif
239 392
240typedef struct etp_worker 393typedef struct etp_worker
241{ 394{
395 struct tmpbuf tmpbuf;
396
242 /* locked by wrklock */ 397 /* locked by wrklock */
243 struct etp_worker *prev, *next; 398 struct etp_worker *prev, *next;
244 399
245 xthread_t tid; 400 xthread_t tid;
246 401
247 /* locked by reslock, reqlock or wrklock */ 402#ifdef ETP_WORKER_COMMON
248 ETP_REQ *req; /* currently processed request */
249
250 ETP_WORKER_COMMON 403 ETP_WORKER_COMMON
404#endif
251} etp_worker; 405} etp_worker;
252 406
253static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 407static etp_worker wrk_first; /* NOT etp */
254 408
255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 409#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 410#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
257 411
258/* worker threads management */ 412/* worker threads management */
259 413
260static void ecb_cold 414static void ecb_cold
261etp_worker_clear (etp_worker *wrk) 415etp_worker_clear (etp_worker *wrk)
262{ 416{
263 ETP_WORKER_CLEAR (wrk);
264} 417}
265 418
266static void ecb_cold 419static void ecb_cold
267etp_worker_free (etp_worker *wrk) 420etp_worker_free (etp_worker *wrk)
268{ 421{
422 free (wrk->tmpbuf.ptr);
423
269 wrk->next->prev = wrk->prev; 424 wrk->next->prev = wrk->prev;
270 wrk->prev->next = wrk->next; 425 wrk->prev->next = wrk->next;
271 426
272 free (wrk); 427 free (wrk);
273} 428}
329} etp_reqq; 484} etp_reqq;
330 485
331static etp_reqq req_queue; 486static etp_reqq req_queue;
332static etp_reqq res_queue; 487static etp_reqq res_queue;
333 488
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
334static int ecb_noinline 500static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req) 501reqq_push (etp_reqq *q, ETP_REQ *req)
336{ 502{
337 int pri = req->pri; 503 int pri = req->pri;
338 req->next = 0; 504 req->next = 0;
372 } 538 }
373 539
374 abort (); 540 abort ();
375} 541}
376 542
377static void ecb_cold 543static int ecb_cold
378etp_thread_init (void) 544etp_init (void (*want_poll)(void), void (*done_poll)(void))
379{ 545{
380 X_MUTEX_CREATE (wrklock); 546 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock); 547 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock); 548 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait); 549 X_COND_CREATE (reqwait);
384}
385 550
386static void ecb_cold 551 reqq_init (&req_queue);
387etp_atfork_prepare (void) 552 reqq_init (&res_queue);
388{
389 X_LOCK (wrklock);
390 X_LOCK (reqlock);
391 X_LOCK (reslock);
392#if !HAVE_PREADWRITE
393 X_LOCK (preadwritelock);
394#endif
395}
396 553
397static void ecb_cold 554 wrk_first.next =
398etp_atfork_parent (void) 555 wrk_first.prev = &wrk_first;
399{
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 ecb_cold
409etp_atfork_child (void)
410{
411 ETP_REQ *prv;
412
413 while ((prv = reqq_shift (&req_queue)))
414 ETP_DESTROY (prv);
415
416 while ((prv = reqq_shift (&res_queue)))
417 ETP_DESTROY (prv);
418
419 while (wrk_first.next != &wrk_first)
420 {
421 etp_worker *wrk = wrk_first.next;
422
423 if (wrk->req)
424 ETP_DESTROY (wrk->req);
425
426 etp_worker_clear (wrk);
427 etp_worker_free (wrk);
428 }
429 556
430 started = 0; 557 started = 0;
431 idle = 0; 558 idle = 0;
432 nreqs = 0; 559 nreqs = 0;
433 nready = 0; 560 nready = 0;
434 npending = 0; 561 npending = 0;
435
436 etp_thread_init ();
437}
438
439static void ecb_cold
440etp_once_init (void)
441{
442 etp_thread_init ();
443 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
444}
445
446static int ecb_cold
447etp_init (void (*want_poll)(void), void (*done_poll)(void))
448{
449 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
450
451 pthread_once (&doinit, etp_once_init);
452 562
453 want_poll_cb = want_poll; 563 want_poll_cb = want_poll;
454 done_poll_cb = done_poll; 564 done_poll_cb = done_poll;
455 565
456 return 0; 566 return 0;
496} 606}
497 607
498static void ecb_cold 608static void ecb_cold
499etp_end_thread (void) 609etp_end_thread (void)
500{ 610{
501 eio_req *req = calloc (1, sizeof (eio_req)); 611 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
502 612
503 req->type = -1; 613 req->type = -1;
504 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 614 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
505 615
506 X_LOCK (reqlock); 616 X_LOCK (reqlock);
583} 693}
584 694
585static void 695static void
586etp_cancel (ETP_REQ *req) 696etp_cancel (ETP_REQ *req)
587{ 697{
588 X_LOCK (wrklock); 698 req->cancelled = 1;
589 req->flags |= EIO_FLAG_CANCELLED;
590 X_UNLOCK (wrklock);
591 699
592 eio_grp_cancel (req); 700 eio_grp_cancel (req);
593} 701}
594 702
595static void 703static void
711 return eio_finish (grp); 819 return eio_finish (grp);
712 else 820 else
713 return 0; 821 return 0;
714} 822}
715 823
716void 824static void
717eio_destroy (eio_req *req) 825eio_destroy (eio_req *req)
718{ 826{
719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 827 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 828 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721 829
739 if (grp->grp_first == req) 847 if (grp->grp_first == req)
740 grp->grp_first = req->grp_next; 848 grp->grp_first = req->grp_next;
741 849
742 res2 = grp_dec (grp); 850 res2 = grp_dec (grp);
743 851
744 if (!res && res2) 852 if (!res)
745 res = res2; 853 res = res2;
746 } 854 }
747 855
748 eio_destroy (req); 856 eio_destroy (req);
749 857
841# undef pread 949# undef pread
842# undef pwrite 950# undef pwrite
843# define pread eio__pread 951# define pread eio__pread
844# define pwrite eio__pwrite 952# define pwrite eio__pwrite
845 953
846static ssize_t 954static eio_ssize_t
847eio__pread (int fd, void *buf, size_t count, off_t offset) 955eio__pread (int fd, void *buf, size_t count, off_t offset)
848{ 956{
849 ssize_t res; 957 eio_ssize_t res;
850 off_t ooffset; 958 off_t ooffset;
851 959
852 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
853 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
854 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
857 X_UNLOCK (preadwritelock); 965 X_UNLOCK (preadwritelock);
858 966
859 return res; 967 return res;
860} 968}
861 969
862static ssize_t 970static eio_ssize_t
863eio__pwrite (int fd, void *buf, size_t count, off_t offset) 971eio__pwrite (int fd, void *buf, size_t count, off_t offset)
864{ 972{
865 ssize_t res; 973 eio_ssize_t res;
866 off_t ooffset; 974 off_t ooffset;
867 975
