xref: /openbmc/linux/drivers/s390/block/dcssblk.c (revision 45cc35e8)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * dcssblk.c -- the S/390 block driver for dcss memory
4  *
5  * Authors: Carsten Otte, Stefan Weinhuber, Gerald Schaefer
6  */
7 
8 #define KMSG_COMPONENT "dcssblk"
9 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
10 
11 #include <linux/module.h>
12 #include <linux/moduleparam.h>
13 #include <linux/ctype.h>
14 #include <linux/errno.h>
15 #include <linux/init.h>
16 #include <linux/slab.h>
17 #include <linux/blkdev.h>
18 #include <linux/completion.h>
19 #include <linux/interrupt.h>
20 #include <linux/pfn_t.h>
21 #include <linux/uio.h>
22 #include <linux/dax.h>
23 #include <asm/extmem.h>
24 #include <asm/io.h>
25 
26 #define DCSSBLK_NAME "dcssblk"
27 #define DCSSBLK_MINORS_PER_DISK 1
28 #define DCSSBLK_PARM_LEN 400
29 #define DCSS_BUS_ID_SIZE 20
30 
31 static int dcssblk_open(struct block_device *bdev, fmode_t mode);
32 static void dcssblk_release(struct gendisk *disk, fmode_t mode);
33 static void dcssblk_submit_bio(struct bio *bio);
34 static long dcssblk_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
35 		long nr_pages, enum dax_access_mode mode, void **kaddr,
36 		pfn_t *pfn);
37 
38 static char dcssblk_segments[DCSSBLK_PARM_LEN] = "\0";
39 
40 static int dcssblk_major;
41 static const struct block_device_operations dcssblk_devops = {
42 	.owner   	= THIS_MODULE,
43 	.submit_bio	= dcssblk_submit_bio,
44 	.open    	= dcssblk_open,
45 	.release 	= dcssblk_release,
46 };
47 
48 static int dcssblk_dax_zero_page_range(struct dax_device *dax_dev,
49 				       pgoff_t pgoff, size_t nr_pages)
50 {
51 	long rc;
52 	void *kaddr;
53 
54 	rc = dax_direct_access(dax_dev, pgoff, nr_pages, DAX_ACCESS,
55 			&kaddr, NULL);
56 	if (rc < 0)
57 		return rc;
58 	memset(kaddr, 0, nr_pages << PAGE_SHIFT);
59 	dax_flush(dax_dev, kaddr, nr_pages << PAGE_SHIFT);
60 	return 0;
61 }
62 
63 static const struct dax_operations dcssblk_dax_ops = {
64 	.direct_access = dcssblk_dax_direct_access,
65 	.zero_page_range = dcssblk_dax_zero_page_range,
66 };
67 
68 struct dcssblk_dev_info {
69 	struct list_head lh;
70 	struct device dev;
71 	char segment_name[DCSS_BUS_ID_SIZE];
72 	atomic_t use_count;
73 	struct gendisk *gd;
74 	unsigned long start;
75 	unsigned long end;
76 	int segment_type;
77 	unsigned char save_pending;
78 	unsigned char is_shared;
79 	int num_of_segments;
80 	struct list_head seg_list;
81 	struct dax_device *dax_dev;
82 };
83 
84 struct segment_info {
85 	struct list_head lh;
86 	char segment_name[DCSS_BUS_ID_SIZE];
87 	unsigned long start;
88 	unsigned long end;
89 	int segment_type;
90 };
91 
92 static ssize_t dcssblk_add_store(struct device * dev, struct device_attribute *attr, const char * buf,
93 				  size_t count);
94 static ssize_t dcssblk_remove_store(struct device * dev, struct device_attribute *attr, const char * buf,
95 				  size_t count);
96 
97 static DEVICE_ATTR(add, S_IWUSR, NULL, dcssblk_add_store);
98 static DEVICE_ATTR(remove, S_IWUSR, NULL, dcssblk_remove_store);
99 
100 static struct device *dcssblk_root_dev;
101 
102 static LIST_HEAD(dcssblk_devices);
103 static struct rw_semaphore dcssblk_devices_sem;
104 
105 /*
106  * release function for segment device.
107  */
108 static void
109 dcssblk_release_segment(struct device *dev)
110 {
111 	struct dcssblk_dev_info *dev_info;
112 	struct segment_info *entry, *temp;
113 
114 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
115 	list_for_each_entry_safe(entry, temp, &dev_info->seg_list, lh) {
116 		list_del(&entry->lh);
117 		kfree(entry);
118 	}
119 	kfree(dev_info);
120 	module_put(THIS_MODULE);
121 }
122 
123 /*
124  * get a minor number. needs to be called with
125  * down_write(&dcssblk_devices_sem) and the
126  * device needs to be enqueued before the semaphore is
127  * freed.
128  */
129 static int
130 dcssblk_assign_free_minor(struct dcssblk_dev_info *dev_info)
131 {
132 	int minor, found;
133 	struct dcssblk_dev_info *entry;
134 
135 	if (dev_info == NULL)
136 		return -EINVAL;
137 	for (minor = 0; minor < (1<<MINORBITS); minor++) {
138 		found = 0;
139 		// test if minor available
140 		list_for_each_entry(entry, &dcssblk_devices, lh)
141 			if (minor == entry->gd->first_minor)
142 				found++;
143 		if (!found) break; // got unused minor
144 	}
145 	if (found)
146 		return -EBUSY;
147 	dev_info->gd->first_minor = minor;
148 	return 0;
149 }
150 
151 /*
152  * get the struct dcssblk_dev_info from dcssblk_devices
153  * for the given name.
154  * down_read(&dcssblk_devices_sem) must be held.
155  */
156 static struct dcssblk_dev_info *
157 dcssblk_get_device_by_name(char *name)
158 {
159 	struct dcssblk_dev_info *entry;
160 
161 	list_for_each_entry(entry, &dcssblk_devices, lh) {
162 		if (!strcmp(name, entry->segment_name)) {
163 			return entry;
164 		}
165 	}
166 	return NULL;
167 }
168 
169 /*
170  * get the struct segment_info from seg_list
171  * for the given name.
172  * down_read(&dcssblk_devices_sem) must be held.
173  */
174 static struct segment_info *
175 dcssblk_get_segment_by_name(char *name)
176 {
177 	struct dcssblk_dev_info *dev_info;
178 	struct segment_info *entry;
179 
180 	list_for_each_entry(dev_info, &dcssblk_devices, lh) {
181 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
182 			if (!strcmp(name, entry->segment_name))
183 				return entry;
184 		}
185 	}
186 	return NULL;
187 }
188 
189 /*
190  * get the highest address of the multi-segment block.
191  */
192 static unsigned long
193 dcssblk_find_highest_addr(struct dcssblk_dev_info *dev_info)
194 {
195 	unsigned long highest_addr;
196 	struct segment_info *entry;
197 
198 	highest_addr = 0;
199 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
200 		if (highest_addr < entry->end)
201 			highest_addr = entry->end;
202 	}
203 	return highest_addr;
204 }
205 
206 /*
207  * get the lowest address of the multi-segment block.
208  */
209 static unsigned long
210 dcssblk_find_lowest_addr(struct dcssblk_dev_info *dev_info)
211 {
212 	int set_first;
213 	unsigned long lowest_addr;
214 	struct segment_info *entry;
215 
216 	set_first = 0;
217 	lowest_addr = 0;
218 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
219 		if (set_first == 0) {
220 			lowest_addr = entry->start;
221 			set_first = 1;
222 		} else {
223 			if (lowest_addr > entry->start)
224 				lowest_addr = entry->start;
225 		}
226 	}
227 	return lowest_addr;
228 }
229 
230 /*
231  * Check continuity of segments.
232  */
233 static int
234 dcssblk_is_continuous(struct dcssblk_dev_info *dev_info)
235 {
236 	int i, j, rc;
237 	struct segment_info *sort_list, *entry, temp;
238 
239 	if (dev_info->num_of_segments <= 1)
240 		return 0;
241 
242 	sort_list = kcalloc(dev_info->num_of_segments,
243 			    sizeof(struct segment_info),
244 			    GFP_KERNEL);
245 	if (sort_list == NULL)
246 		return -ENOMEM;
247 	i = 0;
248 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
249 		memcpy(&sort_list[i], entry, sizeof(struct segment_info));
250 		i++;
251 	}
252 
253 	/* sort segments */
254 	for (i = 0; i < dev_info->num_of_segments; i++)
255 		for (j = 0; j < dev_info->num_of_segments; j++)
256 			if (sort_list[j].start > sort_list[i].start) {
257 				memcpy(&temp, &sort_list[i],
258 					sizeof(struct segment_info));
259 				memcpy(&sort_list[i], &sort_list[j],
260 					sizeof(struct segment_info));
261 				memcpy(&sort_list[j], &temp,
262 					sizeof(struct segment_info));
263 			}
264 
265 	/* check continuity */
266 	for (i = 0; i < dev_info->num_of_segments - 1; i++) {
267 		if ((sort_list[i].end + 1) != sort_list[i+1].start) {
268 			pr_err("Adjacent DCSSs %s and %s are not "
269 			       "contiguous\n", sort_list[i].segment_name,
270 			       sort_list[i+1].segment_name);
271 			rc = -EINVAL;
272 			goto out;
273 		}
274 		/* EN and EW are allowed in a block device */
275 		if (sort_list[i].segment_type != sort_list[i+1].segment_type) {
276 			if (!(sort_list[i].segment_type & SEGMENT_EXCLUSIVE) ||
277 				(sort_list[i].segment_type == SEG_TYPE_ER) ||
278 				!(sort_list[i+1].segment_type &
279 				SEGMENT_EXCLUSIVE) ||
280 				(sort_list[i+1].segment_type == SEG_TYPE_ER)) {
281 				pr_err("DCSS %s and DCSS %s have "
282 				       "incompatible types\n",
283 				       sort_list[i].segment_name,
284 				       sort_list[i+1].segment_name);
285 				rc = -EINVAL;
286 				goto out;
287 			}
288 		}
289 	}
290 	rc = 0;
291 out:
292 	kfree(sort_list);
293 	return rc;
294 }
295 
296 /*
297  * Load a segment
298  */
299 static int
300 dcssblk_load_segment(char *name, struct segment_info **seg_info)
301 {
302 	int rc;
303 
304 	/* already loaded? */
305 	down_read(&dcssblk_devices_sem);
306 	*seg_info = dcssblk_get_segment_by_name(name);
307 	up_read(&dcssblk_devices_sem);
308 	if (*seg_info != NULL)
309 		return -EEXIST;
310 
311 	/* get a struct segment_info */
312 	*seg_info = kzalloc(sizeof(struct segment_info), GFP_KERNEL);
313 	if (*seg_info == NULL)
314 		return -ENOMEM;
315 
316 	strcpy((*seg_info)->segment_name, name);
317 
318 	/* load the segment */
319 	rc = segment_load(name, SEGMENT_SHARED,
320 			&(*seg_info)->start, &(*seg_info)->end);
321 	if (rc < 0) {
322 		segment_warning(rc, (*seg_info)->segment_name);
323 		kfree(*seg_info);
324 	} else {
325 		INIT_LIST_HEAD(&(*seg_info)->lh);
326 		(*seg_info)->segment_type = rc;
327 	}
328 	return rc;
329 }
330 
331 /*
332  * device attribute for switching shared/nonshared (exclusive)
333  * operation (show + store)
334  */
335 static ssize_t
336 dcssblk_shared_show(struct device *dev, struct device_attribute *attr, char *buf)
337 {
338 	struct dcssblk_dev_info *dev_info;
339 
340 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
341 	return sprintf(buf, dev_info->is_shared ? "1\n" : "0\n");
342 }
343 
344 static ssize_t
345 dcssblk_shared_store(struct device *dev, struct device_attribute *attr, const char *inbuf, size_t count)
346 {
347 	struct dcssblk_dev_info *dev_info;
348 	struct segment_info *entry, *temp;
349 	int rc;
350 
351 	if ((count > 1) && (inbuf[1] != '\n') && (inbuf[1] != '\0'))
352 		return -EINVAL;
353 	down_write(&dcssblk_devices_sem);
354 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
355 	if (atomic_read(&dev_info->use_count)) {
356 		rc = -EBUSY;
357 		goto out;
358 	}
359 	if (inbuf[0] == '1') {
360 		/* reload segments in shared mode */
361 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
362 			rc = segment_modify_shared(entry->segment_name,
363 						SEGMENT_SHARED);
364 			if (rc < 0) {
365 				BUG_ON(rc == -EINVAL);
366 				if (rc != -EAGAIN)
367 					goto removeseg;
368 			}
369 		}
370 		dev_info->is_shared = 1;
371 		switch (dev_info->segment_type) {
372 		case SEG_TYPE_SR:
373 		case SEG_TYPE_ER:
374 		case SEG_TYPE_SC:
375 			set_disk_ro(dev_info->gd, 1);
376 		}
377 	} else if (inbuf[0] == '0') {
378 		/* reload segments in exclusive mode */
379 		if (dev_info->segment_type == SEG_TYPE_SC) {
380 			pr_err("DCSS %s is of type SC and cannot be "
381 			       "loaded as exclusive-writable\n",
382 			       dev_info->segment_name);
383 			rc = -EINVAL;
384 			goto out;
385 		}
386 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
387 			rc = segment_modify_shared(entry->segment_name,
388 						   SEGMENT_EXCLUSIVE);
389 			if (rc < 0) {
390 				BUG_ON(rc == -EINVAL);
391 				if (rc != -EAGAIN)
392 					goto removeseg;
393 			}
394 		}
395 		dev_info->is_shared = 0;
396 		set_disk_ro(dev_info->gd, 0);
397 	} else {
398 		rc = -EINVAL;
399 		goto out;
400 	}
401 	rc = count;
402 	goto out;
403 
404 removeseg:
405 	pr_err("DCSS device %s is removed after a failed access mode "
406 	       "change\n", dev_info->segment_name);
407 	temp = entry;
408 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
409 		if (entry != temp)
410 			segment_unload(entry->segment_name);
411 	}
412 	list_del(&dev_info->lh);
413 
414 	kill_dax(dev_info->dax_dev);
415 	put_dax(dev_info->dax_dev);
416 	del_gendisk(dev_info->gd);
417 	put_disk(dev_info->gd);
418 	up_write(&dcssblk_devices_sem);
419 
420 	if (device_remove_file_self(dev, attr)) {
421 		device_unregister(dev);
422 		put_device(dev);
423 	}
424 	return rc;
425 out:
426 	up_write(&dcssblk_devices_sem);
427 	return rc;
428 }
429 static DEVICE_ATTR(shared, S_IWUSR | S_IRUSR, dcssblk_shared_show,
430 		   dcssblk_shared_store);
431 
432 /*
433  * device attribute for save operation on current copy
434  * of the segment. If the segment is busy, saving will
435  * become pending until it gets released, which can be
436  * undone by storing a non-true value to this entry.
437  * (show + store)
438  */
439 static ssize_t
440 dcssblk_save_show(struct device *dev, struct device_attribute *attr, char *buf)
441 {
442 	struct dcssblk_dev_info *dev_info;
443 
444 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
445 	return sprintf(buf, dev_info->save_pending ? "1\n" : "0\n");
446 }
447 
448 static ssize_t
449 dcssblk_save_store(struct device *dev, struct device_attribute *attr, const char *inbuf, size_t count)
450 {
451 	struct dcssblk_dev_info *dev_info;
452 	struct segment_info *entry;
453 
454 	if ((count > 1) && (inbuf[1] != '\n') && (inbuf[1] != '\0'))
455 		return -EINVAL;
456 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
457 
458 	down_write(&dcssblk_devices_sem);
459 	if (inbuf[0] == '1') {
460 		if (atomic_read(&dev_info->use_count) == 0) {
461 			// device is idle => we save immediately
462 			pr_info("All DCSSs that map to device %s are "
463 				"saved\n", dev_info->segment_name);
464 			list_for_each_entry(entry, &dev_info->seg_list, lh) {
465 				if (entry->segment_type == SEG_TYPE_EN ||
466 				    entry->segment_type == SEG_TYPE_SN)
467 					pr_warn("DCSS %s is of type SN or EN"
468 						" and cannot be saved\n",
469 						entry->segment_name);
470 				else
471 					segment_save(entry->segment_name);
472 			}
473 		}  else {
474 			// device is busy => we save it when it becomes
475 			// idle in dcssblk_release
476 			pr_info("Device %s is in use, its DCSSs will be "
477 				"saved when it becomes idle\n",
478 				dev_info->segment_name);
479 			dev_info->save_pending = 1;
480 		}
481 	} else if (inbuf[0] == '0') {
482 		if (dev_info->save_pending) {
483 			// device is busy & the user wants to undo his save
484 			// request
485 			dev_info->save_pending = 0;
486 			pr_info("A pending save request for device %s "
487 				"has been canceled\n",
488 				dev_info->segment_name);
489 		}
490 	} else {
491 		up_write(&dcssblk_devices_sem);
492 		return -EINVAL;
493 	}
494 	up_write(&dcssblk_devices_sem);
495 	return count;
496 }
497 static DEVICE_ATTR(save, S_IWUSR | S_IRUSR, dcssblk_save_show,
498 		   dcssblk_save_store);
499 
500 /*
501  * device attribute for showing all segments in a device
502  */
503 static ssize_t
504 dcssblk_seglist_show(struct device *dev, struct device_attribute *attr,
505 		char *buf)
506 {
507 	int i;
508 
509 	struct dcssblk_dev_info *dev_info;
510 	struct segment_info *entry;
511 
512 	down_read(&dcssblk_devices_sem);
513 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
514 	i = 0;
515 	buf[0] = '\0';
516 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
517 		strcpy(&buf[i], entry->segment_name);
518 		i += strlen(entry->segment_name);
519 		buf[i] = '\n';
520 		i++;
521 	}
522 	up_read(&dcssblk_devices_sem);
523 	return i;
524 }
525 static DEVICE_ATTR(seglist, S_IRUSR, dcssblk_seglist_show, NULL);
526 
527 static struct attribute *dcssblk_dev_attrs[] = {
528 	&dev_attr_shared.attr,
529 	&dev_attr_save.attr,
530 	&dev_attr_seglist.attr,
531 	NULL,
532 };
533 static struct attribute_group dcssblk_dev_attr_group = {
534 	.attrs = dcssblk_dev_attrs,
535 };
536 static const struct attribute_group *dcssblk_dev_attr_groups[] = {
537 	&dcssblk_dev_attr_group,
538 	NULL,
539 };
540 
541 /*
542  * device attribute for adding devices
543  */
544 static ssize_t
545 dcssblk_add_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
546 {
547 	int rc, i, j, num_of_segments;
548 	struct dcssblk_dev_info *dev_info;
549 	struct segment_info *seg_info, *temp;
550 	char *local_buf;
551 	unsigned long seg_byte_size;
552 
553 	dev_info = NULL;
554 	seg_info = NULL;
555 	if (dev != dcssblk_root_dev) {
556 		rc = -EINVAL;
557 		goto out_nobuf;
558 	}
559 	if ((count < 1) || (buf[0] == '\0') || (buf[0] == '\n')) {
560 		rc = -ENAMETOOLONG;
561 		goto out_nobuf;
562 	}
563 
564 	local_buf = kmalloc(count + 1, GFP_KERNEL);
565 	if (local_buf == NULL) {
566 		rc = -ENOMEM;
567 		goto out_nobuf;
568 	}
569 
570 	/*
571 	 * parse input
572 	 */
573 	num_of_segments = 0;
574 	for (i = 0; (i < count && (buf[i] != '\0') && (buf[i] != '\n')); i++) {
575 		for (j = i; j < count &&
576 			(buf[j] != ':') &&
577 			(buf[j] != '\0') &&
578 			(buf[j] != '\n'); j++) {
579 			local_buf[j-i] = toupper(buf[j]);
580 		}
581 		local_buf[j-i] = '\0';
582 		if (((j - i) == 0) || ((j - i) > 8)) {
583 			rc = -ENAMETOOLONG;
584 			goto seg_list_del;
585 		}
586 
587 		rc = dcssblk_load_segment(local_buf, &seg_info);
588 		if (rc < 0)
589 			goto seg_list_del;
590 		/*
591 		 * get a struct dcssblk_dev_info
592 		 */
593 		if (num_of_segments == 0) {
594 			dev_info = kzalloc(sizeof(struct dcssblk_dev_info),
595 					GFP_KERNEL);
596 			if (dev_info == NULL) {
597 				rc = -ENOMEM;
598 				goto out;
599 			}
600 			strcpy(dev_info->segment_name, local_buf);
601 			dev_info->segment_type = seg_info->segment_type;
602 			INIT_LIST_HEAD(&dev_info->seg_list);
603 		}
604 		list_add_tail(&seg_info->lh, &dev_info->seg_list);
605 		num_of_segments++;
606 		i = j;
607 
608 		if ((buf[j] == '\0') || (buf[j] == '\n'))
609 			break;
610 	}
611 
612 	/* no trailing colon at the end of the input */
613 	if ((i > 0) && (buf[i-1] == ':')) {
614 		rc = -ENAMETOOLONG;
615 		goto seg_list_del;
616 	}
617 	strscpy(local_buf, buf, i + 1);
618 	dev_info->num_of_segments = num_of_segments;
619 	rc = dcssblk_is_continuous(dev_info);
620 	if (rc < 0)
621 		goto seg_list_del;
622 
623 	dev_info->start = dcssblk_find_lowest_addr(dev_info);
624 	dev_info->end = dcssblk_find_highest_addr(dev_info);
625 
626 	dev_set_name(&dev_info->dev, "%s", dev_info->segment_name);
627 	dev_info->dev.release = dcssblk_release_segment;
628 	dev_info->dev.groups = dcssblk_dev_attr_groups;
629 	INIT_LIST_HEAD(&dev_info->lh);
630 	dev_info->gd = blk_alloc_disk(NUMA_NO_NODE);
631 	if (dev_info->gd == NULL) {
632 		rc = -ENOMEM;
633 		goto seg_list_del;
634 	}
635 	dev_info->gd->major = dcssblk_major;
636 	dev_info->gd->minors = DCSSBLK_MINORS_PER_DISK;
637 	dev_info->gd->fops = &dcssblk_devops;
638 	dev_info->gd->private_data = dev_info;
639 	dev_info->gd->flags |= GENHD_FL_NO_PART;
640 	blk_queue_logical_block_size(dev_info->gd->queue, 4096);
641 	blk_queue_flag_set(QUEUE_FLAG_DAX, dev_info->gd->queue);
642 
643 	seg_byte_size = (dev_info->end - dev_info->start + 1);
644 	set_capacity(dev_info->gd, seg_byte_size >> 9); // size in sectors
645 	pr_info("Loaded %s with total size %lu bytes and capacity %lu "
646 		"sectors\n", local_buf, seg_byte_size, seg_byte_size >> 9);
647 
648 	dev_info->save_pending = 0;
649 	dev_info->is_shared = 1;
650 	dev_info->dev.parent = dcssblk_root_dev;
651 
652 	/*
653 	 *get minor, add to list
654 	 */
655 	down_write(&dcssblk_devices_sem);
656 	if (dcssblk_get_segment_by_name(local_buf)) {
657 		rc = -EEXIST;
658 		goto release_gd;
659 	}
660 	rc = dcssblk_assign_free_minor(dev_info);
661 	if (rc)
662 		goto release_gd;
663 	sprintf(dev_info->gd->disk_name, "dcssblk%d",
664 		dev_info->gd->first_minor);
665 	list_add_tail(&dev_info->lh, &dcssblk_devices);
666 
667 	if (!try_module_get(THIS_MODULE)) {
668 		rc = -ENODEV;
669 		goto dev_list_del;
670 	}
671 	/*
672 	 * register the device
673 	 */
674 	rc = device_register(&dev_info->dev);
675 	if (rc)
676 		goto put_dev;
677 
678 	dev_info->dax_dev = alloc_dax(dev_info, &dcssblk_dax_ops);
679 	if (IS_ERR(dev_info->dax_dev)) {
680 		rc = PTR_ERR(dev_info->dax_dev);
681 		dev_info->dax_dev = NULL;
682 		goto put_dev;
683 	}
684 	set_dax_synchronous(dev_info->dax_dev);
685 	rc = dax_add_host(dev_info->dax_dev, dev_info->gd);
686 	if (rc)
687 		goto out_dax;
688 
689 	get_device(&dev_info->dev);
690 	rc = device_add_disk(&dev_info->dev, dev_info->gd, NULL);
691 	if (rc)
692 		goto out_dax_host;
693 
694 	switch (dev_info->segment_type) {
695 		case SEG_TYPE_SR:
696 		case SEG_TYPE_ER:
697 		case SEG_TYPE_SC:
698 			set_disk_ro(dev_info->gd,1);
699 			break;
700 		default:
701 			set_disk_ro(dev_info->gd,0);
702 			break;
703 	}
704 	up_write(&dcssblk_devices_sem);
705 	rc = count;
706 	goto out;
707 
708 out_dax_host:
709 	dax_remove_host(dev_info->gd);
710 out_dax:
711 	put_device(&dev_info->dev);
712 	kill_dax(dev_info->dax_dev);
713 	put_dax(dev_info->dax_dev);
714 put_dev:
715 	list_del(&dev_info->lh);
716 	put_disk(dev_info->gd);
717 	list_for_each_entry(seg_info, &dev_info->seg_list, lh) {
718 		segment_unload(seg_info->segment_name);
719 	}
720 	put_device(&dev_info->dev);
721 	up_write(&dcssblk_devices_sem);
722 	goto out;
723 dev_list_del:
724 	list_del(&dev_info->lh);
725 release_gd:
726 	put_disk(dev_info->gd);
727 	up_write(&dcssblk_devices_sem);
728 seg_list_del:
729 	if (dev_info == NULL)
730 		goto out;
731 	list_for_each_entry_safe(seg_info, temp, &dev_info->seg_list, lh) {
732 		list_del(&seg_info->lh);
733 		segment_unload(seg_info->segment_name);
734 		kfree(seg_info);
735 	}
736 	kfree(dev_info);
737 out:
738 	kfree(local_buf);
739 out_nobuf:
740 	return rc;
741 }
742 
743 /*
744  * device attribute for removing devices
745  */
746 static ssize_t
747 dcssblk_remove_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
748 {
749 	struct dcssblk_dev_info *dev_info;
750 	struct segment_info *entry;
751 	int rc, i;
752 	char *local_buf;
753 
754 	if (dev != dcssblk_root_dev) {
755 		return -EINVAL;
756 	}
757 	local_buf = kmalloc(count + 1, GFP_KERNEL);
758 	if (local_buf == NULL) {
759 		return -ENOMEM;
760 	}
761 	/*
762 	 * parse input
763 	 */
764 	for (i = 0; (i < count && (*(buf+i)!='\0') && (*(buf+i)!='\n')); i++) {
765 		local_buf[i] = toupper(buf[i]);
766 	}
767 	local_buf[i] = '\0';
768 	if ((i == 0) || (i > 8)) {
769 		rc = -ENAMETOOLONG;
770 		goto out_buf;
771 	}
772 
773 	down_write(&dcssblk_devices_sem);
774 	dev_info = dcssblk_get_device_by_name(local_buf);
775 	if (dev_info == NULL) {
776 		up_write(&dcssblk_devices_sem);
777 		pr_warn("Device %s cannot be removed because it is not a known device\n",
778 			local_buf);
779 		rc = -ENODEV;
780 		goto out_buf;
781 	}
782 	if (atomic_read(&dev_info->use_count) != 0) {
783 		up_write(&dcssblk_devices_sem);
784 		pr_warn("Device %s cannot be removed while it is in use\n",
785 			local_buf);
786 		rc = -EBUSY;
787 		goto out_buf;
788 	}
789 
790 	list_del(&dev_info->lh);
791 	kill_dax(dev_info->dax_dev);
792 	put_dax(dev_info->dax_dev);
793 	del_gendisk(dev_info->gd);
794 	put_disk(dev_info->gd);
795 
796 	/* unload all related segments */
797 	list_for_each_entry(entry, &dev_info->seg_list, lh)
798 		segment_unload(entry->segment_name);
799 
800 	up_write(&dcssblk_devices_sem);
801 
802 	device_unregister(&dev_info->dev);
803 	put_device(&dev_info->dev);
804 
805 	rc = count;
806 out_buf:
807 	kfree(local_buf);
808 	return rc;
809 }
810 
811 static int
812 dcssblk_open(struct block_device *bdev, fmode_t mode)
813 {
814 	struct dcssblk_dev_info *dev_info;
815 	int rc;
816 
817 	dev_info = bdev->bd_disk->private_data;
818 	if (NULL == dev_info) {
819 		rc = -ENODEV;
820 		goto out;
821 	}
822 	atomic_inc(&dev_info->use_count);
823 	rc = 0;
824 out:
825 	return rc;
826 }
827 
828 static void
829 dcssblk_release(struct gendisk *disk, fmode_t mode)
830 {
831 	struct dcssblk_dev_info *dev_info = disk->private_data;
832 	struct segment_info *entry;
833 
834 	if (!dev_info) {
835 		WARN_ON(1);
836 		return;
837 	}
838 	down_write(&dcssblk_devices_sem);
839 	if (atomic_dec_and_test(&dev_info->use_count)
840 	    && (dev_info->save_pending)) {
841 		pr_info("Device %s has become idle and is being saved "
842 			"now\n", dev_info->segment_name);
843 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
844 			if (entry->segment_type == SEG_TYPE_EN ||
845 			    entry->segment_type == SEG_TYPE_SN)
846 				pr_warn("DCSS %s is of type SN or EN and cannot"
847 					" be saved\n", entry->segment_name);
848 			else
849 				segment_save(entry->segment_name);
850 		}
851 		dev_info->save_pending = 0;
852 	}
853 	up_write(&dcssblk_devices_sem);
854 }
855 
856 static void
857 dcssblk_submit_bio(struct bio *bio)
858 {
859 	struct dcssblk_dev_info *dev_info;
860 	struct bio_vec bvec;
861 	struct bvec_iter iter;
862 	unsigned long index;
863 	unsigned long page_addr;
864 	unsigned long source_addr;
865 	unsigned long bytes_done;
866 
867 	bytes_done = 0;
868 	dev_info = bio->bi_bdev->bd_disk->private_data;
869 	if (dev_info == NULL)
870 		goto fail;
871 	if ((bio->bi_iter.bi_sector & 7) != 0 ||
872 	    (bio->bi_iter.bi_size & 4095) != 0)
873 		/* Request is not page-aligned. */
874 		goto fail;
875 	/* verify data transfer direction */
876 	if (dev_info->is_shared) {
877 		switch (dev_info->segment_type) {
878 		case SEG_TYPE_SR:
879 		case SEG_TYPE_ER:
880 		case SEG_TYPE_SC:
881 			/* cannot write to these segments */
882 			if (bio_data_dir(bio) == WRITE) {
883 				pr_warn("Writing to %s failed because it is a read-only device\n",
884 					dev_name(&dev_info->dev));
885 				goto fail;
886 			}
887 		}
888 	}
889 
890 	index = (bio->bi_iter.bi_sector >> 3);
891 	bio_for_each_segment(bvec, bio, iter) {
892 		page_addr = (unsigned long)bvec_virt(&bvec);
893 		source_addr = dev_info->start + (index<<12) + bytes_done;
894 		if (unlikely((page_addr & 4095) != 0) || (bvec.bv_len & 4095) != 0)
895 			// More paranoia.
896 			goto fail;
897 		if (bio_data_dir(bio) == READ) {
898 			memcpy((void*)page_addr, (void*)source_addr,
899 				bvec.bv_len);
900 		} else {
901 			memcpy((void*)source_addr, (void*)page_addr,
902 				bvec.bv_len);
903 		}
904 		bytes_done += bvec.bv_len;
905 	}
906 	bio_endio(bio);
907 	return;
908 fail:
909 	bio_io_error(bio);
910 }
911 
912 static long
913 __dcssblk_direct_access(struct dcssblk_dev_info *dev_info, pgoff_t pgoff,
914 		long nr_pages, void **kaddr, pfn_t *pfn)
915 {
916 	resource_size_t offset = pgoff * PAGE_SIZE;
917 	unsigned long dev_sz;
918 
919 	dev_sz = dev_info->end - dev_info->start + 1;
920 	if (kaddr)
921 		*kaddr = (void *) dev_info->start + offset;
922 	if (pfn)
923 		*pfn = __pfn_to_pfn_t(PFN_DOWN(dev_info->start + offset),
924 				PFN_DEV|PFN_SPECIAL);
925 
926 	return (dev_sz - offset) / PAGE_SIZE;
927 }
928 
929 static long
930 dcssblk_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
931 		long nr_pages, enum dax_access_mode mode, void **kaddr,
932 		pfn_t *pfn)
933 {
934 	struct dcssblk_dev_info *dev_info = dax_get_private(dax_dev);
935 
936 	return __dcssblk_direct_access(dev_info, pgoff, nr_pages, kaddr, pfn);
937 }
938 
939 static void
940 dcssblk_check_params(void)
941 {
942 	int rc, i, j, k;
943 	char buf[DCSSBLK_PARM_LEN + 1];
944 	struct dcssblk_dev_info *dev_info;
945 
946 	for (i = 0; (i < DCSSBLK_PARM_LEN) && (dcssblk_segments[i] != '\0');
947 	     i++) {
948 		for (j = i; (j < DCSSBLK_PARM_LEN) &&
949 			    (dcssblk_segments[j] != ',')  &&
950 			    (dcssblk_segments[j] != '\0') &&
951 			    (dcssblk_segments[j] != '('); j++)
952 		{
953 			buf[j-i] = dcssblk_segments[j];
954 		}
955 		buf[j-i] = '\0';
956 		rc = dcssblk_add_store(dcssblk_root_dev, NULL, buf, j-i);
957 		if ((rc >= 0) && (dcssblk_segments[j] == '(')) {
958 			for (k = 0; (buf[k] != ':') && (buf[k] != '\0'); k++)
959 				buf[k] = toupper(buf[k]);
960 			buf[k] = '\0';
961 			if (!strncmp(&dcssblk_segments[j], "(local)", 7)) {
962 				down_read(&dcssblk_devices_sem);
963 				dev_info = dcssblk_get_device_by_name(buf);
964 				up_read(&dcssblk_devices_sem);
965 				if (dev_info)
966 					dcssblk_shared_store(&dev_info->dev,
967 							     NULL, "0\n", 2);
968 			}
969 		}
970 		while ((dcssblk_segments[j] != ',') &&
971 		       (dcssblk_segments[j] != '\0'))
972 		{
973 			j++;
974 		}
975 		if (dcssblk_segments[j] == '\0')
976 			break;
977 		i = j;
978 	}
979 }
980 
981 /*
982  * The init/exit functions.
983  */
984 static void __exit
985 dcssblk_exit(void)
986 {
987 	root_device_unregister(dcssblk_root_dev);
988 	unregister_blkdev(dcssblk_major, DCSSBLK_NAME);
989 }
990 
991 static int __init
992 dcssblk_init(void)
993 {
994 	int rc;
995 
996 	dcssblk_root_dev = root_device_register("dcssblk");
997 	if (IS_ERR(dcssblk_root_dev))
998 		return PTR_ERR(dcssblk_root_dev);
999 	rc = device_create_file(dcssblk_root_dev, &dev_attr_add);
1000 	if (rc)
1001 		goto out_root;
1002 	rc = device_create_file(dcssblk_root_dev, &dev_attr_remove);
1003 	if (rc)
1004 		goto out_root;
1005 	rc = register_blkdev(0, DCSSBLK_NAME);
1006 	if (rc < 0)
1007 		goto out_root;
1008 	dcssblk_major = rc;
1009 	init_rwsem(&dcssblk_devices_sem);
1010 
1011 	dcssblk_check_params();
1012 	return 0;
1013 
1014 out_root:
1015 	root_device_unregister(dcssblk_root_dev);
1016 
1017 	return rc;
1018 }
1019 
1020 module_init(dcssblk_init);
1021 module_exit(dcssblk_exit);
1022 
1023 module_param_string(segments, dcssblk_segments, DCSSBLK_PARM_LEN, 0444);
1024 MODULE_PARM_DESC(segments, "Name of DCSS segment(s) to be loaded, "
1025 		 "comma-separated list, names in each set separated "
1026 		 "by commas are separated by colons, each set contains "
1027 		 "names of contiguous segments and each name max. 8 chars.\n"
1028 		 "Adding \"(local)\" to the end of each set equals echoing 0 "
1029 		 "to /sys/devices/dcssblk/<device name>/shared after loading "
1030 		 "the contiguous segments - \n"
1031 		 "e.g. segments=\"mydcss1,mydcss2:mydcss3,mydcss4(local)\"");
1032 
1033 MODULE_LICENSE("GPL");
1034