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

Comparing cvsroot/libeio/eio.c (file contents):
Revision 1.79 by root, Tue Jul 5 16:57:41 2011 UTC vs.
Revision 1.114 by root, Fri Dec 30 07:38:00 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 */
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
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
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
1006 1148
1007 /* according to source inspection, this is correct, and useful behaviour */ 1149 /* according to source inspection, this is correct, and useful behaviour */
1008 if (sbytes) 1150 if (sbytes)
1009 res = sbytes; 1151 res = sbytes;
1010 1152
1011# elif defined (__APPLE__) && 0 /* broken, as everything on os x */ 1153# elif defined (__APPLE__)
1012 off_t sbytes = count; 1154 off_t sbytes = count;
1013 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 1155 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1014 1156
1015 /* according to the manpage, sbytes is always valid */ 1157 /* according to the manpage, sbytes is always valid */
1016 if (sbytes) 1158 if (sbytes)
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;
1120} 1266}
1121 1267
1148 /* round up length */ 1294 /* round up length */
1149 *length = (*length + mask) & ~mask; 1295 *length = (*length + mask) & ~mask;
1150} 1296}
1151 1297
1152#if !_POSIX_MEMLOCK 1298#if !_POSIX_MEMLOCK
1153# define eio__mlockall(a) ((errno = ENOSYS), -1) 1299# define eio__mlockall(a) EIO_ENOSYS ()
1154#else 1300#else
1155 1301
1156static int 1302static int
1157eio__mlockall (int flags) 1303eio__mlockall (int flags)
1158{ 1304{
1172 return mlockall (flags); 1318 return mlockall (flags);
1173} 1319}
1174#endif 1320#endif
1175 1321
1176#if !_POSIX_MEMLOCK_RANGE 1322#if !_POSIX_MEMLOCK_RANGE
1177# define eio__mlock(a,b) ((errno = ENOSYS), -1) 1323# define eio__mlock(a,b) EIO_ENOSYS ()
1178#else 1324#else
1179 1325
1180static int 1326static int
1181eio__mlock (void *addr, size_t length) 1327eio__mlock (void *addr, size_t length)
1182{ 1328{
1186} 1332}
1187 1333
1188#endif 1334#endif
1189 1335
1190#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO) 1336#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1191# define eio__msync(a,b,c) ((errno = ENOSYS), -1) 1337# define eio__msync(a,b,c) EIO_ENOSYS ()
1192#else 1338#else
1193 1339
1194static int 1340static int
1195eio__msync (void *mem, size_t len, int flags) 1341eio__msync (void *mem, size_t len, int flags)
1196{ 1342{
1236} 1382}
1237 1383
1238/*****************************************************************************/ 1384/*****************************************************************************/
1239/* requests implemented outside eio_execute, because they are so large */ 1385/* requests implemented outside eio_execute, because they are so large */
1240 1386
1387/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
1241static void 1388static int
1242eio__realpath (eio_req *req, etp_worker *self) 1389eio__realpath (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
1243{ 1390{
1244 char *rel = req->ptr1; 1391 const char *rel = path;
1245 char *res; 1392 char *res;
1246 char *tmp1, *tmp2; 1393 char *tmp1, *tmp2;
1247#if SYMLOOP_MAX > 32 1394#if SYMLOOP_MAX > 32
1248 int symlinks = SYMLOOP_MAX; 1395 int symlinks = SYMLOOP_MAX;
1249#else 1396#else
1250 int symlinks = 32; 1397 int symlinks = 32;
1251#endif 1398#endif
1252 1399
1253 req->result = -1;
1254
1255 errno = EINVAL; 1400 errno = EINVAL;
1256 if (!rel) 1401 if (!rel)
1257 return; 1402 return -1;
1258 1403
1259 errno = ENOENT; 1404 errno = ENOENT;
1260 if (!*rel) 1405 if (!*rel)
1261 return; 1406 return -1;
1262 1407
1263 if (!req->ptr2) 1408 res = tmpbuf_get (tmpbuf, PATH_MAX * 3);
1264 {
1265 X_LOCK (wrklock);
1266 req->flags |= EIO_FLAG_PTR2_FREE;
1267 X_UNLOCK (wrklock);
1268 req->ptr2 = malloc (PATH_MAX * 3);
1269
1270 errno = ENOMEM;
1271 if (!req->ptr2)
1272 return;
1273 }
1274
1275 res = req->ptr2;
1276 tmp1 = res + PATH_MAX; 1409 tmp1 = res + PATH_MAX;
1277 tmp2 = tmp1 + PATH_MAX; 1410 tmp2 = tmp1 + PATH_MAX;
1278 1411
1279#if 0 /* disabled, the musl way to do things is just too racy */ 1412#if 0 /* disabled, the musl way to do things is just too racy */
1280#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1413#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1299#endif 1432#endif
1300#endif 1433#endif
1301 1434
1302 if (*rel != '/') 1435 if (*rel != '/')
1303 { 1436 {
1437 int len;
1438
1439 errno = ENOENT;
1440 if (wd == EIO_INVALID_WD)
1441 return -1;
1442
1443 if (wd == EIO_CWD)
1444 {
1304 if (!getcwd (res, PATH_MAX)) 1445 if (!getcwd (res, PATH_MAX))
1305 return; 1446 return -1;
1447
1448 len = strlen (res);
1449 }
1450 else
1451 memcpy (res, wd->str, len = wd->len);
1306 1452
1307 if (res [1]) /* only use if not / */ 1453 if (res [1]) /* only use if not / */
1308 res += strlen (res); 1454 res += len;
