aoeblk.c revision 0c6f0e7920f39b28bdbe5f134f3e592542332d87
1/* Copyright (c) 2004 Coraid, Inc.  See COPYING for GPL terms. */
2/*
3 * aoeblk.c
4 * block device routines
5 */
6
7#include <linux/hdreg.h>
8#include <linux/blkdev.h>
9#include <linux/fs.h>
10#include <linux/ioctl.h>
11#include <linux/genhd.h>
12#include <linux/netdevice.h>
13#include "aoe.h"
14
15static kmem_cache_t *buf_pool_cache;
16
17/* add attributes for our block devices in sysfs */
18static ssize_t aoedisk_show_state(struct gendisk * disk, char *page)
19{
20	struct aoedev *d = disk->private_data;
21
22	return snprintf(page, PAGE_SIZE,
23			"%s%s\n",
24			(d->flags & DEVFL_UP) ? "up" : "down",
25			(d->flags & DEVFL_CLOSEWAIT) ? ",closewait" : "");
26}
27static ssize_t aoedisk_show_mac(struct gendisk * disk, char *page)
28{
29	struct aoedev *d = disk->private_data;
30
31	return snprintf(page, PAGE_SIZE, "%012llx\n",
32			(unsigned long long)mac_addr(d->addr));
33}
34static ssize_t aoedisk_show_netif(struct gendisk * disk, char *page)
35{
36	struct aoedev *d = disk->private_data;
37
38	return snprintf(page, PAGE_SIZE, "%s\n", d->ifp->name);
39}
40
41static struct disk_attribute disk_attr_state = {
42	.attr = {.name = "state", .mode = S_IRUGO },
43	.show = aoedisk_show_state
44};
45static struct disk_attribute disk_attr_mac = {
46	.attr = {.name = "mac", .mode = S_IRUGO },
47	.show = aoedisk_show_mac
48};
49static struct disk_attribute disk_attr_netif = {
50	.attr = {.name = "netif", .mode = S_IRUGO },
51	.show = aoedisk_show_netif
52};
53
54static void
55aoedisk_add_sysfs(struct aoedev *d)
56{
57	sysfs_create_file(&d->gd->kobj, &disk_attr_state.attr);
58	sysfs_create_file(&d->gd->kobj, &disk_attr_mac.attr);
59	sysfs_create_file(&d->gd->kobj, &disk_attr_netif.attr);
60}
61void
62aoedisk_rm_sysfs(struct aoedev *d)
63{
64	sysfs_remove_link(&d->gd->kobj, "state");
65	sysfs_remove_link(&d->gd->kobj, "mac");
66	sysfs_remove_link(&d->gd->kobj, "netif");
67}
68
69static int
70aoeblk_open(struct inode *inode, struct file *filp)
71{
72	struct aoedev *d;
73	ulong flags;
74
75	d = inode->i_bdev->bd_disk->private_data;
76
77	spin_lock_irqsave(&d->lock, flags);
78	if (d->flags & DEVFL_UP) {
79		d->nopen++;
80		spin_unlock_irqrestore(&d->lock, flags);
81		return 0;
82	}
83	spin_unlock_irqrestore(&d->lock, flags);
84	return -ENODEV;
85}
86
87static int
88aoeblk_release(struct inode *inode, struct file *filp)
89{
90	struct aoedev *d;
91	ulong flags;
92
93	d = inode->i_bdev->bd_disk->private_data;
94
95	spin_lock_irqsave(&d->lock, flags);
96
97	if (--d->nopen == 0 && (d->flags & DEVFL_CLOSEWAIT)) {
98		d->flags &= ~DEVFL_CLOSEWAIT;
99		spin_unlock_irqrestore(&d->lock, flags);
100		aoecmd_cfg(d->aoemajor, d->aoeminor);
101		return 0;
102	}
103	spin_unlock_irqrestore(&d->lock, flags);
104
105	return 0;
106}
107
108static int
109aoeblk_make_request(request_queue_t *q, struct bio *bio)
110{
111	struct aoedev *d;
112	struct buf *buf;
113	struct sk_buff *sl;
114	ulong flags;
115
116	blk_queue_bounce(q, &bio);
117
118	d = bio->bi_bdev->bd_disk->private_data;
119	buf = mempool_alloc(d->bufpool, GFP_NOIO);
120	if (buf == NULL) {
121		printk(KERN_INFO "aoe: aoeblk_make_request: buf allocation "
122			"failure\n");
123		bio_endio(bio, bio->bi_size, -ENOMEM);
124		return 0;
125	}
126	memset(buf, 0, sizeof(*buf));
127	INIT_LIST_HEAD(&buf->bufs);
128	buf->start_time = jiffies;
129	buf->bio = bio;
130	buf->resid = bio->bi_size;
131	buf->sector = bio->bi_sector;
132	buf->bv = buf->bio->bi_io_vec;
133	buf->bv_resid = buf->bv->bv_len;
134	buf->bufaddr = page_address(buf->bv->bv_page) + buf->bv->bv_offset;
135
136	spin_lock_irqsave(&d->lock, flags);
137
138	if ((d->flags & DEVFL_UP) == 0) {
139		printk(KERN_INFO "aoe: aoeblk_make_request: device %ld.%ld is not up\n",
140			d->aoemajor, d->aoeminor);
141		spin_unlock_irqrestore(&d->lock, flags);
142		mempool_free(buf, d->bufpool);
143		bio_endio(bio, bio->bi_size, -ENXIO);
144		return 0;
145	}
146
147	list_add_tail(&buf->bufs, &d->bufq);
148	aoecmd_work(d);
149
150	sl = d->skblist;
151	d->skblist = NULL;
152
153	spin_unlock_irqrestore(&d->lock, flags);
154
155	aoenet_xmit(sl);
156	return 0;
157}
158
159/* This ioctl implementation expects userland to have the device node
160 * permissions set so that only priviledged users can open an aoe
161 * block device directly.
162 */
163static int
164aoeblk_ioctl(struct inode *inode, struct file *filp, uint cmd, ulong arg)
165{
166	struct aoedev *d;
167
168	if (!arg)
169		return -EINVAL;
170
171	d = inode->i_bdev->bd_disk->private_data;
172	if ((d->flags & DEVFL_UP) == 0) {
173		printk(KERN_ERR "aoe: aoeblk_ioctl: disk not up\n");
174		return -ENODEV;
175	}
176
177	if (cmd == HDIO_GETGEO) {
178		d->geo.start = get_start_sect(inode->i_bdev);
179		if (!copy_to_user((void __user *) arg, &d->geo, sizeof d->geo))
180			return 0;
181		return -EFAULT;
182	}
183	printk(KERN_INFO "aoe: aoeblk_ioctl: unknown ioctl %d\n", cmd);
184	return -EINVAL;
185}
186
187static struct block_device_operations aoe_bdops = {
188	.open = aoeblk_open,
189	.release = aoeblk_release,
190	.ioctl = aoeblk_ioctl,
191	.owner = THIS_MODULE,
192};
193
194/* alloc_disk and add_disk can sleep */
195void
196aoeblk_gdalloc(void *vp)
197{
198	struct aoedev *d = vp;
199	struct gendisk *gd;
200	ulong flags;
201
202	gd = alloc_disk(AOE_PARTITIONS);
203	if (gd == NULL) {
204		printk(KERN_ERR "aoe: aoeblk_gdalloc: cannot allocate disk "
205			"structure for %ld.%ld\n", d->aoemajor, d->aoeminor);
206		spin_lock_irqsave(&d->lock, flags);
207		d->flags &= ~DEVFL_WORKON;
208		spin_unlock_irqrestore(&d->lock, flags);
209		return;
210	}
211
212	d->bufpool = mempool_create(MIN_BUFS,
213				    mempool_alloc_slab, mempool_free_slab,
214				    buf_pool_cache);
215	if (d->bufpool == NULL) {
216		printk(KERN_ERR "aoe: aoeblk_gdalloc: cannot allocate bufpool "
217			"for %ld.%ld\n", d->aoemajor, d->aoeminor);
218		put_disk(gd);
219		spin_lock_irqsave(&d->lock, flags);
220		d->flags &= ~DEVFL_WORKON;
221		spin_unlock_irqrestore(&d->lock, flags);
222		return;
223	}
224
225	spin_lock_irqsave(&d->lock, flags);
226	blk_queue_make_request(&d->blkq, aoeblk_make_request);
227	gd->major = AOE_MAJOR;
228	gd->first_minor = d->sysminor * AOE_PARTITIONS;
229	gd->fops = &aoe_bdops;
230	gd->private_data = d;
231	gd->capacity = d->ssize;
232	snprintf(gd->disk_name, sizeof gd->disk_name, "etherd/e%ld.%ld",
233		d->aoemajor, d->aoeminor);
234
235	gd->queue = &d->blkq;
236	d->gd = gd;
237	d->flags &= ~DEVFL_WORKON;
238	d->flags |= DEVFL_UP;
239
240	spin_unlock_irqrestore(&d->lock, flags);
241
242	add_disk(gd);
243	aoedisk_add_sysfs(d);
244
245	printk(KERN_INFO "aoe: %012llx e%lu.%lu v%04x has %llu "
246		"sectors\n", (unsigned long long)mac_addr(d->addr),
247		d->aoemajor, d->aoeminor,
248		d->fw_ver, (long long)d->ssize);
249}
250
251void
252aoeblk_exit(void)
253{
254	kmem_cache_destroy(buf_pool_cache);
255}
256
257int __init
258aoeblk_init(void)
259{
260	buf_pool_cache = kmem_cache_create("aoe_bufs",
261					   sizeof(struct buf),
262					   0, 0, NULL, NULL);
263	if (buf_pool_cache == NULL)
264		return -ENOMEM;
265
266	return 0;
267}
268
269