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/blk-mq.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 += scnprintf(p, PAGE_SIZE - (p-page), "%s%s",
91			p == page ? "" : ",", (*nd)->name);
92	p += scnprintf(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, 0444, aoedisk_show_state, NULL);
163static DEVICE_ATTR(mac, 0444, aoedisk_show_mac, NULL);
164static DEVICE_ATTR(netif, 0444, aoedisk_show_netif, NULL);
165static struct device_attribute dev_attr_firmware_version = {
166	.attr = { .name = "firmware-version", .mode = 0444 },
167	.show = aoedisk_show_fwver,
168};
169static DEVICE_ATTR(payload, 0444, 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 aoe_attr_group = {
181	.attrs = aoe_attrs,
182};
183
184static const struct attribute_group *aoe_attr_groups[] = {
185	&aoe_attr_group,
186	NULL,
187};
188
189static const struct file_operations aoe_debugfs_fops = {
190	.open = aoe_debugfs_open,
191	.read = seq_read,
192	.llseek = seq_lseek,
193	.release = single_release,
194};
195
196static void
197aoedisk_add_debugfs(struct aoedev *d)
198{
199	char *p;
200
201	if (aoe_debugfs_dir == NULL)
202		return;
203	p = strchr(d->gd->disk_name, '/');
204	if (p == NULL)
205		p = d->gd->disk_name;
206	else
207		p++;
208	BUG_ON(*p == '\0');
209	d->debugfs = debugfs_create_file(p, 0444, aoe_debugfs_dir, d,
210					 &aoe_debugfs_fops);
211}
212void
213aoedisk_rm_debugfs(struct aoedev *d)
214{
215	debugfs_remove(d->debugfs);
216	d->debugfs = NULL;
217}
218
219static int
220aoeblk_open(struct block_device *bdev, fmode_t mode)
221{
222	struct aoedev *d = bdev->bd_disk->private_data;
223	ulong flags;
224
225	if (!virt_addr_valid(d)) {
226		pr_crit("aoe: invalid device pointer in %s\n",
227			__func__);
228		WARN_ON(1);
229		return -ENODEV;
230	}
231	if (!(d->flags & DEVFL_UP) || d->flags & DEVFL_TKILL)
232		return -ENODEV;
233
234	mutex_lock(&aoeblk_mutex);
235	spin_lock_irqsave(&d->lock, flags);
236	if (d->flags & DEVFL_UP && !(d->flags & DEVFL_TKILL)) {
237		d->nopen++;
238		spin_unlock_irqrestore(&d->lock, flags);
239		mutex_unlock(&aoeblk_mutex);
240		return 0;
241	}
242	spin_unlock_irqrestore(&d->lock, flags);
243	mutex_unlock(&aoeblk_mutex);
244	return -ENODEV;
245}
246
247static void
248aoeblk_release(struct gendisk *disk, fmode_t mode)
249{
250	struct aoedev *d = disk->private_data;
251	ulong flags;
252
253	spin_lock_irqsave(&d->lock, flags);
254
255	if (--d->nopen == 0) {
256		spin_unlock_irqrestore(&d->lock, flags);
257		aoecmd_cfg(d->aoemajor, d->aoeminor);
258		return;
259	}
260	spin_unlock_irqrestore(&d->lock, flags);
261}
262
263static blk_status_t aoeblk_queue_rq(struct blk_mq_hw_ctx *hctx,
264				    const struct blk_mq_queue_data *bd)
265{
266	struct aoedev *d = hctx->queue->queuedata;
267
268	spin_lock_irq(&d->lock);
269
270	if ((d->flags & DEVFL_UP) == 0) {
271		pr_info_ratelimited("aoe: device %ld.%d is not up\n",
272			d->aoemajor, d->aoeminor);
273		spin_unlock_irq(&d->lock);
274		blk_mq_start_request(bd->rq);
275		return BLK_STS_IOERR;
276	}
277
278	list_add_tail(&bd->rq->queuelist, &d->rq_list);
279	aoecmd_work(d);
280	spin_unlock_irq(&d->lock);
281	return BLK_STS_OK;
282}
283
284static int
285aoeblk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
286{
287	struct aoedev *d = bdev->bd_disk->private_data;
288
289	if ((d->flags & DEVFL_UP) == 0) {
290		printk(KERN_ERR "aoe: disk not up\n");
291		return -ENODEV;
292	}
293
294	geo->cylinders = d->geo.cylinders;
295	geo->heads = d->geo.heads;
296	geo->sectors = d->geo.sectors;
297	return 0;
298}
299
300static int
301aoeblk_ioctl(struct block_device *bdev, fmode_t mode, uint cmd, ulong arg)
302{
303	struct aoedev *d;
304
305	if (!arg)
306		return -EINVAL;
307
308	d = bdev->bd_disk->private_data;
309	if ((d->flags & DEVFL_UP) == 0) {
310		pr_err("aoe: disk not up\n");
311		return -ENODEV;
312	}
313
314	if (cmd == HDIO_GET_IDENTITY) {
315		if (!copy_to_user((void __user *) arg, &d->ident,
316			sizeof(d->ident)))
317			return 0;
318		return -EFAULT;
319	}
320
321	/* udev calls scsi_id, which uses SG_IO, resulting in noise */
322	if (cmd != SG_IO)
323		pr_info("aoe: unknown ioctl 0x%x\n", cmd);
324
325	return -ENOTTY;
326}
327
328static const struct block_device_operations aoe_bdops = {
329	.open = aoeblk_open,
330	.release = aoeblk_release,
331	.ioctl = aoeblk_ioctl,
332	.compat_ioctl = blkdev_compat_ptr_ioctl,
333	.getgeo = aoeblk_getgeo,
334	.owner = THIS_MODULE,
335};
336
337static const struct blk_mq_ops aoeblk_mq_ops = {
338	.queue_rq	= aoeblk_queue_rq,
339};
340
341/* alloc_disk and add_disk can sleep */
342void
343aoeblk_gdalloc(void *vp)
344{
345	struct aoedev *d = vp;
346	struct gendisk *gd;
347	mempool_t *mp;
348	struct request_queue *q;
349	struct blk_mq_tag_set *set;
350	ulong flags;
351	int late = 0;
352	int err;
353
354	spin_lock_irqsave(&d->lock, flags);
355	if (d->flags & DEVFL_GDALLOC
356	&& !(d->flags & DEVFL_TKILL)
357	&& !(d->flags & DEVFL_GD_NOW))
358		d->flags |= DEVFL_GD_NOW;
359	else
360		late = 1;
361	spin_unlock_irqrestore(&d->lock, flags);
362	if (late)
363		return;
364
365	gd = alloc_disk(AOE_PARTITIONS);
366	if (gd == NULL) {
367		pr_err("aoe: cannot allocate disk structure for %ld.%d\n",
368			d->aoemajor, d->aoeminor);
369		goto err;
370	}
371
372	mp = mempool_create(MIN_BUFS, mempool_alloc_slab, mempool_free_slab,
373		buf_pool_cache);
374	if (mp == NULL) {
375		printk(KERN_ERR "aoe: cannot allocate bufpool for %ld.%d\n",
376			d->aoemajor, d->aoeminor);
377		goto err_disk;
378	}
379
380	set = &d->tag_set;
381	set->ops = &aoeblk_mq_ops;
382	set->cmd_size = sizeof(struct aoe_req);
383	set->nr_hw_queues = 1;
384	set->queue_depth = 128;
385	set->numa_node = NUMA_NO_NODE;
386	set->flags = BLK_MQ_F_SHOULD_MERGE;
387	err = blk_mq_alloc_tag_set(set);
388	if (err) {
389		pr_err("aoe: cannot allocate tag set for %ld.%d\n",
390			d->aoemajor, d->aoeminor);
391		goto err_mempool;
392	}
393
394	q = blk_mq_init_queue(set);
395	if (IS_ERR(q)) {
396		pr_err("aoe: cannot allocate block queue for %ld.%d\n",
397			d->aoemajor, d->aoeminor);
398		blk_mq_free_tag_set(set);
399		goto err_mempool;
400	}
401
402	spin_lock_irqsave(&d->lock, flags);
403	WARN_ON(!(d->flags & DEVFL_GD_NOW));
404	WARN_ON(!(d->flags & DEVFL_GDALLOC));
405	WARN_ON(d->flags & DEVFL_TKILL);
406	WARN_ON(d->gd);
407	WARN_ON(d->flags & DEVFL_UP);
408	blk_queue_max_hw_sectors(q, BLK_DEF_MAX_SECTORS);
409	blk_queue_io_opt(q, SZ_2M);
410	d->bufpool = mp;
411	d->blkq = gd->queue = q;
412	q->queuedata = d;
413	d->gd = gd;
414	if (aoe_maxsectors)
415		blk_queue_max_hw_sectors(q, aoe_maxsectors);
416	gd->major = AOE_MAJOR;
417	gd->first_minor = d->sysminor;
418	gd->fops = &aoe_bdops;
419	gd->private_data = d;
420	set_capacity(gd, d->ssize);
421	snprintf(gd->disk_name, sizeof gd->disk_name, "etherd/e%ld.%d",
422		d->aoemajor, d->aoeminor);
423
424	d->flags &= ~DEVFL_GDALLOC;
425	d->flags |= DEVFL_UP;
426
427	spin_unlock_irqrestore(&d->lock, flags);
428
429	device_add_disk(NULL, gd, aoe_attr_groups);
430	aoedisk_add_debugfs(d);
431
432	spin_lock_irqsave(&d->lock, flags);
433	WARN_ON(!(d->flags & DEVFL_GD_NOW));
434	d->flags &= ~DEVFL_GD_NOW;
435	spin_unlock_irqrestore(&d->lock, flags);
436	return;
437
438err_mempool:
439	mempool_destroy(mp);
440err_disk:
441	put_disk(gd);
442err:
443	spin_lock_irqsave(&d->lock, flags);
444	d->flags &= ~DEVFL_GD_NOW;
445	schedule_work(&d->work);
446	spin_unlock_irqrestore(&d->lock, flags);
447}
448
449void
450aoeblk_exit(void)
451{
452	debugfs_remove_recursive(aoe_debugfs_dir);
453	aoe_debugfs_dir = NULL;
454	kmem_cache_destroy(buf_pool_cache);
455}
456
457int __init
458aoeblk_init(void)
459{
460	buf_pool_cache = kmem_cache_create("aoe_bufs",
461					   sizeof(struct buf),
462					   0, 0, NULL);
463	if (buf_pool_cache == NULL)
464		return -ENOMEM;
465	aoe_debugfs_dir = debugfs_create_dir("aoe", NULL);
466	return 0;
467}
468
469