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

Comparing libeio/eio.c (file contents):
Revision 1.75 by root, Wed Jun 29 10:32:55 2011 UTC vs.
Revision 1.106 by root, Mon Sep 26 20:19:08 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
63#if _POSIX_VERSION >= 200809L
64# define HAVE_AT 1
65#else
66# define HAVE_AT 0
67#endif
68
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 69/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes form stdint.h, says idiot openbsd coder */ 70/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H 71#if HAVE_STDINT_H
67# include <stdint.h> 72# include <stdint.h>
68#endif 73#endif
69 74
75#ifndef ECANCELED
76# define ECANCELED EDOM
77#endif
78#ifndef ELOOP
79# define ELOOP EDOM
80#endif
81
82#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
83# define ENOTSOCK WSAENOTSOCK
84#endif
85
86static void eio_destroy (eio_req *req);
87
70#ifndef EIO_FINISH 88#ifndef EIO_FINISH
71# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 89# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
72#endif 90#endif
73 91
74#ifndef EIO_DESTROY 92#ifndef EIO_DESTROY
77 95
78#ifndef EIO_FEED 96#ifndef EIO_FEED
79# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 97# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
80#endif 98#endif
81 99
100#ifndef EIO_FD_TO_WIN32_HANDLE
101# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
102#endif
103#ifndef EIO_WIN32_HANDLE_TO_FD
104# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
105#endif
106
107#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
108
109#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
110
82#ifdef _WIN32 111#ifdef _WIN32
83 112
84 /*doh*/ 113 #undef PAGESIZE
114 #define PAGESIZE 4096 /* GetSystemInfo? */
115
116 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
117
118 #ifdef EIO_STRUCT_STATI64
119 /* look at perl's non-sloppy stat */
120 #define stat(path,buf) _stati64 (path,buf)
121 #define fstat(fd,buf) _fstati64 (fd,buf)
122 #endif
123 #define lstat(path,buf) stat (path,buf)
124 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
125 #define mkdir(path,mode) _mkdir (path)
126 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
127
128 #define chmod(path,mode) _chmod (path, mode)
129 #define dup(fd) _dup (fd)
130 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
131
132 #define fchmod(fd,mode) EIO_ENOSYS ()
133 #define chown(path,uid,gid) EIO_ENOSYS ()
134 #define fchown(fd,uid,gid) EIO_ENOSYS ()
135 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
136 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
137 #define mknod(path,mode,dev) EIO_ENOSYS ()
138 #define sync() EIO_ENOSYS ()
139 #define readlink(path,buf,s) EIO_ENOSYS ()
140 #define statvfs(path,buf) EIO_ENOSYS ()
141 #define fstatvfs(fd,buf) EIO_ENOSYS ()
142
143 /* rename() uses MoveFile, which fails to overwrite */
144 #define rename(old,neu) eio__rename (old, neu)
145
146 static int
147 eio__rename (const char *old, const char *neu)
148 {
149 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
150 return 0;
151
152 /* should steal _dosmaperr */
153 switch (GetLastError ())
154 {
155 case ERROR_FILE_NOT_FOUND:
156 case ERROR_PATH_NOT_FOUND:
157 case ERROR_INVALID_DRIVE:
158 case ERROR_NO_MORE_FILES:
159 case ERROR_BAD_NETPATH:
160 case ERROR_BAD_NET_NAME:
161 case ERROR_BAD_PATHNAME:
162 case ERROR_FILENAME_EXCED_RANGE:
163 errno = ENOENT;
164 break;
165
166 default:
167 errno = EACCES;
168 break;
169 }
170
171 return -1;
172 }
173
174 /* we could even stat and see if it exists */
175 static int
176 symlink (const char *old, const char *neu)
177 {
178 #if WINVER >= 0x0600
179 if (CreateSymbolicLink (neu, old, 1))
180 return 0;
181
182 if (CreateSymbolicLink (neu, old, 0))
183 return 0;
184 #endif
185
186 return EIO_ERRNO (ENOENT, -1);
187 }
188
189 /* POSIX API only */
190 #define CreateHardLink(neu,old,flags) 0
191 #define CreateSymbolicLink(neu,old,flags) 0
192
193 struct statvfs
194 {
195 int dummy;
196 };
197
198 #define DT_DIR EIO_DT_DIR
199 #define DT_REG EIO_DT_REG
200 #define D_NAME(entp) entp.cFileName
201 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
202
85#else 203#else
86 204
87# include <sys/time.h> 205 #include <sys/time.h>
88# include <sys/select.h> 206 #include <sys/select.h>
207 #include <sys/statvfs.h>
89# include <unistd.h> 208 #include <unistd.h>
209 #include <signal.h>
210 #include <dirent.h>
211
212 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
213 #include <sys/mman.h>
214 #endif
215
216 #define D_NAME(entp) entp->d_name
217
218 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
219 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__
220 #define _DIRENT_HAVE_D_TYPE /* sigh */
221 #define D_INO(de) (de)->d_fileno
222 #define D_NAMLEN(de) (de)->d_namlen
223 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
224 #define D_INO(de) (de)->d_ino
225 #endif
226
227 #ifdef _D_EXACT_NAMLEN
228 #undef D_NAMLEN
229 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
230 #endif
231
232 #ifdef _DIRENT_HAVE_D_TYPE
233 #define D_TYPE(de) (de)->d_type
234 #endif
235
236 #ifndef EIO_STRUCT_DIRENT
237 #define EIO_STRUCT_DIRENT struct dirent
238 #endif
239
240#endif
241
242#if HAVE_UTIMES
90# include <utime.h> 243# 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 244#endif
106 245
107#ifdef _D_EXACT_NAMLEN 246#if HAVE_SYS_SYSCALL_H
108# undef D_NAMLEN 247# 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 248#endif
115 249
116# ifndef EIO_STRUCT_DIRENT 250#if HAVE_SYS_PRCTL_H
117# define EIO_STRUCT_DIRENT struct dirent 251# include <sys/prctl.h>
118# endif
119
120#endif 252#endif
121 253
122#if HAVE_SENDFILE 254#if HAVE_SENDFILE
123# if __linux 255# if __linux
124# include <sys/sendfile.h> 256# include <sys/sendfile.h>
139#endif 271#endif
140#ifndef D_INO 272#ifndef D_INO
141# define D_INO(de) 0 273# define D_INO(de) 0
142#endif 274#endif
143#ifndef D_NAMLEN 275#ifndef D_NAMLEN
144# define D_NAMLEN(de) strlen ((de)->d_name) 276# define D_NAMLEN(entp) strlen (D_NAME (entp))
145#endif 277#endif
146 278
147/* used for struct dirent, AIX doesn't provide it */ 279/* used for struct dirent, AIX doesn't provide it */
148#ifndef NAME_MAX 280#ifndef NAME_MAX
149# define NAME_MAX 4096 281# define NAME_MAX 4096
156 288
157/* buffer size for various temporary buffers */ 289/* buffer size for various temporary buffers */
158#define EIO_BUFSIZE 65536 290#define EIO_BUFSIZE 65536
159 291
160#define dBUF \ 292#define dBUF \
161 char *eio_buf; \
162 ETP_WORKER_LOCK (self); \
163 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 293 char *eio_buf = malloc (EIO_BUFSIZE); \
164 ETP_WORKER_UNLOCK (self); \
165 errno = ENOMEM; \ 294 errno = ENOMEM; \
166 if (!eio_buf) \ 295 if (!eio_buf) \
167 return -1; 296 return -1
297
298#define FUBd \
299 free (eio_buf)
168 300
169#define EIO_TICKS ((1000000 + 1023) >> 10) 301#define EIO_TICKS ((1000000 + 1023) >> 10)
302
303/*****************************************************************************/
304
305struct tmpbuf
306{
307 void *ptr;
308 int len;
309};
310
311static void *
312tmpbuf_get (struct tmpbuf *buf, int len)
313{
314 if (buf->len < len)
315 {
316 free (buf->ptr);
317 buf->ptr = malloc (buf->len = len);
318 }
319
320 return buf->ptr;
321}
322
323/*****************************************************************************/
170 324
171#define ETP_PRI_MIN EIO_PRI_MIN 325#define ETP_PRI_MIN EIO_PRI_MIN
172#define ETP_PRI_MAX EIO_PRI_MAX 326#define ETP_PRI_MAX EIO_PRI_MAX
173 327
174struct etp_worker; 328struct etp_worker;
178static int eio_finish (eio_req *req); 332static int eio_finish (eio_req *req);
179#define ETP_FINISH(req) eio_finish (req) 333#define ETP_FINISH(req) eio_finish (req)
180static void eio_execute (struct etp_worker *self, eio_req *req); 334static void eio_execute (struct etp_worker *self, eio_req *req);
181#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 335#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req)
182 336
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/*****************************************************************************/ 337/*****************************************************************************/
201 338
202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1) 339#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203 340
204/* calculate time difference in ~1/EIO_TICKS of a second */ 341/* calculate time difference in ~1/EIO_TICKS of a second */
215static void (*done_poll_cb) (void); 352static void (*done_poll_cb) (void);
216 353
217static unsigned int max_poll_time; /* reslock */ 354static unsigned int max_poll_time; /* reslock */
218static unsigned int max_poll_reqs; /* reslock */ 355static unsigned int max_poll_reqs; /* reslock */
219 356
220static volatile unsigned int nreqs; /* reqlock */ 357static unsigned int nreqs; /* reqlock */
221static volatile unsigned int nready; /* reqlock */ 358static unsigned int nready; /* reqlock */
222static volatile unsigned int npending; /* reqlock */ 359static unsigned int npending; /* reqlock */
223static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */ 360static 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 */ 361static unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
225 362
226static xmutex_t wrklock; 363static xmutex_t wrklock;
227static xmutex_t reslock; 364static xmutex_t reslock;
228static xmutex_t reqlock; 365static xmutex_t reqlock;
229static xcond_t reqwait; 366static xcond_t reqwait;
232/* 369/*
233 * make our pread/pwrite emulation safe against themselves, but not against 370 * make our pread/pwrite emulation safe against themselves, but not against
234 * normal read/write by using a mutex. slows down execution a lot, 371 * normal read/write by using a mutex. slows down execution a lot,
235 * but that's your problem, not mine. 372 * but that's your problem, not mine.
