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

Comparing libeio/eio.c (file contents):
Revision 1.9 by root, Mon May 12 00:31:43 2008 UTC vs.
Revision 1.144 by root, Tue Dec 27 09:58:44 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 0
322#endif
323
324#ifndef EIO_PATH_MIN
325# define EIO_PATH_MIN 8160
326#endif
327
328#define EIO_PATH_MAX (PATH_MAX <= EIO_PATH_MIN ? EIO_PATH_MIN : PATH_MAX)
329
70/* buffer size for various temporary buffers */ 330/* buffer size for various temporary buffers */
71#define EIO_BUFSIZE 65536 331#define EIO_BUFSIZE 65536
72 332
73#define dBUF \ 333#define dBUF \
74 char *eio_buf; \
75 X_LOCK (wrklock); \
76 self->dbuf = eio_buf = malloc (EIO_BUFSIZE); \ 334 char *eio_buf = malloc (EIO_BUFSIZE); \
77 X_UNLOCK (wrklock); \
78 errno = ENOMEM; \ 335 errno = ENOMEM; \
79 if (!eio_buf) \ 336 if (!eio_buf) \
80 return -1; 337 return -1
81 338
82#define EIO_TICKS ((1000000 + 1023) >> 10) 339#define FUBd \
340 free (eio_buf)
83 341
84static void (*want_poll_cb) (void); 342/*****************************************************************************/
85static void (*done_poll_cb) (void);
86 343
87static unsigned int max_poll_time = 0; 344struct etp_tmpbuf;
88static unsigned int max_poll_reqs = 0;
89 345
90/* calculcate time difference in ~1/EIO_TICKS of a second */ 346#if _POSIX_VERSION >= 200809L
91static int tvdiff (struct timeval *tv1, struct timeval *tv2) 347 #define HAVE_AT 1
92{ 348 #define WD2FD(wd) ((wd) ? (wd)->fd : AT_FDCWD)
93 return (tv2->tv_sec - tv1->tv_sec ) * EIO_TICKS 349 #ifndef O_SEARCH
94 + ((tv2->tv_usec - tv1->tv_usec) >> 10); 350 #define O_SEARCH O_RDONLY
95} 351 #endif
352#else
353 #define HAVE_AT 0
354 static const char *wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path);
355#endif
96 356
97static unsigned int started, idle, wanted = 4; 357struct eio_pwd
98
99/* worker threads management */
100static mutex_t wrklock = X_MUTEX_INIT;
101
102typedef struct worker
103{ 358{
104 /* locked by wrklock */ 359#if HAVE_AT
105 struct worker *prev, *next; 360 int fd;
106 361#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; 362 int len;
193} reqq; 363 char str[1]; /* actually, a 0-terminated canonical path */
364};
194 365
195static reqq req_queue; 366/*****************************************************************************/
196static reqq res_queue;
197 367
198static int reqq_push (reqq *q, eio_req *req) 368#define ETP_PRI_MIN EIO_PRI_MIN
199{ 369#define ETP_PRI_MAX EIO_PRI_MAX
200 int pri = req->pri;
201 req->next = 0;
202 370
203 if (q->qe[pri]) 371#define ETP_TYPE_QUIT -1
204 { 372#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;
210 373
211 return q->size++; 374static void eio_nop_callback (void) { }
212} 375static void (*eio_want_poll_cb)(void) = eio_nop_callback;
376static void (*eio_done_poll_cb)(void) = eio_nop_callback;
213 377
214static eio_req *reqq_shift (reqq *q) 378#define ETP_WANT_POLL(pool) eio_want_poll_cb ()
215{ 379#define ETP_DONE_POLL(pool) eio_done_poll_cb ()
216 int pri;
217 380
218 if (!q->size) 381struct etp_worker;
219 return 0; 382#define ETP_REQ eio_req
383#define ETP_DESTROY(req) eio_destroy (req)
384static int eio_finish (eio_req *req);
385#define ETP_FINISH(req) eio_finish (req)
386static void eio_execute (struct etp_worker *self, eio_req *req);
387#define ETP_EXECUTE(wrk,req) eio_execute (wrk, req)
220 388
221 --q->size; 389#include "etp.c"
222 390
223 for (pri = EIO_NUM_PRI; pri--; ) 391static struct etp_pool eio_pool;
224 { 392#define EIO_POOL (&eio_pool)
225 eio_req *req = q->qs[pri];
226 393
227 if (req) 394/*****************************************************************************/
228 {
229 if (!(q->qs[pri] = (eio_req *)req->next))
230 q->qe[pri] = 0;
231 395
232 return req; 396static void
233 }
234 }
235
236 abort ();
237}
238
239static void grp_try_feed (eio_req *grp) 397grp_try_feed (eio_req *grp)
240{ 398{
241 while (grp->size < grp->int2 && !EIO_CANCELLED (grp)) 399 while (grp->size < grp->int2 && !EIO_CANCELLED (grp))
242 { 400 {
243 int old_len = grp->size; 401 grp->flags &= ~ETP_FLAG_GROUPADD;
244 402
245 EIO_FEED (grp); 403 EIO_FEED (grp);
246 404
247 /* stop if no progress has been made */ 405 /* stop if no progress has been made */
248 if (old_len == grp->size) 406 if (!(grp->flags & ETP_FLAG_GROUPADD))
249 { 407 {
250 grp->feed = 0; 408 grp->feed = 0;
251 break; 409 break;
252 } 410 }
253 } 411 }
254} 412}
255 413
256static int eio_finish (eio_req *req); 414static int
257
258static int grp_dec (eio_req *grp) 415grp_dec (eio_req *grp)
259{ 416{
260 --grp->size; 417 --grp->size;
261 418
262 /* call feeder, if applicable */ 419 /* call feeder, if applicable */
263 grp_try_feed (grp); 420 grp_try_feed (grp);
264 421
265 /* finish, if done */ 422 /* finish, if done */
266 if (!grp->size && grp->int1) 423 if (!grp->size && grp->flags & ETP_FLAG_DELAYED)
267 return eio_finish (grp); 424 return eio_finish (grp);
268 else 425 else
269 return 0; 426 return 0;
270} 427}
271 428
429static void
272void eio_destroy (eio_req *req) 430eio_destroy (eio_req *req)
273{ 431{
274 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1); 432 if ((req)->flags & EIO_FLAG_PTR1_FREE) free (req->ptr1);
275 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2); 433 if ((req)->flags & EIO_FLAG_PTR2_FREE) free (req->ptr2);
276 434
277 EIO_DESTROY (req); 435 EIO_DESTROY (req);
278} 436}
279 437
438static int
280static int eio_finish (eio_req *req) 439eio_finish (eio_req *req)
281{ 440{
282 int res = EIO_FINISH (req); 441 int res = EIO_FINISH (req);
283 442
284 if (req->grp) 443 if (req->grp)
285 { 444 {
… …
293 if (grp->grp_first == req) 452 if (grp->grp_first == req)
294 grp->grp_first = req->grp_next; 453 grp->grp_first = req->grp_next;
295 454
296 res2 = grp_dec (grp); 455 res2 = grp_dec (grp);
297 456
298 if (!res && res2) 457 if (!res)
299 res = res2; 458 res = res2;
300 } 459 }
301 460
302 eio_destroy (req); 461 eio_destroy (req);
303 462
304 return res; 463 return res;
305} 464}
306 465
466void
307void eio_grp_cancel (eio_req *grp) 467eio_grp_cancel (eio_req *grp)
308{ 468{
309 for (grp = grp->grp_first; grp; grp = grp->grp_next) 469 etp_grp_cancel (EIO_POOL, grp);
310 eio_cancel (grp);
311} 470}
312 471
472void
313void eio_cancel (eio_req *req) 473eio_cancel (eio_req *req)
314{ 474{
315 X_LOCK (wrklock); 475 etp_cancel (EIO_POOL, req);
316 req->flags |= EIO_FLAG_CANCELLED;
317 X_UNLOCK (wrklock);
318
319 eio_grp_cancel (req);
320} 476}
321 477
322X_THREAD_PROC (eio_proc); 478void
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) 479eio_submit (eio_req *req)
360{ 480{
361 ++nreqs; 481 etp_submit (EIO_POOL, req);
362
363 X_LOCK (reqlock);
364 ++nready;
365 reqq_push (&req_queue, req);
366 X_COND_SIGNAL (reqwait);
367 X_UNLOCK (reqlock);
368
369 maybe_start_thread ();
370} 482}
371 483
372static void end_thread (void) 484unsigned int
485eio_nreqs (void)
373{ 486{
374 eio_req *req = calloc (1, sizeof (eio_req)); 487 return etp_nreqs (EIO_POOL);
375
376 req->type = EIO_QUIT;
377 req->pri = EIO_PRI_MAX + EIO_PRI_BIAS;
378
379 X_LOCK (reqlock);
380 reqq_push (&req_queue, req);
381 X_COND_SIGNAL (reqwait);
382 X_UNLOCK (reqlock);
383
384 X_LOCK (wrklock);
385 --started;
386 X_UNLOCK (wrklock);
387} 488}
388 489
490unsigned int
491eio_nready (void)
492{
493 return etp_nready (EIO_POOL);
494}
495
496unsigned int
497eio_npending (void)
498{
499 return etp_npending (EIO_POOL);
500}
501
502unsigned int ecb_cold
503eio_nthreads (void)
504{
505 return etp_nthreads (EIO_POOL);
506}
507
508void ecb_cold
389void eio_set_max_poll_time (double nseconds) 509eio_set_max_poll_time (double nseconds)
390{ 510{
391 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 511 etp_set_max_poll_time (EIO_POOL, nseconds);
392 max_poll_time = nseconds;
393 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
394} 512}
395 513
514void ecb_cold
396void eio_set_max_poll_reqs (unsigned int maxreqs) 515eio_set_max_poll_reqs (unsigned int maxreqs)
397{ 516{
398 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 517 etp_set_max_poll_reqs (EIO_POOL, maxreqs);
399 max_poll_reqs = maxreqs;
400 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
401} 518}
402 519
520void ecb_cold
403void eio_set_max_idle (unsigned int nthreads) 521eio_set_max_idle (unsigned int nthreads)
404{ 522{
405 if (WORDACCESS_UNSAFE) X_LOCK (reqlock); 523 etp_set_max_idle (EIO_POOL, nthreads);
406 max_idle = nthreads <= 0 ? 1 : nthreads;
407 if (WORDACCESS_UNSAFE) X_UNLOCK (reqlock);
408} 524}
409 525
526void ecb_cold
527eio_set_idle_timeout (unsigned int seconds)
528{
529 etp_set_idle_timeout (EIO_POOL, seconds);
530}
531
532void ecb_cold
410void eio_set_min_parallel (unsigned int nthreads) 533eio_set_min_parallel (unsigned int nthreads)
411{ 534{
412 if (wanted < nthreads) 535 etp_set_min_parallel (EIO_POOL, nthreads);
413 wanted = nthreads;
414} 536}
415 537
538void ecb_cold
416void eio_set_max_parallel (unsigned int nthreads) 539eio_set_max_parallel (unsigned int nthreads)
417{ 540{
418 if (wanted > nthreads) 541 etp_set_max_parallel (EIO_POOL, nthreads);
419 wanted = nthreads;
420
421 while (started > wanted)
422 end_thread ();
423} 542}
424 543
425int eio_poll (void) 544int eio_poll (void)
426{ 545{
427 int maxreqs = max_poll_reqs; 546 return etp_poll (EIO_POOL);
428 struct timeval tv_start, tv_now;
429 eio_req *req;
430
431 if (max_poll_time)
432 gettimeofday (&tv_start, 0);
433
434 for (;;)
435 {
436 maybe_start_thread ();
437
438 X_LOCK (reslock);
439 req = reqq_shift (&res_queue);
440
441 if (req)
442 {
443 --npending;
444
445 if (!res_queue.size && done_poll_cb)
446 done_poll_cb ();
447 }
448
449 X_UNLOCK (reslock);
450
451 if (!req)
452 return 0;
453
454 --nreqs;
455
456 if (req->type == EIO_GROUP && req->size)
457 {
458 req->int1 = 1; /* mark request as delayed */
459 continue;
460 }
461 else
462 {
463 int res = eio_finish (req);
464 if (res)
465 return res;
466 }
467
468 if (maxreqs && !--maxreqs)
469 break;
470
471 if (max_poll_time)
472 {
473 gettimeofday (&tv_now, 0);
474
475 if (tvdiff (&tv_start, &tv_now) >= max_poll_time)
476 break;
477 }
478 }
479
480 errno = EAGAIN;
481 return -1;
482} 547}
483 548
484/*****************************************************************************/ 549/*****************************************************************************/
485/* work around various missing functions */ 550/* work around various missing functions */
486 551
487#if !HAVE_PREADWRITE
488# define pread eio__pread
489# define pwrite eio__pwrite
490
491/*
492 * make our pread/pwrite safe against themselves, but not against
493 * normal read/write by using a mutex. slows down execution a lot,
494 * but that's your problem, not mine.
495 */
496static mutex_t preadwritelock = X_MUTEX_INIT;
497
498static ssize_t
499eio__pread (int fd, void *buf, size_t count, off_t offset)
500{
501 ssize_t res;
502 off_t ooffset;
503
504 X_LOCK (preadwritelock);
505 ooffset = lseek (fd, 0, SEEK_CUR);
506 lseek (fd, offset, SEEK_SET);
507 res = read (fd, buf, count);
508 lseek (fd, ooffset, SEEK_SET);
509 X_UNLOCK (preadwritelock);
510
511 return res;
512}
513
514static ssize_t
515eio__pwrite (int fd, void *buf, size_t count, off_t offset)
516{
517 ssize_t res;
518 off_t ooffset;
519
520 X_LOCK (preadwritelock);
521 ooffset = lseek (fd, 0, SEEK_CUR);
522 lseek (fd, offset, SEEK_SET);
523 res = write (fd, buf, count);
524 lseek (fd, offset, SEEK_SET);
525 X_UNLOCK (preadwritelock);
526
527 return res;
528}
529#endif
530
531#ifndef HAVE_FUTIMES 552#ifndef HAVE_UTIMES
532 553
554# undef utimes
533# define utimes(path,times) eio__utimes (path, times) 555# define utimes(path,times) eio__utimes (path, times)
534# define futimes(fd,times) eio__futimes (fd, times)
535 556
536static int 557static int
537eio__utimes (const char *filename, const struct timeval times[2]) 558eio__utimes (const char *filename, const struct timeval times[2])
538{ 559{
539 if (times) 560 if (times)
… …
547 } 568 }
548 else 569 else
549 return utime (filename, 0); 570 return utime (filename, 0);
550} 571}
551 572
573#endif
574
575#ifndef HAVE_FUTIMES
576
577# undef futimes
578# define futimes(fd,times) eio__futimes (fd, times)
579
580static int
552static int eio__futimes (int fd, const struct timeval tv[2]) 581eio__futimes (int fd, const struct timeval tv[2])
553{ 582{
554 errno = ENOSYS; 583 errno = ENOSYS;
555 return -1; 584 return -1;
556} 585}
557 586
558#endif 587#endif
559 588
560#if !HAVE_FDATASYNC 589#if !HAVE_FDATASYNC
590# undef fdatasync
561# define fdatasync fsync 591# define fdatasync(fd) fsync (fd)
562#endif 592#endif
593
594static int
595eio__syncfs (int fd)
596{
597 int res;
598
599#if HAVE_SYS_SYNCFS
600 res = (int)syscall (__NR_syncfs, (int)(fd));
601#else
602 res = EIO_ENOSYS ();
603#endif
604
605 if (res < 0 && errno == ENOSYS && fd >= 0)
606 sync ();
607
608 return res;
609}
610
611/* sync_file_range always needs emulation */
612static int
613eio__sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags)
614{
615#if HAVE_SYNC_FILE_RANGE
616 int res;
617
618 if (EIO_SYNC_FILE_RANGE_WAIT_BEFORE != SYNC_FILE_RANGE_WAIT_BEFORE
619 || EIO_SYNC_FILE_RANGE_WRITE != SYNC_FILE_RANGE_WRITE
620 || EIO_SYNC_FILE_RANGE_WAIT_AFTER != SYNC_FILE_RANGE_WAIT_AFTER)
621 {
622 flags = 0
623 | (flags & EIO_SYNC_FILE_RANGE_WAIT_BEFORE ? SYNC_FILE_RANGE_WAIT_BEFORE : 0)
624 | (flags & EIO_SYNC_FILE_RANGE_WRITE ? SYNC_FILE_RANGE_WRITE : 0)
625 | (flags & EIO_SYNC_FILE_RANGE_WAIT_AFTER ? SYNC_FILE_RANGE_WAIT_AFTER : 0);
626 }
627
628 res = sync_file_range (fd, offset, nbytes, flags);
629
630 if (!res || errno != ENOSYS)
631 return res;
632#endif
633
634 /* even though we could play tricks with the flags, it's better to always
635 * call fdatasync, as that matches the expectation of its users best */
636 return fdatasync (fd);
637}
638
639static int
640eio__fallocate (int fd, int mode, off_t offset, size_t len)
641{
642#if HAVE_LINUX_FALLOCATE
643 return fallocate (fd, mode, offset, len);
644#else
645 return EIO_ENOSYS ();
646#endif
647}
563 648
564#if !HAVE_READAHEAD 649#if !HAVE_READAHEAD
650# undef readahead
565# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self) 651# define readahead(fd,offset,count) eio__readahead (fd, offset, count, self)
566 652
567static ssize_t 653static eio_ssize_t
568eio__readahead (int fd, off_t offset, size_t count, worker *self) 654eio__readahead (int fd, off_t offset, size_t count, etp_worker *self)
569{ 655{
570 size_t todo = count; 656 size_t todo = count;
571 dBUF; 657 dBUF;
572 658
573 while (todo > 0) 659 while (todo > 0)
… …
577 pread (fd, eio_buf, len, offset); 663 pread (fd, eio_buf, len, offset);
578 offset += len; 664 offset += len;
579 todo -= len; 665 todo -= len;
580 } 666 }
581 667
582 errno = 0; 668 FUBd;
669
670 /* linux's readahead basically only fails for EBADF or EINVAL (not mmappable) */
671 /* but not for e.g. EIO or eof, so we also never fail */
583 return count; 672 return 0;
584} 673}
585 674
586#endif
587
588#if !HAVE_READDIR_R
589# define readdir_r eio__readdir_r
590
591static mutex_t readdirlock = X_MUTEX_INIT;
592
593static int
594eio__readdir_r (DIR *dirp, EIO_STRUCT_DIRENT *ent, EIO_STRUCT_DIRENT **res)
595{
596 EIO_STRUCT_DIRENT *e;
597 int errorno;
598
599 X_LOCK (readdirlock);
600
601 e = readdir (dirp);
602 errorno = errno;
603
604 if (e)
605 {
606 *res = ent;
607 strcpy (ent->d_name, e->d_name);
608 }
609 else
610 *res = 0;
611
612 X_UNLOCK (readdirlock);
613
614 errno = errorno;
615 return e ? 0 : -1;
616}
617#endif 675#endif
618 676
619/* sendfile always needs emulation */ 677/* sendfile always needs emulation */
620static ssize_t 678static eio_ssize_t
621eio__sendfile (int ofd, int ifd, off_t offset, size_t count, worker *self) 679eio__sendfile (int ofd, int ifd, off_t offset, size_t count)
622{ 680{
681 eio_ssize_t written = 0;
623 ssize_t res; 682 eio_ssize_t res;
624 683
625 if (!count) 684 if (!count)
626 return 0; 685 return 0;
627 686
687 for (;;)
688 {
689#ifdef __APPLE__
690# undef HAVE_SENDFILE /* broken, as everything on os x */
691#endif
628#if HAVE_SENDFILE 692#if HAVE_SENDFILE
629# if __linux 693# if __linux
694 off_t soffset = offset;
630 res = sendfile (ofd, ifd, &offset, count); 695 res = sendfile (ofd, ifd, &soffset, count);
631 696
632# elif __freebsd 697# elif __FreeBSD__
633 /* 698 /*
634 * Of course, the freebsd sendfile is a dire hack with no thoughts 699 * Of course, the freebsd sendfile is a dire hack with no thoughts
635 * wasted on making it similar to other I/O functions. 700 * wasted on making it similar to other I/O functions.
636 */ 701 */
637 {
638 off_t sbytes; 702 off_t sbytes;
639 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0); 703 res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
640 704
641 if (res < 0 && sbytes) 705 #if 0 /* according to the manpage, this is correct, but broken behaviour */
642 /* maybe only on EAGAIN: as usual, the manpage leaves you guessing */ 706 /* freebsd' sendfile will return 0 on success */
707 /* freebsd 8 documents it as only setting *sbytes on EINTR and EAGAIN, but */
708 /* not on e.g. EIO or EPIPE - sounds broken */
709 if ((res < 0 && (errno == EAGAIN || errno == EINTR) && sbytes) || res == 0)
643 res = sbytes; 710 res = sbytes;
644 } 711 #endif
712
713 /* according to source inspection, this is correct, and useful behaviour */
714 if (sbytes)
715 res = sbytes;
716
717# elif defined __APPLE__
718 off_t sbytes = count;
719 res = sendfile (ifd, ofd, offset, &sbytes, 0, 0);
720
721 /* according to the manpage, sbytes is always valid */
722 if (sbytes)
723 res = sbytes;
645 724
646# elif __hpux 725# elif __hpux
647 res = sendfile (ofd, ifd, offset, count, 0, 0); 726 res = sendfile (ofd, ifd, offset, count, 0, 0);
648 727
649# elif __solaris 728# elif __solaris
650 {
651 struct sendfilevec vec; 729 struct sendfilevec vec;
652 size_t sbytes; 730 size_t sbytes;
653 731
654 vec.sfv_fd = ifd; 732 vec.sfv_fd = ifd;
655 vec.sfv_flag = 0; 733 vec.sfv_flag = 0;
656 vec.sfv_off = offset; 734 vec.sfv_off = offset;
657 vec.sfv_len = count; 735 vec.sfv_len = count;
658 736
659 res = sendfilev (ofd, &vec, 1, &sbytes); 737 res = sendfilev (ofd, &vec, 1, &sbytes);
660 738
661 if (res < 0 && sbytes) 739 if (res < 0 && sbytes)
662 res = sbytes; 740 res = sbytes;
663 }
664 741
665# endif 742# endif
743
744#elif defined (_WIN32) && 0
745 /* does not work, just for documentation of what would need to be done */
746 /* actually, cannot be done like this, as TransmitFile changes the file offset, */
747 /* libeio guarantees that the file offset does not change, and windows */
748 /* has no way to get an independent handle to the same file description */
749 HANDLE h = TO_SOCKET (ifd);
750 SetFilePointer (h, offset, 0, FILE_BEGIN);
751 res = TransmitFile (TO_SOCKET (ofd), h, count, 0, 0, 0, 0);
752
666#else 753#else
667 res = -1; 754 res = EIO_ENOSYS ();
668 errno = ENOSYS;
669#endif 755#endif
670 756
757 /* we assume sendfile can copy at least 128mb in one go */
758 if (res <= 128 * 1024 * 1024)
759 {
760 if (res > 0)
761 written += res;
762
763 if (written)
764 return written;
765
766 break;
767 }
768 else
769 {
770 /* if we requested more, then probably the kernel was lazy */
771 written += res;
772 offset += res;
773 count -= res;
774
775 if (!count)
776 return written;
777 }
778 }
779
671 if (res < 0 780 if (res < 0
672 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK 781 && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
782 /* BSDs */
783#ifdef ENOTSUP /* sigh, if the steenking pile called openbsd would only try to at least compile posix code... */
784 || errno == ENOTSUP
785#endif
786#ifdef EOPNOTSUPP /* windows */
787 || errno == EOPNOTSUPP /* BSDs */
788#endif
673#if __solaris 789#if __solaris
674 || errno == EAFNOSUPPORT || errno == EPROTOTYPE 790 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
675#endif 791#endif
676 ) 792 )
677 ) 793 )
… …
681 797
682 res = 0; 798 res = 0;
683 799
684 while (count) 800 while (count)
685 { 801 {
686 ssize_t cnt; 802 eio_ssize_t cnt;
687 803
688 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset); 804 cnt = pread (ifd, eio_buf, count > EIO_BUFSIZE ? EIO_BUFSIZE : count, offset);
689 805
690 if (cnt <= 0) 806 if (cnt <= 0)
691 { 807 {
… …
703 819
704 offset += cnt; 820 offset += cnt;
705 res += cnt; 821 res += cnt;
706 count -= cnt; 822 count -= cnt;
707 } 823 }
824
825 FUBd;
708 } 826 }
709 827
710 return res; 828 return res;
829}
830
831#ifdef PAGESIZE
832# define eio_pagesize() PAGESIZE
833#else
834static intptr_t
835eio_pagesize (void)
836{
837 static intptr_t page;
838
839 if (!page)
840 page = sysconf (_SC_PAGESIZE);
841
842 return page;
843}
844#endif
845
846static void
847eio_page_align (void **addr, size_t *length)
848{
849 intptr_t mask = eio_pagesize () - 1;
850
851 /* round down addr */
852 intptr_t adj = mask & (intptr_t)*addr;
853
854 *addr = (void *)((intptr_t)*addr - adj);
855 *length += adj;
856
857 /* round up length */
858 *length = (*length + mask) & ~mask;
859}
860
861#if !_POSIX_MEMLOCK
862# define eio__mlockall(a) EIO_ENOSYS ()
863#else
864
865static int
866eio__mlockall (int flags)
867{
868 #if __GLIBC__ == 2 && __GLIBC_MINOR__ <= 7
869 extern int mallopt (int, int);
870 mallopt (-6, 238); /* http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=473812 */
871 #endif
872
873 if (EIO_MCL_CURRENT != MCL_CURRENT
874 || EIO_MCL_FUTURE != MCL_FUTURE)
875 {
876 flags = 0
877 | (flags & EIO_MCL_CURRENT ? MCL_CURRENT : 0)
878 | (flags & EIO_MCL_FUTURE ? MCL_FUTURE : 0);
879 }
880
881 return mlockall (flags);
882}
883#endif
884
885#if !_POSIX_MEMLOCK_RANGE
886# define eio__mlock(a,b) EIO_ENOSYS ()
887#else
888
889static int
890eio__mlock (void *addr, size_t length)
891{
892 eio_page_align (&addr, &length);
893
894 return mlock (addr, length);
895}
896
897#endif
898
899#if !(_POSIX_MAPPED_FILES && _POSIX_SYNCHRONIZED_IO)
900# define eio__msync(a,b,c) EIO_ENOSYS ()
901#else
902
903static int
904eio__msync (void *mem, size_t len, int flags)
905{
906 eio_page_align (&mem, &len);
907
908 if (EIO_MS_ASYNC != MS_SYNC
909 || EIO_MS_INVALIDATE != MS_INVALIDATE
910 || EIO_MS_SYNC != MS_SYNC)
911 {
912 flags = 0
913 | (flags & EIO_MS_ASYNC ? MS_ASYNC : 0)
914 | (flags & EIO_MS_INVALIDATE ? MS_INVALIDATE : 0)
915 | (flags & EIO_MS_SYNC ? MS_SYNC : 0);
916 }
917
918 return msync (mem, len, flags);
919}
920
921#endif
922
923static int
924eio__mtouch (eio_req *req)
925{
926 void *mem = req->ptr2;
927 size_t len = req->size;
928 int flags = req->int1;
929
930 eio_page_align (&mem, &len);
931
932 {
933 intptr_t addr = (intptr_t)mem;
934 intptr_t end = addr + len;
935 intptr_t page = eio_pagesize ();
936
937 if (addr < end)
938 if (flags & EIO_MT_MODIFY) /* modify */
939 do { *((volatile sig_atomic_t *)addr) |= 0; } while ((addr += page) < len && !EIO_CANCELLED (req));
940 else
941 do { *((volatile sig_atomic_t *)addr) ; } while ((addr += page) < len && !EIO_CANCELLED (req));
942 }
943
944 return 0;
945}
946
947/*****************************************************************************/
948/* requests implemented outside eio_execute, because they are so large */
949
950static void
951eio__lseek (eio_req *req)
952{
953 /* this usually gets optimised away completely, or your compiler sucks, */
954 /* or the whence constants really are not 0, 1, 2 */
955 int whence = req->int2 == EIO_SEEK_SET ? SEEK_SET
956 : req->int2 == EIO_SEEK_CUR ? SEEK_CUR
957 : req->int2 == EIO_SEEK_END ? SEEK_END
958 : req->int2;
959
960 req->offs = lseek (req->int1, req->offs, whence);
961 req->result = req->offs == (off_t)-1 ? -1 : 0;
962}
963
964/* result will always end up in tmpbuf, there is always space for adding a 0-byte */
965static int
966eio__realpath (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
967{
968 char *res;
969 const char *rel = path;
970 char *tmp1, *tmp2;
971#if SYMLOOP_MAX > 32
972 int symlinks = SYMLOOP_MAX;
973#else
974 int symlinks = 32;
975#endif
976
977 errno = EINVAL;
978 if (!rel)
979 return -1;
980
981 errno = ENOENT;
982 if (!*rel)
983 return -1;
984
985 res = etp_tmpbuf_get (tmpbuf, EIO_PATH_MAX * 3);
986#ifdef _WIN32
987 if (_access (rel, 4) != 0)
988 return -1;
989
990 symlinks = GetFullPathName (rel, EIO_PATH_MAX * 3, res, 0);
991
992 errno = ENAMETOOLONG;
993 if (symlinks >= EIO_PATH_MAX * 3)
994 return -1;
995
996 errno = EIO;
997 if (symlinks <= 0)
998 return -1;
999
1000 return symlinks;
1001
1002#else
1003 tmp1 = res + EIO_PATH_MAX;
1004 tmp2 = tmp1 + EIO_PATH_MAX;
1005
1006#if 0 /* disabled, the musl way to do things is just too racy */
1007#if __linux && defined(O_NONBLOCK) && defined(O_NOATIME)
1008 /* on linux we may be able to ask the kernel */
1009 {
1010 int fd = open (rel, O_RDONLY | O_NONBLOCK | O_NOCTTY | O_NOATIME);
1011
1012 if (fd >= 0)
1013 {
1014 sprintf (tmp1, "/proc/self/fd/%d", fd);
1015 req->result = readlink (tmp1, res, EIO_PATH_MAX);
1016 /* here we should probably stat the open file and the disk file, to make sure they still match */
1017 close (fd);
1018
1019 if (req->result > 0)
1020 goto done;
1021 }
1022 else if (errno == ELOOP || errno == ENAMETOOLONG || errno == ENOENT || errno == ENOTDIR || errno == EIO)
1023 return -1;
1024 }
1025#endif
1026#endif
1027
1028 if (*rel != '/')
1029 {
1030 int len;
1031
1032 errno = ENOENT;
1033 if (wd == EIO_INVALID_WD)
1034 return -1;
1035
1036 if (wd == EIO_CWD)
1037 {
1038 if (!getcwd (res, EIO_PATH_MAX))
1039 return -1;
1040
1041 len = strlen (res);
1042 }
1043 else
1044 memcpy (res, wd->str, len = wd->len);
1045
1046 if (res [1]) /* only use if not / */
1047 res += len;
1048 }
1049
1050 while (*rel)
1051 {
1052 eio_ssize_t len, linklen;
1053 const char *beg = rel;
1054
1055 while (*rel && *rel != '/')
1056 ++rel;
1057
1058 len = rel - beg;
1059
1060 if (!len) /* skip slashes */
1061 {
1062 ++rel;
1063 continue;
1064 }
1065
1066 if (beg [0] == '.')
1067 {
1068 if (len == 1)
1069 continue; /* . - nop */
1070
1071 if (beg [1] == '.' && len == 2)
1072 {
1073 /* .. - back up one component, if possible */
1074
1075 while (res != tmpbuf->ptr)
1076 if (*--res == '/')
1077 break;
1078
1079 continue;
1080 }
1081 }
1082
1083 errno = ENAMETOOLONG;
1084 if (res + 1 + len + 1 >= tmp1)
1085 return -1;
1086
1087 /* copy one component */
1088 *res = '/';
1089 memcpy (res + 1, beg, len);
1090
1091 /* zero-terminate, for readlink */
1092 res [len + 1] = 0;
1093
1094 /* now check if it's a symlink */
1095 linklen = readlink (tmpbuf->ptr, tmp1, EIO_PATH_MAX);
1096
1097 if (linklen < 0)
1098 {
1099 if (errno != EINVAL)
1100 return -1;
1101
1102 /* it's a normal directory. hopefully */
1103 res += len + 1;
1104 }
1105 else
1106 {
1107 /* yay, it was a symlink - build new path in tmp2 */
1108 int rellen = strlen (rel);
1109
1110 errno = ENAMETOOLONG;
1111 if (linklen + 1 + rellen >= EIO_PATH_MAX) /* also catch linklen >= EIO_PATH_MAX */
1112 return -1;
1113
1114 errno = ELOOP;
1115 if (!--symlinks)
1116 return -1;
1117
1118 if (*tmp1 == '/')
1119 res = tmpbuf->ptr; /* symlink resolves to an absolute path */
1120
1121 /* we need to be careful, as rel might point into tmp2 already */
1122 memmove (tmp2 + linklen + 1, rel, rellen + 1);
1123 tmp2 [linklen] = '/';
1124 memcpy (tmp2, tmp1, linklen);
1125
1126 rel = tmp2;
1127 }
1128 }
1129
1130 /* special case for the lone root path */
1131 if (res == tmpbuf->ptr)
1132 *res++ = '/';
1133
1134 return res - (char *)tmpbuf->ptr;
1135#endif
1136}
1137
1138static signed char
1139eio_dent_cmp (const eio_dirent *a, const eio_dirent *b)
1140{
1141 return a->score - b->score ? a->score - b->score /* works because our signed char is always 0..100 */
1142 : a->inode < b->inode ? -1
1143 : a->inode > b->inode ? 1
1144 : 0;
1145}
1146
1147#define EIO_DENT_CMP(i,op,j) eio_dent_cmp (&i, &j) op 0
1148
1149#define EIO_SORT_CUTOFF 30 /* quite high, but performs well on many filesystems */
1150#define EIO_SORT_FAST 60 /* when to only use insertion sort */
1151
1152static void
1153eio_dent_radix_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1154{
1155 unsigned char bits [9 + sizeof (eio_ino_t) * 8];
1156 unsigned char *bit = bits;
1157
1158 assert (CHAR_BIT == 8);
1159 assert (sizeof (eio_dirent) * 8 < 256);
1160 assert (offsetof (eio_dirent, inode)); /* we use bit #0 as sentinel */
1161 assert (offsetof (eio_dirent, score)); /* we use bit #0 as sentinel */
1162
1163 if (size <= EIO_SORT_FAST)
1164 return;
1165
1166 /* first prepare an array of bits to test in our radix sort */
1167 /* try to take endianness into account, as well as differences in eio_ino_t sizes */
1168 /* inode_bits must contain all inodes ORed together */
1169 /* which is used to skip bits that are 0 everywhere, which is very common */
1170 {
1171 eio_ino_t endianness;
1172 int i, j;
1173
1174 /* we store the byte offset of byte n into byte n of "endianness" */
1175 for (i = 0; i < sizeof (eio_ino_t); ++i)
1176 ((unsigned char *)&endianness)[i] = i;
1177
1178 *bit++ = 0;
1179
1180 for (i = 0; i < sizeof (eio_ino_t); ++i)
1181 {
1182 /* shifting off the byte offsets out of "endianness" */
1183 int offs = (offsetof (eio_dirent, inode) + (endianness & 0xff)) * 8;
1184 endianness >>= 8;
1185
1186 for (j = 0; j < 8; ++j)
1187 if (inode_bits & (((eio_ino_t)1) << (i * 8 + j)))
1188 *bit++ = offs + j;
1189 }
1190
1191 for (j = 0; j < 8; ++j)
1192 if (score_bits & (1 << j))
1193 *bit++ = offsetof (eio_dirent, score) * 8 + j;
1194 }
1195
1196 /* now actually do the sorting (a variant of MSD radix sort) */
1197 {
1198 eio_dirent *base_stk [9 + sizeof (eio_ino_t) * 8], *base;
1199 eio_dirent *end_stk [9 + sizeof (eio_ino_t) * 8], *end;
1200 unsigned char *bit_stk [9 + sizeof (eio_ino_t) * 8];
1201 int stk_idx = 0;
1202
1203 base_stk [stk_idx] = dents;
1204 end_stk [stk_idx] = dents + size;
1205 bit_stk [stk_idx] = bit - 1;
1206
1207 do
1208 {
1209 base = base_stk [stk_idx];
1210 end = end_stk [stk_idx];
1211 bit = bit_stk [stk_idx];
1212
1213 for (;;)
1214 {
1215 unsigned char O = *bit >> 3;
1216 unsigned char M = 1 << (*bit & 7);
1217
1218 eio_dirent *a = base;
1219 eio_dirent *b = end;
1220
1221 if (b - a < EIO_SORT_CUTOFF)
1222 break;
1223
1224 /* now bit-partition the array on the bit */
1225 /* this ugly asymmetric loop seems to perform much better than typical */
1226 /* partition algos found in the literature */
1227 do
1228 if (!(((unsigned char *)a)[O] & M))
1229 ++a;
1230 else if (!(((unsigned char *)--b)[O] & M))
1231 {
1232 eio_dirent tmp = *a; *a = *b; *b = tmp;
1233 ++a;
1234 }
1235 while (b > a);
1236
1237 /* next bit, or stop, if no bits left in this path */
1238 if (!*--bit)
1239 break;
1240
1241 base_stk [stk_idx] = a;
1242 end_stk [stk_idx] = end;
1243 bit_stk [stk_idx] = bit;
1244 ++stk_idx;
1245
1246 end = a;
1247 }
1248 }
1249 while (stk_idx--);
1250 }
1251}
1252
1253static void
1254eio_dent_insertion_sort (eio_dirent *dents, int size)
1255{
1256 /* first move the smallest element to the front, to act as a sentinel */
1257 {
1258 int i;
1259 eio_dirent *min = dents;
1260
1261 /* the radix pre-pass ensures that the minimum element is in the first EIO_SORT_CUTOFF + 1 elements */
1262 for (i = size > EIO_SORT_FAST ? EIO_SORT_CUTOFF + 1 : size; --i; )
1263 if (EIO_DENT_CMP (dents [i], <, *min))
1264 min = &dents [i];
1265
1266 /* swap elements 0 and j (minimum) */
1267 {
1268 eio_dirent tmp = *dents; *dents = *min; *min = tmp;
1269 }
1270 }
1271
1272 /* then do standard insertion sort, assuming that all elements are >= dents [0] */
1273 {
1274 eio_dirent *i, *j;
1275
1276 for (i = dents + 1; i < dents + size; ++i)
1277 {
1278 eio_dirent value = *i;
1279
1280 for (j = i - 1; EIO_DENT_CMP (*j, >, value); --j)
1281 j [1] = j [0];
1282
1283 j [1] = value;
1284 }
1285 }
1286}
1287
1288static void
1289eio_dent_sort (eio_dirent *dents, int size, signed char score_bits, eio_ino_t inode_bits)
1290{
1291 if (size <= 1)
1292 return; /* our insertion sort relies on size > 0 */
1293
1294 /* first we use a radix sort, but only for dirs >= EIO_SORT_FAST */
1295 /* and stop sorting when the partitions are <= EIO_SORT_CUTOFF */
1296 eio_dent_radix_sort (dents, size, score_bits, inode_bits);
1297
1298 /* use an insertion sort at the end, or for small arrays, */
1299 /* as insertion sort is more efficient for small partitions */
1300 eio_dent_insertion_sort (dents, size);
711} 1301}
712 1302
713/* read a full directory */ 1303/* read a full directory */
714static void 1304static void
715eio__scandir (eio_req *req, worker *self) 1305eio__scandir (eio_req *req, etp_worker *self)
716{ 1306{
1307 char *name, *names;
1308 int namesalloc = 4096 - sizeof (void *) * 4;
1309 int namesoffs = 0;
1310 int flags = req->int1;
1311 eio_dirent *dents = 0;
1312 int dentalloc = 128;
1313 int dentoffs = 0;
1314 eio_ino_t inode_bits = 0;
1315#ifdef _WIN32
1316 HANDLE dirp;
1317 WIN32_FIND_DATA entp;
1318#else
717 DIR *dirp; 1319 DIR *dirp;
718 union
719 {
720 EIO_STRUCT_DIRENT d;
721 char b [offsetof (EIO_STRUCT_DIRENT, d_name) + NAME_MAX + 1];
722 } *u;
723 EIO_STRUCT_DIRENT *entp; 1320 EIO_STRUCT_DIRENT *entp;
724 char *name, *names; 1321#endif
725 int memlen = 4096;
726 int memofs = 0;
727 int res = 0;
728 1322
729 X_LOCK (wrklock); 1323 req->result = -1;
730 self->dirp = dirp = opendir (req->ptr1);
731 self->dbuf = u = malloc (sizeof (*u));
732 req->flags |= EIO_FLAG_PTR2_FREE;
733 req->ptr2 = names = malloc (memlen);
734 X_UNLOCK (wrklock);
735 1324
736 if (dirp && u && names) 1325 if (!(flags & EIO_READDIR_DENTS))
737 for (;;) 1326 flags &= ~(EIO_READDIR_DIRS_FIRST | EIO_READDIR_STAT_ORDER);
1327
1328#ifdef _WIN32
1329 {
1330 int len = strlen ((const char *)req->ptr1);
1331 char *path = malloc (MAX_PATH);
1332 const char *fmt;
1333 const char *reqpath = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1334
1335 if (!len)
1336 fmt = "./*";
1337 else if (reqpath[len - 1] == '/' || reqpath[len - 1] == '\\')
1338 fmt = "%s*";
1339 else
1340 fmt = "%s/*";
1341
1342 _snprintf (path, MAX_PATH, fmt, reqpath);
1343 dirp = FindFirstFile (path, &entp);
1344 free (path);
1345
1346 if (dirp == INVALID_HANDLE_VALUE)
1347 {
1348 /* should steal _dosmaperr */
1349 switch (GetLastError ())
1350 {
1351 case ERROR_FILE_NOT_FOUND:
1352 req->result = 0;
1353 break;
1354
1355 case ERROR_INVALID_NAME:
1356 case ERROR_PATH_NOT_FOUND:
1357 case ERROR_NO_MORE_FILES:
1358 errno = ENOENT;
1359 break;
1360
1361 case ERROR_NOT_ENOUGH_MEMORY:
1362 errno = ENOMEM;
1363 break;
1364
1365 default:
1366 errno = EINVAL;
1367 break;
1368 }
1369
1370 return;
1371 }
1372 }
1373#else
1374 #if HAVE_AT
1375 if (req->wd)
738 { 1376 {
1377 int fd = openat (WD2FD (req->wd), req->ptr1, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1378
1379 if (fd < 0)
1380 return;
1381
1382 dirp = fdopendir (fd);
1383
1384 if (!dirp)
1385 close (fd);
1386 }
1387 else
1388 dirp = opendir (req->ptr1);
1389 #else
1390 dirp = opendir (wd_expand (&self->tmpbuf, req->wd, req->ptr1));
1391 #endif
1392
1393 if (!dirp)
1394 return;
1395#endif
1396
1397 if (req->flags & EIO_FLAG_PTR1_FREE)
1398 free (req->ptr1);
1399
1400 req->flags |= EIO_FLAG_PTR1_FREE | EIO_FLAG_PTR2_FREE;
1401 req->ptr1 = dents = flags ? malloc (dentalloc * sizeof (eio_dirent)) : 0;
1402 req->ptr2 = names = malloc (namesalloc);
1403
1404 if (!names || (flags && !dents))
1405 return;
1406
1407 for (;;)
1408 {
1409 int done;
1410
1411#ifdef _WIN32
1412 done = !dirp;
1413#else
739 errno = 0; 1414 errno = 0;
740 readdir_r (dirp, &u->d, &entp); 1415 entp = readdir (dirp);
1416 done = !entp;
1417#endif
741 1418
1419 if (done)
1420 {
1421#ifndef _WIN32
1422 int old_errno = errno;
1423 closedir (dirp);
1424 errno = old_errno;
1425
742 if (!entp) 1426 if (errno)
1427 break;
1428#endif
1429
1430 /* sort etc. */
1431 req->int1 = flags;
1432 req->result = dentoffs;
1433
1434 if (flags & EIO_READDIR_STAT_ORDER)
1435 eio_dent_sort (dents, dentoffs, flags & EIO_READDIR_DIRS_FIRST ? 7 : 0, inode_bits);
1436 else if (flags & EIO_READDIR_DIRS_FIRST)
1437 if (flags & EIO_READDIR_FOUND_UNKNOWN)
1438 eio_dent_sort (dents, dentoffs, 7, inode_bits); /* sort by score and inode */
1439 else
1440 {
1441 /* in this case, all is known, and we just put dirs first and sort them */
1442 eio_dirent *oth = dents + dentoffs;
1443 eio_dirent *dir = dents;
1444
1445 /* now partition dirs to the front, and non-dirs to the back */
1446 /* by walking from both sides and swapping if necessary */
1447 while (oth > dir)
1448 {
1449 if (dir->type == EIO_DT_DIR)
1450 ++dir;
1451 else if ((--oth)->type == EIO_DT_DIR)
1452 {
1453 eio_dirent tmp = *dir; *dir = *oth; *oth = tmp;
1454
1455 ++dir;
1456 }
1457 }
1458
1459 /* now sort the dirs only (dirs all have the same score) */
1460 eio_dent_sort (dents, dir - dents, 0, inode_bits);
1461 }
1462
743 break; 1463 break;
1464 }
744 1465
745 name = entp->d_name; 1466 /* now add the entry to our list(s) */
1467 name = D_NAME (entp);
746 1468
1469 /* skip . and .. entries */
747 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2]))) 1470 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
748 { 1471 {
749 int len = strlen (name) + 1; 1472 int len = D_NAMLEN (entp) + 1;
750 1473
751 res++; 1474 while (ecb_expect_false (namesoffs + len > namesalloc))
752
753 while (memofs + len > memlen)
754 { 1475 {
755 memlen *= 2; 1476 namesalloc *= 2;
756 X_LOCK (wrklock);
757 req->ptr2 = names = realloc (names, memlen); 1477 req->ptr2 = names = realloc (names, namesalloc);
758 X_UNLOCK (wrklock);
759 1478
760 if (!names) 1479 if (!names)
1480 break;
1481 }
1482
1483 memcpy (names + namesoffs, name, len);
1484
1485 if (dents)
1486 {
1487 struct eio_dirent *ent;
1488
1489 if (ecb_expect_false (dentoffs == dentalloc))
1490 {
1491 dentalloc *= 2;
1492 req->ptr1 = dents = realloc (dents, dentalloc * sizeof (eio_dirent));
1493
1494 if (!dents)
761 break; 1495 break;
762 } 1496 }
763 1497
764 memcpy (names + memofs, name, len); 1498 ent = dents + dentoffs;
765 memofs += len; 1499
1500 ent->nameofs = namesoffs; /* rather dirtily we store the offset in the pointer */
1501 ent->namelen = len - 1;
1502 ent->inode = D_INO (entp);
1503
1504 inode_bits |= ent->inode;
1505
1506 switch (D_TYPE (entp))
1507 {
1508 default:
1509 ent->type = EIO_DT_UNKNOWN;
1510 flags |= EIO_READDIR_FOUND_UNKNOWN;
1511 break;
1512
1513 #ifdef DT_FIFO
1514 case DT_FIFO: ent->type = EIO_DT_FIFO; break;
1515 #endif
1516 #ifdef DT_CHR
1517 case DT_CHR: ent->type = EIO_DT_CHR; break;
1518 #endif
1519 #ifdef DT_MPC
1520 case DT_MPC: ent->type = EIO_DT_MPC; break;
1521 #endif
1522 #ifdef DT_DIR
1523 case DT_DIR: ent->type = EIO_DT_DIR; break;
1524 #endif
1525 #ifdef DT_NAM
1526 case DT_NAM: ent->type = EIO_DT_NAM; break;
1527 #endif
1528 #ifdef DT_BLK
1529 case DT_BLK: ent->type = EIO_DT_BLK; break;
1530 #endif
1531 #ifdef DT_MPB
1532 case DT_MPB: ent->type = EIO_DT_MPB; break;
1533 #endif
1534 #ifdef DT_REG
1535 case DT_REG: ent->type = EIO_DT_REG; break;
1536 #endif
1537 #ifdef DT_NWK
1538 case DT_NWK: ent->type = EIO_DT_NWK; break;
1539 #endif
1540 #ifdef DT_CMP
1541 case DT_CMP: ent->type = EIO_DT_CMP; break;
1542 #endif
1543 #ifdef DT_LNK
1544 case DT_LNK: ent->type = EIO_DT_LNK; break;
1545 #endif
1546 #ifdef DT_SOCK
1547 case DT_SOCK: ent->type = EIO_DT_SOCK; break;
1548 #endif
1549 #ifdef DT_DOOR
1550 case DT_DOOR: ent->type = EIO_DT_DOOR; break;
1551 #endif
1552 #ifdef DT_WHT
1553 case DT_WHT: ent->type = EIO_DT_WHT; break;
1554 #endif
1555 }
1556
1557 ent->score = 7;
1558
1559 if (flags & EIO_READDIR_DIRS_FIRST)
1560 {
1561 if (ent->type == EIO_DT_UNKNOWN)
1562 {
1563 if (*name == '.') /* leading dots are likely directories, and, in any case, rare */
1564 ent->score = 1;
1565 else if (!strchr (name, '.')) /* absence of dots indicate likely dirs */
1566 ent->score = len <= 2 ? 4 - len : len <= 4 ? 4 : len <= 7 ? 5 : 6; /* shorter == more likely dir, but avoid too many classes */
1567 }
1568 else if (ent->type == EIO_DT_DIR)
1569 ent->score = 0;
1570 }
766 } 1571 }
1572
1573 namesoffs += len;
1574 ++dentoffs;
767 } 1575 }
768 1576
769 if (errno) 1577 if (EIO_CANCELLED (req))
770 res = -1; 1578 {
1579 errno = ECANCELED;
1580 break;
1581 }
1582
1583#ifdef _WIN32
1584 if (!FindNextFile (dirp, &entp))
1585 {
1586 FindClose (dirp);
1587 dirp = 0;
1588 }
1589#endif
1590 }
1591}
1592
1593/*****************************************************************************/
1594/* working directory stuff */
1595/* various deficiencies in the posix 2008 api force us to */
1596/* keep the absolute path in string form at all times */
1597/* fuck yeah. */
1598
1599#if !HAVE_AT
1600
1601/* a bit like realpath, but usually faster because it doesn'T have to return */
1602/* an absolute or canonical path */
1603static const char *
1604wd_expand (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1605{
1606 if (!wd || *path == '/')
1607 return path;
1608
1609 if (path [0] == '.' && !path [1])
1610 return wd->str;
1611
771 1612 {
772 req->result = res; 1613 int l1 = wd->len;
1614 int l2 = strlen (path);
1615
1616 char *res = etp_tmpbuf_get (tmpbuf, l1 + l2 + 2);
1617
1618 memcpy (res, wd->str, l1);
1619 res [l1] = '/';
1620 memcpy (res + l1 + 1, path, l2 + 1);
1621
1622 return res;
1623 }
773} 1624}
1625
1626#endif
1627
1628static eio_wd
1629eio__wd_open_sync (struct etp_tmpbuf *tmpbuf, eio_wd wd, const char *path)
1630{
1631 int fd;
1632 eio_wd res;
1633 int len = eio__realpath (tmpbuf, wd, path);
1634
1635 if (len < 0)
1636 return EIO_INVALID_WD;
1637
1638#if HAVE_AT
1639 fd = openat (WD2FD (wd), path, O_CLOEXEC | O_SEARCH | O_DIRECTORY);
1640
1641 if (fd < 0)
1642 return EIO_INVALID_WD;
1643#endif
1644
1645 res = malloc (sizeof (*res) + len); /* one extra 0-byte */
1646
1647#if HAVE_AT
1648 res->fd = fd;
1649#endif
1650
1651 res->len = len;
1652 memcpy (res->str, tmpbuf->ptr, len);
1653 res->str [len] = 0;
1654
1655 return res;
1656}
1657
1658eio_wd
1659eio_wd_open_sync (eio_wd wd, const char *path)
1660{
1661 struct etp_tmpbuf tmpbuf = { };
1662 wd = eio__wd_open_sync (&tmpbuf, wd, path);
1663 free (tmpbuf.ptr);
1664
1665 return wd;
1666}
1667
1668void
1669eio_wd_close_sync (eio_wd wd)
1670{
1671 if (wd != EIO_INVALID_WD && wd != EIO_CWD)
1672 {
1673 #if HAVE_AT
1674 close (wd->fd);
1675 #endif
1676 free (wd);
1677 }
1678}
1679
1680#if HAVE_AT
1681
1682/* they forgot these */
1683
1684static int
1685eio__truncateat (int dirfd, const char *path, off_t length)
1686{
1687 int fd = openat (dirfd, path, O_WRONLY | O_CLOEXEC);
1688 int res;
1689
1690 if (fd < 0)
1691 return fd;
1692
1693 res = ftruncate (fd, length);
1694 close (fd);
1695 return res;
1696}
1697
1698static int
1699eio__statvfsat (int dirfd, const char *path, struct statvfs *buf)
1700{
1701 int fd = openat (dirfd, path, O_SEARCH | O_CLOEXEC);
1702 int res;
1703
1704 if (fd < 0)
1705 return fd;
1706
1707 res = fstatvfs (fd, buf);
1708 close (fd);
1709 return res;
1710
1711}
1712
1713#endif
774 1714
775/*****************************************************************************/ 1715/*****************************************************************************/
776 1716
777#define ALLOC(len) \ 1717#define ALLOC(len) \
778 if (!req->ptr2) \ 1718 if (!req->ptr2) \
779 { \ 1719 { \
780 X_LOCK (wrklock); \ 1720 X_LOCK (EIO_POOL->wrklock); \
781 req->flags |= EIO_FLAG_PTR2_FREE; \ 1721 req->flags |= EIO_FLAG_PTR2_FREE; \
782 X_UNLOCK (wrklock); \ 1722 X_UNLOCK (EIO_POOL->wrklock); \
783 req->ptr2 = malloc (len); \ 1723 req->ptr2 = malloc (len); \
784 if (!req->ptr2) \ 1724 if (!req->ptr2) \
785 { \ 1725 { \
786 errno = ENOMEM; \ 1726 errno = ENOMEM; \
787 req->result = -1; \ 1727 req->result = -1; \
788 break; \ 1728 break; \
789 } \ 1729 } \
790 } 1730 }
791 1731
792X_THREAD_PROC (eio_proc)
793{
794 eio_req *req;
795 struct timespec ts;
796 worker *self = (worker *)thr_arg;
797
798 /* try to distribute timeouts somewhat randomly */
799 ts.tv_nsec = ((unsigned long)self & 1023UL) * (1000000000UL / 1024UL);
800
801 for (;;)
802 {
803 ts.tv_sec = time (0) + IDLE_TIMEOUT;
804
805 X_LOCK (reqlock);
806
807 for (;;)
808 {
809 self->req = req = reqq_shift (&req_queue);
810
811 if (req)
812 break;
813
814 ++idle;
815
816 if (X_COND_TIMEDWAIT (reqwait, reqlock, ts) == ETIMEDOUT)
817 {
818 if (idle > max_idle)
819 {
820 --idle;
821 X_UNLOCK (reqlock);
822 X_LOCK (wrklock);
823 --started;
824 X_UNLOCK (wrklock);
825 goto quit;
826 }
827
828 /* we are allowed to idle, so do so without any timeout */
829 X_COND_WAIT (reqwait, reqlock);
830 ts.tv_sec = time (0) + IDLE_TIMEOUT;
831 }
832
833 --idle;
834 }
835
836 --nready;
837
838 X_UNLOCK (reqlock);
839
840 errno = 0; /* strictly unnecessary */
841
842 if (!EIO_CANCELLED (req))
843 switch (req->type)
844 {
845 case EIO_READ: ALLOC (req->size);
846 req->result = req->offs >= 0
847 ? pread (req->int1, req->ptr2, req->size, req->offs)
848 : read (req->int1, req->ptr2, req->size); break;
849 case EIO_WRITE: req->result = req->offs >= 0
850 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
851 : write (req->int1, req->ptr2, req->size); break;
852
853 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
854 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size, self); break;
855
856 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
857 req->result = stat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
858 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
859 req->result = lstat (req->ptr1, (EIO_STRUCT_STAT *)req->ptr2); break;
860 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
861 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
862
863 case EIO_CHOWN: req->result = chown (req->ptr1, req->int2, req->int3); break;
864 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
865 case EIO_CHMOD: req->result = chmod (req->ptr1, (mode_t)req->int2); break;
866 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
867 case EIO_TRUNCATE: req->result = truncate (req->ptr1, req->offs); break;
868 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
869
870 case EIO_OPEN: req->result = open (req->ptr1, req->int1, (mode_t)req->int2); break;
871 case EIO_CLOSE: req->result = close (req->int1); break;
872 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
873 case EIO_UNLINK: req->result = unlink (req->ptr1); break;
874 case EIO_RMDIR: req->result = rmdir (req->ptr1); break;
875 case EIO_MKDIR: req->result = mkdir (req->ptr1, (mode_t)req->int2); break;
876 case EIO_RENAME: req->result = rename (req->ptr1, req->ptr2); break;
877 case EIO_LINK: req->result = link (req->ptr1, req->ptr2); break;
878 case EIO_SYMLINK: req->result = symlink (req->ptr1, req->ptr2); break;
879 case EIO_MKNOD: req->result = mknod (req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
880
881 case EIO_READLINK: ALLOC (NAME_MAX);
882 req->result = readlink (req->ptr1, req->ptr2, NAME_MAX); break;
883
884 case EIO_SYNC: req->result = 0; sync (); break;
885 case EIO_FSYNC: req->result = fsync (req->int1); break;
886 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
887
888 case EIO_READDIR: eio__scandir (req, self); break;
889
890 case EIO_BUSY:
891#ifdef _WIN32
892 Sleep (req->nv1 * 1000.);
893#else
894 {
895 struct timeval tv;
896
897 tv.tv_sec = req->nv1;
898 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1000000.;
899
900 req->result = select (0, 0, 0, 0, &tv);
901 }
902#endif
903 break;
904
905 case EIO_UTIME:
906 case EIO_FUTIME:
907 {
908 struct timeval tv[2];
909 struct timeval *times;
910
911 if (req->nv1 != -1. || req->nv2 != -1.)
912 {
913 tv[0].tv_sec = req->nv1;
914 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1000000.;
915 tv[1].tv_sec = req->nv2;
916 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1000000.;
917
918 times = tv;
919 }
920 else
921 times = 0;
922
923
924 req->result = req->type == EIO_FUTIME
925 ? futimes (req->int1, times)
926 : utimes (req->ptr1, times);
927 }
928
929 case EIO_GROUP:
930 case EIO_NOP:
931 req->result = 0;
932 break;
933
934 case EIO_QUIT:
935 goto quit;
936
937 default:
938 req->result = -1;
939 break;
940 }
941
942 req->errorno = errno;
943
944 X_LOCK (reslock);
945
946 ++npending;
947
948 if (!reqq_push (&res_queue, req) && want_poll_cb)
949 want_poll_cb ();
950
951 self->req = 0;
952 worker_clear (self);
953
954 X_UNLOCK (reslock);
955 }
956
957quit:
958 X_LOCK (wrklock);
959 worker_free (self);
960 X_UNLOCK (wrklock);
961
962 return 0;
963}
964
965/*****************************************************************************/ 1732/*****************************************************************************/
966 1733
967static void eio_atfork_prepare (void) 1734int ecb_cold
968{
969 X_LOCK (wrklock);
970 X_LOCK (reqlock);
971 X_LOCK (reslock);
972#if !HAVE_PREADWRITE
973 X_LOCK (preadwritelock);
974#endif
975#if !HAVE_READDIR_R
976 X_LOCK (readdirlock);
977#endif
978}
979
980static void eio_atfork_parent (void)
981{
982#if !HAVE_READDIR_R
983 X_UNLOCK (readdirlock);
984#endif
985#if !HAVE_PREADWRITE
986 X_UNLOCK (preadwritelock);
987#endif
988 X_UNLOCK (reslock);
989 X_UNLOCK (reqlock);
990 X_UNLOCK (wrklock);
991}
992
993static void eio_atfork_child (void)
994{
995 eio_req *prv;
996
997 while (prv = reqq_shift (&req_queue))
998 eio_destroy (prv);
999
1000 while (prv = reqq_shift (&res_queue))
1001 eio_destroy (prv);
1002
1003 while (wrk_first.next != &wrk_first)
1004 {
1005 worker *wrk = wrk_first.next;
1006
1007 if (wrk->req)
1008 eio_destroy (wrk->req);
1009
1010 worker_clear (wrk);
1011 worker_free (wrk);
1012 }
1013
1014 started = 0;
1015 idle = 0;
1016 nreqs = 0;
1017 nready = 0;
1018 npending = 0;
1019
1020 eio_atfork_parent ();
1021}
1022
1023int eio_init (void (*want_poll)(void), void (*done_poll)(void)) 1735eio_init (void (*want_poll)(void), void (*done_poll)(void))
1024{ 1736{
1025 want_poll_cb = want_poll; 1737 eio_want_poll_cb = want_poll;
1026 done_poll_cb = done_poll; 1738 eio_done_poll_cb = done_poll;
1027 1739
1028#ifdef _WIN32 1740 return etp_init (EIO_POOL, 0, 0, 0);
1029 X_MUTEX_CHECK (wrklock);
1030 X_MUTEX_CHECK (reslock);
1031 X_MUTEX_CHECK (reqlock);
1032 X_MUTEX_CHECK (reqwait);
1033 X_MUTEX_CHECK (preadwritelock);
1034 X_MUTEX_CHECK (readdirlock);
1035
1036 X_COND_CHECK (reqwait);
1037#endif
1038
1039 X_THREAD_ATFORK (eio_atfork_prepare, eio_atfork_parent, eio_atfork_child);
1040} 1741}
1041 1742
1743ecb_inline void
1042static void eio_api_destroy (eio_req *req) 1744eio_api_destroy (eio_req *req)
1043{ 1745{
1044 free (req); 1746 free (req);
1045} 1747}
1046 1748
1047#define REQ(rtype) \ 1749#define REQ(rtype) \
1048 eio_req *req; \ 1750 eio_req *req; \
1049 \ 1751 \
1050 req = (eio_req *)calloc (1, sizeof *req); \ 1752 req = (eio_req *)calloc (1, sizeof *req); \
1051 if (!req) \ 1753 if (!req) \
1052 return 0; \ 1754 return 0; \
1053 \ 1755 \
1054 req->type = rtype; \ 1756 req->type = rtype; \
1055 req->pri = EIO_DEFAULT_PRI + EIO_PRI_BIAS; \ 1757 req->pri = pri; \
1056 req->finish = cb; \ 1758 req->finish = cb; \
1759 req->data = data; \
1057 req->destroy = eio_api_destroy; 1760 req->destroy = eio_api_destroy;
1058 1761
1059#define SEND eio_submit (req); return req 1762#define SEND eio_submit (req); return req
1060 1763
1061#define PATH \ 1764#define PATH \
… …
1065 { \ 1768 { \
1066 eio_api_destroy (req); \ 1769 eio_api_destroy (req); \
1067 return 0; \ 1770 return 0; \
1068 } 1771 }
1069 1772
1070eio_req *eio_nop (eio_cb cb) 1773#define SINGLEDOT(ptr) (0[(char *)(ptr)] == '.' && !1[(char *)(ptr)])
1774
1775static void
1776eio_execute (etp_worker *self, eio_req *req)
1777{
1778#if HAVE_AT
1779 int dirfd;
1780#else
1781 const char *path;
1782#endif
1783
1784 if (ecb_expect_false (EIO_CANCELLED (req)))
1785 {
1786 req->result = -1;
1787 req->errorno = ECANCELED;
1788 return;
1789 }
1790
1791 if (ecb_expect_false (req->wd == EIO_INVALID_WD))
1792 {
1793 req->result = -1;
1794 req->errorno = ENOENT;
1795 return;
1796 }
1797
1798 if (req->type >= EIO_OPEN)
1799 {
1800 #if HAVE_AT
1801 dirfd = WD2FD (req->wd);
1802 #else
1803 path = wd_expand (&self->tmpbuf, req->wd, req->ptr1);
1804 #endif
1805 }
1806
1807 switch (req->type)
1808 {
1809 case EIO_WD_OPEN: req->wd = eio__wd_open_sync (&self->tmpbuf, req->wd, req->ptr1);
1810 req->result = req->wd == EIO_INVALID_WD ? -1 : 0;
1811 break;
1812 case EIO_WD_CLOSE: req->result = 0;
1813 eio_wd_close_sync (req->wd); break;
1814
1815 case EIO_SEEK: eio__lseek (req); break;
1816 case EIO_READ: ALLOC (req->size);
1817 req->result = req->offs >= 0
1818 ? pread (req->int1, req->ptr2, req->size, req->offs)
1819 : read (req->int1, req->ptr2, req->size); break;
1820 case EIO_WRITE: req->result = req->offs >= 0
1821 ? pwrite (req->int1, req->ptr2, req->size, req->offs)
1822 : write (req->int1, req->ptr2, req->size); break;
1823
1824 case EIO_FCNTL: req->result = fcntl (req->int1, (int) req->int2, req->ptr2); break;
1825 case EIO_IOCTL: req->result = ioctl (req->int1, (unsigned long)req->int2, req->ptr2); break;
1826
1827 case EIO_READAHEAD: req->result = readahead (req->int1, req->offs, req->size); break;
1828 case EIO_SENDFILE: req->result = eio__sendfile (req->int1, req->int2, req->offs, req->size); break;
1829
1830#if HAVE_AT
1831
1832 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1833 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, 0); break;
1834 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1835 req->result = fstatat (dirfd, req->ptr1, (EIO_STRUCT_STAT *)req->ptr2, AT_SYMLINK_NOFOLLOW); break;
1836 case EIO_CHOWN: req->result = fchownat (dirfd, req->ptr1, req->int2, req->int3, 0); break;
1837 case EIO_CHMOD: req->result = fchmodat (dirfd, req->ptr1, (mode_t)req->int2, 0); break;
1838 case EIO_TRUNCATE: req->result = eio__truncateat (dirfd, req->ptr1, req->offs); break;
1839 case EIO_OPEN: req->result = openat (dirfd, req->ptr1, req->int1, (mode_t)req->int2); break;
1840
1841 case EIO_UNLINK: req->result = unlinkat (dirfd, req->ptr1, 0); break;
1842 case EIO_RMDIR: /* complications arise because "." cannot be removed, so we might have to expand */
1843 req->result = req->wd && SINGLEDOT (req->ptr1)
1844 ? rmdir (req->wd->str)
1845 : unlinkat (dirfd, req->ptr1, AT_REMOVEDIR); break;
1846 case EIO_MKDIR: req->result = mkdirat (dirfd, req->ptr1, (mode_t)req->int2); break;
1847 case EIO_RENAME: /* complications arise because "." cannot be renamed, so we might have to expand */
1848 req->result = req->wd && SINGLEDOT (req->ptr1)
1849 ? rename (req->wd->str, req->ptr2)
1850 : renameat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2); break;
1851 case EIO_LINK: req->result = linkat (dirfd, req->ptr1, WD2FD ((eio_wd)req->int3), req->ptr2, 0); break;
1852 case EIO_SYMLINK: req->result = symlinkat (req->ptr1, dirfd, req->ptr2); break;
1853 case EIO_MKNOD: req->result = mknodat (dirfd, req->ptr1, (mode_t)req->int2, (dev_t)req->offs); break;
1854 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1855 req->result = eio__statvfsat (dirfd, req->ptr1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1856 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1857 req->result = readlinkat (dirfd, req->ptr1, req->ptr2, EIO_PATH_MAX);
1858 if (req->result == EIO_PATH_MAX)
1859 {
1860 req->result = -1;
1861 errno = ENAMETOOLONG;
1862 }
1863 break;
1864 case EIO_UTIME:
1865 case EIO_FUTIME:
1866 {
1867 struct timespec ts[2];
1868 struct timespec *times;
1869
1870 if (req->nv1 != -1. || req->nv2 != -1.)
1871 {
1872 ts[0].tv_sec = req->nv1;
1873 ts[0].tv_nsec = (req->nv1 - ts[0].tv_sec) * 1e9;
1874 ts[1].tv_sec = req->nv2;
1875 ts[1].tv_nsec = (req->nv2 - ts[1].tv_sec) * 1e9;
1876
1877 times = ts;
1878 }
1879 else
1880 times = 0;
1881
1882 req->result = req->type == EIO_FUTIME
1883 ? futimens (req->int1, times)
1884 : utimensat (dirfd, req->ptr1, times, 0);
1885 }
1886 break;
1887
1888#else
1889
1890 case EIO_STAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1891 req->result = stat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1892 case EIO_LSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1893 req->result = lstat (path , (EIO_STRUCT_STAT *)req->ptr2); break;
1894 case EIO_CHOWN: req->result = chown (path , req->int2, req->int3); break;
1895 case EIO_CHMOD: req->result = chmod (path , (mode_t)req->int2); break;
1896 case EIO_TRUNCATE: req->result = truncate (path , req->offs); break;
1897 case EIO_OPEN: req->result = open (path , req->int1, (mode_t)req->int2); break;
1898
1899 case EIO_UNLINK: req->result = unlink (path ); break;
1900 case EIO_RMDIR: req->result = rmdir (path ); break;
1901 case EIO_MKDIR: req->result = mkdir (path , (mode_t)req->int2); break;
1902 case EIO_RENAME: req->result = rename (path , req->ptr2); break;
1903 case EIO_LINK: req->result = link (path , req->ptr2); break;
1904 case EIO_SYMLINK: req->result = symlink (path , req->ptr2); break;
1905 case EIO_MKNOD: req->result = mknod (path , (mode_t)req->int2, (dev_t)req->offs); break;
1906 case EIO_STATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1907 req->result = statvfs (path , (EIO_STRUCT_STATVFS *)req->ptr2); break;
1908 case EIO_READLINK: ALLOC (EIO_PATH_MAX);
1909 req->result = readlink (path, req->ptr2, EIO_PATH_MAX);
1910 if (req->result == EIO_PATH_MAX)
1911 {
1912 req->result = -1;
1913 errno = ENAMETOOLONG;
1914 }
1915 break;
1916
1917 case EIO_UTIME:
1918 case EIO_FUTIME:
1919 {
1920 struct timeval tv[2];
1921 struct timeval *times;
1922
1923 if (req->nv1 != -1. || req->nv2 != -1.)
1924 {
1925 tv[0].tv_sec = req->nv1;
1926 tv[0].tv_usec = (req->nv1 - tv[0].tv_sec) * 1e6;
1927 tv[1].tv_sec = req->nv2;
1928 tv[1].tv_usec = (req->nv2 - tv[1].tv_sec) * 1e6;
1929
1930 times = tv;
1931 }
1932 else
1933 times = 0;
1934
1935 req->result = req->type == EIO_FUTIME
1936 ? futimes (req->int1, times)
1937 : utimes (req->ptr1, times);
1938 }
1939 break;
1940
1941#endif
1942
1943 case EIO_REALPATH: if (0 <= (req->result = eio__realpath (&self->tmpbuf, req->wd, req->ptr1)))
1944 {
1945 ALLOC (req->result);
1946 memcpy (req->ptr2, self->tmpbuf.ptr, req->result);
1947 }
1948 break;
1949
1950 case EIO_FSTAT: ALLOC (sizeof (EIO_STRUCT_STAT));
1951 req->result = fstat (req->int1, (EIO_STRUCT_STAT *)req->ptr2); break;
1952
1953 case EIO_FSTATVFS: ALLOC (sizeof (EIO_STRUCT_STATVFS));
1954 req->result = fstatvfs (req->int1, (EIO_STRUCT_STATVFS *)req->ptr2); break;
1955
1956 case EIO_FCHOWN: req->result = fchown (req->int1, req->int2, req->int3); break;
1957 case EIO_FCHMOD: req->result = fchmod (req->int1, (mode_t)req->int2); break;
1958 case EIO_FTRUNCATE: req->result = ftruncate (req->int1, req->offs); break;
1959
1960 case EIO_CLOSE: req->result = close (req->int1); break;
1961 case EIO_DUP2: req->result = dup2 (req->int1, req->int2); break;
1962 case EIO_SYNC: req->result = 0; sync (); break;
1963 case EIO_FSYNC: req->result = fsync (req->int1); break;
1964 case EIO_FDATASYNC: req->result = fdatasync (req->int1); break;
1965 case EIO_SYNCFS: req->result = eio__syncfs (req->int1); break;
1966 case EIO_SYNC_FILE_RANGE: req->result = eio__sync_file_range (req->int1, req->offs, req->size, req->int2); break;
1967 case EIO_MSYNC: req->result = eio__msync (req->ptr2, req->size, req->int1); break;
1968 case EIO_MTOUCH: req->result = eio__mtouch (req); break;
1969 case EIO_MLOCK: req->result = eio__mlock (req->ptr2, req->size); break;
1970 case EIO_MLOCKALL: req->result = eio__mlockall (req->int1); break;
1971 case EIO_FALLOCATE: req->result = eio__fallocate (req->int1, req->int2, req->offs, req->size); break;
1972
1973 case EIO_READDIR: eio__scandir (req, self); break;
1974
1975 case EIO_BUSY:
1976#ifdef _WIN32
1977 Sleep (req->nv1 * 1e3);
1978#else
1979 {
1980 struct timeval tv;
1981
1982 tv.tv_sec = req->nv1;
1983 tv.tv_usec = (req->nv1 - tv.tv_sec) * 1e6;
1984
1985 req->result = select (0, 0, 0, 0, &tv);
1986 }
1987#endif
1988 break;
1989
1990#if 0
1991 case EIO_GROUP:
1992 abort (); /* handled in eio_request */
1993#endif
1994
1995 case EIO_NOP:
1996 req->result = 0;
1997 break;
1998
1999 case EIO_CUSTOM:
2000 req->feed (req);
2001 break;
2002
2003 default:
2004 req->result = EIO_ENOSYS ();
2005 break;
2006 }
2007
2008 req->errorno = errno;
2009}
2010
2011#ifndef EIO_NO_WRAPPERS
2012
2013eio_req *eio_wd_open (const char *path, int pri, eio_cb cb, void *data)
2014{
2015 REQ (EIO_WD_OPEN); PATH; SEND;
2016}
2017
2018eio_req *eio_wd_close (eio_wd wd, int pri, eio_cb cb, void *data)
2019{
2020 REQ (EIO_WD_CLOSE); req->wd = wd; SEND;
2021}
2022
2023eio_req *eio_nop (int pri, eio_cb cb, void *data)
1071{ 2024{
1072 REQ (EIO_NOP); SEND; 2025 REQ (EIO_NOP); SEND;
1073} 2026}
1074 2027
1075eio_req *eio_busy (double delay, eio_cb cb) 2028eio_req *eio_busy (double delay, int pri, eio_cb cb, void *data)
1076{ 2029{
1077 REQ (EIO_BUSY); req->nv1 = delay; SEND; 2030 REQ (EIO_BUSY); req->nv1 = delay; SEND;
1078} 2031}
1079 2032
1080eio_req *eio_sync (eio_cb cb) 2033eio_req *eio_sync (int pri, eio_cb cb, void *data)
1081{ 2034{
1082 REQ (EIO_SYNC); SEND; 2035 REQ (EIO_SYNC); SEND;
1083} 2036}
1084 2037
1085eio_req *eio_fsync (int fd, eio_cb cb) 2038eio_req *eio_fsync (int fd, int pri, eio_cb cb, void *data)
1086{ 2039{
1087 REQ (EIO_FSYNC); req->int1 = fd; SEND; 2040 REQ (EIO_FSYNC); req->int1 = fd; SEND;
1088} 2041}
1089 2042
2043eio_req *eio_msync (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2044{
2045 REQ (EIO_MSYNC); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2046}
2047
1090eio_req *eio_fdatasync (int fd, eio_cb cb) 2048eio_req *eio_fdatasync (int fd, int pri, eio_cb cb, void *data)
1091{ 2049{
1092 REQ (EIO_FDATASYNC); req->int1 = fd; SEND; 2050 REQ (EIO_FDATASYNC); req->int1 = fd; SEND;
1093} 2051}
1094 2052
2053eio_req *eio_syncfs (int fd, int pri, eio_cb cb, void *data)
2054{
2055 REQ (EIO_SYNCFS); req->int1 = fd; SEND;
2056}
2057
2058eio_req *eio_sync_file_range (int fd, off_t offset, size_t nbytes, unsigned int flags, int pri, eio_cb cb, void *data)
2059{
2060 REQ (EIO_SYNC_FILE_RANGE); req->int1 = fd; req->offs = offset; req->size = nbytes; req->int2 = flags; SEND;
2061}
2062
2063eio_req *eio_mtouch (void *addr, size_t length, int flags, int pri, eio_cb cb, void *data)
2064{
2065 REQ (EIO_MTOUCH); req->ptr2 = addr; req->size = length; req->int1 = flags; SEND;
2066}
2067
2068eio_req *eio_mlock (void *addr, size_t length, int pri, eio_cb cb, void *data)
2069{
2070 REQ (EIO_MLOCK); req->ptr2 = addr; req->size = length; SEND;
2071}
2072
2073eio_req *eio_mlockall (int flags, int pri, eio_cb cb, void *data)
2074{
2075 REQ (EIO_MLOCKALL); req->int1 = flags; SEND;
2076}
2077
2078eio_req *eio_fallocate (int fd, int mode, off_t offset, size_t len, int pri, eio_cb cb, void *data)
2079{
2080 REQ (EIO_FALLOCATE); req->int1 = fd; req->int2 = mode; req->offs = offset; req->size = len; SEND;
2081}
2082
1095eio_req *eio_close (int fd, eio_cb cb) 2083eio_req *eio_close (int fd, int pri, eio_cb cb, void *data)
1096{ 2084{
1097 REQ (EIO_CLOSE); req->int1 = fd; SEND; 2085 REQ (EIO_CLOSE); req->int1 = fd; SEND;
1098} 2086}
1099 2087
1100eio_req *eio_readahead (int fd, off_t offset, size_t length, eio_cb cb) 2088eio_req *eio_readahead (int fd, off_t offset, size_t length, int pri, eio_cb cb, void *data)
1101{ 2089{
1102 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND; 2090 REQ (EIO_READAHEAD); req->int1 = fd; req->offs = offset; req->size = length; SEND;
1103} 2091}
1104 2092
2093eio_req *eio_seek (int fd, off_t offset, int whence, int pri, eio_cb cb, void *data)
2094{
2095 REQ (EIO_SEEK); req->int1 = fd; req->offs = offset; req->int2 = whence; SEND;
2096}
2097
1105eio_req *eio_read (int fd, void *data, size_t length, off_t offset, eio_cb cb) 2098eio_req *eio_read (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1106{ 2099{
1107 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2100 REQ (EIO_READ); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1108} 2101}
1109 2102
1110eio_req *eio_write (int fd, void *data, size_t length, off_t offset, eio_cb cb) 2103eio_req *eio_write (int fd, void *buf, size_t length, off_t offset, int pri, eio_cb cb, void *data)
1111{ 2104{
1112 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = data; SEND; 2105 REQ (EIO_WRITE); req->int1 = fd; req->offs = offset; req->size = length; req->ptr2 = buf; SEND;
1113} 2106}
1114 2107
2108eio_req *eio_fcntl (int fd, int cmd, void *arg, int pri, eio_cb cb, void *data)
2109{
2110 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = cmd; req->ptr2 = arg; SEND;
2111}
2112
2113eio_req *eio_ioctl (int fd, unsigned long request, void *buf, int pri, eio_cb cb, void *data)
2114{
2115 REQ (EIO_IOCTL); req->int1 = fd; req->int2 = request; req->ptr2 = buf; SEND;
2116}
2117
1115eio_req *eio_fstat (int fd, eio_cb cb) 2118eio_req *eio_fstat (int fd, int pri, eio_cb cb, void *data)
1116{ 2119{
1117 REQ (EIO_FSTAT); req->int1 = fd; SEND; 2120 REQ (EIO_FSTAT); req->int1 = fd; SEND;
1118} 2121}
1119 2122
2123eio_req *eio_fstatvfs (int fd, int pri, eio_cb cb, void *data)
2124{
2125 REQ (EIO_FSTATVFS); req->int1 = fd; SEND;
2126}
2127
1120eio_req *eio_futime (int fd, double atime, double mtime, eio_cb cb) 2128eio_req *eio_futime (int fd, double atime, double mtime, int pri, eio_cb cb, void *data)
1121{ 2129{
1122 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND; 2130 REQ (EIO_FUTIME); req->int1 = fd; req->nv1 = atime; req->nv2 = mtime; SEND;
1123} 2131}
1124 2132
1125eio_req *eio_ftruncate (int fd, off_t offset, eio_cb cb) 2133eio_req *eio_ftruncate (int fd, off_t offset, int pri, eio_cb cb, void *data)
1126{ 2134{
1127 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND; 2135 REQ (EIO_FTRUNCATE); req->int1 = fd; req->offs = offset; SEND;
1128} 2136}
1129 2137
1130eio_req *eio_fchmod (int fd, mode_t mode, eio_cb cb) 2138eio_req *eio_fchmod (int fd, mode_t mode, int pri, eio_cb cb, void *data)
1131{ 2139{
1132 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND; 2140 REQ (EIO_FCHMOD); req->int1 = fd; req->int2 = (long)mode; SEND;
1133} 2141}
1134 2142
1135eio_req *eio_fchown (int fd, uid_t uid, gid_t gid, eio_cb cb) 2143eio_req *eio_fchown (int fd, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1136{ 2144{
1137 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2145 REQ (EIO_FCHOWN); req->int1 = fd; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1138} 2146}
1139 2147
1140eio_req *eio_dup2 (int fd, int fd2, eio_cb cb) 2148eio_req *eio_dup2 (int fd, int fd2, int pri, eio_cb cb, void *data)
1141{ 2149{
1142 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND; 2150 REQ (EIO_DUP2); req->int1 = fd; req->int2 = fd2; SEND;
1143} 2151}
1144 2152
1145eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, eio_cb cb) 2153eio_req *eio_sendfile (int out_fd, int in_fd, off_t in_offset, size_t length, int pri, eio_cb cb, void *data)
1146{ 2154{
1147 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND; 2155 REQ (EIO_SENDFILE); req->int1 = out_fd; req->int2 = in_fd; req->offs = in_offset; req->size = length; SEND;
1148} 2156}
1149 2157
1150eio_req *eio_open (const char *path, int flags, mode_t mode, eio_cb cb) 2158eio_req *eio_open (const char *path, int flags, mode_t mode, int pri, eio_cb cb, void *data)
1151{ 2159{
1152 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND; 2160 REQ (EIO_OPEN); PATH; req->int1 = flags; req->int2 = (long)mode; SEND;
1153} 2161}
1154 2162
1155eio_req *eio_utime (const char *path, double atime, double mtime, eio_cb cb) 2163eio_req *eio_utime (const char *path, double atime, double mtime, int pri, eio_cb cb, void *data)
1156{ 2164{
1157 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND; 2165 REQ (EIO_UTIME); PATH; req->nv1 = atime; req->nv2 = mtime; SEND;
1158} 2166}
1159 2167
1160eio_req *eio_truncate (const char *path, off_t offset, eio_cb cb) 2168eio_req *eio_truncate (const char *path, off_t offset, int pri, eio_cb cb, void *data)
1161{ 2169{
1162 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND; 2170 REQ (EIO_TRUNCATE); PATH; req->offs = offset; SEND;
1163} 2171}
1164 2172
1165eio_req *eio_chown (const char *path, uid_t uid, gid_t gid, eio_cb cb) 2173eio_req *eio_chown (const char *path, eio_uid_t uid, eio_gid_t gid, int pri, eio_cb cb, void *data)
1166{ 2174{
1167 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND; 2175 REQ (EIO_CHOWN); PATH; req->int2 = (long)uid; req->int3 = (long)gid; SEND;
1168} 2176}
1169 2177
1170eio_req *eio_chmod (const char *path, mode_t mode, eio_cb cb) 2178eio_req *eio_chmod (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1171{ 2179{
1172 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND; 2180 REQ (EIO_CHMOD); PATH; req->int2 = (long)mode; SEND;
1173} 2181}
1174 2182
1175eio_req *eio_mkdir (const char *path, mode_t mode, eio_cb cb) 2183eio_req *eio_mkdir (const char *path, mode_t mode, int pri, eio_cb cb, void *data)
1176{ 2184{
1177 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND; 2185 REQ (EIO_MKDIR); PATH; req->int2 = (long)mode; SEND;
1178} 2186}
1179 2187
1180static eio_req * 2188static eio_req *
1181eio__1path (int type, const char *path, eio_cb cb) 2189eio__1path (int type, const char *path, int pri, eio_cb cb, void *data)
1182{ 2190{
1183 REQ (type); PATH; SEND; 2191 REQ (type); PATH; SEND;
1184} 2192}
1185 2193
1186eio_req *eio_readlink (const char *path, eio_cb cb) 2194eio_req *eio_readlink (const char *path, int pri, eio_cb cb, void *data)
1187{ 2195{
1188 return eio__1path (EIO_READLINK, path, cb); 2196 return eio__1path (EIO_READLINK, path, pri, cb, data);
1189} 2197}
1190 2198
2199eio_req *eio_realpath (const char *path, int pri, eio_cb cb, void *data)
2200{
2201 return eio__1path (EIO_REALPATH, path, pri, cb, data);
2202}
2203
1191eio_req *eio_stat (const char *path, eio_cb cb) 2204eio_req *eio_stat (const char *path, int pri, eio_cb cb, void *data)
1192{ 2205{
1193 return eio__1path (EIO_STAT, path, cb); 2206 return eio__1path (EIO_STAT, path, pri, cb, data);
1194} 2207}
1195 2208
1196eio_req *eio_lstat (const char *path, eio_cb cb) 2209eio_req *eio_lstat (const char *path, int pri, eio_cb cb, void *data)
1197{ 2210{
1198 return eio__1path (EIO_LSTAT, path, cb); 2211 return eio__1path (EIO_LSTAT, path, pri, cb, data);
1199} 2212}
1200 2213
2214eio_req *eio_statvfs (const char *path, int pri, eio_cb cb, void *data)
2215{
2216 return eio__1path (EIO_STATVFS, path, pri, cb, data);
2217}
2218
1201eio_req *eio_unlink (const char *path, eio_cb cb) 2219eio_req *eio_unlink (const char *path, int pri, eio_cb cb, void *data)
1202{ 2220{
1203 return eio__1path (EIO_UNLINK, path, cb); 2221 return eio__1path (EIO_UNLINK, path, pri, cb, data);
1204} 2222}
1205 2223
1206eio_req *eio_rmdir (const char *path, eio_cb cb) 2224eio_req *eio_rmdir (const char *path, int pri, eio_cb cb, void *data)
1207{ 2225{
1208 return eio__1path (EIO_RMDIR, path, cb); 2226 return eio__1path (EIO_RMDIR, path, pri, cb, data);
1209} 2227}
1210 2228
1211eio_req *eio_readdir (const char *path, eio_cb cb) 2229eio_req *eio_readdir (const char *path, int flags, int pri, eio_cb cb, void *data)
1212{ 2230{
1213 return eio__1path (EIO_READDIR, path, cb); 2231 REQ (EIO_READDIR); PATH; req->int1 = flags; SEND;
1214} 2232}
1215 2233
1216eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, eio_cb cb) 2234eio_req *eio_mknod (const char *path, mode_t mode, dev_t dev, int pri, eio_cb cb, void *data)
1217{ 2235{
1218 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->int2 = (long)dev; SEND; 2236 REQ (EIO_MKNOD); PATH; req->int2 = (long)mode; req->offs = (off_t)dev; SEND;
1219} 2237}
1220 2238
1221static eio_req * 2239static eio_req *
1222eio__2path (int type, const char *path, const char *new_path, eio_cb cb) 2240eio__2path (int type, const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1223{ 2241{
1224 REQ (type); PATH; 2242 REQ (type); PATH;
1225 2243
1226 req->flags |= EIO_FLAG_PTR2_FREE; 2244 req->flags |= EIO_FLAG_PTR2_FREE;
1227 req->ptr2 = strdup (new_path); 2245 req->ptr2 = strdup (new_path);
… …
1232 } 2250 }
1233 2251
1234 SEND; 2252 SEND;
1235} 2253}
1236 2254
1237eio_req *eio_link (const char *path, const char *new_path, eio_cb cb) 2255eio_req *eio_link (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1238{ 2256{
1239 return eio__2path (EIO_LINK, path, new_path, cb); 2257 return eio__2path (EIO_LINK, path, new_path, pri, cb, data);
1240} 2258}
1241 2259
1242eio_req *eio_symlink (const char *path, const char *new_path, eio_cb cb) 2260eio_req *eio_symlink (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1243{ 2261{
1244 return eio__2path (EIO_SYMLINK, path, new_path, cb); 2262 return eio__2path (EIO_SYMLINK, path, new_path, pri, cb, data);
1245} 2263}
1246 2264
1247eio_req *eio_rename (const char *path, const char *new_path, eio_cb cb) 2265eio_req *eio_rename (const char *path, const char *new_path, int pri, eio_cb cb, void *data)
1248{ 2266{
1249 return eio__2path (EIO_RENAME, path, new_path, cb); 2267 return eio__2path (EIO_RENAME, path, new_path, pri, cb, data);
2268}
2269
2270eio_req *eio_custom (void (*execute)(eio_req *), int pri, eio_cb cb, void *data)
2271{
2272 REQ (EIO_CUSTOM); req->feed = execute; SEND;
2273}
2274
2275#endif
2276
2277eio_req *eio_grp (eio_cb cb, void *data)
2278{
2279 const int pri = EIO_PRI_MAX;
2280
2281 REQ (EIO_GROUP); SEND;
1250} 2282}
1251 2283
1252#undef REQ 2284#undef REQ
1253#undef PATH 2285#undef PATH
1254#undef SEND 2286#undef SEND
1255 2287
1256/*****************************************************************************/ 2288/*****************************************************************************/
1257/* grp functions */ 2289/* grp functions */
1258 2290
2291void
1259void eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit) 2292eio_grp_feed (eio_req *grp, void (*feed)(eio_req *req), int limit)
1260{ 2293{
1261 grp->int2 = limit; 2294 grp->int2 = limit;
1262 grp->feed = feed; 2295 grp->feed = feed;
1263 2296
1264 grp_try_feed (grp); 2297 grp_try_feed (grp);
1265} 2298}
1266 2299
2300void
1267void eio_grp_limit (eio_req *grp, int limit) 2301eio_grp_limit (eio_req *grp, int limit)
1268{ 2302{
1269 grp->int2 = limit; 2303 grp->int2 = limit;
1270 2304
1271 grp_try_feed (grp); 2305 grp_try_feed (grp);
1272} 2306}
1273 2307
2308void
1274void eio_grp_add (eio_req *grp, eio_req *req) 2309eio_grp_add (eio_req *grp, eio_req *req)
1275{ 2310{
1276 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2)); 2311 assert (("cannot add requests to IO::AIO::GRP after the group finished", grp->int1 != 2));
2312
2313 grp->flags |= ETP_FLAG_GROUPADD;
1277 2314
1278 ++grp->size; 2315 ++grp->size;
1279 req->grp = grp; 2316 req->grp = grp;
1280 2317
1281 req->grp_prev = 0; 2318 req->grp_prev = 0;
… …
1288} 2325}
1289 2326
1290/*****************************************************************************/ 2327/*****************************************************************************/
1291/* misc garbage */ 2328/* misc garbage */
1292 2329
2330eio_ssize_t
1293ssize_t eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count) 2331eio_sendfile_sync (int ofd, int ifd, off_t offset, size_t count)
1294{ 2332{
1295 worker wrk;
1296
1297 wrk.dbuf = 0;
1298
1299 eio__sendfile (ofd, ifd, offset, count, &wrk); 2333 return eio__sendfile (ofd, ifd, offset, count);
1300
1301 if (wrk.dbuf)
1302 free (wrk.dbuf);
1303} 2334}
1304 2335

Diff Legend

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