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

Comparing libeio/eio.c (file contents):
Revision 1.12 by root, Tue May 13 18:54:52 2008 UTC vs.
Revision 1.143 by root, Sat Dec 3 16:33:46 2016 UTC

1/*
2 * libeio implementation
3 *
4 * Copyright (c) 2007,2008,2009,2010,2011,2012,2013,2016 Marc Alexander Lehmann <libeio@schmorp.de>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without modifica-
8 * tion, are permitted provided that the following conditions are met:
9 *
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
12 *
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
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-
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-
21 * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
22 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
23 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
24 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
25 * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
26 * OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 * Alternatively, the contents of this file may be used under the terms of
29 * the GNU General Public License ("GPL") version 2 or any later version,
30 * in which case the provisions of the GPL are applicable instead of
31 * the above. If you wish to allow the use of your version of this file
32 * only under the terms of the GPL and not to allow others to use your
33 * version of this file under the BSD license, indicate your decision
34 * by deleting the provisions above and replace them with the notice
35 * and other provisions required by the GPL. If you do not delete the
36 * provisions above, a recipient may use your version of this file under
37 * either the BSD or the GPL.
38 */
39
40#ifndef _WIN32
41# include "config.h"
42#endif
43
1#include "eio.h" 44#include "eio.h"
2#include "xthread.h" 45#include "ecb.h"
3 46
4#include <errno.h> 47#include <errno.h>
5#include <stddef.h> 48#include <stddef.h>
6#include <stdlib.h> 49#include <stdlib.h>
7#include <string.h> 50#include <string.h>
… …
10#include <sys/stat.h> 53#include <sys/stat.h>
11#include <limits.h> 54#include <limits.h>
12#include <fcntl.h> 55#include <fcntl.h>
13#include <assert.h> 56#include <assert.h>
14 57
58/* intptr_t comes from unistd.h, says POSIX/UNIX/tradition */
59/* intptr_t only comes from stdint.h, says idiot openbsd coder */
60#if HAVE_STDINT_H
61# include <stdint.h>
62#endif
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
15#ifndef EIO_FINISH 77#ifndef EIO_FINISH
16# 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
17#endif 79#endif
18 80
19#ifndef EIO_DESTROY 81#ifndef EIO_DESTROY
… …
22 84
23#ifndef EIO_FEED 85#ifndef EIO_FEED
24# 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)
25#endif 87#endif
26 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
27#ifdef _WIN32 100#ifdef _WIN32
28 101
29 /*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)
30 233
31#else 234#else
32 235
33# include "config.h"
34# include <sys/time.h> 236 #include <sys/time.h>
35# include <sys/select.h> 237 #include <sys/select.h>
36# include <unistd.h> 238 #include <unistd.h>
239 #include <signal.h>
240 #include <dirent.h>
241
242 #ifdef ANDROID
243 #include <sys/vfs.h>
244 #define statvfs statfs
245 #define fstatvfs fstatfs
246 #include <asm/page.h> /* supposedly limits.h does #define PAGESIZE PAGESIZE */
247 #else
248 #include <sys/statvfs.h>
249 #endif
250
251 #if _POSIX_MEMLOCK || _POSIX_MEMLOCK_RANGE || _POSIX_MAPPED_FILES
252 #include <sys/mman.h>
253 #endif
254
255 #define D_NAME(entp) entp->d_name
256
257 /* POSIX_SOURCE is useless on bsd's, and XOPEN_SOURCE is unreliable there, too */
258 #if __FreeBSD__ || __NetBSD__ || __OpenBSD__
259 #define _DIRENT_HAVE_D_TYPE /* sigh */
260 #define D_INO(de) (de)->d_fileno
261 #define D_NAMLEN(de) (de)->d_namlen
262 #elif __linux || defined d_ino || _XOPEN_SOURCE >= 600
263 #define D_INO(de) (de)->d_ino
264 #endif
265
266 #ifdef _D_EXACT_NAMLEN
267 #undef D_NAMLEN
268 #define D_NAMLEN(de) _D_EXACT_NAMLEN (de)
269 #endif
270
271 #ifdef _DIRENT_HAVE_D_TYPE
272 #define D_TYPE(de) (de)->d_type
273 #endif
274
275 #ifndef EIO_STRUCT_DIRENT
276 #define EIO_STRUCT_DIRENT struct dirent
277 #endif
278
279#endif
280
281#if HAVE_UTIMES
37# include <utime.h> 282# include <utime.h>
38# include <signal.h>
39# include <dirent.h>
40
41# ifndef EIO_STRUCT_DIRENT
42# define EIO_STRUCT_DIRENT struct dirent
43# endif 283#endif
44 284
285#if HAVE_SYS_SYSCALL_H
286# include <sys/syscall.h>
45#endif 287#endif
46 288
47#if HAVE_SENDFILE 289#if HAVE_SENDFILE
48# if __linux 290# if __linux
49# include <sys/sendfile.h> 291# include <sys/sendfile.h>
50# elif __freebsd 292# elif __FreeBSD__ || defined __APPLE__
51# include <sys/socket.h> 293# include <sys/socket.h>
52# include <sys/uio.h> 294# include <sys/uio.h>
53# elif __hpux 295# elif __hpux
54# include <sys/socket.h> 296# include <sys/socket.h>
55# elif __solaris /* not yet */ 297# elif __solaris
56# include <sys/sendfile.h> 298# include <sys/sendfile.h>
57# else 299# else
58# error sendfile support requested but not available 300# error sendfile support requested but not available
59# endif 301# endif
60#endif 302#endif
61 303
62/* number of seconds after which an idle threads exit */ 304#ifndef D_TYPE
63#define IDLE_TIMEOUT 10 305# define D_TYPE(de) 0
306#endif
307#ifndef D_INO
308# define D_INO(de) 0
309#endif
310#ifndef D_NAMLEN
311# define D_NAMLEN(entp) strlen (D_NAME (entp))
312#endif
64 313
65/* used for struct dirent, AIX doesn't provide it */ 314/* used for struct dirent, AIX doesn't provide it */
66#ifndef NAME_MAX 315#ifndef NAME_MAX
67# define NAME_MAX 4096 316# define NAME_MAX 4096
68#endif 317#endif
69 318
319/* used for readlink etc. */
320#ifndef PATH_MAX
321# define PATH_MAX 4096
322#endif
323
70/* buffer size for various temporary buffers */ 324/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 325#define EIO_BUFSIZE 65536
72 326
73#define dBUF \ 327#define dBUF \
74 char *eio_buf; \
75 X_LOCK (wrklock); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 328 char *eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (wrklock); \
78 errno = ENOMEM; \ 329 errno = ENOMEM; \
79 if (!eio_buf) \ 330 if (!eio_buf) \
80 return -1; 331 return -1
81 332
82#define EIO_TICKS ((1000000 + 1023) >> 10) 333#define FUBd \
334 free (eio_buf)
83 335
84static void (*want_poll_cb) (void); 336/*****************************************************************************/
85static void (*done_poll_cb) (void);
86 337
87static unsigned int max_poll_time = 0; 338struct etp_tmpbuf;
88static unsigned int max_poll_reqs = 0;
89 339
90/* calculcate time difference in ~1/EIO_TICKS of a second */ 340#if _POSIX_VERSION >= 200809L
91static int tvdiff (struct timeval *tv1, struct timeval *tv2) 341 #define HAVE_AT 1
92{ 342 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
93 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 343 #ifndef O_SEARCH
94 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 344 #define O_SEARCH O_RDONLY
95} 345 #endif
346#else
347 #define HAVE_AT 0
348 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
349#endif
96 350
97static unsigned int started, idle, wanted = 4; 351struct eio_pwd
98
99/* worker threads management */
100static mutex_t wrklock = X_MUTEX_INIT;
101
102typedef struct worker
103{ 352{
104 /* locked by wrklock */ 353#if HAVE_AT
105 struct worker *prev, *next; 354 int fd;
106 355#endif
107 thread_t tid;
108
109 /* locked by reslock, reqlock or wrklock */
110 eio_req *req; /* currently processed request */
111 void *dbuf;
112 DIR *dirp;
113} worker;
114
115static worker wrk_first = { &wrk_first, &wrk_first, 0 };
116
117static void worker_clear (worker *wrk)
118{
119 if (wrk->dirp)
120 {
121 closedir (wrk->dirp);
122 wrk->dirp = 0;
123 }
124
125 if (wrk->dbuf)
126 {
127 free (wrk->dbuf);
128 wrk->dbuf = 0;
129 }
130}
131
132static void worker_free (worker *wrk)
133{
134 wrk->next->prev = wrk->prev;
135 wrk->prev->next = wrk->next;
136
137 free (wrk);
138}
139
140static volatile unsigned int nreqs, nready, npending;
141static volatile unsigned int max_idle = 4;
142
143static mutex_t reslock = X_MUTEX_INIT;
144static mutex_t reqlock = X_MUTEX_INIT;
145static cond_t reqwait = X_COND_INIT;
146
147unsigned int eio_nreqs (void)
148{
149 return nreqs;
150}
151
152unsigned int eio_nready (void)
153{
154 unsigned int retval;
155
156 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
157 retval = nready;
158 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
159
160 return retval;
161}
162
163unsigned int eio_npending (void)
164{
165 unsigned int retval;
166
167 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
168 retval = npending;
169 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
170
171 return retval;
172}
173
174unsigned int eio_nthreads (void)
175{
176 unsigned int retval;
177
178 if (WORDACCESS_UNSAFE) X_LOCK (reqlock);
179 retval = started;
180 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
181
182 return retval;
183}
184
185/*
186 * a somewhat faster data structure might be nice, but
187 * with 8 priorities this actually needs <20 insns
188 * per shift, the most expensive operation.
189 */
190typedef struct {
191 eio_req *qs[EIO_NUM_PRI], *qe[EIO_NUM_PRI]; /* qstart, qend */
192 int size; 356 int len;
193} reqq; 357 char str[1]; /* actually, a 0-terminated canonical path */
358};
194 359
195static reqq req_queue; 360/*****************************************************************************/
196static reqq res_queue;
197 361
198static int reqq_push (reqq *q, eio_req *req) 362#define ETP_PRI_MIN EIO_PRI_MIN
199{ 363#define ETP_PRI_MAX EIO_PRI_MAX
200 int pri = req->pri;
201 req->next = 0;
202 364
203 if (q->qe[pri]) 365#define ETP_TYPE_QUIT -1
204 { 366#define ETP_TYPE_GROUP EIO_GROUP
205 q->qe[pri]->next = req;
206 q->qe[pri] = req;
207 }
208 else
209 q->qe[pri] = q->qs[pri] = req;
210 367
211 return q->size++; 368static void eio_nop_callback (void) { }
212} 369static void (*eio_want_poll_cb)(void) = eio_nop_callback;
370static void (*eio_done_poll_cb)(void) = eio_nop_callback;
213 371
214static eio_req *reqq_shift (reqq *q) 372#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
215{ 373#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
216 int pri;
217 374
218 if (!q->size) 375struct etp_worker;
219 return 0; 376#define ETP_REQ eio_req
377#define ETP_DESTROY(req) eio_destroy (req)
378static int eio_finish (eio_req *req);
379#define ETP_FINISH(req) eio_finish (req)
380static void eio_execute (struct etp_worker *self, eio_req *req);
381#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
220 382
221 --q->size; 383#include "etp.c"
222 384
223 for (pri = EIO_NUM_PRI; pri--; ) 385static struct etp_pool eio_pool;
224 { 386#define EIO_POOL (&eio_pool)
225 eio_req *req = q->qs[pri];
226 387
227 if (req) 388/*****************************************************************************/
228 {
229 if (!(q->qs[pri] = (eio_req *)req->next))
230 q->qe[pri] = 0;
231 389
232 return req; 390static void
233 }
234 }
235
236 abort ();
237}
238
239static void grp_try_feed (eio_req *grp) 391grp_try_feed (eio_req *grp)
240{ 392{
241 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 393 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
242 { 394 {
243 int old_len = grp->size; 395 grp->flags &= ~ETP_FLAG_GROUPADD;
244 396
245 EIO_FEED (grp); 397 EIO_FEED (grp);
246 398
247 /* stop if no progress has been made */ 399 /* stop if no progress has been made */
248 if (old_len == grp->size) 400 if (!(grp->flags & ETP_FLAG_GROUPADD))
249 { 401 {
250 grp->feed = 0; 402 grp->feed = 0;
251 break; 403 break;
252 } 404 }
253 } 405 }
254} 406}
255 407
256static int eio_finish (eio_req *req); 408static int
257
258static int grp_dec (eio_req *grp) 409grp_dec (eio_req *grp)
259{ 410{
260 --grp->size; 411 --grp->size;
261 412
262 /* call feeder, if applicable */ 413 /* call feeder, if applicable */
263 grp_try_feed (grp); 414 grp_try_feed (grp);
264 415
265 /* finish, if done */ 416 /* finish, if done */
266 if (!grp->size && grp->int1) 417 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
267 return eio_finish (grp); 418 return eio_finish (grp);
268 else 419 else
269 return 0; 420 return 0;
270} 421}
271 422
423static void
272void eio_destroy (eio_req *req) 424eio_destroy (eio_req *req)
273{ 425{
274 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 426 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 427 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276 428
277 EIO_DESTROY (req); 429 EIO_DESTROY (req);
278} 430}
279 431
432static int
280static int eio_finish (eio_req *req) 433eio_finish (eio_req *req)
281{ 434{
282 int res = EIO_FINISH (req); 435 int res = EIO_FINISH (req);
283 436
284 if (req->grp) 437 if (req->grp)
285 { 438 {
… …
293 if (grp->grp_first == req) 446 if (grp->grp_first == req)
294 grp->grp_first = req->grp_next; 447 grp->grp_first = req->grp_next;
295 448
296 res2 = grp_dec (grp); 449 res2 = grp_dec (grp);
297 450
298 if (!res && res2) 451 if (!res)
299 res = res2; 452 res = res2;
300 } 453 }
301 454
302 eio_destroy (req); 455 eio_destroy (req);
303 456
304 return res; 457 return res;
305} 458}
306 459
460void
307void eio_grp_cancel (eio_req *grp) 461eio_grp_cancel (eio_req *grp)
308{ 462{
309 for (grp = grp->grp_first; grp; grp = grp->grp_next) 463 etp_grp_cancel (EIO_POOL, grp);
310 eio_cancel (grp);
311} 464}
312 465
466void
313void eio_cancel (eio_req *req) 467eio_cancel (eio_req *req)
314{ 468{
315 X_LOCK (wrklock); 469 etp_cancel (EIO_POOL, req);
316 req->flags |= EIO_FLAG_CANCELLED;
317 X_UNLOCK (wrklock);
318
319 eio_grp_cancel (req);
320} 470}
321 471
322X_THREAD_PROC (eio_proc); 472void
323
324static void start_thread (void)
325{
326 worker *wrk = calloc (1, sizeof (worker));
327
328 /*TODO*/
329 assert (("unable to allocate worker thread data", wrk));
330
331 X_LOCK (wrklock);
332
333 if (thread_create (&wrk->tid, eio_proc, (void *)wrk))
334 {
335 wrk->prev = &wrk_first;
336 wrk->next = wrk_first.next;
337 wrk_first.next->prev = wrk;
338 wrk_first.next = wrk;
339 ++started;
340 }
341 else
342 free (wrk);
343
344 X_UNLOCK (wrklock);
345}
346
347static void maybe_start_thread (void)
348{
349 if (eio_nthreads () >= wanted)
350 return;
351
352 /* todo: maybe use idle here, but might be less exact */
353 if (0 <= (int)eio_nthreads () + (int)eio_npending () - (int)eio_nreqs ())
354 return;
355
356 start_thread ();
357}
358
359void eio_submit (eio_req *req) 473eio_submit (eio_req *req)
360{ 474{
361 req->pri += EIO_PRI_BIAS; 475 etp_submit (EIO_POOL, req);
362
363 if (req->pri < EIO_PRI_MIN + EIO_PRI_BIAS) req->pri = EIO_PRI_MIN + EIO_PRI_BIAS;
364 if (req->pri > EIO_PRI_MAX + EIO_PRI_BIAS) req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
365
366 ++nreqs;
367
368 X_LOCK (reqlock);
369 ++nready;
370 reqq_push (&req_queue, req);
371 X_COND_SIGNAL (reqwait);
372 X_UNLOCK (reqlock);
373
374 maybe_start_thread ();
375} 476}
376 477
377static void end_thread (void) 478unsigned int
479eio_nreqs (void)
378{ 480{
379 eio_req *req = calloc (1, sizeof (eio_req)); 481 return etp_nreqs (EIO_POOL);
380
381 req->type = EIO_QUIT;
382 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
383
384 X_LOCK (reqlock);
385 reqq_push (&req_queue, req);
386 X_COND_SIGNAL (reqwait);
387 X_UNLOCK (reqlock);
388
389 X_LOCK (wrklock);
390 --started;
391 X_UNLOCK (wrklock);
392} 482}
393 483
484unsigned int
485eio_nready (void)
486{
487 return etp_nready (EIO_POOL);
488}
489
490unsigned int
491eio_npending (void)
492{
493 return etp_npending (EIO_POOL);
494}
495
496unsigned int ecb_cold
497eio_nthreads (void)
498{
499 return etp_nthreads (EIO_POOL);
500}
501
502void ecb_cold
394void eio_set_max_poll_time (double nseconds) 503eio_set_max_poll_time (double nseconds)
395{ 504{
396 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 505 etp_set_max_poll_time (EIO_POOL, nseconds);
397 max_poll_time = nseconds;
398 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
399} 506}
400 507
508void ecb_cold
401void eio_set_max_poll_reqs (unsigned int maxreqs) 509eio_set_max_poll_reqs (unsigned int maxreqs)
402{ 510{
403 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 511 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
404 max_poll_reqs = maxreqs;
405 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
406} 512}
407 513
514void ecb_cold
408void eio_set_max_idle (unsigned int nthreads) 515eio_set_max_idle (unsigned int nthreads)
409{ 516{
410 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 517 etp_set_max_idle (EIO_POOL, nthreads);
411 max_idle = nthreads <= 0 ? 1 : nthreads;
412 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
413} 518}
414 519
520void ecb_cold
521eio_set_idle_timeout (unsigned int seconds)
522{
523 etp_set_idle_timeout (EIO_POOL, seconds);
524}
525
526void ecb_cold
415void eio_set_min_parallel (unsigned int nthreads) 527eio_set_min_parallel (unsigned int nthreads)
416{ 528{
417 if (wanted < nthreads) 529 etp_set_min_parallel (EIO_POOL, nthreads);
418 wanted = nthreads;
419} 530}
420 531
532void ecb_cold
421void eio_set_max_parallel (unsigned int nthreads) 533eio_set_max_parallel (unsigned int nthreads)
422{ 534{
423 if (wanted > nthreads) 535 etp_set_max_parallel (EIO_POOL, nthreads);
424 wanted = nthreads;
425
426 while (started > wanted)
427 end_thread ();
428} 536}
429 537
430int eio_poll (void) 538int eio_poll (void)
431{ 539{
432 int maxreqs = max_poll_reqs; 540 return etp_poll (EIO_POOL);
433 struct timeval tv_start, tv_now;
434 eio_req *req;
435
436 if (max_poll_time)
437 gettimeofday (&tv_start, 0);
438
439 for (;;)
440 {
441 maybe_start_thread ();
442
443 X_LOCK (reslock);
444 req = reqq_shift (&res_queue);
445
446 if (req)
447 {
448 --npending;
449
450 if (!res_queue.size && done_poll_cb)
451 done_poll_cb ();
452 }
453
454 X_UNLOCK (reslock);
455
456 if (!req)
457 return 0;
458
459 --nreqs;
460
461 if (req->type == EIO_GROUP && req->size)
462 {
463 req->int1 = 1; /* mark request as delayed */
464 continue;
465 }
466 else
467 {
468 int res = eio_finish (req);
469 if (res)
470 return res;
471 }
472
473 if (maxreqs && !--maxreqs)
474 break;
475
476 if (max_poll_time)
477 {
478 gettimeofday (&tv_now, 0);
479
480 if (tvdiff (&tv_start, &tv_now) >= max_poll_time)
481 break;
482 }
483 }
484
485 errno = EAGAIN;
486 return -1;
487} 541}
488 542
489/*****************************************************************************/ 543/*****************************************************************************/
490/* work around various missing functions */ 544/* work around various missing functions */
491 545
492#if !HAVE_PREADWRITE
493# define pread eio__pread
494# define pwrite eio__pwrite
495
496/*
497 * make our pread/pwrite safe against themselves, but not against
498 * normal read/write by using a mutex. slows down execution a lot,
499 * but that's your problem, not mine.
500 */
501static mutex_t preadwritelock = X_MUTEX_INIT;
502
503static ssize_t
504eio__pread (int fd, void *buf, size_t count, off_t offset)
505{
506 ssize_t res;
507 off_t ooffset;
508
509 X_LOCK (preadwritelock);
510 ooffset = lseek (fd, 0, SEEK_CUR);
511 lseek (fd, offset, SEEK_SET);
512 res = read (fd, buf, count);
513 lseek (fd, ooffset, SEEK_SET);
514 X_UNLOCK (preadwritelock);
515
516 return res;
517}
518
519static ssize_t
520eio__pwrite (int fd, void *buf, size_t count, off_t offset)
521{
522 ssize_t res;
523 off_t ooffset;
524
525 X_LOCK (preadwritelock);
526 ooffset = lseek (fd, 0, SEEK_CUR);
527 lseek (fd, offset, SEEK_SET);
528 res = write (fd, buf, count);
529 lseek (fd, offset, SEEK_SET);
530 X_UNLOCK (preadwritelock);
531
532 return res;
533}
534#endif
535
536#ifndef HAVE_FUTIMES 546#ifndef HAVE_UTIMES
537 547
548# undef utimes
538# define utimes(path,times) eio__utimes (path, times) 549# define utimes(path,times) eio__utimes (path, times)
539# define futimes(fd,times) eio__futimes (fd, times)
540 550
541static int 551static int
542eio__utimes (const char *filename, const struct timeval times[2]) 552eio__utimes (const char *filename, const struct timeval times[2])
543{ 553{
544 if (times) 554 if (times)
… …
552 } 562 }
553 else 563 else
554 return utime (filename, 0); 564 return utime (filename, 0);
555} 565}
556 566
567#endif
568
569#ifndef HAVE_FUTIMES
570
571# undef futimes
572# define futimes(fd,times) eio__futimes (fd, times)
573
574static int
557static int eio__futimes (int fd, const struct timeval tv[2]) 575eio__futimes (int fd, const struct timeval tv[2])
558{ 576{
559 errno = ENOSYS; 577 errno = ENOSYS;
560 return -1; 578 return -1;
561} 579}
562 580
563#endif 581#endif
564 582
565#if !HAVE_FDATASYNC 583#if !HAVE_FDATASYNC
584# undef fdatasync
566# define fdatasync fsync 585# define fdatasync(fd) fsync (fd)
567#endif 586#endif
587
588static int
589eio__syncfs (int fd)
590{
591 int res;
592
593#if HAVE_SYS_SYNCFS
594 res = (int)syscall (__NR_syncfs, (int)(fd));
595#else
596 res = EIO_ENOSYS ();
597#endif
598
599 if (res < 0 && errno == ENOSYS && fd >= 0)
600 sync ();
601
602 return res;
603}
604
605/* sync_file_range always needs emulation */
606static int
607eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
608{
609#if HAVE_SYNC_FILE_RANGE
610 int res;
611
612 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
613 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
614 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
615 {
616 flags = 0
617 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
618 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
619 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
620 }
621
622 res = sync_file_range (fd, offset, nbytes, flags);
623
624 if (!res || errno != ENOSYS)
625 return res;
626#endif
627
628 /* even though we could play tricks with the flags, it's better to always
629 * call fdatasync, as that matches the expectation of its users best */
630 return fdatasync (fd);
631}
632
633static int
634eio__fallocate (int fd, int mode, off_t offset, size_t len)
635{
636#if HAVE_LINUX_FALLOCATE
637 return fallocate (fd, mode, offset, len);
638#else
639 return EIO_ENOSYS ();
640#endif
641}
568 642
569#if !HAVE_READAHEAD 643#if !HAVE_READAHEAD
644# undef readahead
570# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 645# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
571 646
572static ssize_t 647static eio_ssize_t
573eio__readahead (int fd, off_t offset, size_t count, worker *self) 648eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
574{ 649{
575 size_t todo = count; 650 size_t todo = count;
576 dBUF; 651 dBUF;
577 652
578 while (todo > 0) 653 while (todo > 0)
… …
582 pread (fd, eio_buf, len, offset); 657 pread (fd, eio_buf, len, offset);
583 offset += len; 658 offset += len;
584 todo -= len; 659 todo -= len;
585 } 660 }
586 661
587 errno = 0; 662 FUBd;
663
664 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
665 /* but not for e.g. EIO or eof, so we also never fail */
588 return count; 666 return 0;
589} 667}
590 668
591#endif
592
593#if !HAVE_READDIR_R
594# define readdir_r eio__readdir_r
595
596static mutex_t readdirlock = X_MUTEX_INIT;
597
598static int
599eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
600{
601 EIO_STRUCT_DIRENT *e;
602 int errorno;
603
604 X_LOCK (readdirlock);
605
606 e = readdir (dirp);
607 errorno = errno;
608
609 if (e)
610 {
611 *res = ent;
612 strcpy (ent->d_name, e->d_name);
613 }
614 else
615 *res = 0;
616
617 X_UNLOCK (readdirlock);
618
619 errno = errorno;
620 return e ? 0 : -1;
621}
622#endif 669#endif
623 670
624/* sendfile always needs emulation */ 671/* sendfile always needs emulation */
625static ssize_t 672static eio_ssize_t
626eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 673eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
627{ 674{
675 eio_ssize_t written = 0;
628 ssize_t res; 676 eio_ssize_t res;
629 677
630 if (!count) 678 if (!count)
631 return 0; 679 return 0;
632 680
681 for (;;)
682 {
683#ifdef __APPLE__
684# undef HAVE_SENDFILE /* broken, as everything on os x */
685#endif
633#if HAVE_SENDFILE 686#if HAVE_SENDFILE
634# if __linux 687# if __linux
688 off_t soffset = offset;
635 res = sendfile (ofd, ifd, &offset, count); 689 res = sendfile (ofd, ifd, &soffset, count);
636 690
637# elif __freebsd 691# elif __FreeBSD__
638 /* 692 /*
639 * Of course, the freebsd sendfile is a dire hack with no thoughts 693 * Of course, the freebsd sendfile is a dire hack with no thoughts
640 * wasted on making it similar to other I/O functions. 694 * wasted on making it similar to other I/O functions.
641 */ 695 */
642 {
643 off_t sbytes; 696 off_t sbytes;
644 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 697 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
645 698
646 if (res < 0 && sbytes) 699 #if 0 /* according to the manpage, this is correct, but broken behaviour */
647 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 700 /* freebsd' sendfile will return 0 on success */
701 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
702 /* not on e.g. EIO or EPIPE - sounds broken */
703 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
648 res = sbytes; 704 res = sbytes;
649 } 705 #endif
706
707 /* according to source inspection, this is correct, and useful behaviour */
708 if (sbytes)
709 res = sbytes;
710
711# elif defined __APPLE__
712 off_t sbytes = count;
713 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
714
715 /* according to the manpage, sbytes is always valid */
716 if (sbytes)
717 res = sbytes;
650 718
651# elif __hpux 719# elif __hpux
652 res = sendfile (ofd, ifd, offset, count, 0, 0); 720 res = sendfile (ofd, ifd, offset, count, 0, 0);
653 721
654# elif __solaris 722# elif __solaris
655 {
656 struct sendfilevec vec; 723 struct sendfilevec vec;
657 size_t sbytes; 724 size_t sbytes;
658 725
659 vec.sfv_fd = ifd; 726 vec.sfv_fd = ifd;
660 vec.sfv_flag = 0; 727 vec.sfv_flag = 0;
661 vec.sfv_off = offset; 728 vec.sfv_off = offset;
662 vec.sfv_len = count; 729 vec.sfv_len = count;
663 730
664 res = sendfilev (ofd, &vec, 1, &sbytes); 731 res = sendfilev (ofd, &vec, 1, &sbytes);
665 732
666 if (res < 0 && sbytes) 733 if (res < 0 && sbytes)
667 res = sbytes; 734 res = sbytes;
668 }
669 735
670# endif 736# endif
737
738#elif defined (_WIN32) && 0
739 /* does not work, just for documentation of what would need to be done */
740 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
741 /* libeio guarantees that the file offset does not change, and windows */
742 /* has no way to get an independent handle to the same file description */
743 HANDLE h = TO_SOCKET (ifd);
744 SetFilePointer (h, offset, 0, FILE_BEGIN);
745 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
746
671#else 747#else
672 res = -1; 748 res = EIO_ENOSYS ();
673 errno = ENOSYS;
674#endif 749#endif
675 750
751 /* we assume sendfile can copy at least 128mb in one go */
752 if (res <= 128 * 1024 * 1024)
753 {
754 if (res > 0)
755 written += res;
756
757 if (written)
758 return written;
759
760 break;
761 }
762 else
763 {
764 /* if we requested more, then probably the kernel was lazy */
765 written += res;
766 offset += res;
767 count -= res;
768
769 if (!count)
770 return written;
771 }
772 }
773
676 if (res < 0 774 if (res < 0
677 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 775 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
776 /* BSDs */
777#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
778 || errno == ENOTSUP
779#endif
780#ifdef EOPNOTSUPP /* windows */
781 || errno == EOPNOTSUPP /* BSDs */
782#endif
678#if __solaris 783#if __solaris
679 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 784 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
680#endif 785#endif
681 ) 786 )
682 ) 787 )
… …
686 791
687 res = 0; 792 res = 0;
688 793
689 while (count) 794 while (count)
690 { 795 {
691 ssize_t cnt; 796 eio_ssize_t cnt;
692 797
693 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 798 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
694 799
695 if (cnt <= 0) 800 if (cnt <= 0)
696 { 801 {
… …
708 813
709 offset += cnt; 814 offset += cnt;
710 res += cnt; 815 res += cnt;
711 count -= cnt; 816 count -= cnt;
712 } 817 }
818
819 FUBd;
713 } 820 }
714 821
715 return res; 822 return res;
823}
824
825#ifdef PAGESIZE
826# define eio_pagesize() PAGESIZE
827#else
828static intptr_t
829eio_pagesize (void)
830{
831 static intptr_t page;
832
833 if (!page)
834 page = sysconf (_SC_PAGESIZE);
835
836 return page;
837}
838#endif
839
840static void
841eio_page_align (void **addr, size_t *length)
842{
843 intptr_t mask = eio_pagesize () - 1;
844
845 /* round down addr */
846 intptr_t adj = mask & (intptr_t)*addr;
847
848 *addr = (void *)((intptr_t)*addr - adj);
849 *length += adj;
850
851 /* round up length */
852 *length = (*length + mask) & ~mask;
853}
854
855#if !_POSIX_MEMLOCK
856# define eio__mlockall(a) EIO_ENOSYS ()
857#else
858
859static int
860eio__mlockall (int flags)
861{
862 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
863 extern int mallopt (int, int);
864 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
865 #endif
866
867 if (EIO_MCL_CURRENT != MCL_CURRENT
868 || EIO_MCL_FUTURE != MCL_FUTURE)
869 {
870 flags = 0
871 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
872 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
873 }
874
875 return mlockall (flags);
876}
877#endif
878
879#if !_POSIX_MEMLOCK_RANGE
880# define eio__mlock(a,b) EIO_ENOSYS ()
881#else
882
883static int
884eio__mlock (void *addr, size_t length)
885{
886 eio_page_align (&addr, &length);
887
888 return mlock (addr, length);
889}
890
891#endif
892
893#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
894# define eio__msync(a,b,c) EIO_ENOSYS ()
895#else
896
897static int
898eio__msync (void *mem, size_t len, int flags)
899{
900 eio_page_align (&mem, &len);
901
902 if (EIO_MS_ASYNC != MS_SYNC
903 || EIO_MS_INVALIDATE != MS_INVALIDATE
904 || EIO_MS_SYNC != MS_SYNC)
905 {
906 flags = 0
907 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
908 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
909 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
910 }
911
912 return msync (mem, len, flags);
913}
914
915#endif
916
917static int
918eio__mtouch (eio_req *req)
919{
920 void *mem = req->ptr2;
921 size_t len = req->size;
922 int flags = req->int1;
923
924 eio_page_align (&mem, &len);
925
926 {
927 intptr_t addr = (intptr_t)mem;
928 intptr_t end = addr + len;
929 intptr_t page = eio_pagesize ();
930
931 if (addr < end)
932 if (flags & EIO_MT_MODIFY) /* modify */
933 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
934 else
935 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
936 }
937
938 return 0;
939}
940
941/*****************************************************************************/
942/* requests implemented outside eio_execute, because they are so large */
943
944static void
945eio__lseek (eio_req *req)
946{
947 /* this usually gets optimised away completely, or your compiler sucks, */
948 /* or the whence constants really are not 0, 1, 2 */
949 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
950 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
951 : req->int2 == EIO_SEEK_END ? SEEK_END
952 : req->int2;
953
954 req->offs = lseek (req->int1, req->offs, whence);
955 req->result = req->offs == (off_t)-1 ? -1 : 0;
956}
957
958/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
959static int
960eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
961{
962 char *res;
963 const char *rel = path;
964 char *tmp1, *tmp2;
965#if SYMLOOP_MAX > 32
966 int symlinks = SYMLOOP_MAX;
967#else
968 int symlinks = 32;
969#endif
970
971 errno = EINVAL;
972 if (!rel)
973 return -1;
974
975 errno = ENOENT;
976 if (!*rel)
977 return -1;
978
979 res = etp_tmpbuf_get (tmpbuf, PATH_MAX * 3);
980#ifdef _WIN32
981 if (_access (rel, 4) != 0)
982 return -1;
983
984 symlinks = GetFullPathName (rel, PATH_MAX * 3, res, 0);
985
986 errno = ENAMETOOLONG;
987 if (symlinks >= PATH_MAX * 3)
988 return -1;
989
990 errno = EIO;
991 if (symlinks <= 0)
992 return -1;
993
994 return symlinks;
995
996#else
997 tmp1 = res + PATH_MAX;
998 tmp2 = tmp1 + PATH_MAX;
999
1000#if 0 /* disabled, the musl way to do things is just too racy */
1001#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1002 /* on linux we may be able to ask the kernel */
1003 {
1004 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1005
1006 if (fd >= 0)
1007 {
1008 sprintf (tmp1, "/proc/self/fd/%d", fd);
1009 req->result = readlink (tmp1, res, PATH_MAX);
1010 /* here we should probably stat the open file and the disk file, to make sure they still match */
1011 close (fd);
1012
1013 if (req->result > 0)
1014 goto done;
1015 }
1016 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1017 return -1;
1018 }
1019#endif
1020#endif
1021
1022 if (*rel != '/')
1023 {
1024 int len;
1025
1026 errno = ENOENT;
1027 if (wd == EIO_INVALID_WD)
1028 return -1;
1029
1030 if (wd == EIO_CWD)
1031 {
1032 if (!getcwd (res, PATH_MAX))
1033 return -1;
1034
1035 len = strlen (res);
1036 }
1037 else
1038 memcpy (res, wd->str, len = wd->len);
1039
1040 if (res [1]) /* only use if not / */
1041 res += len;
1042 }
1043
1044 while (*rel)
1045 {
1046 eio_ssize_t len, linklen;
1047 const char *beg = rel;
1048
1049 while (*rel && *rel != '/')
1050 ++rel;
1051
1052 len = rel - beg;
1053
1054 if (!len) /* skip slashes */
1055 {
1056 ++rel;
1057 continue;
1058 }
1059
1060 if (beg [0] == '.')
1061 {
1062 if (len == 1)
1063 continue; /* . - nop */
1064
1065 if (beg [1] == '.' && len == 2)
1066 {
1067 /* .. - back up one component, if possible */
1068
1069 while (res != tmpbuf->ptr)
1070 if (*--res == '/')
1071 break;
1072
1073 continue;
1074 }
1075 }
1076
1077 errno = ENAMETOOLONG;
1078 if (res + 1 + len + 1 >= tmp1)
1079 return -1;
1080
1081 /* copy one component */
1082 *res = '/';
1083 memcpy (res + 1, beg, len);
1084
1085 /* zero-terminate, for readlink */
1086 res [len + 1] = 0;
1087
1088 /* now check if it's a symlink */
1089 linklen = readlink (tmpbuf->ptr, tmp1, PATH_MAX);
1090
1091 if (linklen < 0)
1092 {
1093 if (errno != EINVAL)
1094 return -1;
1095
1096 /* it's a normal directory. hopefully */
1097 res += len + 1;
1098 }
1099 else
1100 {
1101 /* yay, it was a symlink - build new path in tmp2 */
1102 int rellen = strlen (rel);
1103
1104 errno = ENAMETOOLONG;
1105 if (linklen + 1 + rellen >= PATH_MAX)
1106 return -1;
1107
1108 errno = ELOOP;
1109 if (!--symlinks)
1110 return -1;
1111
1112 if (*tmp1 == '/')
1113 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1114
1115 /* we need to be careful, as rel might point into tmp2 already */
1116 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1117 tmp2 [linklen] = '/';
1118 memcpy (tmp2, tmp1, linklen);
1119
1120 rel = tmp2;
1121 }
1122 }
1123
1124 /* special case for the lone root path */
1125 if (res == tmpbuf->ptr)
1126 *res++ = '/';
1127
1128 return res - (char *)tmpbuf->ptr;
1129#endif
1130}
1131
1132static signed char
1133eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1134{
1135 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1136 : a->inode < b->inode ? -1
1137 : a->inode > b->inode ? 1
1138 : 0;
1139}
1140
1141#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1142
1143#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1144#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1145
1146static void
1147eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1148{
1149 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1150 unsigned char *bit = bits;
1151
1152 assert (CHAR_BIT == 8);
1153 assert (sizeof (eio_dirent) * 8 < 256);
1154 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1155 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1156
1157 if (size <= EIO_SORT_FAST)
1158 return;
1159
1160 /* first prepare an array of bits to test in our radix sort */
1161 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1162 /* inode_bits must contain all inodes ORed together */
1163 /* which is used to skip bits that are 0 everywhere, which is very common */
1164 {
1165 eio_ino_t endianness;
1166 int i, j;
1167
1168 /* we store the byte offset of byte n into byte n of "endianness" */
1169 for (i = 0; i < sizeof (eio_ino_t); ++i)
1170 ((unsigned char *)&endianness)[i] = i;
1171
1172 *bit++ = 0;
1173
1174 for (i = 0; i < sizeof (eio_ino_t); ++i)
1175 {
1176 /* shifting off the byte offsets out of "endianness" */
1177 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1178 endianness >>= 8;
1179
1180 for (j = 0; j < 8; ++j)
1181 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1182 *bit++ = offs + j;
1183 }
1184
1185 for (j = 0; j < 8; ++j)
1186 if (score_bits & (1 << j))
1187 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1188 }
1189
1190 /* now actually do the sorting (a variant of MSD radix sort) */
1191 {
1192 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1193 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1194 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1195 int stk_idx = 0;
1196
1197 base_stk [stk_idx] = dents;
1198 end_stk [stk_idx] = dents + size;
1199 bit_stk [stk_idx] = bit - 1;
1200
1201 do
1202 {
1203 base = base_stk [stk_idx];
1204 end = end_stk [stk_idx];
1205 bit = bit_stk [stk_idx];
1206
1207 for (;;)
1208 {
1209 unsigned char O = *bit >> 3;
1210 unsigned char M = 1 << (*bit & 7);
1211
1212 eio_dirent *a = base;
1213 eio_dirent *b = end;
1214
1215 if (b - a < EIO_SORT_CUTOFF)
1216 break;
1217
1218 /* now bit-partition the array on the bit */
1219 /* this ugly asymmetric loop seems to perform much better than typical */
1220 /* partition algos found in the literature */
1221 do
1222 if (!(((unsigned char *)a)[O] & M))
1223 ++a;
1224 else if (!(((unsigned char *)--b)[O] & M))
1225 {
1226 eio_dirent tmp = *a; *a = *b; *b = tmp;
1227 ++a;
1228 }
1229 while (b > a);
1230
1231 /* next bit, or stop, if no bits left in this path */
1232 if (!*--bit)
1233 break;
1234
1235 base_stk [stk_idx] = a;
1236 end_stk [stk_idx] = end;
1237 bit_stk [stk_idx] = bit;
1238 ++stk_idx;
1239
1240 end = a;
1241 }
1242 }
1243 while (stk_idx--);
1244 }
1245}
1246
1247static void
1248eio_dent_insertion_sort (eio_dirent *dents, int size)
1249{
1250 /* first move the smallest element to the front, to act as a sentinel */
1251 {
1252 int i;
1253 eio_dirent *min = dents;
1254
1255 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1256 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1257 if (EIO_DENT_CMP (dents [i], <, *min))
1258 min = &dents [i];
1259
1260 /* swap elements 0 and j (minimum) */
1261 {
1262 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1263 }
1264 }
1265
1266 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1267 {
1268 eio_dirent *i, *j;
1269
1270 for (i = dents + 1; i < dents + size; ++i)
1271 {
1272 eio_dirent value = *i;
1273
1274 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1275 j [1] = j [0];
1276
1277 j [1] = value;
1278 }
1279 }
1280}
1281
1282static void
1283eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1284{
1285 if (size <= 1)
1286 return; /* our insertion sort relies on size > 0 */
1287
1288 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1289 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1290 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1291
1292 /* use an insertion sort at the end, or for small arrays, */
1293 /* as insertion sort is more efficient for small partitions */
1294 eio_dent_insertion_sort (dents, size);
716} 1295}
717 1296
718/* read a full directory */ 1297/* read a full directory */
719static void 1298static void
720eio__scandir (eio_req *req, worker *self) 1299eio__scandir (eio_req *req, etp_worker *self)
721{ 1300{
1301 char *name, *names;
1302 int namesalloc = 4096 - sizeof (void *) * 4;
1303 int namesoffs = 0;
1304 int flags = req->int1;
1305 eio_dirent *dents = 0;
1306 int dentalloc = 128;
1307 int dentoffs = 0;
1308 eio_ino_t inode_bits = 0;
1309#ifdef _WIN32
1310 HANDLE dirp;
1311 WIN32_FIND_DATA entp;
1312#else
722 DIR *dirp; 1313 DIR *dirp;
723 union
724 {
725 EIO_STRUCT_DIRENT d;
726 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
727 } *u;
728 EIO_STRUCT_DIRENT *entp; 1314 EIO_STRUCT_DIRENT *entp;
729 char *name, *names; 1315#endif
730 int memlen = 4096;
731 int memofs = 0;
732 int res = 0;
733 1316
734 X_LOCK (wrklock); 1317 req->result = -1;
735 self->dirp = dirp = opendir (req->ptr1);
736 self->dbuf = u = malloc (sizeof (*u));
737 req->flags |= EIO_FLAG_PTR2_FREE;
738 req->ptr2 = names = malloc (memlen);
739 X_UNLOCK (wrklock);
740 1318
741 if (dirp && u && names) 1319 if (!(flags & EIO_READDIR_DENTS))
742 for (;;) 1320 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1321
1322#ifdef _WIN32
1323 {
1324 int len = strlen ((const char *)req->ptr1);
1325 char *path = malloc (MAX_PATH);
1326 const char *fmt;
1327 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1328
1329 if (!len)
1330 fmt = "./*";
1331 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1332 fmt = "%s*";
1333 else
1334 fmt = "%s/*";
1335
1336 _snprintf (path, MAX_PATH, fmt, reqpath);
1337 dirp = FindFirstFile (path, &entp);
1338 free (path);
1339
1340 if (dirp == INVALID_HANDLE_VALUE)
1341 {
1342 /* should steal _dosmaperr */
1343 switch (GetLastError ())
1344 {
1345 case ERROR_FILE_NOT_FOUND:
1346 req->result = 0;
1347 break;
1348
1349 case ERROR_INVALID_NAME:
1350 case ERROR_PATH_NOT_FOUND:
1351 case ERROR_NO_MORE_FILES:
1352 errno = ENOENT;
1353 break;
1354
1355 case ERROR_NOT_ENOUGH_MEMORY:
1356 errno = ENOMEM;
1357 break;
1358
1359 default:
1360 errno = EINVAL;
1361 break;
1362 }
1363
1364 return;
1365 }
1366 }
1367#else
1368 #if HAVE_AT
1369 if (req->wd)
743 { 1370 {
1371 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1372
1373 if (fd < 0)
1374 return;
1375
1376 dirp = fdopendir (fd);
1377
1378 if (!dirp)
1379 close (fd);
1380 }
1381 else
1382 dirp = opendir (req->ptr1);
1383 #else
1384 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1385 #endif
1386
1387 if (!dirp)
1388 return;
1389#endif
1390
1391 if (req->flags & EIO_FLAG_PTR1_FREE)
1392 free (req->ptr1);
1393
1394 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1395 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1396 req->ptr2 = names = malloc (namesalloc);
1397
1398 if (!names || (flags && !dents))
1399 return;
1400
1401 for (;;)
1402 {
1403 int done;
1404
1405#ifdef _WIN32
1406 done = !dirp;
1407#else
744 errno = 0; 1408 errno = 0;
745 readdir_r (dirp, &u->d, &entp); 1409 entp = readdir (dirp);
1410 done = !entp;
1411#endif
746 1412
1413 if (done)
1414 {
1415#ifndef _WIN32
1416 int old_errno = errno;
1417 closedir (dirp);
1418 errno = old_errno;
1419
747 if (!entp) 1420 if (errno)
1421 break;
1422#endif
1423
1424 /* sort etc. */
1425 req->int1 = flags;
1426 req->result = dentoffs;
1427
1428 if (flags & EIO_READDIR_STAT_ORDER)
1429 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1430 else if (flags & EIO_READDIR_DIRS_FIRST)
1431 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1432 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1433 else
1434 {
1435 /* in this case, all is known, and we just put dirs first and sort them */
1436 eio_dirent *oth = dents + dentoffs;
1437 eio_dirent *dir = dents;
1438
1439 /* now partition dirs to the front, and non-dirs to the back */
1440 /* by walking from both sides and swapping if necessary */
1441 while (oth > dir)
1442 {
1443 if (dir->type == EIO_DT_DIR)
1444 ++dir;
1445 else if ((--oth)->type == EIO_DT_DIR)
1446 {
1447 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1448
1449 ++dir;
1450 }
1451 }
1452
1453 /* now sort the dirs only (dirs all have the same score) */
1454 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1455 }
1456
748 break; 1457 break;
1458 }
749 1459
750 name = entp->d_name; 1460 /* now add the entry to our list(s) */
1461 name = D_NAME (entp);
751 1462
1463 /* skip . and .. entries */
752 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1464 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
753 { 1465 {
754 int len = strlen (name) + 1; 1466 int len = D_NAMLEN (entp) + 1;
755 1467
756 res++; 1468 while (ecb_expect_false (namesoffs + len > namesalloc))
757
758 while (memofs + len > memlen)
759 { 1469 {
760 memlen *= 2; 1470 namesalloc *= 2;
761 X_LOCK (wrklock);
762 req->ptr2 = names = realloc (names, memlen); 1471 req->ptr2 = names = realloc (names, namesalloc);
763 X_UNLOCK (wrklock);
764 1472
765 if (!names) 1473 if (!names)
1474 break;
1475 }
1476
1477 memcpy (names + namesoffs, name, len);
1478
1479 if (dents)
1480 {
1481 struct eio_dirent *ent;
1482
1483 if (ecb_expect_false (dentoffs == dentalloc))
1484 {
1485 dentalloc *= 2;
1486 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1487
1488 if (!dents)
766 break; 1489 break;
767 } 1490 }
768 1491
769 memcpy (names + memofs, name, len); 1492 ent = dents + dentoffs;
770 memofs += len; 1493
1494 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1495 ent->namelen = len - 1;
1496 ent->inode = D_INO (entp);
1497
1498 inode_bits |= ent->inode;
1499
1500 switch (D_TYPE (entp))
1501 {
1502 default:
1503 ent->type = EIO_DT_UNKNOWN;
1504 flags |= EIO_READDIR_FOUND_UNKNOWN;
1505 break;
1506
1507 #ifdef DT_FIFO
1508 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1509 #endif
1510 #ifdef DT_CHR
1511 case DT_CHR: ent->type = EIO_DT_CHR; break;
1512 #endif
1513 #ifdef DT_MPC
1514 case DT_MPC: ent->type = EIO_DT_MPC; break;
1515 #endif
1516 #ifdef DT_DIR
1517 case DT_DIR: ent->type = EIO_DT_DIR; break;
1518 #endif
1519 #ifdef DT_NAM
1520 case DT_NAM: ent->type = EIO_DT_NAM; break;
1521 #endif
1522 #ifdef DT_BLK
1523 case DT_BLK: ent->type = EIO_DT_BLK; break;
1524 #endif
1525 #ifdef DT_MPB
1526 case DT_MPB: ent->type = EIO_DT_MPB; break;
1527 #endif
1528 #ifdef DT_REG
1529 case DT_REG: ent->type = EIO_DT_REG; break;
1530 #endif
1531 #ifdef DT_NWK
1532 case DT_NWK: ent->type = EIO_DT_NWK; break;
1533 #endif
1534 #ifdef DT_CMP
1535 case DT_CMP: ent->type = EIO_DT_CMP; break;
1536 #endif
1537 #ifdef DT_LNK
1538 case DT_LNK: ent->type = EIO_DT_LNK; break;
1539 #endif
1540 #ifdef DT_SOCK
1541 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1542 #endif
1543 #ifdef DT_DOOR
1544 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1545 #endif
1546 #ifdef DT_WHT
1547 case DT_WHT: ent->type = EIO_DT_WHT; break;
1548 #endif
1549 }
1550
1551 ent->score = 7;
1552
1553 if (flags & EIO_READDIR_DIRS_FIRST)
1554 {
1555 if (ent->type == EIO_DT_UNKNOWN)
1556 {
1557 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1558 ent->score = 1;
1559 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1560 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1561 }
1562 else if (ent->type == EIO_DT_DIR)
1563 ent->score = 0;
1564 }
771 } 1565 }
1566
1567 namesoffs += len;
1568 ++dentoffs;
772 } 1569 }
773 1570
774 if (errno) 1571 if (EIO_CANCELLED (req))
775 res = -1; 1572 {
1573 errno = ECANCELED;
1574 break;
1575 }
1576
1577#ifdef _WIN32
1578 if (!FindNextFile (dirp, &entp))
1579 {
1580 FindClose (dirp);
1581 dirp = 0;
1582 }
1583#endif
1584 }
1585}
1586
1587/*****************************************************************************/
1588/* working directory stuff */
1589/* various deficiencies in the posix 2008 api force us to */
1590/* keep the absolute path in string form at all times */
1591/* fuck yeah. */
1592
1593#if !HAVE_AT
1594
1595/* a bit like realpath, but usually faster because it doesn'T have to return */
1596/* an absolute or canonical path */
1597static const char *
1598wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1599{
1600 if (!wd || *path == '/')
1601 return path;
1602
1603 if (path [0] == '.' && !path [1])
1604 return wd->str;
1605
776 1606 {
777 req->result = res; 1607 int l1 = wd->len;
1608 int l2 = strlen (path);
1609
1610 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1611
1612 memcpy (res, wd->str, l1);
1613 res [l1] = '/';
1614 memcpy (res + l1 + 1, path, l2 + 1);
1615
1616 return res;
1617 }
778} 1618}
1619
1620#endif
1621
1622static eio_wd
1623eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1624{
1625 int fd;
1626 eio_wd res;
1627 int len = eio__realpath (tmpbuf, wd, path);
1628
1629 if (len < 0)
1630 return EIO_INVALID_WD;
1631
1632#if HAVE_AT
1633 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1634
1635 if (fd < 0)
1636 return EIO_INVALID_WD;
1637#endif
1638
1639 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1640
1641#if HAVE_AT
1642 res->fd = fd;
1643#endif
1644
1645 res->len = len;
1646 memcpy (res->str, tmpbuf->ptr, len);
1647 res->str [len] = 0;
1648
1649 return res;
1650}
1651
1652eio_wd
1653eio_wd_open_sync (eio_wd wd, const char *path)
1654{
1655 struct etp_tmpbuf tmpbuf = { };
1656 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1657 free (tmpbuf.ptr);
1658
1659 return wd;
1660}
1661
1662void
1663eio_wd_close_sync (eio_wd wd)
1664{
1665 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1666 {
1667 #if HAVE_AT
1668 close (wd->fd);
1669 #endif
1670 free (wd);
1671 }
1672}
1673
1674#if HAVE_AT
1675
1676/* they forgot these */
1677
1678static int
1679eio__truncateat (int dirfd, const char *path, off_t length)
1680{
1681 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1682 int res;
1683
1684 if (fd < 0)
1685 return fd;
1686
1687 res = ftruncate (fd, length);
1688 close (fd);
1689 return res;
1690}
1691
1692static int
1693eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1694{
1695 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1696 int res;
1697
1698 if (fd < 0)
1699 return fd;
1700
1701 res = fstatvfs (fd, buf);
1702 close (fd);
1703 return res;
1704
1705}
1706
1707#endif
779 1708
780/*****************************************************************************/ 1709/*****************************************************************************/
781 1710
782#define ALLOC(len) \ 1711#define ALLOC(len) \
783 if (!req->ptr2) \ 1712 if (!req->ptr2) \
784 { \ 1713 { \
785 X_LOCK (wrklock); \ 1714 X_LOCK (EIO_POOL->wrklock); \
786 req->flags |= EIO_FLAG_PTR2_FREE; \ 1715 req->flags |= EIO_FLAG_PTR2_FREE; \
787 X_UNLOCK (wrklock); \ 1716 X_UNLOCK (EIO_POOL->wrklock); \
788 req->ptr2 = malloc (len); \ 1717 req->ptr2 = malloc (len); \
789 if (!req->ptr2) \ 1718 if (!req->ptr2) \
790 { \ 1719 { \
791 errno = ENOMEM; \ 1720 errno = ENOMEM; \
792 req->result = -1; \ 1721 req->result = -1; \
793 break; \ 1722 break; \
794 } \ 1723 } \
795 } 1724 }
796 1725
797X_THREAD_PROC (eio_proc)
798{
799 eio_req *req;
800 struct timespec ts;
801 worker *self = (worker *)thr_arg;
802
803 /* try to distribute timeouts somewhat randomly */
804 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
805
806 for (;;)
807 {
808 ts.tv_sec = time (0) + IDLE_TIMEOUT;
809
810 X_LOCK (reqlock);
811
812 for (;;)
813 {
814 self->req = req = reqq_shift (&req_queue);
815
816 if (req)
817 break;
818
819 ++idle;
820
821 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
822 {
823 if (idle > max_idle)
824 {
825 --idle;
826 X_UNLOCK (reqlock);
827 X_LOCK (wrklock);
828 --started;
829 X_UNLOCK (wrklock);
830 goto quit;
831 }
832
833 /* we are allowed to idle, so do so without any timeout */
834 X_COND_WAIT (reqwait, reqlock);
835 ts.tv_sec = time (0) + IDLE_TIMEOUT;
836 }
837
838 --idle;
839 }
840
841 --nready;
842
843 X_UNLOCK (reqlock);
844
845 errno = 0; /* strictly unnecessary */
846
847 if (!EIO_CANCELLED (req))
848 switch (req->type)
849 {
850 case EIO_READ: ALLOC (req->size);
851 req->result = req->offs >= 0
852 ? pread (req->int1, req->ptr2, req->size, req->offs)
853 : read (req->int1, req->ptr2, req->size); break;
854 case EIO_WRITE: req->result = req->offs >= 0
855 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
856 : write (req->int1, req->ptr2, req->size); break;
857
858 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
859 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
860
861 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
862 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
863 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
864 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
865 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
866 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
867
868 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
869 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
870 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
871 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
872 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
873 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
874
875 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
876 case EIO_CLOSE: req->result = close (req->int1); break;
877 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
878 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
879 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
880 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
881 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
882 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
883 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
884 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
885
886 case EIO_READLINK: ALLOC (NAME_MAX);
887 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
888
889 case EIO_SYNC: req->result = 0; sync (); break;
890 case EIO_FSYNC: req->result = fsync (req->int1); break;
891 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
892
893 case EIO_READDIR: eio__scandir (req, self); break;
894
895 case EIO_BUSY:
896#ifdef _WIN32
897 Sleep (req->nv1 * 1000.);
898#else
899 {
900 struct timeval tv;
901
902 tv.tv_sec = req->nv1;
903 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
904
905 req->result = select (0, 0, 0, 0, &tv);
906 }
907#endif
908 break;
909
910 case EIO_UTIME:
911 case EIO_FUTIME:
912 {
913 struct timeval tv[2];
914 struct timeval *times;
915
916 if (req->nv1 != -1. || req->nv2 != -1.)
917 {
918 tv[0].tv_sec = req->nv1;
919 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
920 tv[1].tv_sec = req->nv2;
921 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
922
923 times = tv;
924 }
925 else
926 times = 0;
927
928
929 req->result = req->type == EIO_FUTIME
930 ? futimes (req->int1, times)
931 : utimes (req->ptr1, times);
932 }
933
934 case EIO_GROUP:
935 case EIO_NOP:
936 req->result = 0;
937 break;
938
939 case EIO_QUIT:
940 goto quit;
941
942 default:
943 req->result = -1;
944 break;
945 }
946
947 req->errorno = errno;
948
949 X_LOCK (reslock);
950
951 ++npending;
952
953 if (!reqq_push (&res_queue, req) && want_poll_cb)
954 want_poll_cb ();
955
956 self->req = 0;
957 worker_clear (self);
958
959 X_UNLOCK (reslock);
960 }
961
962quit:
963 X_LOCK (wrklock);
964 worker_free (self);
965 X_UNLOCK (wrklock);
966
967 return 0;
968}
969
970/*****************************************************************************/ 1726/*****************************************************************************/
971 1727
972static void eio_atfork_prepare (void) 1728int ecb_cold
973{
974 X_LOCK (wrklock);
975 X_LOCK (reqlock);
976 X_LOCK (reslock);
977#if !HAVE_PREADWRITE
978 X_LOCK (preadwritelock);
979#endif
980#if !HAVE_READDIR_R
981 X_LOCK (readdirlock);
982#endif
983}
984
985static void eio_atfork_parent (void)
986{
987#if !HAVE_READDIR_R
988 X_UNLOCK (readdirlock);
989#endif
990#if !HAVE_PREADWRITE
991 X_UNLOCK (preadwritelock);
992#endif
993 X_UNLOCK (reslock);
994 X_UNLOCK (reqlock);
995 X_UNLOCK (wrklock);
996}
997
998static void eio_atfork_child (void)
999{
1000 eio_req *prv;
1001
1002 while (prv = reqq_shift (&req_queue))
1003 eio_destroy (prv);
1004
1005 while (prv = reqq_shift (&res_queue))
1006 eio_destroy (prv);
1007
1008 while (wrk_first.next != &wrk_first)
1009 {
1010 worker *wrk = wrk_first.next;
1011
1012 if (wrk->req)
1013 eio_destroy (wrk->req);
1014
1015 worker_clear (wrk);
1016 worker_free (wrk);
1017 }
1018
1019 started = 0;
1020 idle = 0;
1021 nreqs = 0;
1022 nready = 0;
1023 npending = 0;
1024
1025 eio_atfork_parent ();
1026}
1027
1028int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1729eio_init (void (*want_poll)(void), void (*done_poll)(void))
1029{ 1730{
1030 want_poll_cb = want_poll; 1731 eio_want_poll_cb = want_poll;
1031 done_poll_cb = done_poll; 1732 eio_done_poll_cb = done_poll;
1032 1733
1033#ifdef _WIN32 1734 return etp_init (EIO_POOL, 0, 0, 0);
1034 X_MUTEX_CHECK (wrklock);
1035 X_MUTEX_CHECK (reslock);
1036 X_MUTEX_CHECK (reqlock);
1037 X_MUTEX_CHECK (reqwait);
1038 X_MUTEX_CHECK (preadwritelock);
1039 X_MUTEX_CHECK (readdirlock);
1040
1041 X_COND_CHECK (reqwait);
1042#endif
1043
1044 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1045} 1735}
1046 1736
1737ecb_inline void
1047static void eio_api_destroy (eio_req *req) 1738eio_api_destroy (eio_req *req)
1048{ 1739{
1049 free (req); 1740 free (req);
1050} 1741}
1051 1742
1052#define REQ(rtype) \ 1743#define REQ(rtype) \
1053 eio_req *req; \ 1744 eio_req *req; \
1054 \ 1745 \
1055 req = (eio_req *)calloc (1, sizeof *req); \ 1746 req = (eio_req *)calloc (1, sizeof *req); \
1056 if (!req) \ 1747 if (!req) \
1057 return 0; \ 1748 return 0; \
… …
1071 { \ 1762 { \
1072 eio_api_destroy (req); \ 1763 eio_api_destroy (req); \
1073 return 0; \ 1764 return 0; \
1074 } 1765 }
1075 1766
1767#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1768
1769static void
1770eio_execute (etp_worker *self, eio_req *req)
1771{
1772#if HAVE_AT
1773 int dirfd;
1774#else
1775 const char *path;
1776#endif
1777
1778 if (ecb_expect_false (EIO_CANCELLED (req)))
1779 {
1780 req->result = -1;
1781 req->errorno = ECANCELED;
1782 return;
1783 }
1784
1785 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1786 {
1787 req->result = -1;
1788 req->errorno = ENOENT;
1789 return;
1790 }
1791
1792 if (req->type >= EIO_OPEN)
1793 {
1794 #if HAVE_AT
1795 dirfd = WD2FD (req->wd);
1796 #else
1797 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1798 #endif
1799 }
1800
1801 switch (req->type)
1802 {
1803 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1804 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1805 break;
1806 case EIO_WD_CLOSE: req->result = 0;
1807 eio_wd_close_sync (req->wd); break;
1808
1809 case EIO_SEEK: eio__lseek (req); break;
1810 case EIO_READ: ALLOC (req->size);
1811 req->result = req->offs >= 0
1812 ? pread (req->int1, req->ptr2, req->size, req->offs)
1813 : read (req->int1, req->ptr2, req->size); break;
1814 case EIO_WRITE: req->result = req->offs >= 0
1815 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1816 : write (req->int1, req->ptr2, req->size); break;
1817
1818 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1819 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1820
1821 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1822 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1823
1824#if HAVE_AT
1825
1826 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1827 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1828 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1829 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1830 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1831 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1832 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1833 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1834
1835 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1836 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1837 req->result = req->wd && SINGLEDOT (req->ptr1)
1838 ? rmdir (req->wd->str)
1839 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1840 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1841 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1842 req->result = req->wd && SINGLEDOT (req->ptr1)
1843 ? rename (req->wd->str, req->ptr2)
1844 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1845 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1846 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1847 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1848 case EIO_READLINK: ALLOC (PATH_MAX);
1849 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, PATH_MAX); break;
1850 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1851 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1852 case EIO_UTIME:
1853 case EIO_FUTIME:
1854 {
1855 struct timespec ts[2];
1856 struct timespec *times;
1857
1858 if (req->nv1 != -1. || req->nv2 != -1.)
1859 {
1860 ts[0].tv_sec = req->nv1;
1861 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1862 ts[1].tv_sec = req->nv2;
1863 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1864
1865 times = ts;
1866 }
1867 else
1868 times = 0;
1869
1870 req->result = req->type == EIO_FUTIME
1871 ? futimens (req->int1, times)
1872 : utimensat (dirfd, req->ptr1, times, 0);
1873 }
1874 break;
1875
1876#else
1877
1878 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1879 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1880 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1881 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1882 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1883 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1884 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1885 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1886
1887 case EIO_UNLINK: req->result = unlink (path ); break;
1888 case EIO_RMDIR: req->result = rmdir (path ); break;
1889 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1890 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1891 case EIO_LINK: req->result = link (path , req->ptr2); break;
1892 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1893 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1894 case EIO_READLINK: ALLOC (PATH_MAX);
1895 req->result = readlink (path, req->ptr2, PATH_MAX); break;
1896 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1897 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1898
1899 case EIO_UTIME:
1900 case EIO_FUTIME:
1901 {
1902 struct timeval tv[2];
1903 struct timeval *times;
1904
1905 if (req->nv1 != -1. || req->nv2 != -1.)
1906 {
1907 tv[0].tv_sec = req->nv1;
1908 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1909 tv[1].tv_sec = req->nv2;
1910 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1911
1912 times = tv;
1913 }
1914 else
1915 times = 0;
1916
1917 req->result = req->type == EIO_FUTIME
1918 ? futimes (req->int1, times)
1919 : utimes (req->ptr1, times);
1920 }
1921 break;
1922
1923#endif
1924
1925 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1926 {
1927 ALLOC (req->result);
1928 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1929 }
1930 break;
1931
1932 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1933 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1934
1935 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1936 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1937
1938 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1939 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1940 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1941
1942 case EIO_CLOSE: req->result = close (req->int1); break;
1943 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1944 case EIO_SYNC: req->result = 0; sync (); break;
1945 case EIO_FSYNC: req->result = fsync (req->int1); break;
1946 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1947 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1948 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1949 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1950 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1951 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1952 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1953 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1954
1955 case EIO_READDIR: eio__scandir (req, self); break;
1956
1957 case EIO_BUSY:
1958#ifdef _WIN32
1959 Sleep (req->nv1 * 1e3);
1960#else
1961 {
1962 struct timeval tv;
1963
1964 tv.tv_sec = req->nv1;
1965 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1966
1967 req->result = select (0, 0, 0, 0, &tv);
1968 }
1969#endif
1970 break;
1971
1972#if 0
1973 case EIO_GROUP:
1974 abort (); /* handled in eio_request */
1975#endif
1976
1977 case EIO_NOP:
1978 req->result = 0;
1979 break;
1980
1981 case EIO_CUSTOM:
1982 req->feed (req);
1983 break;
1984
1985 default:
1986 req->result = EIO_ENOSYS ();
1987 break;
1988 }
1989
1990 req->errorno = errno;
1991}
1992
1076#ifndef EIO_NO_WRAPPERS 1993#ifndef EIO_NO_WRAPPERS
1077 1994
1995eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
1996{
1997 REQ (EIO_WD_OPEN); PATH; SEND;
1998}
1999
2000eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2001{
2002 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2003}
2004
1078eio_req *eio_nop (int pri, eio_cb cb, void *data) 2005eio_req *eio_nop (int pri, eio_cb cb, void *data)
1079{ 2006{
1080 REQ (EIO_NOP); SEND; 2007 REQ (EIO_NOP); SEND;
1081} 2008}
1082 2009
… …
1093eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data) 2020eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1094{ 2021{
1095 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2022 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1096} 2023}
1097 2024
2025eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2026{
2027 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2028}
2029
1098eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data) 2030eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1099{ 2031{
1100 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2032 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1101} 2033}
1102 2034
2035eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2036{
2037 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2038}
2039
2040eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2041{
2042 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2043}
2044
2045eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2046{
2047 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2048}
2049
2050eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2051{
2052 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2053}
2054
2055eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2056{
2057 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2058}
2059
2060eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2061{
2062 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2063}
2064
1103eio_req *eio_close (int fd, int pri, eio_cb cb, void *data) 2065eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1104{ 2066{
1105 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2067 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1106} 2068}
1107 2069
1108eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data) 2070eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1109{ 2071{
1110 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2072 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1111} 2073}
1112 2074
2075eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2076{
2077 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2078}
2079
1113eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2080eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1114{ 2081{
1115 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2082 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1116} 2083}
1117 2084
1118eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data) 2085eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1119{ 2086{
1120 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND; 2087 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1121} 2088}
1122 2089
2090eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2091{
2092 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2093}
2094
2095eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2096{
2097 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2098}
2099
1123eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data) 2100eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1124{ 2101{
1125 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2102 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1126} 2103}
1127 2104
2105eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2106{
2107 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2108}
2109
1128eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data) 2110eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1129{ 2111{
1130 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2112 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1131} 2113}
1132 2114
… …
1138eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data) 2120eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1139{ 2121{
1140 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2122 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1141} 2123}
1142 2124
1143eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2125eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1144{ 2126{
1145 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2127 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1146} 2128}
1147 2129
1148eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data) 2130eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
… …
1168eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data) 2150eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1169{ 2151{
1170 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2152 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1171} 2153}
1172 2154
1173eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, int pri, eio_cb cb, void *data) 2155eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1174{ 2156{
1175 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2157 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1176} 2158}
1177 2159
1178eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data) 2160eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
… …
1194eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data) 2176eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1195{ 2177{
1196 return eio__1path (EIO_READLINK, path, pri, cb, data); 2178 return eio__1path (EIO_READLINK, path, pri, cb, data);
1197} 2179}
1198 2180
2181eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2182{
2183 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2184}
2185
1199eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data) 2186eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1200{ 2187{
1201 return eio__1path (EIO_STAT, path, pri, cb, data); 2188 return eio__1path (EIO_STAT, path, pri, cb, data);
1202} 2189}
1203 2190
1204eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data) 2191eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1205{ 2192{
1206 return eio__1path (EIO_LSTAT, path, pri, cb, data); 2193 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1207} 2194}
1208 2195
2196eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2197{
2198 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2199}
2200
1209eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data) 2201eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1210{ 2202{
1211 return eio__1path (EIO_UNLINK, path, pri, cb, data); 2203 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1212} 2204}
1213 2205
1214eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data) 2206eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1215{ 2207{
1216 return eio__1path (EIO_RMDIR, path, pri, cb, data); 2208 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1217} 2209}
1218 2210
1219eio_req *eio_readdir (const char *path, int pri, eio_cb cb, void *data) 2211eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1220{ 2212{
1221 return eio__1path (EIO_READDIR, path, pri, cb, data); 2213 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1222} 2214}
1223 2215
1224eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data) 2216eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1225{ 2217{
1226 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2218 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1227} 2219}
1228 2220
1229static eio_req * 2221static eio_req *
1230eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2222eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1231{ 2223{
… …
1255eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data) 2247eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1256{ 2248{
1257 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data); 2249 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
1258} 2250}
1259 2251
2252eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2253{
2254 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2255}
2256
1260#endif 2257#endif
1261 2258
1262eio_req *eio_grp (eio_cb cb, void *data) 2259eio_req *eio_grp (eio_cb cb, void *data)
1263{ 2260{
1264 const int pri = EIO_PRI_MAX; 2261 const int pri = EIO_PRI_MAX;
… …
1271#undef SEND 2268#undef SEND
1272 2269
1273/*****************************************************************************/ 2270/*****************************************************************************/
1274/* grp functions */ 2271/* grp functions */
1275 2272
2273void
1276void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2274eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1277{ 2275{
1278 grp->int2 = limit; 2276 grp->int2 = limit;
1279 grp->feed = feed; 2277 grp->feed = feed;
1280 2278
1281 grp_try_feed (grp); 2279 grp_try_feed (grp);
1282} 2280}
1283 2281
2282void
1284void eio_grp_limit (eio_req *grp, int limit) 2283eio_grp_limit (eio_req *grp, int limit)
1285{ 2284{
1286 grp->int2 = limit; 2285 grp->int2 = limit;
1287 2286
1288 grp_try_feed (grp); 2287 grp_try_feed (grp);
1289} 2288}
1290 2289
2290void
1291void eio_grp_add (eio_req *grp, eio_req *req) 2291eio_grp_add (eio_req *grp, eio_req *req)
1292{ 2292{
1293 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2293 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2294
2295 grp->flags |= ETP_FLAG_GROUPADD;
1294 2296
1295 ++grp->size; 2297 ++grp->size;
1296 req->grp = grp; 2298 req->grp = grp;
1297 2299
1298 req->grp_prev = 0; 2300 req->grp_prev = 0;
… …
1305} 2307}
1306 2308
1307/*****************************************************************************/ 2309/*****************************************************************************/
1308/* misc garbage */ 2310/* misc garbage */
1309 2311
2312eio_ssize_t
1310ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2313eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1311{ 2314{
1312 worker wrk;
1313
1314 wrk.dbuf = 0;
1315
1316 eio__sendfile (ofd, ifd, offset, count, &wrk); 2315 return eio__sendfile (ofd, ifd, offset, count);
1317
1318 if (wrk.dbuf)
1319 free (wrk.dbuf);
1320} 2316}
1321 2317

Diff Legend

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