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/debugfs.h>
35 #include <linux/cpuhotplug.h>
39 static DEFINE_IDR(zram_index_idr);
40 /* idr index must be protected */
41 static DEFINE_MUTEX(zram_index_mutex);
43 static int zram_major;
44 static const char *default_compressor = "lzo";
46 /* Module params (documentation at end) */
47 static unsigned int num_devices = 1;
49 * Pages that compress to sizes equals or greater than this are stored
50 * uncompressed in memory.
52 static size_t huge_class_size;
54 static void zram_free_page(struct zram *zram, size_t index);
56 static void zram_slot_lock(struct zram *zram, u32 index)
58 bit_spin_lock(ZRAM_LOCK, &zram->table[index].value);
61 static void zram_slot_unlock(struct zram *zram, u32 index)
63 bit_spin_unlock(ZRAM_LOCK, &zram->table[index].value);
66 static inline bool init_done(struct zram *zram)
68 return zram->disksize;
71 static inline bool zram_allocated(struct zram *zram, u32 index)
74 return (zram->table[index].value >> (ZRAM_FLAG_SHIFT + 1)) ||
75 zram->table[index].handle;
78 static inline struct zram *dev_to_zram(struct device *dev)
80 return (struct zram *)dev_to_disk(dev)->private_data;
83 static unsigned long zram_get_handle(struct zram *zram, u32 index)
85 return zram->table[index].handle;
88 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
90 zram->table[index].handle = handle;
93 /* flag operations require table entry bit_spin_lock() being held */
94 static bool zram_test_flag(struct zram *zram, u32 index,
95 enum zram_pageflags flag)
97 return zram->table[index].value & BIT(flag);
100 static void zram_set_flag(struct zram *zram, u32 index,
101 enum zram_pageflags flag)
103 zram->table[index].value |= BIT(flag);
106 static void zram_clear_flag(struct zram *zram, u32 index,
107 enum zram_pageflags flag)
109 zram->table[index].value &= ~BIT(flag);
112 static inline void zram_set_element(struct zram *zram, u32 index,
113 unsigned long element)
115 zram->table[index].element = element;
118 static unsigned long zram_get_element(struct zram *zram, u32 index)
120 return zram->table[index].element;
123 static size_t zram_get_obj_size(struct zram *zram, u32 index)
125 return zram->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
128 static void zram_set_obj_size(struct zram *zram,
129 u32 index, size_t size)
131 unsigned long flags = zram->table[index].value >> ZRAM_FLAG_SHIFT;
133 zram->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
136 #if PAGE_SIZE != 4096
137 static inline bool is_partial_io(struct bio_vec *bvec)
139 return bvec->bv_len != PAGE_SIZE;
142 static inline bool is_partial_io(struct bio_vec *bvec)
149 * Check if request is within bounds and aligned on zram logical blocks.
151 static inline bool valid_io_request(struct zram *zram,
152 sector_t start, unsigned int size)
156 /* unaligned request */
157 if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
159 if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
162 end = start + (size >> SECTOR_SHIFT);
163 bound = zram->disksize >> SECTOR_SHIFT;
164 /* out of range range */
165 if (unlikely(start >= bound || end > bound || start > end))
168 /* I/O request is valid */
172 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
174 *index += (*offset + bvec->bv_len) / PAGE_SIZE;
175 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
178 static inline void update_used_max(struct zram *zram,
179 const unsigned long pages)
181 unsigned long old_max, cur_max;
183 old_max = atomic_long_read(&zram->stats.max_used_pages);
188 old_max = atomic_long_cmpxchg(
189 &zram->stats.max_used_pages, cur_max, pages);
190 } while (old_max != cur_max);
193 static inline void zram_fill_page(void *ptr, unsigned long len,
196 WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
197 memset_l(ptr, value, len / sizeof(unsigned long));
200 static bool page_same_filled(void *ptr, unsigned long *element)
206 page = (unsigned long *)ptr;
209 for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
210 if (val != page[pos])
219 static ssize_t initstate_show(struct device *dev,
220 struct device_attribute *attr, char *buf)
223 struct zram *zram = dev_to_zram(dev);
225 down_read(&zram->init_lock);
226 val = init_done(zram);
227 up_read(&zram->init_lock);
229 return scnprintf(buf, PAGE_SIZE, "%u\n", val);
232 static ssize_t disksize_show(struct device *dev,
233 struct device_attribute *attr, char *buf)
235 struct zram *zram = dev_to_zram(dev);
237 return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
240 static ssize_t mem_limit_store(struct device *dev,
241 struct device_attribute *attr, const char *buf, size_t len)
245 struct zram *zram = dev_to_zram(dev);
247 limit = memparse(buf, &tmp);
248 if (buf == tmp) /* no chars parsed, invalid input */
251 down_write(&zram->init_lock);
252 zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
253 up_write(&zram->init_lock);
258 static ssize_t mem_used_max_store(struct device *dev,
259 struct device_attribute *attr, const char *buf, size_t len)
263 struct zram *zram = dev_to_zram(dev);
265 err = kstrtoul(buf, 10, &val);
269 down_read(&zram->init_lock);
270 if (init_done(zram)) {
271 atomic_long_set(&zram->stats.max_used_pages,
272 zs_get_total_pages(zram->mem_pool));
274 up_read(&zram->init_lock);
279 #ifdef CONFIG_ZRAM_WRITEBACK
280 static bool zram_wb_enabled(struct zram *zram)
282 return zram->backing_dev;
285 static void reset_bdev(struct zram *zram)
287 struct block_device *bdev;
289 if (!zram_wb_enabled(zram))
293 if (zram->old_block_size)
294 set_blocksize(bdev, zram->old_block_size);
295 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
296 /* hope filp_close flush all of IO */
297 filp_close(zram->backing_dev, NULL);
298 zram->backing_dev = NULL;
299 zram->old_block_size = 0;
301 zram->disk->queue->backing_dev_info->capabilities |=
302 BDI_CAP_SYNCHRONOUS_IO;
303 kvfree(zram->bitmap);
307 static ssize_t backing_dev_show(struct device *dev,
308 struct device_attribute *attr, char *buf)
310 struct zram *zram = dev_to_zram(dev);
311 struct file *file = zram->backing_dev;
315 down_read(&zram->init_lock);
316 if (!zram_wb_enabled(zram)) {
317 memcpy(buf, "none\n", 5);
318 up_read(&zram->init_lock);
322 p = file_path(file, buf, PAGE_SIZE - 1);
329 memmove(buf, p, ret);
332 up_read(&zram->init_lock);
336 static ssize_t backing_dev_store(struct device *dev,
337 struct device_attribute *attr, const char *buf, size_t len)
341 struct file *backing_dev = NULL;
343 struct address_space *mapping;
344 unsigned int bitmap_sz, old_block_size = 0;
345 unsigned long nr_pages, *bitmap = NULL;
346 struct block_device *bdev = NULL;
348 struct zram *zram = dev_to_zram(dev);
350 file_name = kmalloc(PATH_MAX, GFP_KERNEL);
354 down_write(&zram->init_lock);
355 if (init_done(zram)) {
356 pr_info("Can't setup backing device for initialized device\n");
361 strlcpy(file_name, buf, PATH_MAX);
362 /* ignore trailing newline */
363 sz = strlen(file_name);
364 if (sz > 0 && file_name[sz - 1] == '\n')
365 file_name[sz - 1] = 0x00;
367 backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
368 if (IS_ERR(backing_dev)) {
369 err = PTR_ERR(backing_dev);
374 mapping = backing_dev->f_mapping;
375 inode = mapping->host;
377 /* Support only block device in this moment */
378 if (!S_ISBLK(inode->i_mode)) {
383 bdev = bdgrab(I_BDEV(inode));
384 err = blkdev_get(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
388 nr_pages = i_size_read(inode) >> PAGE_SHIFT;
389 bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
390 bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
396 old_block_size = block_size(bdev);
397 err = set_blocksize(bdev, PAGE_SIZE);
402 spin_lock_init(&zram->bitmap_lock);
404 zram->old_block_size = old_block_size;
406 zram->backing_dev = backing_dev;
407 zram->bitmap = bitmap;
408 zram->nr_pages = nr_pages;
410 * With writeback feature, zram does asynchronous IO so it's no longer
411 * synchronous device so let's remove synchronous io flag. Othewise,
412 * upper layer(e.g., swap) could wait IO completion rather than
413 * (submit and return), which will cause system sluggish.
414 * Furthermore, when the IO function returns(e.g., swap_readpage),
415 * upper layer expects IO was done so it could deallocate the page
416 * freely but in fact, IO is going on so finally could cause
417 * use-after-free when the IO is really done.
