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

Comparing libeio/eio.c (file contents):
Revision 1.86 by root, Thu Jul 14 18:30:10 2011 UTC vs.
Revision 1.148 by root, Tue Feb 20 06:05:19 2018 UTC

1/* 1/*
2 * libeio implementation 2 * libeio implementation
3 * 3 *
4 * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libeio@schmorp.de> 4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016,2017 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved. 5 * All rights reserved.
6 * 6 *
7 * Redistribution and use in source and binary forms, with or without modifica- 7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met: 8 * tion, are permitted provided that the following conditions are met:
9 * 9 *
10 * 1. Redistributions of source code must retain the above copyright notice, 10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer. 11 * this list of conditions and the following disclaimer.
12 * 12 *
13 * 2. Redistributions in binary form must reproduce the above copyright 13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the 14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution. 15 * documentation and/or other materials provided with the distribution.
16 * 16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER- 18 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO 19 * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE- 20 * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
42#endif 42#endif
43 43
44#include "eio.h" 44#include "eio.h"
45#include "ecb.h" 45#include "ecb.h"
46 46
47#ifdef EIO_STACKSIZE
48# define XTHREAD_STACKSIZE EIO_STACKSIZE
49#endif
50#include "xthread.h"
51
52#include <errno.h> 47#include <errno.h>
53#include <stddef.h> 48#include <stddef.h>
54#include <stdlib.h> 49#include <stdlib.h>
55#include <string.h> 50#include <string.h>
56#include <errno.h> 51#include <errno.h>
58#include <sys/stat.h> 53#include <sys/stat.h>
59#include <limits.h> 54#include <limits.h>
60#include <fcntl.h> 55#include <fcntl.h>
61#include <assert.h> 56#include <assert.h>
62 57
63#include <sys/statvfs.h>
64/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */ 58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
65/* intptr_t only comes from stdint.h, says idiot openbsd coder */ 59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
66#if HAVE_STDINT_H 60#if HAVE_STDINT_H
67# include <stdint.h> 61# include <stdint.h>
68#endif 62#endif
69 63
70#ifndef ECANCELED 64#ifndef ECANCELED
71# define ECANCELED EDOM 65# define ECANCELED EDOM
72#endif 66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
73 74
74static void eio_destroy (eio_req *req); 75static void eio_destroy (eio_req *req);
75 76
76#ifndef EIO_FINISH 77#ifndef EIO_FINISH
77# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0 78# define EIO_FINISH(req) ((req)->finish) && !EIO_CANCELLED (req) ? (req)->finish (req) : 0
96 97
97#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1) 98#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
98 99
99#ifdef _WIN32 100#ifdef _WIN32
100 101
102 #undef PAGESIZE
101 #define PAGESIZE 4096 /* GetSystemInfo? */ 103 #define PAGESIZE 4096 /* GetSystemInfo? */
102 104
105 #define O_CLOEXEC 0
106
107 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
108
109 #ifdef EIO_STRUCT_STATI64
110 /* look at perl's non-sloppy stat */
103 #define stat(path,buf) _stati64 (path,buf) 111 #define stat(path,buf) _stati64 (path,buf)
112 #define fstat(fd,buf) _fstati64 (fd,buf)
113 #endif
104 #define lstat(path,buf) stat (path,buf) 114 #define lstat(path,buf) stat (path,buf)
105 #define fstat(fd,buf) _fstati64 (path,buf)
106 #define fsync(fd) (FlushFileBuffers (EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1)) 115 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
107 #define mkdir(path,mode) _mkdir (path) 116 #define mkdir(path,mode) _mkdir (path)
108 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1)) 117 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
109 118
119 #define chmod(path,mode) _chmod (path, mode)
120 #define dup(fd) _dup (fd)
121 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
122 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
123
124 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
125 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
126 #define fchmod(fd,mode) EIO_ENOSYS ()
110 #define chown(path,uid,gid) EIO_ENOSYS () 127 #define chown(path,uid,gid) EIO_ENOSYS ()
111 #define fchown(fd,uid,gid) EIO_ENOSYS () 128 #define fchown(fd,uid,gid) EIO_ENOSYS ()
112 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */ 129 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
113 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */ 130 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
114 #define mknod(path,mode,dev) EIO_ENOSYS () 131 #define mknod(path,mode,dev) EIO_ENOSYS ()
115 #define sync() 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 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
138 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
139
140 #if __GNUC__
141 typedef long long eio_off_t; /* signed for compatibility to msvc */
142 #else
143 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
144 #endif
145
146 static eio_ssize_t
147 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
148 {
149 OVERLAPPED o = { 0 };
150 DWORD got;
151
152 o.Offset = offset;
153 o.OffsetHigh = offset >> 32;
154
155 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
156 ? got : -1;
157 }
158
159 static eio_ssize_t
160 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
161 {
162 OVERLAPPED o = { 0 };
163 DWORD got;
164
165 o.Offset = offset;
166 o.OffsetHigh = offset >> 32;
167
168 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
169 ? got : -1;
170 }
171
172 /* rename() uses MoveFile, which fails to overwrite */
173 #define rename(old,neu) eio__rename (old, neu)
174
175 static int
176 eio__rename (const char *old, const char *neu)
177 {
178 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
179 return 0;
180
181 /* should steal _dosmaperr */
182 switch (GetLastError ())
183 {
184 case ERROR_FILE_NOT_FOUND:
185 case ERROR_PATH_NOT_FOUND:
186 case ERROR_INVALID_DRIVE:
187 case ERROR_NO_MORE_FILES:
188 case ERROR_BAD_NETPATH:
189 case ERROR_BAD_NET_NAME:
190 case ERROR_BAD_PATHNAME:
191 case ERROR_FILENAME_EXCED_RANGE:
192 errno = ENOENT;
193 break;
194
195 default:
196 errno = EACCES;
197 break;
198 }
199
200 return -1;
201 }
116 202
117 /* we could even stat and see if it exists */ 203 /* we could even stat and see if it exists */
118 static int 204 static int
119 symlink (const char *old, const char *neu) 205 symlink (const char *old, const char *neu)
120 { 206 {
207 #if WINVER >= 0x0600
208 int flags;
209
210 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
211 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
212 * with directory first.
213 */
214 for (flags = 3; flags >= 0; --flags)
121 if (CreateSymbolicLink (neu, old, 1)) 215 if (CreateSymbolicLink (neu, old, flags))
122 return 0; 216 return 0;
123 217 #endif
124 if (CreateSymbolicLink (neu, old, 0))
125 return 0;
126 218
127 return EIO_ERRNO (ENOENT, -1); 219 return EIO_ERRNO (ENOENT, -1);
128 } 220 }
129 221
222 /* POSIX API only, causing trouble for win32 apps */
223 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
224 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
225
226 struct statvfs
227 {
228 int dummy;
229 };
230
231 #define DT_DIR EIO_DT_DIR
232 #define DT_REG EIO_DT_REG
233 #define D_NAME(entp) entp.cFileName
234 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
235
130#else 236#else
131 237
238 #include <sys/ioctl.h>
132 #include <sys/time.h> 239 #include <sys/time.h>
133 #include <sys/select.h> 240 #include <sys/select.h>
134 #include <sys/statvfs.h>
135 #include <unistd.h> 241 #include <unistd.h>
136 #include <utime.h>
137 #include <signal.h> 242 #include <signal.h>
138 #include <dirent.h> 243 #include <dirent.h>
244
245 #ifdef ANDROID
246 #include <sys/vfs.h>
247 #define statvfs statfs
248 #define fstatvfs fstatfs
249 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
250 #else
251 #include <sys/statvfs.h>
252 #endif
139 253
140 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES 254 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
141 #include <sys/mman.h> 255 #include <sys/mman.h>
142 #endif 256 #endif
143 257
258 #define D_NAME(entp) entp->d_name
259
144 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */ 260 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
145 #if __FreeBSD__ || defined __NetBSD__ || defined __OpenBSD__ 261 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
146 #define _DIRENT_HAVE_D_TYPE /* sigh */ 262 #define _DIRENT_HAVE_D_TYPE /* sigh */
147 #define D_INO(de) (de)->d_fileno 263 #define D_INO(de) (de)->d_fileno
148 #define D_NAMLEN(de) (de)->d_namlen 264 #define D_NAMLEN(de) (de)->d_namlen
149 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600 265 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
150 #define D_INO(de) (de)->d_ino 266 #define D_INO(de) (de)->d_ino
161 277
162 #ifndef EIO_STRUCT_DIRENT 278 #ifndef EIO_STRUCT_DIRENT
163 #define EIO_STRUCT_DIRENT struct dirent 279 #define EIO_STRUCT_DIRENT struct dirent
164 #endif 280 #endif
165 281
282#endif
283
284#if HAVE_UTIMES
285# include <utime.h>
286#endif
287
288#if HAVE_SYS_SYSCALL_H
289# include <sys/syscall.h>
166#endif 290#endif
167 291
168#if HAVE_SENDFILE 292#if HAVE_SENDFILE
169# if __linux 293# if __linux
170# include <sys/sendfile.h> 294# include <sys/sendfile.h>
178# else 302# else
179# error sendfile support requested but not available 303# error sendfile support requested but not available
180# endif 304# endif
181#endif 305#endif
182 306
307#if HAVE_RENAMEAT2
308# include <sys/syscall.h>
309# include <linux/fs.h>
310#endif
311
183#ifndef D_TYPE 312#ifndef D_TYPE
184# define D_TYPE(de) 0 313# define D_TYPE(de) 0
185#endif 314#endif
186#ifndef D_INO 315#ifndef D_INO
187# define D_INO(de) 0 316# define D_INO(de) 0
188#endif 317#endif
189#ifndef D_NAMLEN 318#ifndef D_NAMLEN
190# define D_NAMLEN(de) strlen ((de)->d_name) 319# define D_NAMLEN(entp) strlen (D_NAME (entp))
191#endif 320#endif
192 321
193/* used for struct dirent, AIX doesn't provide it */ 322/* used for struct dirent, AIX doesn't provide it */
194#ifndef NAME_MAX 323#ifndef NAME_MAX
195# define NAME_MAX 4096 324# define NAME_MAX 4096
196#endif 325#endif
197 326
198/* used for readlink etc. */ 327/* used for readlink etc. */
199#ifndef PATH_MAX 328#ifndef PATH_MAX
200# define PATH_MAX 4096 329# define PATH_MAX 0
201#endif 330#endif
331
332#ifndef EIO_PATH_MIN
333# define EIO_PATH_MIN 8160
334#endif
335
336#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
202 337
203/* buffer size for various temporary buffers */ 338/* buffer size for various temporary buffers */
204#define EIO_BUFSIZE 65536 339#define EIO_BUFSIZE 65536
205 340
206#define dBUF \ 341#define dBUF \
207 char *eio_buf; \
208 ETP_WORKER_LOCK (self); \
209 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 342 char *eio_buf = malloc (EIO_BUFSIZE); \
210 ETP_WORKER_UNLOCK (self); \
211 errno = ENOMEM; \ 343 errno = ENOMEM; \
212 if (!eio_buf) \ 344 if (!eio_buf) \
213 return -1; 345 return -1
214 346
215#define EIO_TICKS ((1000000 + 1023) >> 10) 347#define FUBd \
348 free (eio_buf)
349
350/*****************************************************************************/
351
352struct etp_tmpbuf;
353
354#if _POSIX_VERSION >= 200809L
355 #define HAVE_AT 1
356 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
357 #ifndef O_SEARCH
358 #define O_SEARCH O_RDONLY
359 #endif
360#else
361 #define HAVE_AT 0
362 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
363#endif
364
365struct eio_pwd
366{
367#if HAVE_AT
368 int fd;
369#endif
370 int len;
371 char str[1]; /* actually, a 0-terminated canonical path */
372};
373
374/*****************************************************************************/
216 375
217#define ETP_PRI_MIN EIO_PRI_MIN 376#define ETP_PRI_MIN EIO_PRI_MIN
218#define ETP_PRI_MAX EIO_PRI_MAX 377#define ETP_PRI_MAX EIO_PRI_MAX
219 378
379#define ETP_TYPE_QUIT -1
380#define ETP_TYPE_GROUP EIO_GROUP
381
382static void eio_nop_callback (void) { }
383static void (*eio_want_poll_cb)(void) = eio_nop_callback;
384static void (*eio_done_poll_cb)(void) = eio_nop_callback;
385
386#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
387#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
388
220struct etp_worker; 389struct etp_worker;
221
222#define ETP_REQ eio_req 390#define ETP_REQ eio_req
223#define ETP_DESTROY(req) eio_destroy (req) 391#define ETP_DESTROY(req) eio_destroy (req)
224static int eio_finish (eio_req *req); 392static int eio_finish (eio_req *req);
225#define ETP_FINISH(req) eio_finish (req) 393#define ETP_FINISH(req) eio_finish (req)
226static void eio_execute (struct etp_worker *self, eio_req *req); 394static void eio_execute (struct etp_worker *self, eio_req *req);
227#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 395#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
228 396
229#define ETP_WORKER_CLEAR(req) \ 397#include "etp.c"
230 if (wrk->dbuf) \
231 { \
232 free (wrk->dbuf); \
233 wrk->dbuf = 0; \
234 } \
235 \
236 if (wrk->dirp) \
237 { \
238 closedir (wrk->dirp); \
239 wrk->dirp = 0; \
240 }
241 398
242#define ETP_WORKER_COMMON \ 399static struct etp_pool eio_pool;
243 void *dbuf; \ 400#define EIO_POOL (&eio_pool)
244 DIR *dirp;
245
246/*****************************************************************************/
247
248#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
249
250/* calculate time difference in ~1/EIO_TICKS of a second */
251ecb_inline int
252tvdiff (struct timeval *tv1, struct timeval *tv2)
253{
254 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
255 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
256}
257
258static unsigned int started, idle, wanted = 4;
259
260static void (*want_poll_cb) (void);
261static void (*done_poll_cb) (void);
262
263static unsigned int max_poll_time; /* reslock */
264static unsigned int max_poll_reqs; /* reslock */
265
266static volatile unsigned int nreqs; /* reqlock */
267static volatile unsigned int nready; /* reqlock */
268static volatile unsigned int npending; /* reqlock */
269static volatile unsigned int max_idle = 4; /* maximum number of threads that can idle indefinitely */
270static volatile unsigned int idle_timeout = 10; /* number of seconds after which an idle threads exit */
271
272static xmutex_t wrklock;
273static xmutex_t reslock;
274static xmutex_t reqlock;
275static xcond_t reqwait;
276
277#if !HAVE_PREADWRITE
278/*
279 * make our pread/pwrite emulation safe against themselves, but not against
280 * normal read/write by using a mutex. slows down execution a lot,
281 * but that's your problem, not mine.
282 */
283static xmutex_t preadwritelock = X_MUTEX_INIT;
284#endif
285
286typedef struct etp_worker
287{
288 /* locked by wrklock */
289 struct etp_worker *prev, *next;
290
291 xthread_t tid;
292
293 /* locked by reslock, reqlock or wrklock */
294 ETP_REQ *req; /* currently processed request */
295
296 ETP_WORKER_COMMON
297} etp_worker;
298
299static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
300
301#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
302#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
303
304/* worker threads management */
305
306static void ecb_cold
307etp_worker_clear (etp_worker *wrk)
308{
309 ETP_WORKER_CLEAR (wrk);
310}
311
312static void ecb_cold
313etp_worker_free (etp_worker *wrk)
314{
315 wrk->next->prev = wrk->prev;
316 wrk->prev->next = wrk->next;
317
318 free (wrk);
319}
320
321static unsigned int
322etp_nreqs (void)
323{
324 int retval;
325 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
326 retval = nreqs;
327 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
328 return retval;
329}
330
331static unsigned int
332etp_nready (void)
333{
334 unsigned int retval;
335
336 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
337 retval = nready;
338 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
339
340 return retval;
341}
342
343static unsigned int
344etp_npending (void)
345{
346 unsigned int retval;
347
348 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
349 retval = npending;
350 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
351
352 return retval;
353}
354
355static unsigned int
356etp_nthreads (void)
357{
358 unsigned int retval;
359
360 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
361 retval = started;
362 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
363
364 return retval;
365}
366
367/*
368 * a somewhat faster data structure might be nice, but
369 * with 8 priorities this actually needs <20 insns
370 * per shift, the most expensive operation.
371 */
372typedef struct {
373 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
374 int size;
375} etp_reqq;
376
377static etp_reqq req_queue;
378static etp_reqq res_queue;
379
380static int ecb_noinline
381reqq_push (etp_reqq *q, ETP_REQ *req)
382{
383 int pri = req->pri;
384 req->next = 0;
385
386 if (q->qe[pri])
387 {
388 q->qe[pri]->next = req;
389 q->qe[pri] = req;
390 }
391 else
392 q->qe[pri] = q->qs[pri] = req;
393
394 return q->size++;
395}
396
397static ETP_REQ * ecb_noinline
398reqq_shift (etp_reqq *q)
399{
400 int pri;
401
402 if (!q->size)
403 return 0;
404
405 --q->size;
406
407 for (pri = ETP_NUM_PRI; pri--; )
408 {
409 eio_req *req = q->qs[pri];
410
411 if (req)
412 {
413 if (!(q->qs[pri] = (eio_req *)req->next))
414 q->qe[pri] = 0;
415
416 return req;
417 }
418 }
419
420 abort ();
421}
422
423static void ecb_cold
424etp_thread_init (void)
425{
426#if !HAVE_PREADWRITE
427 X_MUTEX_CREATE (preadwritelock);
428#endif
429 X_MUTEX_CREATE (wrklock);
430 X_MUTEX_CREATE (reslock);
431 X_MUTEX_CREATE (reqlock);
432 X_COND_CREATE (reqwait);
433}
434
435static void ecb_cold
436etp_atfork_prepare (void)
437{
438}
439
440static void ecb_cold
441etp_atfork_parent (void)
442{
443}
444
445static void ecb_cold
446etp_atfork_child (void)
447{
448 ETP_REQ *prv;
449
450 while ((prv = reqq_shift (&req_queue)))
451 ETP_DESTROY (prv);
452
453 while ((prv = reqq_shift (&res_queue)))
454 ETP_DESTROY (prv);
455
456 while (wrk_first.next != &wrk_first)
457 {
458 etp_worker *wrk = wrk_first.next;
459
460 if (wrk->req)
461 ETP_DESTROY (wrk->req);
462
463 etp_worker_clear (wrk);
464 etp_worker_free (wrk);
465 }
466
467 started = 0;
468 idle = 0;
469 nreqs = 0;
470 nready = 0;
471 npending = 0;
472
473 etp_thread_init ();
474}
475
476static void ecb_cold
477etp_once_init (void)
478{
479 etp_thread_init ();
480 X_THREAD_ATFORK (etp_atfork_prepare, etp_atfork_parent, etp_atfork_child);
481}
482
483static int ecb_cold
484etp_init (void (*want_poll)(void), void (*done_poll)(void))
485{
486 static pthread_once_t doinit = PTHREAD_ONCE_INIT;
487
488 pthread_once (&doinit, etp_once_init);
489
490 want_poll_cb = want_poll;
491 done_poll_cb = done_poll;
492
493 return 0;
494}
495
496X_THREAD_PROC (etp_proc);
497
498static void ecb_cold
499etp_start_thread (void)
500{
501 etp_worker *wrk = calloc (1, sizeof (etp_worker));
502
503 /*TODO*/
504 assert (("unable to allocate worker thread data", wrk));
505
506 X_LOCK (wrklock);
507
508 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
509 {
510 wrk->prev = &wrk_first;
511 wrk->next = wrk_first.next;
512 wrk_first.next->prev = wrk;
513 wrk_first.next = wrk;
514 ++started;
515 }
516 else
517 free (wrk);
518
519 X_UNLOCK (wrklock);
520}
521
522static void
523etp_maybe_start_thread (void)
524{
525 if (ecb_expect_true (etp_nthreads () >= wanted))
526 return;
527
528 /* todo: maybe use idle here, but might be less exact */
529 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
530 return;
531
532 etp_start_thread ();
533}
534
535static void ecb_cold
536etp_end_thread (void)
537{
538 eio_req *req = calloc (1, sizeof (eio_req));
539
540 req->type = -1;
541 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
542
543 X_LOCK (reqlock);
544 reqq_push (&req_queue, req);
545 X_COND_SIGNAL (reqwait);
546 X_UNLOCK (reqlock);
547
548 X_LOCK (wrklock);
549 --started;
550 X_UNLOCK (wrklock);
551}
552
553static int
554etp_poll (void)
555{
556 unsigned int maxreqs;
557 unsigned int maxtime;
558 struct timeval tv_start, tv_now;
559
560 X_LOCK (reslock);
561 maxreqs = max_poll_reqs;
562 maxtime = max_poll_time;
563 X_UNLOCK (reslock);
564
565 if (maxtime)
566 gettimeofday (&tv_start, 0);
567
568 for (;;)
569 {
570 ETP_REQ *req;
571
572 etp_maybe_start_thread ();
573
574 X_LOCK (reslock);
575 req = reqq_shift (&res_queue);
576
577 if (req)
578 {
579 --npending;
580
581 if (!res_queue.size && done_poll_cb)
582 done_poll_cb ();
583 }
584
585 X_UNLOCK (reslock);
586
587 if (!req)
588 return 0;
589
590 X_LOCK (reqlock);
591 --nreqs;
592 X_UNLOCK (reqlock);
593
594 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
595 {
596 req->int1 = 1; /* mark request as delayed */
597 continue;
598 }
599 else
600 {
601 int res = ETP_FINISH (req);
602 if (ecb_expect_false (res))
603 return res;
604 }
605
606 if (ecb_expect_false (maxreqs && !--maxreqs))
607 break;
608
609 if (maxtime)
610 {
611 gettimeofday (&tv_now, 0);
612
613 if (tvdiff (&tv_start, &tv_now) >= maxtime)
614 break;
615 }
616 }
617
618 errno = EAGAIN;
619 return -1;
620}
621
622static void
623etp_cancel (ETP_REQ *req)
624{
625 req->cancelled = 1;
626
627 eio_grp_cancel (req);
628}
629
630static void
631etp_submit (ETP_REQ *req)
632{
633 req->pri -= ETP_PRI_MIN;
634
635 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
636 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
637
638 if (ecb_expect_false (req->type == EIO_GROUP))
639 {
640 /* I hope this is worth it :/ */
641 X_LOCK (reqlock);
642 ++nreqs;
643 X_UNLOCK (reqlock);
644
645 X_LOCK (reslock);
646
647 ++npending;
648
649 if (!reqq_push (&res_queue, req) && want_poll_cb)
650 want_poll_cb ();
651
652 X_UNLOCK (reslock);
653 }
654 else
655 {
656 X_LOCK (reqlock);
657 ++nreqs;
658 ++nready;
659 reqq_push (&req_queue, req);
660 X_COND_SIGNAL (reqwait);
661 X_UNLOCK (reqlock);
662
663 etp_maybe_start_thread ();
664 }
665}
666
667static void ecb_cold
668etp_set_max_poll_time (double nseconds)
669{
670 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
671 max_poll_time = nseconds * EIO_TICKS;
672 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
673}
674
675static void ecb_cold
676etp_set_max_poll_reqs (unsigned int maxreqs)
677{
678 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
679 max_poll_reqs = maxreqs;
680 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
681}
682
683static void ecb_cold
684etp_set_max_idle (unsigned int nthreads)
685{
686 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
687 max_idle = nthreads;
688 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
689}
690
691static void ecb_cold
692etp_set_idle_timeout (unsigned int seconds)
693{
694 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
695 idle_timeout = seconds;
696 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
697}
698
699static void ecb_cold
700etp_set_min_parallel (unsigned int nthreads)
701{
702 if (wanted < nthreads)
703 wanted = nthreads;
704}
705
706static void ecb_cold
707etp_set_max_parallel (unsigned int nthreads)
708{
709 if (wanted > nthreads)
710 wanted = nthreads;
711
712 while (started > wanted)
713 etp_end_thread ();
714}
715 401
716/*****************************************************************************/ 402/*****************************************************************************/
717 403
718static void 404static void
719grp_try_feed (eio_req *grp) 405grp_try_feed (eio_req *grp)
720{ 406{
721 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 407 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
722 { 408 {
723 grp->flags &= ~EIO_FLAG_GROUPADD; 409 grp->flags &= ~ETP_FLAG_GROUPADD;
724 410
725 EIO_FEED (grp); 411 EIO_FEED (grp);
726 412
727 /* stop if no progress has been made */ 413 /* stop if no progress has been made */
728 if (!(grp->flags & EIO_FLAG_GROUPADD)) 414 if (!(grp->flags & ETP_FLAG_GROUPADD))
729 { 415 {
730 grp->feed = 0; 416 grp->feed = 0;
731 break; 417 break;
732 } 418 }
733 } 419 }
740 426
741 /* call feeder, if applicable */ 427 /* call feeder, if applicable */
742 grp_try_feed (grp); 428 grp_try_feed (grp);
743 429
744 /* finish, if done */ 430 /* finish, if done */
745 if (!grp->size && grp->int1) 431 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
746 return eio_finish (grp); 432 return eio_finish (grp);
747 else 433 else
748 return 0; 434 return 0;
749} 435}
750 436
786} 472}
787 473
788void 474void
789eio_grp_cancel (eio_req *grp) 475eio_grp_cancel (eio_req *grp)
790{ 476{
791 for (grp = grp->grp_first; grp; grp = grp->grp_next) 477 etp_grp_cancel (EIO_POOL, grp);
792 eio_cancel (grp);
793} 478}
794 479
795void 480void
796eio_cancel (eio_req *req) 481eio_cancel (eio_req *req)
797{ 482{
798 etp_cancel (req); 483 etp_cancel (EIO_POOL, req);
799} 484}
800 485
801void 486void
802eio_submit (eio_req *req) 487eio_submit (eio_req *req)
803{ 488{
804 etp_submit (req); 489 etp_submit (EIO_POOL, req);
805} 490}
806 491
807unsigned int 492unsigned int
808eio_nreqs (void) 493eio_nreqs (void)
809{ 494{
810 return etp_nreqs (); 495 return etp_nreqs (EIO_POOL);
811} 496}
812 497
813unsigned int 498unsigned int
814eio_nready (void) 499eio_nready (void)
815{ 500{
816 return etp_nready (); 501 return etp_nready (EIO_POOL);
817} 502}
818 503
819unsigned int 504unsigned int
820eio_npending (void) 505eio_npending (void)
821{ 506{
822 return etp_npending (); 507 return etp_npending (EIO_POOL);
823} 508}
824 509
825unsigned int ecb_cold 510unsigned int ecb_cold
826eio_nthreads (void) 511eio_nthreads (void)
827{ 512{
828 return etp_nthreads (); 513 return etp_nthreads (EIO_POOL);
829} 514}
830 515
831void ecb_cold 516void ecb_cold
832eio_set_max_poll_time (double nseconds) 517eio_set_max_poll_time (double nseconds)
833{ 518{
834 etp_set_max_poll_time (nseconds); 519 etp_set_max_poll_time (EIO_POOL, nseconds);
835} 520}
836 521
837void ecb_cold 522void ecb_cold
838eio_set_max_poll_reqs (unsigned int maxreqs) 523eio_set_max_poll_reqs (unsigned int maxreqs)
839{ 524{
840 etp_set_max_poll_reqs (maxreqs); 525 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
841} 526}
842 527
843void ecb_cold 528void ecb_cold
844eio_set_max_idle (unsigned int nthreads) 529eio_set_max_idle (unsigned int nthreads)
845{ 530{
846 etp_set_max_idle (nthreads); 531 etp_set_max_idle (EIO_POOL, nthreads);
847} 532}
848 533
849void ecb_cold 534void ecb_cold
850eio_set_idle_timeout (unsigned int seconds) 535eio_set_idle_timeout (unsigned int seconds)
851{ 536{
852 etp_set_idle_timeout (seconds); 537 etp_set_idle_timeout (EIO_POOL, seconds);
853} 538}
854 539
855void ecb_cold 540void ecb_cold
856eio_set_min_parallel (unsigned int nthreads) 541eio_set_min_parallel (unsigned int nthreads)
857{ 542{
858 etp_set_min_parallel (nthreads); 543 etp_set_min_parallel (EIO_POOL, nthreads);
859} 544}
860 545
861void ecb_cold 546void ecb_cold
862eio_set_max_parallel (unsigned int nthreads) 547eio_set_max_parallel (unsigned int nthreads)
863{ 548{
864 etp_set_max_parallel (nthreads); 549 etp_set_max_parallel (EIO_POOL, nthreads);
865} 550}
866 551
867int eio_poll (void) 552int eio_poll (void)
868{ 553{
869 return etp_poll (); 554 return etp_poll (EIO_POOL);
870} 555}
871 556
872/*****************************************************************************/ 557/*****************************************************************************/
873/* work around various missing functions */ 558/* work around various missing functions */
874 559
875#if !HAVE_PREADWRITE 560#if HAVE_POSIX_CLOSE && !__linux
876# undef pread 561# define eio__close(fd) posix_close (fd, 0)
877# undef pwrite 562#else
878# define pread eio__pread 563# define eio__close(fd) close (fd)
879# define pwrite eio__pwrite
880
881static ssize_t
882eio__pread (int fd, void *buf, size_t count, off_t offset)
883{
884 ssize_t res;
885 off_t ooffset;
886
887 X_LOCK (preadwritelock);
888 ooffset = lseek (fd, 0, SEEK_CUR);
889 lseek (fd, offset, SEEK_SET);
890 res = read (fd, buf, count);
891 lseek (fd, ooffset, SEEK_SET);
892 X_UNLOCK (preadwritelock);
893
894 return res;
895}
896
897static ssize_t
898eio__pwrite (int fd, void *buf, size_t count, off_t offset)
899{
900 ssize_t res;
901 off_t ooffset;
902
903 X_LOCK (preadwritelock);
904 ooffset = lseek (fd, 0, SEEK_CUR);
905 lseek (fd, offset, SEEK_SET);
906 res = write (fd, buf, count);
907 lseek (fd, ooffset, SEEK_SET);
908 X_UNLOCK (preadwritelock);
909
910 return res;
911}
912#endif 564#endif
565
566/* close() without disturbing errno */
567static void
568silent_close (int fd)
569{
570 int saved_errno = errno;
571 eio__close (fd);
572 errno = saved_errno;
573}
913 574
914#ifndef HAVE_UTIMES 575#ifndef HAVE_UTIMES
915 576
916# undef utimes 577# undef utimes
917# define utimes(path,times) eio__utimes (path, times) 578# define utimes(path,times) eio__utimes (path, times)
950 611
951#if !HAVE_FDATASYNC 612#if !HAVE_FDATASYNC
952# undef fdatasync 613# undef fdatasync
953# define fdatasync(fd) fsync (fd) 614# define fdatasync(fd) fsync (fd)
954#endif 615#endif
616
617static int
618eio__syncfs (int fd)
619{
620 int res;
621
622#if HAVE_SYS_SYNCFS
623 res = (int)syscall (__NR_syncfs, (int)(fd));
624#else
625 res = EIO_ENOSYS ();
626#endif
627
