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

Comparing libeio/eio.c (file contents):
Revision 1.76 by root, Tue Jul 5 09:24:12 2011 UTC vs.
Revision 1.150 by root, Fri Jun 15 02:57:36 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>
57#include <sys/types.h> 52#include <sys/types.h>
58#include <sys/stat.h> 53#include <sys/stat.h>
59#include <sys/statvfs.h>
60#include <limits.h> 54#include <limits.h>
61#include <fcntl.h> 55#include <fcntl.h>
62#include <assert.h> 56#include <assert.h>
63 57
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 form 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
64#ifndef ECANCELED
65# define ECANCELED EDOM
66#endif
67#ifndef ELOOP
68# define ELOOP EDOM
69#endif
70
71#if !defined(ENOTSOCK) && defined(WSAENOTSOCK)
72# define ENOTSOCK WSAENOTSOCK
73#endif
74
75static void eio_destroy (eio_req *req);
76
70#ifndef EIO_FINISH 77#ifndef EIO_FINISH
71# 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
72#endif 79#endif
73 80
74#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
… …
77 84
78#ifndef EIO_FEED 85#ifndef EIO_FEED
79# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0) 86# define EIO_FEED(req) do { if ((req)->feed ) (req)->feed (req); } while (0)
80#endif 87#endif
81 88
89#ifndef EIO_FD_TO_WIN32_HANDLE
90# define EIO_FD_TO_WIN32_HANDLE(fd) _get_osfhandle (fd)
91#endif
92#ifndef EIO_WIN32_HANDLE_TO_FD
93# define EIO_WIN32_HANDLE_TO_FD(handle) _open_osfhandle (handle, 0)
94#endif
95
96#define EIO_ERRNO(errval,retval) ((errno = errval), retval)
97
98#define EIO_ENOSYS() EIO_ERRNO (ENOSYS, -1)
99
82#ifdef _WIN32 100#ifdef _WIN32
83 101
84 /*doh*/ 102 #undef PAGESIZE
103 #define PAGESIZE 4096 /* GetSystemInfo? */
104
105 /* TODO: look at how perl does stat (non-sloppy), unlink (ro-files), utime, link */
106
107 #ifdef EIO_STRUCT_STATI64
108 /* look at perl's non-sloppy stat */
109 #define stat(path,buf) _stati64 (path,buf)
110 #define fstat(fd,buf) _fstati64 (fd,buf)
111 #endif
112 #define lstat(path,buf) stat (path,buf)
113 #define fsync(fd) (FlushFileBuffers ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd)) ? 0 : EIO_ERRNO (EBADF, -1))
114 #define mkdir(path,mode) _mkdir (path)
115 #define link(old,neu) (CreateHardLink (neu, old, 0) ? 0 : EIO_ERRNO (ENOENT, -1))
116
117 #define chmod(path,mode) _chmod (path, mode)
118 #define dup(fd) _dup (fd)
119 #define dup2(fd1,fd2) _dup2 (fd1, fd2)
120 #define pipe(fds) _pipe (fds, 4096, O_BINARY)
121
122 #define fcntl(fd,cmd,arg) EIO_ENOSYS ()
123 #define ioctl(fd,cmd,arg) EIO_ENOSYS ()
124 #define fchmod(fd,mode) EIO_ENOSYS ()
125 #define chown(path,uid,gid) EIO_ENOSYS ()
126 #define fchown(fd,uid,gid) EIO_ENOSYS ()
127 #define truncate(path,offs) EIO_ENOSYS () /* far-miss: SetEndOfFile */
128 #define ftruncate(fd,offs) EIO_ENOSYS () /* near-miss: SetEndOfFile */
129 #define mknod(path,mode,dev) EIO_ENOSYS ()
130 #define sync() EIO_ENOSYS ()
131 #define readlink(path,buf,s) EIO_ENOSYS ()
132 #define statvfs(path,buf) EIO_ENOSYS ()
133 #define fstatvfs(fd,buf) EIO_ENOSYS ()
134
135 #define pread(fd,buf,count,offset) eio__pread (fd, buf, count, offset)
136 #define pwrite(fd,buf,count,offset) eio__pwrite (fd, buf, count, offset)
137
138 #if __GNUC__
139 typedef long long eio_off_t; /* signed for compatibility to msvc */
140 #else
141 typedef __int64 eio_off_t; /* unsigned not supported by msvc */
142 #endif
143
144 static eio_ssize_t
145 eio__pread (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
146 {
147 OVERLAPPED o = { 0 };
148 DWORD got;
149
150 o.Offset = offset;
151 o.OffsetHigh = offset >> 32;
152
153 return ReadFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
154 ? got : -1;
155 }
156
157 static eio_ssize_t
158 eio__pwrite (int fd, void *buf, eio_ssize_t count, eio_off_t offset)
159 {
160 OVERLAPPED o = { 0 };
161 DWORD got;
162
163 o.Offset = offset;
164 o.OffsetHigh = offset >> 32;
165
166 return WriteFile ((HANDLE)EIO_FD_TO_WIN32_HANDLE (fd), buf, count, &got, &o)
167 ? got : -1;
168 }
169
170 /* rename() uses MoveFile, which fails to overwrite */
171 #define rename(old,neu) eio__rename (old, neu)
172
173 static int
174 eio__rename (const char *old, const char *neu)
175 {
176 if (MoveFileEx (old, neu, MOVEFILE_REPLACE_EXISTING))
177 return 0;
178
179 /* should steal _dosmaperr */
180 switch (GetLastError ())
181 {
182 case ERROR_FILE_NOT_FOUND:
183 case ERROR_PATH_NOT_FOUND:
184 case ERROR_INVALID_DRIVE:
185 case ERROR_NO_MORE_FILES:
186 case ERROR_BAD_NETPATH:
187 case ERROR_BAD_NET_NAME:
188 case ERROR_BAD_PATHNAME:
189 case ERROR_FILENAME_EXCED_RANGE:
190 errno = ENOENT;
191 break;
192
193 default:
194 errno = EACCES;
195 break;
196 }
197
198 return -1;
199 }
200
201 /* we could even stat and see if it exists */
202 static int
203 symlink (const char *old, const char *neu)
204 {
205 #if WINVER >= 0x0600
206 int flags;
207
208 /* This tries out all combinations of SYMBOLIC_LINK_FLAG_DIRECTORY
209 * and SYMBOLIC_LINK_FLAG_ALLOW_UNPRIVILEGED_CREATE,
210 * with directory first.
211 */
212 for (flags = 3; flags >= 0; --flags)
213 if (CreateSymbolicLink (neu, old, flags))
214 return 0;
215 #endif
216
217 return EIO_ERRNO (ENOENT, -1);
218 }
219
220 /* POSIX API only, causing trouble for win32 apps */
221 #define CreateHardLink(neu,old,flags) 0 /* not really creating hardlink, still using relative paths? */
222 #define CreateSymbolicLink(neu,old,flags) 0 /* vista+ only */
223
224 struct statvfs
225 {
226 int dummy;
227 };
228
229 #define DT_DIR EIO_DT_DIR
230 #define DT_REG EIO_DT_REG
231 #define D_NAME(entp) entp.cFileName
232 #define D_TYPE(entp) (entp.dwFileAttributes & FILE_ATTRIBUTE_DIRECTORY ? DT_DIR : DT_REG)
233
85#else 234#else
86 235
236 #include <sys/ioctl.h>
87# include <sys/time.h> 237 #include <sys/time.h>
88# include <sys/select.h> 238 #include <sys/select.h>
89# include <unistd.h> 239 #include <unistd.h>
240 #include <signal.h>
241 #include <dirent.h>
242
243 #ifdef ANDROID
244 #include <sys/vfs.h>
245 #define statvfs statfs
246 #define fstatvfs fstatfs
247 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
248 #else
249 #include <sys/statvfs.h>
250 #endif
251
252 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
253 #include <sys/mman.h>
254 #endif
255
256 #define D_NAME(entp) entp->d_name
257
258 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
259 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
260 #define _DIRENT_HAVE_D_TYPE /* sigh */
261 #define D_INO(de) (de)->d_fileno
262 #define D_NAMLEN(de) (de)->d_namlen
263 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
264 #define D_INO(de) (de)->d_ino
265 #endif
266
267 #ifdef _D_EXACT_NAMLEN
268 #undef D_NAMLEN
269 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
270 #endif
271
272 #ifdef _DIRENT_HAVE_D_TYPE
273 #define D_TYPE(de) (de)->d_type
274 #endif
275
276 #ifndef EIO_STRUCT_DIRENT
277 #define EIO_STRUCT_DIRENT struct dirent
278 #endif
279
280#endif
281
282#if HAVE_UTIMES
90# include <utime.h> 283# 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 284#endif
106 285
107#ifdef _D_EXACT_NAMLEN 286#if HAVE_SYS_SYSCALL_H
108# undef D_NAMLEN 287# 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
115
116# ifndef EIO_STRUCT_DIRENT
117# define EIO_STRUCT_DIRENT struct dirent
118# endif
119
120#endif 288#endif
121 289
122#if HAVE_SENDFILE 290#if HAVE_SENDFILE
123# if __linux 291# if __linux
124# include <sys/sendfile.h> 292# include <sys/sendfile.h>
… …
132# else 300# else
133# error sendfile support requested but not available 301# error sendfile support requested but not available
134# endif 302# endif
135#endif 303#endif
136 304
305#if HAVE_RENAMEAT2
306# include <sys/syscall.h>
307# include <linux/fs.h>
308#endif
309
137#ifndef D_TYPE 310#ifndef D_TYPE
138# define D_TYPE(de) 0 311# define D_TYPE(de) 0
139#endif 312#endif
140#ifndef D_INO 313#ifndef D_INO
141# define D_INO(de) 0 314# define D_INO(de) 0
142#endif 315#endif
143#ifndef D_NAMLEN 316#ifndef D_NAMLEN
144# define D_NAMLEN(de) strlen ((de)->d_name) 317# define D_NAMLEN(entp) strlen (D_NAME (entp))
145#endif 318#endif
146 319
147/* used for struct dirent, AIX doesn't provide it */ 320/* used for struct dirent, AIX doesn't provide it */
148#ifndef NAME_MAX 321#ifndef NAME_MAX
149# define NAME_MAX 4096 322# define NAME_MAX 4096
150#endif 323#endif
151 324
152/* used for readlink etc. */ 325/* used for readlink etc. */
153#ifndef PATH_MAX 326#ifndef PATH_MAX
154# define PATH_MAX 4096 327# define PATH_MAX 0
155#endif 328#endif
329
330#ifndef O_CLOEXEC
331 #define O_CLOEXEC 0
332#endif
333
334#ifndef EIO_PATH_MIN
335# define EIO_PATH_MIN 8160
336#endif
337
338#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
156 339
157/* buffer size for various temporary buffers */ 340/* buffer size for various temporary buffers */
158#define EIO_BUFSIZE 65536 341#define EIO_BUFSIZE 65536
159 342
160#define dBUF \ 343#define dBUF \
161 char *eio_buf; \
162 ETP_WORKER_LOCK (self); \
163 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 344 char *eio_buf = malloc (EIO_BUFSIZE); \
164 ETP_WORKER_UNLOCK (self); \
165 errno = ENOMEM; \ 345 errno = ENOMEM; \
166 if (!eio_buf) \ 346 if (!eio_buf) \
167 return -1; 347 return -1
168 348
169#define EIO_TICKS ((1000000 + 1023) >> 10) 349#define FUBd \
350 free (eio_buf)
351
352/*****************************************************************************/
353
354struct etp_tmpbuf;
355
356#if _POSIX_VERSION >= 200809L
357 #define HAVE_AT 1
358 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
359 #ifndef O_SEARCH
360 #define O_SEARCH O_RDONLY
361 #endif
362#else
363 #define HAVE_AT 0
364 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
365#endif
366
367struct eio_pwd
368{
369#if HAVE_AT
370 int fd;
371#endif
372 int len;
373 char str[1]; /* actually, a 0-terminated canonical path */
374};
375
376/*****************************************************************************/
170 377
171#define ETP_PRI_MIN EIO_PRI_MIN 378#define ETP_PRI_MIN EIO_PRI_MIN
172#define ETP_PRI_MAX EIO_PRI_MAX 379#define ETP_PRI_MAX EIO_PRI_MAX
173 380
381#define ETP_TYPE_QUIT -1
382#define ETP_TYPE_GROUP EIO_GROUP
383
384static void eio_nop_callback (void) { }
385static void (*eio_want_poll_cb)(void) = eio_nop_callback;
386static void (*eio_done_poll_cb)(void) = eio_nop_callback;
387
388#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
389#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
390
174struct etp_worker; 391struct etp_worker;
175
176#define ETP_REQ eio_req 392#define ETP_REQ eio_req
177#define ETP_DESTROY(req) eio_destroy (req) 393#define ETP_DESTROY(req) eio_destroy (req)
178static int eio_finish (eio_req *req); 394static int eio_finish (eio_req *req);
179#define ETP_FINISH(req) eio_finish (req) 395#define ETP_FINISH(req) eio_finish (req)
180static void eio_execute (struct etp_worker *self, eio_req *req); 396static void eio_execute (struct etp_worker *self, eio_req *req);
181#define ETP_EXECUTE(wrk,req) eio_execute (wrk,req) 397#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
182 398
183#define ETP_WORKER_CLEAR(req) \ 399#include "etp.c"
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 }
195 400
196#define ETP_WORKER_COMMON \ 401static struct etp_pool eio_pool;
197 void *dbuf; \ 402#define EIO_POOL (&eio_pool)
198 DIR *dirp;
199
200/*****************************************************************************/
201
202#define ETP_NUM_PRI (ETP_PRI_MAX - ETP_PRI_MIN + 1)
203
204/* calculate time difference in ~1/EIO_TICKS of a second */
205ecb_inline int
206tvdiff (struct timeval *tv1, struct timeval *tv2)
207{
208 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS
209 + ((tv2->tv_usec - tv1->tv_usec) >> 10);
210}
211
212static unsigned int started, idle, wanted = 4;
213
214static void (*want_poll_cb) (void);
215static void (*done_poll_cb) (void);
216
217static unsigned int max_poll_time; /* reslock */
218static unsigned int max_poll_reqs; /* reslock */
219
220static volatile unsigned int nreqs; /* reqlock */
221static volatile unsigned int nready; /* reqlock */
222static volatile unsigned int npending; /* reqlock */
223static volatile 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 */
225
226static xmutex_t wrklock;
227static xmutex_t reslock;
228static xmutex_t reqlock;
229static xcond_t reqwait;
230
231#if !HAVE_PREADWRITE
232/*
233 * make our pread/pwrite emulation safe against themselves, but not against
234 * normal read/write by using a mutex. slows down execution a lot,
235 * but that's your problem, not mine.
236 */
237static xmutex_t preadwritelock = X_MUTEX_INIT;
238#endif
239
240typedef struct etp_worker
241{
242 /* locked by wrklock */
243 struct etp_worker *prev, *next;
244
245 xthread_t tid;
246
247 /* locked by reslock, reqlock or wrklock */
248 ETP_REQ *req; /* currently processed request */
249
250 ETP_WORKER_COMMON
251} etp_worker;
252
253static etp_worker wrk_first = { &wrk_first, &wrk_first, 0 }; /* NOT etp */
254
255#define ETP_WORKER_LOCK(wrk) X_LOCK (wrklock)
256#define ETP_WORKER_UNLOCK(wrk) X_UNLOCK (wrklock)
257
258/* worker threads management */
259
260static void ecb_cold
261etp_worker_clear (etp_worker *wrk)
262{
263 ETP_WORKER_CLEAR (wrk);
264}
265
266static void ecb_cold
267etp_worker_free (etp_worker *wrk)
268{
269 wrk->next->prev = wrk->prev;
270 wrk->prev->next = wrk->next;
271
272 free (wrk);
273}
274
275static unsigned int
276etp_nreqs (void)
277{
278 int retval;
279 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
280 retval = nreqs;
281 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
282 return retval;
283}
284
285static unsigned int
286etp_nready (void)
287{
288 unsigned int retval;
289
290 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
291 retval = nready;
292 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
293
294 return retval;
295}
296
297static unsigned int
298etp_npending (void)
299{
300 unsigned int retval;
301
302 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
303 retval = npending;
304 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
305
306 return retval;
307}
308
309static unsigned int
310etp_nthreads (void)
311{
312 unsigned int retval;
313
314 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
315 retval = started;
316 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
317
318 return retval;
319}
320
321/*
322 * a somewhat faster data structure might be nice, but
323 * with 8 priorities this actually needs <20 insns
324 * per shift, the most expensive operation.
325 */
326typedef struct {
327 ETP_REQ *qs[ETP_NUM_PRI], *qe[ETP_NUM_PRI]; /* qstart, qend */
328 int size;
329} etp_reqq;
330
331static etp_reqq req_queue;
332static etp_reqq res_queue;
333
334static int ecb_noinline
335reqq_push (etp_reqq *q, ETP_REQ *req)
336{
337 int pri = req->pri;
338 req->next = 0;
339
340 if (q->qe[pri])
341 {
342 q->qe[pri]->next = req;
343 q->qe[pri] = req;
344 }
345 else
346 q->qe[pri] = q->qs[pri] = req;
347
348 return q->size++;
349}
350
351static ETP_REQ * ecb_noinline
352reqq_shift (etp_reqq *q)
353{
354 int pri;
355
356 if (!q->size)
357 return 0;
358
359 --q->size;
360
361 for (pri = ETP_NUM_PRI; pri--; )
362 {
363 eio_req *req = q->qs[pri];
364
365 if (req)
366 {
367 if (!(q->qs[pri] = (eio_req *)req->next))
368 q->qe[pri] = 0;
369
370 return req;
371 }
372 }
373
374 abort ();
375}
376
377static void ecb_cold
378etp_thread_init (void)
379{
380 X_MUTEX_CREATE (wrklock);
381 X_MUTEX_CREATE (reslock);
382 X_MUTEX_CREATE (reqlock);
383 X_COND_CREATE (reqwait);
384}
385
386static void ecb_cold
387etp_atfork_prepare (void)
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
397static void ecb_cold
398etp_atfork_parent (void)
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
430 started = 0;
431 idle = 0;
432 nreqs = 0;
433 nready = 0;
434 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
453 want_poll_cb = want_poll;
454 done_poll_cb = done_poll;
455
456 return 0;
457}
458
459X_THREAD_PROC (etp_proc);
460
461static void ecb_cold
462etp_start_thread (void)
463{
464 etp_worker *wrk = calloc (1, sizeof (etp_worker));
465
466 /*TODO*/
467 assert (("unable to allocate worker thread data", wrk));
468
469 X_LOCK (wrklock);
470
471 if (thread_create (&wrk->tid, etp_proc, (void *)wrk))
472 {
473 wrk->prev = &wrk_first;
474 wrk->next = wrk_first.next;
475 wrk_first.next->prev = wrk;
476 wrk_first.next = wrk;
477 ++started;
478 }
479 else
480 free (wrk);
481
482 X_UNLOCK (wrklock);
483}
484
485static void
486etp_maybe_start_thread (void)
487{
488 if (ecb_expect_true (etp_nthreads () >= wanted))
489 return;
490
491 /* todo: maybe use idle here, but might be less exact */
492 if (ecb_expect_true (0 <= (int)etp_nthreads () + (int)etp_npending () - (int)etp_nreqs ()))
493 return;
494
495 etp_start_thread ();
496}
497
498static void ecb_cold
499etp_end_thread (void)
500{
501 eio_req *req = calloc (1, sizeof (eio_req));
502
503 req->type = -1;
504 req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
505
506 X_LOCK (reqlock);
507 reqq_push (&req_queue, req);
508 X_COND_SIGNAL (reqwait);
509 X_UNLOCK (reqlock);
510
511 X_LOCK (wrklock);
512 --started;
513 X_UNLOCK (wrklock);
514}
515
516static int
517etp_poll (void)
518{
519 unsigned int maxreqs;
520 unsigned int maxtime;
521 struct timeval tv_start, tv_now;
522
523 X_LOCK (reslock);
524 maxreqs = max_poll_reqs;
525 maxtime = max_poll_time;
526 X_UNLOCK (reslock);
527
528 if (maxtime)
529 gettimeofday (&tv_start, 0);
530
531 for (;;)
532 {
533 ETP_REQ *req;
534
535 etp_maybe_start_thread ();
536
537 X_LOCK (reslock);
538 req = reqq_shift (&res_queue);
539
540 if (req)
541 {
542 --npending;
543
544 if (!res_queue.size && done_poll_cb)
545 done_poll_cb ();
546 }
547
548 X_UNLOCK (reslock);
549
550 if (!req)
551 return 0;
552
553 X_LOCK (reqlock);
554 --nreqs;
555 X_UNLOCK (reqlock);
556
557 if (ecb_expect_false (req->type == EIO_GROUP && req->size))
558 {
559 req->int1 = 1; /* mark request as delayed */
560 continue;
561 }
562 else
563 {
564 int res = ETP_FINISH (req);
565 if (ecb_expect_false (res))
566 return res;
567 }
568
569 if (ecb_expect_false (maxreqs && !--maxreqs))
570 break;
571
572 if (maxtime)
573 {
574 gettimeofday (&tv_now, 0);
575
576 if (tvdiff (&tv_start, &tv_now) >= maxtime)
577 break;
578 }
579 }
580
581 errno = EAGAIN;
582 return -1;
583}
584
585static void
586etp_cancel (ETP_REQ *req)
587{
588 X_LOCK (wrklock);
589 req->flags |= EIO_FLAG_CANCELLED;
590 X_UNLOCK (wrklock);
591
592 eio_grp_cancel (req);
593}
594
595static void
596etp_submit (ETP_REQ *req)
597{
598 req->pri -= ETP_PRI_MIN;
599
600 if (ecb_expect_false (req->pri < ETP_PRI_MIN - ETP_PRI_MIN)) req->pri = ETP_PRI_MIN - ETP_PRI_MIN;
601 if (ecb_expect_false (req->pri > ETP_PRI_MAX - ETP_PRI_MIN)) req->pri = ETP_PRI_MAX - ETP_PRI_MIN;
602
603 if (ecb_expect_false (req->type == EIO_GROUP))
604 {
605 /* I hope this is worth it :/ */
606 X_LOCK (reqlock);
607 ++nreqs;
608 X_UNLOCK (reqlock);
609
610 X_LOCK (reslock);
611
612 ++npending;
613
614 if (!reqq_push (&res_queue, req) && want_poll_cb)
615 want_poll_cb ();
616
617 X_UNLOCK (reslock);
618 }
619 else
620 {
621 X_LOCK (reqlock);
622 ++nreqs;
623 ++nready;
624 reqq_push (&req_queue, req);
625 X_COND_SIGNAL (reqwait);
626 X_UNLOCK (reqlock);
627
628 etp_maybe_start_thread ();
629 }
630}
631
632static void ecb_cold
633etp_set_max_poll_time (double nseconds)
634{
635 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
636 max_poll_time = nseconds * EIO_TICKS;
637 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
638}
639
640static void ecb_cold
641etp_set_max_poll_reqs (unsigned int maxreqs)
642{
643 if (WORDACCESS_UNSAFE) X_LOCK (reslock);
644 max_poll_reqs = maxreqs;
645 if (WORDACCESS_UNSAFE) X_UNLOCK (reslock);
646}
647
648static void ecb_cold
649etp_set_max_idle (unsigned int nthreads)
650{
651 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
652 max_idle = nthreads;
653 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
654}
655
656static void ecb_cold
657etp_set_idle_timeout (unsigned int seconds)
658{
659 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
660 idle_timeout = seconds;
661 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
662}
663
664static void ecb_cold
665etp_set_min_parallel (unsigned int nthreads)
666{
667 if (wanted < nthreads)
668 wanted = nthreads;
669}
670
671static void ecb_cold
672etp_set_max_parallel (unsigned int nthreads)
673{
674 if (wanted > nthreads)
675 wanted = nthreads;
676
677 while (started > wanted)
678 etp_end_thread ();
679}
680 403
681/*****************************************************************************/ 404/*****************************************************************************/
682 405
683static void 406static void
684grp_try_feed (eio_req *grp) 407grp_try_feed (eio_req *grp)
685{ 408{
686 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 409 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
687 { 410 {
688 grp->flags &= ~EIO_FLAG_GROUPADD; 411 grp->flags &= ~ETP_FLAG_GROUPADD;
689 412
690 EIO_FEED (grp); 413 EIO_FEED (grp);
691 414
692 /* stop if no progress has been made */ 415 /* stop if no progress has been made */
693 if (!(grp->flags & EIO_FLAG_GROUPADD)) 416 if (!(grp->flags & ETP_FLAG_GROUPADD))
694 { 417 {
695 grp->feed = 0; 418 grp->feed = 0;
696 break; 419 break;
697 } 420 }
698 } 421 }
… …
705 428
706 /* call feeder, if applicable */ 429 /* call feeder, if applicable */
707 grp_try_feed (grp); 430 grp_try_feed (grp);
708 431
709 /* finish, if done */ 432 /* finish, if done */
710 if (!grp->size && grp->int1) 433 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
711 return eio_finish (grp); 434 return eio_finish (grp);
712 else 435 else
713 return 0; 436 return 0;
714} 437}
715 438
716void 439static void
717eio_destroy (eio_req *req) 440eio_destroy (eio_req *req)
718{ 441{
719 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 442 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
720 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 443 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
721 444
… …
739 if (grp->grp_first == req) 462 if (grp->grp_first == req)
740 grp->grp_first = req->grp_next; 463 grp->grp_first = req->grp_next;
741 464
742 res2 = grp_dec (grp); 465 res2 = grp_dec (grp);
743 466
744 if (!res && res2) 467 if (!res)
745 res = res2; 468 res = res2;
746 } 469 }
747 470
748 eio_destroy (req); 471 eio_destroy (req);
749 472
… …
751} 474}
752 475
753void 476void
754eio_grp_cancel (eio_req *grp) 477eio_grp_cancel (eio_req *grp)
755{ 478{
756 for (grp = grp->grp_first; grp; grp = grp->grp_next) 479 etp_grp_cancel (EIO_POOL, grp);
757 eio_cancel (grp);
758} 480}
759 481
760void 482void
761eio_cancel (eio_req *req) 483eio_cancel (eio_req *req)
762{ 484{
763 etp_cancel (req); 485 etp_cancel (EIO_POOL, req);
764} 486}
765 487
766void 488void
767eio_submit (eio_req *req) 489eio_submit (eio_req *req)
768{ 490{
769 etp_submit (req); 491 etp_submit (EIO_POOL, req);
770} 492}
771 493
772unsigned int 494unsigned int
773eio_nreqs (void) 495eio_nreqs (void)
774{ 496{
775 return etp_nreqs (); 497 return etp_nreqs (EIO_POOL);
776} 498}
777 499
778unsigned int 500unsigned int
779eio_nready (void) 501eio_nready (void)
780{ 502{
781 return etp_nready (); 503 return etp_nready (EIO_POOL);
782} 504}
783 505
784unsigned int 506unsigned int
785eio_npending (void) 507eio_npending (void)
786{ 508{
787 return etp_npending (); 509 return etp_npending (EIO_POOL);
788} 510}
789 511
790unsigned int ecb_cold 512unsigned int ecb_cold
791eio_nthreads (void) 513eio_nthreads (void)
792{ 514{
793 return etp_nthreads (); 515 return etp_nthreads (EIO_POOL);
794} 516}
795 517
796void ecb_cold 518void ecb_cold
797eio_set_max_poll_time (double nseconds) 519eio_set_max_poll_time (double nseconds)
798{ 520{
799 etp_set_max_poll_time (nseconds); 521 etp_set_max_poll_time (EIO_POOL, nseconds);
800} 522}
801 523
802void ecb_cold 524void ecb_cold
803eio_set_max_poll_reqs (unsigned int maxreqs) 525eio_set_max_poll_reqs (unsigned int maxreqs)
804{ 526{
805 etp_set_max_poll_reqs (maxreqs); 527 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
806} 528}
807 529
808void ecb_cold 530void ecb_cold
809eio_set_max_idle (unsigned int nthreads) 531eio_set_max_idle (unsigned int nthreads)
810{ 532{
811 etp_set_max_idle (nthreads); 533 etp_set_max_idle (EIO_POOL, nthreads);
812} 534}
813 535
814void ecb_cold 536void ecb_cold
815eio_set_idle_timeout (unsigned int seconds) 537eio_set_idle_timeout (unsigned int seconds)
816{ 538{
817 etp_set_idle_timeout (seconds); 539 etp_set_idle_timeout (EIO_POOL, seconds);
818} 540}
819 541
820void ecb_cold 542void ecb_cold
821eio_set_min_parallel (unsigned int nthreads) 543eio_set_min_parallel (unsigned int nthreads)
822{ 544{
823 etp_set_min_parallel (nthreads); 545 etp_set_min_parallel (EIO_POOL, nthreads);
824} 546}
825 547
826void ecb_cold 548void ecb_cold
827eio_set_max_parallel (unsigned int nthreads) 549eio_set_max_parallel (unsigned int nthreads)
828{ 550{
829 etp_set_max_parallel (nthreads); 551 etp_set_max_parallel (EIO_POOL, nthreads);
830} 552}
831 553
832int eio_poll (void) 554int eio_poll (void)
833{ 555{
834 return etp_poll (); 556 return etp_poll (EIO_POOL);
835} 557}
836 558
837/*****************************************************************************/ 559/*****************************************************************************/
838/* work around various missing functions */ 560/* work around various missing functions */
839 561
840#if _POSIX_VERSION < 200809L 562#if HAVE_POSIX_CLOSE && !__linux
841# define realpath(path,resolved_path) (errno = ENOSYS, 0) 563# define eio__close(fd) posix_close (fd, 0)
564#else
565# define eio__close(fd) close (fd)
842#endif 566#endif
843 567
844#if !HAVE_PREADWRITE 568/* close() without disturbing errno */
845# undef pread 569static void
846# undef pwrite 570silent_close (int fd)
847# define pread eio__pread
848# define pwrite eio__pwrite
849
850static ssize_t
851eio__pread (int fd, void *buf, size_t count, off_t offset)
852{ 571{
853 ssize_t res; 572 int saved_errno = errno;
854 off_t ooffset; 573 eio__close (fd);
855 574 errno = saved_errno;
856 X_LOCK (preadwritelock);
857 ooffset = lseek (fd, 0, SEEK_CUR);
858 lseek (fd, offset, SEEK_SET);
859 res = read (fd, buf, count);
860 lseek (fd, ooffset, SEEK_SET);
861 X_UNLOCK (preadwritelock);
862
863 return res;
864} 575}
865
866static ssize_t
867eio__pwrite (int fd, void *buf, size_t count, off_t offset)
868{
869 ssize_t res;
870 off_t ooffset;
871
872 X_LOCK (preadwritelock);
873 ooffset = lseek (fd, 0, SEEK_CUR);
874 lseek (fd, offset, SEEK_SET);
875 res = write (fd, buf, count);
876 lseek (fd, ooffset, SEEK_SET);
877 X_UNLOCK (preadwritelock);
878
879 return res;
880}
881#endif
882 576
883#ifndef HAVE_UTIMES 577#ifndef HAVE_UTIMES
884 578
885# undef utimes 579# undef utimes
886# define utimes(path,times) eio__utimes (path, times) 580# define utimes(path,times) eio__utimes (path, times)
… …
919 613
920#if !HAVE_FDATASYNC 614#if !HAVE_FDATASYNC
921# undef fdatasync 615# undef fdatasync
922# define fdatasync(fd) fsync (fd) 616# define fdatasync(fd) fsync (fd)
923#endif 617#endif
618
619static int
620eio__syncfs (int fd)
621{
622 int res;
623
624#if HAVE_SYS_SYNCFS
625 res = (int)syscall (__NR_syncfs, (int)(fd));
626#else
627 res = EIO_ENOSYS ();
628#endif
629
630 if (res < 0 && errno == ENOSYS && fd >= 0)
631 sync ();
632
633 return res;
634}
924 635
925/* sync_file_range always needs emulation */ 636/* sync_file_range always needs emulation */
926static int 637static int
927eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags) 638eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
928{ 639{
… …
948 /* even though we could play tricks with the flags, it's better to always 659 /* even though we could play tricks with the flags, it's better to always
949 * call fdatasync, as that matches the expectation of its users best */ 660 * call fdatasync, as that matches the expectation of its users best */
950 return fdatasync (fd); 661 return fdatasync (fd);
951} 662}
952 663
664static int
665eio__fallocate (int fd, int mode, off_t offset, size_t len)
666{
667#if HAVE_LINUX_FALLOCATE
668 return fallocate (fd, mode, offset, len);
669#else
670 return EIO_ENOSYS ();
671#endif
672}
673
953#if !HAVE_READAHEAD 674#if !HAVE_READAHEAD
954# undef readahead 675# undef readahead
955# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 676# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
956 677
957static ssize_t 678static eio_ssize_t
958eio__readahead (int fd, off_t offset, size_t count, etp_worker *self) 679eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
959{ 680{
960 size_t todo = count; 681 size_t todo = count;
961 dBUF; 682 dBUF;
962 683
… …
967 pread (fd, eio_buf, len, offset); 688 pread (fd, eio_buf, len, offset);
968 offset += len; 689 offset += len;
969 todo -= len; 690 todo -= len;
970 } 691 }
971 692
972 errno = 0; 693 FUBd;
694
695 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
696 /* but not for e.g. EIO or eof, so we also never fail */
973 return count; 697 return 0;
974} 698}
975 699
976#endif 700#endif
977 701
978/* sendfile always needs emulation */ 702/* sendfile always needs emulation */
979static ssize_t 703static eio_ssize_t
980eio__sendfile (int ofd, int ifd, off_t offset, size_t count, etp_worker *self) 704eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
981{ 705{
982 ssize_t written = 0; 706 eio_ssize_t written = 0;
983 ssize_t res; 707 eio_ssize_t res;
984 708
985 if (!count) 709 if (!count)
986 return 0; 710 return 0;
987 711
988 for (;;) 712 for (;;)
989 { 713 {
714#ifdef __APPLE__
715# undef HAVE_SENDFILE /* broken, as everything on os x */
716#endif
990#if HAVE_SENDFILE 717#if HAVE_SENDFILE
991# if __linux 718# if __linux
992 off_t soffset = offset; 719 off_t soffset = offset;
993 res = sendfile (ofd, ifd, &soffset, count); 720 res = sendfile (ofd, ifd, &soffset, count);
994 721
… …
1010 737
1011 /* according to source inspection, this is correct, and useful behaviour */ 738 /* according to source inspection, this is correct, and useful behaviour */
1012 if (sbytes) 739 if (sbytes)
1013 res = sbytes; 740 res = sbytes;
1014 741
1015# elif defined (__APPLE__) && 0 /* broken, as everything on os x */ 742# elif defined __APPLE__
1016 off_t sbytes = count; 743 off_t sbytes = count;
1017 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0); 744 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
1018 745
1019 /* according to the manpage, sbytes is always valid */ 746 /* according to the manpage, sbytes is always valid */
1020 if (sbytes) 747 if (sbytes)
… …
1037 if (res < 0 && sbytes) 764 if (res < 0 && sbytes)
1038 res = sbytes; 765 res = sbytes;
1039 766
1040# endif 767# endif
1041 768
1042#elif defined (_WIN32) 769#elif defined (_WIN32) && 0
1043 /* does not work, just for documentation of what would need to be done */ 770 /* does not work, just for documentation of what would need to be done */
1044 /* actually, cannot be done like this, as TransmitFile changes the file offset, */ 771 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
1045 /* libeio guarantees that the file offset does not change, and windows */ 772 /* libeio guarantees that the file offset does not change, and windows */
1046 /* has no way to get an independent handle to the same file description */ 773 /* has no way to get an independent handle to the same file description */
1047 HANDLE h = TO_SOCKET (ifd); 774 HANDLE h = TO_SOCKET (ifd);
1048 SetFilePointer (h, offset, 0, FILE_BEGIN); 775 SetFilePointer (h, offset, 0, FILE_BEGIN);
1049 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0); 776 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
1050 777
1051#else 778#else
1052 res = -1; 779 res = EIO_ENOSYS ();
1053 errno = ENOSYS;
1054#endif 780#endif
1055 781
1056 /* we assume sendfile can copy at least 128mb in one go */ 782 /* we assume sendfile can copy at least 128mb in one go */
1057 if (res <= 128 * 1024 * 1024) 783 if (res <= 128 * 1024 * 1024)
1058 { 784 {
… …
1080 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 806 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
1081 /* BSDs */ 807 /* BSDs */
1082#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */ 808#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
1083 || errno == ENOTSUP 809 || errno == ENOTSUP
1084#endif 810#endif
811#ifdef EOPNOTSUPP /* windows */
1085 || errno == EOPNOTSUPP /* BSDs */ 812 || errno == EOPNOTSUPP /* BSDs */
813#endif
1086#if __solaris 814#if __solaris
1087 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 815 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
1088#endif 816#endif
1089 ) 817 )
1090 ) 818 )
… …
1094 822
1095 res = 0; 823 res = 0;
1096 824
1097 while (count) 825 while (count)
1098 { 826 {
1099 ssize_t cnt; 827 eio_ssize_t cnt;
1100 828
1101 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 829 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
1102 830
1103 if (cnt <= 0) 831 if (cnt <= 0)
1104 { 832 {
… …
1116 844
1117 offset += cnt; 845 offset += cnt;
1118 res += cnt; 846 res += cnt;
1119 count -= cnt; 847 count -= cnt;
1120 } 848 }
849
850 FUBd;
1121 } 851 }
1122 852
1123 return res; 853 return res;
854}
855
856#ifdef PAGESIZE
857# define eio_pagesize() PAGESIZE
858#else
859static intptr_t
860eio_pagesize (void)
861{
862 static intptr_t page;
863
864 if (!page)
865 page = sysconf (_SC_PAGESIZE);
866
867 return page;
868}
869#endif
870
871static void
872eio_page_align (void **addr, size_t *length)
873{
874 intptr_t mask = eio_pagesize () - 1;
875
876 /* round down addr */
877 intptr_t adj = mask & (intptr_t)*addr;
878
879 *addr = (void *)((intptr_t)*addr - adj);
880 *length += adj;
881
882 /* round up length */
883 *length = (*length + mask) & ~mask;
884}
885
886#if !_POSIX_MEMLOCK
887# define eio__mlockall(a) EIO_ENOSYS ()
888#else
889
890static int
891eio__mlockall (int flags)
892{
893 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
894 extern int mallopt (int, int);
895 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
896 #endif
897
898 if (EIO_MCL_CURRENT != MCL_CURRENT
899 || EIO_MCL_FUTURE != MCL_FUTURE)
900 {
901 flags = 0
902 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
903 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
904 }
905
906 return mlockall (flags);
907}
908#endif
909
910#if !_POSIX_MEMLOCK_RANGE
911# define eio__mlock(a,b) EIO_ENOSYS ()
912#else
913
914static int
915eio__mlock (void *addr, size_t length)
916{
917 eio_page_align (&addr, &length);
918
919 return mlock (addr, length);
920}
921
922#endif
923
924#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
925# define eio__msync(a,b,c) EIO_ENOSYS ()
926#else
927
928static int
929eio__msync (void *mem, size_t len, int flags)
930{
931 eio_page_align (&mem, &len);
932
933 if (EIO_MS_ASYNC != MS_SYNC
934 || EIO_MS_INVALIDATE != MS_INVALIDATE
935 || EIO_MS_SYNC != MS_SYNC)
936 {
937 flags = 0
938 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
939 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
940 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
941 }
942
943 return msync (mem, len, flags);
944}
945
946#endif
947
948static int
949eio__mtouch (eio_req *req)
950{
951 void *mem = req->ptr2;
952 size_t len = req->size;
953 int flags = req->int1;
954
955 eio_page_align (&mem, &len);
956
957 {
958 intptr_t addr = (intptr_t)mem;
959 intptr_t end = addr + len;
960 intptr_t page = eio_pagesize ();
961
962 if (addr < end)
963 if (flags & EIO_MT_MODIFY) /* modify */
964 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < end && !EIO_CANCELLED (req));
965 else
966 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < end && !EIO_CANCELLED (req));
967 }
968
969 return 0;
970}
971
972/*****************************************************************************/
973/* requests implemented outside eio_execute, because they are so large */
974
975static void
976eio__lseek (eio_req *req)
977{
978 /* this usually gets optimised away completely, or your compiler sucks, */
979 /* or the whence constants really are not 0, 1, 2 */
980 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
981 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
982 : req->int2 == EIO_SEEK_END ? SEEK_END
983 : req->int2;
984
985 req->offs = lseek (req->int1, req->offs, whence);
986 req->result = req->offs == (off_t)-1 ? -1 : 0;
987}
988
989/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
990static int
991eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
992{
993 char *res;
994 const char *rel = path;
995 char *tmp1, *tmp2;
996#if SYMLOOP_MAX > 32
997 int symlinks = SYMLOOP_MAX;
998#else
999 int symlinks = 32;
1000#endif
1001
1002 errno = EINVAL;
1003 if (!rel)
1004 return -1;
1005
1006 errno = ENOENT;
1007 if (!*rel)
1008 return -1;
1009
1010 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
1011#ifdef _WIN32
1012 if (_access (rel, 4) != 0)
1013 return -1;
1014
1015 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
1016
1017 errno = ENAMETOOLONG;
1018 if (symlinks >= EIO_PATH_MAX * 3)
1019 return -1;
1020
1021 errno = EIO;
1022 if (symlinks <= 0)
1023 return -1;
1024
1025 return symlinks;
1026
1027#else
1028 tmp1 = res + EIO_PATH_MAX;
1029 tmp2 = tmp1 + EIO_PATH_MAX;
1030
1031#if 0 /* disabled, the musl way to do things is just too racy */
1032#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1033 /* on linux we may be able to ask the kernel */
1034 {
1035 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1036
1037 if (fd >= 0)
1038 {
1039 sprintf (tmp1, "/proc/self/fd/%d", fd);
1040 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1041 /* here we should probably stat the open file and the disk file, to make sure they still match */
1042 eio__close (fd);
1043
1044 if (req->result > 0)
1045 goto done;
1046 }
1047 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1048 return -1;
1049 }
1050#endif
1051#endif
1052
1053 if (*rel != '/')
1054 {
1055 int len;
1056
1057 errno = ENOENT;
1058 if (wd == EIO_INVALID_WD)
1059 return -1;
1060
1061 if (wd == EIO_CWD)
1062 {
1063 if (!getcwd (res, EIO_PATH_MAX))
1064 return -1;
1065
1066 len = strlen (res);
1067 }
1068 else
1069 memcpy (res, wd->str, len = wd->len);
1070
1071 if (res [1]) /* only use if not / */
1072 res += len;
1073 }
1074
1075 while (*rel)
1076 {
1077 eio_ssize_t len, linklen;
1078 const char *beg = rel;
1079
1080 while (*rel && *rel != '/')
1081 ++rel;
1082
1083 len = rel - beg;
1084
1085 if (!len) /* skip slashes */
1086 {
1087 ++rel;
1088 continue;
1089 }
1090
1091 if (beg [0] == '.')
1092 {
1093 if (len == 1)
1094 continue; /* . - nop */
1095
1096 if (beg [1] == '.' && len == 2)
1097 {
1098 /* .. - back up one component, if possible */
1099
1100 while (res != tmpbuf->ptr)
1101 if (*--res == '/')
1102 break;
1103
1104 continue;
1105 }
1106 }
1107
1108 errno = ENAMETOOLONG;
1109 if (res + 1 + len + 1 >= tmp1)
1110 return -1;
1111
1112 /* copy one component */
1113 *res = '/';
1114 memcpy (res + 1, beg, len);
1115
1116 /* zero-terminate, for readlink */
1117 res [len + 1] = 0;
1118
1119 /* now check if it's a symlink */
1120 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1121
1122 if (linklen < 0)
1123 {
1124 if (errno != EINVAL)
1125 return -1;
1126
1127 /* it's a normal directory. hopefully */
1128 res += len + 1;
1129 }
1130 else
1131 {
1132 /* yay, it was a symlink - build new path in tmp2 */
1133 int rellen = strlen (rel);
1134
1135 errno = ENAMETOOLONG;
1136 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1137 return -1;
1138
1139 errno = ELOOP;
1140 if (!--symlinks)
1141 return -1;
1142
1143 if (*tmp1 == '/')
1144 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1145
1146 /* we need to be careful, as rel might point into tmp2 already */
1147 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1148 tmp2 [linklen] = '/';
1149 memcpy (tmp2, tmp1, linklen);
1150
1151 rel = tmp2;
1152 }
1153 }
1154
1155 /* special case for the lone root path */
1156 if (res == tmpbuf->ptr)
1157 *res++ = '/';
1158
1159 return res - (char *)tmpbuf->ptr;
1160#endif
1124} 1161}
1125 1162
1126static signed char 1163static signed char
1127eio_dent_cmp (const eio_dirent *a, const eio_dirent *b) 1164eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1128{ 1165{
… …
1136 1173
1137#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */ 1174#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1138#define EIO_SORT_FAST 60 /* when to only use insertion sort */ 1175#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1139 1176
1140static void 1177static void
1141eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1178eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1142{ 1179{
1143 unsigned char bits [9 + sizeof (ino_t) * 8]; 1180 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1144 unsigned char *bit = bits; 1181 unsigned char *bit = bits;
1145 1182
1146 assert (CHAR_BIT == 8); 1183 assert (CHAR_BIT == 8);
1147 assert (sizeof (eio_dirent) * 8 < 256); 1184 assert (sizeof (eio_dirent) * 8 < 256);
1148 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */ 1185 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
… …
1150 1187
1151 if (size <= EIO_SORT_FAST) 1188 if (size <= EIO_SORT_FAST)
1152 return; 1189 return;
1153 1190
1154 /* first prepare an array of bits to test in our radix sort */ 1191 /* first prepare an array of bits to test in our radix sort */
1155 /* try to take endianness into account, as well as differences in ino_t sizes */ 1192 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1156 /* inode_bits must contain all inodes ORed together */ 1193 /* inode_bits must contain all inodes ORed together */
1157 /* which is used to skip bits that are 0 everywhere, which is very common */ 1194 /* which is used to skip bits that are 0 everywhere, which is very common */
1158 { 1195 {
1159 ino_t endianness; 1196 eio_ino_t endianness;
1160 int i, j; 1197 int i, j;
1161 1198
1162 /* we store the byte offset of byte n into byte n of "endianness" */ 1199 /* we store the byte offset of byte n into byte n of "endianness" */
1163 for (i = 0; i < sizeof (ino_t); ++i) 1200 for (i = 0; i < sizeof (eio_ino_t); ++i)
1164 ((unsigned char *)&endianness)[i] = i; 1201 ((unsigned char *)&endianness)[i] = i;
1165 1202
1166 *bit++ = 0; 1203 *bit++ = 0;
1167 1204
1168 for (i = 0; i < sizeof (ino_t); ++i) 1205 for (i = 0; i < sizeof (eio_ino_t); ++i)
1169 { 1206 {
1170 /* shifting off the byte offsets out of "endianness" */ 1207 /* shifting off the byte offsets out of "endianness" */
1171 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8; 1208 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1172 endianness >>= 8; 1209 endianness >>= 8;
1173 1210
1174 for (j = 0; j < 8; ++j) 1211 for (j = 0; j < 8; ++j)
1175 if (inode_bits & (((ino_t)1) << (i * 8 + j))) 1212 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1176 *bit++ = offs + j; 1213 *bit++ = offs + j;
1177 } 1214 }
1178 1215
1179 for (j = 0; j < 8; ++j) 1216 for (j = 0; j < 8; ++j)
1180 if (score_bits & (1 << j)) 1217 if (score_bits & (1 << j))
1181 *bit++ = offsetof (eio_dirent, score) * 8 + j; 1218 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1182 } 1219 }
1183 1220
1184 /* now actually do the sorting (a variant of MSD radix sort) */ 1221 /* now actually do the sorting (a variant of MSD radix sort) */
1185 { 1222 {
1186 eio_dirent *base_stk [9 + sizeof (ino_t) * 8], *base; 1223 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1187 eio_dirent *end_stk [9 + sizeof (ino_t) * 8], *end; 1224 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1188 unsigned char *bit_stk [9 + sizeof (ino_t) * 8]; 1225 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1189 int stk_idx = 0; 1226 int stk_idx = 0;
1190 1227
1191 base_stk [stk_idx] = dents; 1228 base_stk [stk_idx] = dents;
1192 end_stk [stk_idx] = dents + size; 1229 end_stk [stk_idx] = dents + size;
1193 bit_stk [stk_idx] = bit - 1; 1230 bit_stk [stk_idx] = bit - 1;
… …
1272 } 1309 }
1273 } 1310 }
1274} 1311}
1275 1312
1276static void 1313static void
1277eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, ino_t inode_bits) 1314eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1278{ 1315{
1279 if (size <= 1) 1316 if (size <= 1)
1280 return; /* our insertion sort relies on size > 0 */ 1317 return; /* our insertion sort relies on size > 0 */
1281 1318
1282 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */ 1319 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
… …
1290 1327
1291/* read a full directory */ 1328/* read a full directory */
1292static void 1329static void
1293eio__scandir (eio_req *req, etp_worker *self) 1330eio__scandir (eio_req *req, etp_worker *self)
1294{ 1331{
1295 DIR *dirp;
1296 EIO_STRUCT_DIRENT *entp;
1297 char *name, *names; 1332 char *name, *names;
1298 int namesalloc = 4096; 1333 int namesalloc = 4096 - sizeof (void *) * 4;
1299 int namesoffs = 0; 1334 int namesoffs = 0;
1300 int flags = req->int1; 1335 int flags = req->int1;
1301 eio_dirent *dents = 0; 1336 eio_dirent *dents = 0;
1302 int dentalloc = 128; 1337 int dentalloc = 128;
1303 int dentoffs = 0; 1338 int dentoffs = 0;
1304 ino_t inode_bits = 0; 1339 eio_ino_t inode_bits = 0;
1340#ifdef _WIN32
1341 HANDLE dirp;
1342 WIN32_FIND_DATA entp;
1343#else
1344 DIR *dirp;
1345 EIO_STRUCT_DIRENT *entp;
1346#endif
1305 1347
1306 req->result = -1; 1348 req->result = -1;
1307 1349
1308 if (!(flags & EIO_READDIR_DENTS)) 1350 if (!(flags & EIO_READDIR_DENTS))
1309 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER); 1351 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1310 1352
1311 X_LOCK (wrklock); 1353#ifdef _WIN32
1312 /* the corresponding closedir is in ETP_WORKER_CLEAR */ 1354 {
1355 int len = strlen ((const char *)req->ptr1);
1356 char *path = malloc (MAX_PATH);
1357 const char *fmt;
1358 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1359
1360 if (!len)
1361 fmt = "./*";
1362 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1363 fmt = "%s*";
1364 else
1365 fmt = "%s/*";
1366
1367 _snprintf (path, MAX_PATH, fmt, reqpath);
1368 dirp = FindFirstFile (path, &entp);
1369 free (path);
1370
1371 if (dirp == INVALID_HANDLE_VALUE)
1372 {
1373 /* should steal _dosmaperr */
1374 switch (GetLastError ())
1375 {
1376 case ERROR_FILE_NOT_FOUND:
1377 req->result = 0;
1378 break;
1379
1380 case ERROR_INVALID_NAME:
1381 case ERROR_PATH_NOT_FOUND:
1382 case ERROR_NO_MORE_FILES:
1383 errno = ENOENT;
1384 break;
1385
1386 case ERROR_NOT_ENOUGH_MEMORY:
1387 errno = ENOMEM;
1388 break;
1389
1390 default:
1391 errno = EINVAL;
1392 break;
1393 }
1394
1395 return;
1396 }
1397 }
1398#else
1399 #if HAVE_AT
1400 if (req->wd)
1401 {
1402 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1403
1404 if (fd < 0)
1405 return;
1406
1407 dirp = fdopendir (fd);
1408
1409 if (!dirp)
1410 silent_close (fd);
1411 }
1412 else
1313 self->dirp = dirp = opendir (req->ptr1); 1413 dirp = opendir (req->ptr1);
1414 #else
1415 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1416 #endif
1417
1418 if (!dirp)
1419 return;
1420#endif
1314 1421
1315 if (req->flags & EIO_FLAG_PTR1_FREE) 1422 if (req->flags & EIO_FLAG_PTR1_FREE)
1316 free (req->ptr1); 1423 free (req->ptr1);
1317 1424
1318 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE; 1425 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1319 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0; 1426 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1320 req->ptr2 = names = malloc (namesalloc); 1427 req->ptr2 = names = malloc (namesalloc);
1321 X_UNLOCK (wrklock);
1322 1428
1323 if (dirp && names && (!flags || dents)) 1429 if (!names || (flags && !dents))
1430 return;
1431
1324 for (;;) 1432 for (;;)
1325 { 1433 {
1434 int done;
1435
1436#ifdef _WIN32
1437 done = !dirp;
1438#else
1326 errno = 0; 1439 errno = 0;
1327 entp = readdir (dirp); 1440 entp = readdir (dirp);
1441 done = !entp;
1442#endif
1328 1443
1329 if (!entp) 1444 if (done)
1330 { 1445 {
1446#ifndef _WIN32
1447 int old_errno = errno;
1448 closedir (dirp);
1449 errno = old_errno;
1450
1331 if (errno) 1451 if (errno)
1332 break; 1452 break;
1453#endif
1333 1454
1334 /* sort etc. */ 1455 /* sort etc. */
1335 req->int1 = flags; 1456 req->int1 = flags;
1336 req->result = dentoffs; 1457 req->result = dentoffs;
1337 1458
1338 if (flags & EIO_READDIR_STAT_ORDER) 1459 if (flags & EIO_READDIR_STAT_ORDER)
1339 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits); 1460 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1340 else if (flags & EIO_READDIR_DIRS_FIRST) 1461 else if (flags & EIO_READDIR_DIRS_FIRST)
1341 if (flags & EIO_READDIR_FOUND_UNKNOWN) 1462 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1342 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */ 1463 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1343 else 1464 else
1465 {
1466 /* in this case, all is known, and we just put dirs first and sort them */
1467 eio_dirent *oth = dents + dentoffs;
1468 eio_dirent *dir = dents;
1469
1470 /* now partition dirs to the front, and non-dirs to the back */
1471 /* by walking from both sides and swapping if necessary */
1472 while (oth > dir)
1473 {
1474 if (dir->type == EIO_DT_DIR)
1475 ++dir;
1476 else if ((--oth)->type == EIO_DT_DIR)
1477 {
1478 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1479
1480 ++dir;
1481 }
1482 }
1483
1484 /* now sort the dirs only (dirs all have the same score) */
1485 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1486 }
1487
1488 break;
1489 }
1490
1491 /* now add the entry to our list(s) */
1492 name = D_NAME (entp);
1493
1494 /* skip . and .. entries */
1495 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1496 {
1497 int len = D_NAMLEN (entp) + 1;
1498
1499 while (ecb_expect_false (namesoffs + len > namesalloc))
1500 {
1501 namesalloc *= 2;
1502 req->ptr2 = names = realloc (names, namesalloc);
1503
1504 if (!names)
1505 break;
1506 }
1507
1508 memcpy (names + namesoffs, name, len);
1509
1510 if (dents)
1511 {
1512 struct eio_dirent *ent;
1513
1514 if (ecb_expect_false (dentoffs == dentalloc))
1344 { 1515 {
1345 /* in this case, all is known, and we just put dirs first and sort them */ 1516 dentalloc *= 2;
1517 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1518
1519 if (!dents)
1520 break;
1521 }
1522
1346 eio_dirent *oth = dents + dentoffs; 1523 ent = dents + dentoffs;
1347 eio_dirent *dir = dents;
1348 1524
1349 /* now partition dirs to the front, and non-dirs to the back */ 1525 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1350 /* by walking from both sides and swapping if necessary */ 1526 ent->namelen = len - 1;
1351 while (oth > dir) 1527 ent->inode = D_INO (entp);
1528
1529 inode_bits |= ent->inode;
1530
1531 switch (D_TYPE (entp))
1532 {
1533 default:
1534 ent->type = EIO_DT_UNKNOWN;
1535 flags |= EIO_READDIR_FOUND_UNKNOWN;
1536 break;
1537
1538 #ifdef DT_FIFO
1539 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1540 #endif
1541 #ifdef DT_CHR
1542 case DT_CHR: ent->type = EIO_DT_CHR; break;
1543 #endif
1544 #ifdef DT_MPC
1545 case DT_MPC: ent->type = EIO_DT_MPC; break;
1546 #endif
1547 #ifdef DT_DIR
1548 case DT_DIR: ent->type = EIO_DT_DIR; break;
1549 #endif
1550 #ifdef DT_NAM
1551 case DT_NAM: ent->type = EIO_DT_NAM; break;
1552 #endif
1553 #ifdef DT_BLK
1554 case DT_BLK: ent->type = EIO_DT_BLK; break;
1555 #endif
1556 #ifdef DT_MPB
1557 case DT_MPB: ent->type = EIO_DT_MPB; break;
1558 #endif
1559 #ifdef DT_REG
1560 case DT_REG: ent->type = EIO_DT_REG; break;
1561 #endif
1562 #ifdef DT_NWK
1563 case DT_NWK: ent->type = EIO_DT_NWK; break;
1564 #endif
1565 #ifdef DT_CMP
1566 case DT_CMP: ent->type = EIO_DT_CMP; break;
1567 #endif
1568 #ifdef DT_LNK
1569 case DT_LNK: ent->type = EIO_DT_LNK; break;
1570 #endif
1571 #ifdef DT_SOCK
1572 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1573 #endif
1574 #ifdef DT_DOOR
1575 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1576 #endif
1577 #ifdef DT_WHT
1578 case DT_WHT: ent->type = EIO_DT_WHT; break;
1579 #endif
1580 }
1581
1582 ent->score = 7;
1583
1584 if (flags & EIO_READDIR_DIRS_FIRST)
1585 {
1586 if (ent->type == EIO_DT_UNKNOWN)
1352 { 1587 {
1353 if (dir->type == EIO_DT_DIR) 1588 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1354 ++dir; 1589 ent->score = 1;
1355 else if ((--oth)->type == EIO_DT_DIR) 1590 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1356 { 1591 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1357 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1358
1359 ++dir;
1360 }
1361 } 1592 }
1362 1593 else if (ent->type == EIO_DT_DIR)
1363 /* now sort the dirs only (dirs all have the same score) */ 1594 ent->score = 0;
1364 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1365 } 1595 }
1366
1367 break;
1368 }
1369
1370 /* now add the entry to our list(s) */
1371 name = entp->d_name;
1372
1373 /* skip . and .. entries */
1374 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
1375 {
1376 int len = D_NAMLEN (entp) + 1;
1377
1378 while (ecb_expect_false (namesoffs + len > namesalloc))
1379 {
1380 namesalloc *= 2;
1381 X_LOCK (wrklock);
1382 req->ptr2 = names = realloc (names, namesalloc);
1383 X_UNLOCK (wrklock);
1384
1385 if (!names)
1386 break;
1387 } 1596 }
1388 1597
1389 memcpy (names + namesoffs, name, len);
1390
1391 if (dents)
1392 {
1393 struct eio_dirent *ent;
1394
1395 if (ecb_expect_false (dentoffs == dentalloc))
1396 {
1397 dentalloc *= 2;
1398 X_LOCK (wrklock);
1399 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1400 X_UNLOCK (wrklock);
1401
1402 if (!dents)
1403 break;
1404 }
1405
1406 ent = dents + dentoffs;
1407
1408 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1409 ent->namelen = len - 1;
1410 ent->inode = D_INO (entp);
1411
1412 inode_bits |= ent->inode;
1413
1414 switch (D_TYPE (entp))
1415 {
1416 default:
1417 ent->type = EIO_DT_UNKNOWN;
1418 flags |= EIO_READDIR_FOUND_UNKNOWN;
1419 break;
1420
1421 #ifdef DT_FIFO
1422 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1423 #endif
1424 #ifdef DT_CHR
1425 case DT_CHR: ent->type = EIO_DT_CHR; break;
1426 #endif
1427 #ifdef DT_MPC
1428 case DT_MPC: ent->type = EIO_DT_MPC; break;
1429 #endif
1430 #ifdef DT_DIR
1431 case DT_DIR: ent->type = EIO_DT_DIR; break;
1432 #endif
1433 #ifdef DT_NAM
1434 case DT_NAM: ent->type = EIO_DT_NAM; break;
1435 #endif
1436 #ifdef DT_BLK
1437 case DT_BLK: ent->type = EIO_DT_BLK; break;
1438 #endif
1439 #ifdef DT_MPB
1440 case DT_MPB: ent->type = EIO_DT_MPB; break;
1441 #endif
1442 #ifdef DT_REG
1443 case DT_REG: ent->type = EIO_DT_REG; break;
1444 #endif
1445 #ifdef DT_NWK
1446 case DT_NWK: ent->type = EIO_DT_NWK; break;
1447 #endif
1448 #ifdef DT_CMP
1449 case DT_CMP: ent->type = EIO_DT_CMP; break;
1450 #endif
1451 #ifdef DT_LNK
1452 case DT_LNK: ent->type = EIO_DT_LNK; break;
1453 #endif
1454 #ifdef DT_SOCK
1455 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1456 #endif
1457 #ifdef DT_DOOR
1458 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1459 #endif
1460 #ifdef DT_WHT
1461 case DT_WHT: ent->type = EIO_DT_WHT; break;
1462 #endif
1463 }
1464
1465 ent->score = 7;
1466
1467 if (flags & EIO_READDIR_DIRS_FIRST)
1468 {
1469 if (ent->type == EIO_DT_UNKNOWN)
1470 {
1471 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1472 ent->score = 1;
1473 else if (!strchr (name, '.')) /* absense of dots indicate likely dirs */
1474 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1475 }
1476 else if (ent->type == EIO_DT_DIR)
1477 ent->score = 0;
1478 }
1479 }
1480
1481 namesoffs += len; 1598 namesoffs += len;
1482 ++dentoffs; 1599 ++dentoffs;
1483 } 1600 }
1484 1601
1485 if (EIO_CANCELLED (req)) 1602 if (EIO_CANCELLED (req))
1486 { 1603 {
1487 errno = ECANCELED; 1604 errno = ECANCELED;
1488 break; 1605 break;
1489 } 1606 }
1607
1608#ifdef _WIN32
1609 if (!FindNextFile (dirp, &entp))
1610 {
1611 FindClose (dirp);
1612 dirp = 0;
1490 } 1613 }
1614#endif
1615 }
1491} 1616}
1492 1617
1493#ifdef PAGESIZE 1618/*****************************************************************************/
1494# define eio_pagesize() PAGESIZE 1619/* working directory stuff */
1620/* various deficiencies in the posix 2008 api force us to */
1621/* keep the absolute path in string form at all times */
1622/* fuck yeah. */
1623
1624#if !HAVE_AT
1625
1626/* a bit like realpath, but usually faster because it doesn'T have to return */
1627/* an absolute or canonical path */
1628static const char *
1629wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1630{
1631 if (!wd || *path == '/')
1632 return path;
1633
1634 if (path [0] == '.' && !path [1])
1635 return wd->str;
1636
1637 {
1638 int l1 = wd->len;
1639 int l2 = strlen (path);
1640
1641 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1642
1643 memcpy (res, wd->str, l1);
1644 res [l1] = '/';
1645 memcpy (res + l1 + 1, path, l2 + 1);
1646
1647 return res;
1648 }
1649}
1650
1651#endif
1652
1653static eio_wd
1654eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1655{
1656 int fd;
1657 eio_wd res;
1658 int len = eio__realpath (tmpbuf, wd, path);
1659
1660 if (len < 0)
1661 return EIO_INVALID_WD;
1662
1663#if HAVE_AT
1664 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1665
1666 if (fd < 0)
1667 return EIO_INVALID_WD;
1668#endif
1669
1670 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1671
1672#if HAVE_AT
1673 res->fd = fd;
1674#endif
1675
1676 res->len = len;
1677 memcpy (res->str, tmpbuf->ptr, len);
1678 res->str [len] = 0;
1679
1680 return res;
1681}
1682
1683eio_wd
1684eio_wd_open_sync (eio_wd wd, const char *path)
1685{
1686 struct etp_tmpbuf tmpbuf = { };
1687 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1688 free (tmpbuf.ptr);
1689
1690 return wd;
1691}
1692
1693void
1694eio_wd_close_sync (eio_wd wd)
1695{
1696 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1697 {
1698 #if HAVE_AT
1699 eio__close (wd->fd);
1700 #endif
1701 free (wd);
1702 }
1703}
1704
1705#if HAVE_AT
1706
1707static int
1708eio__renameat2 (int olddirfd, const char *oldpath, int newdirfd, const char *newpath, unsigned int flags)
1709{
1710#if HAVE_RENAMEAT2
1711 return syscall (SYS_renameat2, olddirfd, oldpath, newdirfd, newpath, flags);
1495#else 1712#else
1496static intptr_t
1497eio_pagesize (void)
1498{
1499 static intptr_t page;
1500
1501 if (!page) 1713 if (flags)
1502 page = sysconf (_SC_PAGESIZE); 1714 return EIO_ENOSYS ();
1503 1715
1504 return page; 1716 return renameat (olddirfd, oldpath, newdirfd, newpath);
1505}
1506#endif 1717#endif
1507
1508static void
1509eio_page_align (void **addr, size_t *length)
1510{
1511 intptr_t mask = eio_pagesize () - 1;
1512
1513 /* round down addr */
1514 intptr_t adj = mask & (intptr_t)*addr;
1515
1516 *addr = (void *)((intptr_t)*addr - adj);
1517 *length += adj;
1518
1519 /* round up length */
1520 *length = (*length + mask) & ~mask;
1521} 1718}
1522 1719
1523#if !_POSIX_MEMLOCK 1720/* they forgot these */
1524# define eio__mlockall(a) ((errno = ENOSYS), -1)
1525#else
1526 1721
1527static int 1722static int
1528eio__mlockall (int flags) 1723eio__truncateat (int dirfd, const char *path, off_t length)
1529{ 1724{
1530 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7 1725 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1531 extern int mallopt (int, int); 1726 int res;
1532 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
1533 #endif
1534 1727
1535 if (EIO_MCL_CURRENT != MCL_CURRENT 1728 if (fd < 0)
1536 || EIO_MCL_FUTURE != MCL_FUTURE) 1729 return fd;
1537 {
1538 flags = 0
1539 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
1540 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
1541 }
1542 1730
1543 return mlockall (flags); 1731 res = ftruncate (fd, length);
1732 silent_close (fd);
1733 return res;
1544} 1734}
1545#endif
1546
1547#if !_POSIX_MEMLOCK_RANGE
1548# define eio__mlock(a,b) ((errno = ENOSYS), -1)
1549#else
1550 1735
1551static int 1736static int
1552eio__mlock (void *addr, size_t length) 1737eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1553{ 1738{
1554 eio_page_align (&addr, &length); 1739 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1740 int res;
1555 1741
1556 return mlock (addr, length); 1742 if (fd < 0)
1557} 1743 return fd;
1558 1744
1559#endif 1745 res = fstatvfs (fd, buf);
1560 1746 silent_close (fd);
1561#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
1562# define eio__msync(a,b,c) ((errno = ENOSYS), -1)
1563#else
1564
1565static int
1566eio__msync (void *mem, size_t len, int flags)
1567{
1568 eio_page_align (&mem, &len);
1569
1570 if (EIO_MS_ASYNC != MS_SYNC
1571 || EIO_MS_INVALIDATE != MS_INVALIDATE
1572 || EIO_MS_SYNC != MS_SYNC)
1573 {
1574 flags = 0
1575 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
1576 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
1577 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
1578 }
1579
1580 return msync (mem, len, flags);
1581}
1582
1583#endif
1584
1585static int
1586eio__mtouch (eio_req *req)
1587{
1588 void *mem = req->ptr2;
1589 size_t len = req->size;
1590 int flags = req->int1;
1591
1592 eio_page_align (&mem, &len);
1593
1594 {
1595 intptr_t addr = (intptr_t)mem;
1596 intptr_t end = addr + len;
1597 intptr_t page = eio_pagesize ();
1598
1599 if (addr < end)
1600 if (flags & EIO_MT_MODIFY) /* modify */
1601 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
1602 else
1603 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
1604 }
1605
1606 return 0; 1747 return res;
1607} 1748}
1749
1750#endif
1608 1751
1609/*****************************************************************************/ 1752/*****************************************************************************/
1610 1753
1611#define ALLOC(len) \ 1754#define ALLOC(len) \
1612 if (!req->ptr2) \ 1755 if (!req->ptr2) \
1613 { \ 1756 { \
1614 X_LOCK (wrklock); \ 1757 X_LOCK (EIO_POOL->wrklock); \
1615 req->flags |= EIO_FLAG_PTR2_FREE; \ 1758 req->flags |= EIO_FLAG_PTR2_FREE; \
1616 X_UNLOCK (wrklock); \ 1759 X_UNLOCK (EIO_POOL->wrklock); \
1617 req->ptr2 = malloc (len); \ 1760 req->ptr2 = malloc (len); \
1618 if (!req->ptr2) \ 1761 if (!req->ptr2) \
1619 { \ 1762 { \
1620 errno = ENOMEM; \ 1763 errno = ENOMEM; \
1621 req->result = -1; \ 1764 req->result = -1; \
1622 break; \ 1765 goto alloc_fail; \
1623 } \ 1766 } \
1624 } 1767 }
1625 1768
1626X_THREAD_PROC (etp_proc) 1769/*****************************************************************************/
1627{
1628 ETP_REQ *req;
1629 struct timespec ts;
1630 etp_worker *self = (etp_worker *)thr_arg;
1631 1770
1632 /* try to distribute timeouts somewhat randomly */ 1771static void
1633 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL); 1772eio__slurp (int fd, eio_req *req)
1773{
1774 req->result = fd;
1634 1775
1635 for (;;) 1776 if (fd < 0)
1777 return;
1778
1779 if (req->offs < 0 || !req->size) /* do we need the size? */
1636 { 1780 {
1637 X_LOCK (reqlock); 1781 off_t size = lseek (fd, 0, SEEK_END);
1638 1782
1639 for (;;)
1640 {
1641 self->req = req = reqq_shift (&req_queue);
1642
1643 if (req)
1644 break;
1645
1646 ++idle;
1647
1648 ts.tv_sec = time (0) + idle_timeout;
1649 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
1650 {
1651 if (idle > max_idle)
1652 {
1653 --idle;
1654 X_UNLOCK (reqlock);
1655 X_LOCK (wrklock);
1656 --started;
1657 X_UNLOCK (wrklock);
1658 goto quit;
1659 }
1660
1661 /* we are allowed to idle, so do so without any timeout */
1662 X_COND_WAIT (reqwait, reqlock);
1663 }
1664
1665 --idle;
1666 }
1667
1668 --nready;
1669
1670 X_UNLOCK (reqlock);
1671
1672 if (req->type < 0) 1783 if (req->offs < 0)
1673 goto quit; 1784 req->offs += size;
1674 1785
1675 if (!EIO_CANCELLED (req)) 1786 if (!req->size)
1676 ETP_EXECUTE (self, req); 1787 req->size = size - req->offs;
1677
1678 X_LOCK (reslock);
1679
1680 ++npending;
1681
1682 if (!reqq_push (&res_queue, req) && want_poll_cb)
1683 want_poll_cb ();
1684
1685 self->req = 0;
1686 etp_worker_clear (self);
1687
1688 X_UNLOCK (reslock);
1689 } 1788 }
1690 1789
1691quit: 1790 ALLOC (req->size);
1692 X_LOCK (wrklock); 1791 req->result = pread (fd, req->ptr2, req->size, req->offs);
1693 etp_worker_free (self);
1694 X_UNLOCK (wrklock);
1695 1792
1696 return 0; 1793 silent_close (fd);
1697}
1698 1794
1699/*****************************************************************************/ 1795alloc_fail:
1796 ;
1797}
1700 1798
1701int ecb_cold 1799int ecb_cold
1702eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1800eio_init (void (*want_poll)(void), void (*done_poll)(void))
1703{ 1801{
1704 return etp_init (want_poll, done_poll); 1802 eio_want_poll_cb = want_poll;
1803 eio_done_poll_cb = done_poll;
1804
1805 return etp_init (EIO_POOL, 0, 0, 0);
1705} 1806}
1706 1807
1707ecb_inline void 1808ecb_inline void
1708eio_api_destroy (eio_req *req) 1809eio_api_destroy (eio_req *req)
1709{ 1810{
1710 free (req); 1811 free (req);
1711} 1812}
1712 1813
1713#define REQ(rtype) \ 1814#define REQ(rtype) \
1714 eio_req *req; \ 1815 eio_req *req; \
1715 \ 1816 \
1716 req = (eio_req *)calloc (1, sizeof *req); \ 1817 req = (eio_req *)calloc (1, sizeof *req); \
1717 if (!req) \ 1818 if (!req) \
1718 return 0; \ 1819 return 0; \
… …
1732 { \ 1833 { \
1733 eio_api_destroy (req); \ 1834 eio_api_destroy (req); \
1734 return 0; \ 1835 return 0; \
1735 } 1836 }
1736 1837
1838#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1839
1737static void 1840static void
1738eio_execute (etp_worker *self, eio_req *req) 1841eio_execute (etp_worker *self, eio_req *req)
1739{ 1842{
1843#if HAVE_AT
1844 int dirfd;
1845#else
1846 const char *path;
1847#endif
1848
1849 if (ecb_expect_false (EIO_CANCELLED (req)))
1850 {
1851 req->result = -1;
1852 req->errorno = ECANCELED;
1853 return;
1854 }
1855
1856 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1857 {
1858 req->result = -1;
1859 req->errorno = ENOENT;
1860 return;
1861 }
1862
1863 if (req->type >= EIO_OPEN)
1864 {
1865 #if HAVE_AT
1866 dirfd = WD2FD (req->wd);
1867 #else
1868 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1869 #endif
1870 }
1871
1740 switch (req->type) 1872 switch (req->type)
1741 { 1873 {
1874 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1875 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1876 break;
1877 case EIO_WD_CLOSE: req->result = 0;
1878 eio_wd_close_sync (req->wd); break;
1879
1880 case EIO_SEEK: eio__lseek (req); break;
1742 case EIO_READ: ALLOC (req->size); 1881 case EIO_READ: ALLOC (req->size);
1743 req->result = req->offs >= 0 1882 req->result = req->offs >= 0
1744 ? pread (req->int1, req->ptr2, req->size, req->offs) 1883 ? pread (req->int1, req->ptr2, req->size, req->offs)
1745 : read (req->int1, req->ptr2, req->size); break; 1884 : read (req->int1, req->ptr2, req->size); break;
1746 case EIO_WRITE: req->result = req->offs >= 0 1885 case EIO_WRITE: req->result = req->offs >= 0
1747 ? pwrite (req->int1, req->ptr2, req->size, req->offs) 1886 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1748 : write (req->int1, req->ptr2, req->size); break; 1887 : write (req->int1, req->ptr2, req->size); break;
1749 1888
1889 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1890 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1891
1750 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break; 1892 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1751 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break; 1893 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1894
1895#if HAVE_AT
1752 1896
1753 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1897 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1754 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1898 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1755 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 1899 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1900 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1901 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1902 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1903 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1904 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1905 case EIO_SLURP: eio__slurp ( openat (dirfd, req->ptr1, O_RDONLY | O_CLOEXEC), req); break;
1906
1907 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1908 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1909 req->result = req->wd && SINGLEDOT (req->ptr1)
1910 ? rmdir (req->wd->str)
1911 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1912 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1913 case EIO_RENAME: req->result = eio__renameat2 (
1914 dirfd,
1915 /* complications arise because "." cannot be renamed, so we might have to expand */
1916 req->wd && SINGLEDOT (req->ptr1) ? req->wd->str : req->ptr1,
1917 WD2FD ((eio_wd)req->int3),
1918 req->ptr2,
1919 req->int2
1920 );
1921 break;
1922 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1923 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1924 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1925 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1926 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1927 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1928 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1929 if (req->result == EIO_PATH_MAX)
1930 {
1931 req->result = -1;
1932 errno = ENAMETOOLONG;
1933 }
1934 break;
1935 case EIO_UTIME:
1936 case EIO_FUTIME:
1937 {
1938 struct timespec ts[2];
1939 struct timespec *times;
1940
1941 if (req->nv1 != -1. || req->nv2 != -1.)
1942 {
1943 ts[0].tv_sec = req->nv1;
1944 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1945 ts[1].tv_sec = req->nv2;
1946 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1947
1948 times = ts;
1949 }
1950 else
1951 times = 0;
1952
1953 req->result = req->type == EIO_FUTIME
1954 ? futimens (req->int1, times)
1955 : utimensat (dirfd, req->ptr1, times, 0);
1956 }
1957 break;
1958
1959#else
1960
1961 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1962 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1963 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1756 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break; 1964 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1965 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1966 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1967 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1968 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1969 case EIO_SLURP: eio__slurp ( open (path , O_RDONLY | O_CLOEXEC), req); break;
1970
1971 case EIO_UNLINK: req->result = unlink (path ); break;
1972 case EIO_RMDIR: req->result = rmdir (path ); break;
1973 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1974 case EIO_RENAME: req->result = req->int2 ? EIO_ENOSYS () : rename (path, req->ptr2); break;
1975 case EIO_LINK: req->result = link (path , req->ptr2); break;
1976 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1977 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1978 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1979 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1980 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1981 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1982 if (req->result == EIO_PATH_MAX)
1983 {
1984 req->result = -1;
1985 errno = ENAMETOOLONG;
1986 }
1987 break;
1988
1989 case EIO_UTIME:
1990 case EIO_FUTIME:
1991 {
1992 struct timeval tv[2];
1993 struct timeval *times;
1994
1995 if (req->nv1 != -1. || req->nv2 != -1.)
1996 {
1997 tv[0].tv_sec = req->nv1;
1998 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1999 tv[1].tv_sec = req->nv2;
2000 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
2001
2002 times = tv;
2003 }
2004 else
2005 times = 0;
2006
2007 req->result = req->type == EIO_FUTIME
2008 ? futimes (req->int1, times)
2009 : utimes (req->ptr1, times);
2010 }
2011 break;
2012
2013#endif
2014
2015 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
2016 {
2017 ALLOC (req->result);
2018 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
2019 }
2020 break;
2021
1757 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT)); 2022 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1758 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break; 2023 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1759 2024
1760 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1761 req->result = statvfs (req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1762 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS)); 2025 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1763 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break; 2026 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1764 2027
1765 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
1766 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break; 2028 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1767 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
1768 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break; 2029 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1769 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
1770 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break; 2030 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1771 2031
1772 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
1773 case EIO_CLOSE: req->result = close (req->int1); break; 2032 case EIO_CLOSE: req->result = eio__close (req->int1); break;
1774 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break; 2033 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1775 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
1776 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
1777 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
1778 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
1779 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
1780 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
1781 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1782
1783 case EIO_REALPATH: req->flags |= EIO_FLAG_PTR2_FREE;
1784 req->ptr2 = realpath (req->ptr1, 0);
1785 req->result = req->ptr2 ? strlen (req->ptr2) : -1;
1786 break;
1787
1788 case EIO_READLINK: ALLOC (PATH_MAX);
1789 req->result = readlink (req->ptr1, req->ptr2, PATH_MAX); break;
1790
1791 case EIO_SYNC: req->result = 0; sync (); break; 2034 case EIO_SYNC: req->result = 0; sync (); break;
1792 case EIO_FSYNC: req->result = fsync (req->int1); break; 2035 case EIO_FSYNC: req->result = fsync (req->int1); break;
1793 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break; 2036 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
2037 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
2038 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1794 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break; 2039 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1795 case EIO_MTOUCH: req->result = eio__mtouch (req); break; 2040 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1796 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break; 2041 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1797 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break; 2042 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1798 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break; 2043 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1799 2044
1800 case EIO_READDIR: eio__scandir (req, self); break; 2045 case EIO_READDIR: eio__scandir (req, self); break;
1801 2046
1802 case EIO_BUSY: 2047 case EIO_BUSY:
1803#ifdef _WIN32 2048#ifdef _WIN32
… …
1812 req->result = select (0, 0, 0, 0, &tv); 2057 req->result = select (0, 0, 0, 0, &tv);
1813 } 2058 }
1814#endif 2059#endif
1815 break; 2060 break;
1816 2061
1817 case EIO_UTIME: 2062#if 0
1818 case EIO_FUTIME:
1819 {
1820 struct timeval tv[2];
1821 struct timeval *times;
1822
1823 if (req->nv1 != -1. || req->nv2 != -1.)
1824 {
1825 tv[0].tv_sec = req->nv1;
1826 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
1827 tv[1].tv_sec = req->nv2;
1828 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
1829
1830 times = tv;
1831 }
1832 else
1833 times = 0;
1834
1835 req->result = req->type == EIO_FUTIME
1836 ? futimes (req->int1, times)
1837 : utimes (req->ptr1, times);
1838 }
1839 break;
1840
1841 case EIO_GROUP: 2063 case EIO_GROUP:
1842 abort (); /* handled in eio_request */ 2064 abort (); /* handled in eio_request */
2065#endif
1843 2066
1844 case EIO_NOP: 2067 case EIO_NOP:
1845 req->result = 0; 2068 req->result = 0;
1846 break; 2069 break;
1847 2070
1848 case EIO_CUSTOM: 2071 case EIO_CUSTOM:
1849 req->feed (req); 2072 req->feed (req);
1850 break; 2073 break;
1851 2074
1852 default: 2075 default:
1853 errno = ENOSYS;
1854 req->result = -1; 2076 req->result = EIO_ENOSYS ();
1855 break; 2077 break;
1856 } 2078 }
1857 2079
2080alloc_fail:
1858 req->errorno = errno; 2081 req->errorno = errno;
1859} 2082}
1860 2083
1861#ifndef EIO_NO_WRAPPERS 2084#ifndef EIO_NO_WRAPPERS
1862 2085
2086eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2087{
2088 REQ (EIO_WD_OPEN); PATH; SEND;
2089}
2090
2091eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2092{
2093 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2094}
2095
1863eio_req *eio_nop (int pri, eio_cb cb, void *data) 2096eio_req *eio_nop (int pri, eio_cb cb, void *data)
1864{ 2097{
1865 REQ (EIO_NOP); SEND; 2098 REQ (EIO_NOP); SEND;
1866} 2099}
1867 2100
… …
1883eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2116eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1884{ 2117{
1885 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2118 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1886} 2119}
1887 2120
2121eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
2122{
2123 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
2124}
2125
2126eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2127{
2128 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2129}
2130
2131eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2132{
2133 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2134}
2135
1888eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data) 2136eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
1889{ 2137{
1890 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND; 2138 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
1891} 2139}
1892 2140
… …
1898eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data) 2146eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
1899{ 2147{
1900 REQ (EIO_MLOCKALL); req->int1 = flags; SEND; 2148 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
1901} 2149}
1902 2150
1903eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data) 2151eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
1904{ 2152{
1905 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND; 2153 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
1906}
1907
1908eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1909{
1910 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1911} 2154}
1912 2155
1913eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2156eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1914{ 2157{
1915 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2158 REQ (EIO_CLOSE); req->int1 = fd; SEND;
… …
1918eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2161eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1919{ 2162{
1920 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2163 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1921} 2164}
1922 2165
2166eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2167{
2168 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2169}
2170
1923eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2171eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1924{ 2172{
1925 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2173 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1926} 2174}
1927 2175
1928eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2176eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1929{ 2177{
1930 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2178 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1931} 2179}
1932 2180
2181eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2182{
2183 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2184}
2185
2186eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2187{
2188 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2189}
2190
1933eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2191eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1934{ 2192{
1935 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2193 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1936} 2194}
1937 2195
… …
1953eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2211eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1954{ 2212{
1955 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2213 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1956} 2214}
1957 2215
1958eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2216eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1959{ 2217{
1960 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2218 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1961} 2219}
1962 2220
1963eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2221eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1983eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2241eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1984{ 2242{
1985 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2243 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1986} 2244}
1987 2245
1988eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2246eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1989{ 2247{
1990 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2248 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1991} 2249}
1992 2250
1993eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2251eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
2080eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2338eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
2081{ 2339{
2082 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2340 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2083} 2341}
2084 2342
2343eio_req *eio_slurp (const char *path, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
2344{
2345 REQ (EIO_SLURP); PATH; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
2346}
2347
2085eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data) 2348eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2086{ 2349{
2087 REQ (EIO_CUSTOM); req->feed = execute; SEND; 2350 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2088} 2351}
2089 2352
… …
2123void 2386void
2124eio_grp_add (eio_req *grp, eio_req *req) 2387eio_grp_add (eio_req *grp, eio_req *req)
2125{ 2388{
2126 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2389 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2127 2390
2128 grp->flags |= EIO_FLAG_GROUPADD; 2391 grp->flags |= ETP_FLAG_GROUPADD;
2129 2392
2130 ++grp->size; 2393 ++grp->size;
2131 req->grp = grp; 2394 req->grp = grp;
2132 2395
2133 req->grp_prev = 0; 2396 req->grp_prev = 0;
… …
2140} 2403}
2141 2404
2142/*****************************************************************************/ 2405/*****************************************************************************/
2143/* misc garbage */ 2406/* misc garbage */
2144 2407
2145ssize_t 2408eio_ssize_t
2146eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2409eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
2147{ 2410{
2148 etp_worker wrk;
2149 ssize_t ret;
2150
2151 wrk.dbuf = 0;
2152
2153 ret = eio__sendfile (ofd, ifd, offset, count, &wrk); 2411 return eio__sendfile (ofd, ifd, offset, count);
2154
2155 if (wrk.dbuf)
2156 free (wrk.dbuf);
2157
2158 return ret;
2159} 2412}
2160 2413

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines