1/* Copyright (c) 2013 Coraid, Inc.  See COPYING for GPL terms. */
2/*
3 * aoeblk.c
4 * block device routines
5 */
6
7#include <linux/kernel.h>
8#include <linux/hdreg.h>
9#include <linux/blkdev.h>
10#include <linux/backing-dev.h>
11#include <linux/fs.h>
12#include <linux/ioctl.h>
13#include <linux/slab.h>
14#include <linux/ratelimit.h>
15#include <linux/genhd.h>
16#include <linux/netdevice.h>
17#include <linux/mutex.h>
18#include <linux/export.h>
19#include <linux/moduleparam.h>
20#include <linux/debugfs.h>
21#include <scsi/sg.h>
22#include "aoe.h"
23
24static DEFINE_MUTEX(aoeblk_mutex);
25static struct kmem_cache *buf_pool_cache;
26static struct dentry *aoe_debugfs_dir;
27
28/* GPFS needs a larger value than the default. */
29static int aoe_maxsectors;
30module_param(aoe_maxsectors, int, 0644);
31MODULE_PARM_DESC(aoe_maxsectors,
32	"When nonzero, set the maximum number of sectors per I/O request");
33
34static ssize_t aoedisk_show_state(struct device *dev,
35				  struct device_attribute *attr, char *page)
36{
37	struct gendisk *disk = dev_to_disk(dev);
38	struct aoedev *d = disk->private_data;
39
40	return snprintf(page, PAGE_SIZE,
41			"%s%s\n",
42			(d->flags & DEVFL_UP) ? "up" : "down",
43			(d->flags & DEVFL_KICKME) ? ",kickme" :
44			(d->nopen && !(d->flags & DEVFL_UP)) ? ",closewait" : "");
45	/* I'd rather see nopen exported so we can ditch closewait */
46}
47static ssize_t aoedisk_show_mac(struct device *dev,
48				struct device_attribute *attr, char *page)
49{
50	struct gendisk *disk = dev_to_disk(dev);
51	struct aoedev *d = disk->private_data;
52	struct aoetgt *t = d->targets[0];
53
54	if (t == NULL)
55		return snprintf(page, PAGE_SIZE, "none\n");
56	return snprintf(page, PAGE_SIZE, "%pm\n", t->addr);
57}
58static ssize_t aoedisk_show_netif(struct device *dev,
59				  struct device_attribute *attr, char *page)
60{
61	struct gendisk *disk = dev_to_disk(dev);
62	struct aoedev *d = disk->private_data;
63	struct net_device *nds[8], **nd, **nnd, **ne;
64	struct aoetgt **t, **te;
65	struct aoeif *ifp, *e;
66	char *p;
67
68	memset(nds, 0, sizeof nds);
69	nd = nds;
70	ne = nd + ARRAY_SIZE(nds);
71	t = d->targets;
72	te = t + d->ntargets;
73	for (; t < te && *t; t++) {
74		ifp = (*t)->ifs;
75		e = ifp + NAOEIFS;
76		for (; ifp < e && ifp->nd; ifp++) {
77			for (nnd = nds; nnd < nd; nnd++)
78				if (*nnd == ifp->nd)
79					break;
80			if (nnd == nd && nd != ne)
81				*nd++ = ifp->nd;
82		}
83	}
84
85	ne = nd;
86	nd = nds;
87	if (*nd == NULL)
88		return snprintf(page, PAGE_SIZE, "none\n");
89	for (p = page; nd < ne; nd++)
90		p += snprintf(p, PAGE_SIZE - (p-page), "%s%s",
91			p == page ? "" : ",", (*nd)->name);
92	p += snprintf(p, PAGE_SIZE - (p-page), "\n");
93	return p-page;
94}
95/* firmware version */
96static ssize_t aoedisk_show_fwver(struct device *dev,
97				  struct device_attribute *attr, char *page)
98{
99	struct gendisk *disk = dev_to_disk(dev);
100	struct aoedev *d = disk->private_data;
101
102	return snprintf(page, PAGE_SIZE, "0x%04x\n", (unsigned int) d->fw_ver);
103}
104static ssize_t aoedisk_show_payload(struct device *dev,
105				    struct device_attribute *attr, char *page)
106{
107	struct gendisk *disk = dev_to_disk(dev);
108	struct aoedev *d = disk->private_data;
109
110	return snprintf(page, PAGE_SIZE, "%lu\n", d->maxbcnt);
111}
112
113static int aoedisk_debugfs_show(struct seq_file *s, void *ignored)
114{
115	struct aoedev *d;
116	struct aoetgt **t, **te;
117	struct aoeif *ifp, *ife;
118	unsigned long flags;
119	char c;
120
121	d = s->private;
122	seq_printf(s, "rttavg: %d rttdev: %d\n",
123		d->rttavg >> RTTSCALE,
124		d->rttdev >> RTTDSCALE);
125	seq_printf(s, "nskbpool: %d\n", skb_queue_len(&d->skbpool));
126	seq_printf(s, "kicked: %ld\n", d->kicked);
127	seq_printf(s, "maxbcnt: %ld\n", d->maxbcnt);
128	seq_printf(s, "ref: %ld\n", d->ref);
129
130	spin_lock_irqsave(&d->lock, flags);
131	t = d->targets;
132	te = t + d->ntargets;
133	for (; t < te && *t; t++) {
134		c = '\t';
135		seq_printf(s, "falloc: %ld\n", (*t)->falloc);
136		seq_printf(s, "ffree: %p\n",
137			list_empty(&(*t)->ffree) ? NULL : (*t)->ffree.next);
138		seq_printf(s, "%pm:%d:%d:%d\n", (*t)->addr, (*t)->nout,
139			(*t)->maxout, (*t)->nframes);
140		seq_printf(s, "\tssthresh:%d\n", (*t)->ssthresh);
141		seq_printf(s, "\ttaint:%d\n", (*t)->taint);
142		seq_printf(s, "\tr:%d\n", (*t)->rpkts);
143		seq_printf(s, "\tw:%d\n", (*t)->wpkts);
144		ifp = (*t)->ifs;
145		ife = ifp + ARRAY_SIZE((*t)->ifs);
146		for (; ifp->nd && ifp < ife; ifp++) {
147			seq_printf(s, "%c%s", c, ifp->nd->name);
148			c = ',';
149		}
150		seq_puts(s, "\n");
151	}
152	spin_unlock_irqrestore(&d->lock, flags);
153
154	return 0;
155}
156
157static int aoe_debugfs_open(struct inode *inode, struct file *file)
158{
159	return single_open(file, aoedisk_debugfs_show, inode->i_private);
160}
161
162static DEVICE_ATTR(state, S_IRUGO, aoedisk_show_state, NULL);
163static DEVICE_ATTR(mac, S_IRUGO, aoedisk_show_mac, NULL);
164static DEVICE_ATTR(netif, S_IRUGO, aoedisk_show_netif, NULL);
165static struct device_attribute dev_attr_firmware_version = {
166	.attr = { .name = "firmware-version", .mode = S_IRUGO },
167	.show = aoedisk_show_fwver,
168};
169static DEVICE_ATTR(payload, S_IRUGO, aoedisk_show_payload, NULL);
170
171static struct attribute *aoe_attrs[] = {
172	&dev_attr_state.attr,
173	&dev_attr_mac.attr,
174	&dev_attr_netif.attr,
175	&dev_attr_firmware_version.attr,
176	&dev_attr_payload.attr,
177	NULL,
178};
179
180static const struct attribute_group attr_group = {
181	.attrs = aoe_attrs,
182};
183
184static const struct file_operations aoe_debugfs_fops = {
185	.open = aoe_debugfs_open,
186	.read = seq_read,
187	.llseek = seq_lseek,
188	.release = single_release,
189};
190
191static void
192aoedisk_add_debugfs(struct aoedev *d)
193{
194	struct dentry *entry;
195	char *p;
196
197	if (aoe_debugfs_dir == NULL)
198		return;
199	p = strchr(d->gd->disk_name, '/');
200	if (p == NULL)
201		p = d->gd->disk_name;
202	else
203		p++;
204	BUG_ON(*p == '\0');
205	entry = debugfs_create_file(p, 0444, aoe_debugfs_dir, d,
206				    &aoe_debugfs_fops);
207	if (IS_ERR_OR_NULL(entry)) {
208		pr_info("aoe: cannot create debugfs file for %s\n",
209			d->gd->disk_name);
210		return;
211	}
212	BUG_ON(d->debugfs);
213	d->debugfs = entry;
214}
215void
216aoedisk_rm_debugfs(struct aoedev *d)
217{
218	debugfs_remove(d->debugfs);
219	d->debugfs = NULL;
220}
221
222static int
223aoedisk_add_sysfs(struct aoedev *d)
224{
225	return sysfs_create_group(&disk_to_dev(d->gd)->kobj, &attr_group);
226}
227void
228aoedisk_rm_sysfs(struct aoedev *d)
229{
230	sysfs_remove_group(&disk_to_dev(d->gd)->kobj, &attr_group);
231}
232
233static int
234aoeblk_open(struct block_device *bdev, fmode_t mode)
235{
236	struct aoedev *d = bdev->bd_disk->private_data;
237	ulong flags;
238
239	if (!virt_addr_valid(d)) {
240		pr_crit("aoe: invalid device pointer in %s\n",
241			__func__);
242		WARN_ON(1);
243		return -ENODEV;
244	}
245	if (!(d->flags & DEVFL_UP) || d->flags & DEVFL_TKILL)
246		return -ENODEV;
247
248	mutex_lock(&aoeblk_mutex);
249	spin_lock_irqsave(&d->lock, flags);
250	if (d->flags & DEVFL_UP && !(d->flags & DEVFL_TKILL)) {
251		d->nopen++;
252		spin_unlock_irqrestore(&d->lock, flags);
253		mutex_unlock(&aoeblk_mutex);
254		return 0;
255	}
256	spin_unlock_irqrestore(&d->lock, flags);
257	mutex_unlock(&aoeblk_mutex);
258	return -ENODEV;
259}
260
261static void
262aoeblk_release(struct gendisk *disk, fmode_t mode)
263{
264	struct aoedev *d = disk->private_data;
265	ulong flags;
266
267	spin_lock_irqsave(&d->lock, flags);
268
269	if (--d->nopen == 0) {
270		spin_unlock_irqrestore(&d->lock, flags);
271		aoecmd_cfg(d->aoemajor, d->aoeminor);
272		return;
273	}
274	spin_unlock_irqrestore(&d->lock, flags);
275}
276
277static void
278aoeblk_request(struct request_queue *q)
279{
280	struct aoedev *d;
281	struct request *rq;
282
283	d = q->queuedata;
284	if ((d->flags & DEVFL_UP) == 0) {
285		pr_info_ratelimited("aoe: device %ld.%d is not up\n",
286			d->aoemajor, d->aoeminor);
287		while ((rq = blk_peek_request(q))) {
288			blk_start_request(rq);
289			aoe_end_request(d, rq, 1);
290		}
291		return;
292	}
293	aoecmd_work(d);
294}
295
296static int
297aoeblk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
298{
299	struct aoedev *d = bdev->bd_disk->private_data;
300
301	if ((d->flags & DEVFL_UP) == 0) {
302		printk(KERN_ERR "aoe: disk not up\n");
303		return -ENODEV;
304	}
305
306	geo->cylinders = d->geo.cylinders;
307	geo->heads = d->geo.heads;
308	geo->sectors = d->geo.sectors;
309	return 0;
310}
311
312static int
313aoeblk_ioctl(struct block_device *bdev, fmode_t mode, uint cmd, ulong arg)
314{
315	struct aoedev *d;
316
317	if (!arg)
318		return -EINVAL;
319
320	d = bdev->bd_disk->private_data;
321	if ((d->flags & DEVFL_UP) == 0) {
322		pr_err("aoe: disk not up\n");
323		return -ENODEV;
324	}
325
326	if (cmd == HDIO_GET_IDENTITY) {
327		if (!copy_to_user((void __user *) arg, &d->ident,
328			sizeof(d->ident)))
329			return 0;
330		return -EFAULT;
331	}
332
333	/* udev calls scsi_id, which uses SG_IO, resulting in noise */
334	if (cmd != SG_IO)
335		pr_info("aoe: unknown ioctl 0x%x\n", cmd);
336
337	return -ENOTTY;
338}
339
340static const struct block_device_operations aoe_bdops = {
341	.open = aoeblk_open,
342	.release = aoeblk_release,
343	.ioctl = aoeblk_ioctl,
344	.getgeo = aoeblk_getgeo,
345	.owner = THIS_MODULE,
346};
347
348/* alloc_disk and add_disk can sleep */
349void
350aoeblk_gdalloc(void *vp)
351{
352	struct aoedev *d = vp;
353	struct gendisk *gd;
354	mempool_t *mp;
355	struct request_queue *q;
356	enum { KB = 1024, MB = KB * KB, READ_AHEAD = 2 * MB, };
357	ulong flags;
358	int late = 0;
359
360	spin_lock_irqsave(&d->lock, flags);
361	if (d->flags & DEVFL_GDALLOC
362	&& !(d->flags & DEVFL_TKILL)
363	&& !(d->flags & DEVFL_GD_NOW))
364		d->flags |= DEVFL_GD_NOW;
365	else
366		late = 1;
367	spin_unlock_irqrestore(&d->lock, flags);
368	if (late)
369		return;
370
371	gd = alloc_disk(AOE_PARTITIONS);
372	if (gd == NULL) {
373		pr_err("aoe: cannot allocate disk structure for %ld.%d\n",
374			d->aoemajor, d->aoeminor);
375		goto err;
376	}
377
378	mp = mempool_create(MIN_BUFS, mempool_alloc_slab, mempool_free_slab,
379		buf_pool_cache);
380	if (mp == NULL) {
381		printk(KERN_ERR "aoe: cannot allocate bufpool for %ld.%d\n",
382			d->aoemajor, d->aoeminor);
383		goto err_disk;
384	}
385	q = blk_init_queue(aoeblk_request, &d->lock);
386	if (q == NULL) {
387		pr_err("aoe: cannot allocate block queue for %ld.%d\n",
388			d->aoemajor, d->aoeminor);
389		goto err_mempool;
390	}
391
392	spin_lock_irqsave(&d->lock, flags);
393	WARN_ON(!(d->flags & DEVFL_GD_NOW));
394	WARN_ON(!(d->flags & DEVFL_GDALLOC));
395	WARN_ON(d->flags & DEVFL_TKILL);
396	WARN_ON(d->gd);
397	WARN_ON(d->flags & DEVFL_UP);
398	blk_queue_max_hw_sectors(q, 1024);
399	q->backing_dev_info.name = "aoe";
400	q->backing_dev_info.ra_pages = READ_AHEAD / PAGE_CACHE_SIZE;
401	d->bufpool = mp;
402	d->blkq = gd->queue = q;
403	q->queuedata = d;
404	d->gd = gd;
405	if (aoe_maxsectors)
406		blk_queue_max_hw_sectors(q, aoe_maxsectors);
407	gd->major = AOE_MAJOR;
408	gd->first_minor = d->sysminor;
409	gd->fops = &aoe_bdops;
410	gd->private_data = d;
411	set_capacity(gd, d->ssize);
412	snprintf(gd->disk_name, sizeof gd->disk_name, "etherd/e%ld.%d",
413		d->aoemajor, d->aoeminor);
414
415	d->flags &= ~DEVFL_GDALLOC;
416	d->flags |= DEVFL_UP;
417
418	spin_unlock_irqrestore(&d->lock, flags);
419
420	add_disk(gd);
421	aoedisk_add_sysfs(d);
422	aoedisk_add_debugfs(d);
423
424	spin_lock_irqsave(&d->lock, flags);
425	WARN_ON(!(d->flags & DEVFL_GD_NOW));
426	d->flags &= ~DEVFL_GD_NOW;
427	spin_unlock_irqrestore(&d->lock, flags);
428	return;
429
430err_mempool:
431	mempool_destroy(mp);
432err_disk:
433	put_disk(gd);
434err:
435	spin_lock_irqsave(&d->lock, flags);
436	d->flags &= ~DEVFL_GD_NOW;
437	schedule_work(&d->work);
438	spin_unlock_irqrestore(&d->lock, flags);
439}
440
441void
442aoeblk_exit(void)
443{
444	debugfs_remove_recursive(aoe_debugfs_dir);
445	aoe_debugfs_dir = NULL;
446	kmem_cache_destroy(buf_pool_cache);
447}
448
449int __init
450aoeblk_init(void)
451{
452	buf_pool_cache = kmem_cache_create("aoe_bufs",
453					   sizeof(struct buf),
454					   0, 0, NULL);
455	if (buf_pool_cache == NULL)
456		return -ENOMEM;
457	aoe_debugfs_dir = debugfs_create_dir("aoe", NULL);
458	if (IS_ERR_OR_NULL(aoe_debugfs_dir)) {
459		pr_info("aoe: cannot create debugfs directory\n");
460		aoe_debugfs_dir = NULL;
461	}
462	return 0;
463}
464
465