xref: /openbmc/linux/drivers/block/zram/zram_drv.c (revision a2cce7a9)
1 /*
2  * Compressed RAM block device
3  *
4  * Copyright (C) 2008, 2009, 2010  Nitin Gupta
5  *               2012, 2013 Minchan Kim
6  *
7  * This code is released using a dual license strategy: BSD/GPL
8  * You can choose the licence that better fits your requirements.
9  *
10  * Released under the terms of 3-clause BSD License
11  * Released under the terms of GNU General Public License Version 2.0
12  *
13  */
14 
15 #define KMSG_COMPONENT "zram"
16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
17 
18 #include <linux/module.h>
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/bitops.h>
22 #include <linux/blkdev.h>
23 #include <linux/buffer_head.h>
24 #include <linux/device.h>
25 #include <linux/genhd.h>
26 #include <linux/highmem.h>
27 #include <linux/slab.h>
28 #include <linux/string.h>
29 #include <linux/vmalloc.h>
30 #include <linux/err.h>
31 #include <linux/idr.h>
32 #include <linux/sysfs.h>
33 
34 #include "zram_drv.h"
35 
36 static DEFINE_IDR(zram_index_idr);
37 /* idr index must be protected */
38 static DEFINE_MUTEX(zram_index_mutex);
39 
40 static int zram_major;
41 static const char *default_compressor = "lzo";
42 
43 /* Module params (documentation at end) */
44 static unsigned int num_devices = 1;
45 
46 static inline void deprecated_attr_warn(const char *name)
47 {
48 	pr_warn_once("%d (%s) Attribute %s (and others) will be removed. %s\n",
49 			task_pid_nr(current),
50 			current->comm,
51 			name,
52 			"See zram documentation.");
53 }
54 
55 #define ZRAM_ATTR_RO(name)						\
56 static ssize_t name##_show(struct device *d,				\
57 				struct device_attribute *attr, char *b)	\
58 {									\
59 	struct zram *zram = dev_to_zram(d);				\
60 									\
61 	deprecated_attr_warn(__stringify(name));			\
62 	return scnprintf(b, PAGE_SIZE, "%llu\n",			\
63 		(u64)atomic64_read(&zram->stats.name));			\
64 }									\
65 static DEVICE_ATTR_RO(name);
66 
67 static inline bool init_done(struct zram *zram)
68 {
69 	return zram->disksize;
70 }
71 
72 static inline struct zram *dev_to_zram(struct device *dev)
73 {
74 	return (struct zram *)dev_to_disk(dev)->private_data;
75 }
76 
77 /* flag operations require table entry bit_spin_lock() being held */
78 static int zram_test_flag(struct zram_meta *meta, u32 index,
79 			enum zram_pageflags flag)
80 {
81 	return meta->table[index].value & BIT(flag);
82 }
83 
84 static void zram_set_flag(struct zram_meta *meta, u32 index,
85 			enum zram_pageflags flag)
86 {
87 	meta->table[index].value |= BIT(flag);
88 }
89 
90 static void zram_clear_flag(struct zram_meta *meta, u32 index,
91 			enum zram_pageflags flag)
92 {
93 	meta->table[index].value &= ~BIT(flag);
94 }
95 
96 static size_t zram_get_obj_size(struct zram_meta *meta, u32 index)
97 {
98 	return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
99 }
100 
101 static void zram_set_obj_size(struct zram_meta *meta,
102 					u32 index, size_t size)
103 {
104 	unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT;
105 
106 	meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
107 }
108 
109 static inline int is_partial_io(struct bio_vec *bvec)
110 {
111 	return bvec->bv_len != PAGE_SIZE;
112 }
113 
114 /*
115  * Check if request is within bounds and aligned on zram logical blocks.
116  */
117 static inline int valid_io_request(struct zram *zram,
118 		sector_t start, unsigned int size)
119 {
120 	u64 end, bound;
121 
122 	/* unaligned request */
123 	if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
124 		return 0;
125 	if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
126 		return 0;
127 
128 	end = start + (size >> SECTOR_SHIFT);
129 	bound = zram->disksize >> SECTOR_SHIFT;
130 	/* out of range range */
131 	if (unlikely(start >= bound || end > bound || start > end))
132 		return 0;
133 
134 	/* I/O request is valid */
135 	return 1;
136 }
137 
138 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
139 {
140 	if (*offset + bvec->bv_len >= PAGE_SIZE)
141 		(*index)++;
142 	*offset = (*offset + bvec->bv_len) % PAGE_SIZE;
143 }
144 
145 static inline void update_used_max(struct zram *zram,
146 					const unsigned long pages)
147 {
148 	unsigned long old_max, cur_max;
149 
150 	old_max = atomic_long_read(&zram->stats.max_used_pages);
151 
152 	do {
153 		cur_max = old_max;
154 		if (pages > cur_max)
155 			old_max = atomic_long_cmpxchg(
156 				&zram->stats.max_used_pages, cur_max, pages);
157 	} while (old_max != cur_max);
158 }
159 
160 static int page_zero_filled(void *ptr)
161 {
162 	unsigned int pos;
163 	unsigned long *page;
164 
165 	page = (unsigned long *)ptr;
166 
167 	for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
168 		if (page[pos])
169 			return 0;
170 	}
171 
172 	return 1;
173 }
174 
175 static void handle_zero_page(struct bio_vec *bvec)
176 {
177 	struct page *page = bvec->bv_page;
178 	void *user_mem;
179 
180 	user_mem = kmap_atomic(page);
181 	if (is_partial_io(bvec))
182 		memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
183 	else
184 		clear_page(user_mem);
185 	kunmap_atomic(user_mem);
186 
187 	flush_dcache_page(page);
188 }
189 
190 static ssize_t initstate_show(struct device *dev,
191 		struct device_attribute *attr, char *buf)
192 {
193 	u32 val;
194 	struct zram *zram = dev_to_zram(dev);
195 
196 	down_read(&zram->init_lock);
197 	val = init_done(zram);
198 	up_read(&zram->init_lock);
199 
200 	return scnprintf(buf, PAGE_SIZE, "%u\n", val);
201 }
202 
203 static ssize_t disksize_show(struct device *dev,
204 		struct device_attribute *attr, char *buf)
205 {
206 	struct zram *zram = dev_to_zram(dev);
207 
208 	return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
209 }
210 
211 static ssize_t orig_data_size_show(struct device *dev,
212 		struct device_attribute *attr, char *buf)
213 {
214 	struct zram *zram = dev_to_zram(dev);
215 
216 	deprecated_attr_warn("orig_data_size");
217 	return scnprintf(buf, PAGE_SIZE, "%llu\n",
218 		(u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
219 }
220 
221 static ssize_t mem_used_total_show(struct device *dev,
222 		struct device_attribute *attr, char *buf)
223 {
224 	u64 val = 0;
225 	struct zram *zram = dev_to_zram(dev);
226 
227 	deprecated_attr_warn("mem_used_total");
228 	down_read(&zram->init_lock);
229 	if (init_done(zram)) {
230 		struct zram_meta *meta = zram->meta;
231 		val = zs_get_total_pages(meta->mem_pool);
232 	}
233 	up_read(&zram->init_lock);
234 
235 	return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
236 }
237 
238 static ssize_t mem_limit_show(struct device *dev,
239 		struct device_attribute *attr, char *buf)
240 {
241 	u64 val;
242 	struct zram *zram = dev_to_zram(dev);
243 
244 	deprecated_attr_warn("mem_limit");
245 	down_read(&zram->init_lock);
246 	val = zram->limit_pages;
247 	up_read(&zram->init_lock);
248 
249 	return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
250 }
251 
252 static ssize_t mem_limit_store(struct device *dev,
253 		struct device_attribute *attr, const char *buf, size_t len)
254 {
255 	u64 limit;
256 	char *tmp;
257 	struct zram *zram = dev_to_zram(dev);
258 
259 	limit = memparse(buf, &tmp);
260 	if (buf == tmp) /* no chars parsed, invalid input */
261 		return -EINVAL;
262 
263 	down_write(&zram->init_lock);
264 	zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
265 	up_write(&zram->init_lock);
266 
267 	return len;
268 }
269 
270 static ssize_t mem_used_max_show(struct device *dev,
271 		struct device_attribute *attr, char *buf)
272 {
273 	u64 val = 0;
274 	struct zram *zram = dev_to_zram(dev);
275 
276 	deprecated_attr_warn("mem_used_max");
277 	down_read(&zram->init_lock);
278 	if (init_done(zram))
279 		val = atomic_long_read(&zram->stats.max_used_pages);
280 	up_read(&zram->init_lock);
281 
282 	return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
283 }
284 
285 static ssize_t mem_used_max_store(struct device *dev,
286 		struct device_attribute *attr, const char *buf, size_t len)
287 {
288 	int err;
289 	unsigned long val;
290 	struct zram *zram = dev_to_zram(dev);
291 
292 	err = kstrtoul(buf, 10, &val);
293 	if (err || val != 0)
294 		return -EINVAL;
295 
296 	down_read(&zram->init_lock);
297 	if (init_done(zram)) {
298 		struct zram_meta *meta = zram->meta;
299 		atomic_long_set(&zram->stats.max_used_pages,
300 				zs_get_total_pages(meta->mem_pool));
301 	}
302 	up_read(&zram->init_lock);
303 
304 	return len;
305 }
306 
307 static ssize_t max_comp_streams_show(struct device *dev,
308 		struct device_attribute *attr, char *buf)
309 {
310 	int val;
311 	struct zram *zram = dev_to_zram(dev);
312 
313 	down_read(&zram->init_lock);
314 	val = zram->max_comp_streams;
315 	up_read(&zram->init_lock);
316 
317 	return scnprintf(buf, PAGE_SIZE, "%d\n", val);
318 }
319 
320 static ssize_t max_comp_streams_store(struct device *dev,
321 		struct device_attribute *attr, const char *buf, size_t len)
322 {
323 	int num;
324 	struct zram *zram = dev_to_zram(dev);
325 	int ret;
326 
327 	ret = kstrtoint(buf, 0, &num);
328 	if (ret < 0)
329 		return ret;
330 	if (num < 1)
331 		return -EINVAL;
332 
333 	down_write(&zram->init_lock);
334 	if (init_done(zram)) {
335 		if (!zcomp_set_max_streams(zram->comp, num)) {
336 			pr_info("Cannot change max compression streams\n");
337 			ret = -EINVAL;
338 			goto out;
339 		}
340 	}
341 
342 	zram->max_comp_streams = num;
343 	ret = len;
344 out:
345 	up_write(&zram->init_lock);
346 	return ret;
347 }
348 
349 static ssize_t comp_algorithm_show(struct device *dev,
350 		struct device_attribute *attr, char *buf)
351 {
352 	size_t sz;
353 	struct zram *zram = dev_to_zram(dev);
354 
355 	down_read(&zram->init_lock);
356 	sz = zcomp_available_show(zram->compressor, buf);
357 	up_read(&zram->init_lock);
358 
359 	return sz;
360 }
361 
362 static ssize_t comp_algorithm_store(struct device *dev,
363 		struct device_attribute *attr, const char *buf, size_t len)
364 {
365 	struct zram *zram = dev_to_zram(dev);
366 	size_t sz;
367 
368 	down_write(&zram->init_lock);
369 	if (init_done(zram)) {
370 		up_write(&zram->init_lock);
371 		pr_info("Can't change algorithm for initialized device\n");
372 		return -EBUSY;
373 	}
374 	strlcpy(zram->compressor, buf, sizeof(zram->compressor));
375 
376 	/* ignore trailing newline */
377 	sz = strlen(zram->compressor);
378 	if (sz > 0 && zram->compressor[sz - 1] == '\n')
379 		zram->compressor[sz - 1] = 0x00;
380 
381 	if (!zcomp_available_algorithm(zram->compressor))
382 		len = -EINVAL;
383 
384 	up_write(&zram->init_lock);
385 	return len;
386 }
387 
388 static ssize_t compact_store(struct device *dev,
389 		struct device_attribute *attr, const char *buf, size_t len)
390 {
391 	struct zram *zram = dev_to_zram(dev);
392 	struct zram_meta *meta;
393 
394 	down_read(&zram->init_lock);
395 	if (!init_done(zram)) {
396 		up_read(&zram->init_lock);
397 		return -EINVAL;
398 	}
399 
400 	meta = zram->meta;
401 	zs_compact(meta->mem_pool);
402 	up_read(&zram->init_lock);
403 
404 	return len;
405 }
406 
407 static ssize_t io_stat_show(struct device *dev,
408 		struct device_attribute *attr, char *buf)
409 {
410 	struct zram *zram = dev_to_zram(dev);
411 	ssize_t ret;
412 
413 	down_read(&zram->init_lock);
414 	ret = scnprintf(buf, PAGE_SIZE,
415 			"%8llu %8llu %8llu %8llu\n",
416 			(u64)atomic64_read(&zram->stats.failed_reads),
417 			(u64)atomic64_read(&zram->stats.failed_writes),
418 			(u64)atomic64_read(&zram->stats.invalid_io),
419 			(u64)atomic64_read(&zram->stats.notify_free));
420 	up_read(&zram->init_lock);
421 
422 	return ret;
423 }
424 
425 static ssize_t mm_stat_show(struct device *dev,
426 		struct device_attribute *attr, char *buf)
427 {
428 	struct zram *zram = dev_to_zram(dev);
429 	struct zs_pool_stats pool_stats;
430 	u64 orig_size, mem_used = 0;
431 	long max_used;
432 	ssize_t ret;
433 
434 	memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
435 
436 	down_read(&zram->init_lock);
437 	if (init_done(zram)) {
438 		mem_used = zs_get_total_pages(zram->meta->mem_pool);
439 		zs_pool_stats(zram->meta->mem_pool, &pool_stats);
440 	}
441 
442 	orig_size = atomic64_read(&zram->stats.pages_stored);
443 	max_used = atomic_long_read(&zram->stats.max_used_pages);
444 
445 	ret = scnprintf(buf, PAGE_SIZE,
446 			"%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n",
447 			orig_size << PAGE_SHIFT,
448 			(u64)atomic64_read(&zram->stats.compr_data_size),
449 			mem_used << PAGE_SHIFT,
450 			zram->limit_pages << PAGE_SHIFT,
451 			max_used << PAGE_SHIFT,
452 			(u64)atomic64_read(&zram->stats.zero_pages),
453 			pool_stats.pages_compacted);
454 	up_read(&zram->init_lock);
455 
456 	return ret;
457 }
458 
459 static DEVICE_ATTR_RO(io_stat);
460 static DEVICE_ATTR_RO(mm_stat);
461 ZRAM_ATTR_RO(num_reads);
462 ZRAM_ATTR_RO(num_writes);
463 ZRAM_ATTR_RO(failed_reads);
464 ZRAM_ATTR_RO(failed_writes);
465 ZRAM_ATTR_RO(invalid_io);
466 ZRAM_ATTR_RO(notify_free);
467 ZRAM_ATTR_RO(zero_pages);
468 ZRAM_ATTR_RO(compr_data_size);
469 
470 static inline bool zram_meta_get(struct zram *zram)
471 {
472 	if (atomic_inc_not_zero(&zram->refcount))
473 		return true;
474 	return false;
475 }
476 
477 static inline void zram_meta_put(struct zram *zram)
478 {
479 	atomic_dec(&zram->refcount);
480 }
481 
482 static void zram_meta_free(struct zram_meta *meta, u64 disksize)
483 {
484 	size_t num_pages = disksize >> PAGE_SHIFT;
485 	size_t index;
486 
487 	/* Free all pages that are still in this zram device */
488 	for (index = 0; index < num_pages; index++) {
489 		unsigned long handle = meta->table[index].handle;
490 
491 		if (!handle)
492 			continue;
493 
494 		zs_free(meta->mem_pool, handle);
495 	}
496 
497 	zs_destroy_pool(meta->mem_pool);
498 	vfree(meta->table);
499 	kfree(meta);
500 }
501 
502 static struct zram_meta *zram_meta_alloc(char *pool_name, u64 disksize)
503 {
504 	size_t num_pages;
505 	struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
506 
507 	if (!meta)
508 		return NULL;
509 
510 	num_pages = disksize >> PAGE_SHIFT;
511 	meta->table = vzalloc(num_pages * sizeof(*meta->table));
512 	if (!meta->table) {
513 		pr_err("Error allocating zram address table\n");
514 		goto out_error;
515 	}
516 
517 	meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM);
518 	if (!meta->mem_pool) {
519 		pr_err("Error creating memory pool\n");
520 		goto out_error;
521 	}
522 
523 	return meta;
524 
525 out_error:
526 	vfree(meta->table);
527 	kfree(meta);
528 	return NULL;
529 }
530 
531 /*
532  * To protect concurrent access to the same index entry,
533  * caller should hold this table index entry's bit_spinlock to
534  * indicate this index entry is accessing.
535  */
536 static void zram_free_page(struct zram *zram, size_t index)
537 {
538 	struct zram_meta *meta = zram->meta;
539 	unsigned long handle = meta->table[index].handle;
540 
541 	if (unlikely(!handle)) {
542 		/*
543 		 * No memory is allocated for zero filled pages.
544 		 * Simply clear zero page flag.
545 		 */
546 		if (zram_test_flag(meta, index, ZRAM_ZERO)) {
547 			zram_clear_flag(meta, index, ZRAM_ZERO);
548 			atomic64_dec(&zram->stats.zero_pages);
549 		}
550 		return;
551 	}
552 
553 	zs_free(meta->mem_pool, handle);
554 
555 	atomic64_sub(zram_get_obj_size(meta, index),
556 			&zram->stats.compr_data_size);
557 	atomic64_dec(&zram->stats.pages_stored);
558 
559 	meta->table[index].handle = 0;
560 	zram_set_obj_size(meta, index, 0);
561 }
562 
563 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
564 {
565 	int ret = 0;
566 	unsigned char *cmem;
567 	struct zram_meta *meta = zram->meta;
568 	unsigned long handle;
569 	size_t size;
570 
571 	bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
572 	handle = meta->table[index].handle;
573 	size = zram_get_obj_size(meta, index);
574 
575 	if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
576 		bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
577 		clear_page(mem);
578 		return 0;
579 	}
580 
581 	cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
582 	if (size == PAGE_SIZE)
583 		copy_page(mem, cmem);
584 	else
585 		ret = zcomp_decompress(zram->comp, cmem, size, mem);
586 	zs_unmap_object(meta->mem_pool, handle);
587 	bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
588 
589 	/* Should NEVER happen. Return bio error if it does. */
590 	if (unlikely(ret)) {
591 		pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
592 		return ret;
593 	}
594 
595 	return 0;
596 }
597 
598 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
599 			  u32 index, int offset)
600 {
601 	int ret;
602 	struct page *page;
603 	unsigned char *user_mem, *uncmem = NULL;
604 	struct zram_meta *meta = zram->meta;
605 	page = bvec->bv_page;
606 
607 	bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
608 	if (unlikely(!meta->table[index].handle) ||
609 			zram_test_flag(meta, index, ZRAM_ZERO)) {
610 		bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
611 		handle_zero_page(bvec);
612 		return 0;
613 	}
614 	bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
615 
616 	if (is_partial_io(bvec))
617 		/* Use  a temporary buffer to decompress the page */
618 		uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
619 
620 	user_mem = kmap_atomic(page);
621 	if (!is_partial_io(bvec))
622 		uncmem = user_mem;
623 
624 	if (!uncmem) {
625 		pr_err("Unable to allocate temp memory\n");
626 		ret = -ENOMEM;
627 		goto out_cleanup;
628 	}
629 
630 	ret = zram_decompress_page(zram, uncmem, index);
631 	/* Should NEVER happen. Return bio error if it does. */
632 	if (unlikely(ret))
633 		goto out_cleanup;
634 
635 	if (is_partial_io(bvec))
636 		memcpy(user_mem + bvec->bv_offset, uncmem + offset,
637 				bvec->bv_len);
638 
639 	flush_dcache_page(page);
640 	ret = 0;
641 out_cleanup:
642 	kunmap_atomic(user_mem);
643 	if (is_partial_io(bvec))
644 		kfree(uncmem);
645 	return ret;
646 }
647 
648 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
649 			   int offset)
650 {
651 	int ret = 0;
652 	size_t clen;
653 	unsigned long handle;
654 	struct page *page;
655 	unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
656 	struct zram_meta *meta = zram->meta;
657 	struct zcomp_strm *zstrm = NULL;
658 	unsigned long alloced_pages;
659 
660 	page = bvec->bv_page;
661 	if (is_partial_io(bvec)) {
662 		/*
663 		 * This is a partial IO. We need to read the full page
664 		 * before to write the changes.
665 		 */
666 		uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
667 		if (!uncmem) {
668 			ret = -ENOMEM;
669 			goto out;
670 		}
671 		ret = zram_decompress_page(zram, uncmem, index);
672 		if (ret)
673 			goto out;
674 	}
675 
676 	zstrm = zcomp_strm_find(zram->comp);
677 	user_mem = kmap_atomic(page);
678 
679 	if (is_partial_io(bvec)) {
680 		memcpy(uncmem + offset, user_mem + bvec->bv_offset,
681 		       bvec->bv_len);
682 		kunmap_atomic(user_mem);
683 		user_mem = NULL;
684 	} else {
685 		uncmem = user_mem;
686 	}
687 
688 	if (page_zero_filled(uncmem)) {
689 		if (user_mem)
690 			kunmap_atomic(user_mem);
691 		/* Free memory associated with this sector now. */
692 		bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
693 		zram_free_page(zram, index);
694 		zram_set_flag(meta, index, ZRAM_ZERO);
695 		bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
696 
697 		atomic64_inc(&zram->stats.zero_pages);
698 		ret = 0;
699 		goto out;
700 	}
701 
702 	ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
703 	if (!is_partial_io(bvec)) {
704 		kunmap_atomic(user_mem);
705 		user_mem = NULL;
706 		uncmem = NULL;
707 	}
708 
709 	if (unlikely(ret)) {
710 		pr_err("Compression failed! err=%d\n", ret);
711 		goto out;
712 	}
713 	src = zstrm->buffer;
714 	if (unlikely(clen > max_zpage_size)) {
715 		clen = PAGE_SIZE;
716 		if (is_partial_io(bvec))
717 			src = uncmem;
718 	}
719 
720 	handle = zs_malloc(meta->mem_pool, clen);
721 	if (!handle) {
722 		pr_err("Error allocating memory for compressed page: %u, size=%zu\n",
723 			index, clen);
724 		ret = -ENOMEM;
725 		goto out;
726 	}
727 
728 	alloced_pages = zs_get_total_pages(meta->mem_pool);
729 	if (zram->limit_pages && alloced_pages > zram->limit_pages) {
730 		zs_free(meta->mem_pool, handle);
731 		ret = -ENOMEM;
732 		goto out;
733 	}
734 
735 	update_used_max(zram, alloced_pages);
736 
737 	cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
738 
739 	if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
740 		src = kmap_atomic(page);
741 		copy_page(cmem, src);
742 		kunmap_atomic(src);
743 	} else {
744 		memcpy(cmem, src, clen);
745 	}
746 
747 	zcomp_strm_release(zram->comp, zstrm);
748 	zstrm = NULL;
749 	zs_unmap_object(meta->mem_pool, handle);
750 
751 	/*
752 	 * Free memory associated with this sector
753 	 * before overwriting unused sectors.
754 	 */
755 	bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
756 	zram_free_page(zram, index);
757 
758 	meta->table[index].handle = handle;
759 	zram_set_obj_size(meta, index, clen);
760 	bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
761 
762 	/* Update stats */
763 	atomic64_add(clen, &zram->stats.compr_data_size);
764 	atomic64_inc(&zram->stats.pages_stored);
765 out:
766 	if (zstrm)
767 		zcomp_strm_release(zram->comp, zstrm);
768 	if (is_partial_io(bvec))
769 		kfree(uncmem);
770 	return ret;
771 }
772 
773 /*
774  * zram_bio_discard - handler on discard request
775  * @index: physical block index in PAGE_SIZE units
776  * @offset: byte offset within physical block
777  */
778 static void zram_bio_discard(struct zram *zram, u32 index,
779 			     int offset, struct bio *bio)
780 {
781 	size_t n = bio->bi_iter.bi_size;
782 	struct zram_meta *meta = zram->meta;
783 
784 	/*
785 	 * zram manages data in physical block size units. Because logical block
786 	 * size isn't identical with physical block size on some arch, we
787 	 * could get a discard request pointing to a specific offset within a
788 	 * certain physical block.  Although we can handle this request by
789 	 * reading that physiclal block and decompressing and partially zeroing
790 	 * and re-compressing and then re-storing it, this isn't reasonable
791 	 * because our intent with a discard request is to save memory.  So
792 	 * skipping this logical block is appropriate here.
793 	 */
794 	if (offset) {
795 		if (n <= (PAGE_SIZE - offset))
796 			return;
797 
798 		n -= (PAGE_SIZE - offset);
799 		index++;
800 	}
801 
802 	while (n >= PAGE_SIZE) {
803 		bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
804 		zram_free_page(zram, index);
805 		bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
806 		atomic64_inc(&zram->stats.notify_free);
807 		index++;
808 		n -= PAGE_SIZE;
809 	}
810 }
811 
812 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
813 			int offset, int rw)
814 {
815 	unsigned long start_time = jiffies;
816 	int ret;
817 
818 	generic_start_io_acct(rw, bvec->bv_len >> SECTOR_SHIFT,
819 			&zram->disk->part0);
820 
821 	if (rw == READ) {
822 		atomic64_inc(&zram->stats.num_reads);
823 		ret = zram_bvec_read(zram, bvec, index, offset);
824 	} else {
825 		atomic64_inc(&zram->stats.num_writes);
826 		ret = zram_bvec_write(zram, bvec, index, offset);
827 	}
828 
829 	generic_end_io_acct(rw, &zram->disk->part0, start_time);
830 
831 	if (unlikely(ret)) {
832 		if (rw == READ)
833 			atomic64_inc(&zram->stats.failed_reads);
834 		else
835 			atomic64_inc(&zram->stats.failed_writes);
836 	}
837 
838 	return ret;
839 }
840 
841 static void __zram_make_request(struct zram *zram, struct bio *bio)
842 {
843 	int offset, rw;
844 	u32 index;
845 	struct bio_vec bvec;
846 	struct bvec_iter iter;
847 
848 	index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
849 	offset = (bio->bi_iter.bi_sector &
850 		  (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
851 
852 	if (unlikely(bio->bi_rw & REQ_DISCARD)) {
853 		zram_bio_discard(zram, index, offset, bio);
854 		bio_endio(bio);
855 		return;
856 	}
857 
858 	rw = bio_data_dir(bio);
859 	bio_for_each_segment(bvec, bio, iter) {
860 		int max_transfer_size = PAGE_SIZE - offset;
861 
862 		if (bvec.bv_len > max_transfer_size) {
863 			/*
864 			 * zram_bvec_rw() can only make operation on a single
865 			 * zram page. Split the bio vector.
866 			 */
867 			struct bio_vec bv;
868 
869 			bv.bv_page = bvec.bv_page;
870 			bv.bv_len = max_transfer_size;
871 			bv.bv_offset = bvec.bv_offset;
872 
873 			if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0)
874 				goto out;
875 
876 			bv.bv_len = bvec.bv_len - max_transfer_size;
877 			bv.bv_offset += max_transfer_size;
878 			if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0)
879 				goto out;
880 		} else
881 			if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0)
882 				goto out;
883 
884 		update_position(&index, &offset, &bvec);
885 	}
886 
887 	bio_endio(bio);
888 	return;
889 
890 out:
891 	bio_io_error(bio);
892 }
893 
894 /*
895  * Handler function for all zram I/O requests.
896  */
897 static void zram_make_request(struct request_queue *queue, struct bio *bio)
898 {
899 	struct zram *zram = queue->queuedata;
900 
901 	if (unlikely(!zram_meta_get(zram)))
902 		goto error;
903 
904 	blk_queue_split(queue, &bio, queue->bio_split);
905 
906 	if (!valid_io_request(zram, bio->bi_iter.bi_sector,
907 					bio->bi_iter.bi_size)) {
908 		atomic64_inc(&zram->stats.invalid_io);
909 		goto put_zram;
910 	}
911 
912 	__zram_make_request(zram, bio);
913 	zram_meta_put(zram);
914 	return;
915 put_zram:
916 	zram_meta_put(zram);
917 error:
918 	bio_io_error(bio);
919 }
920 
921 static void zram_slot_free_notify(struct block_device *bdev,
922 				unsigned long index)
923 {
924 	struct zram *zram;
925 	struct zram_meta *meta;
926 
927 	zram = bdev->bd_disk->private_data;
928 	meta = zram->meta;
929 
930 	bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
931 	zram_free_page(zram, index);
932 	bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
933 	atomic64_inc(&zram->stats.notify_free);
934 }
935 
936 static int zram_rw_page(struct block_device *bdev, sector_t sector,
937 		       struct page *page, int rw)
938 {
939 	int offset, err = -EIO;
940 	u32 index;
941 	struct zram *zram;
942 	struct bio_vec bv;
943 
944 	zram = bdev->bd_disk->private_data;
945 	if (unlikely(!zram_meta_get(zram)))
946 		goto out;
947 
948 	if (!valid_io_request(zram, sector, PAGE_SIZE)) {
949 		atomic64_inc(&zram->stats.invalid_io);
950 		err = -EINVAL;
951 		goto put_zram;
952 	}
953 
954 	index = sector >> SECTORS_PER_PAGE_SHIFT;
955 	offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
956 
957 	bv.bv_page = page;
958 	bv.bv_len = PAGE_SIZE;
959 	bv.bv_offset = 0;
960 
961 	err = zram_bvec_rw(zram, &bv, index, offset, rw);
962 put_zram:
963 	zram_meta_put(zram);
964 out:
965 	/*
966 	 * If I/O fails, just return error(ie, non-zero) without
967 	 * calling page_endio.
968 	 * It causes resubmit the I/O with bio request by upper functions
969 	 * of rw_page(e.g., swap_readpage, __swap_writepage) and
970 	 * bio->bi_end_io does things to handle the error
971 	 * (e.g., SetPageError, set_page_dirty and extra works).
972 	 */
973 	if (err == 0)
974 		page_endio(page, rw, 0);
975 	return err;
976 }
977 
978 static void zram_reset_device(struct zram *zram)
979 {
980 	struct zram_meta *meta;
981 	struct zcomp *comp;
982 	u64 disksize;
983 
984 	down_write(&zram->init_lock);
985 
986 	zram->limit_pages = 0;
987 
988 	if (!init_done(zram)) {
989 		up_write(&zram->init_lock);
990 		return;
991 	}
992 
993 	meta = zram->meta;
994 	comp = zram->comp;
995 	disksize = zram->disksize;
996 	/*
997 	 * Refcount will go down to 0 eventually and r/w handler
998 	 * cannot handle further I/O so it will bail out by
999 	 * check zram_meta_get.
1000 	 */
1001 	zram_meta_put(zram);
1002 	/*
1003 	 * We want to free zram_meta in process context to avoid
1004 	 * deadlock between reclaim path and any other locks.
1005 	 */
1006 	wait_event(zram->io_done, atomic_read(&zram->refcount) == 0);
1007 
1008 	/* Reset stats */
1009 	memset(&zram->stats, 0, sizeof(zram->stats));
1010 	zram->disksize = 0;
1011 	zram->max_comp_streams = 1;
1012 
1013 	set_capacity(zram->disk, 0);
1014 	part_stat_set_all(&zram->disk->part0, 0);
1015 
1016 	up_write(&zram->init_lock);
1017 	/* I/O operation under all of CPU are done so let's free */
1018 	zram_meta_free(meta, disksize);
1019 	zcomp_destroy(comp);
1020 }
1021 
1022 static ssize_t disksize_store(struct device *dev,
1023 		struct device_attribute *attr, const char *buf, size_t len)
1024 {
1025 	u64 disksize;
1026 	struct zcomp *comp;
1027 	struct zram_meta *meta;
1028 	struct zram *zram = dev_to_zram(dev);
1029 	int err;
1030 
1031 	disksize = memparse(buf, NULL);
1032 	if (!disksize)
1033 		return -EINVAL;
1034 
1035 	disksize = PAGE_ALIGN(disksize);
1036 	meta = zram_meta_alloc(zram->disk->disk_name, disksize);
1037 	if (!meta)
1038 		return -ENOMEM;
1039 
1040 	comp = zcomp_create(zram->compressor, zram->max_comp_streams);
1041 	if (IS_ERR(comp)) {
1042 		pr_err("Cannot initialise %s compressing backend\n",
1043 				zram->compressor);
1044 		err = PTR_ERR(comp);
1045 		goto out_free_meta;
1046 	}
1047 
1048 	down_write(&zram->init_lock);
1049 	if (init_done(zram)) {
1050 		pr_info("Cannot change disksize for initialized device\n");
1051 		err = -EBUSY;
1052 		goto out_destroy_comp;
1053 	}
1054 
1055 	init_waitqueue_head(&zram->io_done);
1056 	atomic_set(&zram->refcount, 1);
1057 	zram->meta = meta;
1058 	zram->comp = comp;
1059 	zram->disksize = disksize;
1060 	set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1061 	up_write(&zram->init_lock);
1062 
1063 	/*
1064 	 * Revalidate disk out of the init_lock to avoid lockdep splat.
1065 	 * It's okay because disk's capacity is protected by init_lock
1066 	 * so that revalidate_disk always sees up-to-date capacity.
1067 	 */
1068 	revalidate_disk(zram->disk);
1069 
1070 	return len;
1071 
1072 out_destroy_comp:
1073 	up_write(&zram->init_lock);
1074 	zcomp_destroy(comp);
1075 out_free_meta:
1076 	zram_meta_free(meta, disksize);
1077 	return err;
1078 }
1079 
1080 static ssize_t reset_store(struct device *dev,
1081 		struct device_attribute *attr, const char *buf, size_t len)
1082 {
1083 	int ret;
1084 	unsigned short do_reset;
1085 	struct zram *zram;
1086 	struct block_device *bdev;
1087 
1088 	ret = kstrtou16(buf, 10, &do_reset);
1089 	if (ret)
1090 		return ret;
1091 
1092 	if (!do_reset)
1093 		return -EINVAL;
1094 
1095 	zram = dev_to_zram(dev);
1096 	bdev = bdget_disk(zram->disk, 0);
1097 	if (!bdev)
1098 		return -ENOMEM;
1099 
1100 	mutex_lock(&bdev->bd_mutex);
1101 	/* Do not reset an active device or claimed device */
1102 	if (bdev->bd_openers || zram->claim) {
1103 		mutex_unlock(&bdev->bd_mutex);
1104 		bdput(bdev);
1105 		return -EBUSY;
1106 	}
1107 
1108 	/* From now on, anyone can't open /dev/zram[0-9] */
1109 	zram->claim = true;
1110 	mutex_unlock(&bdev->bd_mutex);
1111 
1112 	/* Make sure all the pending I/O are finished */
1113 	fsync_bdev(bdev);
1114 	zram_reset_device(zram);
1115 	revalidate_disk(zram->disk);
1116 	bdput(bdev);
1117 
1118 	mutex_lock(&bdev->bd_mutex);
1119 	zram->claim = false;
1120 	mutex_unlock(&bdev->bd_mutex);
1121 
1122 	return len;
1123 }
1124 
1125 static int zram_open(struct block_device *bdev, fmode_t mode)
1126 {
1127 	int ret = 0;
1128 	struct zram *zram;
1129 
1130 	WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1131 
1132 	zram = bdev->bd_disk->private_data;
1133 	/* zram was claimed to reset so open request fails */
1134 	if (zram->claim)
1135 		ret = -EBUSY;
1136 
1137 	return ret;
1138 }
1139 
1140 static const struct block_device_operations zram_devops = {
1141 	.open = zram_open,
1142 	.swap_slot_free_notify = zram_slot_free_notify,
1143 	.rw_page = zram_rw_page,
1144 	.owner = THIS_MODULE
1145 };
1146 
1147 static DEVICE_ATTR_WO(compact);
1148 static DEVICE_ATTR_RW(disksize);
1149 static DEVICE_ATTR_RO(initstate);
1150 static DEVICE_ATTR_WO(reset);
1151 static DEVICE_ATTR_RO(orig_data_size);
1152 static DEVICE_ATTR_RO(mem_used_total);
1153 static DEVICE_ATTR_RW(mem_limit);
1154 static DEVICE_ATTR_RW(mem_used_max);
1155 static DEVICE_ATTR_RW(max_comp_streams);
1156 static DEVICE_ATTR_RW(comp_algorithm);
1157 
1158 static struct attribute *zram_disk_attrs[] = {
1159 	&dev_attr_disksize.attr,
1160 	&dev_attr_initstate.attr,
1161 	&dev_attr_reset.attr,
1162 	&dev_attr_num_reads.attr,
1163 	&dev_attr_num_writes.attr,
1164 	&dev_attr_failed_reads.attr,
1165 	&dev_attr_failed_writes.attr,
1166 	&dev_attr_compact.attr,
1167 	&dev_attr_invalid_io.attr,
1168 	&dev_attr_notify_free.attr,
1169 	&dev_attr_zero_pages.attr,
1170 	&dev_attr_orig_data_size.attr,
1171 	&dev_attr_compr_data_size.attr,
1172 	&dev_attr_mem_used_total.attr,
1173 	&dev_attr_mem_limit.attr,
1174 	&dev_attr_mem_used_max.attr,
1175 	&dev_attr_max_comp_streams.attr,
1176 	&dev_attr_comp_algorithm.attr,
1177 	&dev_attr_io_stat.attr,
1178 	&dev_attr_mm_stat.attr,
1179 	NULL,
1180 };
1181 
1182 static struct attribute_group zram_disk_attr_group = {
1183 	.attrs = zram_disk_attrs,
1184 };
1185 
1186 /*
1187  * Allocate and initialize new zram device. the function returns
1188  * '>= 0' device_id upon success, and negative value otherwise.
1189  */
1190 static int zram_add(void)
1191 {
1192 	struct zram *zram;
1193 	struct request_queue *queue;
1194 	int ret, device_id;
1195 
1196 	zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1197 	if (!zram)
1198 		return -ENOMEM;
1199 
1200 	ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1201 	if (ret < 0)
1202 		goto out_free_dev;
1203 	device_id = ret;
1204 
1205 	init_rwsem(&zram->init_lock);
1206 
1207 	queue = blk_alloc_queue(GFP_KERNEL);
1208 	if (!queue) {
1209 		pr_err("Error allocating disk queue for device %d\n",
1210 			device_id);
1211 		ret = -ENOMEM;
1212 		goto out_free_idr;
1213 	}
1214 
1215 	blk_queue_make_request(queue, zram_make_request);
1216 
1217 	/* gendisk structure */
1218 	zram->disk = alloc_disk(1);
1219 	if (!zram->disk) {
1220 		pr_err("Error allocating disk structure for device %d\n",
1221 			device_id);
1222 		ret = -ENOMEM;
1223 		goto out_free_queue;
1224 	}
1225 
1226 	zram->disk->major = zram_major;
1227 	zram->disk->first_minor = device_id;
1228 	zram->disk->fops = &zram_devops;
1229 	zram->disk->queue = queue;
1230 	zram->disk->queue->queuedata = zram;
1231 	zram->disk->private_data = zram;
1232 	snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1233 
1234 	/* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1235 	set_capacity(zram->disk, 0);
1236 	/* zram devices sort of resembles non-rotational disks */
1237 	queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1238 	queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1239 	/*
1240 	 * To ensure that we always get PAGE_SIZE aligned
1241 	 * and n*PAGE_SIZED sized I/O requests.
1242 	 */
1243 	blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1244 	blk_queue_logical_block_size(zram->disk->queue,
1245 					ZRAM_LOGICAL_BLOCK_SIZE);
1246 	blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1247 	blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1248 	zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1249 	blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1250 	/*
1251 	 * zram_bio_discard() will clear all logical blocks if logical block
1252 	 * size is identical with physical block size(PAGE_SIZE). But if it is
1253 	 * different, we will skip discarding some parts of logical blocks in
1254 	 * the part of the request range which isn't aligned to physical block
1255 	 * size.  So we can't ensure that all discarded logical blocks are
1256 	 * zeroed.
1257 	 */
1258 	if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1259 		zram->disk->queue->limits.discard_zeroes_data = 1;
1260 	else
1261 		zram->disk->queue->limits.discard_zeroes_data = 0;
1262 	queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1263 
1264 	add_disk(zram->disk);
1265 
1266 	ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1267 				&zram_disk_attr_group);
1268 	if (ret < 0) {
1269 		pr_err("Error creating sysfs group for device %d\n",
1270 				device_id);
1271 		goto out_free_disk;
1272 	}
1273 	strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1274 	zram->meta = NULL;
1275 	zram->max_comp_streams = 1;
1276 
1277 	pr_info("Added device: %s\n", zram->disk->disk_name);
1278 	return device_id;
1279 
1280 out_free_disk:
1281 	del_gendisk(zram->disk);
1282 	put_disk(zram->disk);
1283 out_free_queue:
1284 	blk_cleanup_queue(queue);
1285 out_free_idr:
1286 	idr_remove(&zram_index_idr, device_id);
1287 out_free_dev:
1288 	kfree(zram);
1289 	return ret;
1290 }
1291 
1292 static int zram_remove(struct zram *zram)
1293 {
1294 	struct block_device *bdev;
1295 
1296 	bdev = bdget_disk(zram->disk, 0);
1297 	if (!bdev)
1298 		return -ENOMEM;
1299 
1300 	mutex_lock(&bdev->bd_mutex);
1301 	if (bdev->bd_openers || zram->claim) {
1302 		mutex_unlock(&bdev->bd_mutex);
1303 		bdput(bdev);
1304 		return -EBUSY;
1305 	}
1306 
1307 	zram->claim = true;
1308 	mutex_unlock(&bdev->bd_mutex);
1309 
1310 	/*
1311 	 * Remove sysfs first, so no one will perform a disksize
1312 	 * store while we destroy the devices. This also helps during
1313 	 * hot_remove -- zram_reset_device() is the last holder of
1314 	 * ->init_lock, no later/concurrent disksize_store() or any
1315 	 * other sysfs handlers are possible.
1316 	 */
1317 	sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1318 			&zram_disk_attr_group);
1319 
1320 	/* Make sure all the pending I/O are finished */
1321 	fsync_bdev(bdev);
1322 	zram_reset_device(zram);
1323 	bdput(bdev);
1324 
1325 	pr_info("Removed device: %s\n", zram->disk->disk_name);
1326 
1327 	idr_remove(&zram_index_idr, zram->disk->first_minor);
1328 	blk_cleanup_queue(zram->disk->queue);
1329 	del_gendisk(zram->disk);
1330 	put_disk(zram->disk);
1331 	kfree(zram);
1332 	return 0;
1333 }
1334 
1335 /* zram-control sysfs attributes */
1336 static ssize_t hot_add_show(struct class *class,
1337 			struct class_attribute *attr,
1338 			char *buf)
1339 {
1340 	int ret;
1341 
1342 	mutex_lock(&zram_index_mutex);
1343 	ret = zram_add();
1344 	mutex_unlock(&zram_index_mutex);
1345 
1346 	if (ret < 0)
1347 		return ret;
1348 	return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1349 }
1350 
1351 static ssize_t hot_remove_store(struct class *class,
1352 			struct class_attribute *attr,
1353 			const char *buf,
1354 			size_t count)
1355 {
1356 	struct zram *zram;
1357 	int ret, dev_id;
1358 
1359 	/* dev_id is gendisk->first_minor, which is `int' */
1360 	ret = kstrtoint(buf, 10, &dev_id);
1361 	if (ret)
1362 		return ret;
1363 	if (dev_id < 0)
1364 		return -EINVAL;
1365 
1366 	mutex_lock(&zram_index_mutex);
1367 
1368 	zram = idr_find(&zram_index_idr, dev_id);
1369 	if (zram)
1370 		ret = zram_remove(zram);
1371 	else
1372 		ret = -ENODEV;
1373 
1374 	mutex_unlock(&zram_index_mutex);
1375 	return ret ? ret : count;
1376 }
1377 
1378 static struct class_attribute zram_control_class_attrs[] = {
1379 	__ATTR_RO(hot_add),
1380 	__ATTR_WO(hot_remove),
1381 	__ATTR_NULL,
1382 };
1383 
1384 static struct class zram_control_class = {
1385 	.name		= "zram-control",
1386 	.owner		= THIS_MODULE,
1387 	.class_attrs	= zram_control_class_attrs,
1388 };
1389 
1390 static int zram_remove_cb(int id, void *ptr, void *data)
1391 {
1392 	zram_remove(ptr);
1393 	return 0;
1394 }
1395 
1396 static void destroy_devices(void)
1397 {
1398 	class_unregister(&zram_control_class);
1399 	idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1400 	idr_destroy(&zram_index_idr);
1401 	unregister_blkdev(zram_major, "zram");
1402 }
1403 
1404 static int __init zram_init(void)
1405 {
1406 	int ret;
1407 
1408 	ret = class_register(&zram_control_class);
1409 	if (ret) {
1410 		pr_err("Unable to register zram-control class\n");
1411 		return ret;
1412 	}
1413 
1414 	zram_major = register_blkdev(0, "zram");
1415 	if (zram_major <= 0) {
1416 		pr_err("Unable to get major number\n");
1417 		class_unregister(&zram_control_class);
1418 		return -EBUSY;
1419 	}
1420 
1421 	while (num_devices != 0) {
1422 		mutex_lock(&zram_index_mutex);
1423 		ret = zram_add();
1424 		mutex_unlock(&zram_index_mutex);
1425 		if (ret < 0)
1426 			goto out_error;
1427 		num_devices--;
1428 	}
1429 
1430 	return 0;
1431 
1432 out_error:
1433 	destroy_devices();
1434 	return ret;
1435 }
1436 
1437 static void __exit zram_exit(void)
1438 {
1439 	destroy_devices();
1440 }
1441 
1442 module_init(zram_init);
1443 module_exit(zram_exit);
1444 
1445 module_param(num_devices, uint, 0);
1446 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1447 
1448 MODULE_LICENSE("Dual BSD/GPL");
1449 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1450 MODULE_DESCRIPTION("Compressed RAM Block Device");
1451