1309 } 1455 }
1310 1456
1311 while (*rel) 1457 while (*rel)
1312 { 1458 {
1313 ssize_t len, linklen; 1459 eio_ssize_t len, linklen;
1314 char *beg = rel; 1460 const char *beg = rel;
1315 1461
1316 while (*rel && *rel != '/') 1462 while (*rel && *rel != '/')
1317 ++rel; 1463 ++rel;
1318 1464
1319 len = rel - beg; 1465 len = rel - beg;
1331 1477
1332 if (beg [1] == '.' && len == 2) 1478 if (beg [1] == '.' && len == 2)
1333 { 1479 {
1334 /* .. - back up one component, if possible */ 1480 /* .. - back up one component, if possible */
1335 1481
1336 while (res != req->ptr2) 1482 while (res != tmpbuf->ptr)
1337 if (*--res == '/') 1483 if (*--res == '/')
1338 break; 1484 break;
1339 1485
1340 continue; 1486 continue;
1341 } 1487 }
1342 } 1488 }
1343 1489
1344 errno = ENAMETOOLONG; 1490 errno = ENAMETOOLONG;
1345 if (res + 1 + len + 1 >= tmp1) 1491 if (res + 1 + len + 1 >= tmp1)
1346 return; 1492 return -1;
1347 1493
1348 /* copy one component */ 1494 /* copy one component */
1349 *res = '/'; 1495 *res = '/';
1350 memcpy (res + 1, beg, len); 1496 memcpy (res + 1, beg, len);
1351 1497
1352 /* zero-terminate, for readlink */ 1498 /* zero-terminate, for readlink */
1353 res [len + 1] = 0; 1499 res [len + 1] = 0;
1354 1500
1355 /* now check if it's a symlink */ 1501 /* now check if it's a symlink */
1356 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1502 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1357 1503
1358 if (linklen < 0) 1504 if (linklen < 0)
1359 { 1505 {
1360 if (errno != EINVAL) 1506 if (errno != EINVAL)
1361 return; 1507 return -1;
1362 1508
1363 /* it's a normal directory. hopefully */ 1509 /* it's a normal directory. hopefully */
1364 res += len + 1; 1510 res += len + 1;
1365 } 1511 }
1366 else 1512 else
1368 /* yay, it was a symlink - build new path in tmp2 */ 1514 /* yay, it was a symlink - build new path in tmp2 */
1369 int rellen = strlen (rel); 1515 int rellen = strlen (rel);
1370 1516
1371 errno = ENAMETOOLONG; 1517 errno = ENAMETOOLONG;
1372 if (linklen + 1 + rellen >= PATH_MAX) 1518 if (linklen + 1 + rellen >= PATH_MAX)
1373 return; 1519 return -1;
1374 1520
1375 errno = ELOOP; 1521 errno = ELOOP;
1376 if (!--symlinks) 1522 if (!--symlinks)
1377 return; 1523 return -1;
1378 1524
1379 if (*tmp1 == '/') 1525 if (*tmp1 == '/')
1380 res = req->ptr2; /* symlink resolves to an absolute path */ 1526 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1381 1527
1382 /* we need to be careful, as rel might point into tmp2 already */ 1528 /* we need to be careful, as rel might point into tmp2 already */
1383 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1529 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1384 tmp2 [linklen] = '/'; 1530 tmp2 [linklen] = '/';
1385 memcpy (tmp2, tmp1, linklen); 1531 memcpy (tmp2, tmp1, linklen);
1387 rel = tmp2; 1533 rel = tmp2;
1388 } 1534 }
1389 } 1535 }
1390 1536
1391 /* special case for the lone root path */ 1537 /* special case for the lone root path */
1392 if (res == req->ptr2) 1538 if (res == tmpbuf->ptr)
1393 *res++ = '/'; 1539 *res++ = '/';
1394 1540
1395 req->result = res - (char *)req->ptr2; 1541 return res - (char *)tmpbuf->ptr;
1396
1397done:
1398 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1399} 1542}
1400 1543
1401static signed char 1544static signed char
1402eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1545eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1403{ 1546{
1411 1554
1412#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 */
1413#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1556#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1414 1557
1415static void 1558static void
1416eio_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)
1417{ 1560{
1418 unsigned char bits [9 + sizeof (ino_t) * 8]; 1561 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1419 unsigned char *bit = bits; 1562 unsigned char *bit = bits;
1420 1563
1421 assert (CHAR_BIT == 8); 1564 assert (CHAR_BIT == 8);
1422 assert (sizeof (eio_dirent) * 8 < 256); 1565 assert (sizeof (eio_dirent) * 8 < 256);
1423 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1566 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1425 1568
1426 if (size <= EIO_SORT_FAST) 1569 if (size <= EIO_SORT_FAST)
1427 return; 1570 return;
1428 1571
1429 /* 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 */
1430 /* 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 */
1431 /* inode_bits must contain all inodes ORed together */ 1574 /* inode_bits must contain all inodes ORed together */
1432 /* 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 */
1433 { 1576 {
1434 ino_t endianness; 1577 eio_ino_t endianness;
1435 int i, j; 1578 int i, j;
1436 1579
1437 /* 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" */
1438 for (i = 0; i < sizeof (ino_t); ++i) 1581 for (i = 0; i < sizeof (eio_ino_t); ++i)