628 if (res < 0 && errno == ENOSYS && fd >= 0)
629 sync ();
630
631 return res;
632}
955 633
956/* sync_file_range always needs emulation */ 634/* sync_file_range always needs emulation */
957static int 635static int
958eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 636eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
959{ 637{
982} 660}
983 661
984static int 662static int
985eio__fallocate (int fd, int mode, off_t offset, size_t len) 663eio__fallocate (int fd, int mode, off_t offset, size_t len)
986{ 664{
987#if HAVE_FALLOCATE 665#if HAVE_LINUX_FALLOCATE
988 return fallocate (fd, mode, offset, len); 666 return fallocate (fd, mode, offset, len);
989#else 667#else
990 errno = ENOSYS; 668 return EIO_ENOSYS ();
991 return -1;
992#endif 669#endif
993} 670}
994 671
995#if !HAVE_READAHEAD 672#if !HAVE_READAHEAD
996# undef readahead 673# undef readahead
997# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 674# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
998 675
999static ssize_t 676static eio_ssize_t
1000eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 677eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
1001{ 678{
1002 size_t todo = count; 679 size_t todo = count;
1003 dBUF; 680 dBUF;
1004 681
1009 pread (fd, eio_buf, len, offset); 686 pread (fd, eio_buf, len, offset);
1010 offset += len; 687 offset += len;
1011 todo -= len; 688 todo -= len;
1012 } 689 }
1013 690
1014 errno = 0; 691 FUBd;
692
693 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
694 /* but not for e.g. EIO or eof, so we also never fail */
1015 return count; 695 return 0;
1016} 696}
1017 697
1018#endif 698#endif
1019 699
1020/* sendfile always needs emulation */ 700/* sendfile always needs emulation */
1021static ssize_t 701static eio_ssize_t
1022eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 702eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
1023{ 703{
1024 ssize_t written = 0; 704 eio_ssize_t written = 0;
1025 ssize_t res; 705 eio_ssize_t res;
1026 706
1027 if (!count) 707 if (!count)
1028 return 0; 708 return 0;
1029 709
1030 for (;;) 710 for (;;)
1055 735
1056 /* according to source inspection, this is correct, and useful behaviour */ 736 /* according to source inspection, this is correct, and useful behaviour */
1057 if (sbytes) 737 if (sbytes)
1058 res = sbytes; 738 res = sbytes;
1059 739
1060# elif defined (__APPLE__) 740# elif defined __APPLE__
1061 off_t sbytes = count; 741 off_t sbytes = count;
1062 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 742 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1063 743
1064 /* according to the manpage, sbytes is always valid */ 744 /* according to the manpage, sbytes is always valid */
1065 if (sbytes) 745 if (sbytes)
1092 HANDLE h = TO_SOCKET (ifd); 772 HANDLE h = TO_SOCKET (ifd);
1093 SetFilePointer (h, offset, 0, FILE_BEGIN); 773 SetFilePointer (h, offset, 0, FILE_BEGIN);
1094 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 774 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1095 775
1096#else 776#else
1097 res = -1; 777 res = EIO_ENOSYS ();
1098 errno = ENOSYS;
1099#endif 778#endif
1100 779
1101 /* we assume sendfile can copy at least 128mb in one go */ 780 /* we assume sendfile can copy at least 128mb in one go */
1102 if (res <= 128 * 1024 * 1024) 781 if (res <= 128 * 1024 * 1024)
1103 { 782 {
1125 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 804 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1126 /* BSDs */ 805 /* BSDs */
1127#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 806#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1128 || errno == ENOTSUP 807 || errno == ENOTSUP
1129#endif 808#endif
809#ifdef EOPNOTSUPP /* windows */
1130 || errno == EOPNOTSUPP /* BSDs */ 810 || errno == EOPNOTSUPP /* BSDs */
811#endif
1131#if __solaris 812#if __solaris
1132 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 813 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1133#endif 814#endif
1134 ) 815 )
1135 ) 816 )
1139 820
1140 res = 0; 821 res = 0;
1141 822
1142 while (count) 823 while (count)
1143 { 824 {
1144 ssize_t cnt; 825 eio_ssize_t cnt;
1145 826
1146 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 827 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1147 828
1148 if (cnt <= 0) 829 if (cnt <= 0)
1149 { 830 {
1161 842
1162 offset += cnt; 843 offset += cnt;
1163 res += cnt; 844 res += cnt;
1164 count -= cnt; 845 count -= cnt;
1165 } 846 }
847
848 FUBd;
1166 } 849 }
1167 850
1168 return res; 851 return res;
1169} 852}
1170 853
1197 /* round up length */ 880 /* round up length */
1198 *length = (*length + mask) & ~mask; 881 *length = (*length + mask) & ~mask;
1199} 882}
1200 883
1201#if !_POSIX_MEMLOCK 884#if !_POSIX_MEMLOCK
1202# define eio__mlockall(a) eio_nosyscall() 885# define eio__mlockall(a) EIO_ENOSYS ()
1203#else 886#else
1204 887
1205static int 888static int
1206eio__mlockall (int flags) 889eio__mlockall (int flags)
1207{ 890{
1286 969
1287/*****************************************************************************/ 970/*****************************************************************************/
1288/* requests implemented outside eio_execute, because they are so large */ 971/* requests implemented outside eio_execute, because they are so large */
1289 972
1290static void 973static void
1291eio__realpath (eio_req *req, etp_worker *self) 974eio__lseek (eio_req *req)
1292{ 975{
1293 char *rel = req->ptr1; 976 /* this usually gets optimised away completely, or your compiler sucks, */
977 /* or the whence constants really are not 0, 1, 2 */
978 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
979 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
980 : req->int2 == EIO_SEEK_END ? SEEK_END
981 : req->int2;
982
983 req->offs = lseek (req->int1, req->offs, whence);
984 req->result = req->offs == (off_t)-1 ? -1 : 0;
985}
986
987/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
988static int
989eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
990{
1294 char *res; 991 char *res;
992 const char *rel = path;
1295 char *tmp1, *tmp2; 993 char *tmp1, *tmp2;
1296#if SYMLOOP_MAX > 32 994#if SYMLOOP_MAX > 32
1297 int symlinks = SYMLOOP_MAX; 995 int symlinks = SYMLOOP_MAX;
1298#else 996#else
1299 int symlinks = 32; 997 int symlinks = 32;
1300#endif 998#endif
1301 999
1302 req->result = -1;
1303
1304 errno = EINVAL; 1000 errno = EINVAL;
1305 if (!rel) 1001 if (!rel)
1306 return; 1002 return -1;
1307 1003
1308 errno = ENOENT; 1004 errno = ENOENT;
1309 if (!*rel) 1005 if (!*rel)
1310 return; 1006 return -1;
1311 1007
1312 if (!req->ptr2) 1008 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1313 { 1009#ifdef _WIN32
1314 X_LOCK (wrklock); 1010 if (_access (rel, 4) != 0)
1315 req->flags |= EIO_FLAG_PTR2_FREE;
1316 X_UNLOCK (wrklock);
1317 req->ptr2 = malloc (PATH_MAX * 3);
1318
1319 errno = ENOMEM;
1320 if (!req->ptr2)
1321 return; 1011 return -1;
1322 }
1323 1012
1324 res = req->ptr2; 1013 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1014
1015 errno = ENAMETOOLONG;
1016 if (symlinks >= EIO_PATH_MAX * 3)
1017 return -1;
1018
1019 errno = EIO;
1020 if (symlinks <= 0)
1021 return -1;
1022
1023 return symlinks;
1024
1025#else
1325 tmp1 = res + PATH_MAX; 1026 tmp1 = res + EIO_PATH_MAX;
1326 tmp2 = tmp1 + PATH_MAX; 1027 tmp2 = tmp1 + EIO_PATH_MAX;
1327 1028
1328#if 0 /* disabled, the musl way to do things is just too racy */ 1029#if 0 /* disabled, the musl way to do things is just too racy */
1329#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME) 1030#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1330 /* on linux we may be able to ask the kernel */ 1031 /* on linux we may be able to ask the kernel */
1331 { 1032 {
1332 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME); 1033 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1333 1034
1334 if (fd >= 0) 1035 if (fd >= 0)
1335 { 1036 {
1336 sprintf (tmp1, "/proc/self/fd/%d", fd); 1037 sprintf (tmp1, "/proc/self/fd/%d", fd);
1337 req->result = readlink (tmp1, res, PATH_MAX); 1038 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1338 close (fd);
1339
1340 /* here we should probably stat the open file and the disk file, to make sure they still match */ 1039 /* here we should probably stat the open file and the disk file, to make sure they still match */
1040 eio__close (fd);
1341 1041
1342 if (req->result > 0) 1042 if (req->result > 0)
1343 goto done; 1043 goto done;
1344 } 1044 }
1345 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO) 1045 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1346 return; 1046 return -1;
1347 } 1047 }
1348#endif 1048#endif
1349#endif 1049#endif
1350 1050
1351 if (*rel != '/') 1051 if (*rel != '/')
1352 { 1052 {
1353 if (!getcwd (res, PATH_MAX)) 1053 int len;
1054
1055 errno = ENOENT;
1056 if (wd == EIO_INVALID_WD)
1354 return; 1057 return -1;
1058
1059 if (wd == EIO_CWD)
1060 {
1061 if (!getcwd (res, EIO_PATH_MAX))
1062 return -1;
1063
1064 len = strlen (res);
1065 }
1066 else
1067 memcpy (res, wd->str, len = wd->len);
1355 1068
1356 if (res [1]) /* only use if not / */ 1069 if (res [1]) /* only use if not / */
1357 res += strlen (res); 1070 res += len;
1358 } 1071 }
1359 1072
1360 while (*rel) 1073 while (*rel)
1361 { 1074 {
1362 ssize_t len, linklen; 1075 eio_ssize_t len, linklen;
1363 char *beg = rel; 1076 const char *beg = rel;
1364 1077
1365 while (*rel && *rel != '/') 1078 while (*rel && *rel != '/')
1366 ++rel; 1079 ++rel;
1367 1080
1368 len = rel - beg; 1081 len = rel - beg;
1380 1093
1381 if (beg [1] == '.' && len == 2) 1094 if (beg [1] == '.' && len == 2)
1382 { 1095 {
1383 /* .. - back up one component, if possible */ 1096 /* .. - back up one component, if possible */
1384 1097
1385 while (res != req->ptr2) 1098 while (res != tmpbuf->ptr)
1386 if (*--res == '/') 1099 if (*--res == '/')
