ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/IO-AIO/AIO.xs
Revision: 1.70
Committed: Tue Oct 24 15:15:56 2006 UTC (17 years, 6 months ago) by root
Branch: MAIN
Changes since 1.69: +3 -0 lines
Log Message:
*** empty log message ***

File Contents

# User Rev Content
1 root 1.70 /* solaris */
2     #define _POSIX_PTHREAD_SEMANTICS 1
3    
4 root 1.63 #if __linux
5     # define _GNU_SOURCE
6     #endif
7    
8 root 1.20 #define _REENTRANT 1
9 root 1.63
10 root 1.19 #include <errno.h>
11    
12 root 1.15 #include "EXTERN.h"
13 root 1.1 #include "perl.h"
14     #include "XSUB.h"
15    
16 root 1.16 #include "autoconf/config.h"
17    
18 root 1.32 #include <pthread.h>
19    
20 root 1.37 #include <stddef.h>
21 root 1.41 #include <errno.h>
22 root 1.45 #include <sys/time.h>
23     #include <sys/select.h>
24 root 1.1 #include <sys/types.h>
25     #include <sys/stat.h>
26 root 1.37 #include <limits.h>
27 root 1.1 #include <unistd.h>
28     #include <fcntl.h>
29     #include <signal.h>
30     #include <sched.h>
31    
32 root 1.32 #if HAVE_SENDFILE
33     # if __linux
34     # include <sys/sendfile.h>
35     # elif __freebsd
36     # include <sys/socket.h>
37     # include <sys/uio.h>
38     # elif __hpux
39     # include <sys/socket.h>
40 root 1.35 # elif __solaris /* not yet */
41     # include <sys/sendfile.h>
42 root 1.34 # else
43     # error sendfile support requested but not available
44 root 1.32 # endif
45     #endif
46 root 1.1
47 root 1.39 /* used for struct dirent, AIX doesn't provide it */
48     #ifndef NAME_MAX
49     # define NAME_MAX 4096
50     #endif
51    
52 root 1.4 #if __ia64
53     # define STACKSIZE 65536
54 root 1.65 #elif __i386 || __x86_64 /* 16k is unreasonably high :( */
55     # define STACKSIZE PTHREAD_STACK_MIN
56 root 1.1 #else
57 root 1.65 # define STACKSIZE 16384
58 root 1.1 #endif
59    
60 root 1.65 /* buffer size for various temporary buffers */
61     #define AIO_BUFSIZE 65536
62    
63     #define dBUF \
64     char *aio_buf = malloc (AIO_BUFSIZE); \
65     if (!aio_buf) \
66     return -1;
67    
68     #define fBUF free (aio_buf)
69    
70 root 1.1 enum {
71     REQ_QUIT,
72     REQ_OPEN, REQ_CLOSE,
73     REQ_READ, REQ_WRITE, REQ_READAHEAD,
74 root 1.32 REQ_SENDFILE,
75 root 1.22 REQ_STAT, REQ_LSTAT, REQ_FSTAT,
76 root 1.1 REQ_FSYNC, REQ_FDATASYNC,
77 root 1.40 REQ_UNLINK, REQ_RMDIR, REQ_RENAME,
78 root 1.37 REQ_READDIR,
79 root 1.40 REQ_LINK, REQ_SYMLINK,
80 root 1.54 REQ_GROUP, REQ_NOP,
81 root 1.69 REQ_BUSY,
82 root 1.1 };
83    
84 root 1.44 #define AIO_REQ_KLASS "IO::AIO::REQ"
85     #define AIO_GRP_KLASS "IO::AIO::GRP"
86 root 1.43
87     typedef struct aio_cb
88     {
89 root 1.49 struct aio_cb *volatile next;
90 root 1.43
91     SV *data, *callback;
92     SV *fh, *fh2;
93     void *dataptr, *data2ptr;
94     Stat_t *statdata;
95 root 1.1 off_t offset;
96     size_t length;
97     ssize_t result;
98 root 1.43
99 root 1.60 STRLEN dataoffset;
100 root 1.43 int type;
101     int fd, fd2;
102 root 1.1 int errorno;
103 root 1.43 mode_t mode; /* open */
104 root 1.60
105     unsigned char flags;
106 root 1.58 unsigned char pri;
107 root 1.60
108     SV *self; /* the perl counterpart of this request, if any */
109     struct aio_cb *grp, *grp_prev, *grp_next, *grp_first;
110 root 1.1 } aio_cb;
111    
112 root 1.58 enum {
113     FLAG_CANCELLED = 0x01,
114     };
115    
116 root 1.1 typedef aio_cb *aio_req;
117 root 1.43 typedef aio_cb *aio_req_ornot;
118 root 1.1
119 root 1.60 enum {
120 root 1.61 PRI_MIN = -4,
121     PRI_MAX = 4,
122 root 1.60
123     DEFAULT_PRI = 0,
124 root 1.61 PRI_BIAS = -PRI_MIN,
125 root 1.67 NUM_PRI = PRI_MAX + PRI_BIAS + 1,
126 root 1.60 };
127    
128     static int next_pri = DEFAULT_PRI + PRI_BIAS;
129    
130 root 1.30 static int started, wanted;
131 root 1.11 static volatile int nreqs;
132 root 1.4 static int max_outstanding = 1<<30;
133 root 1.3 static int respipe [2];
134 root 1.1
135 root 1.63 #if __linux && defined (PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP)
136     # define AIO_MUTEX_INIT PTHREAD_ADAPTIVE_MUTEX_INITIALIZER_NP
137     #else
138     # define AIO_MUTEX_INIT PTHREAD_MUTEX_INITIALIZER
139     #endif
140    
141     static pthread_mutex_t reslock = AIO_MUTEX_INIT;
142     static pthread_mutex_t reqlock = AIO_MUTEX_INIT;
143 root 1.3 static pthread_cond_t reqwait = PTHREAD_COND_INITIALIZER;
144    
145 root 1.67 /*
146     * a somewhat faster data structure might be nice, but
147     * with 8 priorities this actually needs <20 insns
148     * per shift, the most expensive operation.
149     */
150     typedef struct {
151     aio_req qs[NUM_PRI], qe[NUM_PRI]; /* qstart, qend */
152     int size;
153     } reqq;
154    
155     static reqq req_queue;
156     static reqq res_queue;
157    
158     int reqq_push (reqq *q, aio_req req)
159     {
160     int pri = req->pri;
161     req->next = 0;
162    
163     if (q->qe[pri])
164     {
165     q->qe[pri]->next = req;
166     q->qe[pri] = req;
167     }
168     else
169     q->qe[pri] = q->qs[pri] = req;
170    
171     return q->size++;
172     }
173    
174     aio_req reqq_shift (reqq *q)
175     {
176     int pri;
177    
178     if (!q->size)
179     return 0;
180    
181     --q->size;
182    
183     for (pri = NUM_PRI; pri--; )
184     {
185     aio_req req = q->qs[pri];
186    
187     if (req)
188     {
189     if (!(q->qs[pri] = req->next))
190     q->qe[pri] = 0;
191    
192     return req;
193     }
194     }
195    
196     abort ();
197     }
198 root 1.1
199 root 1.50 static void req_invoke (aio_req req);
200 root 1.45 static void req_free (aio_req req);
201    
202 root 1.43 /* must be called at most once */
203 root 1.44 static SV *req_sv (aio_req req, const char *klass)
204 root 1.43 {
205 root 1.49 if (!req->self)
206     {
207     req->self = (SV *)newHV ();
208     sv_magic (req->self, 0, PERL_MAGIC_ext, (char *)req, 0);
209     }
210 root 1.43
211 root 1.45 return sv_2mortal (sv_bless (newRV_inc (req->self), gv_stashpv (klass, 1)));
212 root 1.43 }
213    
214 root 1.45 static aio_req SvAIO_REQ (SV *sv)
215 root 1.26 {
216 root 1.53 MAGIC *mg;
217    
218 root 1.45 if (!sv_derived_from (sv, AIO_REQ_KLASS) || !SvROK (sv))
219     croak ("object of class " AIO_REQ_KLASS " expected");
220 root 1.43
221 root 1.53 mg = mg_find (SvRV (sv), PERL_MAGIC_ext);
222 root 1.43
223     return mg ? (aio_req)mg->mg_ptr : 0;
224     }
225    
226 root 1.49 static void aio_grp_feed (aio_req grp)
227     {
228 root 1.58 while (grp->length < grp->fd2 && !(grp->flags & FLAG_CANCELLED))
229 root 1.49 {
230     int old_len = grp->length;
231    
232     if (grp->fh2 && SvOK (grp->fh2))
233     {
234     dSP;
235    
236     ENTER;
237     SAVETMPS;
238     PUSHMARK (SP);
239     XPUSHs (req_sv (grp, AIO_GRP_KLASS));
240     PUTBACK;
241 root 1.67 call_sv (grp->fh2, G_VOID | G_EVAL | G_KEEPERR);
242 root 1.49 SPAGAIN;
243     FREETMPS;
244     LEAVE;
245     }
246    
247     /* stop if no progress has been made */
248     if (old_len == grp->length)
249     {
250     SvREFCNT_dec (grp->fh2);
251     grp->fh2 = 0;
252     break;
253     }
254     }
255     }
256    
257 root 1.50 static void aio_grp_dec (aio_req grp)
258     {
259     --grp->length;
260    
261     /* call feeder, if applicable */
262     aio_grp_feed (grp);
263    
264     /* finish, if done */
265     if (!grp->length && grp->fd)
266     {
267     req_invoke (grp);
268     req_free (grp);
269     }
270     }
271    
272 root 1.45 static void poll_wait ()
273     {
274 root 1.62 fd_set rfd;
275    
276 root 1.60 while (nreqs)
277 root 1.45 {
278 root 1.67 int size;
279 root 1.65 #if !(__i386 || __x86_64) /* safe without sempahore on this archs */
280 root 1.60 pthread_mutex_lock (&reslock);
281 root 1.64 #endif
282 root 1.67 size = res_queue.size;
283 root 1.65 #if !(__i386 || __x86_64) /* safe without sempahore on this archs */
284 root 1.60 pthread_mutex_unlock (&reslock);
285 root 1.64 #endif
286 root 1.60
287 root 1.67 if (size)
288 root 1.60 return;
289    
290 root 1.45 FD_ZERO(&rfd);
291     FD_SET(respipe [0], &rfd);
292    
293     select (respipe [0] + 1, &rfd, 0, 0, 0);
294     }
295     }
296    
297     static void req_invoke (aio_req req)
298     {
299     dSP;
300    
301 root 1.67 if (!(req->flags & FLAG_CANCELLED) && SvOK (req->callback))
302     {
303     errno = req->errorno;
304 root 1.45
305 root 1.67 ENTER;
306     SAVETMPS;
307     PUSHMARK (SP);
308     EXTEND (SP, 1);
309 root 1.45
310 root 1.67 switch (req->type)
311 root 1.45 {
312 root 1.67 case REQ_READDIR:
313 root 1.45 {
314 root 1.67 SV *rv = &PL_sv_undef;
315 root 1.45
316 root 1.67 if (req->result >= 0)
317 root 1.45 {
318 root 1.67 char *buf = req->data2ptr;
319     AV *av = newAV ();
320    
321     while (req->result)
322     {
323     SV *sv = newSVpv (buf, 0);
324    
325     av_push (av, sv);
326     buf += SvCUR (sv) + 1;
327     req->result--;
328     }
329 root 1.45
330 root 1.67 rv = sv_2mortal (newRV_noinc ((SV *)av));
331 root 1.45 }
332    
333 root 1.67 PUSHs (rv);
334 root 1.45 }
335 root 1.67 break;
336 root 1.45
337 root 1.67 case REQ_OPEN:
338     {
339     /* convert fd to fh */
340     SV *fh;
341 root 1.45
342 root 1.67 PUSHs (sv_2mortal (newSViv (req->result)));
343     PUTBACK;
344     call_pv ("IO::AIO::_fd2fh", G_SCALAR | G_EVAL);
345     SPAGAIN;
346 root 1.45
347 root 1.67 fh = SvREFCNT_inc (POPs);
348 root 1.45
349 root 1.67 PUSHMARK (SP);
350     XPUSHs (sv_2mortal (fh));
351     }
352     break;
353 root 1.45
354 root 1.67 case REQ_GROUP:
355     req->fd = 2; /* mark group as finished */
356 root 1.45
357 root 1.67 if (req->data)
358     {
359     int i;
360     AV *av = (AV *)req->data;
361 root 1.49
362 root 1.67 EXTEND (SP, AvFILL (av) + 1);
363     for (i = 0; i <= AvFILL (av); ++i)
364     PUSHs (*av_fetch (av, i, 0));
365     }
366     break;
367 root 1.48
368 root 1.67 case REQ_NOP:
369 root 1.69 case REQ_BUSY:
370 root 1.67 break;
371 root 1.48
372 root 1.67 default:
373     PUSHs (sv_2mortal (newSViv (req->result)));
374     break;
375     }
376 root 1.45
377 root 1.51
378 root 1.67 PUTBACK;
379     call_sv (req->callback, G_VOID | G_EVAL);
380     SPAGAIN;
381 root 1.51
382 root 1.67 FREETMPS;
383     LEAVE;
384 root 1.45 }
385    
386     if (req->grp)
387     {
388     aio_req grp = req->grp;
389    
390     /* unlink request */
391 root 1.49 if (req->grp_next) req->grp_next->grp_prev = req->grp_prev;
392     if (req->grp_prev) req->grp_prev->grp_next = req->grp_next;
393    
394     if (grp->grp_first == req)
395     grp->grp_first = req->grp_next;
396    
397 root 1.50 aio_grp_dec (grp);
398 root 1.45 }
399    
400 root 1.67 if (SvTRUE (ERRSV))
401     {
402     req_free (req);
403     croak (0);
404     }
405     }
406    
407     static void req_free (aio_req req)
408     {
409 root 1.43 if (req->self)
410     {
411     sv_unmagic (req->self, PERL_MAGIC_ext);
412     SvREFCNT_dec (req->self);
413     }
414    
415 root 1.49 SvREFCNT_dec (req->data);
416     SvREFCNT_dec (req->fh);
417     SvREFCNT_dec (req->fh2);
418     SvREFCNT_dec (req->callback);
419     Safefree (req->statdata);
420 root 1.26
421 root 1.37 if (req->type == REQ_READDIR && req->result >= 0)
422     free (req->data2ptr);
423    
424 root 1.26 Safefree (req);
425     }
426    
427 root 1.45 static void req_cancel (aio_req req)
428 root 1.1 {
429 root 1.58 req->flags |= FLAG_CANCELLED;
430 root 1.45
431     if (req->type == REQ_GROUP)
432 root 1.11 {
433 root 1.45 aio_req sub;
434 root 1.3
435 root 1.49 for (sub = req->grp_first; sub; sub = sub->grp_next)
436 root 1.45 req_cancel (sub);
437 root 1.11 }
438 root 1.1 }
439    
440 root 1.45 static int poll_cb ()
441 root 1.1 {
442     dSP;
443     int count = 0;
444 root 1.24 int do_croak = 0;
445 root 1.27 aio_req req;
446    
447     for (;;)
448     {
449     pthread_mutex_lock (&reslock);
450 root 1.67 req = reqq_shift (&res_queue);
451 root 1.27
452     if (req)
453     {
454 root 1.67 if (!res_queue.size)
455 root 1.27 {
456     /* read any signals sent by the worker threads */
457     char buf [32];
458     while (read (respipe [0], buf, 32) == 32)
459     ;
460     }
461     }
462 root 1.1
463 root 1.27 pthread_mutex_unlock (&reslock);
464 root 1.3
465 root 1.27 if (!req)
466     break;
467 root 1.3
468 root 1.50 --nreqs;
469 root 1.1
470     if (req->type == REQ_QUIT)
471     started--;
472 root 1.49 else if (req->type == REQ_GROUP && req->length)
473 root 1.46 {
474     req->fd = 1; /* mark request as delayed */
475     continue;
476     }
477 root 1.1 else
478     {
479     if (req->type == REQ_READ)
480 root 1.27 SvCUR_set (req->data, req->dataoffset + (req->result > 0 ? req->result : 0));
481 root 1.1
482 root 1.28 if (req->data2ptr && (req->type == REQ_READ || req->type == REQ_WRITE))
483     SvREADONLY_off (req->data);
484    
485 root 1.27 if (req->statdata)
486 root 1.1 {
487     PL_laststype = req->type == REQ_LSTAT ? OP_LSTAT : OP_STAT;
488     PL_laststatval = req->result;
489     PL_statcache = *(req->statdata);
490     }
491    
492 root 1.45 req_invoke (req);
493 root 1.26
494 root 1.1 count++;
495     }
496    
497 root 1.43 req_free (req);
498 root 1.1 }
499    
500     return count;
501     }
502    
503 root 1.4 static void *aio_proc(void *arg);
504    
505 root 1.45 static void start_thread (void)
506 root 1.4 {
507     sigset_t fullsigset, oldsigset;
508     pthread_t tid;
509     pthread_attr_t attr;
510    
511     pthread_attr_init (&attr);
512     pthread_attr_setstacksize (&attr, STACKSIZE);
513 root 1.31 pthread_attr_setdetachstate (&attr, PTHREAD_CREATE_DETACHED);
514 root 1.4
515     sigfillset (&fullsigset);
516     sigprocmask (SIG_SETMASK, &fullsigset, &oldsigset);
517    
518     if (pthread_create (&tid, &attr, aio_proc, 0) == 0)
519     started++;
520    
521     sigprocmask (SIG_SETMASK, &oldsigset, 0);
522     }
523    
524 root 1.45 static void req_send (aio_req req)
525 root 1.4 {
526 root 1.30 while (started < wanted && nreqs >= started)
527     start_thread ();
528    
529 root 1.50 ++nreqs;
530 root 1.4
531     pthread_mutex_lock (&reqlock);
532 root 1.67 reqq_push (&req_queue, req);
533 root 1.4 pthread_cond_signal (&reqwait);
534     pthread_mutex_unlock (&reqlock);
535    
536 root 1.27 if (nreqs > max_outstanding)
537     for (;;)
538     {
539     poll_cb ();
540    
541     if (nreqs <= max_outstanding)
542     break;
543    
544     poll_wait ();
545     }
546 root 1.4 }
547    
548 root 1.45 static void end_thread (void)
549 root 1.4 {
550     aio_req req;
551 root 1.67
552 root 1.26 Newz (0, req, 1, aio_cb);
553 root 1.67
554 root 1.4 req->type = REQ_QUIT;
555 root 1.67 req->pri = PRI_MAX + PRI_BIAS;
556 root 1.4
557 root 1.43 req_send (req);
558 root 1.4 }
559    
560 root 1.22 static void min_parallel (int nthreads)
561     {
562 root 1.30 if (wanted < nthreads)
563     wanted = nthreads;
564 root 1.22 }
565    
566     static void max_parallel (int nthreads)
567     {
568     int cur = started;
569 root 1.27
570 root 1.30 if (wanted > nthreads)
571     wanted = nthreads;
572    
573     while (cur > wanted)
574     {
575 root 1.22 end_thread ();
576     cur--;
577     }
578    
579 root 1.30 while (started > wanted)
580 root 1.22 {
581     poll_wait ();
582     poll_cb ();
583     }
584     }
585    
586 root 1.26 static void create_pipe ()
587 root 1.22 {
588 root 1.26 if (pipe (respipe))
589     croak ("unable to initialize result pipe");
590 root 1.22
591 root 1.26 if (fcntl (respipe [0], F_SETFL, O_NONBLOCK))
592     croak ("cannot set result pipe to nonblocking mode");
593 root 1.22
594 root 1.26 if (fcntl (respipe [1], F_SETFL, O_NONBLOCK))
595     croak ("cannot set result pipe to nonblocking mode");
596     }
597 root 1.22
598     /*****************************************************************************/
599 root 1.17 /* work around various missing functions */
600    
601     #if !HAVE_PREADWRITE
602     # define pread aio_pread
603     # define pwrite aio_pwrite
604    
605     /*
606     * make our pread/pwrite safe against themselves, but not against
607     * normal read/write by using a mutex. slows down execution a lot,
608     * but that's your problem, not mine.
609     */
610 root 1.37 static pthread_mutex_t preadwritelock = PTHREAD_MUTEX_INITIALIZER;
611 root 1.17
612 root 1.45 static ssize_t pread (int fd, void *buf, size_t count, off_t offset)
613 root 1.17 {
614     ssize_t res;
615     off_t ooffset;
616    
617 root 1.37 pthread_mutex_lock (&preadwritelock);
618 root 1.17 ooffset = lseek (fd, 0, SEEK_CUR);
619     lseek (fd, offset, SEEK_SET);
620     res = read (fd, buf, count);
621     lseek (fd, ooffset, SEEK_SET);
622 root 1.37 pthread_mutex_unlock (&preadwritelock);
623 root 1.17
624     return res;
625     }
626    
627 root 1.45 static ssize_t pwrite (int fd, void *buf, size_t count, off_t offset)
628 root 1.17 {
629     ssize_t res;
630     off_t ooffset;
631    
632 root 1.37 pthread_mutex_lock (&preadwritelock);
633 root 1.17 ooffset = lseek (fd, 0, SEEK_CUR);
634     lseek (fd, offset, SEEK_SET);
635     res = write (fd, buf, count);
636     lseek (fd, offset, SEEK_SET);
637 root 1.37 pthread_mutex_unlock (&preadwritelock);
638 root 1.17
639     return res;
640     }
641     #endif
642    
643     #if !HAVE_FDATASYNC
644     # define fdatasync fsync
645     #endif
646    
647     #if !HAVE_READAHEAD
648     # define readahead aio_readahead
649    
650 root 1.45 static ssize_t readahead (int fd, off_t offset, size_t count)
651 root 1.17 {
652 root 1.65 dBUF;
653 root 1.37
654 root 1.17 while (count > 0)
655     {
656 root 1.65 size_t len = count < AIO_BUFSIZE ? count : AIO_BUFSIZE;
657 root 1.17
658 root 1.65 pread (fd, aio_buf, len, offset);
659 root 1.17 offset += len;
660     count -= len;
661     }
662    
663 root 1.65 fBUF;
664    
665 root 1.17 errno = 0;
666     }
667     #endif
668    
669 root 1.37 #if !HAVE_READDIR_R
670     # define readdir_r aio_readdir_r
671    
672     static pthread_mutex_t readdirlock = PTHREAD_MUTEX_INITIALIZER;
673    
674 root 1.45 static int readdir_r (DIR *dirp, struct dirent *ent, struct dirent **res)
675 root 1.37 {
676     struct dirent *e;
677     int errorno;
678    
679     pthread_mutex_lock (&readdirlock);
680    
681     e = readdir (dirp);
682     errorno = errno;
683    
684     if (e)
685     {
686     *res = ent;
687     strcpy (ent->d_name, e->d_name);
688     }
689     else
690     *res = 0;
691    
692     pthread_mutex_unlock (&readdirlock);
693    
694     errno = errorno;
695     return e ? 0 : -1;
696     }
697     #endif
698    
699 root 1.32 /* sendfile always needs emulation */
700 root 1.45 static ssize_t sendfile_ (int ofd, int ifd, off_t offset, size_t count)
701 root 1.32 {
702 root 1.37 ssize_t res;
703 root 1.32
704 root 1.37 if (!count)
705     return 0;
706 root 1.32
707 root 1.35 #if HAVE_SENDFILE
708     # if __linux
709 root 1.37 res = sendfile (ofd, ifd, &offset, count);
710 root 1.32
711 root 1.35 # elif __freebsd
712 root 1.37 /*
713     * Of course, the freebsd sendfile is a dire hack with no thoughts
714     * wasted on making it similar to other I/O functions.
715     */
716     {
717     off_t sbytes;
718     res = sendfile (ifd, ofd, offset, count, 0, &sbytes, 0);
719    
720     if (res < 0 && sbytes)
721     /* maybe only on EAGAIN only: as usual, the manpage leaves you guessing */
722     res = sbytes;
723     }
724 root 1.32
725 root 1.35 # elif __hpux
726 root 1.37 res = sendfile (ofd, ifd, offset, count, 0, 0);
727 root 1.32
728 root 1.35 # elif __solaris
729 root 1.37 {
730     struct sendfilevec vec;
731     size_t sbytes;
732    
733     vec.sfv_fd = ifd;
734     vec.sfv_flag = 0;
735     vec.sfv_off = offset;
736     vec.sfv_len = count;
737    
738     res = sendfilev (ofd, &vec, 1, &sbytes);
739    
740     if (res < 0 && sbytes)
741     res = sbytes;
742     }
743 root 1.35
744 root 1.38 # endif
745     #else
746 root 1.37 res = -1;
747     errno = ENOSYS;
748 root 1.32 #endif
749    
750 root 1.37 if (res < 0
751     && (errno == ENOSYS || errno == EINVAL || errno == ENOTSOCK
752 root 1.35 #if __solaris
753 root 1.37 || errno == EAFNOSUPPORT || errno == EPROTOTYPE
754 root 1.35 #endif
755 root 1.37 )
756     )
757     {
758     /* emulate sendfile. this is a major pain in the ass */
759 root 1.65 dBUF;
760    
761 root 1.37 res = 0;
762    
763 root 1.38 while (count)
764 root 1.37 {
765     ssize_t cnt;
766    
767 root 1.65 cnt = pread (ifd, aio_buf, count > AIO_BUFSIZE ? AIO_BUFSIZE : count, offset);
768 root 1.32
769 root 1.37 if (cnt <= 0)
770     {
771     if (cnt && !res) res = -1;
772     break;
773     }
774    
775 root 1.65 cnt = write (ofd, aio_buf, cnt);
776 root 1.37
777     if (cnt <= 0)
778     {
779     if (cnt && !res) res = -1;
780     break;
781     }
782    
783     offset += cnt;
784     res += cnt;
785 root 1.38 count -= cnt;
786 root 1.37 }
787 root 1.65
788     fBUF;
789 root 1.37 }
790    
791     return res;
792     }
793    
794     /* read a full directory */
795 root 1.45 static int scandir_ (const char *path, void **namesp)
796 root 1.37 {
797 root 1.65 DIR *dirp;
798 root 1.37 union
799     {
800     struct dirent d;
801     char b [offsetof (struct dirent, d_name) + NAME_MAX + 1];
802 root 1.65 } *u;
803 root 1.37 struct dirent *entp;
804     char *name, *names;
805     int memlen = 4096;
806     int memofs = 0;
807     int res = 0;
808     int errorno;
809    
810 root 1.65 dirp = opendir (path);
811 root 1.37 if (!dirp)
812     return -1;
813    
814 root 1.65 u = malloc (sizeof (*u));
815 root 1.37 names = malloc (memlen);
816    
817 root 1.65 if (u && names)
818     for (;;)
819     {
820     errno = 0;
821     readdir_r (dirp, &u->d, &entp);
822 root 1.37
823 root 1.65 if (!entp)
824     break;
825 root 1.37
826 root 1.65 name = entp->d_name;
827 root 1.37
828 root 1.65 if (name [0] != '.' || (name [1] && (name [1] != '.' || name [2])))
829     {
830     int len = strlen (name) + 1;
831 root 1.37
832 root 1.65 res++;
833 root 1.37
834 root 1.65 while (memofs + len > memlen)
835     {
836     memlen *= 2;
837     names = realloc (names, memlen);
838     if (!names)
839     break;
840     }
841 root 1.37
842 root 1.65 memcpy (names + memofs, name, len);
843     memofs += len;
844     }
845     }
846 root 1.37
847     errorno = errno;
848 root 1.65 free (u);
849 root 1.37 closedir (dirp);
850    
851     if (errorno)
852     {
853     free (names);
854     errno = errorno;
855     res = -1;
856     }
857    
858     *namesp = (void *)names;
859     return res;
860 root 1.32 }
861    
862 root 1.22 /*****************************************************************************/
863    
864 root 1.45 static void *aio_proc (void *thr_arg)
865 root 1.1 {
866     aio_req req;
867 root 1.3 int type;
868 root 1.1
869 root 1.3 do
870 root 1.1 {
871 root 1.3 pthread_mutex_lock (&reqlock);
872    
873     for (;;)
874     {
875 root 1.67 req = reqq_shift (&req_queue);
876 root 1.3
877     if (req)
878     break;
879    
880     pthread_cond_wait (&reqwait, &reqlock);
881     }
882    
883     pthread_mutex_unlock (&reqlock);
884    
885 root 1.1 errno = 0; /* strictly unnecessary */
886 root 1.60 type = req->type; /* remember type for QUIT check */
887 root 1.1
888 root 1.58 if (!(req->flags & FLAG_CANCELLED))
889 root 1.60 switch (type)
890 root 1.43 {
891     case REQ_READ: req->result = pread (req->fd, req->dataptr, req->length, req->offset); break;
892     case REQ_WRITE: req->result = pwrite (req->fd, req->dataptr, req->length, req->offset); break;
893 root 1.3
894 root 1.43 case REQ_READAHEAD: req->result = readahead (req->fd, req->offset, req->length); break;
895     case REQ_SENDFILE: req->result = sendfile_ (req->fd, req->fd2, req->offset, req->length); break;
896 root 1.16
897 root 1.43 case REQ_STAT: req->result = stat (req->dataptr, req->statdata); break;
898     case REQ_LSTAT: req->result = lstat (req->dataptr, req->statdata); break;
899     case REQ_FSTAT: req->result = fstat (req->fd , req->statdata); break;
900    
901     case REQ_OPEN: req->result = open (req->dataptr, req->fd, req->mode); break;
902     case REQ_CLOSE: req->result = close (req->fd); break;
903     case REQ_UNLINK: req->result = unlink (req->dataptr); break;
904     case REQ_RMDIR: req->result = rmdir (req->dataptr); break;
905     case REQ_RENAME: req->result = rename (req->data2ptr, req->dataptr); break;
906     case REQ_LINK: req->result = link (req->data2ptr, req->dataptr); break;
907     case REQ_SYMLINK: req->result = symlink (req->data2ptr, req->dataptr); break;
908    
909     case REQ_FDATASYNC: req->result = fdatasync (req->fd); break;
910     case REQ_FSYNC: req->result = fsync (req->fd); break;
911     case REQ_READDIR: req->result = scandir_ (req->dataptr, &req->data2ptr); break;
912 root 1.1
913 root 1.69 case REQ_BUSY:
914 root 1.45 {
915     struct timeval tv;
916    
917     tv.tv_sec = req->fd;
918     tv.tv_usec = req->fd2;
919    
920     req->result = select (0, 0, 0, 0, &tv);
921     }
922    
923 root 1.55 case REQ_GROUP:
924     case REQ_NOP:
925 root 1.43 case REQ_QUIT:
926     break;
927 root 1.1
928 root 1.43 default:
929     req->result = ENOSYS;
930     break;
931     }
932 root 1.1
933     req->errorno = errno;
934 root 1.3
935     pthread_mutex_lock (&reslock);
936    
937 root 1.67 if (!reqq_push (&res_queue, req))
938     /* write a dummy byte to the pipe so fh becomes ready */
939     write (respipe [1], &respipe, 1);
940 root 1.3
941     pthread_mutex_unlock (&reslock);
942 root 1.1 }
943 root 1.3 while (type != REQ_QUIT);
944 root 1.1
945     return 0;
946     }
947    
948 root 1.37 /*****************************************************************************/
949    
950     static void atfork_prepare (void)
951     {
952     pthread_mutex_lock (&reqlock);
953     pthread_mutex_lock (&reslock);
954     #if !HAVE_PREADWRITE
955     pthread_mutex_lock (&preadwritelock);
956     #endif
957     #if !HAVE_READDIR_R
958     pthread_mutex_lock (&readdirlock);
959     #endif
960     }
961    
962     static void atfork_parent (void)
963     {
964     #if !HAVE_READDIR_R
965     pthread_mutex_unlock (&readdirlock);
966     #endif
967     #if !HAVE_PREADWRITE
968     pthread_mutex_unlock (&preadwritelock);
969     #endif
970     pthread_mutex_unlock (&reslock);
971     pthread_mutex_unlock (&reqlock);
972     }
973    
974     static void atfork_child (void)
975     {
976     aio_req prv;
977    
978     started = 0;
979    
980 root 1.67 while (prv = reqq_shift (&req_queue))
981     req_free (prv);
982 root 1.37
983 root 1.67 while (prv = reqq_shift (&res_queue))
984     req_free (prv);
985    
986 root 1.37 close (respipe [0]);
987     close (respipe [1]);
988     create_pipe ();
989    
990     atfork_parent ();
991     }
992    
993 root 1.22 #define dREQ \
994     aio_req req; \
995 root 1.60 int req_pri = next_pri; \
996     next_pri = DEFAULT_PRI + PRI_BIAS; \
997 root 1.22 \
998     if (SvOK (callback) && !SvROK (callback)) \
999 root 1.43 croak ("callback must be undef or of reference type"); \
1000 root 1.22 \
1001     Newz (0, req, 1, aio_cb); \
1002     if (!req) \
1003     croak ("out of memory during aio_req allocation"); \
1004     \
1005 root 1.60 req->callback = newSVsv (callback); \
1006     req->pri = req_pri
1007 root 1.43
1008     #define REQ_SEND \
1009     req_send (req); \
1010     \
1011     if (GIMME_V != G_VOID) \
1012 root 1.44 XPUSHs (req_sv (req, AIO_REQ_KLASS));
1013 root 1.22
1014 root 1.1 MODULE = IO::AIO PACKAGE = IO::AIO
1015    
1016 root 1.8 PROTOTYPES: ENABLE
1017    
1018 root 1.1 BOOT:
1019     {
1020 root 1.41 HV *stash = gv_stashpv ("IO::AIO", 1);
1021     newCONSTSUB (stash, "EXDEV", newSViv (EXDEV));
1022     newCONSTSUB (stash, "O_RDONLY", newSViv (O_RDONLY));
1023     newCONSTSUB (stash, "O_WRONLY", newSViv (O_WRONLY));
1024    
1025 root 1.26 create_pipe ();
1026 root 1.22 pthread_atfork (atfork_prepare, atfork_parent, atfork_child);
1027 root 1.1 }
1028    
1029     void
1030 root 1.40 min_parallel (nthreads)
1031 root 1.1 int nthreads
1032     PROTOTYPE: $
1033    
1034     void
1035 root 1.40 max_parallel (nthreads)
1036 root 1.1 int nthreads
1037     PROTOTYPE: $
1038    
1039 root 1.4 int
1040 root 1.40 max_outstanding (nreqs)
1041 root 1.4 int nreqs
1042     PROTOTYPE: $
1043     CODE:
1044     RETVAL = max_outstanding;
1045     max_outstanding = nreqs;
1046    
1047 root 1.1 void
1048 root 1.40 aio_open (pathname,flags,mode,callback=&PL_sv_undef)
1049 root 1.1 SV * pathname
1050     int flags
1051     int mode
1052     SV * callback
1053 root 1.8 PROTOTYPE: $$$;$
1054 root 1.43 PPCODE:
1055 root 1.1 {
1056 root 1.22 dREQ;
1057 root 1.1
1058     req->type = REQ_OPEN;
1059     req->data = newSVsv (pathname);
1060 root 1.22 req->dataptr = SvPVbyte_nolen (req->data);
1061 root 1.1 req->fd = flags;
1062     req->mode = mode;
1063    
1064 root 1.43 REQ_SEND;
1065 root 1.1 }
1066    
1067     void
1068 root 1.40 aio_close (fh,callback=&PL_sv_undef)
1069 root 1.13 SV * fh
1070     SV * callback
1071 root 1.8 PROTOTYPE: $;$
1072 root 1.1 ALIAS:
1073     aio_close = REQ_CLOSE
1074     aio_fsync = REQ_FSYNC
1075     aio_fdatasync = REQ_FDATASYNC
1076 root 1.43 PPCODE:
1077 root 1.1 {
1078 root 1.22 dREQ;
1079 root 1.1
1080     req->type = ix;
1081 root 1.13 req->fh = newSVsv (fh);
1082     req->fd = PerlIO_fileno (IoIFP (sv_2io (fh)));
1083 root 1.1
1084 root 1.43 REQ_SEND (req);
1085 root 1.1 }
1086    
1087     void
1088 root 1.40 aio_read (fh,offset,length,data,dataoffset,callback=&PL_sv_undef)
1089 root 1.13 SV * fh
1090     UV offset
1091 root 1.32 UV length
1092 root 1.13 SV * data
1093 root 1.32 UV dataoffset
1094 root 1.13 SV * callback
1095     ALIAS:
1096     aio_read = REQ_READ
1097     aio_write = REQ_WRITE
1098 root 1.8 PROTOTYPE: $$$$$;$
1099 root 1.43 PPCODE:
1100 root 1.13 {
1101     aio_req req;
1102     STRLEN svlen;
1103 root 1.21 char *svptr = SvPVbyte (data, svlen);
1104 root 1.13
1105     SvUPGRADE (data, SVt_PV);
1106     SvPOK_on (data);
1107 root 1.1
1108 root 1.13 if (dataoffset < 0)
1109     dataoffset += svlen;
1110    
1111     if (dataoffset < 0 || dataoffset > svlen)
1112     croak ("data offset outside of string");
1113    
1114     if (ix == REQ_WRITE)
1115     {
1116     /* write: check length and adjust. */
1117     if (length < 0 || length + dataoffset > svlen)
1118     length = svlen - dataoffset;
1119     }
1120     else
1121     {
1122     /* read: grow scalar as necessary */
1123     svptr = SvGROW (data, length + dataoffset);
1124     }
1125    
1126     if (length < 0)
1127     croak ("length must not be negative");
1128    
1129 root 1.22 {
1130     dREQ;
1131 root 1.13
1132 root 1.22 req->type = ix;
1133     req->fh = newSVsv (fh);
1134     req->fd = PerlIO_fileno (ix == REQ_READ ? IoIFP (sv_2io (fh))
1135     : IoOFP (sv_2io (fh)));
1136     req->offset = offset;
1137     req->length = length;
1138     req->data = SvREFCNT_inc (data);
1139     req->dataptr = (char *)svptr + dataoffset;
1140 root 1.13
1141 root 1.28 if (!SvREADONLY (data))
1142     {
1143     SvREADONLY_on (data);
1144     req->data2ptr = (void *)data;
1145     }
1146    
1147 root 1.43 REQ_SEND;
1148 root 1.22 }
1149 root 1.13 }
1150 root 1.1
1151     void
1152 root 1.40 aio_sendfile (out_fh,in_fh,in_offset,length,callback=&PL_sv_undef)
1153 root 1.32 SV * out_fh
1154     SV * in_fh
1155     UV in_offset
1156     UV length
1157     SV * callback
1158     PROTOTYPE: $$$$;$
1159 root 1.43 PPCODE:
1160 root 1.32 {
1161     dREQ;
1162    
1163     req->type = REQ_SENDFILE;
1164     req->fh = newSVsv (out_fh);
1165     req->fd = PerlIO_fileno (IoIFP (sv_2io (out_fh)));
1166     req->fh2 = newSVsv (in_fh);
1167     req->fd2 = PerlIO_fileno (IoIFP (sv_2io (in_fh)));
1168     req->offset = in_offset;
1169     req->length = length;
1170    
1171 root 1.43 REQ_SEND;
1172 root 1.32 }
1173    
1174     void
1175 root 1.40 aio_readahead (fh,offset,length,callback=&PL_sv_undef)
1176 root 1.13 SV * fh
1177     UV offset
1178     IV length
1179     SV * callback
1180 root 1.8 PROTOTYPE: $$$;$
1181 root 1.43 PPCODE:
1182 root 1.1 {
1183 root 1.22 dREQ;
1184 root 1.1
1185     req->type = REQ_READAHEAD;
1186 root 1.13 req->fh = newSVsv (fh);
1187     req->fd = PerlIO_fileno (IoIFP (sv_2io (fh)));
1188 root 1.1 req->offset = offset;
1189     req->length = length;
1190    
1191 root 1.43 REQ_SEND;
1192 root 1.1 }
1193    
1194     void
1195 root 1.40 aio_stat (fh_or_path,callback=&PL_sv_undef)
1196 root 1.1 SV * fh_or_path
1197     SV * callback
1198     ALIAS:
1199 root 1.8 aio_stat = REQ_STAT
1200     aio_lstat = REQ_LSTAT
1201 root 1.43 PPCODE:
1202 root 1.1 {
1203 root 1.22 dREQ;
1204 root 1.1
1205     New (0, req->statdata, 1, Stat_t);
1206     if (!req->statdata)
1207 root 1.27 {
1208 root 1.43 req_free (req);
1209 root 1.27 croak ("out of memory during aio_req->statdata allocation");
1210     }
1211 root 1.1
1212     if (SvPOK (fh_or_path))
1213     {
1214 root 1.8 req->type = ix;
1215 root 1.1 req->data = newSVsv (fh_or_path);
1216 root 1.22 req->dataptr = SvPVbyte_nolen (req->data);
1217 root 1.1 }
1218     else
1219     {
1220     req->type = REQ_FSTAT;
1221 root 1.13 req->fh = newSVsv (fh_or_path);
1222 root 1.1 req->fd = PerlIO_fileno (IoIFP (sv_2io (fh_or_path)));
1223     }
1224    
1225 root 1.43 REQ_SEND;
1226 root 1.1 }
1227    
1228     void
1229 root 1.40 aio_unlink (pathname,callback=&PL_sv_undef)
1230 root 1.1 SV * pathname
1231     SV * callback
1232 root 1.22 ALIAS:
1233 root 1.40 aio_unlink = REQ_UNLINK
1234     aio_rmdir = REQ_RMDIR
1235     aio_readdir = REQ_READDIR
1236 root 1.43 PPCODE:
1237 root 1.1 {
1238 root 1.22 dREQ;
1239 root 1.1
1240 root 1.22 req->type = ix;
1241     req->data = newSVsv (pathname);
1242     req->dataptr = SvPVbyte_nolen (req->data);
1243 root 1.1
1244 root 1.43 REQ_SEND;
1245 root 1.22 }
1246    
1247     void
1248 root 1.40 aio_link (oldpath,newpath,callback=&PL_sv_undef)
1249 root 1.22 SV * oldpath
1250     SV * newpath
1251     SV * callback
1252 root 1.40 ALIAS:
1253     aio_link = REQ_LINK
1254     aio_symlink = REQ_SYMLINK
1255     aio_rename = REQ_RENAME
1256 root 1.43 PPCODE:
1257 root 1.22 {
1258     dREQ;
1259 root 1.1
1260 root 1.40 req->type = ix;
1261 root 1.22 req->fh = newSVsv (oldpath);
1262     req->data2ptr = SvPVbyte_nolen (req->fh);
1263     req->data = newSVsv (newpath);
1264     req->dataptr = SvPVbyte_nolen (req->data);
1265 root 1.1
1266 root 1.43 REQ_SEND;
1267 root 1.1 }
1268    
1269 root 1.42 void
1270 root 1.69 aio_busy (delay,callback=&PL_sv_undef)
1271 root 1.45 double delay
1272     SV * callback
1273     PPCODE:
1274     {
1275     dREQ;
1276    
1277 root 1.69 req->type = REQ_BUSY;
1278 root 1.45 req->fd = delay < 0. ? 0 : delay;
1279     req->fd2 = delay < 0. ? 0 : 1000. * (delay - req->fd);
1280    
1281     REQ_SEND;
1282     }
1283    
1284     void
1285 root 1.44 aio_group (callback=&PL_sv_undef)
1286     SV * callback
1287 root 1.46 PROTOTYPE: ;$
1288 root 1.44 PPCODE:
1289 root 1.42 {
1290 root 1.44 dREQ;
1291 root 1.60
1292 root 1.44 req->type = REQ_GROUP;
1293 root 1.45 req_send (req);
1294 root 1.60
1295 root 1.44 XPUSHs (req_sv (req, AIO_GRP_KLASS));
1296 root 1.42 }
1297    
1298 root 1.6 void
1299 root 1.54 aio_nop (callback=&PL_sv_undef)
1300     SV * callback
1301     PPCODE:
1302     {
1303     dREQ;
1304    
1305     req->type = REQ_NOP;
1306    
1307     REQ_SEND;
1308     }
1309    
1310 root 1.60 void
1311 root 1.67 aioreq_pri (int pri = DEFAULT_PRI)
1312 root 1.68 CODE:
1313     if (pri < PRI_MIN) pri = PRI_MIN;
1314     if (pri > PRI_MAX) pri = PRI_MAX;
1315     next_pri = pri + PRI_BIAS;
1316    
1317     void
1318     aioreq_nice (int nice = 0)
1319     CODE:
1320     nice = next_pri - nice;
1321     if (nice < PRI_MIN) nice = PRI_MIN;
1322     if (nice > PRI_MAX) nice = PRI_MAX;
1323     next_pri = nice + PRI_BIAS;
1324 root 1.60
1325 root 1.54 void
1326 root 1.40 flush ()
1327 root 1.6 PROTOTYPE:
1328     CODE:
1329     while (nreqs)
1330     {
1331     poll_wait ();
1332     poll_cb ();
1333     }
1334    
1335 root 1.7 void
1336     poll()
1337     PROTOTYPE:
1338     CODE:
1339     if (nreqs)
1340     {
1341     poll_wait ();
1342     poll_cb ();
1343     }
1344    
1345 root 1.1 int
1346     poll_fileno()
1347     PROTOTYPE:
1348     CODE:
1349 root 1.3 RETVAL = respipe [0];
1350 root 1.1 OUTPUT:
1351     RETVAL
1352    
1353     int
1354     poll_cb(...)
1355     PROTOTYPE:
1356     CODE:
1357 root 1.5 RETVAL = poll_cb ();
1358 root 1.1 OUTPUT:
1359     RETVAL
1360    
1361     void
1362     poll_wait()
1363     PROTOTYPE:
1364     CODE:
1365 root 1.3 if (nreqs)
1366     poll_wait ();
1367 root 1.1
1368     int
1369     nreqs()
1370     PROTOTYPE:
1371     CODE:
1372     RETVAL = nreqs;
1373     OUTPUT:
1374     RETVAL
1375    
1376 root 1.48 PROTOTYPES: DISABLE
1377    
1378 root 1.44 MODULE = IO::AIO PACKAGE = IO::AIO::REQ
1379 root 1.43
1380     void
1381     cancel (aio_req_ornot req)
1382     CODE:
1383 root 1.45 req_cancel (req);
1384    
1385 root 1.56 void
1386 root 1.59 cb (aio_req_ornot req, SV *callback=&PL_sv_undef)
1387 root 1.56 CODE:
1388     SvREFCNT_dec (req->callback);
1389     req->callback = newSVsv (callback);
1390    
1391 root 1.45 MODULE = IO::AIO PACKAGE = IO::AIO::GRP
1392    
1393     void
1394     add (aio_req grp, ...)
1395     PPCODE:
1396     {
1397     int i;
1398 root 1.53 aio_req req;
1399 root 1.45
1400 root 1.49 if (grp->fd == 2)
1401     croak ("cannot add requests to IO::AIO::GRP after the group finished");
1402    
1403 root 1.45 for (i = 1; i < items; ++i )
1404     {
1405 root 1.46 if (GIMME_V != G_VOID)
1406     XPUSHs (sv_2mortal (newSVsv (ST (i))));
1407    
1408 root 1.53 req = SvAIO_REQ (ST (i));
1409 root 1.45
1410 root 1.46 if (req)
1411     {
1412 root 1.49 ++grp->length;
1413     req->grp = grp;
1414    
1415     req->grp_prev = 0;
1416     req->grp_next = grp->grp_first;
1417 root 1.45
1418 root 1.49 if (grp->grp_first)
1419     grp->grp_first->grp_prev = req;
1420    
1421     grp->grp_first = req;
1422 root 1.46 }
1423 root 1.45 }
1424     }
1425 root 1.43
1426 root 1.48 void
1427 root 1.51 result (aio_req grp, ...)
1428     CODE:
1429     {
1430     int i;
1431     AV *av = newAV ();
1432    
1433     for (i = 1; i < items; ++i )
1434     av_push (av, newSVsv (ST (i)));
1435    
1436     SvREFCNT_dec (grp->data);
1437     grp->data = (SV *)av;
1438     }
1439    
1440     void
1441 root 1.67 limit (aio_req grp, int limit)
1442 root 1.49 CODE:
1443     grp->fd2 = limit;
1444     aio_grp_feed (grp);
1445    
1446     void
1447 root 1.56 feed (aio_req grp, SV *callback=&PL_sv_undef)
1448 root 1.49 CODE:
1449     {
1450     SvREFCNT_dec (grp->fh2);
1451     grp->fh2 = newSVsv (callback);
1452    
1453     if (grp->fd2 <= 0)
1454     grp->fd2 = 2;
1455    
1456     aio_grp_feed (grp);
1457     }
1458