1439 ((unsigned char *)&endianness)[i] = i; 1582 ((unsigned char *)&endianness)[i] = i;
1440 1583
1441 *bit++ = 0; 1584 *bit++ = 0;
1442 1585
1443 for (i = 0; i < sizeof (ino_t); ++i) 1586 for (i = 0; i < sizeof (eio_ino_t); ++i)
1444 { 1587 {
1445 /* shifting off the byte offsets out of "endianness" */ 1588 /* shifting off the byte offsets out of "endianness" */
1446 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1589 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1447 endianness >>= 8; 1590 endianness >>= 8;
1448 1591
1449 for (j = 0; j < 8; ++j) 1592 for (j = 0; j < 8; ++j)
1450 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1593 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1451 *bit++ = offs + j; 1594 *bit++ = offs + j;
1452 } 1595 }
1453 1596
1454 for (j = 0; j < 8; ++j) 1597 for (j = 0; j < 8; ++j)
1455 if (score_bits & (1 << j)) 1598 if (score_bits & (1 << j))
1456 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1599 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1457 } 1600 }
1458 1601
1459 /* now actually do the sorting (a variant of MSD radix sort) */ 1602 /* now actually do the sorting (a variant of MSD radix sort) */
1460 { 1603 {
1461 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1604 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1462 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1605 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1463 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1606 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1464 int stk_idx = 0; 1607 int stk_idx = 0;
1465 1608
1466 base_stk [stk_idx] = dents; 1609 base_stk [stk_idx] = dents;
1467 end_stk [stk_idx] = dents + size; 1610 end_stk [stk_idx] = dents + size;
1468 bit_stk [stk_idx] = bit - 1; 1611 bit_stk [stk_idx] = bit - 1;
1547 } 1690 }
1548 } 1691 }
1549} 1692}
1550 1693
1551static void 1694static void
1552eio_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)
1553{ 1696{
1554 if (size <= 1) 1697 if (size <= 1)
1555 return; /* our insertion sort relies on size > 0 */ 1698 return; /* our insertion sort relies on size > 0 */
1556 1699
1557 /* 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 */
1565 1708
1566/* read a full directory */ 1709/* read a full directory */
1567static void 1710static void
1568eio__scandir (eio_req *req, etp_worker *self) 1711eio__scandir (eio_req *req, etp_worker *self)
1569{ 1712{
1570 DIR *dirp;
1571 EIO_STRUCT_DIRENT *entp;
1572 char *name, *names; 1713 char *name, *names;
1573 int namesalloc = 4096; 1714 int namesalloc = 4096 - sizeof (void *) * 4;
1574 int namesoffs = 0; 1715 int namesoffs = 0;
1575 int flags = req->int1; 1716 int flags = req->int1;
1576 eio_dirent *dents = 0; 1717 eio_dirent *dents = 0;
1577 int dentalloc = 128; 1718 int dentalloc = 128;
1578 int dentoffs = 0; 1719 int dentoffs = 0;
1579 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
1580 1728
1581 req->result = -1; 1729 req->result = -1;
1582 1730
1583 if (!(flags & EIO_READDIR_DENTS)) 1731 if (!(flags & EIO_READDIR_DENTS))
1584 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1732 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1585 1733
1586 X_LOCK (wrklock); 1734#ifdef _WIN32
1587 /* 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 if (!dirp)
1791 close (fd);
1792 }
1793 else
1588 self->dirp = dirp = opendir (req->ptr1); 1794 dirp = opendir (req->ptr1);
1795 #else
1796 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1797 #endif
1798
1799 if (!dirp)
1800 return;
1801#endif
1589 1802
1590 if (req->flags & EIO_FLAG_PTR1_FREE) 1803 if (req->flags & EIO_FLAG_PTR1_FREE)
1591 free (req->ptr1); 1804 free (req->ptr1);
1592 1805
1593 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1806 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1594 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1807 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1595 req->ptr2 = names = malloc (namesalloc); 1808 req->ptr2 = names = malloc (namesalloc);
1596 X_UNLOCK (wrklock);
1597 1809
1598 if (dirp && names && (!flags || dents)) 1810 if (!names || (flags && !dents))
1811 return;
1812
1599 for (;;) 1813 for (;;)
1600 { 1814 {
1815 int done;
1816
1817#ifdef _WIN32
1818 done = !dirp;
1819#else
1601 errno = 0; 1820 errno = 0;
1602 entp = readdir (dirp); 1821 entp = readdir (dirp);
1822 done = !entp;
1823#endif
1603 1824
1604 if (!entp) 1825 if (done)
1605 { 1826 {
1827#ifndef _WIN32
1828 int old_errno = errno;
1829 closedir (dirp);
1830 errno = old_errno;
1831
1606 if (errno) 1832 if (errno)
1607 break; 1833 break;
1834#endif
1608 1835
1609 /* sort etc. */ 1836 /* sort etc. */
1610 req->int1 = flags; 1837 req->int1 = flags;
1611 req->result = dentoffs; 1838 req->result = dentoffs;
1612 1839
1613 if (flags & EIO_READDIR_STAT_ORDER) 1840 if (flags & EIO_READDIR_STAT_ORDER)