868 X_LOCK (preadwritelock); 976 X_LOCK (preadwritelock);
869 ooffset = lseek (fd, 0, SEEK_CUR); 977 ooffset = lseek (fd, 0, SEEK_CUR);
870 lseek (fd, offset, SEEK_SET); 978 lseek (fd, offset, SEEK_SET);
915 1023
916#if !HAVE_FDATASYNC 1024#if !HAVE_FDATASYNC
917# undef fdatasync 1025# undef fdatasync
918# define fdatasync(fd) fsync (fd) 1026# define fdatasync(fd) fsync (fd)
919#endif 1027#endif
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}
920 1046
921/* sync_file_range always needs emulation */ 1047/* sync_file_range always needs emulation */
922static int 1048static int
923eio__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)
924{ 1050{
944 /* 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
945 * call fdatasync, as that matches the expectation of its users best */ 1071 * call fdatasync, as that matches the expectation of its users best */
946 return fdatasync (fd); 1072 return fdatasync (fd);
947} 1073}
948 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
949#if !HAVE_READAHEAD 1086#if !HAVE_READAHEAD
950# undef readahead 1087# undef readahead
951# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1088# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
952 1089
953static ssize_t 1090static eio_ssize_t
954eio__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)
955{ 1092{
956 size_t todo = count; 1093 size_t todo = count;
957 dBUF; 1094 dBUF;
958 1095
963 pread (fd, eio_buf, len, offset); 1100 pread (fd, eio_buf, len, offset);
964 offset += len; 1101 offset += len;
965 todo -= len; 1102 todo -= len;
966 } 1103 }
967 1104
1105 FUBd;
1106
968 errno = 0; 1107 errno = 0;
969 return count; 1108 return count;
970} 1109}
971 1110
972#endif 1111#endif
973 1112
974/* sendfile always needs emulation */ 1113/* sendfile always needs emulation */
975static ssize_t 1114static eio_ssize_t
976eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1115eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
977{ 1116{
978 ssize_t written = 0; 1117 eio_ssize_t written = 0;
979 ssize_t res; 1118 eio_ssize_t res;
980 1119
981 if (!count) 1120 if (!count)
982 return 0; 1121 return 0;
983 1122
984 for (;;) 1123 for (;;)
985 { 1124 {
1125#ifdef __APPLE__
1126# undef HAVE_SENDFILE /* broken, as everything on os x */
1127#endif
986#if HAVE_SENDFILE 1128#if HAVE_SENDFILE
987# if __linux 1129# if __linux
988 off_t soffset = offset; 1130 off_t soffset = offset;
989 res = sendfile (ofd, ifd, &soffset, count); 1131 res = sendfile (ofd, ifd, &soffset, count);
990 1132
1033 if (res < 0 && sbytes) 1175 if (res < 0 && sbytes)
1034 res = sbytes; 1176 res = sbytes;
1035 1177
1036# endif 1178# endif
1037 1179
1038#elif defined (_WIN32) 1180#elif defined (_WIN32) && 0
1039 /* does not work, just for documentation of what would need to be done */ 1181 /* does not work, just for documentation of what would need to be done */
1040 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 1182 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1041 /* libeio guarantees that the file offset does not change, and windows */ 1183 /* libeio guarantees that the file offset does not change, and windows */
1042 /* has no way to get an independent handle to the same file description */ 1184 /* has no way to get an independent handle to the same file description */
1043 HANDLE h = TO_SOCKET (ifd); 1185 HANDLE h = TO_SOCKET (ifd);
1076 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1218 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1077 /* BSDs */ 1219 /* BSDs */
1078#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1220#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1079 || errno == ENOTSUP 1221 || errno == ENOTSUP
1080#endif 1222#endif
1223#ifdef EOPNOTSUPP /* windows */
1081 || errno == EOPNOTSUPP /* BSDs */ 1224 || errno == EOPNOTSUPP /* BSDs */
1225#endif
1082#if __solaris 1226#if __solaris
1083 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1227 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1084#endif 1228#endif
1085 ) 1229 )
1086 ) 1230 )
1090 1234
1091 res = 0; 1235 res = 0;
1092 1236
1093 while (count) 1237 while (count)
1094 { 1238 {
1095 ssize_t cnt; 1239 eio_ssize_t cnt;
1096 1240
1097 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1241 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1098 1242
1099 if (cnt <= 0) 1243 if (cnt <= 0)
1100 { 1244 {
1112 1256
1113 offset += cnt; 1257 offset += cnt;
1114 res += cnt; 1258 res += cnt;
1115 count -= cnt; 1259 count -= cnt;
1116 } 1260 }
1261
1262 FUBd;
1117 } 1263 }
1118 1264
1119 return res; 1265 return res;
1266}
1267
1268#ifdef PAGESIZE
1269# define eio_pagesize() PAGESIZE
1270#else
1271static intptr_t
1272eio_pagesize (void)
1273{
1274 static intptr_t page;
1275
1276 if (!page)
1277 page = sysconf (_SC_PAGESIZE);
1278
1279 return page;
1280}
1281#endif
1282
1283static void
1284eio_page_align (void **addr, size_t *length)
1285{
1286 intptr_t mask = eio_pagesize () - 1;
1287
1288 /* round down addr */
1289 intptr_t adj = mask & (intptr_t)*addr;
1290
1291 *addr = (void *)((intptr_t)*addr - adj);
1292 *length += adj;
1293
1294 /* round up length */
1295 *length = (*length + mask) & ~mask;
1296}
1297
1298#if !_POSIX_MEMLOCK
1299# define eio__mlockall(a) EIO_ENOSYS ()
1300#else
1301
1302static int
1303eio__mlockall (int flags)
1304{
1305 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1306 extern int mallopt (int, int);
1307 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1308 #endif
1309
1310 if (EIO_MCL_CURRENT != MCL_CURRENT
1311 || EIO_MCL_FUTURE != MCL_FUTURE)
1312 {
1313 flags = 0
1314 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1315 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1316 }
1317
1318 return mlockall (flags);
1319}
1320#endif
1321
1322#if !_POSIX_MEMLOCK_RANGE
1323# define eio__mlock(a,b) EIO_ENOSYS ()
1324#else
1325
1326static int
1327eio__mlock (void *addr, size_t length)
1328{
1329 eio_page_align (&addr, &length);
1330
1331 return mlock (addr, length);
1332}
1333
1334#endif
1335
1336#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1337# define eio__msync(a,b,c) EIO_ENOSYS ()
1338#else
1339
1340static int
1341eio__msync (void *mem, size_t len, int flags)
1342{
1343 eio_page_align (&mem, &len);
1344
1345 if (EIO_MS_ASYNC != MS_SYNC
1346 || EIO_MS_INVALIDATE != MS_INVALIDATE
1347 || EIO_MS_SYNC != MS_SYNC)
1348 {
1349 flags = 0
1350 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1351 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1352 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1353 }
1354
1355 return msync (mem, len, flags);
1356}
1357
1358#endif
1359
1360static int
1361eio__mtouch (eio_req *req)
1362{
1363 void *mem = req->ptr2;
1364 size_t len = req->size;
1365 int flags = req->int1;
1366
1367 eio_page_align (&mem, &len);
1368
1369 {
1370 intptr_t addr = (intptr_t)mem;
1371 intptr_t end = addr + len;
1372 intptr_t page = eio_pagesize ();
1373
1374 if (addr < end)
1375 if (flags & EIO_MT_MODIFY) /* modify */
1376 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1377 else
1378 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1379 }
1380
1381 return 0;
1382}
1383
1384/*****************************************************************************/
1385/* requests implemented outside eio_execute, because they are so large */
1386
1387/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1388static int
1389eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1390{
1391 const char *rel = path;
1392 char *res;
1393 char *tmp1, *tmp2;
1394#if SYMLOOP_MAX > 32
1395 int symlinks = SYMLOOP_MAX;
1396#else
1397 int symlinks = 32;
1398#endif
1399
1400 errno = EINVAL;
1401 if (!rel)
1402 return -1;
1403
1404 errno = ENOENT;
1405 if (!*rel)
1406 return -1;
1407
1408 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1409 tmp1 = res + PATH_MAX;
1410 tmp2 = tmp1 + PATH_MAX;
1411
1412#if 0 /* disabled, the musl way to do things is just too racy */
1413#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1414 /* on linux we may be able to ask the kernel */
1415 {
1416 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1417
1418 if (fd >= 0)
1419 {
1420 sprintf (tmp1, "/proc/self/fd/%d", fd);
1421 req->result = readlink (tmp1, res, PATH_MAX);
1422 close (fd);
1423
1424 /* here we should probably stat the open file and the disk file, to make sure they still match */
1425
1426 if (req->result > 0)
1427 goto done;
1428 }
1429 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1430 return;
1431 }
1432#endif
1433#endif
1434
1435 if (*rel != '/')
1436 {
1437 int len;
1438
1439 errno = ENOENT;
1440 if (wd == EIO_INVALID_WD)
1441 return -1;
1442
1443 if (wd == EIO_CWD)
1444 {
1445 if (!getcwd (res, PATH_MAX))
1446 return -1;
1447
1448 len = strlen (res);
1449 }
1450 else
1451 memcpy (res, wd->str, len = wd->len);
1452
1453 if (res [1]) /* only use if not / */
1454 res += len;
1455 }
1456
1457 while (*rel)
1458 {
1459 eio_ssize_t len, linklen;
1460 const char *beg = rel;
1461
1462 while (*rel && *rel != '/')
1463 ++rel;
1464
1465 len = rel - beg;
1466
1467 if (!len) /* skip slashes */
1468 {
1469 ++rel;
1470 continue;
1471 }
1472
1473 if (beg [0] == '.')
1474 {
1475 if (len == 1)
1476 continue; /* . - nop */
1477
1478 if (beg [1] == '.' && len == 2)
1479 {
1480 /* .. - back up one component, if possible */
1481
1482 while (res != tmpbuf->ptr)
1483 if (*--res == '/')
1484 break;
1485
1486 continue;
1487 }
1488 }
1489
1490 errno = ENAMETOOLONG;
1491 if (res + 1 + len + 1 >= tmp1)
1492 return;
1493
1494 /* copy one component */
1495 *res = '/';
1496 memcpy (res + 1, beg, len);
1497
1498 /* zero-terminate, for readlink */
1499 res [len + 1] = 0;
1500
1501 /* now check if it's a symlink */
1502 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1503
1504 if (linklen < 0)
1505 {
1506 if (errno != EINVAL)
1507 return -1;
1508
1509 /* it's a normal directory. hopefully */
1510 res += len + 1;
1511 }
1512 else
1513 {
1514 /* yay, it was a symlink - build new path in tmp2 */
1515 int rellen = strlen (rel);
1516
1517 errno = ENAMETOOLONG;
1518 if (linklen + 1 + rellen >= PATH_MAX)
1519 return -1;
1520
1521 errno = ELOOP;
1522 if (!--symlinks)
1523 return -1;
1524
1525 if (*tmp1 == '/')
1526 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1527
1528 /* we need to be careful, as rel might point into tmp2 already */
1529 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1530 tmp2 [linklen] = '/';
1531 memcpy (tmp2, tmp1, linklen);
1532
1533 rel = tmp2;
1534 }
1535 }
1536
1537 /* special case for the lone root path */
1538 if (res == tmpbuf->ptr)
1539 *res++ = '/';
1540
1541 return res - (char *)tmpbuf->ptr;
1120} 1542}
1121 1543
1122static signed char 1544static signed char
1123eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1545eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1124{ 1546{
1132 1554
1133#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1555#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1134#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1556#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1135 1557
1136static void 1558static void
1137eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1559eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1138{ 1560{
1139 unsigned char bits [9 + sizeof (ino_t) * 8]; 1561 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1140 unsigned char *bit = bits; 1562 unsigned char *bit = bits;
1141 1563
1142 assert (CHAR_BIT == 8); 1564 assert (CHAR_BIT == 8);
1143 assert (sizeof (eio_dirent) * 8 < 256); 1565 assert (sizeof (eio_dirent) * 8 < 256);
1144 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1566 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1146 1568
1147 if (size <= EIO_SORT_FAST) 1569 if (size <= EIO_SORT_FAST)
1148 return; 1570 return;
1149 1571
1150 /* first prepare an array of bits to test in our radix sort */ 1572 /* first prepare an array of bits to test in our radix sort */
1151 /* try to take endianness into account, as well as differences in ino_t sizes */ 1573 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1152 /* inode_bits must contain all inodes ORed together */ 1574 /* inode_bits must contain all inodes ORed together */
1153 /* which is used to skip bits that are 0 everywhere, which is very common */ 1575 /* which is used to skip bits that are 0 everywhere, which is very common */
1154 { 1576 {
1155 ino_t endianness; 1577 eio_ino_t endianness;
1156 int i, j; 1578 int i, j;
1157 1579
1158 /* we store the byte offset of byte n into byte n of "endianness" */ 1580 /* we store the byte offset of byte n into byte n of "endianness" */
1159 for (i = 0; i < sizeof (ino_t); ++i) 1581 for (i = 0; i < sizeof (eio_ino_t); ++i)
1160 ((unsigned char *)&endianness)[i] = i; 1582 ((unsigned char *)&endianness)[i] = i;
1161 1583
1162 *bit++ = 0; 1584 *bit++ = 0;
1163 1585
1164 for (i = 0; i < sizeof (ino_t); ++i) 1586 for (i = 0; i < sizeof (eio_ino_t); ++i)
1165 { 1587 {
1166 /* shifting off the byte offsets out of "endianness" */ 1588 /* shifting off the byte offsets out of "endianness" */
1167 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1589 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1168 endianness >>= 8; 1590 endianness >>= 8;
1169 1591
1170 for (j = 0; j < 8; ++j) 1592 for (j = 0; j < 8; ++j)
1171 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1593 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1172 *bit++ = offs + j; 1594 *bit++ = offs + j;
1173 } 1595 }
1174 1596
1175 for (j = 0; j < 8; ++j) 1597 for (j = 0; j < 8; ++j)
1176 if (score_bits & (1 << j)) 1598 if (score_bits & (1 << j))
1177 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1599 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1178 } 1600 }
1179 1601
1180 /* now actually do the sorting (a variant of MSD radix sort) */ 1602 /* now actually do the sorting (a variant of MSD radix sort) */
1181 { 1603 {
1182 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1604 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1183 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1605 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1184 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1606 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1185 int stk_idx = 0; 1607 int stk_idx = 0;
1186 1608
1187 base_stk [stk_idx] = dents; 1609 base_stk [stk_idx] = dents;
1188 end_stk [stk_idx] = dents + size; 1610 end_stk [stk_idx] = dents + size;
1189 bit_stk [stk_idx] = bit - 1; 1611 bit_stk [stk_idx] = bit - 1;
1268 } 1690 }
1269 } 1691 }
1270} 1692}
1271 1693
1272static void 1694static void
1273eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1695eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1274{ 1696{
1275 if (size <= 1) 1697 if (size <= 1)
1276 return; /* our insertion sort relies on size > 0 */ 1698 return; /* our insertion sort relies on size > 0 */
1277 1699
1278 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1700 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1286 1708
1287/* read a full directory */ 1709/* read a full directory */
1288static void 1710static void
1289eio__scandir (eio_req *req, etp_worker *self) 1711eio__scandir (eio_req *req, etp_worker *self)
1290{ 1712{
1291 DIR *dirp;
1292 EIO_STRUCT_DIRENT *entp;
1293 char *name, *names; 1713 char *name, *names;
1294 int namesalloc = 4096; 1714 int namesalloc = 4096 - sizeof (void *) * 4;
1295 int namesoffs = 0; 1715 int namesoffs = 0;
1296 int flags = req->int1; 1716 int flags = req->int1;
1297 eio_dirent *dents = 0; 1717 eio_dirent *dents = 0;
1298 int dentalloc = 128; 1718 int dentalloc = 128;
1299 int dentoffs = 0; 1719 int dentoffs = 0;
1300 ino_t inode_bits = 0; 1720 eio_ino_t inode_bits = 0;
1721#ifdef _WIN32
1722 HANDLE dirp;
1723 WIN32_FIND_DATA entp;
1724#else
1725 DIR *dirp;
1726 EIO_STRUCT_DIRENT *entp;
1727#endif
1301 1728
1302 req->result = -1; 1729 req->result = -1;
1303 1730
1304 if (!(flags & EIO_READDIR_DENTS)) 1731 if (!(flags & EIO_READDIR_DENTS))
1305 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1732 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1306 1733
1307 X_LOCK (wrklock); 1734#ifdef _WIN32
1308 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1735 {
1736 int len = strlen ((const char *)req->ptr1);
1737 char *path = malloc (MAX_PATH);
1738 const char *fmt;
1739 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1740
1741 if (!len)
1742 fmt = "./*";
1743 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1744 fmt = "%s*";
1745 else
1746 fmt = "%s/*";
1747
1748 _snprintf (path, MAX_PATH, fmt, reqpath);
1749 dirp = FindFirstFile (path, &entp);
1750 free (path);
1751
1752 if (dirp == INVALID_HANDLE_VALUE)
1753 {
1754 /* should steal _dosmaperr */
1755 switch (GetLastError ())
1756 {
1757 case ERROR_FILE_NOT_FOUND:
1758 req->result = 0;
1759 break;
1760
1761 case ERROR_INVALID_NAME:
1762 case ERROR_PATH_NOT_FOUND:
1763 case ERROR_NO_MORE_FILES:
1764 errno = ENOENT;
1765 break;
1766
1767 case ERROR_NOT_ENOUGH_MEMORY:
1768 errno = ENOMEM;
1769 break;
1770
1771 default:
1772 errno = EINVAL;
1773 break;
1774 }
1775
1776 return;
1777 }
1778 }
1779#else
1780 #if HAVE_AT
1781 if (req->wd)
1782 {
1783 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1784
1785 if (fd < 0)
1786 return;
1787
1788 dirp = fdopendir (fd);
1789 }
1790 else
1309 self->dirp = dirp = opendir (req->ptr1); 1791 dirp = opendir (req->ptr1);
1792 #else
1793 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1794 #endif
1795
1796 if (!dirp)
1797 return;
1798#endif
1799
1800 if (req->flags & EIO_FLAG_PTR1_FREE)
1801 free (req->ptr1);
1310 1802
1311 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1803 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1312 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1804 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1313 req->ptr2 = names = malloc (namesalloc); 1805 req->ptr2 = names = malloc (namesalloc);
1314 X_UNLOCK (wrklock);
1315 1806
1316 if (dirp && names && (!flags || dents)) 1807 if (!names || (flags && !dents))
1808 return;
1809
1317 for (;;) 1810 for (;;)
1318 { 1811 {
1812 int done;
1813
1814#ifdef _WIN32
1815 done = !dirp;
1816#else
1319 errno = 0; 1817 errno = 0;
1320 entp = readdir (dirp); 1818 entp = readdir (dirp);
1819 done = !entp;
1820#endif
1321 1821
1322 if (!entp) 1822 if (done)
1323 { 1823 {
1824#ifndef _WIN32
1825 int old_errno = errno;
1826 closedir (dirp);
1827 errno = old_errno;
1828
1324 if (errno) 1829 if (errno)
1325 break; 1830 break;
1831#endif
1326 1832
1327 /* sort etc. */ 1833 /* sort etc. */
1328 req->int1 = flags; 1834 req->int1 = flags;
1329 req->result = dentoffs; 1835 req->result = dentoffs;
1330 1836
1331 if (flags & EIO_READDIR_STAT_ORDER) 1837 if (flags & EIO_READDIR_STAT_ORDER)
1332 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1838 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1333 else if (flags & EIO_READDIR_DIRS_FIRST) 1839 else if (flags & EIO_READDIR_DIRS_FIRST)
1334 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1840 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1335 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1841 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1336 else 1842 else
1843 {
1844 /* in this case, all is known, and we just put dirs first and sort them */