1387 break; 1100 break;
1388 1101
1389 continue; 1102 continue;
1390 } 1103 }
1391 } 1104 }
1392 1105
1393 errno = ENAMETOOLONG; 1106 errno = ENAMETOOLONG;
1394 if (res + 1 + len + 1 >= tmp1) 1107 if (res + 1 + len + 1 >= tmp1)
1395 return; 1108 return -1;
1396 1109
1397 /* copy one component */ 1110 /* copy one component */
1398 *res = '/'; 1111 *res = '/';
1399 memcpy (res + 1, beg, len); 1112 memcpy (res + 1, beg, len);
1400 1113
1401 /* zero-terminate, for readlink */ 1114 /* zero-terminate, for readlink */
1402 res [len + 1] = 0; 1115 res [len + 1] = 0;
1403 1116
1404 /* now check if it's a symlink */ 1117 /* now check if it's a symlink */
1405 linklen = readlink (req->ptr2, tmp1, PATH_MAX); 1118 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1406 1119
1407 if (linklen < 0) 1120 if (linklen < 0)
1408 { 1121 {
1409 if (errno != EINVAL) 1122 if (errno != EINVAL)
1410 return; 1123 return -1;
1411 1124
1412 /* it's a normal directory. hopefully */ 1125 /* it's a normal directory. hopefully */
1413 res += len + 1; 1126 res += len + 1;
1414 } 1127 }
1415 else 1128 else
1416 { 1129 {
1417 /* yay, it was a symlink - build new path in tmp2 */ 1130 /* yay, it was a symlink - build new path in tmp2 */
1418 int rellen = strlen (rel); 1131 int rellen = strlen (rel);
1419 1132
1420 errno = ENAMETOOLONG; 1133 errno = ENAMETOOLONG;
1421 if (linklen + 1 + rellen >= PATH_MAX) 1134 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1422 return; 1135 return -1;
1423 1136
1424 errno = ELOOP; 1137 errno = ELOOP;
1425 if (!--symlinks) 1138 if (!--symlinks)
1426 return; 1139 return -1;
1427 1140
1428 if (*tmp1 == '/') 1141 if (*tmp1 == '/')
1429 res = req->ptr2; /* symlink resolves to an absolute path */ 1142 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1430 1143
1431 /* we need to be careful, as rel might point into tmp2 already */ 1144 /* we need to be careful, as rel might point into tmp2 already */
1432 memmove (tmp2 + linklen + 1, rel, rellen + 1); 1145 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1433 tmp2 [linklen] = '/'; 1146 tmp2 [linklen] = '/';
1434 memcpy (tmp2, tmp1, linklen); 1147 memcpy (tmp2, tmp1, linklen);
1436 rel = tmp2; 1149 rel = tmp2;
1437 } 1150 }
1438 } 1151 }
1439 1152
1440 /* special case for the lone root path */ 1153 /* special case for the lone root path */
1441 if (res == req->ptr2) 1154 if (res == tmpbuf->ptr)
1442 *res++ = '/'; 1155 *res++ = '/';
1443 1156
1444 req->result = res - (char *)req->ptr2; 1157 return res - (char *)tmpbuf->ptr;
1445 1158#endif
1446done:
1447 req->ptr2 = realloc (req->ptr2, req->result); /* trade time for space savings */
1448} 1159}
1449 1160
1450static signed char 1161static signed char
1451eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1162eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1452{ 1163{
1460 1171
1461#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1172#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1462#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1173#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1463 1174
1464static void 1175static void
1465eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1176eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1466{ 1177{
1467 unsigned char bits [9 + sizeof (ino_t) * 8]; 1178 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1468 unsigned char *bit = bits; 1179 unsigned char *bit = bits;
1469 1180
1470 assert (CHAR_BIT == 8); 1181 assert (CHAR_BIT == 8);
1471 assert (sizeof (eio_dirent) * 8 < 256); 1182 assert (sizeof (eio_dirent) * 8 < 256);
1472 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1183 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1474 1185
1475 if (size <= EIO_SORT_FAST) 1186 if (size <= EIO_SORT_FAST)
1476 return; 1187 return;
1477 1188
1478 /* first prepare an array of bits to test in our radix sort */ 1189 /* first prepare an array of bits to test in our radix sort */
1479 /* try to take endianness into account, as well as differences in ino_t sizes */ 1190 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1480 /* inode_bits must contain all inodes ORed together */ 1191 /* inode_bits must contain all inodes ORed together */
1481 /* which is used to skip bits that are 0 everywhere, which is very common */ 1192 /* which is used to skip bits that are 0 everywhere, which is very common */
1482 { 1193 {
1483 ino_t endianness; 1194 eio_ino_t endianness;
1484 int i, j; 1195 int i, j;
1485 1196
1486 /* we store the byte offset of byte n into byte n of "endianness" */ 1197 /* we store the byte offset of byte n into byte n of "endianness" */
1487 for (i = 0; i < sizeof (ino_t); ++i) 1198 for (i = 0; i < sizeof (eio_ino_t); ++i)
1488 ((unsigned char *)&endianness)[i] = i; 1199 ((unsigned char *)&endianness)[i] = i;
1489 1200
1490 *bit++ = 0; 1201 *bit++ = 0;
1491 1202
1492 for (i = 0; i < sizeof (ino_t); ++i) 1203 for (i = 0; i < sizeof (eio_ino_t); ++i)
1493 { 1204 {
1494 /* shifting off the byte offsets out of "endianness" */ 1205 /* shifting off the byte offsets out of "endianness" */
1495 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1206 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1496 endianness >>= 8; 1207 endianness >>= 8;
1497 1208
1498 for (j = 0; j < 8; ++j) 1209 for (j = 0; j < 8; ++j)
1499 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1210 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1500 *bit++ = offs + j; 1211 *bit++ = offs + j;
1501 } 1212 }
1502 1213
1503 for (j = 0; j < 8; ++j) 1214 for (j = 0; j < 8; ++j)
1504 if (score_bits & (1 << j)) 1215 if (score_bits & (1 << j))
1505 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1216 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1506 } 1217 }
1507 1218
1508 /* now actually do the sorting (a variant of MSD radix sort) */ 1219 /* now actually do the sorting (a variant of MSD radix sort) */
1509 { 1220 {
1510 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1221 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1511 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1222 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1512 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1223 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1513 int stk_idx = 0; 1224 int stk_idx = 0;
1514 1225
1515 base_stk [stk_idx] = dents; 1226 base_stk [stk_idx] = dents;
1516 end_stk [stk_idx] = dents + size; 1227 end_stk [stk_idx] = dents + size;
1517 bit_stk [stk_idx] = bit - 1; 1228 bit_stk [stk_idx] = bit - 1;
1596 } 1307 }
1597 } 1308 }
1598} 1309}
1599 1310
1600static void 1311static void
1601eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1312eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1602{ 1313{
1603 if (size <= 1) 1314 if (size <= 1)
1604 return; /* our insertion sort relies on size > 0 */ 1315 return; /* our insertion sort relies on size > 0 */
1605 1316
1606 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1317 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1614 1325
1615/* read a full directory */ 1326/* read a full directory */
1616static void 1327static void
1617eio__scandir (eio_req *req, etp_worker *self) 1328eio__scandir (eio_req *req, etp_worker *self)
1618{ 1329{
1619 DIR *dirp;
1620 EIO_STRUCT_DIRENT *entp;
1621 char *name, *names; 1330 char *name, *names;
1622 int namesalloc = 4096; 1331 int namesalloc = 4096 - sizeof (void *) * 4;
1623 int namesoffs = 0; 1332 int namesoffs = 0;
1624 int flags = req->int1; 1333 int flags = req->int1;
1625 eio_dirent *dents = 0; 1334 eio_dirent *dents = 0;
1626 int dentalloc = 128; 1335 int dentalloc = 128;
1627 int dentoffs = 0; 1336 int dentoffs = 0;
1628 ino_t inode_bits = 0; 1337 eio_ino_t inode_bits = 0;
1338#ifdef _WIN32
1339 HANDLE dirp;
1340 WIN32_FIND_DATA entp;
1341#else
1342 DIR *dirp;
1343 EIO_STRUCT_DIRENT *entp;
1344#endif
1629 1345
1630 req->result = -1; 1346 req->result = -1;
1631 1347
1632 if (!(flags & EIO_READDIR_DENTS)) 1348 if (!(flags & EIO_READDIR_DENTS))
1633 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1349 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1634 1350
1635 X_LOCK (wrklock); 1351#ifdef _WIN32
1636 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1352 {
1353 int len = strlen ((const char *)req->ptr1);
1354 char *path = malloc (MAX_PATH);
1355 const char *fmt;
1356 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1357
1358 if (!len)
1359 fmt = "./*";
1360 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1361 fmt = "%s*";
1362 else
1363 fmt = "%s/*";
1364
1365 _snprintf (path, MAX_PATH, fmt, reqpath);
1366 dirp = FindFirstFile (path, &entp);
1367 free (path);
1368
1369 if (dirp == INVALID_HANDLE_VALUE)
1370 {
1371 /* should steal _dosmaperr */
1372 switch (GetLastError ())
1373 {
1374 case ERROR_FILE_NOT_FOUND:
1375 req->result = 0;
1376 break;
1377
1378 case ERROR_INVALID_NAME:
1379 case ERROR_PATH_NOT_FOUND:
1380 case ERROR_NO_MORE_FILES:
1381 errno = ENOENT;
1382 break;
1383
1384 case ERROR_NOT_ENOUGH_MEMORY:
1385 errno = ENOMEM;
1386 break;
1387
1388 default:
1389 errno = EINVAL;
1390 break;
1391 }
1392
1393 return;
1394 }
1395 }
1396#else
1397 #if HAVE_AT
1398 if (req->wd)
1399 {
1400 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1401
1402 if (fd < 0)
1403 return;
1404
1405 dirp = fdopendir (fd);
1406
1407 if (!dirp)
1408 silent_close (fd);
1409 }
1410 else
1637 self->dirp = dirp = opendir (req->ptr1); 1411 dirp = opendir (req->ptr1);
1412 #else
1413 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1414 #endif
1415
1416 if (!dirp)
1417 return;
1418#endif
1638 1419
1639 if (req->flags & EIO_FLAG_PTR1_FREE) 1420 if (req->flags & EIO_FLAG_PTR1_FREE)
1640 free (req->ptr1); 1421 free (req->ptr1);
1641 1422
1642 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1423 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1643 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1424 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1644 req->ptr2 = names = malloc (namesalloc); 1425 req->ptr2 = names = malloc (namesalloc);
1645 X_UNLOCK (wrklock);
1646 1426
1647 if (dirp && names && (!flags || dents)) 1427 if (!names || (flags && !dents))
1428 return;
1429
1648 for (;;) 1430 for (;;)
1649 { 1431 {
1432 int done;
1433
1434#ifdef _WIN32
1435 done = !dirp;
1436#else
1650 errno = 0; 1437 errno = 0;
1651 entp = readdir (dirp); 1438 entp = readdir (dirp);
1439 done = !entp;
1440#endif
1652 1441
1653 if (!entp) 1442 if (done)
1654 { 1443 {
1444#ifndef _WIN32
1445 int old_errno = errno;
1446 closedir (dirp);
1447 errno = old_errno;
1448
1655 if (errno) 1449 if (errno)
1656 break; 1450 break;
1451#endif
1657 1452
1658 /* sort etc. */ 1453 /* sort etc. */
1659 req->int1 = flags; 1454 req->int1 = flags;
1660 req->result = dentoffs; 1455 req->result = dentoffs;
1661 1456
1662 if (flags & EIO_READDIR_STAT_ORDER) 1457 if (flags & EIO_READDIR_STAT_ORDER)
1663 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1458 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1664 else if (flags & EIO_READDIR_DIRS_FIRST) 1459 else if (flags & EIO_READDIR_DIRS_FIRST)
1665 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1460 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1666 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1461 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1667 else 1462 else
1463 {
1464 /* in this case, all is known, and we just put dirs first and sort them */
1465 eio_dirent *oth = dents + dentoffs;
1466 eio_dirent *dir = dents;
1467
1468 /* now partition dirs to the front, and non-dirs to the back */
1469 /* by walking from both sides and swapping if necessary */
1470 while (oth > dir)
1471 {
1472 if (dir->type == EIO_DT_DIR)
1473 ++dir;
1474 else if ((--oth)->type == EIO_DT_DIR)
1475 {
1476 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1477
1478 ++dir;
1479 }
1480 }
1481
1482 /* now sort the dirs only (dirs all have the same score) */
1483 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1484 }
1485
1486 break;
1487 }
1488
1489 /* now add the entry to our list(s) */
1490 name = D_NAME (entp);
1491
1492 /* skip . and .. entries */
1493 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1494 {
1495 int len = D_NAMLEN (entp) + 1;
1496
1497 while (ecb_expect_false (namesoffs + len > namesalloc))
1498 {
1499 namesalloc *= 2;
1500 req->ptr2 = names = realloc (names, namesalloc);
1501
1502 if (!names)
1503 break;
1504 }
1505
1506 memcpy (names + namesoffs, name, len);
1507
1508 if (dents)
1509 {
1510 struct eio_dirent *ent;
1511
1512 if (ecb_expect_false (dentoffs == dentalloc))
1668 { 1513 {
1669 /* in this case, all is known, and we just put dirs first and sort them */ 1514 dentalloc *= 2;
1515 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1516
1517 if (!dents)
1518 break;
1519 }
1520
1670 eio_dirent *oth = dents + dentoffs; 1521 ent = dents + dentoffs;
1671 eio_dirent *dir = dents;
1672 1522
1673 /* now partition dirs to the front, and non-dirs to the back */ 1523 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1674 /* by walking from both sides and swapping if necessary */ 1524 ent->namelen = len - 1;
1675 while (oth > dir) 1525 ent->inode = D_INO (entp);
1526
1527 inode_bits |= ent->inode;
1528
1529 switch (D_TYPE (entp))
1530 {
1531 default:
1532 ent->type = EIO_DT_UNKNOWN;
1533 flags |= EIO_READDIR_FOUND_UNKNOWN;
1534 break;
1535
1536 #ifdef DT_FIFO
1537 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1538 #endif
1539 #ifdef DT_CHR
1540 case DT_CHR: ent->type = EIO_DT_CHR; break;
1541 #endif
1542 #ifdef DT_MPC
1543 case DT_MPC: ent->type = EIO_DT_MPC; break;
1544 #endif
1545 #ifdef DT_DIR
1546 case DT_DIR: ent->type = EIO_DT_DIR; break;
1547 #endif
1548 #ifdef DT_NAM
1549 case DT_NAM: ent->type = EIO_DT_NAM; break;
1550 #endif
1551 #ifdef DT_BLK
1552 case DT_BLK: ent->type = EIO_DT_BLK; break;
1553 #endif
1554 #ifdef DT_MPB
1555 case DT_MPB: ent->type = EIO_DT_MPB; break;
1556 #endif
1557 #ifdef DT_REG
1558 case DT_REG: ent->type = EIO_DT_REG; break;
1559 #endif
1560 #ifdef DT_NWK
1561 case DT_NWK: ent->type = EIO_DT_NWK; break;
1562 #endif
1563 #ifdef DT_CMP
1564 case DT_CMP: ent->type = EIO_DT_CMP; break;
1565 #endif
1566 #ifdef DT_LNK
1567 case DT_LNK: ent->type = EIO_DT_LNK; break;
1568 #endif
1569 #ifdef DT_SOCK
1570 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1571 #endif
1572 #ifdef DT_DOOR
1573 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1574 #endif
1575 #ifdef DT_WHT
1576 case DT_WHT: ent->type = EIO_DT_WHT; break;
1577 #endif
1578 }
1579
1580 ent->score = 7;
1581
1582 if (flags & EIO_READDIR_DIRS_FIRST)
1583 {
1584 if (ent->type == EIO_DT_UNKNOWN)
1676 { 1585 {
1677 if (dir->type == EIO_DT_DIR) 1586 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1678 ++dir; 1587 ent->score = 1;
1679 else if ((--oth)->type == EIO_DT_DIR) 1588 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1680 { 1589 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1681 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1682
1683 ++dir;
1684 }
1685 } 1590 }
1686 1591 else if (ent->type == EIO_DT_DIR)
1687 /* now sort the dirs only (dirs all have the same score) */ 1592 ent->score = 0;
1688 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1689 } 1593 }
1690
1691 break;
1692 }
1693
1694 /* now add the entry to our list(s) */
1695 name = entp->d_name;
1696
1697 /* skip . and .. entries */
1698 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1699 {
1700 int len = D_NAMLEN (entp) + 1;
1701
1702 while (ecb_expect_false (namesoffs + len > namesalloc))
1703 {
1704 namesalloc *= 2;
1705 X_LOCK (wrklock);
1706 req->ptr2 = names = realloc (names, namesalloc);
1707 X_UNLOCK (wrklock);
1708
1709 if (!names)
1710 break;
1711 } 1594 }
1712 1595
1713 memcpy (names + namesoffs, name, len);
1714
1715 if (dents)
1716 {
1717 struct eio_dirent *ent;
1718
1719 if (ecb_expect_false (dentoffs == dentalloc))
1720 {
1721 dentalloc *= 2;
1722 X_LOCK (wrklock);
1723 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1724 X_UNLOCK (wrklock);
1725
1726 if (!dents)
1727 break;
1728 }
1729
1730 ent = dents + dentoffs;
1731
1732 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1733 ent->namelen = len - 1;
1734 ent->inode = D_INO (entp);
1735
1736 inode_bits |= ent->inode;
1737
1738 switch (D_TYPE (entp))
1739 {
1740 default:
1741 ent->type = EIO_DT_UNKNOWN;
1742 flags |= EIO_READDIR_FOUND_UNKNOWN;
1743 break;
1744
1745 #ifdef DT_FIFO
1746 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1747 #endif
1748 #ifdef DT_CHR
1749 case DT_CHR: ent->type = EIO_DT_CHR; break;
1750 #endif
1751 #ifdef DT_MPC
1752 case DT_MPC: ent->type = EIO_DT_MPC; break;
1753 #endif
1754 #ifdef DT_DIR
1755 case DT_DIR: ent->type = EIO_DT_DIR; break;
1756 #endif
1757 #ifdef DT_NAM
1758 case DT_NAM: ent->type = EIO_DT_NAM; break;
1759 #endif
1760 #ifdef DT_BLK
1761 case DT_BLK: ent->type = EIO_DT_BLK; break;
1762 #endif
1763 #ifdef DT_MPB
1764 case DT_MPB: ent->type = EIO_DT_MPB; break;
1765 #endif
1766 #ifdef DT_REG
1767 case DT_REG: ent->type = EIO_DT_REG; break;
1768 #endif
1769 #ifdef DT_NWK
1770 case DT_NWK: ent->type = EIO_DT_NWK; break;
1771 #endif
1772 #ifdef DT_CMP
1773 case DT_CMP: ent->type = EIO_DT_CMP; break;
1774 #endif
1775 #ifdef DT_LNK
1776 case DT_LNK: ent->type = EIO_DT_LNK; break;
1777 #endif
1778 #ifdef DT_SOCK
1779 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1780 #endif
1781 #ifdef DT_DOOR
1782 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1783 #endif
1784 #ifdef DT_WHT
1785 case DT_WHT: ent->type = EIO_DT_WHT; break;
1786 #endif
1787 }
1788
1789 ent->score = 7;
1790
1791 if (flags & EIO_READDIR_DIRS_FIRST)
1792 {
1793 if (ent->type == EIO_DT_UNKNOWN)
1794 {
1795 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1796 ent->score = 1;
1797 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1798 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1799 }
1800 else if (ent->type == EIO_DT_DIR)
1801 ent->score = 0;
1802 }
1803 }
1804
1805 namesoffs += len; 1596 namesoffs += len;
1806 ++dentoffs; 1597 ++dentoffs;
1807 } 1598 }
1808 1599
1809 if (EIO_CANCELLED (req)) 1600 if (EIO_CANCELLED (req))
1810 { 1601 {
1811 errno = ECANCELED; 1602 errno = ECANCELED;
1812 break; 1603 break;
1813 } 1604 }
1605
1606#ifdef _WIN32
1607 if (!FindNextFile (dirp, &entp))
1608 {
1609 FindClose (dirp);
1610 dirp = 0;
1814 } 1611 }
1612#endif
1613 }
1815} 1614}
1615
1616/*****************************************************************************/
1617/* working directory stuff */
1618/* various deficiencies in the posix 2008 api force us to */
1619/* keep the absolute path in string form at all times */
1620/* fuck yeah. */
1621
1622#if !HAVE_AT
1623
1624/* a bit like realpath, but usually faster because it doesn'T have to return */
1625/* an absolute or canonical path */
1626static const char *
1627wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1628{
1629 if (!wd || *path == '/')
1630 return path;
1631
1632 if (path [0] == '.' && !path [1])
1633 return wd->str;
1634
1635 {
1636 int l1 = wd->len;
1637 int l2 = strlen (path);
1638
1639 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1640
1641 memcpy (res, wd->str, l1);
1642 res [l1] = '/';
1643 memcpy (res + l1 + 1, path, l2 + 1);
1644
1645 return res;
1646 }
1647}
1648
1649#endif
1650
1651static eio_wd
1652eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1653{
1654 int fd;
1655 eio_wd res;
1656 int len = eio__realpath (tmpbuf, wd, path);
1657
1658 if (len < 0)
1659 return EIO_INVALID_WD;
1660
1661#if HAVE_AT
1662 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1663
1664 if (fd < 0)
1665 return EIO_INVALID_WD;