419 zram->disk->queue->backing_dev_info->capabilities &=
420 ~BDI_CAP_SYNCHRONOUS_IO;
421 up_write(&zram->init_lock);
423 pr_info("setup backing device %s\n", file_name);
432 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
435 filp_close(backing_dev, NULL);
437 up_write(&zram->init_lock);
444 static unsigned long get_entry_bdev(struct zram *zram)
448 spin_lock(&zram->bitmap_lock);
449 /* skip 0 bit to confuse zram.handle = 0 */
450 entry = find_next_zero_bit(zram->bitmap, zram->nr_pages, 1);
451 if (entry == zram->nr_pages) {
452 spin_unlock(&zram->bitmap_lock);
456 set_bit(entry, zram->bitmap);
457 spin_unlock(&zram->bitmap_lock);
462 static void put_entry_bdev(struct zram *zram, unsigned long entry)
466 spin_lock(&zram->bitmap_lock);
467 was_set = test_and_clear_bit(entry, zram->bitmap);
468 spin_unlock(&zram->bitmap_lock);
469 WARN_ON_ONCE(!was_set);
472 static void zram_page_end_io(struct bio *bio)
474 struct page *page = bio_first_page_all(bio);
476 page_endio(page, op_is_write(bio_op(bio)),
477 blk_status_to_errno(bio->bi_status));
482 * Returns 1 if the submission is successful.
484 static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
485 unsigned long entry, struct bio *parent)
489 bio = bio_alloc(GFP_ATOMIC, 1);
493 bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
494 bio_set_dev(bio, zram->bdev);
495 if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
501 bio->bi_opf = REQ_OP_READ;
502 bio->bi_end_io = zram_page_end_io;
504 bio->bi_opf = parent->bi_opf;
505 bio_chain(bio, parent);
513 struct work_struct work;
519 #if PAGE_SIZE != 4096
520 static void zram_sync_read(struct work_struct *work)
523 struct zram_work *zw = container_of(work, struct zram_work, work);
524 struct zram *zram = zw->zram;
525 unsigned long entry = zw->entry;
526 struct bio *bio = zw->bio;
528 read_from_bdev_async(zram, &bvec, entry, bio);
532 * Block layer want one ->make_request_fn to be active at a time
533 * so if we use chained IO with parent IO in same context,
534 * it's a deadlock. To avoid, it, it uses worker thread context.
536 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
537 unsigned long entry, struct bio *bio)
539 struct zram_work work;
545 INIT_WORK_ONSTACK(&work.work, zram_sync_read);
546 queue_work(system_unbound_wq, &work.work);
547 flush_work(&work.work);
548 destroy_work_on_stack(&work.work);
553 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
554 unsigned long entry, struct bio *bio)
561 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
562 unsigned long entry, struct bio *parent, bool sync)
565 return read_from_bdev_sync(zram, bvec, entry, parent);
567 return read_from_bdev_async(zram, bvec, entry, parent);
570 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
571 u32 index, struct bio *parent,
572 unsigned long *pentry)
577 bio = bio_alloc(GFP_ATOMIC, 1);
581 entry = get_entry_bdev(zram);
587 bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
588 bio_set_dev(bio, zram->bdev);
589 if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len,
592 put_entry_bdev(zram, entry);
597 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
598 bio->bi_end_io = zram_page_end_io;
600 bio->bi_opf = parent->bi_opf;
601 bio_chain(bio, parent);
610 static void zram_wb_clear(struct zram *zram, u32 index)
614 zram_clear_flag(zram, index, ZRAM_WB);
615 entry = zram_get_element(zram, index);
616 zram_set_element(zram, index, 0);
617 put_entry_bdev(zram, entry);
621 static bool zram_wb_enabled(struct zram *zram) { return false; }
622 static inline void reset_bdev(struct zram *zram) {};
623 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
624 u32 index, struct bio *parent,
625 unsigned long *pentry)
631 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
632 unsigned long entry, struct bio *parent, bool sync)