1845 eio_dirent *oth = dents + dentoffs;
1846 eio_dirent *dir = dents;
1847
1848 /* now partition dirs to the front, and non-dirs to the back */
1849 /* by walking from both sides and swapping if necessary */
1850 while (oth > dir)
1851 {
1852 if (dir->type == EIO_DT_DIR)
1853 ++dir;
1854 else if ((--oth)->type == EIO_DT_DIR)
1855 {
1856 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1857
1858 ++dir;
1859 }
1860 }
1861
1862 /* now sort the dirs only (dirs all have the same score) */
1863 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1864 }
1865
1866 break;
1867 }
1868
1869 /* now add the entry to our list(s) */
1870 name = D_NAME (entp);
1871
1872 /* skip . and .. entries */
1873 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1874 {
1875 int len = D_NAMLEN (entp) + 1;
1876
1877 while (ecb_expect_false (namesoffs + len > namesalloc))
1878 {
1879 namesalloc *= 2;
1880 req->ptr2 = names = realloc (names, namesalloc);
1881
1882 if (!names)
1883 break;
1884 }
1885
1886 memcpy (names + namesoffs, name, len);
1887
1888 if (dents)
1889 {
1890 struct eio_dirent *ent;
1891
1892 if (ecb_expect_false (dentoffs == dentalloc))
1337 { 1893 {
1338 /* in this case, all is known, and we just put dirs first and sort them */ 1894 dentalloc *= 2;
1895 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1896
1897 if (!dents)
1898 break;
1899 }
1900
1339 eio_dirent *oth = dents + dentoffs; 1901 ent = dents + dentoffs;
1340 eio_dirent *dir = dents;
1341 1902
1342 /* now partition dirs to the front, and non-dirs to the back */ 1903 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1343 /* by walking from both sides and swapping if necessary */ 1904 ent->namelen = len - 1;
1344 while (oth > dir) 1905 ent->inode = D_INO (entp);
1906
1907 inode_bits |= ent->inode;
1908
1909 switch (D_TYPE (entp))
1910 {
1911 default:
1912 ent->type = EIO_DT_UNKNOWN;
1913 flags |= EIO_READDIR_FOUND_UNKNOWN;
1914 break;
1915
1916 #ifdef DT_FIFO
1917 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1918 #endif
1919 #ifdef DT_CHR
1920 case DT_CHR: ent->type = EIO_DT_CHR; break;
1921 #endif
1922 #ifdef DT_MPC
1923 case DT_MPC: ent->type = EIO_DT_MPC; break;
1924 #endif
1925 #ifdef DT_DIR
1926 case DT_DIR: ent->type = EIO_DT_DIR; break;
1927 #endif
1928 #ifdef DT_NAM
1929 case DT_NAM: ent->type = EIO_DT_NAM; break;
1930 #endif
1931 #ifdef DT_BLK
1932 case DT_BLK: ent->type = EIO_DT_BLK; break;
1933 #endif
1934 #ifdef DT_MPB
1935 case DT_MPB: ent->type = EIO_DT_MPB; break;
1936 #endif
1937 #ifdef DT_REG
1938 case DT_REG: ent->type = EIO_DT_REG; break;
1939 #endif
1940 #ifdef DT_NWK
1941 case DT_NWK: ent->type = EIO_DT_NWK; break;
1942 #endif
1943 #ifdef DT_CMP
1944 case DT_CMP: ent->type = EIO_DT_CMP; break;
1945 #endif
1946 #ifdef DT_LNK
1947 case DT_LNK: ent->type = EIO_DT_LNK; break;
1948 #endif
1949 #ifdef DT_SOCK
1950 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1951 #endif
1952 #ifdef DT_DOOR
1953 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1954 #endif
1955 #ifdef DT_WHT
1956 case DT_WHT: ent->type = EIO_DT_WHT; break;
1957 #endif
1958 }
1959
1960 ent->score = 7;
1961
1962 if (flags & EIO_READDIR_DIRS_FIRST)
1963 {
1964 if (ent->type == EIO_DT_UNKNOWN)
1345 { 1965 {
1346 if (dir->type == EIO_DT_DIR) 1966 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1347 ++dir; 1967 ent->score = 1;
1348 else if ((--oth)->type == EIO_DT_DIR) 1968 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1349 { 1969 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1350 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1351
1352 ++dir;
1353 }
1354 } 1970 }
1355 1971 else if (ent->type == EIO_DT_DIR)
1356 /* now sort the dirs only (dirs all have the same score) */ 1972 ent->score = 0;
1357 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1358 } 1973 }
1359
1360 break;
1361 }
1362
1363 /* now add the entry to our list(s) */
1364 name = entp->d_name;
1365
1366 /* skip . and .. entries */
1367 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1368 {
1369 int len = D_NAMLEN (entp) + 1;
1370
1371 while (ecb_expect_false (namesoffs + len > namesalloc))
1372 {
1373 namesalloc *= 2;
1374 X_LOCK (wrklock);
1375 req->ptr2 = names = realloc (names, namesalloc);
1376 X_UNLOCK (wrklock);
1377
1378 if (!names)
1379 break;
1380 } 1974 }
1381 1975
1382 memcpy (names + namesoffs, name, len);
1383
1384 if (dents)
1385 {
1386 struct eio_dirent *ent;
1387
1388 if (ecb_expect_false (dentoffs == dentalloc))
1389 {
1390 dentalloc *= 2;
1391 X_LOCK (wrklock);
1392 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1393 X_UNLOCK (wrklock);
1394
1395 if (!dents)
1396 break;
1397 }
1398
1399 ent = dents + dentoffs;
1400
1401 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1402 ent->namelen = len - 1;
1403 ent->inode = D_INO (entp);
1404
1405 inode_bits |= ent->inode;
1406
1407 switch (D_TYPE (entp))
1408 {
1409 default:
1410 ent->type = EIO_DT_UNKNOWN;
1411 flags |= EIO_READDIR_FOUND_UNKNOWN;
1412 break;
1413
1414 #ifdef DT_FIFO
1415 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1416 #endif
1417 #ifdef DT_CHR
1418 case DT_CHR: ent->type = EIO_DT_CHR; break;
1419 #endif
1420 #ifdef DT_MPC
1421 case DT_MPC: ent->type = EIO_DT_MPC; break;
1422 #endif
1423 #ifdef DT_DIR
1424 case DT_DIR: ent->type = EIO_DT_DIR; break;
1425 #endif
1426 #ifdef DT_NAM
1427 case DT_NAM: ent->type = EIO_DT_NAM; break;
1428 #endif
1429 #ifdef DT_BLK
1430 case DT_BLK: ent->type = EIO_DT_BLK; break;
1431 #endif
1432 #ifdef DT_MPB
1433 case DT_MPB: ent->type = EIO_DT_MPB; break;
1434 #endif
1435 #ifdef DT_REG
1436 case DT_REG: ent->type = EIO_DT_REG; break;
1437 #endif
1438 #ifdef DT_NWK
1439 case DT_NWK: ent->type = EIO_DT_NWK; break;
1440 #endif
1441 #ifdef DT_CMP
1442 case DT_CMP: ent->type = EIO_DT_CMP; break;
1443 #endif
1444 #ifdef DT_LNK
1445 case DT_LNK: ent->type = EIO_DT_LNK; break;
1446 #endif
1447 #ifdef DT_SOCK
1448 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1449 #endif
1450 #ifdef DT_DOOR
1451 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1452 #endif
1453 #ifdef DT_WHT
1454 case DT_WHT: ent->type = EIO_DT_WHT; break;
1455 #endif
1456 }
1457
1458 ent->score = 7;
1459
1460 if (flags & EIO_READDIR_DIRS_FIRST)
1461 {
1462 if (ent->type == EIO_DT_UNKNOWN)
1463 {
1464 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1465 ent->score = 1;
1466 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1467 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1468 }