1614 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1841 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1615 else if (flags & EIO_READDIR_DIRS_FIRST) 1842 else if (flags & EIO_READDIR_DIRS_FIRST)
1616 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1843 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1617 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1844 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1618 else 1845 else
1846 {
1847 /* in this case, all is known, and we just put dirs first and sort them */
1848 eio_dirent *oth = dents + dentoffs;
1849 eio_dirent *dir = dents;
1850
1851 /* now partition dirs to the front, and non-dirs to the back */
1852 /* by walking from both sides and swapping if necessary */
1853 while (oth > dir)
1854 {
1855 if (dir->type == EIO_DT_DIR)
1856 ++dir;
1857 else if ((--oth)->type == EIO_DT_DIR)
1858 {
1859 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1860
1861 ++dir;
1862 }
1863 }
1864
1865 /* now sort the dirs only (dirs all have the same score) */
1866 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1867 }
1868
1869 break;
1870 }
1871
1872 /* now add the entry to our list(s) */
1873 name = D_NAME (entp);
1874
1875 /* skip . and .. entries */
1876 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1877 {
1878 int len = D_NAMLEN (entp) + 1;
1879
1880 while (ecb_expect_false (namesoffs + len > namesalloc))
1881 {
1882 namesalloc *= 2;
1883 req->ptr2 = names = realloc (names, namesalloc);
1884
1885 if (!names)
1886 break;
1887 }
1888
1889 memcpy (names + namesoffs, name, len);
1890
1891 if (dents)
1892 {
1893 struct eio_dirent *ent;
1894
1895 if (ecb_expect_false (dentoffs == dentalloc))
1619 { 1896 {
1620 /* in this case, all is known, and we just put dirs first and sort them */ 1897 dentalloc *= 2;
1898 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1899
1900 if (!dents)
1901 break;
1902 }
1903
1621 eio_dirent *oth = dents + dentoffs; 1904 ent = dents + dentoffs;
1622 eio_dirent *dir = dents;
1623 1905
1624 /* now partition dirs to the front, and non-dirs to the back */ 1906 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1625 /* by walking from both sides and swapping if necessary */ 1907 ent->namelen = len - 1;
1626 while (oth > dir) 1908 ent->inode = D_INO (entp);
1909
1910 inode_bits |= ent->inode;
1911
1912 switch (D_TYPE (entp))
1913 {
1914 default:
1915 ent->type = EIO_DT_UNKNOWN;
1916 flags |= EIO_READDIR_FOUND_UNKNOWN;
1917 break;
1918
1919 #ifdef DT_FIFO
1920 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1921 #endif
1922 #ifdef DT_CHR
1923 case DT_CHR: ent->type = EIO_DT_CHR; break;
1924 #endif
1925 #ifdef DT_MPC
1926 case DT_MPC: ent->type = EIO_DT_MPC; break;
1927 #endif
1928 #ifdef DT_DIR
1929 case DT_DIR: ent->type = EIO_DT_DIR; break;
1930 #endif
1931 #ifdef DT_NAM
1932 case DT_NAM: ent->type = EIO_DT_NAM; break;
1933 #endif
1934 #ifdef DT_BLK
1935 case DT_BLK: ent->type = EIO_DT_BLK; break;
1936 #endif
1937 #ifdef DT_MPB
1938 case DT_MPB: ent->type = EIO_DT_MPB; break;
1939 #endif
1940 #ifdef DT_REG
1941 case DT_REG: ent->type = EIO_DT_REG; break;
1942 #endif
1943 #ifdef DT_NWK
1944 case DT_NWK: ent->type = EIO_DT_NWK; break;
1945 #endif
1946 #ifdef DT_CMP
1947 case DT_CMP: ent->type = EIO_DT_CMP; break;
1948 #endif
1949 #ifdef DT_LNK
1950 case DT_LNK: ent->type = EIO_DT_LNK; break;
1951 #endif
1952 #ifdef DT_SOCK
1953 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1954 #endif
1955 #ifdef DT_DOOR
1956 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1957 #endif
1958 #ifdef DT_WHT
1959 case DT_WHT: ent->type = EIO_DT_WHT; break;
1960 #endif
1961 }
1962
1963 ent->score = 7;
1964
1965 if (flags & EIO_READDIR_DIRS_FIRST)
1966 {
1967 if (ent->type == EIO_DT_UNKNOWN)
1627 { 1968 {
1628 if (dir->type == EIO_DT_DIR) 1969 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1629 ++dir; 1970 ent->score = 1;
1630 else if ((--oth)->type == EIO_DT_DIR) 1971 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1631 { 1972 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1632 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1633
1634 ++dir;
1635 }
1636 } 1973 }
1637 1974 else if (ent->type == EIO_DT_DIR)
1638 /* now sort the dirs only (dirs all have the same score) */ 1975 ent->score = 0;
1639 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1640 } 1976 }
1641
1642 break;
1643 }
1644
1645 /* now add the entry to our list(s) */
1646 name = entp->d_name;
1647
1648 /* skip . and .. entries */
1649 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1650 {
1651 int len = D_NAMLEN (entp) + 1;
1652
1653 while (ecb_expect_false (namesoffs + len > namesalloc))
1654 {
1655 namesalloc *= 2;
1656 X_LOCK (wrklock);
1657 req->ptr2 = names = realloc (names, namesalloc);
1658 X_UNLOCK (wrklock);
1659
1660 if (!names)
1661 break;
1662 } 1977 }
1663 1978
1664 memcpy (names + namesoffs, name, len);
1665
1666 if (dents)
1667 {