636 static void zram_wb_clear(struct zram *zram, u32 index) {}
639 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
641 static struct dentry *zram_debugfs_root;
643 static void zram_debugfs_create(void)
645 zram_debugfs_root = debugfs_create_dir("zram", NULL);
648 static void zram_debugfs_destroy(void)
650 debugfs_remove_recursive(zram_debugfs_root);
653 static void zram_accessed(struct zram *zram, u32 index)
655 zram->table[index].ac_time = ktime_get_boottime();
658 static void zram_reset_access(struct zram *zram, u32 index)
660 zram->table[index].ac_time = 0;
663 static ssize_t read_block_state(struct file *file, char __user *buf,
664 size_t count, loff_t *ppos)
667 ssize_t index, written = 0;
668 struct zram *zram = file->private_data;
669 unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
670 struct timespec64 ts;
672 kbuf = kvmalloc(count, GFP_KERNEL);
676 down_read(&zram->init_lock);
677 if (!init_done(zram)) {
678 up_read(&zram->init_lock);
683 for (index = *ppos; index < nr_pages; index++) {
686 zram_slot_lock(zram, index);
687 if (!zram_allocated(zram, index))
690 ts = ktime_to_timespec64(zram->table[index].ac_time);
691 copied = snprintf(kbuf + written, count,
692 "%12zd %12lld.%06lu %c%c%c\n",
693 index, (s64)ts.tv_sec,
694 ts.tv_nsec / NSEC_PER_USEC,
695 zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
696 zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
697 zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.');
699 if (count < copied) {
700 zram_slot_unlock(zram, index);
706 zram_slot_unlock(zram, index);
710 up_read(&zram->init_lock);
711 if (copy_to_user(buf, kbuf, written))
718 static const struct file_operations proc_zram_block_state_op = {
720 .read = read_block_state,
721 .llseek = default_llseek,
724 static void zram_debugfs_register(struct zram *zram)
726 if (!zram_debugfs_root)
729 zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
731 debugfs_create_file("block_state", 0400, zram->debugfs_dir,
732 zram, &proc_zram_block_state_op);
735 static void zram_debugfs_unregister(struct zram *zram)
737 debugfs_remove_recursive(zram->debugfs_dir);
740 static void zram_debugfs_create(void) {};
741 static void zram_debugfs_destroy(void) {};
742 static void zram_accessed(struct zram *zram, u32 index) {};
743 static void zram_reset_access(struct zram *zram, u32 index) {};
744 static void zram_debugfs_register(struct zram *zram) {};
745 static void zram_debugfs_unregister(struct zram *zram) {};
749 * We switched to per-cpu streams and this attr is not needed anymore.
750 * However, we will keep it around for some time, because:
751 * a) we may revert per-cpu streams in the future
752 * b) it's visible to user space and we need to follow our 2 years
753 * retirement rule; but we already have a number of 'soon to be
754 * altered' attrs, so max_comp_streams need to wait for the next
757 static ssize_t max_comp_streams_show(struct device *dev,
758 struct device_attribute *attr, char *buf)
760 return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
763 static ssize_t max_comp_streams_store(struct device *dev,
764 struct device_attribute *attr, const char *buf, size_t len)
769 static ssize_t comp_algorithm_show(struct device *dev,
770 struct device_attribute *attr, char *buf)
773 struct zram *zram = dev_to_zram(dev);
775 down_read(&zram->init_lock);
776 sz = zcomp_available_show(zram->compressor, buf);
777 up_read(&zram->init_lock);
782 static ssize_t comp_algorithm_store(struct device *dev,
783 struct device_attribute *attr, const char *buf, size_t len)
785 struct zram *zram = dev_to_zram(dev);
786 char compressor[ARRAY_SIZE(zram->compressor)];
789 strlcpy(compressor, buf, sizeof(compressor));
790 /* ignore trailing newline */
791 sz = strlen(compressor);
792 if (sz > 0 && compressor[sz - 1] == '\n')
793 compressor[sz - 1] = 0x00;
795 if (!zcomp_available_algorithm(compressor))
798 down_write(&zram->init_lock);
799 if (init_done(zram)) {
800 up_write(&zram->init_lock);
801 pr_info("Can't change algorithm for initialized device\n");
805 strcpy(zram->compressor, compressor);
806 up_write(&zram->init_lock);
810 static ssize_t compact_store(struct device *dev,
811 struct device_attribute *attr, const char *buf, size_t len)
813 struct zram *zram = dev_to_zram(dev);
815 down_read(&zram->init_lock);
816 if (!init_done(zram)) {
817 up_read(&zram->init_lock);
821 zs_compact(zram->mem_pool);
822 up_read(&zram->init_lock);
827 static ssize_t io_stat_show(struct device *dev,
828 struct device_attribute *attr, char *buf)
830 struct zram *zram = dev_to_zram(dev);
833 down_read(&zram->init_lock);
834 ret = scnprintf(buf, PAGE_SIZE,
835 "%8llu %8llu %8llu %8llu\n",
836 (u64)atomic64_read(&zram->stats.failed_reads),
837 (u64)atomic64_read(&zram->stats.failed_writes),
838 (u64)atomic64_read(&zram->stats.invalid_io),
839 (u64)atomic64_read(&zram->stats.notify_free));
840 up_read(&zram->init_lock);
845 static ssize_t mm_stat_show(struct device *dev,
846 struct device_attribute *attr, char *buf)
848 struct zram *zram = dev_to_zram(dev);
849 struct zs_pool_stats pool_stats;
850 u64 orig_size, mem_used = 0;
854 memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
856 down_read(&zram->init_lock);
857 if (init_done(zram)) {
858 mem_used = zs_get_total_pages(zram->mem_pool);
859 zs_pool_stats(zram->mem_pool, &pool_stats);
862 orig_size = atomic64_read(&zram->stats.pages_stored);
863 max_used = atomic_long_read(&zram->stats.max_used_pages);
865 ret = scnprintf(buf, PAGE_SIZE,
866 "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu\n",
867 orig_size << PAGE_SHIFT,
868 (u64)atomic64_read(&zram->stats.compr_data_size),
869 mem_used << PAGE_SHIFT,
870 zram->limit_pages << PAGE_SHIFT,
871 max_used << PAGE_SHIFT,
872 (u64)atomic64_read(&zram->stats.same_pages),
873 pool_stats.pages_compacted,
874 (u64)atomic64_read(&zram->stats.huge_pages));
875 up_read(&zram->init_lock);
880 static ssize_t debug_stat_show(struct device *dev,
881 struct device_attribute *attr, char *buf)
884 struct zram *zram = dev_to_zram(dev);
887 down_read(&zram->init_lock);
888 ret = scnprintf(buf, PAGE_SIZE,
889 "version: %d\n%8llu\n",
891 (u64)atomic64_read(&zram->stats.writestall));
892 up_read(&zram->init_lock);
897 static DEVICE_ATTR_RO(io_stat);
898 static DEVICE_ATTR_RO(mm_stat);
899 static DEVICE_ATTR_RO(debug_stat);
901 static void zram_meta_free(struct zram *zram, u64 disksize)
903 size_t num_pages = disksize >> PAGE_SHIFT;
906 /* Free all pages that are still in this zram device */
907 for (index = 0; index < num_pages; index++)
908 zram_free_page(zram, index);
910 zs_destroy_pool(zram->mem_pool);
914 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
918 num_pages = disksize >> PAGE_SHIFT;
919 zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
923 zram->mem_pool = zs_create_pool(zram->disk->disk_name);
924 if (!zram->mem_pool) {
929 if (!huge_class_size)
930 huge_class_size = zs_huge_class_size(zram->mem_pool);
935 * To protect concurrent access to the same index entry,
936 * caller should hold this table index entry's bit_spinlock to
937 * indicate this index entry is accessing.
939 static void zram_free_page(struct zram *zram, size_t index)
941 unsigned long handle;
943 zram_reset_access(zram, index);
945 if (zram_test_flag(zram, index, ZRAM_HUGE)) {
946 zram_clear_flag(zram, index, ZRAM_HUGE);
947 atomic64_dec(&zram->stats.huge_pages);
950 if (zram_wb_enabled(zram) && zram_test_flag(zram, index, ZRAM_WB)) {
951 zram_wb_clear(zram, index);
952 atomic64_dec(&zram->stats.pages_stored);
957 * No memory is allocated for same element filled pages.
958 * Simply clear same page flag.
960 if (zram_test_flag(zram, index, ZRAM_SAME)) {
961 zram_clear_flag(zram, index, ZRAM_SAME);
962 zram_set_element(zram, index, 0);
963 atomic64_dec(&zram->stats.same_pages);
964 atomic64_dec(&zram->stats.pages_stored);
968 handle = zram_get_handle(zram, index);
972 zs_free(zram->mem_pool, handle);
974 atomic64_sub(zram_get_obj_size(zram, index),
975 &zram->stats.compr_data_size);
976 atomic64_dec(&zram->stats.pages_stored);
978 zram_set_handle(zram, index, 0);
979 zram_set_obj_size(zram, index, 0);
982 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
983 struct bio *bio, bool partial_io)
986 unsigned long handle;
990 if (zram_wb_enabled(zram)) {
991 zram_slot_lock(zram, index);
992 if (zram_test_flag(zram, index, ZRAM_WB)) {
995 zram_slot_unlock(zram, index);
998 bvec.bv_len = PAGE_SIZE;
1000 return read_from_bdev(zram, &bvec,
1001 zram_get_element(zram, index),
1004 zram_slot_unlock(zram, index);
1007 zram_slot_lock(zram, index);
1008 handle = zram_get_handle(zram, index);
1009 if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1010 unsigned long value;
1013 value = handle ? zram_get_element(zram, index) : 0;
1014 mem = kmap_atomic(page);
1015 zram_fill_page(mem, PAGE_SIZE, value);
1017 zram_slot_unlock(zram, index);
1021 size = zram_get_obj_size(zram, index);
1023 src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1024 if (size == PAGE_SIZE) {
1025 dst = kmap_atomic(page);
1026 memcpy(dst, src, PAGE_SIZE);
1030 struct zcomp_strm *zstrm = zcomp_stream_get(zram->comp);
1032 dst = kmap_atomic(page);
1033 ret = zcomp_decompress(zstrm, src, size, dst);
1035 zcomp_stream_put(zram->comp);
1037 zs_unmap_object(zram->mem_pool, handle);
1038 zram_slot_unlock(zram, index);
1040 /* Should NEVER happen. Return bio error if it does. */
1042 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1047 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1048 u32 index, int offset, struct bio *bio)
1053 page = bvec->bv_page;
1054 if (is_partial_io(bvec)) {
1055 /* Use a temporary buffer to decompress the page */
1056 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1061 ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
1065 if (is_partial_io(bvec)) {
1066 void *dst = kmap_atomic(bvec->bv_page);
1067 void *src = kmap_atomic(page);
1069 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
1074 if (is_partial_io(bvec))
1080 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1081 u32 index, struct bio *bio)
1084 unsigned long alloced_pages;
1085 unsigned long handle = 0;
1086 unsigned int comp_len = 0;
1087 void *src, *dst, *mem;
1088 struct zcomp_strm *zstrm;
1089 struct page *page = bvec->bv_page;
1090 unsigned long element = 0;
1091 enum zram_pageflags flags = 0;
1092 bool allow_wb = true;
1094 mem = kmap_atomic(page);
1095 if (page_same_filled(mem, &element)) {
1097 /* Free memory associated with this sector now. */
1099 atomic64_inc(&zram->stats.same_pages);
1105 zstrm = zcomp_stream_get(zram->comp);
1106 src = kmap_atomic(page);
1107 ret = zcomp_compress(zstrm, src, &comp_len);
1110 if (unlikely(ret)) {
1111 zcomp_stream_put(zram->comp);
1112 pr_err("Compression failed! err=%d\n", ret);
1113 zs_free(zram->mem_pool, handle);
1117 if (unlikely(comp_len >= huge_class_size)) {
1118 comp_len = PAGE_SIZE;
1119 if (zram_wb_enabled(zram) && allow_wb) {
1120 zcomp_stream_put(zram->comp);
1121 ret = write_to_bdev(zram, bvec, index, bio, &element);
1128 goto compress_again;
1133 * handle allocation has 2 paths:
1134 * a) fast path is executed with preemption disabled (for
1135 * per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1136 * since we can't sleep;
1137 * b) slow path enables preemption and attempts to allocate
1138 * the page with __GFP_DIRECT_RECLAIM bit set. we have to
1139 * put per-cpu compression stream and, thus, to re-do
1140 * the compression once handle is allocated.
1142 * if we have a 'non-null' handle here then we are coming
1143 * from the slow path and handle has already been allocated.
1146 handle = zs_malloc(zram->mem_pool, comp_len,
1147 __GFP_KSWAPD_RECLAIM |
1152 zcomp_stream_put(zram->comp);
1153 atomic64_inc(&zram->stats.writestall);
1154 handle = zs_malloc(zram->mem_pool, comp_len,
1155 GFP_NOIO | __GFP_HIGHMEM |
1158 goto compress_again;
1162 alloced_pages = zs_get_total_pages(zram->mem_pool);
1163 update_used_max(zram, alloced_pages);
1165 if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1166 zcomp_stream_put(zram->comp);
1167 zs_free(zram->mem_pool, handle);
1171 dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1173 src = zstrm->buffer;
1174 if (comp_len == PAGE_SIZE)
1175 src = kmap_atomic(page);
1176 memcpy(dst, src, comp_len);
1177 if (comp_len == PAGE_SIZE)
1180 zcomp_stream_put(zram->comp);
1181 zs_unmap_object(zram->mem_pool, handle);
1182 atomic64_add(comp_len, &zram->stats.compr_data_size);
1185 * Free memory associated with this sector
1186 * before overwriting unused sectors.
1188 zram_slot_lock(zram, index);
1189 zram_free_page(zram, index);
1191 if (comp_len == PAGE_SIZE) {
1192 zram_set_flag(zram, index, ZRAM_HUGE);
1193 atomic64_inc(&zram->stats.huge_pages);
1197 zram_set_flag(zram, index, flags);
1198 zram_set_element(zram, index, element);
1200 zram_set_handle(zram, index, handle);
1201 zram_set_obj_size(zram, index, comp_len);
1203 zram_slot_unlock(zram, index);
1206 atomic64_inc(&zram->stats.pages_stored);
1210 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1211 u32 index, int offset, struct bio *bio)
1214 struct page *page = NULL;
1219 if (is_partial_io(bvec)) {
1222 * This is a partial IO. We need to read the full page
1223 * before to write the changes.
1225 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1229 ret = __zram_bvec_read(zram, page, index, bio, true);
1233 src = kmap_atomic(bvec->bv_page);
1234 dst = kmap_atomic(page);
1235 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1240 vec.bv_len = PAGE_SIZE;
1244 ret = __zram_bvec_write(zram, &vec, index, bio);
1246 if (is_partial_io(bvec))
1252 * zram_bio_discard - handler on discard request
1253 * @index: physical block index in PAGE_SIZE units
1254 * @offset: byte offset within physical block
1256 static void zram_bio_discard(struct zram *zram, u32 index,
1257 int offset, struct bio *bio)
1259 size_t n = bio->bi_iter.bi_size;
1262 * zram manages data in physical block size units. Because logical block
1263 * size isn't identical with physical block size on some arch, we
1264 * could get a discard request pointing to a specific offset within a
1265 * certain physical block. Although we can handle this request by
1266 * reading that physiclal block and decompressing and partially zeroing
1267 * and re-compressing and then re-storing it, this isn't reasonable
1268 * because our intent with a discard request is to save memory. So
1269 * skipping this logical block is appropriate here.
1272 if (n <= (PAGE_SIZE - offset))
1275 n -= (PAGE_SIZE - offset);
1279 while (n >= PAGE_SIZE) {
1280 zram_slot_lock(zram, index);
1281 zram_free_page(zram, index);
1282 zram_slot_unlock(zram, index);
1283 atomic64_inc(&zram->stats.notify_free);
1290 * Returns errno if it has some problem. Otherwise return 0 or 1.
1291 * Returns 0 if IO request was done synchronously
1292 * Returns 1 if IO request was successfully submitted.
1294 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1295 int offset, unsigned int op, struct bio *bio)
1297 unsigned long start_time = jiffies;
1298 struct request_queue *q = zram->disk->queue;
1301 generic_start_io_acct(q, op, bvec->bv_len >> SECTOR_SHIFT,
1302 &zram->disk->part0);
1304 if (!op_is_write(op)) {
1305 atomic64_inc(&zram->stats.num_reads);
1306 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1307 flush_dcache_page(bvec->bv_page);
1309 atomic64_inc(&zram->stats.num_writes);
1310 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1313 generic_end_io_acct(q, op, &zram->disk->part0, start_time);
1315 zram_slot_lock(zram, index);
1316 zram_accessed(zram, index);
1317 zram_slot_unlock(zram, index);
1319 if (unlikely(ret < 0)) {
1320 if (!op_is_write(op))
1321 atomic64_inc(&zram->stats.failed_reads);
1323 atomic64_inc(&zram->stats.failed_writes);
1329 static void __zram_make_request(struct zram *zram, struct bio *bio)
1333 struct bio_vec bvec;
1334 struct bvec_iter iter;
1336 index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1337 offset = (bio->bi_iter.bi_sector &
1338 (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1340 switch (bio_op(bio)) {
1341 case REQ_OP_DISCARD:
1342 case REQ_OP_WRITE_ZEROES:
1343 zram_bio_discard(zram, index, offset, bio);
1350 bio_for_each_segment(bvec, bio, iter) {
1351 struct bio_vec bv = bvec;
1352 unsigned int unwritten = bvec.bv_len;
1355 bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1357 if (zram_bvec_rw(zram, &bv, index, offset,
1358 bio_op(bio), bio) < 0)
1361 bv.bv_offset += bv.bv_len;
1362 unwritten -= bv.bv_len;
1364 update_position(&index, &offset, &bv);
1365 } while (unwritten);
1376 * Handler function for all zram I/O requests.
