store_dir_diskd.c

一个功能非常全面的代理服务器源代码程序,

	&diskdinfo->hIpc);
    if (pid < 0)
	fatalf("execl: %s", Config.Program.diskd);
    fd_note(diskdinfo->rfd, "diskd -> squid health monitor");
    fd_note(diskdinfo->wfd, "squid -> diskd health monitor");
    commSetSelect(diskdinfo->rfd, COMM_SELECT_READ, diskdExited, NULL, 0);
    storeDiskdDirInitBitmap(sd);
    if (storeDiskdDirVerifyCacheDirs(sd) < 0)
	fatal(errmsg);
    storeDiskdDirOpenSwapLog(sd);
    storeDiskdDirRebuild(sd);
    if (!started_clean_event) {
	eventAdd("storeDirClean", storeDiskdDirCleanEvent, NULL, 15.0, 1);
	started_clean_event = 1;
    }
    (void) storeDirGetBlkSize(sd->path, &sd->fs.blksize);
    comm_quick_poll_required();
}


static void
storeDiskdStats(StoreEntry * sentry)
{
    storeAppendPrintf(sentry, "sent_count: %d\n", diskd_stats.sent_count);
    storeAppendPrintf(sentry, "recv_count: %d\n", diskd_stats.recv_count);
    storeAppendPrintf(sentry, "max_away: %d\n", diskd_stats.max_away);
    storeAppendPrintf(sentry, "max_shmuse: %d\n", diskd_stats.max_shmuse);
    storeAppendPrintf(sentry, "open_fail_queue_len: %d\n", diskd_stats.open_fail_queue_len);
    storeAppendPrintf(sentry, "block_queue_len: %d\n", diskd_stats.block_queue_len);
    diskd_stats.max_away = diskd_stats.max_shmuse = 0;
    storeAppendPrintf(sentry, "\n              OPS   SUCCESS    FAIL\n");
    storeAppendPrintf(sentry, "%7s %9d %9d %7d\n",
	"open", diskd_stats.open.ops, diskd_stats.open.success, diskd_stats.open.fail);
    storeAppendPrintf(sentry, "%7s %9d %9d %7d\n",
	"create", diskd_stats.create.ops, diskd_stats.create.success, diskd_stats.create.fail);
    storeAppendPrintf(sentry, "%7s %9d %9d %7d\n",
	"close", diskd_stats.close.ops, diskd_stats.close.success, diskd_stats.close.fail);
    storeAppendPrintf(sentry, "%7s %9d %9d %7d\n",
	"unlink", diskd_stats.unlink.ops, diskd_stats.unlink.success, diskd_stats.unlink.fail);
    storeAppendPrintf(sentry, "%7s %9d %9d %7d\n",
	"read", diskd_stats.read.ops, diskd_stats.read.success, diskd_stats.read.fail);
    storeAppendPrintf(sentry, "%7s %9d %9d %7d\n",
	"write", diskd_stats.write.ops, diskd_stats.write.success, diskd_stats.write.fail);
}

/*
 * storeDiskdDirSync
 *
 * Sync any pending data. We just sit around and read the queue
 * until the data has finished writing.
 */
static void
storeDiskdDirSync(SwapDir * SD)
{
    static time_t lastmsg = 0;
    diskdinfo_t *diskdinfo = SD->fsdata;
    while (diskdinfo->away > 0) {
	if (squid_curtime > lastmsg) {
	    debug(47, 1) ("storeDiskdDirSync: %d messages away\n",
		diskdinfo->away);
	    lastmsg = squid_curtime;
	}
	storeDiskdDirCallback(SD);
    }
}


/*
 * storeDiskdDirCallback
 *
 * Handle callbacks. If we have more than magic2 requests away, we block
 * until the queue is below magic2. Otherwise, we simply return when we
 * don't get a message.
 */
int
storeDiskdDirCallback(SwapDir * SD)
{
    diomsg M;
    int x;
    diskdinfo_t *diskdinfo = SD->fsdata;
    int retval = 0;

    if (diskdinfo->away >= diskdinfo->magic2) {
	diskd_stats.block_queue_len++;
	retval = 1;		/* We might not have anything to do, but our queue
				 * is full.. */
    }
    if (diskd_stats.sent_count - diskd_stats.recv_count >
	diskd_stats.max_away) {
	diskd_stats.max_away = diskd_stats.sent_count - diskd_stats.recv_count;
    }
    while (1) {
	memset(&M, '\0', sizeof(M));
	x = msgrcv(diskdinfo->rmsgid, &M, msg_snd_rcv_sz, 0, IPC_NOWAIT);
	if (x < 0)
	    break;
	else if (x != msg_snd_rcv_sz) {
	    debug(79, 1) ("storeDiskdDirCallback: msgget returns %d\n",
		x);
	    break;
	}
	diskd_stats.recv_count++;
	diskdinfo->away--;
	storeDiskdHandle(&M);
	retval = 1;		/* Return that we've actually done some work */
	if (M.shm_offset > -1)
	    storeDiskdShmPut(SD, M.shm_offset);
    }
    return retval;
}



static void
storeDiskdDirRebuildComplete(RebuildState * rb)
{
    if (rb->log) {
	debug(47, 1) ("Done reading %s swaplog (%d entries)\n",
	    rb->sd->path, rb->n_read);
	fclose(rb->log);
	rb->log = NULL;
    } else {
	debug(47, 1) ("Done scanning %s (%d entries)\n",
	    rb->sd->path, rb->counts.scancount);
    }
    store_dirs_rebuilding--;
    storeDiskdDirCloseTmpSwapLog(rb->sd);
    storeRebuildComplete(&rb->counts);
    cbdataFree(rb);
}

static void
storeDiskdDirRebuildFromDirectory(void *data)
{
    RebuildState *rb = data;
    SwapDir *SD = rb->sd;
    LOCAL_ARRAY(char, hdr_buf, SM_PAGE_SIZE);
    StoreEntry *e = NULL;
    StoreEntry tmpe;
    cache_key key[MD5_DIGEST_CHARS];
    sfileno filn = 0;
    int count;
    int size;
    struct stat sb;
    int swap_hdr_len;
    int fd = -1;
    tlv *tlv_list;
    tlv *t;
    assert(rb != NULL);
    debug(20, 3) ("storeDiskdDirRebuildFromDirectory: DIR #%d\n", rb->sd->index);
    for (count = 0; count < rb->speed; count++) {
	assert(fd == -1);
	fd = storeDiskdDirGetNextFile(rb, &filn, &size);
	if (fd == -2) {
	    storeDiskdDirRebuildComplete(rb);
	    return;
	} else if (fd < 0) {
	    continue;
	}
	assert(fd > -1);
	/* lets get file stats here */
	if (fstat(fd, &sb) < 0) {
	    debug(20, 1) ("storeDiskdDirRebuildFromDirectory: fstat(FD %d): %s\n",
		fd, xstrerror());
	    file_close(fd);
	    store_open_disk_fd--;
	    fd = -1;
	    continue;
	}
	if ((++rb->counts.scancount & 0xFFFF) == 0)
	    debug(20, 3) ("  %s %7d files opened so far.\n",
		rb->sd->path, rb->counts.scancount);
	debug(20, 9) ("file_in: fd=%d %08X\n", fd, filn);
	statCounter.syscalls.disk.reads++;
	if (FD_READ_METHOD(fd, hdr_buf, SM_PAGE_SIZE) < 0) {
	    debug(20, 1) ("storeDiskdDirRebuildFromDirectory: read(FD %d): %s\n",
		fd, xstrerror());
	    file_close(fd);
	    store_open_disk_fd--;
	    fd = -1;
	    continue;
	}
	file_close(fd);
	store_open_disk_fd--;
	fd = -1;
	swap_hdr_len = 0;
#if USE_TRUNCATE
	if (sb.st_size == 0)
	    continue;
#endif
	tlv_list = storeSwapMetaUnpack(hdr_buf, &swap_hdr_len);
	if (tlv_list == NULL) {
	    debug(20, 1) ("storeDiskdDirRebuildFromDirectory: failed to get meta data\n");
	    /* XXX shouldn't this be a call to storeDiskdUnlink ? */
	    storeDiskdDirUnlinkFile(SD, filn);
	    continue;
	}
	debug(20, 3) ("storeDiskdDirRebuildFromDirectory: successful swap meta unpacking\n");
	memset(key, '\0', MD5_DIGEST_CHARS);
	memset(&tmpe, '\0', sizeof(StoreEntry));
	for (t = tlv_list; t; t = t->next) {
	    switch (t->type) {
	    case STORE_META_KEY:
		assert(t->length == MD5_DIGEST_CHARS);
		xmemcpy(key, t->value, MD5_DIGEST_CHARS);
		break;
#if SIZEOF_SQUID_FILE_SZ == SIZEOF_SIZE_T
	    case STORE_META_STD:
		assert(t->length == STORE_HDR_METASIZE);
		xmemcpy(&tmpe.timestamp, t->value, STORE_HDR_METASIZE);
		break;
#else
	    case STORE_META_STD_LFS:
		assert(t->length == STORE_HDR_METASIZE);
		xmemcpy(&tmpe.timestamp, t->value, STORE_HDR_METASIZE);
		break;
	    case STORE_META_STD:
		assert(t->length == STORE_HDR_METASIZE_OLD);
		{
		    struct {
			time_t timestamp;
			time_t lastref;
			time_t expires;
			time_t lastmod;
			size_t swap_file_sz;
			u_short refcount;
			u_short flags;
		    }     *tmp = t->value;
		    assert(sizeof(*tmp) == STORE_HDR_METASIZE_OLD);
		    tmpe.timestamp = tmp->timestamp;
		    tmpe.lastref = tmp->lastref;
		    tmpe.expires = tmp->expires;
		    tmpe.lastmod = tmp->lastmod;
		    tmpe.swap_file_sz = tmp->swap_file_sz;
		    tmpe.refcount = tmp->refcount;
		    tmpe.flags = tmp->flags;
		}
		break;
#endif
	    default:
		break;
	    }
	}
	storeSwapTLVFree(tlv_list);
	tlv_list = NULL;
	if (storeKeyNull(key)) {
	    debug(20, 1) ("storeDiskdDirRebuildFromDirectory: NULL key\n");
	    storeDiskdDirUnlinkFile(SD, filn);
	    continue;
	}
	tmpe.hash.key = key;
	/* check sizes */
	if (tmpe.swap_file_sz == 0) {
	    tmpe.swap_file_sz = sb.st_size;
	} else if (tmpe.swap_file_sz == sb.st_size - swap_hdr_len) {
	    tmpe.swap_file_sz = sb.st_size;
	} else if (tmpe.swap_file_sz != sb.st_size) {
	    debug(20, 1) ("storeDiskdDirRebuildFromDirectory: SIZE MISMATCH %ld!=%ld\n",
		(long int) tmpe.swap_file_sz, (long int) sb.st_size);
	    storeDiskdDirUnlinkFile(SD, filn);
	    continue;
	}
	if (EBIT_TEST(tmpe.flags, KEY_PRIVATE)) {
	    storeDiskdDirUnlinkFile(SD, filn);
	    rb->counts.badflags++;
	    continue;
	}
	e = storeGet(key);
	if (e && e->lastref >= tmpe.lastref) {
	    /* key already exists, current entry is newer */
	    /* keep old, ignore new */
	    rb->counts.dupcount++;
	    continue;
	} else if (NULL != e) {
	    /* URL already exists, this swapfile not being used */
	    /* junk old, load new */
	    storeRelease(e);	/* release old entry */
	    rb->counts.dupcount++;
	}
	rb->counts.objcount++;
	storeEntryDump(&tmpe, 5);
	e = storeDiskdDirAddDiskRestore(SD, key,
	    filn,
	    tmpe.swap_file_sz,
	    tmpe.expires,
	    tmpe.timestamp,
	    tmpe.lastref,
	    tmpe.lastmod,
	    tmpe.refcount,	/* refcount */
	    tmpe.flags,		/* flags */
	    (int) rb->flags.clean);
	storeDirSwapLog(e, SWAP_LOG_ADD);
    }
    eventAdd("storeRebuild", storeDiskdDirRebuildFromDirectory, rb, 0.0, 1);
}

static void
storeDiskdDirRebuildFromSwapLog(void *data)
{
    RebuildState *rb = data;
    SwapDir *SD = rb->sd;
    StoreEntry *e = NULL;
    storeSwapLogData s;
    size_t ss = sizeof(storeSwapLogData);
    int count;
    int used;			/* is swapfile already in use? */
    int disk_entry_newer;	/* is the log entry newer than current entry? */
    double x;
    assert(rb != NULL);
    /* load a number of objects per invocation */
    for (count = 0; count < rb->speed; count++) {
	if (fread(&s, ss, 1, rb->log) != 1) {
	    storeDiskdDirRebuildComplete(rb);
	    return;
	}
	rb->n_read++;
	/*
	 * BC: during 2.4 development, we changed the way swap file
	 * numbers are assigned and stored.  The high 16 bits used
	 * to encode the SD index number.  There used to be a call
	 * to storeDirProperFileno here that re-assigned the index
	 * bits.  Now, for backwards compatibility, we just need
	 * to mask it off.
	 */
	s.swap_filen &= 0x00FFFFFF;
	debug(20, 3) ("storeDiskdDirRebuildFromSwapLog: %s %s %08X\n",
	    swap_log_op_str[(int) s.op],
	    storeKeyText(s.key),
	    s.swap_filen);
	if (s.op == SWAP_LOG_ADD) {
	    /*
	     * Here we have some special checks for large files.
	     * I've been seeing a system crash followed by a reboot
	     * that seems to corrupt the swap log.  Squid believes
	     * that the disk holds some really large files.  It
	     * complains about using being over the high water mark
	     * and proceeds to delete files as fast as it can.  To
	     * prevent that, we call stat() on sufficiently large
	     * files (>128KB) and reject those that are missing or
	     * have the wrong size.
	     */
	    struct stat sb;
	    char *p = storeDiskdDirFullPath(SD, s.swap_filen, NULL);
	    if (s.swap_file_sz < (1 << 17)) {
		(void) 0;
	    } else if (stat(p, &sb) < 0) {
		debug(47, 2) ("its missing!: %s\n", p);
		continue;
	    } else if (sb.st_size != s.swap_file_sz) {
		debug(47, 2) ("size mismatch!: stat=%d, log=%d\n",
		    (int) sb.st_size, (int) s.swap_file_sz);
		continue;
	    } else {
		debug(47, 2) ("big file (%d bytes) checks out\n",
		    (int) s.swap_file_sz);
	    }
	} else if (s.op == SWAP_LOG_DEL) {
	    /* Delete unless we already have a newer copy */
	    if ((e = storeGet(s.key)) != NULL && s.lastref > e->lastref) {
		/*
		 * Make sure we don't unlink the file, it might be
		 * in use by a subsequent entry.  Also note that
		 * we don't have to subtract from store_swap_size
		 * because adding to store_swap_size happens in
		 * the cleanup procedure.
		 */
		storeRecycle(e);
		rb->counts.objcount--;
		rb->counts.cancelcount++;
	    }
	    continue;
	} else {
	    x = log(++rb->counts.bad_log_op) / log(10.0);
	    if (0.0 == x - (double) (int) x)
		debug(20, 1) ("WARNING: %d invalid swap log entries found\n",
		    rb->counts.bad_log_op);
	    rb->counts.invalid++;
	    continue;
	}
	if ((++rb->counts.scancount & 0xFFF) == 0) {
	    struct stat sb;
	    if (0 == fstat(fileno(rb->log), &sb))
		storeRebuildProgress(SD->index,
		    (int) sb.st_size / ss, rb->n_read);
	}
	if (!storeDiskdDirValidFileno(SD, s.swap_filen, 0)) {
	    rb->counts.invalid++;
	    continue;
	}
	if (EBIT_TEST(s.flags, KEY_PRIVATE)) {
	    rb->counts.badflags++;
	    continue;
	}
	e = storeGet(s.key);
	used = storeDiskdDirMapBitTest(SD, s.swap_filen);
	/* If this URL already exists in the cache, does the swap log
	 * appear to have a newer entry?  Compare 'lastref' from the
	 * swap log to e->lastref. */
	disk_entry_newer = e ? (s.lastref > e->lastref ? 1 : 0) : 0;
	if (used && !disk_entry_newer) {
	    /* log entry is old, ignore it */
	    rb->counts.clashcount++;
	    continue;
	} else if (used && e && e->swap_filen == s.swap_filen && e->swap_dirn == SD->index) {
	    /* swapfile taken, same URL, newer, update meta */
	    if (e->store_status == STORE_OK) {
		e->lastref = s.timestamp;
		e->timestamp = s.timestamp;
		e->expires = s.expires;
		e->lastmod = s.lastmod;
		e->flags = s.flags;
		e->refcount += s.refcount;
		storeDiskdDirUnrefObj(SD, e);
	    } else {
		debug_trap("storeDiskdDirRebuildFromSwapLog: bad condition");
		debug(20, 1) ("\tSee %s:%d\n", __FILE__, __LINE__);
	    }
	    continue;
	} else if (used) {
	    /* swapfile in use, not by this URL, log entry is newer */
	    /* This is sorta bad: the log entry should NOT be newer at this
	     * point.  If the log is dirty, the filesize check should have
	     * caught this.  If the log is clean, there should never be a
	     * newer entry. */
	    debug(20, 1) ("WARNING: newer swaplog entry for dirno %d, fileno %08X\n",
		SD->index, s.swap_filen);
	    /* I'm tempted to remove the swapfile here just to be safe,
	     * but there is a bad race condition in the NOVM version if
	     * the swapfile has recently been opened for writing, but
	     * not yet opened for reading.  Because we can't map
	     * swapfiles back to StoreEntrys, we don't know the state
	     * of the entry using that file.  */
	    /* We'll assume the existing entry is valid, probably because
	     * were in a slow rebuild and the the swap file number got taken