1469 else if (ent->type == EIO_DT_DIR)
1470 ent->score = 0;
1471 }
1472 }
1473
1474 namesoffs += len; 1976 namesoffs += len;
1475 ++dentoffs; 1977 ++dentoffs;
1476 } 1978 }
1477 1979
1478 if (EIO_CANCELLED (req)) 1980 if (EIO_CANCELLED (req))
1479 { 1981 {
1480 errno = ECANCELED; 1982 errno = ECANCELED;
1481 break; 1983 break;
1482 } 1984 }
1985
1986#ifdef _WIN32
1987 if (!FindNextFile (dirp, &entp))
1988 {
1989 FindClose (dirp);
1990 dirp = 0;
1483 } 1991 }
1992#endif
1993 }
1484} 1994}
1485 1995
1486#ifdef PAGESIZE 1996/*****************************************************************************/
1487# define eio_pagesize() PAGESIZE 1997/* working directory stuff */
1488#else 1998/* various deficiencies in the posix 2008 api force us to */
1489static intptr_t 1999/* keep the absolute path in string form at all times */
1490eio_pagesize (void) 2000/* fuck yeah. */
1491{
1492 static intptr_t page;
1493 2001
1494 if (!page) 2002#if !HAVE_AT
1495 page = sysconf (_SC_PAGESIZE);
1496 2003
2004/* a bit like realpath, but usually faster because it doesn'T have to return */
2005/* an absolute or canonical path */
2006static const char *
2007wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2008{
2009 if (!wd || *path == '/')
2010 return path;
2011
2012 if (path [0] == '.' && !path [1])
2013 return wd->str;
2014
2015 {
2016 int l1 = wd->len;
2017 int l2 = strlen (path);
2018
2019 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2020
2021 memcpy (res, wd->str, l1);
2022 res [l1] = '/';
2023 memcpy (res + l1 + 1, path, l2 + 1);
2024
2025 return res;
2026 }
2027}
2028
2029#endif
2030
2031static eio_wd
2032eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2033{
2034 int fd;
2035 eio_wd res;
2036 int len = eio__realpath (tmpbuf, wd, path);
2037
2038 if (len < 0)
2039 return EIO_INVALID_WD;
2040
2041#if HAVE_AT
2042 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2043
2044 if (fd < 0)
2045 return EIO_INVALID_WD;
2046#endif
2047
2048 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2049
2050#if HAVE_AT
2051 res->fd = fd;
2052#endif
2053
2054 res->len = len;
2055 memcpy (res->str, tmpbuf->ptr, len);
2056 res->str [len] = 0;
2057
1497 return page; 2058 return res;
1498} 2059}
1499#endif
1500 2060
1501static void 2061eio_wd
1502eio_page_align (void **addr, size_t *length) 2062eio_wd_open_sync (eio_wd wd, const char *path)
1503{ 2063{
1504 intptr_t mask = eio_pagesize () - 1; 2064 struct tmpbuf tmpbuf = { 0 };
2065 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2066 free (tmpbuf.ptr);
1505 2067
1506 /* round down addr */ 2068 return wd;
1507 intptr_t adj = mask & (intptr_t)*addr;
1508
1509 *addr = (void *)((intptr_t)*addr - adj);
1510 *length += adj;
1511
1512 /* round up length */
1513 *length = (*length + mask) & ~mask;
1514} 2069}
1515 2070
1516#if !_POSIX_MEMLOCK 2071void
1517# define eio__mlockall(a) ((errno = ENOSYS), -1) 2072eio_wd_close_sync (eio_wd wd)
1518#else 2073{
2074 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2075 {
2076 #if HAVE_AT
2077 close (wd->fd);
2078 #endif
2079 free (wd);
2080 }
2081}
2082
2083#if HAVE_AT
2084
2085/* they forgot these */
1519 2086
1520static int 2087static int
1521eio__mlockall (int flags) 2088eio__truncateat (int dirfd, const char *path, off_t length)
1522{ 2089{
1523 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 2090 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1524 extern int mallopt (int, int); 2091 int res;
1525 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1526 #endif
1527 2092
1528 if (EIO_MCL_CURRENT != MCL_CURRENT 2093 if (fd < 0)
1529 || EIO_MCL_FUTURE != MCL_FUTURE) 2094 return fd;
1530 {
1531 flags = 0
1532 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1533 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1534 }
1535 2095
1536 return mlockall (flags); 2096 res = ftruncate (fd, length);
2097 close (fd);
2098 return res;
1537} 2099}
1538#endif
1539
1540#if !_POSIX_MEMLOCK_RANGE
1541# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1542#else
1543 2100
1544static int 2101static int
1545eio__mlock (void *addr, size_t length) 2102eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1546{ 2103{
1547 eio_page_align (&addr, &length); 2104 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2105 int res;
1548 2106
1549 return mlock (addr, length); 2107 if (fd < 0)
1550} 2108 return fd;
1551 2109
1552#endif 2110 res = fstatvfs (fd, buf);
1553 2111 close (fd);
1554#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1555# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1556#else
1557
1558static int
1559eio__msync (void *mem, size_t len, int flags)
1560{
1561 eio_page_align (&mem, &len);
1562
1563 if (EIO_MS_ASYNC != MS_SYNC
1564 || EIO_MS_INVALIDATE != MS_INVALIDATE
1565 || EIO_MS_SYNC != MS_SYNC)
1566 {
1567 flags = 0
1568 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1569 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1570 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1571 }
1572
1573 return msync (mem, len, flags);
1574}
1575
1576#endif
1577
1578static int
1579eio__mtouch (eio_req *req)
1580{
1581 void *mem = req->ptr2;
1582 size_t len = req->size;
1583 int flags = req->int1;
1584
1585 eio_page_align (&mem, &len);
1586
1587 {
1588 intptr_t addr = (intptr_t)mem;
1589 intptr_t end = addr + len;
1590 intptr_t page = eio_pagesize ();
1591
1592 if (addr < end)
1593 if (flags & EIO_MT_MODIFY) /* modify */
1594 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1595 else
1596 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1597 }
1598
1599 return 0; 2112 return res;
2113
1600} 2114}
2115
2116#endif
1601 2117
1602/*****************************************************************************/ 2118/*****************************************************************************/
1603 2119
1604#define ALLOC(len) \ 2120#define ALLOC(len) \
1605 if (!req->ptr2) \ 2121 if (!req->ptr2) \
1620{ 2136{
1621 ETP_REQ *req; 2137 ETP_REQ *req;
1622 struct timespec ts; 2138 struct timespec ts;
1623 etp_worker *self = (etp_worker *)thr_arg; 2139 etp_worker *self = (etp_worker *)thr_arg;
1624 2140
2141#if HAVE_PRCTL_SET_NAME
2142 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2143#endif
2144
1625 /* try to distribute timeouts somewhat randomly */ 2145 /* try to distribute timeouts somewhat evenly */
1626 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2146 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1627 2147
1628 for (;;) 2148 for (;;)
1629 { 2149 {
2150 ts.tv_sec = 0;
2151
1630 X_LOCK (reqlock); 2152 X_LOCK (reqlock);
1631 2153
1632 for (;;) 2154 for (;;)
1633 { 2155 {
1634 self->req = req = reqq_shift (&req_queue); 2156 req = reqq_shift (&req_queue);
1635 2157
1636 if (req) 2158 if (req)
1637 break; 2159 break;
1638 2160
2161 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2162 {
2163 X_UNLOCK (reqlock);
2164 X_LOCK (wrklock);
2165 --started;
2166 X_UNLOCK (wrklock);
2167 goto quit;
2168 }
2169
1639 ++idle; 2170 ++idle;
1640 2171
1641 ts.tv_sec = time (0) + idle_timeout; 2172 if (idle <= max_idle)
1642 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2173 /* we are allowed to idle, so do so without any timeout */
2174 X_COND_WAIT (reqwait, reqlock);
2175 else
1643 { 2176 {
1644 if (idle > max_idle) 2177 /* initialise timeout once */
1645 { 2178 if (!ts.tv_sec)
1646 --idle; 2179 ts.tv_sec = time (0) + idle_timeout;
1647 X_UNLOCK (reqlock);
1648 X_LOCK (wrklock);
1649 --started;
1650 X_UNLOCK (wrklock);
1651 goto quit;
1652 }
1653 2180
1654 /* we are allowed to idle, so do so without any timeout */
1655 X_COND_WAIT (reqwait, reqlock); 2181 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2182 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1656 } 2183 }
1657 2184
1658 --idle; 2185 --idle;
1659 } 2186 }
1660 2187
1663 X_UNLOCK (reqlock); 2190 X_UNLOCK (reqlock);
1664 2191
1665 if (req->type < 0) 2192 if (req->type < 0)
1666 goto quit; 2193 goto quit;
1667 2194
1668 if (!EIO_CANCELLED (req))
1669 ETP_EXECUTE (self, req); 2195 ETP_EXECUTE (self, req);
1670 2196
1671 X_LOCK (reslock); 2197 X_LOCK (reslock);
1672 2198
1673 ++npending; 2199 ++npending;
1674 2200
1675 if (!reqq_push (&res_queue, req) && want_poll_cb) 2201 if (!reqq_push (&res_queue, req) && want_poll_cb)
1676 want_poll_cb (); 2202 want_poll_cb ();
1677 2203
1678 self->req = 0;
1679 etp_worker_clear (self); 2204 etp_worker_clear (self);
1680 2205
1681 X_UNLOCK (reslock); 2206 X_UNLOCK (reslock);
1682 } 2207 }
1683 2208
1684quit: 2209quit:
2210 free (req);
2211
1685 X_LOCK (wrklock); 2212 X_LOCK (wrklock);
1686 etp_worker_free (self); 2213 etp_worker_free (self);
1687 X_UNLOCK (wrklock); 2214 X_UNLOCK (wrklock);
1688 2215
1689 return 0; 2216 return 0;
1692/*****************************************************************************/ 2219/*****************************************************************************/
1693 2220
1694int ecb_cold 2221int ecb_cold
1695eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2222eio_init (void (*want_poll)(void), void (*done_poll)(void))
1696{ 2223{
2224#if !HAVE_PREADWRITE
2225 X_MUTEX_CREATE (preadwritelock);
2226#endif
2227
1697 return etp_init (want_poll, done_poll); 2228 return etp_init (want_poll, done_poll);
1698} 2229}
1699 2230
1700ECB_INLINE void 2231ecb_inline void
1701eio_api_destroy (eio_req *req) 2232eio_api_destroy (eio_req *req)
1702{ 2233{
1703 free (req); 2234 free (req);
1704} 2235}
1705 2236
1728 } 2259 }
1729 2260
1730static void 2261static void
1731eio_execute (etp_worker *self, eio_req *req) 2262eio_execute (etp_worker *self, eio_req *req)
1732{ 2263{
2264#if HAVE_AT
2265 int dirfd;
2266#else
2267 const char *path;
2268#endif
2269
2270 if (ecb_expect_false (EIO_CANCELLED (req)))
2271 {
2272 req->result = -1;
2273 req->errorno = ECANCELED;
2274 return;
2275 }
2276
2277 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2278 {
2279 req->result = -1;
2280 req->errorno = ENOENT;
2281 return;
2282 }
2283
2284 if (req->type >= EIO_OPEN)
2285 {
2286 #if HAVE_AT
2287 dirfd = WD2FD (req->wd);
2288 #else
2289 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2290 #endif
2291 }
2292
1733 switch (req->type) 2293 switch (req->type)
1734 { 2294 {
2295 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2296 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2297 break;
2298 case EIO_WD_CLOSE: req->result = 0;
2299 eio_wd_close_sync (req->wd); break;
2300
1735 case EIO_READ: ALLOC (req->size); 2301 case EIO_READ: ALLOC (req->size);
1736 req->result = req->offs >= 0 2302 req->result = req->offs >= 0
1737 ? pread (req->int1, req->ptr2, req->size, req->offs) 2303 ? pread (req->int1, req->ptr2, req->size, req->offs)
1738 : read (req->int1, req->ptr2, req->size); break; 2304 : read (req->int1, req->ptr2, req->size); break;
1739 case EIO_WRITE: req->result = req->offs >= 0 2305 case EIO_WRITE: req->result = req->offs >= 0
1740 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2306 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1741 : write (req->int1, req->ptr2, req->size); break; 2307 : write (req->int1, req->ptr2, req->size); break;
1742 2308
1743 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2309 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1744 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2310 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2311
2312#if HAVE_AT
1745 2313
1746 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2314 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1747 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2315 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1748 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2316 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2317 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2318 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2319 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2320 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2321 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2322
2323 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2324 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2325 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2326 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2327 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2328 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2329 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2330 case EIO_READLINK: ALLOC (PATH_MAX);
2331 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2332 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2333 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2334 case EIO_UTIME:
2335 case EIO_FUTIME:
2336 {
2337 struct timespec ts[2];
2338 struct timespec *times;
2339
2340 if (req->nv1 != -1. || req->nv2 != -1.)