1666#endif
1667
1668 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1669
1670#if HAVE_AT
1671 res->fd = fd;
1672#endif
1673
1674 res->len = len;
1675 memcpy (res->str, tmpbuf->ptr, len);
1676 res->str [len] = 0;
1677
1678 return res;
1679}
1680
1681eio_wd
1682eio_wd_open_sync (eio_wd wd, const char *path)
1683{
1684 struct etp_tmpbuf tmpbuf = { };
1685 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1686 free (tmpbuf.ptr);
1687
1688 return wd;
1689}
1690
1691void
1692eio_wd_close_sync (eio_wd wd)
1693{
1694 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1695 {
1696 #if HAVE_AT
1697 eio__close (wd->fd);
1698 #endif
1699 free (wd);
1700 }
1701}
1702
1703#if HAVE_AT
1704
1705static int
1706eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1707{
1708#if HAVE_RENAMEAT2
1709 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1710#else
1711 if (flags)
1712 return EIO_ENOSYS ();
1713
1714 return renameat (olddirfd, oldpath, newdirfd, newpath);
1715#endif
1716}
1717
1718/* they forgot these */
1719
1720static int
1721eio__truncateat (int dirfd, const char *path, off_t length)
1722{
1723 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1724 int res;
1725
1726 if (fd < 0)
1727 return fd;
1728
1729 res = ftruncate (fd, length);
1730 silent_close (fd);
1731 return res;
1732}
1733
1734static int
1735eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1736{
1737 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1738 int res;
1739
1740 if (fd < 0)
1741 return fd;
1742
1743 res = fstatvfs (fd, buf);
1744 silent_close (fd);
1745 return res;
1746}
1747
1748#endif
1816 1749
1817/*****************************************************************************/ 1750/*****************************************************************************/
1818 1751
1819#define ALLOC(len) \ 1752#define ALLOC(len) \
1820 if (!req->ptr2) \ 1753 if (!req->ptr2) \
1821 { \ 1754 { \
1822 X_LOCK (wrklock); \ 1755 X_LOCK (EIO_POOL->wrklock); \
1823 req->flags |= EIO_FLAG_PTR2_FREE; \ 1756 req->flags |= EIO_FLAG_PTR2_FREE; \
1824 X_UNLOCK (wrklock); \ 1757 X_UNLOCK (EIO_POOL->wrklock); \
1825 req->ptr2 = malloc (len); \ 1758 req->ptr2 = malloc (len); \
1826 if (!req->ptr2) \ 1759 if (!req->ptr2) \
1827 { \ 1760 { \
1828 errno = ENOMEM; \ 1761 errno = ENOMEM; \
1829 req->result = -1; \ 1762 req->result = -1; \
1830 break; \ 1763 goto alloc_fail; \
1831 } \ 1764 } \
1832 } 1765 }
1833 1766
1834X_THREAD_PROC (etp_proc) 1767/*****************************************************************************/
1835{
1836 ETP_REQ *req;
1837 struct timespec ts;
1838 etp_worker *self = (etp_worker *)thr_arg;
1839 int timeout;
1840 1768
1841 /* try to distribute timeouts somewhat evenly */ 1769static void
1842 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1770eio__slurp (int fd, eio_req *req)
1771{
1772 req->result = fd;
1843 1773
1844 for (;;) 1774 if (fd < 0)
1775 return;
1776
1777 if (req->offs < 0 || !req->size) /* do we need the size? */
1845 { 1778 {
1846 ts.tv_sec = 0; 1779 off_t size = lseek (fd, 0, SEEK_END);
1847 1780
1848 X_LOCK (reqlock);
1849
1850 for (;;)
1851 {
1852 self->req = req = reqq_shift (&req_queue);
1853
1854 if (req)
1855 break;
1856
1857 if (ts.tv_sec == 1) /* no request, but timeout detected, let's quit */
1858 {
1859 X_UNLOCK (reqlock);
1860 X_LOCK (wrklock);
1861 --started;
1862 X_UNLOCK (wrklock);
1863 goto quit;
1864 }
1865
1866 ++idle;
1867
1868 if (idle <= max_idle)
1869 /* we are allowed to idle, so do so without any timeout */
1870 X_COND_WAIT (reqwait, reqlock);
1871 else
1872 {
1873 /* initialise timeout once */
1874 if (!ts.tv_sec)
1875 ts.tv_sec = time (0) + idle_timeout;
1876
1877 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1878 ts.tv_sec = 1; /* assuming this is not a value computed above.,.. */
1879 }
1880
1881 --idle;
1882 }
1883
1884 --nready;
1885
1886 X_UNLOCK (reqlock);
1887
1888 if (req->type < 0) 1781 if (req->offs < 0)
1889 goto quit; 1782 req->offs += size;
1890 1783
1891 ETP_EXECUTE (self, req); 1784 if (!req->size)
1892 1785 req->size = size - req->offs;
1893 X_LOCK (reslock);
1894
1895 ++npending;
1896
1897 if (!reqq_push (&res_queue, req) && want_poll_cb)
1898 want_poll_cb ();
1899
1900 self->req = 0;
1901 etp_worker_clear (self);
1902
1903 X_UNLOCK (reslock);
1904 } 1786 }
1905 1787
1906quit: 1788 ALLOC (req->size);
1907 X_LOCK (wrklock); 1789 req->result = pread (fd, req->ptr2, req->size, req->offs);
1908 etp_worker_free (self);
1909 X_UNLOCK (wrklock);
1910 1790
1911 return 0; 1791 silent_close (fd);
1912}
1913 1792
1914/*****************************************************************************/ 1793alloc_fail:
1794 ;
1795}
1915 1796
1916int ecb_cold 1797int ecb_cold
1917eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1798eio_init (void (*want_poll)(void), void (*done_poll)(void))
1918{ 1799{
1919 return etp_init (want_poll, done_poll); 1800 eio_want_poll_cb = want_poll;
1801 eio_done_poll_cb = done_poll;
1802
1803 return etp_init (EIO_POOL, 0, 0, 0);
1920} 1804}
1921 1805
1922ecb_inline void 1806ecb_inline void
1923eio_api_destroy (eio_req *req) 1807eio_api_destroy (eio_req *req)
1924{ 1808{
1925 free (req); 1809 free (req);
1926} 1810}
1927 1811
1928#define REQ(rtype) \ 1812#define REQ(rtype) \
1929 eio_req *req; \ 1813 eio_req *req; \
1930 \ 1814 \
1931 req = (eio_req *)calloc (1, sizeof *req); \ 1815 req = (eio_req *)calloc (1, sizeof *req); \
1932 if (!req) \ 1816 if (!req) \
1933 return 0; \ 1817 return 0; \
1947 { \ 1831 { \
1948 eio_api_destroy (req); \ 1832 eio_api_destroy (req); \
1949 return 0; \ 1833 return 0; \
1950 } 1834 }
1951 1835
1836#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1837
1952static void 1838static void
1953eio_execute (etp_worker *self, eio_req *req) 1839eio_execute (etp_worker *self, eio_req *req)
1954{ 1840{
1841#if HAVE_AT
1842 int dirfd;
1843#else
1844 const char *path;
1845#endif
1846
1955 if (ecb_expect_false (EIO_CANCELLED (req))) 1847 if (ecb_expect_false (EIO_CANCELLED (req)))
1956 { 1848 {
1957 req->result = -1; 1849 req->result = -1;
1958 req->errorno = ECANCELED; 1850 req->errorno = ECANCELED;
1959 return; 1851 return;
1960 } 1852 }
1961 1853
1854 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1855 {
1856 req->result = -1;
1857 req->errorno = ENOENT;
1858 return;
1859 }
1860
1861 if (req->type >= EIO_OPEN)
1862 {
1863 #if HAVE_AT
1864 dirfd = WD2FD (req->wd);
1865 #else
1866 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1867 #endif
1868 }
1869
1962 switch (req->type) 1870 switch (req->type)
1963 { 1871 {
1872 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1873 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1874 break;
1875 case EIO_WD_CLOSE: req->result = 0;
1876 eio_wd_close_sync (req->wd); break;
1877
1878 case EIO_SEEK: eio__lseek (req); break;
1964 case EIO_READ: ALLOC (req->size); 1879 case EIO_READ: ALLOC (req->size);
1965 req->result = req->offs >= 0 1880 req->result = req->offs >= 0
1966 ? pread (req->int1, req->ptr2, req->size, req->offs) 1881 ? pread (req->int1, req->ptr2, req->size, req->offs)
1967 : read (req->int1, req->ptr2, req->size); break; 1882 : read (req->int1, req->ptr2, req->size); break;
1968 case EIO_WRITE: req->result = req->offs >= 0 1883 case EIO_WRITE: req->result = req->offs >= 0
1969 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1884 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1970 : write (req->int1, req->ptr2, req->size); break; 1885 : write (req->int1, req->ptr2, req->size); break;
1971 1886
1887 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1888 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1889
1972 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1890 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1973 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1891 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1892
1893#if HAVE_AT
1974 1894
1975 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1895 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1976 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1896 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1977 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1898 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1899 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1900 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1901 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1902 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1903 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1904
1905 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1906 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1907 req->result = req->wd && SINGLEDOT (req->ptr1)
1908 ? rmdir (req->wd->str)
1909 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1910 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1911 case EIO_RENAME: req->result = eio__renameat2 (
1912 dirfd,
1913 /* complications arise because "." cannot be renamed, so we might have to expand */
1914 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1915 WD2FD ((eio_wd)req->int3),
1916 req->ptr2,
1917 req->int2
1918 );
1919 break;
1920 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1921 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1922 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1923 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1924 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1925 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1926 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1927 if (req->result == EIO_PATH_MAX)
1928 {
1929 req->result = -1;
1930 errno = ENAMETOOLONG;
1931 }
1932 break;
1933 case EIO_UTIME:
1934 case EIO_FUTIME:
1935 {
1936 struct timespec ts[2];
1937 struct timespec *times;
1938
1939 if (req->nv1 != -1. || req->nv2 != -1.)
