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/backing-dev.h>
29 #include <linux/string.h>
30 #include <linux/vmalloc.h>
31 #include <linux/err.h>
32 #include <linux/idr.h>
33 #include <linux/sysfs.h>
34 #include <linux/cpuhotplug.h>
38 static DEFINE_IDR(zram_index_idr);
39 /* idr index must be protected */
40 static DEFINE_MUTEX(zram_index_mutex);
42 static int zram_major;
43 static const char *default_compressor = "lzo";
45 /* Module params (documentation at end) */
46 static unsigned int num_devices = 1;
48 static void zram_free_page(struct zram *zram, size_t index);
50 static inline bool init_done(struct zram *zram)
52 return zram->disksize;
55 static inline struct zram *dev_to_zram(struct device *dev)
57 return (struct zram *)dev_to_disk(dev)->private_data;
60 static unsigned long zram_get_handle(struct zram *zram, u32 index)
62 return zram->table[index].handle;
65 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
67 zram->table[index].handle = handle;
70 /* flag operations require table entry bit_spin_lock() being held */
71 static int zram_test_flag(struct zram *zram, u32 index,
72 enum zram_pageflags flag)
74 return zram->table[index].value & BIT(flag);
77 static void zram_set_flag(struct zram *zram, u32 index,
78 enum zram_pageflags flag)
80 zram->table[index].value |= BIT(flag);
83 static void zram_clear_flag(struct zram *zram, u32 index,
84 enum zram_pageflags flag)
86 zram->table[index].value &= ~BIT(flag);
89 static inline void zram_set_element(struct zram *zram, u32 index,
90 unsigned long element)
92 zram->table[index].element = element;
95 static unsigned long zram_get_element(struct zram *zram, u32 index)
97 return zram->table[index].element;
100 static size_t zram_get_obj_size(struct zram *zram, u32 index)
102 return zram->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
105 static void zram_set_obj_size(struct zram *zram,
106 u32 index, size_t size)
108 unsigned long flags = zram->table[index].value >> ZRAM_FLAG_SHIFT;
110 zram->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
113 #if PAGE_SIZE != 4096
114 static inline bool is_partial_io(struct bio_vec *bvec)
116 return bvec->bv_len != PAGE_SIZE;
119 static inline bool is_partial_io(struct bio_vec *bvec)
125 static void zram_revalidate_disk(struct zram *zram)
127 revalidate_disk(zram->disk);
128 /* revalidate_disk reset the BDI_CAP_STABLE_WRITES so set again */
129 zram->disk->queue->backing_dev_info->capabilities |=
130 BDI_CAP_STABLE_WRITES;
134 * Check if request is within bounds and aligned on zram logical blocks.
136 static inline bool valid_io_request(struct zram *zram,
137 sector_t start, unsigned int size)
141 /* unaligned request */
142 if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
144 if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
147 end = start + (size >> SECTOR_SHIFT);
148 bound = zram->disksize >> SECTOR_SHIFT;
149 /* out of range range */
150 if (unlikely(start >= bound || end > bound || start > end))
153 /* I/O request is valid */
157 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
159 *index += (*offset + bvec->bv_len) / PAGE_SIZE;
160 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
163 static inline void update_used_max(struct zram *zram,
164 const unsigned long pages)
166 unsigned long old_max, cur_max;
168 old_max = atomic_long_read(&zram->stats.max_used_pages);
173 old_max = atomic_long_cmpxchg(
174 &zram->stats.max_used_pages, cur_max, pages);
175 } while (old_max != cur_max);
178 static inline void zram_fill_page(char *ptr, unsigned long len,
182 unsigned long *page = (unsigned long *)ptr;
184 WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
186 if (likely(value == 0)) {
189 for (i = 0; i < len / sizeof(*page); i++)
194 static bool page_same_filled(void *ptr, unsigned long *element)
200 page = (unsigned long *)ptr;
203 for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
204 if (val != page[pos])
213 static ssize_t initstate_show(struct device *dev,
214 struct device_attribute *attr, char *buf)
217 struct zram *zram = dev_to_zram(dev);
219 down_read(&zram->init_lock);
220 val = init_done(zram);
221 up_read(&zram->init_lock);
223 return scnprintf(buf, PAGE_SIZE, "%u\n", val);
226 static ssize_t disksize_show(struct device *dev,
227 struct device_attribute *attr, char *buf)
229 struct zram *zram = dev_to_zram(dev);
231 return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
234 static ssize_t mem_limit_store(struct device *dev,
235 struct device_attribute *attr, const char *buf, size_t len)
239 struct zram *zram = dev_to_zram(dev);
241 limit = memparse(buf, &tmp);
242 if (buf == tmp) /* no chars parsed, invalid input */
245 down_write(&zram->init_lock);
246 zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
247 up_write(&zram->init_lock);
252 static ssize_t mem_used_max_store(struct device *dev,
253 struct device_attribute *attr, const char *buf, size_t len)
257 struct zram *zram = dev_to_zram(dev);
259 err = kstrtoul(buf, 10, &val);
263 down_read(&zram->init_lock);
264 if (init_done(zram)) {
265 atomic_long_set(&zram->stats.max_used_pages,
266 zs_get_total_pages(zram->mem_pool));
268 up_read(&zram->init_lock);
273 #ifdef CONFIG_ZRAM_WRITEBACK
274 static bool zram_wb_enabled(struct zram *zram)
276 return zram->backing_dev;
279 static void reset_bdev(struct zram *zram)
281 struct block_device *bdev;
283 if (!zram_wb_enabled(zram))
287 if (zram->old_block_size)
288 set_blocksize(bdev, zram->old_block_size);
289 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
290 /* hope filp_close flush all of IO */
291 filp_close(zram->backing_dev, NULL);
292 zram->backing_dev = NULL;
293 zram->old_block_size = 0;
296 kvfree(zram->bitmap);
300 static ssize_t backing_dev_show(struct device *dev,
301 struct device_attribute *attr, char *buf)
303 struct zram *zram = dev_to_zram(dev);
304 struct file *file = zram->backing_dev;
308 down_read(&zram->init_lock);
309 if (!zram_wb_enabled(zram)) {
310 memcpy(buf, "none\n", 5);
311 up_read(&zram->init_lock);
315 p = file_path(file, buf, PAGE_SIZE - 1);
322 memmove(buf, p, ret);
325 up_read(&zram->init_lock);
329 static ssize_t backing_dev_store(struct device *dev,
330 struct device_attribute *attr, const char *buf, size_t len)
333 struct file *backing_dev = NULL;
335 struct address_space *mapping;
336 unsigned int bitmap_sz, old_block_size = 0;
337 unsigned long nr_pages, *bitmap = NULL;
338 struct block_device *bdev = NULL;
340 struct zram *zram = dev_to_zram(dev);
342 file_name = kmalloc(PATH_MAX, GFP_KERNEL);
346 down_write(&zram->init_lock);
347 if (init_done(zram)) {
348 pr_info("Can't setup backing device for initialized device\n");
353 strlcpy(file_name, buf, len);
355 backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
356 if (IS_ERR(backing_dev)) {
357 err = PTR_ERR(backing_dev);
362 mapping = backing_dev->f_mapping;
363 inode = mapping->host;
365 /* Support only block device in this moment */
366 if (!S_ISBLK(inode->i_mode)) {
371 bdev = bdgrab(I_BDEV(inode));
372 err = blkdev_get(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
376 nr_pages = i_size_read(inode) >> PAGE_SHIFT;
377 bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
378 bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
384 old_block_size = block_size(bdev);
385 err = set_blocksize(bdev, PAGE_SIZE);
390 spin_lock_init(&zram->bitmap_lock);
392 zram->old_block_size = old_block_size;
394 zram->backing_dev = backing_dev;
395 zram->bitmap = bitmap;
396 zram->nr_pages = nr_pages;
397 up_write(&zram->init_lock);
399 pr_info("setup backing device %s\n", file_name);
408 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
411 filp_close(backing_dev, NULL);
413 up_write(&zram->init_lock);
420 static unsigned long get_entry_bdev(struct zram *zram)
424 spin_lock(&zram->bitmap_lock);
425 /* skip 0 bit to confuse zram.handle = 0 */
426 entry = find_next_zero_bit(zram->bitmap, zram->nr_pages, 1);
427 if (entry == zram->nr_pages) {
428 spin_unlock(&zram->bitmap_lock);
432 set_bit(entry, zram->bitmap);
433 spin_unlock(&zram->bitmap_lock);
438 static void put_entry_bdev(struct zram *zram, unsigned long entry)
442 spin_lock(&zram->bitmap_lock);
443 was_set = test_and_clear_bit(entry, zram->bitmap);
444 spin_unlock(&zram->bitmap_lock);
445 WARN_ON_ONCE(!was_set);
448 void zram_page_end_io(struct bio *bio)
450 struct page *page = bio->bi_io_vec[0].bv_page;
452 page_endio(page, op_is_write(bio_op(bio)),
453 blk_status_to_errno(bio->bi_status));
457 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
458 u32 index, struct bio *parent,
459 unsigned long *pentry)
464 bio = bio_alloc(GFP_ATOMIC, 1);
468 entry = get_entry_bdev(zram);
474 bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
475 bio->bi_bdev = zram->bdev;
476 if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len,
479 put_entry_bdev(zram, entry);
484 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
485 bio->bi_end_io = zram_page_end_io;
487 bio->bi_opf = parent->bi_opf;
488 bio_chain(bio, parent);
497 static void zram_wb_clear(struct zram *zram, u32 index)
501 zram_clear_flag(zram, index, ZRAM_WB);
502 entry = zram_get_element(zram, index);
503 zram_set_element(zram, index, 0);
504 put_entry_bdev(zram, entry);
508 static bool zram_wb_enabled(struct zram *zram) { return false; }
509 static inline void reset_bdev(struct zram *zram) {};
510 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
511 u32 index, struct bio *parent,
512 unsigned long *pentry)
517 static void zram_wb_clear(struct zram *zram, u32 index) {}
522 * We switched to per-cpu streams and this attr is not needed anymore.
