2 * DEBUG: section 79 Squid-side DISKD I/O functions.
3 * AUTHOR: Duane Wessels
5 * SQUID Web Proxy Cache http://www.squid-cache.org/
6 * ----------------------------------------------------------
8 * Squid is the result of efforts by numerous individuals from
9 * the Internet community; see the CONTRIBUTORS file for full
10 * details. Many organizations have provided support for Squid's
11 * development; see the SPONSORS file for full details. Squid is
12 * Copyrighted (C) 2001 by the Regents of the University of
13 * California; see the COPYRIGHT file for full details. Squid
14 * incorporates software developed and/or copyrighted by other
15 * sources; see the CREDITS file for full details.
17 * This program is free software; you can redistribute it and/or modify
18 * it under the terms of the GNU General Public License as published by
19 * the Free Software Foundation; either version 2 of the License, or
20 * (at your option) any later version.
22 * This program is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
25 * GNU General Public License for more details.
27 * You should have received a copy of the GNU General Public License
28 * along with this program; if not, write to the Free Software
29 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
31 * Copyright (c) 2003, Robert Collins <robertc@squid-cache.org>
35 #include "comm/Loops.h"
36 #include "ConfigOption.h"
37 #include "DiskdIOStrategy.h"
38 #include "DiskIO/DiskFile.h"
39 #include "DiskdFile.h"
43 #include "StatCounters.h"
44 #include "SquidConfig.h"
46 #include "SquidTime.h"
62 diskd_stats_t diskd_stats
;
64 size_t DiskdIOStrategy::nextInstanceID (0);
65 const int diomsg::msg_snd_rcv_sz
= sizeof(diomsg
) - sizeof(mtyp_t
);
68 DiskdIOStrategy::newInstance()
70 return ++nextInstanceID
;
74 DiskdIOStrategy::shedLoad()
77 * Fail on open() if there are too many requests queued.
81 debugs(79, 3, "storeDiskdIO::shedLoad: Shedding, too many requests away");
90 DiskdIOStrategy::load()
92 /* Calculate the storedir load relative to magic2 on a scale of 0 .. 1000 */
93 /* the parse function guarantees magic2 is positivie */
94 return away
* 1000 / magic2
;
98 DiskdIOStrategy::openFailed()
100 ++diskd_stats
.open_fail_queue_len
;
104 DiskdIOStrategy::newFile(char const *path
)
111 return new DiskdFile (path
, this);
114 DiskdIOStrategy::DiskdIOStrategy() : magic1(64), magic2(72), away(0) , smsgid(-1), rmsgid(-1), wfd(-1) , instanceID(newInstance())
118 DiskdIOStrategy::unlinkdUseful() const
124 DiskdIOStrategy::unlinkFile(char const *path
)
127 /* Damn, we need to issue a sync unlink here :( */
128 debugs(79, 2, "storeDiskUnlink: Out of queue space, sync unlink");
133 /* We can attempt a diskd unlink */
140 buf
= (char *)shm
.get(&shm_offset
);
142 xstrncpy(buf
, path
, SHMBUF_BLKSZ
);
144 x
= send(_MQD_UNLINK
,
146 (StoreIOState::Pointer
)NULL
,
152 debugs(79, DBG_IMPORTANT
, "storeDiskdSend UNLINK: " << xstrerror());
153 ::unlink(buf
); /* XXX EWW! */
154 // shm.put (shm_offset);
157 ++diskd_stats
.unlink
.ops
;
161 DiskdIOStrategy::init()
171 Ip::Address localhost
;
173 ikey
= (getpid() << 10) + (instanceID
<< 2);
175 smsgid
= msgget((key_t
) ikey
, 0700 | IPC_CREAT
);
178 debugs(50, DBG_CRITICAL
, "storeDiskdInit: msgget: " << xstrerror());
179 fatal("msgget failed");
182 rmsgid
= msgget((key_t
) (ikey
+ 1), 0700 | IPC_CREAT
);
185 debugs(50, DBG_CRITICAL
, "storeDiskdInit: msgget: " << xstrerror());
186 fatal("msgget failed");
189 shm
.init(ikey
, magic2
);
190 snprintf(skey1
, 32, "%d", ikey
);
191 snprintf(skey2
, 32, "%d", ikey
+ 1);
192 snprintf(skey3
, 32, "%d", ikey
+ 2);
198 localhost
.SetLocalhost();
199 pid
= ipcCreate(IPC_STREAM
,
200 Config
.Program
.diskd
,
209 fatalf("execl: %s", Config
.Program
.diskd
);
214 fd_note(wfd
, "squid -> diskd");
216 commUnsetFdTimeout(wfd
);
217 commSetNonBlocking(wfd
);
218 Comm::QuickPollRequired();
222 * SHM manipulation routines
225 SharedMemory::put(ssize_t offset
)
229 assert(offset
< nbufs
* SHMBUF_BLKSZ
);
230 i
= offset
/ SHMBUF_BLKSZ
;
232 assert(CBIT_TEST(inuse_map
, i
));
233 CBIT_CLR(inuse_map
, i
);
234 --diskd_stats
.shmbuf_count
;
239 SharedMemory::get(ssize_t
* shm_offset
)
244 for (i
= 0; i
< nbufs
; ++i
) {
245 if (CBIT_TEST(inuse_map
, i
))
248 CBIT_SET(inuse_map
, i
);
250 *shm_offset
= i
* SHMBUF_BLKSZ
;
252 aBuf
= buf
+ (*shm_offset
);
259 assert(aBuf
< buf
+ (nbufs
* SHMBUF_BLKSZ
));
260 ++diskd_stats
.shmbuf_count
;
262 if (diskd_stats
.max_shmuse
< diskd_stats
.shmbuf_count
)
263 diskd_stats
.max_shmuse
= diskd_stats
.shmbuf_count
;
269 SharedMemory::init(int ikey
, int magic2
)
271 nbufs
= (int)(magic2
* 1.3);
272 id
= shmget((key_t
) (ikey
+ 2),
273 nbufs
* SHMBUF_BLKSZ
, 0600 | IPC_CREAT
);
276 debugs(50, DBG_CRITICAL
, "storeDiskdInit: shmget: " << xstrerror());
277 fatal("shmget failed");
280 buf
= (char *)shmat(id
, NULL
, 0);
282 if (buf
== (void *) -1) {
283 debugs(50, DBG_CRITICAL
, "storeDiskdInit: shmat: " << xstrerror());
284 fatal("shmat failed");
287 inuse_map
= (char *)xcalloc((nbufs
+ 7) / 8, 1);
288 diskd_stats
.shmbuf_count
+= nbufs
;
290 for (int i
= 0; i
< nbufs
; ++i
) {
291 CBIT_SET(inuse_map
, i
);
292 put (i
* SHMBUF_BLKSZ
);
297 DiskdIOStrategy::unlinkDone(diomsg
* M
)
299 debugs(79, 3, "storeDiskdUnlinkDone: file " << shm
.buf
+ M
->shm_offset
<< " status " << M
->status
);
300 ++statCounter
.syscalls
.disk
.unlinks
;
303 ++diskd_stats
.unlink
.fail
;
305 ++diskd_stats
.unlink
.success
;
309 DiskdIOStrategy::handle(diomsg
* M
)
311 if (!cbdataReferenceValid (M
->callback_data
)) {
312 /* I.e. already closed file
313 * - say when we have a error opening after
314 * a read was already queued
316 debugs(79, 3, "storeDiskdHandle: Invalid callback_data " << M
->callback_data
);
317 cbdataReferenceDone (M
->callback_data
);
321 /* set errno passed from diskd. makes debugging more meaningful */
326 DiskdFile
*theFile
= (DiskdFile
*)M
->callback_data
;
328 theFile
->completed (M
);
353 cbdataReferenceDone (M
->callback_data
);
357 DiskdIOStrategy::send(int mtype
, int id
, DiskdFile
*theFile
, size_t size
, off_t offset
, ssize_t shm_offset
, Lock
*requestor
)
360 M
.callback_data
= cbdataReference(theFile
);
362 M
.requestor
= requestor
;
368 return SEND(&M
, mtype
, id
, size
, offset
, shm_offset
);
372 DiskdIOStrategy::send(int mtype
, int id
, RefCount
<StoreIOState
> sio
, size_t size
, off_t offset
, ssize_t shm_offset
)
375 M
.callback_data
= cbdataReference(sio
.getRaw());
378 return SEND(&M
, mtype
, id
, size
, offset
, shm_offset
);
382 DiskdIOStrategy::SEND(diomsg
*M
, int mtype
, int id
, size_t size
, off_t offset
, ssize_t shm_offset
)
384 static int send_errors
= 0;
385 static int last_seq_no
= 0;
386 static int seq_no
= 0;
393 M
->shm_offset
= (int) shm_offset
;
395 M
->seq_no
= ++seq_no
;
397 if (M
->seq_no
< last_seq_no
)
398 debugs(79, DBG_IMPORTANT
, "WARNING: sequencing out of order");
400 x
= msgsnd(smsgid
, M
, diomsg::msg_snd_rcv_sz
, IPC_NOWAIT
);
402 last_seq_no
= M
->seq_no
;
405 ++diskd_stats
.sent_count
;
408 debugs(79, DBG_IMPORTANT
, "storeDiskdSend: msgsnd: " << xstrerror());
409 cbdataReferenceDone(M
->callback_data
);
410 assert(++send_errors
< 100);
416 * We have to drain the queue here if necessary. If we don't,
417 * then we can have a lot of messages in the queue (probably
418 * up to 2*magic1) and we can run out of shared memory buffers.
421 * Note that we call Store::Root().callbackk (for all SDs), rather
422 * than callback for just this SD, so that while
423 * we're "blocking" on this SD we can also handle callbacks
424 * from other SDs that might be ready.
427 struct timeval delay
= {0, 1};
429 while (away
> magic2
) {
430 select(0, NULL
, NULL
, NULL
, &delay
);
431 Store::Root().callback();
433 if (delay
.tv_usec
< 1000000)
441 DiskdIOStrategy::getOptionTree() const
443 ConfigOptionVector
*result
= new ConfigOptionVector
;
444 result
->options
.push_back(new ConfigOptionAdapter
<DiskdIOStrategy
>(*const_cast<DiskdIOStrategy
*>(this), &DiskdIOStrategy::optionQ1Parse
, &DiskdIOStrategy::optionQ1Dump
));
445 result
->options
.push_back(new ConfigOptionAdapter
<DiskdIOStrategy
>(*const_cast<DiskdIOStrategy
*>(this), &DiskdIOStrategy::optionQ2Parse
, &DiskdIOStrategy::optionQ2Dump
));
450 DiskdIOStrategy::optionQ1Parse(const char *name
, const char *value
, int isaReconfig
)
452 if (strcmp(name
, "Q1") != 0)
455 int old_magic1
= magic1
;
457 magic1
= atoi(value
);
462 if (old_magic1
< magic1
) {
464 * This is because shm.nbufs is computed at startup, when
465 * we call shmget(). We can't increase the Q1/Q2 parameters
466 * beyond their initial values because then we might have
467 * more "Q2 messages" than shared memory chunks, and this
468 * will cause an assertion in storeDiskdShmGet().
470 /* TODO: have DiskdIO hold a link to the swapdir, to allow detailed reporting again */
471 debugs(3, DBG_IMPORTANT
, "WARNING: cannot increase cache_dir Q1 value while Squid is running.");
476 if (old_magic1
!= magic1
)
477 debugs(3, DBG_IMPORTANT
, "cache_dir new Q1 value '" << magic1
<< "'");
483 DiskdIOStrategy::optionQ1Dump(StoreEntry
* e
) const
485 storeAppendPrintf(e
, " Q1=%d", magic1
);
489 DiskdIOStrategy::optionQ2Parse(const char *name
, const char *value
, int isaReconfig
)
491 if (strcmp(name
, "Q2") != 0)
494 int old_magic2
= magic2
;
496 magic2
= atoi(value
);
501 if (old_magic2
< magic2
) {
502 /* See comments in Q1 function above */
503 debugs(3, DBG_IMPORTANT
, "WARNING: cannot increase cache_dir Q2 value while Squid is running.");
508 if (old_magic2
!= magic2
)
509 debugs(3, DBG_IMPORTANT
, "cache_dir new Q2 value '" << magic2
<< "'");
515 DiskdIOStrategy::optionQ2Dump(StoreEntry
* e
) const
517 storeAppendPrintf(e
, " Q2=%d", magic2
);
521 * Sync any pending data. We just sit around and read the queue
522 * until the data has finished writing.
525 DiskdIOStrategy::sync()
527 static time_t lastmsg
= 0;
530 if (squid_curtime
> lastmsg
) {
531 debugs(47, DBG_IMPORTANT
, "storeDiskdDirSync: " << away
<< " messages away");
532 lastmsg
= squid_curtime
;
540 * Handle callbacks. If we have more than magic2 requests away, we block
541 * until the queue is below magic2. Otherwise, we simply return when we
542 * don't get a message.
546 DiskdIOStrategy::callback()
552 if (away
>= magic2
) {
553 ++diskd_stats
.block_queue_len
;
555 /* We might not have anything to do, but our queue
559 if (diskd_stats
.sent_count
- diskd_stats
.recv_count
>
560 diskd_stats
.max_away
) {
561 diskd_stats
.max_away
= diskd_stats
.sent_count
- diskd_stats
.recv_count
;
565 #ifdef ALWAYS_ZERO_BUFFERS
566 memset(&M
, '\0', sizeof(M
));
569 x
= msgrcv(rmsgid
, &M
, diomsg::msg_snd_rcv_sz
, 0, IPC_NOWAIT
);
573 else if (x
!= diomsg::msg_snd_rcv_sz
) {
574 debugs(47, DBG_IMPORTANT
, "storeDiskdDirCallback: msgget returns " << x
);
578 ++diskd_stats
.recv_count
;
581 retval
= 1; /* Return that we've actually done some work */
583 if (M
.shm_offset
> -1)
584 shm
.put ((off_t
) M
.shm_offset
);
591 DiskdIOStrategy::statfs(StoreEntry
& sentry
)const
593 storeAppendPrintf(&sentry
, "Pending operations: %d\n", away
);