236 */ 373 */
237static xmutex_t preadwritelock = X_MUTEX_INIT; 374static xmutex_t preadwritelock;
238#endif 375#endif
239 376
240typedef struct etp_worker 377typedef struct etp_worker
241{ 378{
379 struct tmpbuf tmpbuf;
380
242 /* locked by wrklock */ 381 /* locked by wrklock */
243 struct etp_worker *prev, *next; 382 struct etp_worker *prev, *next;
244 383
245 xthread_t tid; 384 xthread_t tid;
246 385
247 /* locked by reslock, reqlock or wrklock */ 386#ifdef ETP_WORKER_COMMON
248 ETP_REQ *req; /* currently processed request */
249
250 ETP_WORKER_COMMON 387 ETP_WORKER_COMMON
388#endif
251} etp_worker; 389} etp_worker;
252 390
253static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */ 391static etp_worker wrk_first; /* NOT etp */
254 392
255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock) 393#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock) 394#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
257 395
258/* worker threads management */ 396/* worker threads management */
259 397
260static void ecb_cold 398static void ecb_cold
261etp_worker_clear (etp_worker *wrk) 399etp_worker_clear (etp_worker *wrk)
262{ 400{
263 ETP_WORKER_CLEAR (wrk);
264} 401}
265 402
266static void ecb_cold 403static void ecb_cold
267etp_worker_free (etp_worker *wrk) 404etp_worker_free (etp_worker *wrk)
268{ 405{
406 free (wrk->tmpbuf.ptr);
407
269 wrk->next->prev = wrk->prev; 408 wrk->next->prev = wrk->prev;
270 wrk->prev->next = wrk->next; 409 wrk->prev->next = wrk->next;
271 410
272 free (wrk); 411 free (wrk);
273} 412}
329} etp_reqq; 468} etp_reqq;
330 469
331static etp_reqq req_queue; 470static etp_reqq req_queue;
332static etp_reqq res_queue; 471static etp_reqq res_queue;
333 472
473static void ecb_noinline ecb_cold
474reqq_init (etp_reqq *q)
475{
476 int pri;
477
478 for (pri = 0; pri < ETP_NUM_PRI; ++pri)
479 q->qs[pri] = q->qe[pri] = 0;
480
481 q->size = 0;
482}
483
334static int ecb_noinline 484static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req) 485reqq_push (etp_reqq *q, ETP_REQ *req)
336{ 486{
337 int pri = req->pri; 487 int pri = req->pri;
338 req->next = 0; 488 req->next = 0;
372 } 522 }
373 523
374 abort (); 524 abort ();
375} 525}
376 526
377static void ecb_cold 527static int ecb_cold
378etp_thread_init (void) 528etp_init (void (*want_poll)(void), void (*done_poll)(void))
379{ 529{
380 X_MUTEX_CREATE (wrklock); 530 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock); 531 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock); 532 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait); 533 X_COND_CREATE (reqwait);
384}
385 534
386static void ecb_cold 535 reqq_init (&req_queue);
387etp_atfork_prepare (void) 536 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}
396 537
397static void ecb_cold 538 wrk_first.next =
398etp_atfork_parent (void) 539 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 }
429 540
430 started = 0; 541 started = 0;
431 idle = 0; 542 idle = 0;
432 nreqs = 0; 543 nreqs = 0;
433 nready = 0; 544 nready = 0;
434 npending = 0; 545 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);
452 546
453 want_poll_cb = want_poll; 547 want_poll_cb = want_poll;
454 done_poll_cb = done_poll; 548 done_poll_cb = done_poll;
455 549
456 return 0; 550 return 0;
496} 590}
497 591
498static void ecb_cold 592static void ecb_cold
499etp_end_thread (void) 593etp_end_thread (void)
500{ 594{
501 eio_req *req = calloc (1, sizeof (eio_req)); 595 eio_req *req = calloc (1, sizeof (eio_req)); /* will be freed by worker */
502 596
503 req->type = -1; 597 req->type = -1;
504 req->pri = ETP_PRI_MAX - ETP_PRI_MIN; 598 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
505 599
506 X_LOCK (reqlock); 600 X_LOCK (reqlock);
583} 677}
584 678
585static void 679static void
586etp_cancel (ETP_REQ *req) 680etp_cancel (ETP_REQ *req)
587{ 681{
588 X_LOCK (wrklock); 682 req->cancelled = 1;
589 req->flags |= EIO_FLAG_CANCELLED;
590 X_UNLOCK (wrklock);
591 683
592 eio_grp_cancel (req); 684 eio_grp_cancel (req);
593} 685}
594 686
595static void 687static void
711 return eio_finish (grp); 803 return eio_finish (grp);
712 else 804 else
713 return 0; 805 return 0;
714} 806}
715 807
716void 808static void
717eio_destroy (eio_req *req) 809eio_destroy (eio_req *req)
718{ 810{
719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 811 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 812 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721 813
739 if (grp->grp_first == req) 831 if (grp->grp_first == req)
740 grp->grp_first = req->grp_next; 832 grp->grp_first = req->grp_next;
741 833
742 res2 = grp_dec (grp); 834 res2 = grp_dec (grp);
743 835
744 if (!res && res2) 836 if (!res)
745 res = res2; 837 res = res2;
746 } 838 }
747 839
748 eio_destroy (req); 840 eio_destroy (req);
749 841
841# undef pread 933# undef pread
842# undef pwrite 934# undef pwrite
843# define pread eio__pread 935# define pread eio__pread
844# define pwrite eio__pwrite 936# define pwrite eio__pwrite
845 937
846static ssize_t 938static eio_ssize_t
847eio__pread (int fd, void *buf, size_t count, off_t offset) 939eio__pread (int fd, void *buf, size_t count, off_t offset)
848{ 940{
849 ssize_t res; 941 eio_ssize_t res;
850 off_t ooffset; 942 off_t ooffset;
851 943
852 X_LOCK (preadwritelock); 944 X_LOCK (preadwritelock);
853 ooffset = lseek (fd, 0, SEEK_CUR); 945 ooffset = lseek (fd, 0, SEEK_CUR);
854 lseek (fd, offset, SEEK_SET); 946 lseek (fd, offset, SEEK_SET);
857 X_UNLOCK (preadwritelock); 949 X_UNLOCK (preadwritelock);
858 950
859 return res; 951 return res;
860} 952}
861 953
862static ssize_t 954static eio_ssize_t
863eio__pwrite (int fd, void *buf, size_t count, off_t offset) 955eio__pwrite (int fd, void *buf, size_t count, off_t offset)
864{ 956{
865 ssize_t res; 957 eio_ssize_t res;
866 off_t ooffset; 958 off_t ooffset;
867 959
868 X_LOCK (preadwritelock); 960 X_LOCK (preadwritelock);
869 ooffset = lseek (fd, 0, SEEK_CUR); 961 ooffset = lseek (fd, 0, SEEK_CUR);
870 lseek (fd, offset, SEEK_SET); 962 lseek (fd, offset, SEEK_SET);
915 1007
916#if !HAVE_FDATASYNC 1008#if !HAVE_FDATASYNC
917# undef fdatasync 1009# undef fdatasync
918# define fdatasync(fd) fsync (fd) 1010# define fdatasync(fd) fsync (fd)
919#endif 1011#endif
1012
1013static int
1014eio__syncfs (int fd)
1015{
1016 int res;
1017
1018#if HAVE_SYS_SYNCFS
1019 res = (int)syscall (__NR_syncfs, (int)(fd));
1020#else
1021 res = -1;
1022 errno = ENOSYS;
1023#endif
1024
1025 if (res < 0 && errno == ENOSYS && fd >= 0)
1026 sync ();
1027
1028 return res;
1029}
920 1030
921/* sync_file_range always needs emulation */ 1031/* sync_file_range always needs emulation */
922static int 1032static int
923eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 1033eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
924{ 1034{
944 /* even though we could play tricks with the flags, it's better to always 1054 /* 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 */ 1055 * call fdatasync, as that matches the expectation of its users best */
946 return fdatasync (fd); 1056 return fdatasync (fd);
947} 1057}
948 1058
1059static int
1060eio__fallocate (int fd, int mode, off_t offset, size_t len)
1061{
1062#if HAVE_FALLOCATE
1063 return fallocate (fd, mode, offset, len);
1064#else
1065 errno = ENOSYS;
1066 return -1;
1067#endif
1068}
1069
949#if !HAVE_READAHEAD 1070#if !HAVE_READAHEAD
950# undef readahead 1071# undef readahead
951# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 1072# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
952 1073
953static ssize_t 1074static eio_ssize_t
954eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 1075eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
955{ 1076{
956 size_t todo = count; 1077 size_t todo = count;
957 dBUF; 1078 dBUF;
958 1079
963 pread (fd, eio_buf, len, offset); 1084 pread (fd, eio_buf, len, offset);
964 offset += len; 1085 offset += len;
965 todo -= len; 1086 todo -= len;
966 } 1087 }
967 1088
1089 FUBd;
1090
968 errno = 0; 1091 errno = 0;
969 return count; 1092 return count;
970} 1093}
971 1094
972#endif 1095#endif
973 1096
974/* sendfile always needs emulation */ 1097/* sendfile always needs emulation */
975static ssize_t 1098static eio_ssize_t
976eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 1099eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
977{ 1100{
978 ssize_t written = 0; 1101 eio_ssize_t written = 0;
979 ssize_t res; 1102 eio_ssize_t res;
980 1103
981 if (!count) 1104 if (!count)
982 return 0; 1105 return 0;
983 1106
984 for (;;) 1107 for (;;)
985 { 1108 {
1109#ifdef __APPLE__
1110# undef HAVE_SENDFILE /* broken, as everything on os x */
1111#endif
986#if HAVE_SENDFILE 1112#if HAVE_SENDFILE
987# if __linux 1113# if __linux
988 off_t soffset = offset; 1114 off_t soffset = offset;
989 res = sendfile (ofd, ifd, &soffset, count); 1115 res = sendfile (ofd, ifd, &soffset, count);
990 1116
1006 1132
1007 /* according to source inspection, this is correct, and useful behaviour */ 1133 /* according to source inspection, this is correct, and useful behaviour */
1008 if (sbytes) 1134 if (sbytes)
1009 res = sbytes; 1135 res = sbytes;
1010 1136
1011# elif defined (__APPLE__) && 0 /* broken, as everything on os x */ 1137# elif defined (__APPLE__)
1012 off_t sbytes = count; 1138 off_t sbytes = count;
1013 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1139 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1014 1140
1015 /* according to the manpage, sbytes is always valid */ 1141 /* according to the manpage, sbytes is always valid */
1016 if (sbytes) 1142 if (sbytes)
1033 if (res < 0 && sbytes) 1159 if (res < 0 && sbytes)
1034 res = sbytes; 1160 res = sbytes;
1035 1161
1036# endif 1162# endif
1037 1163
1038#elif defined (_WIN32) 1164#elif defined (_WIN32) && 0
1039 /* does not work, just for documentation of what would need to be done */ 1165 /* 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, */ 1166 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1041 /* libeio guarantees that the file offset does not change, and windows */ 1167 /* 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 */ 1168 /* has no way to get an independent handle to the same file description */
1043 HANDLE h = TO_SOCKET (ifd); 1169 HANDLE h = TO_SOCKET (ifd);
1076 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 1202 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1077 /* BSDs */ 1203 /* BSDs */
1078#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 1204#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1079 || errno == ENOTSUP 1205 || errno == ENOTSUP
1080#endif 1206#endif
1207#ifdef EOPNOTSUPP /* windows */
1081 || errno == EOPNOTSUPP /* BSDs */ 1208 || errno == EOPNOTSUPP /* BSDs */
1209#endif
1082#if __solaris 1210#if __solaris
1083 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 1211 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1084#endif 1212#endif
1085 ) 1213 )
1086 ) 1214 )
1090 1218
1091 res = 0; 1219 res = 0;
1092 1220
1093 while (count) 1221 while (count)
1094 { 1222 {
1095 ssize_t cnt; 1223 eio_ssize_t cnt;
1096 1224
1097 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 1225 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1098 1226
1099 if (cnt <= 0) 1227 if (cnt <= 0)
1100 { 1228 {
1112 1240
1113 offset += cnt; 1241 offset += cnt;
1114 res += cnt; 1242 res += cnt;
1115 count -= cnt; 1243 count -= cnt;
1116 } 1244 }
1245
1246 FUBd;
1117 } 1247 }
1118 1248
1119 return res; 1249 return res;
1250}
1251
1252#ifdef PAGESIZE
1253# define eio_pagesize() PAGESIZE
1254#else
1255static intptr_t
1256eio_pagesize (void)
1257{
1258 static intptr_t page;
1259
1260 if (!page)
1261 page = sysconf (_SC_PAGESIZE);
1262
1263 return page;
1264}
1265#endif
1266
1267static void
1268eio_page_align (void **addr, size_t *length)
1269{
1270 intptr_t mask = eio_pagesize () - 1;
1271
1272 /* round down addr */
1273 intptr_t adj = mask & (intptr_t)*addr;
1274
1275 *addr = (void *)((intptr_t)*addr - adj);
1276 *length += adj;
1277
1278 /* round up length */
1279 *length = (*length + mask) & ~mask;
1280}
1281
1282#if !_POSIX_MEMLOCK
1283# define eio__mlockall(a) EIO_ENOSYS ()
1284#else
1285
1286static int
1287eio__mlockall (int flags)
1288{
1289 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1290 extern int mallopt (int, int);
1291 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1292 #endif
1293
1294 if (EIO_MCL_CURRENT != MCL_CURRENT
1295 || EIO_MCL_FUTURE != MCL_FUTURE)
1296 {
1297 flags = 0
1298 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1299 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1300 }
1301
1302 return mlockall (flags);
1303}
1304#endif
1305
1306#if !_POSIX_MEMLOCK_RANGE
1307# define eio__mlock(a,b) EIO_ENOSYS ()
1308#else
1309
1310static int
1311eio__mlock (void *addr, size_t length)
1312{
1313 eio_page_align (&addr, &length);
1314
1315 return mlock (addr, length);
1316}
1317
1318#endif
1319
1320#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1321# define eio__msync(a,b,c) EIO_ENOSYS ()
1322#else
1323
1324static int
1325eio__msync (void *mem, size_t len, int flags)
1326{
1327 eio_page_align (&mem, &len);
1328
1329 if (EIO_MS_ASYNC != MS_SYNC
1330 || EIO_MS_INVALIDATE != MS_INVALIDATE
1331 || EIO_MS_SYNC != MS_SYNC)
1332 {
1333 flags = 0
1334 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1335 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1336 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1337 }
1338
1339 return msync (mem, len, flags);
1340}
1341
1342#endif
1343
1344static int
1345eio__mtouch (eio_req *req)
1346{
1347 void *mem = req->ptr2;
1348 size_t len = req->size;
1349 int flags = req->int1;
1350
1351 eio_page_align (&mem, &len);
1352
1353 {
1354 intptr_t addr = (intptr_t)mem;
1355 intptr_t end = addr + len;
1356 intptr_t page = eio_pagesize ();
1357
1358 if (addr < end)
1359 if (flags & EIO_MT_MODIFY) /* modify */
1360 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1361 else
1362 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1363 }
1364
1365 return 0;
1366}
1367
1368/*****************************************************************************/
1369/* requests implemented outside eio_execute, because they are so large */
1370
1371/* copies some absolute path to tmpbuf */
1372static char *
1373eio__getwd (struct tmpbuf *tmpbuf, eio_wd wd)
1374{
1375 if (wd == EIO_CWD)
1376 return getcwd (tmpbuf->ptr, PATH_MAX);
1377
1378#if HAVE_AT
1379 abort (); /*TODO*/
1380#else
1381 strcpy (tmpbuf->ptr, wd);
1382#endif
1383 return tmpbuf->ptr;
1384}
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 /*D*/ /*TODO: wd ignored */
1400
1401 errno = EINVAL;
1402 if (!rel)
1403 return -1;
1404
1405 errno = ENOENT;
1406 if (!*rel)
1407 return -1;
1408
1409 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1410 tmp1 = res + PATH_MAX;
1411 tmp2 = tmp1 + PATH_MAX;
1412
1413#if 0 /* disabled, the musl way to do things is just too racy */
1414#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1415 /* on linux we may be able to ask the kernel */
1416 {
1417 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1418
1419 if (fd >= 0)
1420 {
1421 sprintf (tmp1, "/proc/self/fd/%d", fd);
1422 req->result = readlink (tmp1, res, PATH_MAX);
1423 close (fd);
1424
1425 /* here we should probably stat the open file and the disk file, to make sure they still match */
1426
1427 if (req->result > 0)
1428 goto done;
1429 }
1430 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1431 return;
1432 }
1433#endif
1434#endif
1435
1436 if (*rel != '/')
1437 {
1438 if (!eio__getwd (tmpbuf, wd))
1439 return -1;
1440
1441 if (res [1]) /* only use if not / */
1442 res += strlen (res);
1443 }
1444
1445 while (*rel)
1446 {
1447 eio_ssize_t len, linklen;
1448 const char *beg = rel;
1449
1450 while (*rel && *rel != '/')
1451 ++rel;
1452
1453 len = rel - beg;
1454
1455 if (!len) /* skip slashes */
1456 {
1457 ++rel;
1458 continue;
1459 }
1460
1461 if (beg [0] == '.')
1462 {
1463 if (len == 1)
1464 continue; /* . - nop */
1465
1466 if (beg [1] == '.' && len == 2)
1467 {
1468 /* .. - back up one component, if possible */
1469
1470 while (res != tmpbuf->ptr)
1471 if (*--res == '/')
1472 break;
1473
1474 continue;
1475 }
1476 }
1477
1478 errno = ENAMETOOLONG;
1479 if (res + 1 + len + 1 >= tmp1)
1480 return;
1481
1482 /* copy one component */
1483 *res = '/';
1484 memcpy (res + 1, beg, len);
1485
1486 /* zero-terminate, for readlink */
1487 res [len + 1] = 0;
1488
1489 /* now check if it's a symlink */
1490 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1491
1492 if (linklen < 0)
1493 {
1494 if (errno != EINVAL)
1495 return -1;
1496
1497 /* it's a normal directory. hopefully */
1498 res += len + 1;
1499 }
1500 else
1501 {
1502 /* yay, it was a symlink - build new path in tmp2 */
1503 int rellen = strlen (rel);
1504
1505 errno = ENAMETOOLONG;
1506 if (linklen + 1 + rellen >= PATH_MAX)
1507 return -1;
1508
1509 errno = ELOOP;
1510 if (!--symlinks)
1511 return -1;
1512
1513 if (*tmp1 == '/')
1514 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1515
1516 /* we need to be careful, as rel might point into tmp2 already */
1517 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1518 tmp2 [linklen] = '/';
1519 memcpy (tmp2, tmp1, linklen);
1520
1521 rel = tmp2;
1522 }
1523 }
1524
1525 /* special case for the lone root path */
1526 if (res == tmpbuf->ptr)
1527 *res++ = '/';
1528
1529 return res - (char *)tmpbuf->ptr;
1120} 1530}
1121 1531
1122static signed char 1532static signed char
1123eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1533eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1124{ 1534{
1132 1542
1133#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1543#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 */ 1544#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1135 1545
1136static void 1546static void
1137eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1547eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1138{ 1548{
1139 unsigned char bits [9 + sizeof (ino_t) * 8]; 1549 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1140 unsigned char *bit = bits; 1550 unsigned char *bit = bits;
1141 1551
1142 assert (CHAR_BIT == 8); 1552 assert (CHAR_BIT == 8);
1143 assert (sizeof (eio_dirent) * 8 < 256); 1553 assert (sizeof (eio_dirent) * 8 < 256);
1144 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1554 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1146 1556
1147 if (size <= EIO_SORT_FAST) 1557 if (size <= EIO_SORT_FAST)
1148 return; 1558 return;
1149 1559
1150 /* first prepare an array of bits to test in our radix sort */ 1560 /* 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 */ 1561 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1152 /* inode_bits must contain all inodes ORed together */ 1562 /* inode_bits must contain all inodes ORed together */
1153 /* which is used to skip bits that are 0 everywhere, which is very common */ 1563 /* which is used to skip bits that are 0 everywhere, which is very common */
1154 { 1564 {
1155 ino_t endianness; 1565 eio_ino_t endianness;
1156 int i, j; 1566 int i, j;
1157 1567
1158 /* we store the byte offset of byte n into byte n of "endianness" */ 1568 /* we store the byte offset of byte n into byte n of "endianness" */
1159 for (i = 0; i < sizeof (ino_t); ++i) 1569 for (i = 0; i < sizeof (eio_ino_t); ++i)
1160 ((unsigned char *)&endianness)[i] = i; 1570 ((unsigned char *)&endianness)[i] = i;
1161 1571
1162 *bit++ = 0; 1572 *bit++ = 0;
1163 1573
1164 for (i = 0; i < sizeof (ino_t); ++i) 1574 for (i = 0; i < sizeof (eio_ino_t); ++i)
1165 { 1575 {
1166 /* shifting off the byte offsets out of "endianness" */ 1576 /* shifting off the byte offsets out of "endianness" */
1167 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1577 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1168 endianness >>= 8; 1578 endianness >>= 8;
1169 1579
1170 for (j = 0; j < 8; ++j) 1580 for (j = 0; j < 8; ++j)
1171 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1581 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1172 *bit++ = offs + j; 1582 *bit++ = offs + j;
1173 } 1583 }
1174 1584
1175 for (j = 0; j < 8; ++j) 1585 for (j = 0; j < 8; ++j)
1176 if (score_bits & (1 << j)) 1586 if (score_bits & (1 << j))
1177 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1587 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1178 } 1588 }
1179 1589
1180 /* now actually do the sorting (a variant of MSD radix sort) */ 1590 /* now actually do the sorting (a variant of MSD radix sort) */
1181 { 1591 {
1182 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1592 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1183 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1593 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1184 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1594 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1185 int stk_idx = 0; 1595 int stk_idx = 0;
1186 1596
1187 base_stk [stk_idx] = dents; 1597 base_stk [stk_idx] = dents;
1188 end_stk [stk_idx] = dents + size; 1598 end_stk [stk_idx] = dents + size;
1189 bit_stk [stk_idx] = bit - 1; 1599 bit_stk [stk_idx] = bit - 1;
1268 } 1678 }
1269 } 1679 }
1270} 1680}
1271 1681
1272static void 1682static void
1273eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1683eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1274{ 1684{
1275 if (size <= 1) 1685 if (size <= 1)
1276 return; /* our insertion sort relies on size > 0 */ 1686 return; /* our insertion sort relies on size > 0 */
1277 1687
1278 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1688 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1284 eio_dent_insertion_sort (dents, size); 1694 eio_dent_insertion_sort (dents, size);
1285} 1695}
1286 1696
1287/* read a full directory */ 1697/* read a full directory */
1288static void 1698static void
1289eio__scandir (eio_req *req, etp_worker *self) 1699eio__scandir (eio_req *req)
1290{ 1700{
1291 DIR *dirp;
1292 EIO_STRUCT_DIRENT *entp;
1293 char *name, *names; 1701 char *name, *names;
1294 int namesalloc = 4096; 1702 int namesalloc = 4096 - sizeof (void *) * 4;
1295 int namesoffs = 0; 1703 int namesoffs = 0;
1296 int flags = req->int1; 1704 int flags = req->int1;
1297 eio_dirent *dents = 0; 1705 eio_dirent *dents = 0;
1298 int dentalloc = 128; 1706 int dentalloc = 128;
1299 int dentoffs = 0; 1707 int dentoffs = 0;
1300 ino_t inode_bits = 0; 1708 eio_ino_t inode_bits = 0;
1709#ifdef _WIN32
1710 HANDLE dirp;
1711 WIN32_FIND_DATA entp;
1712#else
1713 DIR *dirp;
1714 EIO_STRUCT_DIRENT *entp;
1715#endif
1301 1716
1302 req->result = -1; 1717 req->result = -1;
1303 1718
1304 if (!(flags & EIO_READDIR_DENTS)) 1719 if (!(flags & EIO_READDIR_DENTS))
1305 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1720 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1306 1721
1307 X_LOCK (wrklock); 1722#ifdef _WIN32
1308 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1723 {
1724 int len = strlen ((const char *)req->ptr1);
1725 char *path = malloc (MAX_PATH);
1726 const char *fmt;
1727
1728 if (!len)
1729 fmt = "./*";
1730 else if (((const char *)req->ptr1)[len - 1] == '/' || ((const char *)req->ptr1)[len - 1] == '\\')
1731 fmt = "%s*";
1732 else
1733 fmt = "%s/*";
1734
1735 _snprintf (path, MAX_PATH, fmt, (const char *)req->ptr1);
1736 dirp = FindFirstFile (path, &entp);
1737 free (path);
1738
1739 if (dirp == INVALID_HANDLE_VALUE)
1740 {
1741 /* should steal _dosmaperr */
1742 switch (GetLastError ())
1743 {
1744 case ERROR_FILE_NOT_FOUND:
1745 req->result = 0;
1746 break;
1747
1748 case ERROR_INVALID_NAME:
1749 case ERROR_PATH_NOT_FOUND:
1750 case ERROR_NO_MORE_FILES:
1751 errno = ENOENT;
1752 break;
1753
1754 case ERROR_NOT_ENOUGH_MEMORY:
1755 errno = ENOMEM;
1756 break;
1757
1758 default:
1759 errno = EINVAL;
1760 break;
1761 }
1762
1763 return;
1764 }
1765 }
1766#else
1309 self->dirp = dirp = opendir (req->ptr1); 1767 dirp = opendir (req->ptr1);
1768
1769 if (!dirp)
1770 return;
1771#endif
1310 1772
1311 if (req->flags & EIO_FLAG_PTR1_FREE) 1773 if (req->flags & EIO_FLAG_PTR1_FREE)
1312 free (req->ptr1); 1774 free (req->ptr1);
1313 1775
1314 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1776 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1315 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1777 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1316 req->ptr2 = names = malloc (namesalloc); 1778 req->ptr2 = names = malloc (namesalloc);
1317 X_UNLOCK (wrklock);
1318 1779
1319 if (dirp && names && (!flags || dents)) 1780 if (!names || (flags && !dents))
1781 return;
1782
1320 for (;;) 1783 for (;;)
1321 { 1784 {
1785 int done;
1786
1787#ifdef _WIN32
1788 done = !dirp;
1789#else
1322 errno = 0; 1790 errno = 0;
1323 entp = readdir (dirp); 1791 entp = readdir (dirp);
1792 done = !entp;
1793#endif
1324 1794
1325 if (!entp) 1795 if (done)
1326 { 1796 {
1797#ifndef _WIN32
1798 int old_errno = errno;
1799 closedir (dirp);
1800 errno = old_errno;
1801
1327 if (errno) 1802 if (errno)
1328 break; 1803 break;
1804#endif
1329 1805
1330 /* sort etc. */ 1806 /* sort etc. */
1331 req->int1 = flags; 1807 req->int1 = flags;
1332 req->result = dentoffs; 1808 req->result = dentoffs;
1333 1809
1334 if (flags & EIO_READDIR_STAT_ORDER) 1810 if (flags & EIO_READDIR_STAT_ORDER)
1335 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1811 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1336 else if (flags & EIO_READDIR_DIRS_FIRST) 1812 else if (flags & EIO_READDIR_DIRS_FIRST)
1337 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1813 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1338 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1814 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1339 else 1815 else
1816 {
1817 /* in this case, all is known, and we just put dirs first and sort them */
1818 eio_dirent *oth = dents + dentoffs;
1819 eio_dirent *dir = dents;
1820
1821 /* now partition dirs to the front, and non-dirs to the back */
1822 /* by walking from both sides and swapping if necessary */
1823 while (oth > dir)
1824 {
1825 if (dir->type == EIO_DT_DIR)
1826 ++dir;
1827 else if ((--oth)->type == EIO_DT_DIR)
1828 {
1829 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1830
1831 ++dir;
1832 }
1833 }
1834
1835 /* now sort the dirs only (dirs all have the same score) */
1836 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1837 }
1838
1839 break;
1840 }
1841
1842 /* now add the entry to our list(s) */
1843 name = D_NAME (entp);
1844
1845 /* skip . and .. entries */
1846 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1847 {
1848 int len = D_NAMLEN (entp) + 1;
1849
1850 while (ecb_expect_false (namesoffs + len > namesalloc))
1851 {
1852 namesalloc *= 2;
1853 req->ptr2 = names = realloc (names, namesalloc);
1854
1855 if (!names)
1856 break;
1857 }
1858
1859 memcpy (names + namesoffs, name, len);
1860
1861 if (dents)
1862 {
1863 struct eio_dirent *ent;
1864
1865 if (ecb_expect_false (dentoffs == dentalloc))
1340 { 1866 {
1341 /* in this case, all is known, and we just put dirs first and sort them */ 1867 dentalloc *= 2;
1868 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1869
1870 if (!dents)
1871 break;
1872 }
1873
1342 eio_dirent *oth = dents + dentoffs; 1874 ent = dents + dentoffs;
1343 eio_dirent *dir = dents;
1344 1875
1345 /* now partition dirs to the front, and non-dirs to the back */ 1876 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1346 /* by walking from both sides and swapping if necessary */ 1877 ent->namelen = len - 1;
1347 while (oth > dir) 1878 ent->inode = D_INO (entp);
1879
1880 inode_bits |= ent->inode;
1881
1882 switch (D_TYPE (entp))
1883 {
1884 default:
1885 ent->type = EIO_DT_UNKNOWN;
1886 flags |= EIO_READDIR_FOUND_UNKNOWN;
1887 break;
1888
1889 #ifdef DT_FIFO
1890 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1891 #endif
1892 #ifdef DT_CHR
1893 case DT_CHR: ent->type = EIO_DT_CHR; break;
1894 #endif
1895 #ifdef DT_MPC
1896 case DT_MPC: ent->type = EIO_DT_MPC; break;
1897 #endif
1898 #ifdef DT_DIR
1899 case DT_DIR: ent->type = EIO_DT_DIR; break;
1900 #endif
1901 #ifdef DT_NAM
1902 case DT_NAM: ent->type = EIO_DT_NAM; break;
1903 #endif
1904 #ifdef DT_BLK
1905 case DT_BLK: ent->type = EIO_DT_BLK; break;
1906 #endif
1907 #ifdef DT_MPB
1908 case DT_MPB: ent->type = EIO_DT_MPB; break;
1909 #endif
1910 #ifdef DT_REG
1911 case DT_REG: ent->type = EIO_DT_REG; break;
1912 #endif
1913 #ifdef DT_NWK
1914 case DT_NWK: ent->type = EIO_DT_NWK; break;
1915 #endif
1916 #ifdef DT_CMP
1917 case DT_CMP: ent->type = EIO_DT_CMP; break;
1918 #endif
1919 #ifdef DT_LNK
1920 case DT_LNK: ent->type = EIO_DT_LNK; break;
1921 #endif
1922 #ifdef DT_SOCK
1923 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1924 #endif
1925 #ifdef DT_DOOR
1926 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1927 #endif
1928 #ifdef DT_WHT
1929 case DT_WHT: ent->type = EIO_DT_WHT; break;
1930 #endif
1931 }
1932
1933 ent->score = 7;
1934
1935 if (flags & EIO_READDIR_DIRS_FIRST)
1936 {
1937 if (ent->type == EIO_DT_UNKNOWN)
1348 { 1938 {
1349 if (dir->type == EIO_DT_DIR) 1939 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1350 ++dir; 1940 ent->score = 1;
1351 else if ((--oth)->type == EIO_DT_DIR) 1941 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1352 { 1942 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1353 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1354
1355 ++dir;
1356 }
1357 } 1943 }
1358 1944 else if (ent->type == EIO_DT_DIR)
1359 /* now sort the dirs only (dirs all have the same score) */ 1945 ent->score = 0;
1360 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1361 } 1946 }
1362
1363 break;
1364 }
1365
1366 /* now add the entry to our list(s) */
1367 name = entp->d_name;
1368
1369 /* skip . and .. entries */
1370 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1371 {
1372 int len = D_NAMLEN (entp) + 1;
1373
1374 while (ecb_expect_false (namesoffs + len > namesalloc))
1375 {
1376 namesalloc *= 2;
1377 X_LOCK (wrklock);
1378 req->ptr2 = names = realloc (names, namesalloc);
1379 X_UNLOCK (wrklock);
1380
1381 if (!names)
1382 break;
1383 } 1947 }
1384 1948
1385 memcpy (names + namesoffs, name, len);
1386
1387 if (dents)
1388 {
1389 struct eio_dirent *ent;
1390
1391 if (ecb_expect_false (dentoffs == dentalloc))
1392 {
1393 dentalloc *= 2;
1394 X_LOCK (wrklock);
1395 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1396 X_UNLOCK (wrklock);
1397
1398 if (!dents)
1399 break;
1400 }
1401
1402 ent = dents + dentoffs;
1403
1404 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1405 ent->namelen = len - 1;
1406 ent->inode = D_INO (entp);
1407
1408 inode_bits |= ent->inode;
1409
1410 switch (D_TYPE (entp))
1411 {
1412 default:
1413 ent->type = EIO_DT_UNKNOWN;
1414 flags |= EIO_READDIR_FOUND_UNKNOWN;
1415 break;
1416
1417 #ifdef DT_FIFO
1418 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1419 #endif
1420 #ifdef DT_CHR
1421 case DT_CHR: ent->type = EIO_DT_CHR; break;
1422 #endif
1423 #ifdef DT_MPC
1424 case DT_MPC: ent->type = EIO_DT_MPC; break;
1425 #endif
1426 #ifdef DT_DIR
1427 case DT_DIR: ent->type = EIO_DT_DIR; break;
1428 #endif
1429 #ifdef DT_NAM
1430 case DT_NAM: ent->type = EIO_DT_NAM; break;
1431 #endif
1432 #ifdef DT_BLK
1433 case DT_BLK: ent->type = EIO_DT_BLK; break;
1434 #endif
1435 #ifdef DT_MPB
1436 case DT_MPB: ent->type = EIO_DT_MPB; break;
1437 #endif
1438 #ifdef DT_REG
1439 case DT_REG: ent->type = EIO_DT_REG; break;
1440 #endif
1441 #ifdef DT_NWK
1442 case DT_NWK: ent->type = EIO_DT_NWK; break;
1443 #endif
1444 #ifdef DT_CMP
1445 case DT_CMP: ent->type = EIO_DT_CMP; break;
1446 #endif
1447 #ifdef DT_LNK
1448 case DT_LNK: ent->type = EIO_DT_LNK; break;
1449 #endif
1450 #ifdef DT_SOCK
1451 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1452 #endif
1453 #ifdef DT_DOOR
1454 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1455 #endif
1456 #ifdef DT_WHT
1457 case DT_WHT: ent->type = EIO_DT_WHT; break;
1458 #endif
1459 }
1460
1461 ent->score = 7;
1462
1463 if (flags & EIO_READDIR_DIRS_FIRST)
1464 {
1465 if (ent->type == EIO_DT_UNKNOWN)
1466 {
1467 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1468 ent->score = 1;
1469 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1470 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1471 }
1472 else if (ent->type == EIO_DT_DIR)
1473 ent->score = 0;
1474 }
1475 }
1476
1477 namesoffs += len; 1949 namesoffs += len;
1478 ++dentoffs; 1950 ++dentoffs;
1479 } 1951 }
1480 1952
1481 if (EIO_CANCELLED (req)) 1953 if (EIO_CANCELLED (req))
1482 { 1954 {
1483 errno = ECANCELED; 1955 errno = ECANCELED;
1484 break; 1956 break;
1485 } 1957 }
1958
1959#ifdef _WIN32
1960 if (!FindNextFile (dirp, &entp))
1961 {
1962 FindClose (dirp);
1963 dirp = 0;
1486 } 1964 }
1965#endif
1966 }
1487} 1967}
1488 1968
1489#ifdef PAGESIZE 1969/*****************************************************************************/
1490# define eio_pagesize() PAGESIZE 1970/* working directory stuff */
1971
1972#if HAVE_AT
1973
1974#define WD2FD(wd) ((wd) ? ((int)(long)(wd)) - 1 : AT_FDCWD)
1975
1976#ifndef O_SEARCH
1977# define O_SEARCH O_RDONLY
1978#endif
1979
1980eio_wd
1981eio_wd_open_sync (eio_wd wd, const char *path)
1982{
1983 int fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1984
1985 return fd >= 0 ? (eio_wd)(long)(fd + 1) : EIO_INVALID_WD;
1986}
1987
1988static eio_wd
1989eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1990{
1991 return eio_wd_open_sync (wd, path);
1992}
1993
1994void
1995eio_wd_close_sync (eio_wd wd)
1996{
1997 int fd = WD2FD (wd);
1998
1999 if (fd >= 0)
2000 close (fd);
2001}
2002
2003static int
2004eio__truncateat (int dirfd, const char *path, off_t length)
2005{
2006 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2007 int res;
2008
2009 if (fd < 0)
2010 return fd;
2011
2012 res = ftruncate (fd, length);
2013 close (fd);
2014 return res;
2015}
2016
2017static int
2018eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2019{
2020 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2021 int res;
2022
2023 if (fd < 0)
2024 return fd;
2025
2026 res = fstatvfs (fd, buf);
2027 close (fd);
2028 return res;
2029
2030}
2031
1491#else 2032#else
1492static intptr_t
1493eio_pagesize (void)
1494{
1495 static intptr_t page;
1496 2033
1497 if (!page) 2034/* on legacy systems, we represent the working directories simply by their path strings */
1498 page = sysconf (_SC_PAGESIZE);
1499 2035
2036static const char *
2037wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2038{
2039 if (!wd || *path == '/')
1500 return page; 2040 return path;
1501}
1502#endif
1503 2041
2042 {
2043 int l1 = strlen ((const char *)wd);
2044 int l2 = strlen (path);
2045
2046 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2047
2048 memcpy (res, wd, l1);
2049 res [l1] = '/';
2050 memcpy (res + l1 + 1, path, l2 + 1);
2051
2052 return res;
2053 }
2054}
2055
1504static void 2056static eio_wd
1505eio_page_align (void **addr, size_t *length) 2057eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1506{ 2058{
1507 intptr_t mask = eio_pagesize () - 1; 2059 if (*path == '/') /* absolute paths ignore wd */
1508 2060 path = strdup (path);
1509 /* round down addr */ 2061 else if (path [0] == '.' && !path [1]) /* special case '.', as it is common */
1510 intptr_t adj = mask & (intptr_t)*addr; 2062 return wd;
1511 2063 else
1512 *addr = (void *)((intptr_t)*addr - adj);
1513 *length += adj;
1514
1515 /* round up length */
1516 *length = (*length + mask) & ~mask;
1517}
1518
1519#if !_POSIX_MEMLOCK
1520# define eio__mlockall(a) ((errno = ENOSYS), -1)
1521#else
1522
1523static int
1524eio__mlockall (int flags)
1525{
1526 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
1527 extern int mallopt (int, int);
1528 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1529 #endif
1530
1531 if (EIO_MCL_CURRENT != MCL_CURRENT
1532 || EIO_MCL_FUTURE != MCL_FUTURE)
1533 { 2064 {
1534 flags = 0 2065 int len = eio__realpath (tmpbuf, wd, path);
1535 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1536 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1537 }
1538 2066
1539 return mlockall (flags); 2067 path = EIO_INVALID_WD;
1540}
1541#endif
1542 2068
1543#if !_POSIX_MEMLOCK_RANGE 2069 if (len >= 0)
1544# define eio__mlock(a,b) ((errno = ENOSYS), -1) 2070 {
1545#else 2071 ((char *)tmpbuf->ptr)[len] = 0;
1546 2072 path = strdup (tmpbuf->ptr);
1547static int 2073 }
1548eio__mlock (void *addr, size_t length)
1549{
1550 eio_page_align (&addr, &length);
1551
1552 return mlock (addr, length);
1553}
1554
1555#endif
1556
1557#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1558# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1559#else
1560
1561static int
1562eio__msync (void *mem, size_t len, int flags)
1563{
1564 eio_page_align (&mem, &len);
1565
1566 if (EIO_MS_ASYNC != MS_SYNC
1567 || EIO_MS_INVALIDATE != MS_INVALIDATE
1568 || EIO_MS_SYNC != MS_SYNC)
1569 { 2074 }
1570 flags = 0
1571 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1572 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1573 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1574 }
1575 2075
1576 return msync (mem, len, flags); 2076 if (!path)
1577} 2077 path = EIO_INVALID_WD;
1578 2078
1579#endif 2079 return (eio_wd)path;
2080}
1580 2081
1581static int 2082eio_wd
1582eio__mtouch (eio_req *req) 2083eio_wd_open_sync (eio_wd wd, const char *path)
1583{ 2084{
1584 void *mem = req->ptr2; 2085 struct tmpbuf tmpbuf = { 0 };
1585 size_t len = req->size; 2086 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1586 int flags = req->int1; 2087 free (tmpbuf.ptr);
1587 2088
1588 eio_page_align (&mem, &len);
1589
1590 {
1591 intptr_t addr = (intptr_t)mem;
1592 intptr_t end = addr + len;
1593 intptr_t page = eio_pagesize ();
1594
1595 if (addr < end)
1596 if (flags & EIO_MT_MODIFY) /* modify */
1597 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1598 else
1599 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1600 }
1601
1602 return 0; 2089 return wd;
1603} 2090}
2091
2092void
2093eio_wd_close_sync (eio_wd wd)
2094{
2095 if (wd != EIO_INVALID_WD)
2096 free (wd);
2097}
2098
2099#endif
1604 2100
1605/*****************************************************************************/ 2101/*****************************************************************************/
1606 2102
1607#define ALLOC(len) \ 2103#define ALLOC(len) \
1608 if (!req->ptr2) \ 2104 if (!req->ptr2) \
1623{ 2119{
1624 ETP_REQ *req; 2120 ETP_REQ *req;
1625 struct timespec ts; 2121 struct timespec ts;
1626 etp_worker *self = (etp_worker *)thr_arg; 2122 etp_worker *self = (etp_worker *)thr_arg;
1627 2123
2124#if HAVE_PRCTL_SET_NAME
2125 prctl (PR_SET_NAME, (unsigned long)"eio_thread", 0, 0, 0);
2126#endif
2127
1628 /* try to distribute timeouts somewhat randomly */ 2128 /* try to distribute timeouts somewhat evenly */
1629 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2129 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1630 2130
1631 for (;;) 2131 for (;;)
1632 { 2132 {
2133 ts.tv_sec = 0;
2134
1633 X_LOCK (reqlock); 2135 X_LOCK (reqlock);
1634 2136
1635 for (;;) 2137 for (;;)
1636 { 2138 {
1637 self->req = req = reqq_shift (&req_queue); 2139 req = reqq_shift (&req_queue);
1638 2140
1639 if (req) 2141 if (req)
1640 break; 2142 break;
1641 2143
2144 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2145 {
2146 X_UNLOCK (reqlock);
2147 X_LOCK (wrklock);
2148 --started;
2149 X_UNLOCK (wrklock);
2150 goto quit;
2151 }
2152
1642 ++idle; 2153 ++idle;
1643 2154
1644 ts.tv_sec = time (0) + idle_timeout; 2155 if (idle <= max_idle)
1645 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2156 /* we are allowed to idle, so do so without any timeout */
2157 X_COND_WAIT (reqwait, reqlock);
2158 else
1646 { 2159 {
1647 if (idle > max_idle) 2160 /* initialise timeout once */
1648 { 2161 if (!ts.tv_sec)
1649 --idle; 2162 ts.tv_sec = time (0) + idle_timeout;
1650 X_UNLOCK (reqlock);
1651 X_LOCK (wrklock);
1652 --started;
1653 X_UNLOCK (wrklock);
1654 goto quit;
1655 }
1656 2163
1657 /* we are allowed to idle, so do so without any timeout */
1658 X_COND_WAIT (reqwait, reqlock); 2164 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2165 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1659 } 2166 }
1660 2167
1661 --idle; 2168 --idle;
1662 } 2169 }
1663 2170
1666 X_UNLOCK (reqlock); 2173 X_UNLOCK (reqlock);
1667 2174
1668 if (req->type < 0) 2175 if (req->type < 0)
1669 goto quit; 2176 goto quit;
1670 2177
1671 if (!EIO_CANCELLED (req))
1672 ETP_EXECUTE (self, req); 2178 ETP_EXECUTE (self, req);
1673 2179
1674 X_LOCK (reslock); 2180 X_LOCK (reslock);
1675 2181
1676 ++npending; 2182 ++npending;
1677 2183
1678 if (!reqq_push (&res_queue, req) && want_poll_cb) 2184 if (!reqq_push (&res_queue, req) && want_poll_cb)
1679 want_poll_cb (); 2185 want_poll_cb ();
1680 2186
1681 self->req = 0;
1682 etp_worker_clear (self); 2187 etp_worker_clear (self);
1683 2188
1684 X_UNLOCK (reslock); 2189 X_UNLOCK (reslock);
1685 } 2190 }
1686 2191
1687quit: 2192quit:
2193 free (req);
2194
1688 X_LOCK (wrklock); 2195 X_LOCK (wrklock);
1689 etp_worker_free (self); 2196 etp_worker_free (self);
1690 X_UNLOCK (wrklock); 2197 X_UNLOCK (wrklock);
1691 2198
1692 return 0; 2199 return 0;
1695/*****************************************************************************/ 2202/*****************************************************************************/
1696 2203
1697int ecb_cold 2204int ecb_cold
1698eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2205eio_init (void (*want_poll)(void), void (*done_poll)(void))
1699{ 2206{
2207#if !HAVE_PREADWRITE
2208 X_MUTEX_CREATE (preadwritelock);
2209#endif
2210
1700 return etp_init (want_poll, done_poll); 2211 return etp_init (want_poll, done_poll);
1701} 2212}
1702 2213
1703ecb_inline void 2214ecb_inline void
1704eio_api_destroy (eio_req *req) 2215eio_api_destroy (eio_req *req)
1731 } 2242 }
1732 2243
1733static void 2244static void
1734eio_execute (etp_worker *self, eio_req *req) 2245eio_execute (etp_worker *self, eio_req *req)
1735{ 2246{
2247#if HAVE_AT
2248 int dirfd;
2249#else
2250 const char *path;
2251#endif
2252
2253 if (ecb_expect_false (EIO_CANCELLED (req)))
2254 {
2255 req->result = -1;
2256 req->errorno = ECANCELED;
2257 return;
2258 }
2259
2260 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2261 {
2262 req->result = -1;
2263 req->errorno = ENOENT;
2264 return;
2265 }
2266
2267 if (req->type >= EIO_OPEN)
2268 {
2269 #if HAVE_AT
2270 dirfd = WD2FD (req->wd);
2271 #else
2272 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2273 #endif
2274 }
2275
1736 switch (req->type) 2276 switch (req->type)
1737 { 2277 {
2278 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2279 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2280 break;
2281 case EIO_WD_CLOSE: req->result = 0;
2282 eio_wd_close_sync (req->wd); break;
2283
1738 case EIO_READ: ALLOC (req->size); 2284 case EIO_READ: ALLOC (req->size);
1739 req->result = req->offs >= 0 2285 req->result = req->offs >= 0
1740 ? pread (req->int1, req->ptr2, req->size, req->offs) 2286 ? pread (req->int1, req->ptr2, req->size, req->offs)
1741 : read (req->int1, req->ptr2, req->size); break; 2287 : read (req->int1, req->ptr2, req->size); break;
1742 case EIO_WRITE: req->result = req->offs >= 0 2288 case EIO_WRITE: req->result = req->offs >= 0
1743 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2289 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1744 : write (req->int1, req->ptr2, req->size); break; 2290 : write (req->int1, req->ptr2, req->size); break;
1745 2291
1746 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2292 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1747 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2293 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2294
2295#if HAVE_AT
1748 2296
1749 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2297 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1750 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2298 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1751 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2299 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2300 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2301 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2302 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2303 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2304 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2305
2306 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2307 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2308 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2309 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2); break;
2310 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD (req->int3), req->ptr2, 0); break;
2311 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2312 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2313 case EIO_READLINK: ALLOC (PATH_MAX);
2314 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2315 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2316 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2317 case EIO_UTIME:
2318 case EIO_FUTIME:
2319 {
2320 struct timespec ts[2];
2321 struct timespec *times;
2322
2323 if (req->nv1 != -1. || req->nv2 != -1.)
2324 {
2325 ts[0].tv_sec = req->nv1;
2326 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2327 ts[1].tv_sec = req->nv2;
2328 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2329
2330 times = ts;
2331 }
2332 else
2333 times = 0;
2334
2335 req->result = req->type == EIO_FUTIME
2336 ? futimens (req->int1, times)
2337 : utimensat (dirfd, req->ptr1, times, 0);
2338 }
2339 break;
2340
2341#else
2342
2343 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2344 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2345 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1752 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2346 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2347 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2348 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2349 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2350 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2351
2352 case EIO_UNLINK: req->result = unlink (path ); break;
2353 case EIO_RMDIR: req->result = rmdir (path ); break;
2354 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2355 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2356 case EIO_LINK: req->result = link (path , req->ptr2); break;
2357 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2358 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2359 case EIO_READLINK: ALLOC (PATH_MAX);
2360 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2361 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2362 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2363
2364 case EIO_UTIME:
2365 case EIO_FUTIME:
2366 {
2367 struct timeval tv[2];
2368 struct timeval *times;
2369
2370 if (req->nv1 != -1. || req->nv2 != -1.)
2371 {
2372 tv[0].tv_sec = req->nv1;
2373 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2374 tv[1].tv_sec = req->nv2;
2375 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2376
2377 times = tv;
2378 }
2379 else
2380 times = 0;
2381
2382 req->result = req->type == EIO_FUTIME
2383 ? futimes (req->int1, times)
2384 : utimes (req->ptr1, times);
2385 }
2386 break;
2387
2388#endif
2389
2390 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2391 {
2392 ALLOC (req->result);
2393 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2394 }
2395 break;
2396
1753 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2397 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1754 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2398 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1755 2399
1756 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1757 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1758 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2400 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1759 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2401 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1760 2402
1761 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1762 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2403 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1763 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1764 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2404 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1765 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1766 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2405 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1767 2406
1768 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1769 case EIO_CLOSE: req->result = close (req->int1); break; 2407 case EIO_CLOSE: req->result = close (req->int1); break;
1770 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2408 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1771 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1772 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1773 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1774 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1775 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1776 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1777 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1778
1779 case EIO_READLINK: ALLOC (PATH_MAX);
1780 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1781
1782 case EIO_SYNC: req->result = 0; sync (); break; 2409 case EIO_SYNC: req->result = 0; sync (); break;
1783 case EIO_FSYNC: req->result = fsync (req->int1); break; 2410 case EIO_FSYNC: req->result = fsync (req->int1); break;
1784 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2411 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2412 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2413 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1785 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2414 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1786 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2415 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1787 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2416 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1788 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2417 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1789 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2418 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1790 2419
1791 case EIO_READDIR: eio__scandir (req, self); break; 2420 case EIO_READDIR: eio__scandir (req); break;
1792 2421
1793 case EIO_BUSY: 2422 case EIO_BUSY:
1794#ifdef _WIN32 2423#ifdef _WIN32
1795 Sleep (req->nv1 * 1e3); 2424 Sleep (req->nv1 * 1e3);
1796#else 2425#else
1803 req->result = select (0, 0, 0, 0, &tv); 2432 req->result = select (0, 0, 0, 0, &tv);
1804 } 2433 }
1805#endif 2434#endif
1806 break; 2435 break;
1807 2436
1808 case EIO_UTIME:
1809 case EIO_FUTIME:
1810 {
1811 struct timeval tv[2];
1812 struct timeval *times;
1813
1814 if (req->nv1 != -1. || req->nv2 != -1.)
1815 {
1816 tv[0].tv_sec = req->nv1;
1817 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1818 tv[1].tv_sec = req->nv2;
1819 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1820
1821 times = tv;
1822 }
1823 else
1824 times = 0;
1825
1826 req->result = req->type == EIO_FUTIME
1827 ? futimes (req->int1, times)
1828 : utimes (req->ptr1, times);
1829 }
1830 break;
1831
1832 case EIO_GROUP: 2437 case EIO_GROUP:
1833 abort (); /* handled in eio_request */ 2438 abort (); /* handled in eio_request */
1834 2439
1835 case EIO_NOP: 2440 case EIO_NOP:
1836 req->result = 0; 2441 req->result = 0;
1874eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2479eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1875{ 2480{
1876 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2481 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1877} 2482}
1878 2483
2484eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2485{
2486 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2487}
2488
2489eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2490{
2491 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2492}
2493
2494eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2495{
2496 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2497}
2498
1879eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2499eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1880{ 2500{
1881 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2501 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1882} 2502}
1883 2503
1889eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2509eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1890{ 2510{
1891 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2511 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1892} 2512}
1893 2513
1894eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2514eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1895{ 2515{
1896 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2516 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1897}
1898
1899eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1900{
1901 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1902} 2517}
1903 2518
1904eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2519eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1905{ 2520{
1906 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2521 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1944eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2559eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1945{ 2560{
1946 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2561 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1947} 2562}
1948 2563
1949eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2564eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1950{ 2565{
1951 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2566 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1952} 2567}
1953 2568
1954eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2569eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1974eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2589eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1975{ 2590{
1976 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2591 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1977} 2592}
1978 2593
1979eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2594eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1980{ 2595{
1981 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2596 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1982} 2597}
1983 2598
1984eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2599eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1998} 2613}
1999 2614
2000eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2615eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
2001{ 2616{
2002 return eio__1path (EIO_READLINK, path, pri, cb, data); 2617 return eio__1path (EIO_READLINK, path, pri, cb, data);
2618}
2619
2620eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2621{
2622 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2003} 2623}
2004 2624
2005eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2625eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
2006{ 2626{
2007 return eio__1path (EIO_STAT, path, pri, cb, data); 2627 return eio__1path (EIO_STAT, path, pri, cb, data);
2126} 2746}
2127 2747
2128/*****************************************************************************/ 2748/*****************************************************************************/
2129/* misc garbage */ 2749/* misc garbage */
2130 2750
2131ssize_t 2751eio_ssize_t
2132eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2752eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2133{ 2753{
2134 etp_worker wrk;
2135 ssize_t ret;
2136
2137 wrk.dbuf = 0;
2138
2139 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2754 return eio__sendfile (ofd, ifd, offset, count);
2140
2141 if (wrk.dbuf)
2142 free (wrk.dbuf);
2143
2144 return ret;
2145} 2755}
2146 2756

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines