2 * Compressed RAM block device
4 * Copyright (C) 2008, 2009, 2010 Nitin Gupta
5 * 2012, 2013 Minchan Kim
7 * This code is released using a dual license strategy: BSD/GPL
8 * You can choose the licence that better fits your requirements.
10 * Released under the terms of 3-clause BSD License
11 * Released under the terms of GNU General Public License Version 2.0
15 #define KMSG_COMPONENT "zram"
16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
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>
36 static DEFINE_IDR(zram_index_idr);
37 /* idr index must be protected */
38 static DEFINE_MUTEX(zram_index_mutex);
40 static int zram_major;
41 static const char *default_compressor = "lzo";
43 /* Module params (documentation at end) */
44 static unsigned int num_devices = 1;
46 static inline void deprecated_attr_warn(const char *name)
48 pr_warn_once("%d (%s) Attribute %s (and others) will be removed. %s\n",
52 "See zram documentation.");
55 #define ZRAM_ATTR_RO(name) \
56 static ssize_t name##_show(struct device *d, \
57 struct device_attribute *attr, char *b) \
59 struct zram *zram = dev_to_zram(d); \
61 deprecated_attr_warn(__stringify(name)); \
62 return scnprintf(b, PAGE_SIZE, "%llu\n", \
63 (u64)atomic64_read(&zram->stats.name)); \
65 static DEVICE_ATTR_RO(name);
67 static inline bool init_done(struct zram *zram)
69 return zram->disksize;
72 static inline struct zram *dev_to_zram(struct device *dev)
74 return (struct zram *)dev_to_disk(dev)->private_data;
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)
81 return meta->table[index].value & BIT(flag);
84 static void zram_set_flag(struct zram_meta *meta, u32 index,
85 enum zram_pageflags flag)
87 meta->table[index].value |= BIT(flag);
90 static void zram_clear_flag(struct zram_meta *meta, u32 index,
91 enum zram_pageflags flag)
93 meta->table[index].value &= ~BIT(flag);
96 static size_t zram_get_obj_size(struct zram_meta *meta, u32 index)
98 return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
101 static void zram_set_obj_size(struct zram_meta *meta,
102 u32 index, size_t size)
104 unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT;
106 meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
109 static inline int is_partial_io(struct bio_vec *bvec)
111 return bvec->bv_len != PAGE_SIZE;
115 * Check if request is within bounds and aligned on zram logical blocks.
117 static inline int valid_io_request(struct zram *zram,
118 sector_t start, unsigned int size)
122 /* unaligned request */
123 if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
125 if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
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))
134 /* I/O request is valid */
138 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
140 if (*offset + bvec->bv_len >= PAGE_SIZE)
142 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
145 static inline void update_used_max(struct zram *zram,
146 const unsigned long pages)
148 unsigned long old_max, cur_max;
150 old_max = atomic_long_read(&zram->stats.max_used_pages);
155 old_max = atomic_long_cmpxchg(
156 &zram->stats.max_used_pages, cur_max, pages);
157 } while (old_max != cur_max);
160 static int page_zero_filled(void *ptr)
165 page = (unsigned long *)ptr;
167 for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
175 static void handle_zero_page(struct bio_vec *bvec)
177 struct page *page = bvec->bv_page;
180 user_mem = kmap_atomic(page);
181 if (is_partial_io(bvec))
182 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
184 clear_page(user_mem);
185 kunmap_atomic(user_mem);
187 flush_dcache_page(page);
190 static ssize_t initstate_show(struct device *dev,
191 struct device_attribute *attr, char *buf)
194 struct zram *zram = dev_to_zram(dev);
196 down_read(&zram->init_lock);
197 val = init_done(zram);
198 up_read(&zram->init_lock);
200 return scnprintf(buf, PAGE_SIZE, "%u\n", val);
203 static ssize_t disksize_show(struct device *dev,
204 struct device_attribute *attr, char *buf)
206 struct zram *zram = dev_to_zram(dev);
208 return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
211 static ssize_t orig_data_size_show(struct device *dev,
212 struct device_attribute *attr, char *buf)
214 struct zram *zram = dev_to_zram(dev);
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);
221 static ssize_t mem_used_total_show(struct device *dev,
222 struct device_attribute *attr, char *buf)
225 struct zram *zram = dev_to_zram(dev);
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);
233 up_read(&zram->init_lock);
235 return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
238 static ssize_t mem_limit_show(struct device *dev,
239 struct device_attribute *attr, char *buf)
242 struct zram *zram = dev_to_zram(dev);
244 deprecated_attr_warn("mem_limit");
245 down_read(&zram->init_lock);
246 val = zram->limit_pages;
247 up_read(&zram->init_lock);
249 return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
252 static ssize_t mem_limit_store(struct device *dev,
253 struct device_attribute *attr, const char *buf, size_t len)
257 struct zram *zram = dev_to_zram(dev);
259 limit = memparse(buf, &tmp);
260 if (buf == tmp) /* no chars parsed, invalid input */
263 down_write(&zram->init_lock);
264 zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
265 up_write(&zram->init_lock);
270 static ssize_t mem_used_max_show(struct device *dev,
271 struct device_attribute *attr, char *buf)
274 struct zram *zram = dev_to_zram(dev);
276 deprecated_attr_warn("mem_used_max");
277 down_read(&zram->init_lock);
279 val = atomic_long_read(&zram->stats.max_used_pages);
280 up_read(&zram->init_lock);
282 return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
285 static ssize_t mem_used_max_store(struct device *dev,
286 struct device_attribute *attr, const char *buf, size_t len)
290 struct zram *zram = dev_to_zram(dev);
292 err = kstrtoul(buf, 10, &val);
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));
302 up_read(&zram->init_lock);
307 static ssize_t max_comp_streams_show(struct device *dev,
308 struct device_attribute *attr, char *buf)
311 struct zram *zram = dev_to_zram(dev);
313 down_read(&zram->init_lock);
314 val = zram->max_comp_streams;
315 up_read(&zram->init_lock);
317 return scnprintf(buf, PAGE_SIZE, "%d\n", val);
320 static ssize_t max_comp_streams_store(struct device *dev,
321 struct device_attribute *attr, const char *buf, size_t len)
324 struct zram *zram = dev_to_zram(dev);
327 ret = kstrtoint(buf, 0, &num);
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");
342 zram->max_comp_streams = num;
345 up_write(&zram->init_lock);
349 static ssize_t comp_algorithm_show(struct device *dev,
350 struct device_attribute *attr, char *buf)
353 struct zram *zram = dev_to_zram(dev);
355 down_read(&zram->init_lock);
356 sz = zcomp_available_show(zram->compressor, buf);
357 up_read(&zram->init_lock);
362 static ssize_t comp_algorithm_store(struct device *dev,
363 struct device_attribute *attr, const char *buf, size_t len)
365 struct zram *zram = dev_to_zram(dev);
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");
374 strlcpy(zram->compressor, buf, sizeof(zram->compressor));
376 /* ignore trailing newline */
377 sz = strlen(zram->compressor);
378 if (sz > 0 && zram->compressor[sz - 1] == '\n')
379 zram->compressor[sz - 1] = 0x00;
381 if (!zcomp_available_algorithm(zram->compressor))
384 up_write(&zram->init_lock);
388 static ssize_t compact_store(struct device *dev,
389 struct device_attribute *attr, const char *buf, size_t len)
391 unsigned long nr_migrated;
392 struct zram *zram = dev_to_zram(dev);
393 struct zram_meta *meta;
395 down_read(&zram->init_lock);
396 if (!init_done(zram)) {
397 up_read(&zram->init_lock);
402 nr_migrated = zs_compact(meta->mem_pool);
403 atomic64_add(nr_migrated, &zram->stats.num_migrated);
404 up_read(&zram->init_lock);
409 static ssize_t io_stat_show(struct device *dev,
410 struct device_attribute *attr, char *buf)
412 struct zram *zram = dev_to_zram(dev);
415 down_read(&zram->init_lock);
416 ret = scnprintf(buf, PAGE_SIZE,
417 "%8llu %8llu %8llu %8llu\n",
418 (u64)atomic64_read(&zram->stats.failed_reads),
419 (u64)atomic64_read(&zram->stats.failed_writes),
420 (u64)atomic64_read(&zram->stats.invalid_io),
421 (u64)atomic64_read(&zram->stats.notify_free));
422 up_read(&zram->init_lock);
427 static ssize_t mm_stat_show(struct device *dev,
428 struct device_attribute *attr, char *buf)
430 struct zram *zram = dev_to_zram(dev);
431 u64 orig_size, mem_used = 0;
435 down_read(&zram->init_lock);
437 mem_used = zs_get_total_pages(zram->meta->mem_pool);
439 orig_size = atomic64_read(&zram->stats.pages_stored);
440 max_used = atomic_long_read(&zram->stats.max_used_pages);
442 ret = scnprintf(buf, PAGE_SIZE,
443 "%8llu %8llu %8llu %8lu %8ld %8llu %8llu\n",
444 orig_size << PAGE_SHIFT,
445 (u64)atomic64_read(&zram->stats.compr_data_size),
446 mem_used << PAGE_SHIFT,
447 zram->limit_pages << PAGE_SHIFT,
448 max_used << PAGE_SHIFT,
449 (u64)atomic64_read(&zram->stats.zero_pages),
450 (u64)atomic64_read(&zram->stats.num_migrated));
451 up_read(&zram->init_lock);
456 static DEVICE_ATTR_RO(io_stat);
457 static DEVICE_ATTR_RO(mm_stat);
458 ZRAM_ATTR_RO(num_reads);
459 ZRAM_ATTR_RO(num_writes);
460 ZRAM_ATTR_RO(failed_reads);
461 ZRAM_ATTR_RO(failed_writes);
462 ZRAM_ATTR_RO(invalid_io);
463 ZRAM_ATTR_RO(notify_free);
464 ZRAM_ATTR_RO(zero_pages);
465 ZRAM_ATTR_RO(compr_data_size);
467 static inline bool zram_meta_get(struct zram *zram)
469 if (atomic_inc_not_zero(&zram->refcount))
474 static inline void zram_meta_put(struct zram *zram)
476 atomic_dec(&zram->refcount);
479 static void zram_meta_free(struct zram_meta *meta, u64 disksize)
481 size_t num_pages = disksize >> PAGE_SHIFT;
484 /* Free all pages that are still in this zram device */
485 for (index = 0; index < num_pages; index++) {
486 unsigned long handle = meta->table[index].handle;
491 zs_free(meta->mem_pool, handle);
494 zs_destroy_pool(meta->mem_pool);
499 static struct zram_meta *zram_meta_alloc(int device_id, u64 disksize)
503 struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
508 num_pages = disksize >> PAGE_SHIFT;
509 meta->table = vzalloc(num_pages * sizeof(*meta->table));
511 pr_err("Error allocating zram address table\n");
515 snprintf(pool_name, sizeof(pool_name), "zram%d", device_id);
516 meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM);
517 if (!meta->mem_pool) {
518 pr_err("Error creating memory pool\n");
531 * To protect concurrent access to the same index entry,
532 * caller should hold this table index entry's bit_spinlock to
533 * indicate this index entry is accessing.
535 static void zram_free_page(struct zram *zram, size_t index)
537 struct zram_meta *meta = zram->meta;
538 unsigned long handle = meta->table[index].handle;
540 if (unlikely(!handle)) {
542 * No memory is allocated for zero filled pages.
543 * Simply clear zero page flag.
545 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
546 zram_clear_flag(meta, index, ZRAM_ZERO);
547 atomic64_dec(&zram->stats.zero_pages);
552 zs_free(meta->mem_pool, handle);
554 atomic64_sub(zram_get_obj_size(meta, index),
555 &zram->stats.compr_data_size);
556 atomic64_dec(&zram->stats.pages_stored);
558 meta->table[index].handle = 0;
559 zram_set_obj_size(meta, index, 0);
562 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
566 struct zram_meta *meta = zram->meta;
567 unsigned long handle;
570 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
571 handle = meta->table[index].handle;
572 size = zram_get_obj_size(meta, index);
574 if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
575 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
580 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
581 if (size == PAGE_SIZE)
582 copy_page(mem, cmem);
584 ret = zcomp_decompress(zram->comp, cmem, size, mem);
585 zs_unmap_object(meta->mem_pool, handle);
586 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
588 /* Should NEVER happen. Return bio error if it does. */
590 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
597 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
598 u32 index, int offset)
602 unsigned char *user_mem, *uncmem = NULL;
603 struct zram_meta *meta = zram->meta;
604 page = bvec->bv_page;
606 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
607 if (unlikely(!meta->table[index].handle) ||
608 zram_test_flag(meta, index, ZRAM_ZERO)) {
609 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
610 handle_zero_page(bvec);
613 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
615 if (is_partial_io(bvec))
616 /* Use a temporary buffer to decompress the page */
617 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
619 user_mem = kmap_atomic(page);
620 if (!is_partial_io(bvec))
624 pr_info("Unable to allocate temp memory\n");
629 ret = zram_decompress_page(zram, uncmem, index);
630 /* Should NEVER happen. Return bio error if it does. */
634 if (is_partial_io(bvec))
635 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
638 flush_dcache_page(page);
641 kunmap_atomic(user_mem);
642 if (is_partial_io(bvec))
647 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
652 unsigned long handle;
654 unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
655 struct zram_meta *meta = zram->meta;
656 struct zcomp_strm *zstrm = NULL;
657 unsigned long alloced_pages;
659 page = bvec->bv_page;
660 if (is_partial_io(bvec)) {
662 * This is a partial IO. We need to read the full page
663 * before to write the changes.
665 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
670 ret = zram_decompress_page(zram, uncmem, index);
675 zstrm = zcomp_strm_find(zram->comp);
676 user_mem = kmap_atomic(page);
678 if (is_partial_io(bvec)) {
679 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
681 kunmap_atomic(user_mem);
687 if (page_zero_filled(uncmem)) {
689 kunmap_atomic(user_mem);
690 /* Free memory associated with this sector now. */
691 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
692 zram_free_page(zram, index);
693 zram_set_flag(meta, index, ZRAM_ZERO);
694 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
696 atomic64_inc(&zram->stats.zero_pages);
701 ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
702 if (!is_partial_io(bvec)) {
703 kunmap_atomic(user_mem);
709 pr_err("Compression failed! err=%d\n", ret);
713 if (unlikely(clen > max_zpage_size)) {
715 if (is_partial_io(bvec))
719 handle = zs_malloc(meta->mem_pool, clen);
721 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
727 alloced_pages = zs_get_total_pages(meta->mem_pool);
728 if (zram->limit_pages && alloced_pages > zram->limit_pages) {
729 zs_free(meta->mem_pool, handle);
734 update_used_max(zram, alloced_pages);
736 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
738 if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
739 src = kmap_atomic(page);
740 copy_page(cmem, src);
743 memcpy(cmem, src, clen);
746 zcomp_strm_release(zram->comp, zstrm);
748 zs_unmap_object(meta->mem_pool, handle);
751 * Free memory associated with this sector
752 * before overwriting unused sectors.
754 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
755 zram_free_page(zram, index);
757 meta->table[index].handle = handle;
758 zram_set_obj_size(meta, index, clen);
759 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
762 atomic64_add(clen, &zram->stats.compr_data_size);
763 atomic64_inc(&zram->stats.pages_stored);
766 zcomp_strm_release(zram->comp, zstrm);
767 if (is_partial_io(bvec))
773 * zram_bio_discard - handler on discard request
774 * @index: physical block index in PAGE_SIZE units
775 * @offset: byte offset within physical block
777 static void zram_bio_discard(struct zram *zram, u32 index,
778 int offset, struct bio *bio)
780 size_t n = bio->bi_iter.bi_size;
781 struct zram_meta *meta = zram->meta;
784 * zram manages data in physical block size units. Because logical block
785 * size isn't identical with physical block size on some arch, we
786 * could get a discard request pointing to a specific offset within a
787 * certain physical block. Although we can handle this request by
788 * reading that physiclal block and decompressing and partially zeroing
789 * and re-compressing and then re-storing it, this isn't reasonable
790 * because our intent with a discard request is to save memory. So
791 * skipping this logical block is appropriate here.
794 if (n <= (PAGE_SIZE - offset))
797 n -= (PAGE_SIZE - offset);
801 while (n >= PAGE_SIZE) {
802 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
803 zram_free_page(zram, index);
804 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
805 atomic64_inc(&zram->stats.notify_free);
811 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
814 unsigned long start_time = jiffies;
817 generic_start_io_acct(rw, bvec->bv_len >> SECTOR_SHIFT,
821 atomic64_inc(&zram->stats.num_reads);
822 ret = zram_bvec_read(zram, bvec, index, offset);
824 atomic64_inc(&zram->stats.num_writes);
825 ret = zram_bvec_write(zram, bvec, index, offset);
828 generic_end_io_acct(rw, &zram->disk->part0, start_time);
832 atomic64_inc(&zram->stats.failed_reads);
834 atomic64_inc(&zram->stats.failed_writes);
840 static void __zram_make_request(struct zram *zram, struct bio *bio)
845 struct bvec_iter iter;
847 index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
848 offset = (bio->bi_iter.bi_sector &
849 (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
851 if (unlikely(bio->bi_rw & REQ_DISCARD)) {
852 zram_bio_discard(zram, index, offset, bio);
857 rw = bio_data_dir(bio);
858 bio_for_each_segment(bvec, bio, iter) {
859 int max_transfer_size = PAGE_SIZE - offset;
861 if (bvec.bv_len > max_transfer_size) {
863 * zram_bvec_rw() can only make operation on a single
864 * zram page. Split the bio vector.
868 bv.bv_page = bvec.bv_page;
869 bv.bv_len = max_transfer_size;
870 bv.bv_offset = bvec.bv_offset;
872 if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0)
875 bv.bv_len = bvec.bv_len - max_transfer_size;
876 bv.bv_offset += max_transfer_size;
877 if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0)
880 if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0)
883 update_position(&index, &offset, &bvec);
886 set_bit(BIO_UPTODATE, &bio->bi_flags);
895 * Handler function for all zram I/O requests.
897 static void zram_make_request(struct request_queue *queue, struct bio *bio)
899 struct zram *zram = queue->queuedata;
901 if (unlikely(!zram_meta_get(zram)))
904 if (!valid_io_request(zram, bio->bi_iter.bi_sector,
905 bio->bi_iter.bi_size)) {
906 atomic64_inc(&zram->stats.invalid_io);
910 __zram_make_request(zram, bio);
919 static void zram_slot_free_notify(struct block_device *bdev,
923 struct zram_meta *meta;
925 zram = bdev->bd_disk->private_data;
928 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
929 zram_free_page(zram, index);
930 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
931 atomic64_inc(&zram->stats.notify_free);
934 static int zram_rw_page(struct block_device *bdev, sector_t sector,
935 struct page *page, int rw)
937 int offset, err = -EIO;
942 zram = bdev->bd_disk->private_data;
943 if (unlikely(!zram_meta_get(zram)))
946 if (!valid_io_request(zram, sector, PAGE_SIZE)) {
947 atomic64_inc(&zram->stats.invalid_io);
952 index = sector >> SECTORS_PER_PAGE_SHIFT;
953 offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
956 bv.bv_len = PAGE_SIZE;
959 err = zram_bvec_rw(zram, &bv, index, offset, rw);
964 * If I/O fails, just return error(ie, non-zero) without
965 * calling page_endio.
966 * It causes resubmit the I/O with bio request by upper functions
967 * of rw_page(e.g., swap_readpage, __swap_writepage) and
968 * bio->bi_end_io does things to handle the error
969 * (e.g., SetPageError, set_page_dirty and extra works).
972 page_endio(page, rw, 0);
976 static void zram_reset_device(struct zram *zram)
978 struct zram_meta *meta;
982 down_write(&zram->init_lock);
984 zram->limit_pages = 0;
986 if (!init_done(zram)) {
987 up_write(&zram->init_lock);
993 disksize = zram->disksize;
995 * Refcount will go down to 0 eventually and r/w handler
996 * cannot handle further I/O so it will bail out by
997 * check zram_meta_get.
1001 * We want to free zram_meta in process context to avoid
1002 * deadlock between reclaim path and any other locks.
1004 wait_event(zram->io_done, atomic_read(&zram->refcount) == 0);
1007 memset(&zram->stats, 0, sizeof(zram->stats));
1009 zram->max_comp_streams = 1;
1011 set_capacity(zram->disk, 0);
1012 part_stat_set_all(&zram->disk->part0, 0);
1014 up_write(&zram->init_lock);
1015 /* I/O operation under all of CPU are done so let's free */
1016 zram_meta_free(meta, disksize);
1017 zcomp_destroy(comp);
1020 static ssize_t disksize_store(struct device *dev,
1021 struct device_attribute *attr, const char *buf, size_t len)
1025 struct zram_meta *meta;
1026 struct zram *zram = dev_to_zram(dev);
1029 disksize = memparse(buf, NULL);
1033 disksize = PAGE_ALIGN(disksize);
1034 meta = zram_meta_alloc(zram->disk->first_minor, disksize);
1038 comp = zcomp_create(zram->compressor, zram->max_comp_streams);
1040 pr_info("Cannot initialise %s compressing backend\n",
1042 err = PTR_ERR(comp);
1046 down_write(&zram->init_lock);
1047 if (init_done(zram)) {
1048 pr_info("Cannot change disksize for initialized device\n");
1050 goto out_destroy_comp;
1053 init_waitqueue_head(&zram->io_done);
1054 atomic_set(&zram->refcount, 1);
1057 zram->disksize = disksize;
1058 set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1059 up_write(&zram->init_lock);
1062 * Revalidate disk out of the init_lock to avoid lockdep splat.
1063 * It's okay because disk's capacity is protected by init_lock
1064 * so that revalidate_disk always sees up-to-date capacity.
1066 revalidate_disk(zram->disk);
1071 up_write(&zram->init_lock);
1072 zcomp_destroy(comp);
1074 zram_meta_free(meta, disksize);
1078 static ssize_t reset_store(struct device *dev,
1079 struct device_attribute *attr, const char *buf, size_t len)
1082 unsigned short do_reset;
1084 struct block_device *bdev;
1086 ret = kstrtou16(buf, 10, &do_reset);
1093 zram = dev_to_zram(dev);
1094 bdev = bdget_disk(zram->disk, 0);
1098 mutex_lock(&bdev->bd_mutex);
1099 /* Do not reset an active device or claimed device */
1100 if (bdev->bd_openers || zram->claim) {
1101 mutex_unlock(&bdev->bd_mutex);
1106 /* From now on, anyone can't open /dev/zram[0-9] */
1108 mutex_unlock(&bdev->bd_mutex);
1110 /* Make sure all the pending I/O are finished */
1112 zram_reset_device(zram);
1113 revalidate_disk(zram->disk);
1116 mutex_lock(&bdev->bd_mutex);
1117 zram->claim = false;
1118 mutex_unlock(&bdev->bd_mutex);
1123 static int zram_open(struct block_device *bdev, fmode_t mode)
1128 WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1130 zram = bdev->bd_disk->private_data;
1131 /* zram was claimed to reset so open request fails */
1138 static const struct block_device_operations zram_devops = {
1140 .swap_slot_free_notify = zram_slot_free_notify,
1141 .rw_page = zram_rw_page,
1142 .owner = THIS_MODULE
1145 static DEVICE_ATTR_WO(compact);
1146 static DEVICE_ATTR_RW(disksize);
1147 static DEVICE_ATTR_RO(initstate);
1148 static DEVICE_ATTR_WO(reset);
1149 static DEVICE_ATTR_RO(orig_data_size);
1150 static DEVICE_ATTR_RO(mem_used_total);
1151 static DEVICE_ATTR_RW(mem_limit);
1152 static DEVICE_ATTR_RW(mem_used_max);
1153 static DEVICE_ATTR_RW(max_comp_streams);
1154 static DEVICE_ATTR_RW(comp_algorithm);
1156 static struct attribute *zram_disk_attrs[] = {
1157 &dev_attr_disksize.attr,
1158 &dev_attr_initstate.attr,
1159 &dev_attr_reset.attr,
1160 &dev_attr_num_reads.attr,
1161 &dev_attr_num_writes.attr,
1162 &dev_attr_failed_reads.attr,
1163 &dev_attr_failed_writes.attr,
1164 &dev_attr_compact.attr,
1165 &dev_attr_invalid_io.attr,
1166 &dev_attr_notify_free.attr,
1167 &dev_attr_zero_pages.attr,
1168 &dev_attr_orig_data_size.attr,
1169 &dev_attr_compr_data_size.attr,
1170 &dev_attr_mem_used_total.attr,
1171 &dev_attr_mem_limit.attr,
1172 &dev_attr_mem_used_max.attr,
1173 &dev_attr_max_comp_streams.attr,
1174 &dev_attr_comp_algorithm.attr,
1175 &dev_attr_io_stat.attr,
1176 &dev_attr_mm_stat.attr,
1180 static struct attribute_group zram_disk_attr_group = {
1181 .attrs = zram_disk_attrs,
1185 * Allocate and initialize new zram device. the function returns
1186 * '>= 0' device_id upon success, and negative value otherwise.
1188 static int zram_add(void)
1191 struct request_queue *queue;
1194 zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1198 ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1203 init_rwsem(&zram->init_lock);
1205 queue = blk_alloc_queue(GFP_KERNEL);
1207 pr_err("Error allocating disk queue for device %d\n",
1213 blk_queue_make_request(queue, zram_make_request);
1215 /* gendisk structure */
1216 zram->disk = alloc_disk(1);
1218 pr_warn("Error allocating disk structure for device %d\n",
1221 goto out_free_queue;
1224 zram->disk->major = zram_major;
1225 zram->disk->first_minor = device_id;
1226 zram->disk->fops = &zram_devops;
1227 zram->disk->queue = queue;
1228 zram->disk->queue->queuedata = zram;
1229 zram->disk->private_data = zram;
1230 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1232 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1233 set_capacity(zram->disk, 0);
1234 /* zram devices sort of resembles non-rotational disks */
1235 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1236 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1238 * To ensure that we always get PAGE_SIZE aligned
1239 * and n*PAGE_SIZED sized I/O requests.
1241 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1242 blk_queue_logical_block_size(zram->disk->queue,
1243 ZRAM_LOGICAL_BLOCK_SIZE);
1244 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1245 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1246 zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1247 zram->disk->queue->limits.max_discard_sectors = UINT_MAX;
1249 * zram_bio_discard() will clear all logical blocks if logical block
1250 * size is identical with physical block size(PAGE_SIZE). But if it is
1251 * different, we will skip discarding some parts of logical blocks in
1252 * the part of the request range which isn't aligned to physical block
1253 * size. So we can't ensure that all discarded logical blocks are
1256 if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1257 zram->disk->queue->limits.discard_zeroes_data = 1;
1259 zram->disk->queue->limits.discard_zeroes_data = 0;
1260 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1262 add_disk(zram->disk);
1264 ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1265 &zram_disk_attr_group);
1267 pr_warn("Error creating sysfs group");
1270 strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1272 zram->max_comp_streams = 1;
1274 pr_info("Added device: %s\n", zram->disk->disk_name);
1278 del_gendisk(zram->disk);
1279 put_disk(zram->disk);
1281 blk_cleanup_queue(queue);
1283 idr_remove(&zram_index_idr, device_id);
1289 static int zram_remove(struct zram *zram)
1291 struct block_device *bdev;
1293 bdev = bdget_disk(zram->disk, 0);
1297 mutex_lock(&bdev->bd_mutex);
1298 if (bdev->bd_openers || zram->claim) {
1299 mutex_unlock(&bdev->bd_mutex);
1305 mutex_unlock(&bdev->bd_mutex);
1308 * Remove sysfs first, so no one will perform a disksize
1309 * store while we destroy the devices. This also helps during
1310 * hot_remove -- zram_reset_device() is the last holder of
1311 * ->init_lock, no later/concurrent disksize_store() or any
1312 * other sysfs handlers are possible.
1314 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1315 &zram_disk_attr_group);
1317 /* Make sure all the pending I/O are finished */
1319 zram_reset_device(zram);
1322 pr_info("Removed device: %s\n", zram->disk->disk_name);
1324 idr_remove(&zram_index_idr, zram->disk->first_minor);
1325 blk_cleanup_queue(zram->disk->queue);
1326 del_gendisk(zram->disk);
1327 put_disk(zram->disk);
1332 /* zram-control sysfs attributes */
1333 static ssize_t hot_add_show(struct class *class,
1334 struct class_attribute *attr,
1339 mutex_lock(&zram_index_mutex);
1341 mutex_unlock(&zram_index_mutex);
1345 return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1348 static ssize_t hot_remove_store(struct class *class,
1349 struct class_attribute *attr,
1356 /* dev_id is gendisk->first_minor, which is `int' */
1357 ret = kstrtoint(buf, 10, &dev_id);
1363 mutex_lock(&zram_index_mutex);
1365 zram = idr_find(&zram_index_idr, dev_id);
1367 ret = zram_remove(zram);
1371 mutex_unlock(&zram_index_mutex);
1372 return ret ? ret : count;
1375 static struct class_attribute zram_control_class_attrs[] = {
1377 __ATTR_WO(hot_remove),
1381 static struct class zram_control_class = {
1382 .name = "zram-control",
1383 .owner = THIS_MODULE,
1384 .class_attrs = zram_control_class_attrs,
1387 static int zram_remove_cb(int id, void *ptr, void *data)
1393 static void destroy_devices(void)
1395 class_unregister(&zram_control_class);
1396 idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1397 idr_destroy(&zram_index_idr);
1398 unregister_blkdev(zram_major, "zram");
1401 static int __init zram_init(void)
1405 ret = class_register(&zram_control_class);
1407 pr_warn("Unable to register zram-control class\n");
1411 zram_major = register_blkdev(0, "zram");
1412 if (zram_major <= 0) {
1413 pr_warn("Unable to get major number\n");
1414 class_unregister(&zram_control_class);
1418 while (num_devices != 0) {
1419 mutex_lock(&zram_index_mutex);
1421 mutex_unlock(&zram_index_mutex);
1434 static void __exit zram_exit(void)
1439 module_init(zram_init);
1440 module_exit(zram_exit);
1442 module_param(num_devices, uint, 0);
1443 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1445 MODULE_LICENSE("Dual BSD/GPL");
1446 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1447 MODULE_DESCRIPTION("Compressed RAM Block Device");