1668 struct eio_dirent *ent;
1669
1670 if (ecb_expect_false (dentoffs == dentalloc))
1671 {
1672 dentalloc *= 2;
1673 X_LOCK (wrklock);
1674 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1675 X_UNLOCK (wrklock);
1676
1677 if (!dents)
1678 break;
1679 }
1680
1681 ent = dents + dentoffs;
1682
1683 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1684 ent->namelen = len - 1;
1685 ent->inode = D_INO (entp);
1686
1687 inode_bits |= ent->inode;
1688
1689 switch (D_TYPE (entp))
1690 {
1691 default:
1692 ent->type = EIO_DT_UNKNOWN;
1693 flags |= EIO_READDIR_FOUND_UNKNOWN;
1694 break;
1695
1696 #ifdef DT_FIFO
1697 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1698 #endif
1699 #ifdef DT_CHR
1700 case DT_CHR: ent->type = EIO_DT_CHR; break;
1701 #endif
1702 #ifdef DT_MPC
1703 case DT_MPC: ent->type = EIO_DT_MPC; break;
1704 #endif
1705 #ifdef DT_DIR
1706 case DT_DIR: ent->type = EIO_DT_DIR; break;
1707 #endif
1708 #ifdef DT_NAM
1709 case DT_NAM: ent->type = EIO_DT_NAM; break;
1710 #endif
1711 #ifdef DT_BLK
1712 case DT_BLK: ent->type = EIO_DT_BLK; break;
1713 #endif
1714 #ifdef DT_MPB
1715 case DT_MPB: ent->type = EIO_DT_MPB; break;
1716 #endif
1717 #ifdef DT_REG
1718 case DT_REG: ent->type = EIO_DT_REG; break;
1719 #endif
1720 #ifdef DT_NWK
1721 case DT_NWK: ent->type = EIO_DT_NWK; break;
1722 #endif
1723 #ifdef DT_CMP
1724 case DT_CMP: ent->type = EIO_DT_CMP; break;
1725 #endif
1726 #ifdef DT_LNK
1727 case DT_LNK: ent->type = EIO_DT_LNK; break;
1728 #endif
1729 #ifdef DT_SOCK
1730 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1731 #endif
1732 #ifdef DT_DOOR
1733 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1734 #endif
1735 #ifdef DT_WHT
1736 case DT_WHT: ent->type = EIO_DT_WHT; break;
1737 #endif
1738 }
1739
1740 ent->score = 7;
1741
1742 if (flags & EIO_READDIR_DIRS_FIRST)
1743 {
1744 if (ent->type == EIO_DT_UNKNOWN)
1745 {
1746 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1747 ent->score = 1;
1748 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1749 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1750 }
1751 else if (ent->type == EIO_DT_DIR)
1752 ent->score = 0;
1753 }
1754 }
1755
1756 namesoffs += len; 1979 namesoffs += len;
1757 ++dentoffs; 1980 ++dentoffs;
1758 } 1981 }
1759 1982
1760 if (EIO_CANCELLED (req)) 1983 if (EIO_CANCELLED (req))
1761 { 1984 {
1762 errno = ECANCELED; 1985 errno = ECANCELED;
1763 break; 1986 break;
1764 } 1987 }
1988
1989#ifdef _WIN32
1990 if (!FindNextFile (dirp, &entp))
1991 {
1992 FindClose (dirp);
1993 dirp = 0;
1765 } 1994 }
1995#endif
1996 }
1766} 1997}
1998
1999/*****************************************************************************/
2000/* working directory stuff */
2001/* various deficiencies in the posix 2008 api force us to */
2002/* keep the absolute path in string form at all times */
2003/* fuck yeah. */
2004
2005#if !HAVE_AT
2006
2007/* a bit like realpath, but usually faster because it doesn'T have to return */
2008/* an absolute or canonical path */
2009static const char *
2010wd_expand (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2011{
2012 if (!wd || *path == '/')
2013 return path;
2014
2015 if (path [0] == '.' && !path [1])
2016 return wd->str;
2017
2018 {
2019 int l1 = wd->len;
2020 int l2 = strlen (path);
2021
2022 char *res = tmpbuf_get (tmpbuf, l1 + l2 + 2);
2023
2024 memcpy (res, wd->str, l1);
2025 res [l1] = '/';
2026 memcpy (res + l1 + 1, path, l2 + 1);
2027
2028 return res;
2029 }
2030}
2031
2032#endif
2033
2034static eio_wd
2035eio__wd_open_sync (struct tmpbuf *tmpbuf, eio_wd wd, const char *path)
2036{
2037 int fd;
2038 eio_wd res;
2039 int len = eio__realpath (tmpbuf, wd, path);
2040
2041 if (len < 0)
2042 return EIO_INVALID_WD;
2043
2044#if HAVE_AT
2045 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
2046
2047 if (fd < 0)
2048 return EIO_INVALID_WD;
2049#endif
2050
2051 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
2052
2053#if HAVE_AT
2054 res->fd = fd;
2055#endif
2056
2057 res->len = len;
2058 memcpy (res->str, tmpbuf->ptr, len);
2059 res->str [len] = 0;
2060
2061 return res;
2062}
2063
2064eio_wd
2065eio_wd_open_sync (eio_wd wd, const char *path)
2066{
2067 struct tmpbuf tmpbuf = { 0 };
2068 wd = eio__wd_open_sync (&tmpbuf, wd, path);
2069 free (tmpbuf.ptr);
2070
2071 return wd;
2072}
2073
2074void
2075eio_wd_close_sync (eio_wd wd)
2076{
2077 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
2078 {
2079 #if HAVE_AT
2080 close (wd->fd);
2081 #endif
2082 free (wd);
2083 }
2084}
2085
2086#if HAVE_AT
2087
2088/* they forgot these */
2089
2090static int
2091eio__truncateat (int dirfd, const char *path, off_t length)
2092{
2093 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
2094 int res;
2095
2096 if (fd < 0)
2097 return fd;
2098
2099 res = ftruncate (fd, length);
2100 close (fd);
2101 return res;
2102}
2103
2104static int
2105eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
