2 * 2007+ Copyright (c) Evgeniy Polyakov <zbr@ioremap.net>
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
16 #include <linux/module.h>
17 #include <linux/kernel.h>
18 #include <linux/blkdev.h>
19 #include <linux/bio.h>
20 #include <linux/buffer_head.h>
21 #include <linux/connector.h>
22 #include <linux/dst.h>
23 #include <linux/device.h>
24 #include <linux/jhash.h>
25 #include <linux/idr.h>
26 #include <linux/init.h>
27 #include <linux/namei.h>
28 #include <linux/slab.h>
29 #include <linux/socket.h>
32 #include <linux/in6.h>
38 static DEFINE_MUTEX(dst_hash_lock);
39 static struct list_head *dst_hashtable;
40 static unsigned int dst_hashtable_size = 128;
41 module_param(dst_hashtable_size, uint, 0644);
43 static char dst_name[] = "Dementianting goldfish";
45 static DEFINE_IDR(dst_index_idr);
46 static struct cb_id cn_dst_id = { CN_DST_IDX, CN_DST_VAL };
49 * DST sysfs tree for device called 'storage':
51 * /sys/bus/dst/devices/storage/
52 * /sys/bus/dst/devices/storage/type : 192.168.4.80:1025
53 * /sys/bus/dst/devices/storage/size : 800
54 * /sys/bus/dst/devices/storage/name : storage
57 static int dst_dev_match(struct device *dev, struct device_driver *drv)
62 static struct bus_type dst_dev_bus_type = {
64 .match = &dst_dev_match,
67 static void dst_node_release(struct device *dev)
69 struct dst_info *info = container_of(dev, struct dst_info, device);
74 static struct device dst_node_dev = {
75 .bus = &dst_dev_bus_type,
76 .release = &dst_node_release
80 * Setting size of the node after it was changed.
82 static void dst_node_set_size(struct dst_node *n)
84 struct block_device *bdev;
86 set_capacity(n->disk, n->size >> 9);
88 bdev = bdget_disk(n->disk, 0);
90 mutex_lock(&bdev->bd_inode->i_mutex);
91 i_size_write(bdev->bd_inode, n->size);
92 mutex_unlock(&bdev->bd_inode->i_mutex);
98 * Distributed storage request processing function.
100 static int dst_request(struct request_queue *q, struct bio *bio)
102 struct dst_node *n = q->queuedata;
105 if (bio_empty_barrier(bio) && !blk_queue_discard(q)) {
107 * This is a dirty^Wnice hack, but if we complete this
108 * operation with -EOPNOTSUPP like intended, XFS
109 * will stuck and freeze the machine. This may be
110 * not particulary XFS problem though, but it is the
111 * only FS which sends empty barrier at umount time
114 * Empty barriers are not allowed anyway, see 51fd77bd9f512
115 * for example, although later it was changed to
116 * bio_rw_flagged(bio, BIO_RW_DISCARD) only, which does not
119 /* err = -EOPNOTSUPP; */
126 return dst_process_bio(n, bio);
134 * Open/close callbacks for appropriate block device.
136 static int dst_bdev_open(struct block_device *bdev, fmode_t mode)
138 struct dst_node *n = bdev->bd_disk->private_data;
144 static int dst_bdev_release(struct gendisk *disk, fmode_t mode)
146 struct dst_node *n = disk->private_data;
152 static struct block_device_operations dst_blk_ops = {
153 .open = dst_bdev_open,
154 .release = dst_bdev_release,
155 .owner = THIS_MODULE,
159 * Block layer binding - disk is created when array is fully configured
160 * by userspace request.
162 static int dst_node_create_disk(struct dst_node *n)
167 n->queue = blk_init_queue(NULL, NULL);
171 n->queue->queuedata = n;
172 blk_queue_make_request(n->queue, dst_request);
173 blk_queue_max_phys_segments(n->queue, n->max_pages);
174 blk_queue_max_hw_segments(n->queue, n->max_pages);
177 n->disk = alloc_disk(1);
179 goto err_out_free_queue;
181 if (!(n->state->permissions & DST_PERM_WRITE)) {
182 printk(KERN_INFO "DST node %s attached read-only.\n", n->name);
183 set_disk_ro(n->disk, 1);
186 if (!idr_pre_get(&dst_index_idr, GFP_KERNEL))
189 mutex_lock(&dst_hash_lock);
190 err = idr_get_new(&dst_index_idr, NULL, &index);
191 mutex_unlock(&dst_hash_lock);
195 n->disk->major = dst_major;
196 n->disk->first_minor = index;
197 n->disk->fops = &dst_blk_ops;
198 n->disk->queue = n->queue;
199 n->disk->private_data = n;
200 snprintf(n->disk->disk_name, sizeof(n->disk->disk_name),
208 blk_cleanup_queue(n->queue);
214 * Sysfs machinery: show device's size.
216 static ssize_t dst_show_size(struct device *dev,
217 struct device_attribute *attr, char *buf)
219 struct dst_info *info = container_of(dev, struct dst_info, device);
221 return sprintf(buf, "%llu\n", info->size);
225 * Show local exported device.
227 static ssize_t dst_show_local(struct device *dev,
228 struct device_attribute *attr, char *buf)
230 struct dst_info *info = container_of(dev, struct dst_info, device);
232 return sprintf(buf, "%s\n", info->local);
236 * Shows type of the remote node - device major/minor number
237 * for local nodes and address (af_inet ipv4/ipv6 only) for remote nodes.
239 static ssize_t dst_show_type(struct device *dev,
240 struct device_attribute *attr, char *buf)
242 struct dst_info *info = container_of(dev, struct dst_info, device);
243 int family = info->net.addr.sa_family;
245 if (family == AF_INET) {
246 struct sockaddr_in *sin = (struct sockaddr_in *)&info->net.addr;
247 return sprintf(buf, "%u.%u.%u.%u:%d\n",
248 NIPQUAD(sin->sin_addr.s_addr), ntohs(sin->sin_port));
249 } else if (family == AF_INET6) {
250 struct sockaddr_in6 *sin = (struct sockaddr_in6 *)
254 &sin->sin6_addr, ntohs(sin->sin6_port));
256 int i, sz = PAGE_SIZE - 2; /* 0 symbol and '\n' below */
257 int size, addrlen = info->net.addr.sa_data_len;
258 unsigned char *a = (unsigned char *)&info->net.addr.sa_data;
259 char *buf_orig = buf;
261 size = snprintf(buf, sz, "family: %d, addrlen: %u, addr: ",
266 for (i = 0; i < addrlen; ++i) {
270 size = snprintf(buf, sz, "%02x ", a[i]);
274 buf += sprintf(buf, "\n");
276 return buf - buf_orig;
281 static struct device_attribute dst_node_attrs[] = {
282 __ATTR(size, 0444, dst_show_size, NULL),
283 __ATTR(type, 0444, dst_show_type, NULL),
284 __ATTR(local, 0444, dst_show_local, NULL),
287 static int dst_create_node_attributes(struct dst_node *n)
291 for (i = 0; i < ARRAY_SIZE(dst_node_attrs); ++i) {
292 err = device_create_file(&n->info->device,
295 goto err_out_remove_all;
301 device_remove_file(&n->info->device,
307 static void dst_remove_node_attributes(struct dst_node *n)
311 for (i = 0; i < ARRAY_SIZE(dst_node_attrs); ++i)
312 device_remove_file(&n->info->device,
317 * Sysfs cleanup and initialization.
318 * Shows number of useful parameters.
320 static void dst_node_sysfs_exit(struct dst_node *n)
323 dst_remove_node_attributes(n);
324 device_unregister(&n->info->device);
329 static int dst_node_sysfs_init(struct dst_node *n)
333 n->info = kzalloc(sizeof(struct dst_info), GFP_KERNEL);
337 memcpy(&n->info->device, &dst_node_dev, sizeof(struct device));
338 n->info->size = n->size;
340 dev_set_name(&n->info->device, "dst-%s", n->name);
341 err = device_register(&n->info->device);
343 dprintk(KERN_ERR "Failed to register node '%s', err: %d.\n",
348 dst_create_node_attributes(n);
359 * DST node hash tables machinery.
361 static inline unsigned int dst_hash(char *str, unsigned int size)
363 return jhash(str, size, 0) % dst_hashtable_size;
366 static void dst_node_remove(struct dst_node *n)
368 mutex_lock(&dst_hash_lock);
369 list_del_init(&n->node_entry);
370 mutex_unlock(&dst_hash_lock);
373 static void dst_node_add(struct dst_node *n)
375 unsigned hash = dst_hash(n->name, sizeof(n->name));
377 mutex_lock(&dst_hash_lock);
378 list_add_tail(&n->node_entry, &dst_hashtable[hash]);
379 mutex_unlock(&dst_hash_lock);
383 * Cleaning node when it is about to be freed.
384 * There are still users of the socket though,
385 * so connection cleanup should be protected.
387 static void dst_node_cleanup(struct dst_node *n)
389 struct dst_state *st = n->state;
395 blk_cleanup_queue(n->queue);
397 mutex_lock(&dst_hash_lock);
398 idr_remove(&dst_index_idr, n->disk->first_minor);
399 mutex_unlock(&dst_hash_lock);
405 sync_blockdev(n->bdev);
406 close_bdev_exclusive(n->bdev, FMODE_READ|FMODE_WRITE);
411 dst_state_exit_connected(st);
412 dst_state_unlock(st);
414 wake_up(&st->thread_wait);
421 * Free security attributes attached to given node.
423 static void dst_security_exit(struct dst_node *n)
425 struct dst_secure *s, *tmp;
427 list_for_each_entry_safe(s, tmp, &n->security_list, sec_entry) {
428 list_del(&s->sec_entry);
434 * Free node when there are no more users.
435 * Actually node has to be freed on behalf od userspace process,
436 * since there are number of threads, which are embedded in the
437 * node, so they can not exit and free node from there, that is
438 * why there is a wakeup if reference counter is not equal to zero.
440 void dst_node_put(struct dst_node *n)
445 dprintk("%s: n: %p, refcnt: %d.\n",
446 __func__, n, atomic_read(&n->refcnt));
448 if (atomic_dec_and_test(&n->refcnt)) {
450 n->trans_scan_timeout = 0;
452 thread_pool_destroy(n->pool);
453 dst_node_sysfs_exit(n);
454 dst_node_crypto_exit(n);
455 dst_security_exit(n);
456 dst_node_trans_exit(n);
460 dprintk("%s: freed n: %p.\n", __func__, n);
467 * Setting up export device: lookup by the name, get its size
468 * and setup listening socket, which will accept clients, which
469 * will submit IO for given storage.
471 static int dst_setup_export(struct dst_node *n, struct dst_ctl *ctl,
472 struct dst_export_ctl *le)
476 snprintf(n->info->local, sizeof(n->info->local), "%s", le->device);
478 n->bdev = open_bdev_exclusive(le->device, FMODE_READ|FMODE_WRITE, NULL);
480 return PTR_ERR(n->bdev);
483 n->size = min_t(loff_t, n->bdev->bd_inode->i_size, n->size);
485 n->size = n->bdev->bd_inode->i_size;
487 n->info->size = n->size;
488 err = dst_node_init_listened(n, le);
490 goto err_out_cleanup;
495 close_bdev_exclusive(n->bdev, FMODE_READ|FMODE_WRITE);
501 /* Empty thread pool callbacks for the network processing threads. */
502 static inline void *dst_thread_network_init(void *data)
504 dprintk("%s: data: %p.\n", __func__, data);
508 static inline void dst_thread_network_cleanup(void *data)
510 dprintk("%s: data: %p.\n", __func__, data);
514 * Allocate DST node and initialize some of its parameters.
516 static struct dst_node *dst_alloc_node(struct dst_ctl *ctl,
517 int (*start)(struct dst_node *),
523 n = kzalloc(sizeof(struct dst_node), GFP_KERNEL);
527 INIT_LIST_HEAD(&n->node_entry);
529 INIT_LIST_HEAD(&n->security_list);
530 mutex_init(&n->security_lock);
532 init_waitqueue_head(&n->wait);
534 n->trans_scan_timeout = msecs_to_jiffies(ctl->trans_scan_timeout);
535 if (!n->trans_scan_timeout)
536 n->trans_scan_timeout = HZ;
538 n->trans_max_retries = ctl->trans_max_retries;
539 if (!n->trans_max_retries)
540 n->trans_max_retries = 10;
543 * Pretty much arbitrary default numbers.
544 * 32 matches maximum number of pages in bio originated from ext3 (31).
546 n->max_pages = ctl->max_pages;
550 if (n->max_pages > 1024)
556 atomic_set(&n->refcnt, 1);
557 atomic_long_set(&n->gen, 0);
558 snprintf(n->name, sizeof(n->name), "%s", ctl->name);
560 err = dst_node_sysfs_init(n);
564 n->pool = thread_pool_create(num, n->name, dst_thread_network_init,
565 dst_thread_network_cleanup, n);
566 if (IS_ERR(n->pool)) {
567 err = PTR_ERR(n->pool);
568 goto err_out_sysfs_exit;
571 dprintk("%s: n: %p, name: %s.\n", __func__, n, n->name);
576 dst_node_sysfs_exit(n);
583 * Starting a node, connected to the remote server:
584 * register block device and initialize transaction mechanism.
585 * In revers order though.
587 * It will autonegotiate some parameters with the remote node
588 * and update local if needed.
590 * Transaction initialization should be the last thing before
591 * starting the node, since transaction should include not only
592 * block IO, but also crypto related data (if any), which are
593 * initialized separately.
595 static int dst_start_remote(struct dst_node *n)
599 err = dst_node_trans_init(n, sizeof(struct dst_trans));
603 err = dst_node_create_disk(n);
607 dst_node_set_size(n);
610 dprintk("DST: started remote node '%s', minor: %d.\n",
611 n->name, n->disk->first_minor);
617 * Adding remote node and initialize connection.
619 static int dst_add_remote(struct dst_node *n, struct dst_ctl *ctl,
620 void *data, unsigned int size)
623 struct dst_network_ctl *rctl = data;
628 if (size != sizeof(struct dst_network_ctl))
631 n = dst_alloc_node(ctl, dst_start_remote, 1);
635 memcpy(&n->info->net, rctl, sizeof(struct dst_network_ctl));
636 err = dst_node_init_connected(n, rctl);
650 * Adding export node: initializing block device and listening socket.
652 static int dst_add_export(struct dst_node *n, struct dst_ctl *ctl,
653 void *data, unsigned int size)
656 struct dst_export_ctl *le = data;
661 if (size != sizeof(struct dst_export_ctl))
664 n = dst_alloc_node(ctl, dst_start_export, 2);
668 err = dst_setup_export(n, ctl, le);
681 static int dst_node_remove_unload(struct dst_node *n)
683 printk(KERN_INFO "STOPPED name: '%s', size: %llu.\n",
687 del_gendisk(n->disk);
690 dst_node_sysfs_exit(n);
693 * This is not a hack. Really.
694 * Node's reference counter allows to implement fine grained
695 * node freeing, but since all transactions (which hold node's
696 * reference counter) are processed in the dedicated thread,
697 * it is possible that reference will hit zero in that thread,
698 * so we will not be able to exit thread and cleanup the node.
700 * So, we remove disk, so no new activity is possible, and
701 * wait until all pending transaction are completed (either
702 * in receiving thread or by timeout in workqueue), in this
703 * case reference counter will be less or equal to 2 (once set in
704 * dst_alloc_node() and then in connector message parser;
705 * or when we force module unloading, and connector message
706 * parser does not hold a reference, in this case reference
707 * counter will be equal to 1),
708 * and subsequent dst_node_put() calls will free the node.
710 dprintk("%s: going to sleep with %d refcnt.\n",
711 __func__, atomic_read(&n->refcnt));
712 wait_event(n->wait, atomic_read(&n->refcnt) <= 2);
719 * Remove node from the hash table.
721 static int dst_del_node(struct dst_node *n, struct dst_ctl *ctl,
722 void *data, unsigned int size)
727 return dst_node_remove_unload(n);
731 * Initialize crypto processing for given node.
733 static int dst_crypto_init(struct dst_node *n, struct dst_ctl *ctl,
734 void *data, unsigned int size)
736 struct dst_crypto_ctl *crypto = data;
741 if (size != sizeof(struct dst_crypto_ctl) + crypto->hash_keysize +
742 crypto->cipher_keysize)
748 return dst_node_crypto_init(n, crypto);
752 * Security attributes for given node.
754 static int dst_security_init(struct dst_node *n, struct dst_ctl *ctl,
755 void *data, unsigned int size)
757 struct dst_secure *s;
762 if (size != sizeof(struct dst_secure_user))
765 s = kmalloc(sizeof(struct dst_secure), GFP_KERNEL);
769 memcpy(&s->sec, data, size);
771 mutex_lock(&n->security_lock);
772 list_add_tail(&s->sec_entry, &n->security_list);
773 mutex_unlock(&n->security_lock);
781 static int dst_start_node(struct dst_node *n, struct dst_ctl *ctl,
782 void *data, unsigned int size)
796 printk(KERN_INFO "STARTED name: '%s', size: %llu.\n", n->name, n->size);
800 typedef int (*dst_command_func)(struct dst_node *n, struct dst_ctl *ctl,
801 void *data, unsigned int size);
804 * List of userspace commands.
806 static dst_command_func dst_commands[] = {
807 [DST_ADD_REMOTE] = &dst_add_remote,
808 [DST_ADD_EXPORT] = &dst_add_export,
809 [DST_DEL_NODE] = &dst_del_node,
810 [DST_CRYPTO] = &dst_crypto_init,
811 [DST_SECURITY] = &dst_security_init,
812 [DST_START] = &dst_start_node,
816 * Configuration parser.
818 static void cn_dst_callback(struct cn_msg *msg, struct netlink_skb_parms *nsp)
822 struct dst_ctl_ack ack;
823 struct dst_node *n = NULL, *tmp;
826 if (!cap_raised(nsp->eff_cap, CAP_SYS_ADMIN)) {
831 if (msg->len < sizeof(struct dst_ctl)) {
836 ctl = (struct dst_ctl *)msg->data;
838 if (ctl->cmd >= DST_CMD_MAX) {
842 hash = dst_hash(ctl->name, sizeof(ctl->name));
844 mutex_lock(&dst_hash_lock);
845 list_for_each_entry(tmp, &dst_hashtable[hash], node_entry) {
846 if (!memcmp(tmp->name, ctl->name, sizeof(tmp->name))) {
852 mutex_unlock(&dst_hash_lock);
854 err = dst_commands[ctl->cmd](n, ctl, msg->data + sizeof(struct dst_ctl),
855 msg->len - sizeof(struct dst_ctl));
859 memcpy(&ack.msg, msg, sizeof(struct cn_msg));
861 ack.msg.ack = msg->ack + 1;
862 ack.msg.len = sizeof(struct dst_ctl_ack) - sizeof(struct cn_msg);
866 cn_netlink_send(&ack.msg, 0, GFP_KERNEL);
870 * Global initialization: sysfs, hash table, block device registration,
871 * connector and various caches.
873 static int __init dst_sysfs_init(void)
875 return bus_register(&dst_dev_bus_type);
878 static void dst_sysfs_exit(void)
880 bus_unregister(&dst_dev_bus_type);
883 static int __init dst_hashtable_init(void)
887 dst_hashtable = kcalloc(dst_hashtable_size, sizeof(struct list_head),
892 for (i = 0; i < dst_hashtable_size; ++i)
893 INIT_LIST_HEAD(&dst_hashtable[i]);
898 static void dst_hashtable_exit(void)
901 struct dst_node *n, *tmp;
903 for (i = 0; i < dst_hashtable_size; ++i) {
904 list_for_each_entry_safe(n, tmp, &dst_hashtable[i], node_entry) {
905 dst_node_remove_unload(n);
909 kfree(dst_hashtable);
912 static int __init dst_sys_init(void)
916 err = dst_hashtable_init();
920 err = dst_export_init();
922 goto err_out_hashtable_exit;
924 err = register_blkdev(dst_major, DST_NAME);
926 goto err_out_export_exit;
930 err = dst_sysfs_init();
932 goto err_out_unregister;
934 err = cn_add_callback(&cn_dst_id, "DST", cn_dst_callback);
936 goto err_out_sysfs_exit;
938 printk(KERN_INFO "Distributed storage, '%s' release.\n", dst_name);
945 unregister_blkdev(dst_major, DST_NAME);
948 err_out_hashtable_exit:
949 dst_hashtable_exit();
954 static void __exit dst_sys_exit(void)
956 cn_del_callback(&cn_dst_id);
957 unregister_blkdev(dst_major, DST_NAME);
958 dst_hashtable_exit();
963 module_init(dst_sys_init);
964 module_exit(dst_sys_exit);
966 MODULE_DESCRIPTION("Distributed storage");
967 MODULE_AUTHOR("Evgeniy Polyakov <zbr@ioremap.net>");
968 MODULE_LICENSE("GPL");