2341 {
2342 ts[0].tv_sec = req->nv1;
2343 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2344 ts[1].tv_sec = req->nv2;
2345 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2346
2347 times = ts;
2348 }
2349 else
2350 times = 0;
2351
2352 req->result = req->type == EIO_FUTIME
2353 ? futimens (req->int1, times)
2354 : utimensat (dirfd, req->ptr1, times, 0);
2355 }
2356 break;
2357
2358#else
2359
2360 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2361 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2362 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1749 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2363 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2364 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2365 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2366 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2367 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2368
2369 case EIO_UNLINK: req->result = unlink (path ); break;
2370 case EIO_RMDIR: req->result = rmdir (path ); break;
2371 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2372 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2373 case EIO_LINK: req->result = link (path , req->ptr2); break;
2374 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2375 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2376 case EIO_READLINK: ALLOC (PATH_MAX);
2377 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2378 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2379 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2380
2381 case EIO_UTIME:
2382 case EIO_FUTIME:
2383 {
2384 struct timeval tv[2];
2385 struct timeval *times;
2386
2387 if (req->nv1 != -1. || req->nv2 != -1.)
2388 {
2389 tv[0].tv_sec = req->nv1;
2390 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2391 tv[1].tv_sec = req->nv2;
2392 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2393
2394 times = tv;
2395 }
2396 else
2397 times = 0;
2398
2399 req->result = req->type == EIO_FUTIME
2400 ? futimes (req->int1, times)
2401 : utimes (req->ptr1, times);
2402 }
2403 break;
2404
2405#endif
2406
2407 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2408 {
2409 ALLOC (req->result);
2410 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2411 }
2412 break;
2413
1750 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2414 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1751 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2415 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1752 2416
1753 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1754 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1755 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2417 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1756 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2418 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1757 2419
1758 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1759 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2420 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1760 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1761 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2421 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1762 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1763 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2422 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1764 2423
1765 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1766 case EIO_CLOSE: req->result = close (req->int1); break; 2424 case EIO_CLOSE: req->result = close (req->int1); break;
1767 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2425 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1768 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1769 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1770 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1771 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1772 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1773 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1774 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1775
1776 case EIO_READLINK: ALLOC (PATH_MAX);
1777 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1778
1779 case EIO_SYNC: req->result = 0; sync (); break; 2426 case EIO_SYNC: req->result = 0; sync (); break;
1780 case EIO_FSYNC: req->result = fsync (req->int1); break; 2427 case EIO_FSYNC: req->result = fsync (req->int1); break;
1781 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2428 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2429 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2430 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1782 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2431 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1783 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2432 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1784 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2433 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1785 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2434 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1786 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2435 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1787 2436
1788 case EIO_READDIR: eio__scandir (req, self); break; 2437 case EIO_READDIR: eio__scandir (req, self); break;
1789 2438
1790 case EIO_BUSY: 2439 case EIO_BUSY:
1791#ifdef _WIN32 2440#ifdef _WIN32
1800 req->result = select (0, 0, 0, 0, &tv); 2449 req->result = select (0, 0, 0, 0, &tv);
1801 } 2450 }
1802#endif 2451#endif
1803 break; 2452 break;
1804 2453
1805 case EIO_UTIME:
1806 case EIO_FUTIME:
1807 {
1808 struct timeval tv[2];
1809 struct timeval *times;
1810
1811 if (req->nv1 != -1. || req->nv2 != -1.)
1812 {
1813 tv[0].tv_sec = req->nv1;
1814 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1815 tv[1].tv_sec = req->nv2;
1816 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1817
1818 times = tv;
1819 }
1820 else
1821 times = 0;
1822
1823 req->result = req->type == EIO_FUTIME
1824 ? futimes (req->int1, times)
1825 : utimes (req->ptr1, times);
1826 }
1827 break;
1828
1829 case EIO_GROUP: 2454 case EIO_GROUP:
1830 abort (); /* handled in eio_request */ 2455 abort (); /* handled in eio_request */
1831 2456
1832 case EIO_NOP: 2457 case EIO_NOP:
1833 req->result = 0; 2458 req->result = 0;
1871eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2496eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1872{ 2497{
1873 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2498 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1874} 2499}
1875 2500
2501eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2502{
2503 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2504}
2505
2506eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2507{
2508 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2509}
2510
2511eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2512{
2513 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2514}
2515
1876eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2516eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1877{ 2517{
1878 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2518 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1879} 2519}
1880 2520
1886eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2526eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1887{ 2527{
1888 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2528 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1889} 2529}
1890 2530
1891eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2531eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1892{ 2532{
1893 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2533 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1894}
1895
1896eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1897{
1898 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1899} 2534}
1900 2535
1901eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2536eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1902{ 2537{
1903 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2538 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1941eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2576eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1942{ 2577{
1943 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2578 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1944} 2579}
1945 2580
1946eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2581eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1947{ 2582{
1948 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2583 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1949} 2584}
1950 2585
1951eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2586eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1971eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2606eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1972{ 2607{
1973 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2608 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1974} 2609}
1975 2610
1976eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2611eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1977{ 2612{
1978 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2613 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1979} 2614}
1980 2615
1981eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2616eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1995} 2630}
1996 2631
1997eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2632eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1998{ 2633{
1999 return eio__1path (EIO_READLINK, path, pri, cb, data); 2634 return eio__1path (EIO_READLINK, path, pri, cb, data);
2635}
2636
2637eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2638{
2639 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2000} 2640}
2001 2641
2002eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2642eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2003{ 2643{
2004 return eio__1path (EIO_STAT, path, pri, cb, data); 2644 return eio__1path (EIO_STAT, path, pri, cb, data);
2123} 2763}
2124 2764
2125/*****************************************************************************/ 2765/*****************************************************************************/
2126/* misc garbage */ 2766/* misc garbage */
2127 2767
2128ssize_t 2768eio_ssize_t
2129eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2769eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2130{ 2770{
2131 etp_worker wrk;
2132 ssize_t ret;
2133
2134 wrk.dbuf = 0;
2135
2136 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2771 return eio__sendfile (ofd, ifd, offset, count);
2137
2138 if (wrk.dbuf)
2139 free (wrk.dbuf);
2140
2141 return ret;
2142} 2772}
2143 2773

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines