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 up_write(&zram->init_lock);
385 static ssize_t compact_store(struct device *dev,
386 struct device_attribute *attr, const char *buf, size_t len)
388 unsigned long nr_migrated;
389 struct zram *zram = dev_to_zram(dev);
390 struct zram_meta *meta;
392 down_read(&zram->init_lock);
393 if (!init_done(zram)) {
394 up_read(&zram->init_lock);
399 nr_migrated = zs_compact(meta->mem_pool);
400 atomic64_add(nr_migrated, &zram->stats.num_migrated);
401 up_read(&zram->init_lock);
406 static ssize_t io_stat_show(struct device *dev,
407 struct device_attribute *attr, char *buf)
409 struct zram *zram = dev_to_zram(dev);
412 down_read(&zram->init_lock);
413 ret = scnprintf(buf, PAGE_SIZE,
414 "%8llu %8llu %8llu %8llu\n",
415 (u64)atomic64_read(&zram->stats.failed_reads),
416 (u64)atomic64_read(&zram->stats.failed_writes),
417 (u64)atomic64_read(&zram->stats.invalid_io),
418 (u64)atomic64_read(&zram->stats.notify_free));
419 up_read(&zram->init_lock);
424 static ssize_t mm_stat_show(struct device *dev,
425 struct device_attribute *attr, char *buf)
427 struct zram *zram = dev_to_zram(dev);
428 u64 orig_size, mem_used = 0;
432 down_read(&zram->init_lock);
434 mem_used = zs_get_total_pages(zram->meta->mem_pool);
436 orig_size = atomic64_read(&zram->stats.pages_stored);
437 max_used = atomic_long_read(&zram->stats.max_used_pages);
439 ret = scnprintf(buf, PAGE_SIZE,
440 "%8llu %8llu %8llu %8lu %8ld %8llu %8llu\n",
441 orig_size << PAGE_SHIFT,
442 (u64)atomic64_read(&zram->stats.compr_data_size),
443 mem_used << PAGE_SHIFT,
444 zram->limit_pages << PAGE_SHIFT,
445 max_used << PAGE_SHIFT,
446 (u64)atomic64_read(&zram->stats.zero_pages),
447 (u64)atomic64_read(&zram->stats.num_migrated));
448 up_read(&zram->init_lock);
453 static DEVICE_ATTR_RO(io_stat);
454 static DEVICE_ATTR_RO(mm_stat);
455 ZRAM_ATTR_RO(num_reads);
456 ZRAM_ATTR_RO(num_writes);
457 ZRAM_ATTR_RO(failed_reads);
458 ZRAM_ATTR_RO(failed_writes);
459 ZRAM_ATTR_RO(invalid_io);
460 ZRAM_ATTR_RO(notify_free);
461 ZRAM_ATTR_RO(zero_pages);
462 ZRAM_ATTR_RO(compr_data_size);
464 static inline bool zram_meta_get(struct zram *zram)
466 if (atomic_inc_not_zero(&zram->refcount))
471 static inline void zram_meta_put(struct zram *zram)
473 atomic_dec(&zram->refcount);
476 static void zram_meta_free(struct zram_meta *meta, u64 disksize)
478 size_t num_pages = disksize >> PAGE_SHIFT;
481 /* Free all pages that are still in this zram device */
482 for (index = 0; index < num_pages; index++) {
483 unsigned long handle = meta->table[index].handle;
488 zs_free(meta->mem_pool, handle);
491 zs_destroy_pool(meta->mem_pool);
496 static struct zram_meta *zram_meta_alloc(int device_id, u64 disksize)
500 struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
505 num_pages = disksize >> PAGE_SHIFT;
506 meta->table = vzalloc(num_pages * sizeof(*meta->table));
508 pr_err("Error allocating zram address table\n");
512 snprintf(pool_name, sizeof(pool_name), "zram%d", device_id);
513 meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM);
514 if (!meta->mem_pool) {
515 pr_err("Error creating memory pool\n");
528 * To protect concurrent access to the same index entry,
529 * caller should hold this table index entry's bit_spinlock to
530 * indicate this index entry is accessing.
532 static void zram_free_page(struct zram *zram, size_t index)
534 struct zram_meta *meta = zram->meta;
535 unsigned long handle = meta->table[index].handle;
537 if (unlikely(!handle)) {
539 * No memory is allocated for zero filled pages.
540 * Simply clear zero page flag.
542 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
543 zram_clear_flag(meta, index, ZRAM_ZERO);
544 atomic64_dec(&zram->stats.zero_pages);
549 zs_free(meta->mem_pool, handle);
551 atomic64_sub(zram_get_obj_size(meta, index),
552 &zram->stats.compr_data_size);
553 atomic64_dec(&zram->stats.pages_stored);
555 meta->table[index].handle = 0;
556 zram_set_obj_size(meta, index, 0);
559 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
563 struct zram_meta *meta = zram->meta;
564 unsigned long handle;
567 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
568 handle = meta->table[index].handle;
569 size = zram_get_obj_size(meta, index);
571 if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
572 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
577 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
578 if (size == PAGE_SIZE)
579 copy_page(mem, cmem);
581 ret = zcomp_decompress(zram->comp, cmem, size, mem);
582 zs_unmap_object(meta->mem_pool, handle);
583 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
585 /* Should NEVER happen. Return bio error if it does. */
587 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
594 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
595 u32 index, int offset)
599 unsigned char *user_mem, *uncmem = NULL;
600 struct zram_meta *meta = zram->meta;
601 page = bvec->bv_page;
603 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
604 if (unlikely(!meta->table[index].handle) ||
605 zram_test_flag(meta, index, ZRAM_ZERO)) {
606 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
607 handle_zero_page(bvec);
610 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
612 if (is_partial_io(bvec))
613 /* Use a temporary buffer to decompress the page */
614 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
616 user_mem = kmap_atomic(page);
617 if (!is_partial_io(bvec))
621 pr_info("Unable to allocate temp memory\n");
626 ret = zram_decompress_page(zram, uncmem, index);
627 /* Should NEVER happen. Return bio error if it does. */
631 if (is_partial_io(bvec))
632 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
635 flush_dcache_page(page);
638 kunmap_atomic(user_mem);
639 if (is_partial_io(bvec))
644 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
649 unsigned long handle;
651 unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
652 struct zram_meta *meta = zram->meta;
653 struct zcomp_strm *zstrm = NULL;
654 unsigned long alloced_pages;
656 page = bvec->bv_page;
657 if (is_partial_io(bvec)) {
659 * This is a partial IO. We need to read the full page
660 * before to write the changes.
662 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
667 ret = zram_decompress_page(zram, uncmem, index);
672 zstrm = zcomp_strm_find(zram->comp);
673 user_mem = kmap_atomic(page);
675 if (is_partial_io(bvec)) {
676 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
678 kunmap_atomic(user_mem);
684 if (page_zero_filled(uncmem)) {
686 kunmap_atomic(user_mem);
687 /* Free memory associated with this sector now. */
688 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
689 zram_free_page(zram, index);
690 zram_set_flag(meta, index, ZRAM_ZERO);
691 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
693 atomic64_inc(&zram->stats.zero_pages);
698 ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
699 if (!is_partial_io(bvec)) {
700 kunmap_atomic(user_mem);
706 pr_err("Compression failed! err=%d\n", ret);
710 if (unlikely(clen > max_zpage_size)) {
712 if (is_partial_io(bvec))
716 handle = zs_malloc(meta->mem_pool, clen);
718 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
724 alloced_pages = zs_get_total_pages(meta->mem_pool);
725 if (zram->limit_pages && alloced_pages > zram->limit_pages) {
726 zs_free(meta->mem_pool, handle);
731 update_used_max(zram, alloced_pages);
733 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
735 if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
736 src = kmap_atomic(page);
737 copy_page(cmem, src);
740 memcpy(cmem, src, clen);
743 zcomp_strm_release(zram->comp, zstrm);
745 zs_unmap_object(meta->mem_pool, handle);
748 * Free memory associated with this sector
749 * before overwriting unused sectors.
751 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
752 zram_free_page(zram, index);
754 meta->table[index].handle = handle;
755 zram_set_obj_size(meta, index, clen);
756 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
759 atomic64_add(clen, &zram->stats.compr_data_size);
760 atomic64_inc(&zram->stats.pages_stored);
763 zcomp_strm_release(zram->comp, zstrm);
764 if (is_partial_io(bvec))
770 * zram_bio_discard - handler on discard request
771 * @index: physical block index in PAGE_SIZE units
772 * @offset: byte offset within physical block
774 static void zram_bio_discard(struct zram *zram, u32 index,
775 int offset, struct bio *bio)
777 size_t n = bio->bi_iter.bi_size;
778 struct zram_meta *meta = zram->meta;
781 * zram manages data in physical block size units. Because logical block
782 * size isn't identical with physical block size on some arch, we
783 * could get a discard request pointing to a specific offset within a
784 * certain physical block. Although we can handle this request by
785 * reading that physiclal block and decompressing and partially zeroing
786 * and re-compressing and then re-storing it, this isn't reasonable
787 * because our intent with a discard request is to save memory. So
788 * skipping this logical block is appropriate here.
791 if (n <= (PAGE_SIZE - offset))
794 n -= (PAGE_SIZE - offset);
798 while (n >= PAGE_SIZE) {
799 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
800 zram_free_page(zram, index);
801 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
802 atomic64_inc(&zram->stats.notify_free);
808 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
811 unsigned long start_time = jiffies;
814 generic_start_io_acct(rw, bvec->bv_len >> SECTOR_SHIFT,
818 atomic64_inc(&zram->stats.num_reads);
819 ret = zram_bvec_read(zram, bvec, index, offset);
821 atomic64_inc(&zram->stats.num_writes);
822 ret = zram_bvec_write(zram, bvec, index, offset);
825 generic_end_io_acct(rw, &zram->disk->part0, start_time);
829 atomic64_inc(&zram->stats.failed_reads);
831 atomic64_inc(&zram->stats.failed_writes);
837 static void __zram_make_request(struct zram *zram, struct bio *bio)
842 struct bvec_iter iter;
844 index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
845 offset = (bio->bi_iter.bi_sector &
846 (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
848 if (unlikely(bio->bi_rw & REQ_DISCARD)) {
849 zram_bio_discard(zram, index, offset, bio);
854 rw = bio_data_dir(bio);
855 bio_for_each_segment(bvec, bio, iter) {
856 int max_transfer_size = PAGE_SIZE - offset;
858 if (bvec.bv_len > max_transfer_size) {
860 * zram_bvec_rw() can only make operation on a single
861 * zram page. Split the bio vector.
865 bv.bv_page = bvec.bv_page;
866 bv.bv_len = max_transfer_size;
867 bv.bv_offset = bvec.bv_offset;
869 if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0)
872 bv.bv_len = bvec.bv_len - max_transfer_size;
873 bv.bv_offset += max_transfer_size;
874 if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0)
877 if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0)
880 update_position(&index, &offset, &bvec);
883 set_bit(BIO_UPTODATE, &bio->bi_flags);
892 * Handler function for all zram I/O requests.
894 static void zram_make_request(struct request_queue *queue, struct bio *bio)
896 struct zram *zram = queue->queuedata;
898 if (unlikely(!zram_meta_get(zram)))
901 if (!valid_io_request(zram, bio->bi_iter.bi_sector,
902 bio->bi_iter.bi_size)) {
903 atomic64_inc(&zram->stats.invalid_io);
907 __zram_make_request(zram, bio);
916 static void zram_slot_free_notify(struct block_device *bdev,
920 struct zram_meta *meta;
922 zram = bdev->bd_disk->private_data;
925 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
926 zram_free_page(zram, index);
927 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
928 atomic64_inc(&zram->stats.notify_free);
931 static int zram_rw_page(struct block_device *bdev, sector_t sector,
932 struct page *page, int rw)
934 int offset, err = -EIO;
939 zram = bdev->bd_disk->private_data;
940 if (unlikely(!zram_meta_get(zram)))
943 if (!valid_io_request(zram, sector, PAGE_SIZE)) {
944 atomic64_inc(&zram->stats.invalid_io);
949 index = sector >> SECTORS_PER_PAGE_SHIFT;
950 offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
953 bv.bv_len = PAGE_SIZE;
956 err = zram_bvec_rw(zram, &bv, index, offset, rw);
961 * If I/O fails, just return error(ie, non-zero) without
962 * calling page_endio.
963 * It causes resubmit the I/O with bio request by upper functions
964 * of rw_page(e.g., swap_readpage, __swap_writepage) and
965 * bio->bi_end_io does things to handle the error
966 * (e.g., SetPageError, set_page_dirty and extra works).
969 page_endio(page, rw, 0);
973 static void zram_reset_device(struct zram *zram)
975 struct zram_meta *meta;
979 down_write(&zram->init_lock);
981 zram->limit_pages = 0;
983 if (!init_done(zram)) {
984 up_write(&zram->init_lock);
990 disksize = zram->disksize;
992 * Refcount will go down to 0 eventually and r/w handler
993 * cannot handle further I/O so it will bail out by
994 * check zram_meta_get.
998 * We want to free zram_meta in process context to avoid
999 * deadlock between reclaim path and any other locks.
1001 wait_event(zram->io_done, atomic_read(&zram->refcount) == 0);
1004 memset(&zram->stats, 0, sizeof(zram->stats));
1006 zram->max_comp_streams = 1;
1008 set_capacity(zram->disk, 0);
1009 part_stat_set_all(&zram->disk->part0, 0);
1011 up_write(&zram->init_lock);
1012 /* I/O operation under all of CPU are done so let's free */
1013 zram_meta_free(meta, disksize);
1014 zcomp_destroy(comp);
1017 static ssize_t disksize_store(struct device *dev,
1018 struct device_attribute *attr, const char *buf, size_t len)
1022 struct zram_meta *meta;
1023 struct zram *zram = dev_to_zram(dev);
1026 disksize = memparse(buf, NULL);
1030 disksize = PAGE_ALIGN(disksize);
1031 meta = zram_meta_alloc(zram->disk->first_minor, disksize);
1035 comp = zcomp_create(zram->compressor, zram->max_comp_streams);
1037 pr_info("Cannot initialise %s compressing backend\n",
1039 err = PTR_ERR(comp);
1043 down_write(&zram->init_lock);
1044 if (init_done(zram)) {
1045 pr_info("Cannot change disksize for initialized device\n");
1047 goto out_destroy_comp;
1050 init_waitqueue_head(&zram->io_done);
1051 atomic_set(&zram->refcount, 1);
1054 zram->disksize = disksize;
1055 set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1056 up_write(&zram->init_lock);
1059 * Revalidate disk out of the init_lock to avoid lockdep splat.
1060 * It's okay because disk's capacity is protected by init_lock
1061 * so that revalidate_disk always sees up-to-date capacity.
1063 revalidate_disk(zram->disk);
1068 up_write(&zram->init_lock);
1069 zcomp_destroy(comp);
1071 zram_meta_free(meta, disksize);
1075 static ssize_t reset_store(struct device *dev,
1076 struct device_attribute *attr, const char *buf, size_t len)
1079 unsigned short do_reset;
1081 struct block_device *bdev;
1083 ret = kstrtou16(buf, 10, &do_reset);
1090 zram = dev_to_zram(dev);
1091 bdev = bdget_disk(zram->disk, 0);
1095 mutex_lock(&bdev->bd_mutex);
1096 /* Do not reset an active device or claimed device */
1097 if (bdev->bd_openers || zram->claim) {
1098 mutex_unlock(&bdev->bd_mutex);
1103 /* From now on, anyone can't open /dev/zram[0-9] */
1105 mutex_unlock(&bdev->bd_mutex);
1107 /* Make sure all the pending I/O are finished */
1109 zram_reset_device(zram);
1110 revalidate_disk(zram->disk);
1113 mutex_lock(&bdev->bd_mutex);
1114 zram->claim = false;
1115 mutex_unlock(&bdev->bd_mutex);
1120 static int zram_open(struct block_device *bdev, fmode_t mode)
1125 WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1127 zram = bdev->bd_disk->private_data;
1128 /* zram was claimed to reset so open request fails */
1135 static const struct block_device_operations zram_devops = {
1137 .swap_slot_free_notify = zram_slot_free_notify,
1138 .rw_page = zram_rw_page,
1139 .owner = THIS_MODULE
1142 static DEVICE_ATTR_WO(compact);
1143 static DEVICE_ATTR_RW(disksize);
1144 static DEVICE_ATTR_RO(initstate);
1145 static DEVICE_ATTR_WO(reset);
1146 static DEVICE_ATTR_RO(orig_data_size);
1147 static DEVICE_ATTR_RO(mem_used_total);
1148 static DEVICE_ATTR_RW(mem_limit);
1149 static DEVICE_ATTR_RW(mem_used_max);
1150 static DEVICE_ATTR_RW(max_comp_streams);
1151 static DEVICE_ATTR_RW(comp_algorithm);
1153 static struct attribute *zram_disk_attrs[] = {
1154 &dev_attr_disksize.attr,
1155 &dev_attr_initstate.attr,
1156 &dev_attr_reset.attr,
1157 &dev_attr_num_reads.attr,
1158 &dev_attr_num_writes.attr,
1159 &dev_attr_failed_reads.attr,
1160 &dev_attr_failed_writes.attr,
1161 &dev_attr_compact.attr,
1162 &dev_attr_invalid_io.attr,
1163 &dev_attr_notify_free.attr,
1164 &dev_attr_zero_pages.attr,
1165 &dev_attr_orig_data_size.attr,
1166 &dev_attr_compr_data_size.attr,
1167 &dev_attr_mem_used_total.attr,
1168 &dev_attr_mem_limit.attr,
1169 &dev_attr_mem_used_max.attr,
1170 &dev_attr_max_comp_streams.attr,
1171 &dev_attr_comp_algorithm.attr,
1172 &dev_attr_io_stat.attr,
1173 &dev_attr_mm_stat.attr,
1177 static struct attribute_group zram_disk_attr_group = {
1178 .attrs = zram_disk_attrs,
1182 * Allocate and initialize new zram device. the function returns
1183 * '>= 0' device_id upon success, and negative value otherwise.
1185 static int zram_add(void)
1188 struct request_queue *queue;
1191 zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1195 ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1200 init_rwsem(&zram->init_lock);
1202 queue = blk_alloc_queue(GFP_KERNEL);
1204 pr_err("Error allocating disk queue for device %d\n",
1210 blk_queue_make_request(queue, zram_make_request);
1212 /* gendisk structure */
1213 zram->disk = alloc_disk(1);
1215 pr_warn("Error allocating disk structure for device %d\n",
1218 goto out_free_queue;
1221 zram->disk->major = zram_major;
1222 zram->disk->first_minor = device_id;
1223 zram->disk->fops = &zram_devops;
1224 zram->disk->queue = queue;
1225 zram->disk->queue->queuedata = zram;
1226 zram->disk->private_data = zram;
1227 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1229 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1230 set_capacity(zram->disk, 0);
1231 /* zram devices sort of resembles non-rotational disks */
1232 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1233 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1235 * To ensure that we always get PAGE_SIZE aligned
1236 * and n*PAGE_SIZED sized I/O requests.