523 * However, we will keep it around for some time, because:
524 * a) we may revert per-cpu streams in the future
525 * b) it's visible to user space and we need to follow our 2 years
526 * retirement rule; but we already have a number of 'soon to be
527 * altered' attrs, so max_comp_streams need to wait for the next
530 static ssize_t max_comp_streams_show(struct device *dev,
531 struct device_attribute *attr, char *buf)
533 return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
536 static ssize_t max_comp_streams_store(struct device *dev,
537 struct device_attribute *attr, const char *buf, size_t len)
542 static ssize_t comp_algorithm_show(struct device *dev,
543 struct device_attribute *attr, char *buf)
546 struct zram *zram = dev_to_zram(dev);
548 down_read(&zram->init_lock);
549 sz = zcomp_available_show(zram->compressor, buf);
550 up_read(&zram->init_lock);
555 static ssize_t comp_algorithm_store(struct device *dev,
556 struct device_attribute *attr, const char *buf, size_t len)
558 struct zram *zram = dev_to_zram(dev);
559 char compressor[ARRAY_SIZE(zram->compressor)];
562 strlcpy(compressor, buf, sizeof(compressor));
563 /* ignore trailing newline */
564 sz = strlen(compressor);
565 if (sz > 0 && compressor[sz - 1] == '\n')
566 compressor[sz - 1] = 0x00;
568 if (!zcomp_available_algorithm(compressor))
571 down_write(&zram->init_lock);
572 if (init_done(zram)) {
573 up_write(&zram->init_lock);
574 pr_info("Can't change algorithm for initialized device\n");
578 strcpy(zram->compressor, compressor);
579 up_write(&zram->init_lock);
583 static ssize_t compact_store(struct device *dev,
584 struct device_attribute *attr, const char *buf, size_t len)
586 struct zram *zram = dev_to_zram(dev);
588 down_read(&zram->init_lock);
589 if (!init_done(zram)) {
590 up_read(&zram->init_lock);
594 zs_compact(zram->mem_pool);
595 up_read(&zram->init_lock);
600 static ssize_t io_stat_show(struct device *dev,
601 struct device_attribute *attr, char *buf)
603 struct zram *zram = dev_to_zram(dev);
606 down_read(&zram->init_lock);
607 ret = scnprintf(buf, PAGE_SIZE,
608 "%8llu %8llu %8llu %8llu\n",
609 (u64)atomic64_read(&zram->stats.failed_reads),
610 (u64)atomic64_read(&zram->stats.failed_writes),
611 (u64)atomic64_read(&zram->stats.invalid_io),
612 (u64)atomic64_read(&zram->stats.notify_free));
613 up_read(&zram->init_lock);
618 static ssize_t mm_stat_show(struct device *dev,
619 struct device_attribute *attr, char *buf)
621 struct zram *zram = dev_to_zram(dev);
622 struct zs_pool_stats pool_stats;
623 u64 orig_size, mem_used = 0;
627 memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
629 down_read(&zram->init_lock);
630 if (init_done(zram)) {
631 mem_used = zs_get_total_pages(zram->mem_pool);
632 zs_pool_stats(zram->mem_pool, &pool_stats);
635 orig_size = atomic64_read(&zram->stats.pages_stored);
636 max_used = atomic_long_read(&zram->stats.max_used_pages);
638 ret = scnprintf(buf, PAGE_SIZE,
639 "%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n",
640 orig_size << PAGE_SHIFT,
641 (u64)atomic64_read(&zram->stats.compr_data_size),
642 mem_used << PAGE_SHIFT,
643 zram->limit_pages << PAGE_SHIFT,
644 max_used << PAGE_SHIFT,
645 (u64)atomic64_read(&zram->stats.same_pages),
646 pool_stats.pages_compacted);
647 up_read(&zram->init_lock);
652 static ssize_t debug_stat_show(struct device *dev,
653 struct device_attribute *attr, char *buf)
656 struct zram *zram = dev_to_zram(dev);
659 down_read(&zram->init_lock);
660 ret = scnprintf(buf, PAGE_SIZE,
661 "version: %d\n%8llu\n",
663 (u64)atomic64_read(&zram->stats.writestall));
664 up_read(&zram->init_lock);
669 static DEVICE_ATTR_RO(io_stat);
670 static DEVICE_ATTR_RO(mm_stat);
671 static DEVICE_ATTR_RO(debug_stat);
673 static void zram_slot_lock(struct zram *zram, u32 index)
675 bit_spin_lock(ZRAM_ACCESS, &zram->table[index].value);
678 static void zram_slot_unlock(struct zram *zram, u32 index)
680 bit_spin_unlock(ZRAM_ACCESS, &zram->table[index].value);
683 static bool zram_same_page_read(struct zram *zram, u32 index,
685 unsigned int offset, unsigned int len)
687 zram_slot_lock(zram, index);
688 if (unlikely(!zram_get_handle(zram, index) ||
689 zram_test_flag(zram, index, ZRAM_SAME))) {
692 zram_slot_unlock(zram, index);
693 mem = kmap_atomic(page);
694 zram_fill_page(mem + offset, len,
695 zram_get_element(zram, index));
699 zram_slot_unlock(zram, index);
704 static void zram_meta_free(struct zram *zram, u64 disksize)
706 size_t num_pages = disksize >> PAGE_SHIFT;
709 /* Free all pages that are still in this zram device */
710 for (index = 0; index < num_pages; index++)
711 zram_free_page(zram, index);
713 zs_destroy_pool(zram->mem_pool);
717 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
721 num_pages = disksize >> PAGE_SHIFT;
722 zram->table = vzalloc(num_pages * sizeof(*zram->table));
726 zram->mem_pool = zs_create_pool(zram->disk->disk_name);
727 if (!zram->mem_pool) {
736 * To protect concurrent access to the same index entry,
737 * caller should hold this table index entry's bit_spinlock to
738 * indicate this index entry is accessing.
740 static void zram_free_page(struct zram *zram, size_t index)
742 unsigned long handle;
744 if (zram_wb_enabled(zram) && zram_test_flag(zram, index, ZRAM_WB)) {
745 zram_wb_clear(zram, index);
746 atomic64_dec(&zram->stats.pages_stored);
751 * No memory is allocated for same element filled pages.
752 * Simply clear same page flag.
754 if (zram_test_flag(zram, index, ZRAM_SAME)) {
755 zram_clear_flag(zram, index, ZRAM_SAME);
756 zram_set_element(zram, index, 0);
757 atomic64_dec(&zram->stats.same_pages);
758 atomic64_dec(&zram->stats.pages_stored);
762 handle = zram_get_handle(zram, index);
766 zs_free(zram->mem_pool, handle);
768 atomic64_sub(zram_get_obj_size(zram, index),
769 &zram->stats.compr_data_size);
770 atomic64_dec(&zram->stats.pages_stored);
772 zram_set_handle(zram, index, 0);
773 zram_set_obj_size(zram, index, 0);
776 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index)
779 unsigned long handle;
783 if (zram_same_page_read(zram, index, page, 0, PAGE_SIZE))
786 zram_slot_lock(zram, index);
787 handle = zram_get_handle(zram, index);
788 size = zram_get_obj_size(zram, index);
790 src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
791 if (size == PAGE_SIZE) {
792 dst = kmap_atomic(page);
793 memcpy(dst, src, PAGE_SIZE);
797 struct zcomp_strm *zstrm = zcomp_stream_get(zram->comp);
799 dst = kmap_atomic(page);
800 ret = zcomp_decompress(zstrm, src, size, dst);
802 zcomp_stream_put(zram->comp);
804 zs_unmap_object(zram->mem_pool, handle);
805 zram_slot_unlock(zram, index);
807 /* Should NEVER happen. Return bio error if it does. */
809 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
814 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
815 u32 index, int offset)
820 page = bvec->bv_page;
821 if (is_partial_io(bvec)) {
822 /* Use a temporary buffer to decompress the page */
823 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
828 ret = __zram_bvec_read(zram, page, index);
832 if (is_partial_io(bvec)) {
833 void *dst = kmap_atomic(bvec->bv_page);
834 void *src = kmap_atomic(page);
836 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
841 if (is_partial_io(bvec))
847 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
848 u32 index, struct bio *bio)
851 unsigned long alloced_pages;
852 unsigned long handle = 0;
853 unsigned int comp_len = 0;
854 void *src, *dst, *mem;
855 struct zcomp_strm *zstrm;
856 struct page *page = bvec->bv_page;
857 unsigned long element = 0;
858 enum zram_pageflags flags = 0;
859 bool allow_wb = true;