2106{
2107 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
2108 int res;
2109
2110 if (fd < 0)
2111 return fd;
2112
2113 res = fstatvfs (fd, buf);
2114 close (fd);
2115 return res;
2116
2117}
2118
2119#endif
1767 2120
1768/*****************************************************************************/ 2121/*****************************************************************************/
1769 2122
1770#define ALLOC(len) \ 2123#define ALLOC(len) \
1771 if (!req->ptr2) \ 2124 if (!req->ptr2) \
1780 req->result = -1; \ 2133 req->result = -1; \
1781 break; \ 2134 break; \
1782 } \ 2135 } \
1783 } 2136 }
1784 2137
2138static void ecb_noinline ecb_cold
2139etp_proc_init (void)
2140{
2141#if HAVE_PRCTL_SET_NAME
2142 /* provide a more sensible "thread name" */
2143 char name[16 + 1];
2144 const int namelen = sizeof (name) - 1;
2145 int len;
2146
2147 prctl (PR_GET_NAME, (unsigned long)name, 0, 0, 0);
2148 name [namelen] = 0;
2149 len = strlen (name);
2150 strcpy (name + (len <= namelen - 4 ? len : namelen - 4), "/eio");
2151 prctl (PR_SET_NAME, (unsigned long)name, 0, 0, 0);
2152#endif
2153}
2154
1785X_THREAD_PROC (etp_proc) 2155X_THREAD_PROC (etp_proc)
1786{ 2156{
1787 ETP_REQ *req; 2157 ETP_REQ *req;
1788 struct timespec ts; 2158 struct timespec ts;
1789 etp_worker *self = (etp_worker *)thr_arg; 2159 etp_worker *self = (etp_worker *)thr_arg;
1790 2160
2161 etp_proc_init ();
2162
1791 /* try to distribute timeouts somewhat randomly */ 2163 /* try to distribute timeouts somewhat evenly */
1792 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 2164 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
1793 2165
1794 for (;;) 2166 for (;;)
1795 { 2167 {
2168 ts.tv_sec = 0;
2169
1796 X_LOCK (reqlock); 2170 X_LOCK (reqlock);
1797 2171
1798 for (;;) 2172 for (;;)
1799 { 2173 {
1800 self->req = req = reqq_shift (&req_queue); 2174 req = reqq_shift (&req_queue);
1801 2175
1802 if (req) 2176 if (req)
1803 break; 2177 break;
1804 2178
2179 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
2180 {
2181 X_UNLOCK (reqlock);
2182 X_LOCK (wrklock);
2183 --started;
2184 X_UNLOCK (wrklock);
2185 goto quit;
2186 }
2187
1805 ++idle; 2188 ++idle;
1806 2189
1807 ts.tv_sec = time (0) + idle_timeout; 2190 if (idle <= max_idle)
1808 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT) 2191 /* we are allowed to idle, so do so without any timeout */
2192 X_COND_WAIT (reqwait, reqlock);
2193 else
1809 { 2194 {
1810 if (idle > max_idle) 2195 /* initialise timeout once */
1811 { 2196 if (!ts.tv_sec)
1812 --idle; 2197 ts.tv_sec = time (0) + idle_timeout;
1813 X_UNLOCK (reqlock);
1814 X_LOCK (wrklock);
1815 --started;
1816 X_UNLOCK (wrklock);
1817 goto quit;
1818 }
1819 2198
1820 /* we are allowed to idle, so do so without any timeout */
1821 X_COND_WAIT (reqwait, reqlock); 2199 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
2200 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1822 } 2201 }
1823 2202
1824 --idle; 2203 --idle;
1825 } 2204 }
1826 2205
1829 X_UNLOCK (reqlock); 2208 X_UNLOCK (reqlock);
1830 2209
1831 if (req->type < 0) 2210 if (req->type < 0)
1832 goto quit; 2211 goto quit;
1833 2212
1834 if (!EIO_CANCELLED (req))
1835 ETP_EXECUTE (self, req); 2213 ETP_EXECUTE (self, req);
1836 2214
1837 X_LOCK (reslock); 2215 X_LOCK (reslock);
1838 2216
1839 ++npending; 2217 ++npending;
1840 2218
1841 if (!reqq_push (&res_queue, req) && want_poll_cb) 2219 if (!reqq_push (&res_queue, req) && want_poll_cb)
1842 want_poll_cb (); 2220 want_poll_cb ();
1843 2221
1844 self->req = 0;
1845 etp_worker_clear (self); 2222 etp_worker_clear (self);
1846 2223
1847 X_UNLOCK (reslock); 2224 X_UNLOCK (reslock);
1848 } 2225 }
1849 2226
1850quit: 2227quit:
2228 free (req);
2229
1851 X_LOCK (wrklock); 2230 X_LOCK (wrklock);
1852 etp_worker_free (self); 2231 etp_worker_free (self);
1853 X_UNLOCK (wrklock); 2232 X_UNLOCK (wrklock);
1854 2233
1855 return 0; 2234 return 0;
1858/*****************************************************************************/ 2237/*****************************************************************************/
1859 2238
1860int ecb_cold 2239int ecb_cold
1861eio_init (void (*want_poll)(void), void (*done_poll)(void)) 2240eio_init (void (*want_poll)(void), void (*done_poll)(void))
1862{ 2241{
2242#if !HAVE_PREADWRITE
2243 X_MUTEX_CREATE (preadwritelock);
2244#endif
2245
1863 return etp_init (want_poll, done_poll); 2246 return etp_init (want_poll, done_poll);
1864} 2247}
1865 2248
1866ecb_inline void 2249ecb_inline void
1867eio_api_destroy (eio_req *req) 2250eio_api_destroy (eio_req *req)
1894 } 2277 }
1895 2278
1896static void 2279static void
1897eio_execute (etp_worker *self, eio_req *req) 2280eio_execute (etp_worker *self, eio_req *req)
1898{ 2281{
2282#if HAVE_AT
2283 int dirfd;
2284#else
2285 const char *path;
2286#endif
2287
2288 if (ecb_expect_false (EIO_CANCELLED (req)))
2289 {
2290 req->result = -1;
