xref: /openbmc/linux/drivers/s390/block/dcssblk.c (revision 2326f3cd)
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 blk_qc_t 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, void **kaddr, pfn_t *pfn);
36 
37 static char dcssblk_segments[DCSSBLK_PARM_LEN] = "\0";
38 
39 static int dcssblk_major;
40 static const struct block_device_operations dcssblk_devops = {
41 	.owner   	= THIS_MODULE,
42 	.submit_bio	= dcssblk_submit_bio,
43 	.open    	= dcssblk_open,
44 	.release 	= dcssblk_release,
45 };
46 
47 static size_t dcssblk_dax_copy_from_iter(struct dax_device *dax_dev,
48 		pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i)
49 {
50 	return copy_from_iter(addr, bytes, i);
51 }
52 
53 static size_t dcssblk_dax_copy_to_iter(struct dax_device *dax_dev,
54 		pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i)
55 {
56 	return copy_to_iter(addr, bytes, i);
57 }
58 
59 static int dcssblk_dax_zero_page_range(struct dax_device *dax_dev,
60 				       pgoff_t pgoff, size_t nr_pages)
61 {
62 	long rc;
63 	void *kaddr;
64 
65 	rc = dax_direct_access(dax_dev, pgoff, nr_pages, &kaddr, NULL);
66 	if (rc < 0)
67 		return rc;
68 	memset(kaddr, 0, nr_pages << PAGE_SHIFT);
69 	dax_flush(dax_dev, kaddr, nr_pages << PAGE_SHIFT);
70 	return 0;
71 }
72 
73 static const struct dax_operations dcssblk_dax_ops = {
74 	.direct_access = dcssblk_dax_direct_access,
75 	.dax_supported = generic_fsdax_supported,
76 	.copy_from_iter = dcssblk_dax_copy_from_iter,
77 	.copy_to_iter = dcssblk_dax_copy_to_iter,
78 	.zero_page_range = dcssblk_dax_zero_page_range,
79 };
80 
81 struct dcssblk_dev_info {
82 	struct list_head lh;
83 	struct device dev;
84 	char segment_name[DCSS_BUS_ID_SIZE];
85 	atomic_t use_count;
86 	struct gendisk *gd;
87 	unsigned long start;
88 	unsigned long end;
89 	int segment_type;
90 	unsigned char save_pending;
91 	unsigned char is_shared;
92 	int num_of_segments;
93 	struct list_head seg_list;
94 	struct dax_device *dax_dev;
95 };
96 
97 struct segment_info {
98 	struct list_head lh;
99 	char segment_name[DCSS_BUS_ID_SIZE];
100 	unsigned long start;
101 	unsigned long end;
102 	int segment_type;
103 };
104 
105 static ssize_t dcssblk_add_store(struct device * dev, struct device_attribute *attr, const char * buf,
106 				  size_t count);
107 static ssize_t dcssblk_remove_store(struct device * dev, struct device_attribute *attr, const char * buf,
108 				  size_t count);
109 
110 static DEVICE_ATTR(add, S_IWUSR, NULL, dcssblk_add_store);
111 static DEVICE_ATTR(remove, S_IWUSR, NULL, dcssblk_remove_store);
112 
113 static struct device *dcssblk_root_dev;
114 
115 static LIST_HEAD(dcssblk_devices);
116 static struct rw_semaphore dcssblk_devices_sem;
117 
118 /*
119  * release function for segment device.
120  */
121 static void
122 dcssblk_release_segment(struct device *dev)
123 {
124 	struct dcssblk_dev_info *dev_info;
125 	struct segment_info *entry, *temp;
126 
127 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
128 	list_for_each_entry_safe(entry, temp, &dev_info->seg_list, lh) {
129 		list_del(&entry->lh);
130 		kfree(entry);
131 	}
132 	kfree(dev_info);
133 	module_put(THIS_MODULE);
134 }
135 
136 /*
137  * get a minor number. needs to be called with
138  * down_write(&dcssblk_devices_sem) and the
139  * device needs to be enqueued before the semaphore is
140  * freed.
141  */
142 static int
143 dcssblk_assign_free_minor(struct dcssblk_dev_info *dev_info)
144 {
145 	int minor, found;
146 	struct dcssblk_dev_info *entry;
147 
148 	if (dev_info == NULL)
149 		return -EINVAL;
150 	for (minor = 0; minor < (1<<MINORBITS); minor++) {
151 		found = 0;
152 		// test if minor available
153 		list_for_each_entry(entry, &dcssblk_devices, lh)
154 			if (minor == entry->gd->first_minor)
155 				found++;
156 		if (!found) break; // got unused minor
157 	}
158 	if (found)
159 		return -EBUSY;
160 	dev_info->gd->first_minor = minor;
161 	return 0;
162 }
163 
164 /*
165  * get the struct dcssblk_dev_info from dcssblk_devices
166  * for the given name.
167  * down_read(&dcssblk_devices_sem) must be held.
168  */
169 static struct dcssblk_dev_info *
170 dcssblk_get_device_by_name(char *name)
171 {
172 	struct dcssblk_dev_info *entry;
173 
174 	list_for_each_entry(entry, &dcssblk_devices, lh) {
175 		if (!strcmp(name, entry->segment_name)) {
176 			return entry;
177 		}
178 	}
179 	return NULL;
180 }
181 
182 /*
183  * get the struct segment_info from seg_list
184  * for the given name.
185  * down_read(&dcssblk_devices_sem) must be held.
186  */
187 static struct segment_info *
188 dcssblk_get_segment_by_name(char *name)
189 {
190 	struct dcssblk_dev_info *dev_info;
191 	struct segment_info *entry;
192 
193 	list_for_each_entry(dev_info, &dcssblk_devices, lh) {
194 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
195 			if (!strcmp(name, entry->segment_name))
196 				return entry;
197 		}
198 	}
199 	return NULL;
200 }
201 
202 /*
203  * get the highest address of the multi-segment block.
204  */
205 static unsigned long
206 dcssblk_find_highest_addr(struct dcssblk_dev_info *dev_info)
207 {
208 	unsigned long highest_addr;
209 	struct segment_info *entry;
210 
211 	highest_addr = 0;
212 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
213 		if (highest_addr < entry->end)
214 			highest_addr = entry->end;
215 	}
216 	return highest_addr;
217 }
218 
219 /*
220  * get the lowest address of the multi-segment block.
221  */
222 static unsigned long
223 dcssblk_find_lowest_addr(struct dcssblk_dev_info *dev_info)
224 {
225 	int set_first;
226 	unsigned long lowest_addr;
227 	struct segment_info *entry;
228 
229 	set_first = 0;
230 	lowest_addr = 0;
231 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
232 		if (set_first == 0) {
233 			lowest_addr = entry->start;
234 			set_first = 1;
235 		} else {
236 			if (lowest_addr > entry->start)
237 				lowest_addr = entry->start;
238 		}
239 	}
240 	return lowest_addr;
241 }
242 
243 /*
244  * Check continuity of segments.
245  */
246 static int
247 dcssblk_is_continuous(struct dcssblk_dev_info *dev_info)
248 {
249 	int i, j, rc;
250 	struct segment_info *sort_list, *entry, temp;
251 
252 	if (dev_info->num_of_segments <= 1)
253 		return 0;
254 
255 	sort_list = kcalloc(dev_info->num_of_segments,
256 			    sizeof(struct segment_info),
257 			    GFP_KERNEL);
258 	if (sort_list == NULL)
259 		return -ENOMEM;
260 	i = 0;
261 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
262 		memcpy(&sort_list[i], entry, sizeof(struct segment_info));
263 		i++;
264 	}
265 
266 	/* sort segments */
267 	for (i = 0; i < dev_info->num_of_segments; i++)
268 		for (j = 0; j < dev_info->num_of_segments; j++)
269 			if (sort_list[j].start > sort_list[i].start) {
270 				memcpy(&temp, &sort_list[i],
271 					sizeof(struct segment_info));
272 				memcpy(&sort_list[i], &sort_list[j],
273 					sizeof(struct segment_info));
274 				memcpy(&sort_list[j], &temp,
275 					sizeof(struct segment_info));
276 			}
277 
278 	/* check continuity */
279 	for (i = 0; i < dev_info->num_of_segments - 1; i++) {
280 		if ((sort_list[i].end + 1) != sort_list[i+1].start) {
281 			pr_err("Adjacent DCSSs %s and %s are not "
282 			       "contiguous\n", sort_list[i].segment_name,
283 			       sort_list[i+1].segment_name);
284 			rc = -EINVAL;
285 			goto out;
286 		}
287 		/* EN and EW are allowed in a block device */
288 		if (sort_list[i].segment_type != sort_list[i+1].segment_type) {
289 			if (!(sort_list[i].segment_type & SEGMENT_EXCLUSIVE) ||
290 				(sort_list[i].segment_type == SEG_TYPE_ER) ||
291 				!(sort_list[i+1].segment_type &
292 				SEGMENT_EXCLUSIVE) ||
293 				(sort_list[i+1].segment_type == SEG_TYPE_ER)) {
294 				pr_err("DCSS %s and DCSS %s have "
295 				       "incompatible types\n",
296 				       sort_list[i].segment_name,
297 				       sort_list[i+1].segment_name);
298 				rc = -EINVAL;
299 				goto out;
300 			}
301 		}
302 	}
303 	rc = 0;
304 out:
305 	kfree(sort_list);
306 	return rc;
307 }
308 
309 /*
310  * Load a segment
311  */
312 static int
313 dcssblk_load_segment(char *name, struct segment_info **seg_info)
314 {
315 	int rc;
316 
317 	/* already loaded? */
318 	down_read(&dcssblk_devices_sem);
319 	*seg_info = dcssblk_get_segment_by_name(name);
320 	up_read(&dcssblk_devices_sem);
321 	if (*seg_info != NULL)
322 		return -EEXIST;
323 
324 	/* get a struct segment_info */
325 	*seg_info = kzalloc(sizeof(struct segment_info), GFP_KERNEL);
326 	if (*seg_info == NULL)
327 		return -ENOMEM;
328 
329 	strcpy((*seg_info)->segment_name, name);
330 
331 	/* load the segment */
332 	rc = segment_load(name, SEGMENT_SHARED,
333 			&(*seg_info)->start, &(*seg_info)->end);
334 	if (rc < 0) {
335 		segment_warning(rc, (*seg_info)->segment_name);
336 		kfree(*seg_info);
337 	} else {
338 		INIT_LIST_HEAD(&(*seg_info)->lh);
339 		(*seg_info)->segment_type = rc;
340 	}
341 	return rc;
342 }
343 
344 /*
345  * device attribute for switching shared/nonshared (exclusive)
346  * operation (show + store)
347  */
348 static ssize_t
349 dcssblk_shared_show(struct device *dev, struct device_attribute *attr, char *buf)
350 {
351 	struct dcssblk_dev_info *dev_info;
352 
353 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
354 	return sprintf(buf, dev_info->is_shared ? "1\n" : "0\n");
355 }
356 
357 static ssize_t
358 dcssblk_shared_store(struct device *dev, struct device_attribute *attr, const char *inbuf, size_t count)
359 {
360 	struct dcssblk_dev_info *dev_info;
361 	struct segment_info *entry, *temp;
362 	int rc;
363 
364 	if ((count > 1) && (inbuf[1] != '\n') && (inbuf[1] != '\0'))
365 		return -EINVAL;
366 	down_write(&dcssblk_devices_sem);
367 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
368 	if (atomic_read(&dev_info->use_count)) {
369 		rc = -EBUSY;
370 		goto out;
371 	}
372 	if (inbuf[0] == '1') {
373 		/* reload segments in shared mode */
374 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
375 			rc = segment_modify_shared(entry->segment_name,
376 						SEGMENT_SHARED);
377 			if (rc < 0) {
378 				BUG_ON(rc == -EINVAL);
379 				if (rc != -EAGAIN)
380 					goto removeseg;
381 			}
382 		}
383 		dev_info->is_shared = 1;
384 		switch (dev_info->segment_type) {
385 		case SEG_TYPE_SR:
386 		case SEG_TYPE_ER:
387 		case SEG_TYPE_SC:
388 			set_disk_ro(dev_info->gd, 1);
389 		}
390 	} else if (inbuf[0] == '0') {
391 		/* reload segments in exclusive mode */
392 		if (dev_info->segment_type == SEG_TYPE_SC) {
393 			pr_err("DCSS %s is of type SC and cannot be "
394 			       "loaded as exclusive-writable\n",
395 			       dev_info->segment_name);
396 			rc = -EINVAL;
397 			goto out;
398 		}
399 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
400 			rc = segment_modify_shared(entry->segment_name,
401 						   SEGMENT_EXCLUSIVE);
402 			if (rc < 0) {
403 				BUG_ON(rc == -EINVAL);
404 				if (rc != -EAGAIN)
405 					goto removeseg;
406 			}
407 		}
408 		dev_info->is_shared = 0;
409 		set_disk_ro(dev_info->gd, 0);
410 	} else {
411 		rc = -EINVAL;
412 		goto out;
413 	}
414 	rc = count;
415 	goto out;
416 
417 removeseg:
418 	pr_err("DCSS device %s is removed after a failed access mode "
419 	       "change\n", dev_info->segment_name);
420 	temp = entry;
421 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
422 		if (entry != temp)
423 			segment_unload(entry->segment_name);
424 	}
425 	list_del(&dev_info->lh);
426 
427 	kill_dax(dev_info->dax_dev);
428 	put_dax(dev_info->dax_dev);
429 	del_gendisk(dev_info->gd);
430 	blk_cleanup_disk(dev_info->gd);
431 	up_write(&dcssblk_devices_sem);
432 
433 	if (device_remove_file_self(dev, attr)) {
434 		device_unregister(dev);
435 		put_device(dev);
436 	}
437 	return rc;
438 out:
439 	up_write(&dcssblk_devices_sem);
440 	return rc;
441 }
442 static DEVICE_ATTR(shared, S_IWUSR | S_IRUSR, dcssblk_shared_show,
443 		   dcssblk_shared_store);
444 
445 /*
446  * device attribute for save operation on current copy
447  * of the segment. If the segment is busy, saving will
448  * become pending until it gets released, which can be
449  * undone by storing a non-true value to this entry.
450  * (show + store)
451  */
452 static ssize_t
453 dcssblk_save_show(struct device *dev, struct device_attribute *attr, char *buf)
454 {
455 	struct dcssblk_dev_info *dev_info;
456 
457 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
458 	return sprintf(buf, dev_info->save_pending ? "1\n" : "0\n");
459 }
460 
461 static ssize_t
462 dcssblk_save_store(struct device *dev, struct device_attribute *attr, const char *inbuf, size_t count)
463 {
464 	struct dcssblk_dev_info *dev_info;
465 	struct segment_info *entry;
466 
467 	if ((count > 1) && (inbuf[1] != '\n') && (inbuf[1] != '\0'))
468 		return -EINVAL;
469 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
470 
471 	down_write(&dcssblk_devices_sem);
472 	if (inbuf[0] == '1') {
473 		if (atomic_read(&dev_info->use_count) == 0) {
474 			// device is idle => we save immediately
475 			pr_info("All DCSSs that map to device %s are "
476 				"saved\n", dev_info->segment_name);
477 			list_for_each_entry(entry, &dev_info->seg_list, lh) {
478 				if (entry->segment_type == SEG_TYPE_EN ||
479 				    entry->segment_type == SEG_TYPE_SN)
480 					pr_warn("DCSS %s is of type SN or EN"
481 						" and cannot be saved\n",
482 						entry->segment_name);
483 				else
484 					segment_save(entry->segment_name);
485 			}
486 		}  else {
487 			// device is busy => we save it when it becomes
488 			// idle in dcssblk_release
489 			pr_info("Device %s is in use, its DCSSs will be "
490 				"saved when it becomes idle\n",
491 				dev_info->segment_name);
492 			dev_info->save_pending = 1;
493 		}
494 	} else if (inbuf[0] == '0') {
495 		if (dev_info->save_pending) {
496 			// device is busy & the user wants to undo his save
497 			// request
498 			dev_info->save_pending = 0;
499 			pr_info("A pending save request for device %s "
500 				"has been canceled\n",
501 				dev_info->segment_name);
502 		}
503 	} else {
504 		up_write(&dcssblk_devices_sem);
505 		return -EINVAL;
506 	}
507 	up_write(&dcssblk_devices_sem);
508 	return count;
509 }
510 static DEVICE_ATTR(save, S_IWUSR | S_IRUSR, dcssblk_save_show,
511 		   dcssblk_save_store);
512 
513 /*
514  * device attribute for showing all segments in a device
515  */
516 static ssize_t
517 dcssblk_seglist_show(struct device *dev, struct device_attribute *attr,
518 		char *buf)
519 {
520 	int i;
521 
522 	struct dcssblk_dev_info *dev_info;
523 	struct segment_info *entry;
524 
525 	down_read(&dcssblk_devices_sem);
526 	dev_info = container_of(dev, struct dcssblk_dev_info, dev);
527 	i = 0;
528 	buf[0] = '\0';
529 	list_for_each_entry(entry, &dev_info->seg_list, lh) {
530 		strcpy(&buf[i], entry->segment_name);
531 		i += strlen(entry->segment_name);
532 		buf[i] = '\n';
533 		i++;
534 	}
535 	up_read(&dcssblk_devices_sem);
536 	return i;
537 }
538 static DEVICE_ATTR(seglist, S_IRUSR, dcssblk_seglist_show, NULL);
539 
540 static struct attribute *dcssblk_dev_attrs[] = {
541 	&dev_attr_shared.attr,
542 	&dev_attr_save.attr,
543 	&dev_attr_seglist.attr,
544 	NULL,
545 };
546 static struct attribute_group dcssblk_dev_attr_group = {
547 	.attrs = dcssblk_dev_attrs,
548 };
549 static const struct attribute_group *dcssblk_dev_attr_groups[] = {
550 	&dcssblk_dev_attr_group,
551 	NULL,
552 };
553 
554 /*
555  * device attribute for adding devices
556  */
557 static ssize_t
558 dcssblk_add_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
559 {
560 	int rc, i, j, num_of_segments;
561 	struct dcssblk_dev_info *dev_info;
562 	struct segment_info *seg_info, *temp;
563 	char *local_buf;
564 	unsigned long seg_byte_size;
565 
566 	dev_info = NULL;
567 	seg_info = NULL;
568 	if (dev != dcssblk_root_dev) {
569 		rc = -EINVAL;
570 		goto out_nobuf;
571 	}
572 	if ((count < 1) || (buf[0] == '\0') || (buf[0] == '\n')) {
573 		rc = -ENAMETOOLONG;
574 		goto out_nobuf;
575 	}
576 
577 	local_buf = kmalloc(count + 1, GFP_KERNEL);
578 	if (local_buf == NULL) {
579 		rc = -ENOMEM;
580 		goto out_nobuf;
581 	}
582 
583 	/*
584 	 * parse input
585 	 */
586 	num_of_segments = 0;
587 	for (i = 0; (i < count && (buf[i] != '\0') && (buf[i] != '\n')); i++) {
588 		for (j = i; j < count &&
589 			(buf[j] != ':') &&
590 			(buf[j] != '\0') &&
591 			(buf[j] != '\n'); j++) {
592 			local_buf[j-i] = toupper(buf[j]);
593 		}
594 		local_buf[j-i] = '\0';
595 		if (((j - i) == 0) || ((j - i) > 8)) {
596 			rc = -ENAMETOOLONG;
597 			goto seg_list_del;
598 		}
599 
600 		rc = dcssblk_load_segment(local_buf, &seg_info);
601 		if (rc < 0)
602 			goto seg_list_del;
603 		/*
604 		 * get a struct dcssblk_dev_info
605 		 */
606 		if (num_of_segments == 0) {
607 			dev_info = kzalloc(sizeof(struct dcssblk_dev_info),
608 					GFP_KERNEL);
609 			if (dev_info == NULL) {
610 				rc = -ENOMEM;
611 				goto out;
612 			}
613 			strcpy(dev_info->segment_name, local_buf);
614 			dev_info->segment_type = seg_info->segment_type;
615 			INIT_LIST_HEAD(&dev_info->seg_list);
616 		}
617 		list_add_tail(&seg_info->lh, &dev_info->seg_list);
618 		num_of_segments++;
619 		i = j;
620 
621 		if ((buf[j] == '\0') || (buf[j] == '\n'))
622 			break;
623 	}
624 
625 	/* no trailing colon at the end of the input */
626 	if ((i > 0) && (buf[i-1] == ':')) {
627 		rc = -ENAMETOOLONG;
628 		goto seg_list_del;
629 	}
630 	strlcpy(local_buf, buf, i + 1);
631 	dev_info->num_of_segments = num_of_segments;
632 	rc = dcssblk_is_continuous(dev_info);
633 	if (rc < 0)
634 		goto seg_list_del;
635 
636 	dev_info->start = dcssblk_find_lowest_addr(dev_info);
637 	dev_info->end = dcssblk_find_highest_addr(dev_info);
638 
639 	dev_set_name(&dev_info->dev, "%s", dev_info->segment_name);
640 	dev_info->dev.release = dcssblk_release_segment;
641 	dev_info->dev.groups = dcssblk_dev_attr_groups;
642 	INIT_LIST_HEAD(&dev_info->lh);
643 	dev_info->gd = blk_alloc_disk(NUMA_NO_NODE);
644 	if (dev_info->gd == NULL) {
645 		rc = -ENOMEM;
646 		goto seg_list_del;
647 	}
648 	dev_info->gd->major = dcssblk_major;
649 	dev_info->gd->minors = DCSSBLK_MINORS_PER_DISK;
650 	dev_info->gd->fops = &dcssblk_devops;
651 	dev_info->gd->private_data = dev_info;
652 	blk_queue_logical_block_size(dev_info->gd->queue, 4096);
653 	blk_queue_flag_set(QUEUE_FLAG_DAX, dev_info->gd->queue);
654 
655 	seg_byte_size = (dev_info->end - dev_info->start + 1);
656 	set_capacity(dev_info->gd, seg_byte_size >> 9); // size in sectors
657 	pr_info("Loaded %s with total size %lu bytes and capacity %lu "
658 		"sectors\n", local_buf, seg_byte_size, seg_byte_size >> 9);
659 
660 	dev_info->save_pending = 0;
661 	dev_info->is_shared = 1;
662 	dev_info->dev.parent = dcssblk_root_dev;
663 
664 	/*
665 	 *get minor, add to list
666 	 */
667 	down_write(&dcssblk_devices_sem);
668 	if (dcssblk_get_segment_by_name(local_buf)) {
669 		rc = -EEXIST;
670 		goto release_gd;
671 	}
672 	rc = dcssblk_assign_free_minor(dev_info);
673 	if (rc)
674 		goto release_gd;
675 	sprintf(dev_info->gd->disk_name, "dcssblk%d",
676 		dev_info->gd->first_minor);
677 	list_add_tail(&dev_info->lh, &dcssblk_devices);
678 
679 	if (!try_module_get(THIS_MODULE)) {
680 		rc = -ENODEV;
681 		goto dev_list_del;
682 	}
683 	/*
684 	 * register the device
685 	 */
686 	rc = device_register(&dev_info->dev);
687 	if (rc)
688 		goto put_dev;
689 
690 	dev_info->dax_dev = alloc_dax(dev_info, dev_info->gd->disk_name,
691 			&dcssblk_dax_ops, DAXDEV_F_SYNC);
692 	if (IS_ERR(dev_info->dax_dev)) {
693 		rc = PTR_ERR(dev_info->dax_dev);
694 		dev_info->dax_dev = NULL;
695 		goto put_dev;
696 	}
697 
698 	get_device(&dev_info->dev);
699 	device_add_disk(&dev_info->dev, dev_info->gd, NULL);
700 
701 	switch (dev_info->segment_type) {
702 		case SEG_TYPE_SR:
703 		case SEG_TYPE_ER:
704 		case SEG_TYPE_SC:
705 			set_disk_ro(dev_info->gd,1);
706 			break;
707 		default:
708 			set_disk_ro(dev_info->gd,0);
709 			break;
710 	}
711 	up_write(&dcssblk_devices_sem);
712 	rc = count;
713 	goto out;
714 
715 put_dev:
716 	list_del(&dev_info->lh);
717 	blk_cleanup_disk(dev_info->gd);
718 	list_for_each_entry(seg_info, &dev_info->seg_list, lh) {
719 		segment_unload(seg_info->segment_name);
720 	}
721 	put_device(&dev_info->dev);
722 	up_write(&dcssblk_devices_sem);
723 	goto out;
724 dev_list_del:
725 	list_del(&dev_info->lh);
726 release_gd:
727 	blk_cleanup_disk(dev_info->gd);
728 	up_write(&dcssblk_devices_sem);
729 seg_list_del:
730 	if (dev_info == NULL)
731 		goto out;
732 	list_for_each_entry_safe(seg_info, temp, &dev_info->seg_list, lh) {
733 		list_del(&seg_info->lh);
734 		segment_unload(seg_info->segment_name);
735 		kfree(seg_info);
736 	}
737 	kfree(dev_info);
738 out:
739 	kfree(local_buf);
740 out_nobuf:
741 	return rc;
742 }
743 
744 /*
745  * device attribute for removing devices
746  */
747 static ssize_t
748 dcssblk_remove_store(struct device *dev, struct device_attribute *attr, const char *buf, size_t count)
749 {
750 	struct dcssblk_dev_info *dev_info;
751 	struct segment_info *entry;
752 	int rc, i;
753 	char *local_buf;
754 
755 	if (dev != dcssblk_root_dev) {
756 		return -EINVAL;
757 	}
758 	local_buf = kmalloc(count + 1, GFP_KERNEL);
759 	if (local_buf == NULL) {
760 		return -ENOMEM;
761 	}
762 	/*
763 	 * parse input
764 	 */
765 	for (i = 0; (i < count && (*(buf+i)!='\0') && (*(buf+i)!='\n')); i++) {
766 		local_buf[i] = toupper(buf[i]);
767 	}
768 	local_buf[i] = '\0';
769 	if ((i == 0) || (i > 8)) {
770 		rc = -ENAMETOOLONG;
771 		goto out_buf;
772 	}
773 
774 	down_write(&dcssblk_devices_sem);
775 	dev_info = dcssblk_get_device_by_name(local_buf);
776 	if (dev_info == NULL) {
777 		up_write(&dcssblk_devices_sem);
778 		pr_warn("Device %s cannot be removed because it is not a known device\n",
779 			local_buf);
780 		rc = -ENODEV;
781 		goto out_buf;
782 	}
783 	if (atomic_read(&dev_info->use_count) != 0) {
784 		up_write(&dcssblk_devices_sem);
785 		pr_warn("Device %s cannot be removed while it is in use\n",
786 			local_buf);
787 		rc = -EBUSY;
788 		goto out_buf;
789 	}
790 
791 	list_del(&dev_info->lh);
792 	kill_dax(dev_info->dax_dev);
793 	put_dax(dev_info->dax_dev);
794 	del_gendisk(dev_info->gd);
795 	blk_cleanup_disk(dev_info->gd);
796 
797 	/* unload all related segments */
798 	list_for_each_entry(entry, &dev_info->seg_list, lh)
799 		segment_unload(entry->segment_name);
800 
801 	up_write(&dcssblk_devices_sem);
802 
803 	device_unregister(&dev_info->dev);
804 	put_device(&dev_info->dev);
805 
806 	rc = count;
807 out_buf:
808 	kfree(local_buf);
809 	return rc;
810 }
811 
812 static int
813 dcssblk_open(struct block_device *bdev, fmode_t mode)
814 {
815 	struct dcssblk_dev_info *dev_info;
816 	int rc;
817 
818 	dev_info = bdev->bd_disk->private_data;
819 	if (NULL == dev_info) {
820 		rc = -ENODEV;
821 		goto out;
822 	}
823 	atomic_inc(&dev_info->use_count);
824 	rc = 0;
825 out:
826 	return rc;
827 }
828 
829 static void
830 dcssblk_release(struct gendisk *disk, fmode_t mode)
831 {
832 	struct dcssblk_dev_info *dev_info = disk->private_data;
833 	struct segment_info *entry;
834 
835 	if (!dev_info) {
836 		WARN_ON(1);
837 		return;
838 	}
839 	down_write(&dcssblk_devices_sem);
840 	if (atomic_dec_and_test(&dev_info->use_count)
841 	    && (dev_info->save_pending)) {
842 		pr_info("Device %s has become idle and is being saved "
843 			"now\n", dev_info->segment_name);
844 		list_for_each_entry(entry, &dev_info->seg_list, lh) {
845 			if (entry->segment_type == SEG_TYPE_EN ||
846 			    entry->segment_type == SEG_TYPE_SN)
847 				pr_warn("DCSS %s is of type SN or EN and cannot"
848 					" be saved\n", entry->segment_name);
849 			else
850 				segment_save(entry->segment_name);
851 		}
852 		dev_info->save_pending = 0;
853 	}
854 	up_write(&dcssblk_devices_sem);
855 }
856 
857 static blk_qc_t
858 dcssblk_submit_bio(struct bio *bio)
859 {
860 	struct dcssblk_dev_info *dev_info;
861 	struct bio_vec bvec;
862 	struct bvec_iter iter;
863 	unsigned long index;
864 	unsigned long page_addr;
865 	unsigned long source_addr;
866 	unsigned long bytes_done;
867 
868 	blk_queue_split(&bio);
869 
870 	bytes_done = 0;
871 	dev_info = bio->bi_bdev->bd_disk->private_data;
872 	if (dev_info == NULL)
873 		goto fail;
874 	if ((bio->bi_iter.bi_sector & 7) != 0 ||
875 	    (bio->bi_iter.bi_size & 4095) != 0)
876 		/* Request is not page-aligned. */
877 		goto fail;
878 	/* verify data transfer direction */
879 	if (dev_info->is_shared) {
880 		switch (dev_info->segment_type) {
881 		case SEG_TYPE_SR:
882 		case SEG_TYPE_ER:
883 		case SEG_TYPE_SC:
884 			/* cannot write to these segments */
885 			if (bio_data_dir(bio) == WRITE) {
886 				pr_warn("Writing to %s failed because it is a read-only device\n",
887 					dev_name(&dev_info->dev));
888 				goto fail;
889 			}
890 		}
891 	}
892 
893 	index = (bio->bi_iter.bi_sector >> 3);
894 	bio_for_each_segment(bvec, bio, iter) {
895 		page_addr = (unsigned long)bvec_virt(&bvec);
896 		source_addr = dev_info->start + (index<<12) + bytes_done;
897 		if (unlikely((page_addr & 4095) != 0) || (bvec.bv_len & 4095) != 0)
898 			// More paranoia.
899 			goto fail;
900 		if (bio_data_dir(bio) == READ) {
901 			memcpy((void*)page_addr, (void*)source_addr,
902 				bvec.bv_len);
903 		} else {
904 			memcpy((void*)source_addr, (void*)page_addr,
905 				bvec.bv_len);
906 		}
907 		bytes_done += bvec.bv_len;
908 	}
909 	bio_endio(bio);
910 	return BLK_QC_T_NONE;
911 fail:
912 	bio_io_error(bio);
913 	return BLK_QC_T_NONE;
914 }
915 
916 static long
917 __dcssblk_direct_access(struct dcssblk_dev_info *dev_info, pgoff_t pgoff,
918 		long nr_pages, void **kaddr, pfn_t *pfn)
919 {
920 	resource_size_t offset = pgoff * PAGE_SIZE;
921 	unsigned long dev_sz;
922 
923 	dev_sz = dev_info->end - dev_info->start + 1;
924 	if (kaddr)
925 		*kaddr = (void *) dev_info->start + offset;
926 	if (pfn)
927 		*pfn = __pfn_to_pfn_t(PFN_DOWN(dev_info->start + offset),
928 				PFN_DEV|PFN_SPECIAL);
929 
930 	return (dev_sz - offset) / PAGE_SIZE;
931 }
932 
933 static long
934 dcssblk_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
935 		long nr_pages, void **kaddr, pfn_t *pfn)
936 {
937 	struct dcssblk_dev_info *dev_info = dax_get_private(dax_dev);
938 
939 	return __dcssblk_direct_access(dev_info, pgoff, nr_pages, kaddr, pfn);
940 }
941 
942 static void
943 dcssblk_check_params(void)
944 {
945 	int rc, i, j, k;
946 	char buf[DCSSBLK_PARM_LEN + 1];
947 	struct dcssblk_dev_info *dev_info;
948 
949 	for (i = 0; (i < DCSSBLK_PARM_LEN) && (dcssblk_segments[i] != '\0');
950 	     i++) {
951 		for (j = i; (j < DCSSBLK_PARM_LEN) &&
952 			    (dcssblk_segments[j] != ',')  &&
953 			    (dcssblk_segments[j] != '\0') &&
954 			    (dcssblk_segments[j] != '('); j++)
955 		{
956 			buf[j-i] = dcssblk_segments[j];
957 		}
958 		buf[j-i] = '\0';
959 		rc = dcssblk_add_store(dcssblk_root_dev, NULL, buf, j-i);
960 		if ((rc >= 0) && (dcssblk_segments[j] == '(')) {
961 			for (k = 0; (buf[k] != ':') && (buf[k] != '\0'); k++)
962 				buf[k] = toupper(buf[k]);
963 			buf[k] = '\0';
964 			if (!strncmp(&dcssblk_segments[j], "(local)", 7)) {
965 				down_read(&dcssblk_devices_sem);
966 				dev_info = dcssblk_get_device_by_name(buf);
967 				up_read(&dcssblk_devices_sem);
968 				if (dev_info)
969 					dcssblk_shared_store(&dev_info->dev,
970 							     NULL, "0\n", 2);
971 			}
972 		}
973 		while ((dcssblk_segments[j] != ',') &&
974 		       (dcssblk_segments[j] != '\0'))
975 		{
976 			j++;
977 		}
978 		if (dcssblk_segments[j] == '\0')
979 			break;
980 		i = j;
981 	}
982 }
983 
984 /*
985  * The init/exit functions.
986  */
987 static void __exit
988 dcssblk_exit(void)
989 {
990 	root_device_unregister(dcssblk_root_dev);
991 	unregister_blkdev(dcssblk_major, DCSSBLK_NAME);
992 }
993 
994 static int __init
995 dcssblk_init(void)
996 {
997 	int rc;
998 
999 	dcssblk_root_dev = root_device_register("dcssblk");
1000 	if (IS_ERR(dcssblk_root_dev))
1001 		return PTR_ERR(dcssblk_root_dev);
1002 	rc = device_create_file(dcssblk_root_dev, &dev_attr_add);
1003 	if (rc)
1004 		goto out_root;
1005 	rc = device_create_file(dcssblk_root_dev, &dev_attr_remove);
1006 	if (rc)
1007 		goto out_root;
1008 	rc = register_blkdev(0, DCSSBLK_NAME);
1009 	if (rc < 0)
1010 		goto out_root;
1011 	dcssblk_major = rc;
1012 	init_rwsem(&dcssblk_devices_sem);
1013 
1014 	dcssblk_check_params();
1015 	return 0;
1016 
1017 out_root:
1018 	root_device_unregister(dcssblk_root_dev);
1019 
1020 	return rc;
1021 }
1022 
1023 module_init(dcssblk_init);
1024 module_exit(dcssblk_exit);
1025 
1026 module_param_string(segments, dcssblk_segments, DCSSBLK_PARM_LEN, 0444);
1027 MODULE_PARM_DESC(segments, "Name of DCSS segment(s) to be loaded, "
1028 		 "comma-separated list, names in each set separated "
1029 		 "by commas are separated by colons, each set contains "
1030 		 "names of contiguous segments and each name max. 8 chars.\n"
1031 		 "Adding \"(local)\" to the end of each set equals echoing 0 "
1032 		 "to /sys/devices/dcssblk/<device name>/shared after loading "
1033 		 "the contiguous segments - \n"
1034 		 "e.g. segments=\"mydcss1,mydcss2:mydcss3,mydcss4(local)\"");
1035 
1036 MODULE_LICENSE("GPL");
1037