1940 {
1941 ts[0].tv_sec = req->nv1;
1942 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1943 ts[1].tv_sec = req->nv2;
1944 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1945
1946 times = ts;
1947 }
1948 else
1949 times = 0;
1950
1951 req->result = req->type == EIO_FUTIME
1952 ? futimens (req->int1, times)
1953 : utimensat (dirfd, req->ptr1, times, 0);
1954 }
1955 break;
1956
1957#else
1958
1959 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1960 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1961 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1978 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1962 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1964 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1965 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1966 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1967 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1968
1969 case EIO_UNLINK: req->result = unlink (path ); break;
1970 case EIO_RMDIR: req->result = rmdir (path ); break;
1971 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1972 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1973 case EIO_LINK: req->result = link (path , req->ptr2); break;
1974 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1975 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1976 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1977 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1978 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1979 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1980 if (req->result == EIO_PATH_MAX)
1981 {
1982 req->result = -1;
1983 errno = ENAMETOOLONG;
1984 }
1985 break;
1986
1987 case EIO_UTIME:
1988 case EIO_FUTIME:
1989 {
1990 struct timeval tv[2];
1991 struct timeval *times;
1992
1993 if (req->nv1 != -1. || req->nv2 != -1.)
1994 {
1995 tv[0].tv_sec = req->nv1;
1996 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1997 tv[1].tv_sec = req->nv2;
1998 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1999
2000 times = tv;
2001 }
2002 else
2003 times = 0;
2004
2005 req->result = req->type == EIO_FUTIME
2006 ? futimes (req->int1, times)
2007 : utimes (req->ptr1, times);
2008 }
2009 break;
2010
2011#endif
2012
2013 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2014 {
2015 ALLOC (req->result);
2016 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2017 }
2018 break;
2019
1979 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2020 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1980 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2021 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1981 2022
1982 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1983 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1984 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2023 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1985 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2024 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1986 2025
1987 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1988 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2026 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1989 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1990 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2027 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1991 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1992 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2028 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1993 2029
1994 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1995 case EIO_CLOSE: req->result = close (req->int1); break; 2030 case EIO_CLOSE: req->result = eio__close (req->int1); break;
1996 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2031 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1997 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1998 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1999 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
2000 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
2001 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
2002 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
2003 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
2004
2005 case EIO_REALPATH: eio__realpath (req, self); break;
2006
2007 case EIO_READLINK: ALLOC (PATH_MAX);
2008 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
2009
2010 case EIO_SYNC: req->result = 0; sync (); break; 2032 case EIO_SYNC: req->result = 0; sync (); break;
2011 case EIO_FSYNC: req->result = fsync (req->int1); break; 2033 case EIO_FSYNC: req->result = fsync (req->int1); break;
2012 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2034 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2035 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2036 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2013 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2037 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
2014 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2038 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
2015 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2039 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
2016 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2040 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
2017 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
2018 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break; 2041 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
2019 2042
2020 case EIO_READDIR: eio__scandir (req, self); break; 2043 case EIO_READDIR: eio__scandir (req, self); break;
2021 2044
2022 case EIO_BUSY: 2045 case EIO_BUSY:
2032 req->result = select (0, 0, 0, 0, &tv); 2055 req->result = select (0, 0, 0, 0, &tv);
2033 } 2056 }
2034#endif 2057#endif
2035 break; 2058 break;
2036 2059
2037 case EIO_UTIME: 2060#if 0
2038 case EIO_FUTIME:
2039 {
2040 struct timeval tv[2];
2041 struct timeval *times;
2042
2043 if (req->nv1 != -1. || req->nv2 != -1.)
2044 {
2045 tv[0].tv_sec = req->nv1;
2046 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
2047 tv[1].tv_sec = req->nv2;
2048 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
2049
2050 times = tv;
2051 }
2052 else
2053 times = 0;
2054
2055 req->result = req->type == EIO_FUTIME
2056 ? futimes (req->int1, times)
2057 : utimes (req->ptr1, times);
2058 }
2059 break;
2060
2061 case EIO_GROUP: 2061 case EIO_GROUP:
2062 abort (); /* handled in eio_request */ 2062 abort (); /* handled in eio_request */
2063#endif
2063 2064
2064 case EIO_NOP: 2065 case EIO_NOP:
2065 req->result = 0; 2066 req->result = 0;
2066 break; 2067 break;
2067 2068
2068 case EIO_CUSTOM: 2069 case EIO_CUSTOM:
2069 req->feed (req); 2070 req->feed (req);
2070 break; 2071 break;
2071 2072
2072 default: 2073 default:
2073 errno = ENOSYS;
2074 req->result = -1; 2074 req->result = EIO_ENOSYS ();
2075 break; 2075 break;
2076 } 2076 }
2077 2077
2078alloc_fail:
2078 req->errorno = errno; 2079 req->errorno = errno;
2079} 2080}
2080 2081
2081#ifndef EIO_NO_WRAPPERS 2082#ifndef EIO_NO_WRAPPERS
2082 2083
2084eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2085{
2086 REQ (EIO_WD_OPEN); PATH; SEND;
2087}
2088
2089eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2090{
2091 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2092}
2093
2083eio_req *eio_nop (int pri, eio_cb cb, void *data) 2094eio_req *eio_nop (int pri, eio_cb cb, void *data)
2084{ 2095{
2085 REQ (EIO_NOP); SEND; 2096 REQ (EIO_NOP); SEND;
2086} 2097}
2087 2098
2103eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2114eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2104{ 2115{
2105 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2116 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2106} 2117}
2107 2118
2119eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2120{
2121 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2122}
2123
2124eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2125{
2126 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2127}
2128
2129eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2130{
2131 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2132}
2133
2108eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2134eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2109{ 2135{
2110 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2136 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2111} 2137}
2112 2138
2118eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2144eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2119{ 2145{
2120 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2146 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2121} 2147}
2122 2148
2123eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2124{
2125 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2126}
2127
2128eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data) 2149eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2129{ 2150{
2130 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND; 2151 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2131} 2152}
2132 2153
2133eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2134{
2135 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2136}
2137
2138eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2154eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
2139{ 2155{
2140 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2156 REQ (EIO_CLOSE); req->int1 = fd; SEND;
2141} 2157}
2142 2158
2143eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2159eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
2144{ 2160{
2145 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2161 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
2146} 2162}
2147 2163
2164eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2165{
2166 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2167}
2168
2148eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2169eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2149{ 2170{
2150 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2171 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2151} 2172}
2152 2173
2153eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2174eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2154{ 2175{
2155 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2176 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2156} 2177}
2157 2178
2179eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2180{
2181 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2182}
2183
2184eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2185{
2186 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2187}
2188
2158eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2189eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
2159{ 2190{
2160 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2191 REQ (EIO_FSTAT); req->int1 = fd; SEND;
2161} 2192}
2162 2193
2178eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2209eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
2179{ 2210{
2180 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2211 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
2181} 2212}
2182 2213
2183eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2214eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2184{ 2215{
2185 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2216 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2186} 2217}
2187 2218
2188eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2219eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
2208eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2239eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
2209{ 2240{
2210 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2241 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
2211} 2242}
2212 2243
2213eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2244eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
2214{ 2245{
2215 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2246 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
2216} 2247}
2217 2248
2218eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2249eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
2305eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2336eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2306{ 2337{
2307 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2338 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2308} 2339}
2309 2340
2341eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2342{
2343 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2344}
2345
2310eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) 2346eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2311{ 2347{
2312 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2348 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2313} 2349}
2314 2350
2348void 2384void
2349eio_grp_add (eio_req *grp, eio_req *req) 2385eio_grp_add (eio_req *grp, eio_req *req)
2350{ 2386{
2351 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2387 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2352 2388
2353 grp->flags |= EIO_FLAG_GROUPADD; 2389 grp->flags |= ETP_FLAG_GROUPADD;
2354 2390
2355 ++grp->size; 2391 ++grp->size;
2356 req->grp = grp; 2392 req->grp = grp;
2357 2393
2358 req->grp_prev = 0; 2394 req->grp_prev = 0;
2365} 2401}
2366 2402
2367/*****************************************************************************/ 2403/*****************************************************************************/
2368/* misc garbage */ 2404/* misc garbage */
2369 2405
2370ssize_t 2406eio_ssize_t
2371eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2407eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2372{ 2408{
2373 etp_worker wrk;
2374 ssize_t ret;
2375
2376 wrk.dbuf = 0;
2377
2378 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2409 return eio__sendfile (ofd, ifd, offset, count);
2379
2380 if (wrk.dbuf)
2381 free (wrk.dbuf);
2382
2383 return ret;
2384} 2410}
2385 2411

Diff Legend

Removed lines
+ Added lines
< Changed lines
> Changed lines