2291 req->errorno = ECANCELED;
2292 return;
2293 }
2294
2295 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
2296 {
2297 req->result = -1;
2298 req->errorno = ENOENT;
2299 return;
2300 }
2301
2302 if (req->type >= EIO_OPEN)
2303 {
2304 #if HAVE_AT
2305 dirfd = WD2FD (req->wd);
2306 #else
2307 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
2308 #endif
2309 }
2310
1899 switch (req->type) 2311 switch (req->type)
1900 { 2312 {
2313 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
2314 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
2315 break;
2316 case EIO_WD_CLOSE: req->result = 0;
2317 eio_wd_close_sync (req->wd); break;
2318
1901 case EIO_READ: ALLOC (req->size); 2319 case EIO_READ: ALLOC (req->size);
1902 req->result = req->offs >= 0 2320 req->result = req->offs >= 0
1903 ? pread (req->int1, req->ptr2, req->size, req->offs) 2321 ? pread (req->int1, req->ptr2, req->size, req->offs)
1904 : read (req->int1, req->ptr2, req->size); break; 2322 : read (req->int1, req->ptr2, req->size); break;
1905 case EIO_WRITE: req->result = req->offs >= 0 2323 case EIO_WRITE: req->result = req->offs >= 0
1906 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 2324 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1907 : write (req->int1, req->ptr2, req->size); break; 2325 : write (req->int1, req->ptr2, req->size); break;
1908 2326
1909 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 2327 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1910 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 2328 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
2329
2330#if HAVE_AT
1911 2331
1912 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2332 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1913 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2333 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1914 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2334 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2335 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
2336 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
2337 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
2338 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
2339 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
2340
2341 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
2342 case EIO_RMDIR: req->result = unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
2343 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
2344 case EIO_RENAME: req->result = renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
2345 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
2346 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
2347 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2348 case EIO_READLINK: ALLOC (PATH_MAX);
2349 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
2350 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2351 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
2352 case EIO_UTIME:
2353 case EIO_FUTIME:
2354 {
2355 struct timespec ts[2];
2356 struct timespec *times;
2357
2358 if (req->nv1 != -1. || req->nv2 != -1.)
2359 {
2360 ts[0].tv_sec = req->nv1;
2361 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
2362 ts[1].tv_sec = req->nv2;
2363 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
2364
2365 times = ts;
2366 }
2367 else
2368 times = 0;
2369
2370 req->result = req->type == EIO_FUTIME
2371 ? futimens (req->int1, times)
2372 : utimensat (dirfd, req->ptr1, times, 0);
2373 }
2374 break;
2375
2376#else
2377
2378 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
2379 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2380 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1915 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 2381 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
2382 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
2383 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
2384 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
2385 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
2386
2387 case EIO_UNLINK: req->result = unlink (path ); break;
2388 case EIO_RMDIR: req->result = rmdir (path ); break;
2389 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
2390 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
2391 case EIO_LINK: req->result = link (path , req->ptr2); break;
2392 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
2393 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
2394 case EIO_READLINK: ALLOC (PATH_MAX);
2395 req->result = readlink (path, req->ptr2, PATH_MAX); break;
2396 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
2397 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
2398
2399 case EIO_UTIME:
2400 case EIO_FUTIME:
2401 {
2402 struct timeval tv[2];
2403 struct timeval *times;
2404
2405 if (req->nv1 != -1. || req->nv2 != -1.)
