xref: /openbmc/linux/drivers/block/aoe/aoeblk.c (revision 6d99a79c) 
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 
24 static DEFINE_MUTEX(aoeblk_mutex);
25 static struct kmem_cache *buf_pool_cache;
26 static struct dentry *aoe_debugfs_dir;
27 
28 /* GPFS needs a larger value than the default. */
29 static int aoe_maxsectors;
30 module_param(aoe_maxsectors, int, 0644);
31 MODULE_PARM_DESC(aoe_maxsectors,
32 	"When nonzero, set the maximum number of sectors per I/O request");
33 
34 static 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 }
47 static 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 }
58 static 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 */
96 static 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 }
104 static 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 
113 static 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 
157 static int aoe_debugfs_open(struct inode *inode, struct file *file)
158 {
159 	return single_open(file, aoedisk_debugfs_show, inode->i_private);
160 }
161 
162 static DEVICE_ATTR(state, 0444, aoedisk_show_state, NULL);
163 static DEVICE_ATTR(mac, 0444, aoedisk_show_mac, NULL);
164 static DEVICE_ATTR(netif, 0444, aoedisk_show_netif, NULL);
165 static struct device_attribute dev_attr_firmware_version = {
166 	.attr = { .name = "firmware-version", .mode = 0444 },
167 	.show = aoedisk_show_fwver,
168 };
169 static DEVICE_ATTR(payload, 0444, aoedisk_show_payload, NULL);
170 
171 static 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 
180 static const struct attribute_group aoe_attr_group = {
181 	.attrs = aoe_attrs,
182 };
183 
184 static const struct attribute_group *aoe_attr_groups[] = {
185 	&aoe_attr_group,
186 	NULL,
187 };
188 
189 static 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 
196 static void
197 aoedisk_add_debugfs(struct aoedev *d)
198 {
199 	struct dentry *entry;
200 	char *p;
201 
202 	if (aoe_debugfs_dir == NULL)
203 		return;
204 	p = strchr(d->gd->disk_name, '/');
205 	if (p == NULL)
206 		p = d->gd->disk_name;
207 	else
208 		p++;
209 	BUG_ON(*p == '\0');
210 	entry = debugfs_create_file(p, 0444, aoe_debugfs_dir, d,
211 				    &aoe_debugfs_fops);
212 	if (IS_ERR_OR_NULL(entry)) {
213 		pr_info("aoe: cannot create debugfs file for %s\n",
214 			d->gd->disk_name);
215 		return;
216 	}
217 	BUG_ON(d->debugfs);
218 	d->debugfs = entry;
219 }
220 void
221 aoedisk_rm_debugfs(struct aoedev *d)
222 {
223 	debugfs_remove(d->debugfs);
224 	d->debugfs = NULL;
225 }
226 
227 static int
228 aoeblk_open(struct block_device *bdev, fmode_t mode)
229 {
230 	struct aoedev *d = bdev->bd_disk->private_data;
231 	ulong flags;
232 
233 	if (!virt_addr_valid(d)) {
234 		pr_crit("aoe: invalid device pointer in %s\n",
235 			__func__);
236 		WARN_ON(1);
237 		return -ENODEV;
238 	}
239 	if (!(d->flags & DEVFL_UP) || d->flags & DEVFL_TKILL)
240 		return -ENODEV;
241 
242 	mutex_lock(&aoeblk_mutex);
243 	spin_lock_irqsave(&d->lock, flags);
244 	if (d->flags & DEVFL_UP && !(d->flags & DEVFL_TKILL)) {
245 		d->nopen++;
246 		spin_unlock_irqrestore(&d->lock, flags);
247 		mutex_unlock(&aoeblk_mutex);
248 		return 0;
249 	}
250 	spin_unlock_irqrestore(&d->lock, flags);
251 	mutex_unlock(&aoeblk_mutex);
252 	return -ENODEV;
253 }
254 
255 static void
256 aoeblk_release(struct gendisk *disk, fmode_t mode)
257 {
258 	struct aoedev *d = disk->private_data;
259 	ulong flags;
260 
261 	spin_lock_irqsave(&d->lock, flags);
262 
263 	if (--d->nopen == 0) {
264 		spin_unlock_irqrestore(&d->lock, flags);
265 		aoecmd_cfg(d->aoemajor, d->aoeminor);
266 		return;
267 	}
268 	spin_unlock_irqrestore(&d->lock, flags);
269 }
270 
271 static blk_status_t aoeblk_queue_rq(struct blk_mq_hw_ctx *hctx,
272 				    const struct blk_mq_queue_data *bd)
273 {
274 	struct aoedev *d = hctx->queue->queuedata;
275 
276 	spin_lock_irq(&d->lock);
277 
278 	if ((d->flags & DEVFL_UP) == 0) {
279 		pr_info_ratelimited("aoe: device %ld.%d is not up\n",
280 			d->aoemajor, d->aoeminor);
281 		spin_unlock_irq(&d->lock);
282 		blk_mq_start_request(bd->rq);
283 		return BLK_STS_IOERR;
284 	}
285 
286 	list_add_tail(&bd->rq->queuelist, &d->rq_list);
287 	aoecmd_work(d);
288 	spin_unlock_irq(&d->lock);
289 	return BLK_STS_OK;
290 }
291 
292 static int
293 aoeblk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
294 {
295 	struct aoedev *d = bdev->bd_disk->private_data;
296 
297 	if ((d->flags & DEVFL_UP) == 0) {
298 		printk(KERN_ERR "aoe: disk not up\n");
299 		return -ENODEV;
300 	}
301 
302 	geo->cylinders = d->geo.cylinders;
303 	geo->heads = d->geo.heads;
304 	geo->sectors = d->geo.sectors;
305 	return 0;
306 }
307 
308 static int
309 aoeblk_ioctl(struct block_device *bdev, fmode_t mode, uint cmd, ulong arg)
310 {
311 	struct aoedev *d;
312 
313 	if (!arg)
314 		return -EINVAL;
315 
316 	d = bdev->bd_disk->private_data;
317 	if ((d->flags & DEVFL_UP) == 0) {
318 		pr_err("aoe: disk not up\n");
319 		return -ENODEV;
320 	}
321 
322 	if (cmd == HDIO_GET_IDENTITY) {
323 		if (!copy_to_user((void __user *) arg, &d->ident,
324 			sizeof(d->ident)))
325 			return 0;
326 		return -EFAULT;
327 	}
328 
329 	/* udev calls scsi_id, which uses SG_IO, resulting in noise */
330 	if (cmd != SG_IO)
331 		pr_info("aoe: unknown ioctl 0x%x\n", cmd);
332 
333 	return -ENOTTY;
334 }
335 
336 static const struct block_device_operations aoe_bdops = {
337 	.open = aoeblk_open,
338 	.release = aoeblk_release,
339 	.ioctl = aoeblk_ioctl,
340 	.getgeo = aoeblk_getgeo,
341 	.owner = THIS_MODULE,
342 };
343 
344 static const struct blk_mq_ops aoeblk_mq_ops = {
345 	.queue_rq	= aoeblk_queue_rq,
346 };
347 
348 /* alloc_disk and add_disk can sleep */
349 void
350 aoeblk_gdalloc(void *vp)
351 {
352 	struct aoedev *d = vp;
353 	struct gendisk *gd;
354 	mempool_t *mp;
355 	struct request_queue *q;
356 	struct blk_mq_tag_set *set;
357 	enum { KB = 1024, MB = KB * KB, READ_AHEAD = 2 * MB, };
358 	ulong flags;
359 	int late = 0;
360 	int err;
361 
362 	spin_lock_irqsave(&d->lock, flags);
363 	if (d->flags & DEVFL_GDALLOC
364 	&& !(d->flags & DEVFL_TKILL)
365 	&& !(d->flags & DEVFL_GD_NOW))
366 		d->flags |= DEVFL_GD_NOW;
367 	else
368 		late = 1;
369 	spin_unlock_irqrestore(&d->lock, flags);
370 	if (late)
371 		return;
372 
373 	gd = alloc_disk(AOE_PARTITIONS);
374 	if (gd == NULL) {
375 		pr_err("aoe: cannot allocate disk structure for %ld.%d\n",
376 			d->aoemajor, d->aoeminor);
377 		goto err;
378 	}
379 
380 	mp = mempool_create(MIN_BUFS, mempool_alloc_slab, mempool_free_slab,
381 		buf_pool_cache);
382 	if (mp == NULL) {
383 		printk(KERN_ERR "aoe: cannot allocate bufpool for %ld.%d\n",
384 			d->aoemajor, d->aoeminor);
385 		goto err_disk;
386 	}
387 
388 	set = &d->tag_set;
389 	set->ops = &aoeblk_mq_ops;
390 	set->nr_hw_queues = 1;
391 	set->queue_depth = 128;
392 	set->numa_node = NUMA_NO_NODE;
393 	set->flags = BLK_MQ_F_SHOULD_MERGE;
394 	err = blk_mq_alloc_tag_set(set);
395 	if (err) {
396 		pr_err("aoe: cannot allocate tag set for %ld.%d\n",
397 			d->aoemajor, d->aoeminor);
398 		goto err_mempool;
399 	}
400 
401 	q = blk_mq_init_queue(set);
402 	if (IS_ERR(q)) {
403 		pr_err("aoe: cannot allocate block queue for %ld.%d\n",
404 			d->aoemajor, d->aoeminor);
405 		blk_mq_free_tag_set(set);
406 		goto err_mempool;
407 	}
408 
409 	spin_lock_irqsave(&d->lock, flags);
410 	WARN_ON(!(d->flags & DEVFL_GD_NOW));
411 	WARN_ON(!(d->flags & DEVFL_GDALLOC));
412 	WARN_ON(d->flags & DEVFL_TKILL);
413 	WARN_ON(d->gd);
414 	WARN_ON(d->flags & DEVFL_UP);
415 	blk_queue_max_hw_sectors(q, BLK_DEF_MAX_SECTORS);
416 	q->backing_dev_info->name = "aoe";
417 	q->backing_dev_info->ra_pages = READ_AHEAD / PAGE_SIZE;
418 	d->bufpool = mp;
419 	d->blkq = gd->queue = q;
420 	q->queuedata = d;
421 	d->gd = gd;
422 	if (aoe_maxsectors)
423 		blk_queue_max_hw_sectors(q, aoe_maxsectors);
424 	gd->major = AOE_MAJOR;
425 	gd->first_minor = d->sysminor;
426 	gd->fops = &aoe_bdops;
427 	gd->private_data = d;
428 	set_capacity(gd, d->ssize);
429 	snprintf(gd->disk_name, sizeof gd->disk_name, "etherd/e%ld.%d",
430 		d->aoemajor, d->aoeminor);
431 
432 	d->flags &= ~DEVFL_GDALLOC;
433 	d->flags |= DEVFL_UP;
434 
435 	spin_unlock_irqrestore(&d->lock, flags);
436 
437 	device_add_disk(NULL, gd, aoe_attr_groups);
438 	aoedisk_add_debugfs(d);
439 
440 	spin_lock_irqsave(&d->lock, flags);
441 	WARN_ON(!(d->flags & DEVFL_GD_NOW));
442 	d->flags &= ~DEVFL_GD_NOW;
443 	spin_unlock_irqrestore(&d->lock, flags);
444 	return;
445 
446 err_mempool:
447 	mempool_destroy(mp);
448 err_disk:
449 	put_disk(gd);
450 err:
451 	spin_lock_irqsave(&d->lock, flags);
452 	d->flags &= ~DEVFL_GD_NOW;
453 	schedule_work(&d->work);
454 	spin_unlock_irqrestore(&d->lock, flags);
455 }
456 
457 void
458 aoeblk_exit(void)
459 {
460 	debugfs_remove_recursive(aoe_debugfs_dir);
461 	aoe_debugfs_dir = NULL;
462 	kmem_cache_destroy(buf_pool_cache);
463 }
464 
465 int __init
466 aoeblk_init(void)
467 {
468 	buf_pool_cache = kmem_cache_create("aoe_bufs",
469 					   sizeof(struct buf),
470 					   0, 0, NULL);
471 	if (buf_pool_cache == NULL)
472 		return -ENOMEM;
473 	aoe_debugfs_dir = debugfs_create_dir("aoe", NULL);
474 	if (IS_ERR_OR_NULL(aoe_debugfs_dir)) {
475 		pr_info("aoe: cannot create debugfs directory\n");
476 		aoe_debugfs_dir = NULL;
477 	}
478 	return 0;
479 }
480 
481