861 mem = kmap_atomic(page);
862 if (page_same_filled(mem, &element)) {
864 /* Free memory associated with this sector now. */
866 atomic64_inc(&zram->stats.same_pages);
872 zstrm = zcomp_stream_get(zram->comp);
873 src = kmap_atomic(page);
874 ret = zcomp_compress(zstrm, src, &comp_len);
878 zcomp_stream_put(zram->comp);
879 pr_err("Compression failed! err=%d\n", ret);
880 zs_free(zram->mem_pool, handle);
884 if (unlikely(comp_len > max_zpage_size)) {
885 if (zram_wb_enabled(zram) && allow_wb) {
886 zcomp_stream_put(zram->comp);
887 ret = write_to_bdev(zram, bvec, index, bio, &element);
896 comp_len = PAGE_SIZE;
900 * handle allocation has 2 paths:
901 * a) fast path is executed with preemption disabled (for
902 * per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
903 * since we can't sleep;
904 * b) slow path enables preemption and attempts to allocate
905 * the page with __GFP_DIRECT_RECLAIM bit set. we have to
906 * put per-cpu compression stream and, thus, to re-do
907 * the compression once handle is allocated.
909 * if we have a 'non-null' handle here then we are coming
910 * from the slow path and handle has already been allocated.
913 handle = zs_malloc(zram->mem_pool, comp_len,
914 __GFP_KSWAPD_RECLAIM |
919 zcomp_stream_put(zram->comp);
920 atomic64_inc(&zram->stats.writestall);
921 handle = zs_malloc(zram->mem_pool, comp_len,
922 GFP_NOIO | __GFP_HIGHMEM |
929 alloced_pages = zs_get_total_pages(zram->mem_pool);
930 update_used_max(zram, alloced_pages);
932 if (zram->limit_pages && alloced_pages > zram->limit_pages) {
933 zcomp_stream_put(zram->comp);
934 zs_free(zram->mem_pool, handle);
938 dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
941 if (comp_len == PAGE_SIZE)
942 src = kmap_atomic(page);
943 memcpy(dst, src, comp_len);
944 if (comp_len == PAGE_SIZE)
947 zcomp_stream_put(zram->comp);
948 zs_unmap_object(zram->mem_pool, handle);
949 atomic64_add(comp_len, &zram->stats.compr_data_size);
952 * Free memory associated with this sector
953 * before overwriting unused sectors.
955 zram_slot_lock(zram, index);
956 zram_free_page(zram, index);
959 zram_set_flag(zram, index, flags);
960 zram_set_element(zram, index, element);
962 zram_set_handle(zram, index, handle);
963 zram_set_obj_size(zram, index, comp_len);
965 zram_slot_unlock(zram, index);
968 atomic64_inc(&zram->stats.pages_stored);
972 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
973 u32 index, int offset, struct bio *bio)
976 struct page *page = NULL;
981 if (is_partial_io(bvec)) {
984 * This is a partial IO. We need to read the full page
985 * before to write the changes.
987 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
991 ret = __zram_bvec_read(zram, page, index);
995 src = kmap_atomic(bvec->bv_page);
996 dst = kmap_atomic(page);
997 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1002 vec.bv_len = PAGE_SIZE;
1006 ret = __zram_bvec_write(zram, &vec, index, bio);
1008 if (is_partial_io(bvec))
1014 * zram_bio_discard - handler on discard request
1015 * @index: physical block index in PAGE_SIZE units
1016 * @offset: byte offset within physical block
1018 static void zram_bio_discard(struct zram *zram, u32 index,
1019 int offset, struct bio *bio)
1021 size_t n = bio->bi_iter.bi_size;
1024 * zram manages data in physical block size units. Because logical block
1025 * size isn't identical with physical block size on some arch, we
1026 * could get a discard request pointing to a specific offset within a
1027 * certain physical block. Although we can handle this request by
1028 * reading that physiclal block and decompressing and partially zeroing
1029 * and re-compressing and then re-storing it, this isn't reasonable
1030 * because our intent with a discard request is to save memory. So
1031 * skipping this logical block is appropriate here.
1034 if (n <= (PAGE_SIZE - offset))
1037 n -= (PAGE_SIZE - offset);
1041 while (n >= PAGE_SIZE) {
1042 zram_slot_lock(zram, index);
1043 zram_free_page(zram, index);
1044 zram_slot_unlock(zram, index);
1045 atomic64_inc(&zram->stats.notify_free);
1052 * Returns errno if it has some problem. Otherwise return 0 or 1.
1053 * Returns 0 if IO request was done synchronously
1054 * Returns 1 if IO request was successfully submitted.
1056 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1057 int offset, bool is_write, struct bio *bio)
1059 unsigned long start_time = jiffies;
1060 int rw_acct = is_write ? REQ_OP_WRITE : REQ_OP_READ;
1063 generic_start_io_acct(rw_acct, bvec->bv_len >> SECTOR_SHIFT,
1064 &zram->disk->part0);
1067 atomic64_inc(&zram->stats.num_reads);
1068 ret = zram_bvec_read(zram, bvec, index, offset);
1069 flush_dcache_page(bvec->bv_page);
1071 atomic64_inc(&zram->stats.num_writes);
1072 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1075 generic_end_io_acct(rw_acct, &zram->disk->part0, start_time);
1077 if (unlikely(ret < 0)) {
1079 atomic64_inc(&zram->stats.failed_reads);
1081 atomic64_inc(&zram->stats.failed_writes);
1087 static void __zram_make_request(struct zram *zram, struct bio *bio)
1091 struct bio_vec bvec;
1092 struct bvec_iter iter;
1094 index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1095 offset = (bio->bi_iter.bi_sector &
1096 (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1098 switch (bio_op(bio)) {
1099 case REQ_OP_DISCARD:
1100 case REQ_OP_WRITE_ZEROES:
1101 zram_bio_discard(zram, index, offset, bio);
1108 bio_for_each_segment(bvec, bio, iter) {
1109 struct bio_vec bv = bvec;
1110 unsigned int unwritten = bvec.bv_len;
1113 bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1115 if (zram_bvec_rw(zram, &bv, index, offset,
1116 op_is_write(bio_op(bio)), bio) < 0)
1119 bv.bv_offset += bv.bv_len;
1120 unwritten -= bv.bv_len;
1122 update_position(&index, &offset, &bv);
1123 } while (unwritten);
1134 * Handler function for all zram I/O requests.
1136 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
1138 struct zram *zram = queue->queuedata;
1140 if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1141 bio->bi_iter.bi_size)) {
1142 atomic64_inc(&zram->stats.invalid_io);
1146 __zram_make_request(zram, bio);
1147 return BLK_QC_T_NONE;
1151 return BLK_QC_T_NONE;
1154 static void zram_slot_free_notify(struct block_device *bdev,
1155 unsigned long index)
1159 zram = bdev->bd_disk->private_data;
1161 zram_slot_lock(zram, index);
1162 zram_free_page(zram, index);
1163 zram_slot_unlock(zram, index);
1164 atomic64_inc(&zram->stats.notify_free);
1167 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1168 struct page *page, bool is_write)
1175 zram = bdev->bd_disk->private_data;
1177 if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1178 atomic64_inc(&zram->stats.invalid_io);
1183 index = sector >> SECTORS_PER_PAGE_SHIFT;
1184 offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1187 bv.bv_len = PAGE_SIZE;
1190 ret = zram_bvec_rw(zram, &bv, index, offset, is_write, NULL);
1193 * If I/O fails, just return error(ie, non-zero) without
1194 * calling page_endio.
1195 * It causes resubmit the I/O with bio request by upper functions
1196 * of rw_page(e.g., swap_readpage, __swap_writepage) and
1197 * bio->bi_end_io does things to handle the error
1198 * (e.g., SetPageError, set_page_dirty and extra works).
1200 if (unlikely(ret < 0))
1205 page_endio(page, is_write, 0);
1216 static void zram_reset_device(struct zram *zram)
1221 down_write(&zram->init_lock);
1223 zram->limit_pages = 0;
1225 if (!init_done(zram)) {
1226 up_write(&zram->init_lock);
1231 disksize = zram->disksize;
1234 set_capacity(zram->disk, 0);
1235 part_stat_set_all(&zram->disk->part0, 0);
1237 up_write(&zram->init_lock);
1238 /* I/O operation under all of CPU are done so let's free */
1239 zram_meta_free(zram, disksize);
1240 memset(&zram->stats, 0, sizeof(zram->stats));
1241 zcomp_destroy(comp);
1245 static ssize_t disksize_store(struct device *dev,
1246 struct device_attribute *attr, const char *buf, size_t len)
1250 struct zram *zram = dev_to_zram(dev);
1253 disksize = memparse(buf, NULL);
1257 down_write(&zram->init_lock);
1258 if (init_done(zram)) {
1259 pr_info("Cannot change disksize for initialized device\n");
1264 disksize = PAGE_ALIGN(disksize);
1265 if (!zram_meta_alloc(zram, disksize)) {
1270 comp = zcomp_create(zram->compressor);
1272 pr_err("Cannot initialise %s compressing backend\n",
1274 err = PTR_ERR(comp);
1279 zram->disksize = disksize;
1280 set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1281 zram_revalidate_disk(zram);
1282 up_write(&zram->init_lock);
1287 zram_meta_free(zram, disksize);
1289 up_write(&zram->init_lock);
1293 static ssize_t reset_store(struct device *dev,
1294 struct device_attribute *attr, const char *buf, size_t len)
1297 unsigned short do_reset;
1299 struct block_device *bdev;
1301 ret = kstrtou16(buf, 10, &do_reset);
1308 zram = dev_to_zram(dev);
1309 bdev = bdget_disk(zram->disk, 0);
1313 mutex_lock(&bdev->bd_mutex);
1314 /* Do not reset an active device or claimed device */
1315 if (bdev->bd_openers || zram->claim) {
1316 mutex_unlock(&bdev->bd_mutex);
1321 /* From now on, anyone can't open /dev/zram[0-9] */
1323 mutex_unlock(&bdev->bd_mutex);
1325 /* Make sure all the pending I/O are finished */
1327 zram_reset_device(zram);
1328 zram_revalidate_disk(zram);
1331 mutex_lock(&bdev->bd_mutex);
1332 zram->claim = false;
1333 mutex_unlock(&bdev->bd_mutex);
1338 static int zram_open(struct block_device *bdev, fmode_t mode)
1343 WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1345 zram = bdev->bd_disk->private_data;
1346 /* zram was claimed to reset so open request fails */
1353 static const struct block_device_operations zram_devops = {
1355 .swap_slot_free_notify = zram_slot_free_notify,
1356 .rw_page = zram_rw_page,
1357 .owner = THIS_MODULE
1360 static DEVICE_ATTR_WO(compact);
1361 static DEVICE_ATTR_RW(disksize);
1362 static DEVICE_ATTR_RO(initstate);
1363 static DEVICE_ATTR_WO(reset);
1364 static DEVICE_ATTR_WO(mem_limit);
1365 static DEVICE_ATTR_WO(mem_used_max);
1366 static DEVICE_ATTR_RW(max_comp_streams);
1367 static DEVICE_ATTR_RW(comp_algorithm);
1368 #ifdef CONFIG_ZRAM_WRITEBACK
1369 static DEVICE_ATTR_RW(backing_dev);
1372 static struct attribute *zram_disk_attrs[] = {
1373 &dev_attr_disksize.attr,
1374 &dev_attr_initstate.attr,
1375 &dev_attr_reset.attr,
1376 &dev_attr_compact.attr,
1377 &dev_attr_mem_limit.attr,
1378 &dev_attr_mem_used_max.attr,
1379 &dev_attr_max_comp_streams.attr,
1380 &dev_attr_comp_algorithm.attr,
1381 #ifdef CONFIG_ZRAM_WRITEBACK
1382 &dev_attr_backing_dev.attr,
1384 &dev_attr_io_stat.attr,
1385 &dev_attr_mm_stat.attr,
1386 &dev_attr_debug_stat.attr,
1390 static const struct attribute_group zram_disk_attr_group = {
1391 .attrs = zram_disk_attrs,
1395 * Allocate and initialize new zram device. the function returns
1396 * '>= 0' device_id upon success, and negative value otherwise.
1398 static int zram_add(void)
1401 struct request_queue *queue;
1404 zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1408 ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1413 init_rwsem(&zram->init_lock);
1415 queue = blk_alloc_queue(GFP_KERNEL);
1417 pr_err("Error allocating disk queue for device %d\n",
1423 blk_queue_make_request(queue, zram_make_request);
1425 /* gendisk structure */
1426 zram->disk = alloc_disk(1);
1428 pr_err("Error allocating disk structure for device %d\n",
1431 goto out_free_queue;
1434 zram->disk->major = zram_major;
1435 zram->disk->first_minor = device_id;
1436 zram->disk->fops = &zram_devops;
1437 zram->disk->queue = queue;
1438 zram->disk->queue->queuedata = zram;
1439 zram->disk->private_data = zram;
1440 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1442 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1443 set_capacity(zram->disk, 0);
1444 /* zram devices sort of resembles non-rotational disks */
1445 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1446 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1448 * To ensure that we always get PAGE_SIZE aligned
1449 * and n*PAGE_SIZED sized I/O requests.
1451 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1452 blk_queue_logical_block_size(zram->disk->queue,
1453 ZRAM_LOGICAL_BLOCK_SIZE);
1454 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1455 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1456 zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1457 blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1458 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1461 * zram_bio_discard() will clear all logical blocks if logical block
1462 * size is identical with physical block size(PAGE_SIZE). But if it is
1463 * different, we will skip discarding some parts of logical blocks in
1464 * the part of the request range which isn't aligned to physical block
1465 * size. So we can't ensure that all discarded logical blocks are
1468 if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1469 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1471 add_disk(zram->disk);
1473 ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1474 &zram_disk_attr_group);
1476 pr_err("Error creating sysfs group for device %d\n",
1480 strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1482 pr_info("Added device: %s\n", zram->disk->disk_name);
1486 del_gendisk(zram->disk);
1487 put_disk(zram->disk);
1489 blk_cleanup_queue(queue);
1491 idr_remove(&zram_index_idr, device_id);
1497 static int zram_remove(struct zram *zram)
1499 struct block_device *bdev;
1501 bdev = bdget_disk(zram->disk, 0);
1505 mutex_lock(&bdev->bd_mutex);
1506 if (bdev->bd_openers || zram->claim) {
1507 mutex_unlock(&bdev->bd_mutex);
1513 mutex_unlock(&bdev->bd_mutex);
1516 * Remove sysfs first, so no one will perform a disksize
1517 * store while we destroy the devices. This also helps during
1518 * hot_remove -- zram_reset_device() is the last holder of
1519 * ->init_lock, no later/concurrent disksize_store() or any
1520 * other sysfs handlers are possible.
1522 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1523 &zram_disk_attr_group);
1525 /* Make sure all the pending I/O are finished */
1527 zram_reset_device(zram);
1530 pr_info("Removed device: %s\n", zram->disk->disk_name);
1532 blk_cleanup_queue(zram->disk->queue);
1533 del_gendisk(zram->disk);
1534 put_disk(zram->disk);
1539 /* zram-control sysfs attributes */
1542 * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
1543 * sense that reading from this file does alter the state of your system -- it
1544 * creates a new un-initialized zram device and returns back this device's
1545 * device_id (or an error code if it fails to create a new device).
1547 static ssize_t hot_add_show(struct class *class,
1548 struct class_attribute *attr,
1553 mutex_lock(&zram_index_mutex);
1555 mutex_unlock(&zram_index_mutex);
1559 return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1561 static CLASS_ATTR_RO(hot_add);
1563 static ssize_t hot_remove_store(struct class *class,
1564 struct class_attribute *attr,
1571 /* dev_id is gendisk->first_minor, which is `int' */
1572 ret = kstrtoint(buf, 10, &dev_id);
1578 mutex_lock(&zram_index_mutex);
1580 zram = idr_find(&zram_index_idr, dev_id);
1582 ret = zram_remove(zram);
1584 idr_remove(&zram_index_idr, dev_id);
1589 mutex_unlock(&zram_index_mutex);
1590 return ret ? ret : count;
1592 static CLASS_ATTR_WO(hot_remove);
1594 static struct attribute *zram_control_class_attrs[] = {
1595 &class_attr_hot_add.attr,
1596 &class_attr_hot_remove.attr,
1599 ATTRIBUTE_GROUPS(zram_control_class);
1601 static struct class zram_control_class = {
1602 .name = "zram-control",
1603 .owner = THIS_MODULE,
1604 .class_groups = zram_control_class_groups,
1607 static int zram_remove_cb(int id, void *ptr, void *data)
1613 static void destroy_devices(void)
1615 class_unregister(&zram_control_class);
1616 idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1617 idr_destroy(&zram_index_idr);
1618 unregister_blkdev(zram_major, "zram");
1619 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1622 static int __init zram_init(void)
1626 ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
1627 zcomp_cpu_up_prepare, zcomp_cpu_dead);
1631 ret = class_register(&zram_control_class);
1633 pr_err("Unable to register zram-control class\n");
1634 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1638 zram_major = register_blkdev(0, "zram");
1639 if (zram_major <= 0) {
1640 pr_err("Unable to get major number\n");
1641 class_unregister(&zram_control_class);
1642 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1646 while (num_devices != 0) {
1647 mutex_lock(&zram_index_mutex);
1649 mutex_unlock(&zram_index_mutex);
1662 static void __exit zram_exit(void)
1667 module_init(zram_init);
1668 module_exit(zram_exit);
1670 module_param(num_devices, uint, 0);
1671 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1673 MODULE_LICENSE("Dual BSD/GPL");
1674 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1675 MODULE_DESCRIPTION("Compressed RAM Block Device");