2406 {
2407 tv[0].tv_sec = req->nv1;
2408 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
2409 tv[1].tv_sec = req->nv2;
2410 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2411
2412 times = tv;
2413 }
2414 else
2415 times = 0;
2416
2417 req->result = req->type == EIO_FUTIME
2418 ? futimes (req->int1, times)
2419 : utimes (req->ptr1, times);
2420 }
2421 break;
2422
2423#endif
2424
2425 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2426 {
2427 ALLOC (req->result);
2428 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2429 }
2430 break;
2431
1916 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2432 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1917 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2433 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1918 2434
1919 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1920 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1921 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2435 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1922 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2436 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1923 2437
1924 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1925 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2438 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1926 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1927 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2439 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1928 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1929 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2440 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1930 2441
1931 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1932 case EIO_CLOSE: req->result = close (req->int1); break; 2442 case EIO_CLOSE: req->result = close (req->int1); break;
1933 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2443 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1934 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1935 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1936 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1937 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1938 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1939 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1940 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1941
1942 case EIO_REALPATH: eio__realpath (req, self); break;
1943
1944 case EIO_READLINK: ALLOC (PATH_MAX);
1945 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1946
1947 case EIO_SYNC: req->result = 0; sync (); break; 2444 case EIO_SYNC: req->result = 0; sync (); break;
1948 case EIO_FSYNC: req->result = fsync (req->int1); break; 2445 case EIO_FSYNC: req->result = fsync (req->int1); break;
1949 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2446 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2447 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2448 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1950 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2449 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1951 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2450 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1952 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2451 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1953 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2452 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1954 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2453 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1955 2454
1956 case EIO_READDIR: eio__scandir (req, self); break; 2455 case EIO_READDIR: eio__scandir (req, self); break;
1957 2456
1958 case EIO_BUSY: 2457 case EIO_BUSY:
1959#ifdef _WIN32 2458#ifdef _WIN32
1968 req->result = select (0, 0, 0, 0, &tv); 2467 req->result = select (0, 0, 0, 0, &tv);
1969 } 2468 }
1970#endif 2469#endif
1971 break; 2470 break;
1972 2471
1973 case EIO_UTIME:
1974 case EIO_FUTIME:
1975 {
1976 struct timeval tv[2];
1977 struct timeval *times;
1978
1979 if (req->nv1 != -1. || req->nv2 != -1.)
1980 {
1981 tv[0].tv_sec = req->nv1;
1982 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1983 tv[1].tv_sec = req->nv2;
1984 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1985
1986 times = tv;
1987 }
1988 else
1989 times = 0;
1990
1991 req->result = req->type == EIO_FUTIME
1992 ? futimes (req->int1, times)
1993 : utimes (req->ptr1, times);
1994 }
1995 break;
1996
1997 case EIO_GROUP: 2472 case EIO_GROUP:
1998 abort (); /* handled in eio_request */ 2473 abort (); /* handled in eio_request */
1999 2474
2000 case EIO_NOP: 2475 case EIO_NOP:
2001 req->result = 0; 2476 req->result = 0;
2014 req->errorno = errno; 2489 req->errorno = errno;
2015} 2490}
2016 2491
2017#ifndef EIO_NO_WRAPPERS 2492#ifndef EIO_NO_WRAPPERS
2018 2493
2494eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2495{
2496 REQ (EIO_WD_OPEN); PATH; SEND;
2497}
2498
2499eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2500{
2501 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2502}
2503
2019eio_req *eio_nop (int pri, eio_cb cb, void *data) 2504eio_req *eio_nop (int pri, eio_cb cb, void *data)
2020{ 2505{
2021 REQ (EIO_NOP); SEND; 2506 REQ (EIO_NOP); SEND;
2022} 2507}
2023 2508
2039eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2524eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2040{ 2525{
2041 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2526 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2042} 2527}
2043 2528
2529eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2530{
2531 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2532}
2533
2534eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2535{
2536 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2537}
2538
2539eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2540{
2541 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2542}
2543
2044eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2544eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2045{ 2545{
2046 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2546 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2047} 2547}
2048 2548
2054eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2554eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2055{ 2555{
2056 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2556 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2057} 2557}
2058 2558
2059eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2559eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2060{ 2560{
2061 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2561 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2062}
2063
2064eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2065{
2066 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2067} 2562}
2068 2563
2069eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2564eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2070{ 2565{
2071 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2566 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2109eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2604eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2110{ 2605{
2111 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2606 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2112} 2607}
2113 2608
2114eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2609eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2115{ 2610{
2116 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2611 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2117} 2612}
2118 2613
2119eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2614eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2139eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2634eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2140{ 2635{
2141 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2636 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2142} 2637}
2143 2638
2144eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2639eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2145{ 2640{
2146 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2641 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2147} 2642}
2148 2643
2149eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2644eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2296} 2791}
2297 2792
2298/*****************************************************************************/ 2793/*****************************************************************************/
2299/* misc garbage */ 2794/* misc garbage */
2300 2795
2301ssize_t 2796eio_ssize_t
2302eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2797eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2303{ 2798{
2304 etp_worker wrk;
2305 ssize_t ret;
2306
2307 wrk.dbuf = 0;
2308
2309 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2799 return eio__sendfile (ofd, ifd, offset, count);
2310
2311 if (wrk.dbuf)
2312 free (wrk.dbuf);
2313
2314 return ret;
2315} 2800}
2316 2801

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines