store_io_diskd.c

功能强大的代理服务器

    if (diskdinfo->away >= diskdinfo->magic1) {
	/* Damn, we need to issue a sync unlink here :( */
	debug(79, 2) ("storeDiskUnlink: Out of queue space, sync unlink\n");
	storeDiskdDirUnlinkFile(SD, e->swap_filen);
	return;
    }
    /* We can attempt a diskd unlink */
    buf = storeDiskdShmGet(SD, &shm_offset);
    xstrncpy(buf, storeDiskdDirFullPath(SD, e->swap_filen, NULL), SHMBUF_BLKSZ);
    x = storeDiskdSend(_MQD_UNLINK,
	SD,
	e->swap_filen,
	NULL,
	0,
	0,
	shm_offset);
    if (x < 0) {
	debug(79, 1) ("storeDiskdSend UNLINK: %s\n", xstrerror());
	unlink(buf);		/* XXX EWW! */
	storeDiskdShmPut(SD, shm_offset);
    }
    diskd_stats.unlink.ops++;
}

void
storeDiskdRecycle(SwapDir * SD, StoreEntry * e)
{
    debug(79, 3) ("storeDiskdUnlink: fileno %08X\n", e->swap_filen);

    /* Detach from the underlying physical object */
    if (e->swap_filen > -1) {
	storeDiskdDirReplRemove(e);
	storeDiskdDirMapBitReset(SD, e->swap_filen);
	e->swap_filen = -1;
	e->swap_dirn = -1;
    }
}



/*  === STATIC =========================================================== */

static void
storeDiskdOpenDone(diomsg * M)
{
    storeIOState *sio = M->callback_data;
    statCounter.syscalls.disk.opens++;
    debug(79, 3) ("storeDiskdOpenDone: dirno %d, fileno %08x status %d\n",
	sio->swap_dirn, sio->swap_filen, M->status);
    if (M->status < 0) {
	FILE_MODE(sio->mode) == O_RDONLY ? diskd_stats.open.fail++ : diskd_stats.create.fail++;
	storeDiskdIOCallback(sio, DISK_ERROR);
    } else {
	FILE_MODE(sio->mode) == O_RDONLY ? diskd_stats.open.success++ : diskd_stats.create.success++;
    }
}

static void
storeDiskdCloseDone(diomsg * M)
{
    storeIOState *sio = M->callback_data;
    statCounter.syscalls.disk.closes++;
    debug(79, 3) ("storeDiskdCloseDone: dirno %d, fileno %08x status %d\n",
	sio->swap_dirn, sio->swap_filen, M->status);
    if (M->status < 0) {
	diskd_stats.close.fail++;
	storeDiskdIOCallback(sio, DISK_ERROR);
	return;
    }
    diskd_stats.close.success++;
    storeDiskdIOCallback(sio, DISK_OK);
}

static void
storeDiskdReadDone(diomsg * M)
{
    storeIOState *sio = M->callback_data;
    STRCB *callback = sio->read.callback;
    SwapDir *sd = INDEXSD(sio->swap_dirn);
    diskdstate_t *diskdstate = sio->fsstate;
    diskdinfo_t *diskdinfo = sd->fsdata;
    void *their_data = sio->read.callback_data;
    char *their_buf = diskdstate->read_buf;
    char *sbuf;
    size_t len;
    int valid;
    statCounter.syscalls.disk.reads++;
    diskdstate->flags.reading = 0;
    if (diskdstate->flags.close_request) {
	debug(79, 2) ("storeDiskReadDone: closing, so ignore!\n");
	return;
    }
    valid = cbdataValid(sio->read.callback_data);
    cbdataUnlock(sio->read.callback_data);
    debug(79, 3) ("storeDiskdReadDone: dirno %d, fileno %08x status %d\n",
	sio->swap_dirn, sio->swap_filen, M->status);
    if (M->status < 0) {
	diskd_stats.read.fail++;
	storeDiskdIOCallback(sio, DISK_ERROR);
	return;
    }
    diskd_stats.read.success++;
    sbuf = diskdinfo->shm.buf + M->shm_offset;
    len = M->status;
    sio->offset += len;
    assert(callback);
    assert(their_data);
    sio->read.callback = NULL;
    sio->read.callback_data = NULL;
    if (valid) {
	assert(!diskdstate->flags.close_request);
	/*
	 * Only copy the data if the callback is still valid,
	 * if it isn't valid then the request should have been
	 * aborted.
	 *   -- adrian
	 */
	xmemcpy(their_buf, sbuf, len);	/* yucky copy */
	callback(their_data, their_buf, len);
    }
}

static void
storeDiskdWriteDone(diomsg * M)
{
    storeIOState *sio = M->callback_data;
    diskdstate_t *diskdstate = sio->fsstate;
    statCounter.syscalls.disk.writes++;
    diskdstate->flags.writing = 0;
    debug(79, 3) ("storeDiskdWriteDone: dirno %d, fileno %08x status %d\n",
	sio->swap_dirn, sio->swap_filen, M->status);
    if (M->status < 0) {
	diskd_stats.write.fail++;
	if (!diskdstate->flags.close_request)
	    storeDiskdIOCallback(sio, DISK_ERROR);
	return;
    }
    diskd_stats.write.success++;
    sio->offset += M->status;
}

static void
storeDiskdUnlinkDone(diomsg * M)
{
    debug(79, 3) ("storeDiskdUnlinkDone: fileno %08x status %d\n",
	M->id, M->status);
    statCounter.syscalls.disk.unlinks++;
    if (M->status < 0)
	diskd_stats.unlink.fail++;
    else
	diskd_stats.unlink.success++;
}

void
storeDiskdHandle(diomsg * M)
{
    int valid = M->callback_data ? cbdataValid(M->callback_data) : 1;
    if (M->callback_data)
	cbdataUnlock(M->callback_data);
    if (!valid) {
	debug(79, 3) ("storeDiskdHandle: Invalid callback_data %p\n",
	    M->callback_data);
	/*
	 * The read operation has its own callback.  If we don't
	 * call storeDiskdReadDone(), then we must make sure the
	 * callback_data gets unlocked!
	 */
	if (_MQD_READ == M->mtype) {
	    storeIOState *sio = M->callback_data;
	    cbdataUnlock(sio->read.callback_data);
	}
	return;
    }
    /* set errno passed from diskd.  makes debugging more meaningful */
    if (M->status < 0)
	errno = -M->status;
    switch (M->mtype) {
    case _MQD_OPEN:
	storeDiskdOpenDone(M);
	break;
    case _MQD_CLOSE:
	storeDiskdCloseDone(M);
	break;
    case _MQD_READ:
	storeDiskdReadDone(M);
	break;
    case _MQD_WRITE:
	storeDiskdWriteDone(M);
	break;
    case _MQD_UNLINK:
	storeDiskdUnlinkDone(M);
	break;
    default:
	assert(0);
	break;
    }
}

static void
storeDiskdIOCallback(storeIOState * sio, int errflag)
{
    void *p = sio->callback_data;
    debug(79, 3) ("storeDiskdIOCallback: errflag=%d\n", errflag);
    if (cbdataValid(p))
	sio->callback(p, errflag, sio);
    cbdataUnlock(p);
    cbdataFree(sio);
}

static int
storeDiskdSend(int mtype, SwapDir * sd, int id, storeIOState * sio, int size, off_t offset, int shm_offset)
{
    int x;
    diomsg M;
    static int send_errors = 0;
    static int last_seq_no = 0;
    static int seq_no = 0;
    diskdinfo_t *diskdinfo = sd->fsdata;
    struct timeval delay =
    {0, 1};

    M.mtype = mtype;
    M.callback_data = sio;
    M.size = size;
    M.offset = offset;
    M.status = -1;
    M.shm_offset = (int) shm_offset;
    M.id = id;
    M.seq_no = ++seq_no;
    if (M.callback_data)
	cbdataLock(M.callback_data);
    if (M.seq_no < last_seq_no)
	debug(79, 1) ("WARNING: sequencing out of order\n");

    /*
     * We have to drain the queue here if necessary.  If we don't,
     * then we can have a lot of messages in the queue (probably
     * up to 2*magic1) and we can run out of shared memory buffers.
     */
    /*
     * NOTE that it is important that we call storeDirCallback AFTER
     * locking the callback data M.callback_data because we need
     * to make sure the cbdata lock count doesn't go to zero (and
     * get freed) before we have a chance to send the current message
     * M!
     */
    /*
     * Note that we call storeDirCallback (for all SDs), rather
     * than storeDiskdDirCallback for just this SD, so that while
     * we're "blocking" on this SD we can also handle callbacks
     * from other SDs that might be ready.
     */
    while (diskdinfo->away > diskdinfo->magic2) {
	select(0, NULL, NULL, NULL, &delay);
	storeDirCallback();
	if (delay.tv_usec < 1000000)
	    delay.tv_usec <<= 1;
    }

    x = msgsnd(diskdinfo->smsgid, &M, msg_snd_rcv_sz, IPC_NOWAIT);
    last_seq_no = M.seq_no;
    if (0 == x) {
	diskd_stats.sent_count++;
	diskdinfo->away++;
    } else {
	debug(79, 1) ("storeDiskdSend: msgsnd: %s\n", xstrerror());
	if (M.callback_data)
	    cbdataUnlock(M.callback_data);
	assert(++send_errors < 100);
    }
    return x;
}


/*
 * We can't pass memFree() as a free function here, because we need to free
 * the fsstate variable ..
 */
static void
storeDiskdIOFreeEntry(void *sio)
{
    memPoolFree(diskd_state_pool, ((storeIOState *) sio)->fsstate);
}