1238 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1239 blk_queue_logical_block_size(zram->disk->queue,
1240 ZRAM_LOGICAL_BLOCK_SIZE);
1241 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1242 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1243 zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1244 zram->disk->queue->limits.max_discard_sectors = UINT_MAX;
1246 * zram_bio_discard() will clear all logical blocks if logical block
1247 * size is identical with physical block size(PAGE_SIZE). But if it is
1248 * different, we will skip discarding some parts of logical blocks in
1249 * the part of the request range which isn't aligned to physical block
1250 * size. So we can't ensure that all discarded logical blocks are
1253 if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1254 zram->disk->queue->limits.discard_zeroes_data = 1;
1256 zram->disk->queue->limits.discard_zeroes_data = 0;
1257 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1259 add_disk(zram->disk);
1261 ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1262 &zram_disk_attr_group);
1264 pr_warn("Error creating sysfs group");
1267 strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1269 zram->max_comp_streams = 1;
1271 pr_info("Added device: %s\n", zram->disk->disk_name);
1275 del_gendisk(zram->disk);
1276 put_disk(zram->disk);
1278 blk_cleanup_queue(queue);
1280 idr_remove(&zram_index_idr, device_id);
1286 static int zram_remove(struct zram *zram)
1288 struct block_device *bdev;
1290 bdev = bdget_disk(zram->disk, 0);
1294 mutex_lock(&bdev->bd_mutex);
1295 if (bdev->bd_openers || zram->claim) {
1296 mutex_unlock(&bdev->bd_mutex);
1302 mutex_unlock(&bdev->bd_mutex);
1305 * Remove sysfs first, so no one will perform a disksize
1306 * store while we destroy the devices. This also helps during
1307 * hot_remove -- zram_reset_device() is the last holder of
1308 * ->init_lock, no later/concurrent disksize_store() or any
1309 * other sysfs handlers are possible.
1311 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1312 &zram_disk_attr_group);
1314 /* Make sure all the pending I/O are finished */
1316 zram_reset_device(zram);
1319 pr_info("Removed device: %s\n", zram->disk->disk_name);
1321 idr_remove(&zram_index_idr, zram->disk->first_minor);
1322 blk_cleanup_queue(zram->disk->queue);
1323 del_gendisk(zram->disk);
1324 put_disk(zram->disk);
1329 /* zram-control sysfs attributes */
1330 static ssize_t hot_add_show(struct class *class,
1331 struct class_attribute *attr,
1336 mutex_lock(&zram_index_mutex);
1338 mutex_unlock(&zram_index_mutex);
1342 return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1345 static ssize_t hot_remove_store(struct class *class,
1346 struct class_attribute *attr,
1353 /* dev_id is gendisk->first_minor, which is `int' */
1354 ret = kstrtoint(buf, 10, &dev_id);
1360 mutex_lock(&zram_index_mutex);
1362 zram = idr_find(&zram_index_idr, dev_id);
1364 ret = zram_remove(zram);
1368 mutex_unlock(&zram_index_mutex);
1369 return ret ? ret : count;
1372 static struct class_attribute zram_control_class_attrs[] = {
1374 __ATTR_WO(hot_remove),
1378 static struct class zram_control_class = {
1379 .name = "zram-control",
1380 .owner = THIS_MODULE,
1381 .class_attrs = zram_control_class_attrs,
1384 static int zram_remove_cb(int id, void *ptr, void *data)
1390 static void destroy_devices(void)
1392 class_unregister(&zram_control_class);
1393 idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1394 idr_destroy(&zram_index_idr);
1395 unregister_blkdev(zram_major, "zram");
1398 static int __init zram_init(void)
1402 ret = class_register(&zram_control_class);
1404 pr_warn("Unable to register zram-control class\n");
1408 zram_major = register_blkdev(0, "zram");
1409 if (zram_major <= 0) {
1410 pr_warn("Unable to get major number\n");
1411 class_unregister(&zram_control_class);
1415 while (num_devices != 0) {
1416 mutex_lock(&zram_index_mutex);
1418 mutex_unlock(&zram_index_mutex);
1431 static void __exit zram_exit(void)
1436 module_init(zram_init);
1437 module_exit(zram_exit);
1439 module_param(num_devices, uint, 0);
1440 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1442 MODULE_LICENSE("Dual BSD/GPL");
1443 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1444 MODULE_DESCRIPTION("Compressed RAM Block Device");