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/netdevice.h>
16 #include <linux/mutex.h>
17 #include <linux/export.h>
18 #include <linux/moduleparam.h>
19 #include <linux/debugfs.h>
20 #include <scsi/sg.h>
21 #include "aoe.h"
22 
23 static DEFINE_MUTEX(aoeblk_mutex);
24 static struct kmem_cache *buf_pool_cache;
25 static struct dentry *aoe_debugfs_dir;
26 
27 /* GPFS needs a larger value than the default. */
28 static int aoe_maxsectors;
29 module_param(aoe_maxsectors, int, 0644);
30 MODULE_PARM_DESC(aoe_maxsectors,
31 	"When nonzero, set the maximum number of sectors per I/O request");
32 
aoedisk_show_state(struct device * dev,struct device_attribute * attr,char * page)33 static ssize_t aoedisk_show_state(struct device *dev,
34 				  struct device_attribute *attr, char *page)
35 {
36 	struct gendisk *disk = dev_to_disk(dev);
37 	struct aoedev *d = disk->private_data;
38 
39 	return sysfs_emit(page, "%s%s\n",
40 			(d->flags & DEVFL_UP) ? "up" : "down",
41 			(d->flags & DEVFL_KICKME) ? ",kickme" :
42 			(d->nopen && !(d->flags & DEVFL_UP)) ? ",closewait" : "");
43 	/* I'd rather see nopen exported so we can ditch closewait */
44 }
aoedisk_show_mac(struct device * dev,struct device_attribute * attr,char * page)45 static ssize_t aoedisk_show_mac(struct device *dev,
46 				struct device_attribute *attr, char *page)
47 {
48 	struct gendisk *disk = dev_to_disk(dev);
49 	struct aoedev *d = disk->private_data;
50 	struct aoetgt *t = d->targets[0];
51 
52 	if (t == NULL)
53 		return sysfs_emit(page, "none\n");
54 	return sysfs_emit(page, "%pm\n", t->addr);
55 }
aoedisk_show_netif(struct device * dev,struct device_attribute * attr,char * page)56 static ssize_t aoedisk_show_netif(struct device *dev,
57 				  struct device_attribute *attr, char *page)
58 {
59 	struct gendisk *disk = dev_to_disk(dev);
60 	struct aoedev *d = disk->private_data;
61 	struct net_device *nds[8], **nd, **nnd, **ne;
62 	struct aoetgt **t, **te;
63 	struct aoeif *ifp, *e;
64 	char *p;
65 
66 	memset(nds, 0, sizeof nds);
67 	nd = nds;
68 	ne = nd + ARRAY_SIZE(nds);
69 	t = d->targets;
70 	te = t + d->ntargets;
71 	for (; t < te && *t; t++) {
72 		ifp = (*t)->ifs;
73 		e = ifp + NAOEIFS;
74 		for (; ifp < e && ifp->nd; ifp++) {
75 			for (nnd = nds; nnd < nd; nnd++)
76 				if (*nnd == ifp->nd)
77 					break;
78 			if (nnd == nd && nd != ne)
79 				*nd++ = ifp->nd;
80 		}
81 	}
82 
83 	ne = nd;
84 	nd = nds;
85 	if (*nd == NULL)
86 		return sysfs_emit(page, "none\n");
87 	for (p = page; nd < ne; nd++)
88 		p += scnprintf(p, PAGE_SIZE - (p-page), "%s%s",
89 			p == page ? "" : ",", (*nd)->name);
90 	p += scnprintf(p, PAGE_SIZE - (p-page), "\n");
91 	return p-page;
92 }
93 /* firmware version */
aoedisk_show_fwver(struct device * dev,struct device_attribute * attr,char * page)94 static ssize_t aoedisk_show_fwver(struct device *dev,
95 				  struct device_attribute *attr, char *page)
96 {
97 	struct gendisk *disk = dev_to_disk(dev);
98 	struct aoedev *d = disk->private_data;
99 
100 	return sysfs_emit(page, "0x%04x\n", (unsigned int) d->fw_ver);
101 }
aoedisk_show_payload(struct device * dev,struct device_attribute * attr,char * page)102 static ssize_t aoedisk_show_payload(struct device *dev,
103 				    struct device_attribute *attr, char *page)
104 {
105 	struct gendisk *disk = dev_to_disk(dev);
106 	struct aoedev *d = disk->private_data;
107 
108 	return sysfs_emit(page, "%lu\n", d->maxbcnt);
109 }
110 
aoedisk_debugfs_show(struct seq_file * s,void * ignored)111 static int aoedisk_debugfs_show(struct seq_file *s, void *ignored)
112 {
113 	struct aoedev *d;
114 	struct aoetgt **t, **te;
115 	struct aoeif *ifp, *ife;
116 	unsigned long flags;
117 	char c;
118 
119 	d = s->private;
120 	seq_printf(s, "rttavg: %d rttdev: %d\n",
121 		d->rttavg >> RTTSCALE,
122 		d->rttdev >> RTTDSCALE);
123 	seq_printf(s, "nskbpool: %d\n", skb_queue_len(&d->skbpool));
124 	seq_printf(s, "kicked: %ld\n", d->kicked);
125 	seq_printf(s, "maxbcnt: %ld\n", d->maxbcnt);
126 	seq_printf(s, "ref: %ld\n", d->ref);
127 
128 	spin_lock_irqsave(&d->lock, flags);
129 	t = d->targets;
130 	te = t + d->ntargets;
131 	for (; t < te && *t; t++) {
132 		c = '\t';
133 		seq_printf(s, "falloc: %ld\n", (*t)->falloc);
134 		seq_printf(s, "ffree: %p\n",
135 			list_empty(&(*t)->ffree) ? NULL : (*t)->ffree.next);
136 		seq_printf(s, "%pm:%d:%d:%d\n", (*t)->addr, (*t)->nout,
137 			(*t)->maxout, (*t)->nframes);
138 		seq_printf(s, "\tssthresh:%d\n", (*t)->ssthresh);
139 		seq_printf(s, "\ttaint:%d\n", (*t)->taint);
140 		seq_printf(s, "\tr:%d\n", (*t)->rpkts);
141 		seq_printf(s, "\tw:%d\n", (*t)->wpkts);
142 		ifp = (*t)->ifs;
143 		ife = ifp + ARRAY_SIZE((*t)->ifs);
144 		for (; ifp->nd && ifp < ife; ifp++) {
145 			seq_printf(s, "%c%s", c, ifp->nd->name);
146 			c = ',';
147 		}
148 		seq_puts(s, "\n");
149 	}
150 	spin_unlock_irqrestore(&d->lock, flags);
151 
152 	return 0;
153 }
154 
aoe_debugfs_open(struct inode * inode,struct file * file)155 static int aoe_debugfs_open(struct inode *inode, struct file *file)
156 {
157 	return single_open(file, aoedisk_debugfs_show, inode->i_private);
158 }
159 
160 static DEVICE_ATTR(state, 0444, aoedisk_show_state, NULL);
161 static DEVICE_ATTR(mac, 0444, aoedisk_show_mac, NULL);
162 static DEVICE_ATTR(netif, 0444, aoedisk_show_netif, NULL);
163 static struct device_attribute dev_attr_firmware_version = {
164 	.attr = { .name = "firmware-version", .mode = 0444 },
165 	.show = aoedisk_show_fwver,
166 };
167 static DEVICE_ATTR(payload, 0444, aoedisk_show_payload, NULL);
168 
169 static struct attribute *aoe_attrs[] = {
170 	&dev_attr_state.attr,
171 	&dev_attr_mac.attr,
172 	&dev_attr_netif.attr,
173 	&dev_attr_firmware_version.attr,
174 	&dev_attr_payload.attr,
175 	NULL,
176 };
177 
178 static const struct attribute_group aoe_attr_group = {
179 	.attrs = aoe_attrs,
180 };
181 
182 static const struct attribute_group *aoe_attr_groups[] = {
183 	&aoe_attr_group,
184 	NULL,
185 };
186 
187 static const struct file_operations aoe_debugfs_fops = {
188 	.open = aoe_debugfs_open,
189 	.read = seq_read,
190 	.llseek = seq_lseek,
191 	.release = single_release,
192 };
193 
194 static void
aoedisk_add_debugfs(struct aoedev * d)195 aoedisk_add_debugfs(struct aoedev *d)
196 {
197 	char *p;
198 
199 	if (aoe_debugfs_dir == NULL)
200 		return;
201 	p = strchr(d->gd->disk_name, '/');
202 	if (p == NULL)
203 		p = d->gd->disk_name;
204 	else
205 		p++;
206 	BUG_ON(*p == '\0');
207 	d->debugfs = debugfs_create_file(p, 0444, aoe_debugfs_dir, d,
208 					 &aoe_debugfs_fops);
209 }
210 void
aoedisk_rm_debugfs(struct aoedev * d)211 aoedisk_rm_debugfs(struct aoedev *d)
212 {
213 	debugfs_remove(d->debugfs);
214 	d->debugfs = NULL;
215 }
216 
217 static int
aoeblk_open(struct block_device * bdev,fmode_t mode)218 aoeblk_open(struct block_device *bdev, fmode_t mode)
219 {
220 	struct aoedev *d = bdev->bd_disk->private_data;
221 	ulong flags;
222 
223 	if (!virt_addr_valid(d)) {
224 		pr_crit("aoe: invalid device pointer in %s\n",
225 			__func__);
226 		WARN_ON(1);
227 		return -ENODEV;
228 	}
229 	if (!(d->flags & DEVFL_UP) || d->flags & DEVFL_TKILL)
230 		return -ENODEV;
231 
232 	mutex_lock(&aoeblk_mutex);
233 	spin_lock_irqsave(&d->lock, flags);
234 	if (d->flags & DEVFL_UP && !(d->flags & DEVFL_TKILL)) {
235 		d->nopen++;
236 		spin_unlock_irqrestore(&d->lock, flags);
237 		mutex_unlock(&aoeblk_mutex);
238 		return 0;
239 	}
240 	spin_unlock_irqrestore(&d->lock, flags);
241 	mutex_unlock(&aoeblk_mutex);
242 	return -ENODEV;
243 }
244 
245 static void
aoeblk_release(struct gendisk * disk,fmode_t mode)246 aoeblk_release(struct gendisk *disk, fmode_t mode)
247 {
248 	struct aoedev *d = disk->private_data;
249 	ulong flags;
250 
251 	spin_lock_irqsave(&d->lock, flags);
252 
253 	if (--d->nopen == 0) {
254 		spin_unlock_irqrestore(&d->lock, flags);
255 		aoecmd_cfg(d->aoemajor, d->aoeminor);
256 		return;
257 	}
258 	spin_unlock_irqrestore(&d->lock, flags);
259 }
260 
aoeblk_queue_rq(struct blk_mq_hw_ctx * hctx,const struct blk_mq_queue_data * bd)261 static blk_status_t aoeblk_queue_rq(struct blk_mq_hw_ctx *hctx,
262 				    const struct blk_mq_queue_data *bd)
263 {
264 	struct aoedev *d = hctx->queue->queuedata;
265 
266 	spin_lock_irq(&d->lock);
267 
268 	if ((d->flags & DEVFL_UP) == 0) {
269 		pr_info_ratelimited("aoe: device %ld.%d is not up\n",
270 			d->aoemajor, d->aoeminor);
271 		spin_unlock_irq(&d->lock);
272 		blk_mq_start_request(bd->rq);
273 		return BLK_STS_IOERR;
274 	}
275 
276 	list_add_tail(&bd->rq->queuelist, &d->rq_list);
277 	aoecmd_work(d);
278 	spin_unlock_irq(&d->lock);
279 	return BLK_STS_OK;
280 }
281 
282 static int
aoeblk_getgeo(struct block_device * bdev,struct hd_geometry * geo)283 aoeblk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
284 {
285 	struct aoedev *d = bdev->bd_disk->private_data;
286 
287 	if ((d->flags & DEVFL_UP) == 0) {
288 		printk(KERN_ERR "aoe: disk not up\n");
289 		return -ENODEV;
290 	}
291 
292 	geo->cylinders = d->geo.cylinders;
293 	geo->heads = d->geo.heads;
294 	geo->sectors = d->geo.sectors;
295 	return 0;
296 }
297 
298 static int
aoeblk_ioctl(struct block_device * bdev,fmode_t mode,uint cmd,ulong arg)299 aoeblk_ioctl(struct block_device *bdev, fmode_t mode, uint cmd, ulong arg)
300 {
301 	struct aoedev *d;
302 
303 	if (!arg)
304 		return -EINVAL;
305 
306 	d = bdev->bd_disk->private_data;
307 	if ((d->flags & DEVFL_UP) == 0) {
308 		pr_err("aoe: disk not up\n");
309 		return -ENODEV;
310 	}
311 
312 	if (cmd == HDIO_GET_IDENTITY) {
313 		if (!copy_to_user((void __user *) arg, &d->ident,
314 			sizeof(d->ident)))
315 			return 0;
316 		return -EFAULT;
317 	}
318 
319 	/* udev calls scsi_id, which uses SG_IO, resulting in noise */
320 	if (cmd != SG_IO)
321 		pr_info("aoe: unknown ioctl 0x%x\n", cmd);
322 
323 	return -ENOTTY;
324 }
325 
326 static const struct block_device_operations aoe_bdops = {
327 	.open = aoeblk_open,
328 	.release = aoeblk_release,
329 	.ioctl = aoeblk_ioctl,
330 	.compat_ioctl = blkdev_compat_ptr_ioctl,
331 	.getgeo = aoeblk_getgeo,
332 	.owner = THIS_MODULE,
333 };
334 
335 static const struct blk_mq_ops aoeblk_mq_ops = {
336 	.queue_rq	= aoeblk_queue_rq,
337 };
338 
339 /* blk_mq_alloc_disk and add_disk can sleep */
340 void
aoeblk_gdalloc(void * vp)341 aoeblk_gdalloc(void *vp)
342 {
343 	struct aoedev *d = vp;
344 	struct gendisk *gd;
345 	mempool_t *mp;
346 	struct blk_mq_tag_set *set;
347 	ulong flags;
348 	int late = 0;
349 	int err;
350 
351 	spin_lock_irqsave(&d->lock, flags);
352 	if (d->flags & DEVFL_GDALLOC
353 	&& !(d->flags & DEVFL_TKILL)
354 	&& !(d->flags & DEVFL_GD_NOW))
355 		d->flags |= DEVFL_GD_NOW;
356 	else
357 		late = 1;
358 	spin_unlock_irqrestore(&d->lock, flags);
359 	if (late)
360 		return;
361 
362 	mp = mempool_create(MIN_BUFS, mempool_alloc_slab, mempool_free_slab,
363 		buf_pool_cache);
364 	if (mp == NULL) {
365 		printk(KERN_ERR "aoe: cannot allocate bufpool for %ld.%d\n",
366 			d->aoemajor, d->aoeminor);
367 		goto err;
368 	}
369 
370 	set = &d->tag_set;
371 	set->ops = &aoeblk_mq_ops;
372 	set->cmd_size = sizeof(struct aoe_req);
373 	set->nr_hw_queues = 1;
374 	set->queue_depth = 128;
375 	set->numa_node = NUMA_NO_NODE;
376 	set->flags = BLK_MQ_F_SHOULD_MERGE;
377 	err = blk_mq_alloc_tag_set(set);
378 	if (err) {
379 		pr_err("aoe: cannot allocate tag set for %ld.%d\n",
380 			d->aoemajor, d->aoeminor);
381 		goto err_mempool;
382 	}
383 
384 	gd = blk_mq_alloc_disk(set, d);
385 	if (IS_ERR(gd)) {
386 		pr_err("aoe: cannot allocate block queue for %ld.%d\n",
387 			d->aoemajor, d->aoeminor);
388 		goto err_tagset;
389 	}
390 
391 	spin_lock_irqsave(&d->lock, flags);
392 	WARN_ON(!(d->flags & DEVFL_GD_NOW));
393 	WARN_ON(!(d->flags & DEVFL_GDALLOC));
394 	WARN_ON(d->flags & DEVFL_TKILL);
395 	WARN_ON(d->gd);
396 	WARN_ON(d->flags & DEVFL_UP);
397 	blk_queue_max_hw_sectors(gd->queue, BLK_DEF_MAX_SECTORS);
398 	blk_queue_io_opt(gd->queue, SZ_2M);
399 	d->bufpool = mp;
400 	d->blkq = gd->queue;
401 	d->gd = gd;
402 	if (aoe_maxsectors)
403 		blk_queue_max_hw_sectors(gd->queue, aoe_maxsectors);
404 	gd->major = AOE_MAJOR;
405 	gd->first_minor = d->sysminor;
406 	gd->minors = AOE_PARTITIONS;
407 	gd->fops = &aoe_bdops;
408 	gd->private_data = d;
409 	set_capacity(gd, d->ssize);
410 	snprintf(gd->disk_name, sizeof gd->disk_name, "etherd/e%ld.%d",
411 		d->aoemajor, d->aoeminor);
412 
413 	d->flags &= ~DEVFL_GDALLOC;
414 	d->flags |= DEVFL_UP;
415 
416 	spin_unlock_irqrestore(&d->lock, flags);
417 
418 	err = device_add_disk(NULL, gd, aoe_attr_groups);
419 	if (err)
420 		goto out_disk_cleanup;
421 	aoedisk_add_debugfs(d);
422 
423 	spin_lock_irqsave(&d->lock, flags);
424 	WARN_ON(!(d->flags & DEVFL_GD_NOW));
425 	d->flags &= ~DEVFL_GD_NOW;
426 	spin_unlock_irqrestore(&d->lock, flags);
427 	return;
428 
429 out_disk_cleanup:
430 	blk_cleanup_disk(gd);
431 err_tagset:
432 	blk_mq_free_tag_set(set);
433 err_mempool:
434 	mempool_destroy(mp);
435 err:
436 	spin_lock_irqsave(&d->lock, flags);
437 	d->flags &= ~DEVFL_GD_NOW;
438 	queue_work(aoe_wq, &d->work);
439 	spin_unlock_irqrestore(&d->lock, flags);
440 }
441 
442 void
aoeblk_exit(void)443 aoeblk_exit(void)
444 {
445 	debugfs_remove_recursive(aoe_debugfs_dir);
446 	aoe_debugfs_dir = NULL;
447 	kmem_cache_destroy(buf_pool_cache);
448 }
449 
450 int __init
aoeblk_init(void)451 aoeblk_init(void)
452 {
453 	buf_pool_cache = kmem_cache_create("aoe_bufs",
454 					   sizeof(struct buf),
455 					   0, 0, NULL);
456 	if (buf_pool_cache == NULL)
457 		return -ENOMEM;
458 	aoe_debugfs_dir = debugfs_create_dir("aoe", NULL);
459 	return 0;
460 }
461 
462