1378 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
1380 struct zram *zram = queue->queuedata;
1382 if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1383 bio->bi_iter.bi_size)) {
1384 atomic64_inc(&zram->stats.invalid_io);
1388 __zram_make_request(zram, bio);
1389 return BLK_QC_T_NONE;
1393 return BLK_QC_T_NONE;
1396 static void zram_slot_free_notify(struct block_device *bdev,
1397 unsigned long index)
1401 zram = bdev->bd_disk->private_data;
1403 zram_slot_lock(zram, index);
1404 zram_free_page(zram, index);
1405 zram_slot_unlock(zram, index);
1406 atomic64_inc(&zram->stats.notify_free);
1409 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1410 struct page *page, unsigned int op)
1417 if (PageTransHuge(page))
1419 zram = bdev->bd_disk->private_data;
1421 if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1422 atomic64_inc(&zram->stats.invalid_io);
1427 index = sector >> SECTORS_PER_PAGE_SHIFT;
1428 offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1431 bv.bv_len = PAGE_SIZE;
1434 ret = zram_bvec_rw(zram, &bv, index, offset, op, NULL);
1437 * If I/O fails, just return error(ie, non-zero) without
1438 * calling page_endio.
1439 * It causes resubmit the I/O with bio request by upper functions
1440 * of rw_page(e.g., swap_readpage, __swap_writepage) and
1441 * bio->bi_end_io does things to handle the error
1442 * (e.g., SetPageError, set_page_dirty and extra works).
1444 if (unlikely(ret < 0))
1449 page_endio(page, op_is_write(op), 0);
1460 static void zram_reset_device(struct zram *zram)
1465 down_write(&zram->init_lock);
1467 zram->limit_pages = 0;
1469 if (!init_done(zram)) {
1470 up_write(&zram->init_lock);
1475 disksize = zram->disksize;
1478 set_capacity(zram->disk, 0);
1479 part_stat_set_all(&zram->disk->part0, 0);
1481 up_write(&zram->init_lock);
1482 /* I/O operation under all of CPU are done so let's free */
1483 zram_meta_free(zram, disksize);
1484 memset(&zram->stats, 0, sizeof(zram->stats));
1485 zcomp_destroy(comp);
1489 static ssize_t disksize_store(struct device *dev,
1490 struct device_attribute *attr, const char *buf, size_t len)
1494 struct zram *zram = dev_to_zram(dev);
1497 disksize = memparse(buf, NULL);
1501 down_write(&zram->init_lock);
1502 if (init_done(zram)) {
1503 pr_info("Cannot change disksize for initialized device\n");
1508 disksize = PAGE_ALIGN(disksize);
1509 if (!zram_meta_alloc(zram, disksize)) {
1514 comp = zcomp_create(zram->compressor);
1516 pr_err("Cannot initialise %s compressing backend\n",
1518 err = PTR_ERR(comp);
1523 zram->disksize = disksize;
1524 set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1526 revalidate_disk(zram->disk);
1527 up_write(&zram->init_lock);
1532 zram_meta_free(zram, disksize);
1534 up_write(&zram->init_lock);
1538 static ssize_t reset_store(struct device *dev,
1539 struct device_attribute *attr, const char *buf, size_t len)
1542 unsigned short do_reset;
1544 struct block_device *bdev;
1546 ret = kstrtou16(buf, 10, &do_reset);
1553 zram = dev_to_zram(dev);
1554 bdev = bdget_disk(zram->disk, 0);
1558 mutex_lock(&bdev->bd_mutex);
1559 /* Do not reset an active device or claimed device */
1560 if (bdev->bd_openers || zram->claim) {
1561 mutex_unlock(&bdev->bd_mutex);
1566 /* From now on, anyone can't open /dev/zram[0-9] */
1568 mutex_unlock(&bdev->bd_mutex);
1570 /* Make sure all the pending I/O are finished */
1572 zram_reset_device(zram);
1573 revalidate_disk(zram->disk);
1576 mutex_lock(&bdev->bd_mutex);
1577 zram->claim = false;
1578 mutex_unlock(&bdev->bd_mutex);
1583 static int zram_open(struct block_device *bdev, fmode_t mode)
1588 WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1590 zram = bdev->bd_disk->private_data;
1591 /* zram was claimed to reset so open request fails */
1598 static const struct block_device_operations zram_devops = {
1600 .swap_slot_free_notify = zram_slot_free_notify,
1601 .rw_page = zram_rw_page,
1602 .owner = THIS_MODULE
1605 static DEVICE_ATTR_WO(compact);
1606 static DEVICE_ATTR_RW(disksize);
1607 static DEVICE_ATTR_RO(initstate);
1608 static DEVICE_ATTR_WO(reset);
1609 static DEVICE_ATTR_WO(mem_limit);
1610 static DEVICE_ATTR_WO(mem_used_max);
1611 static DEVICE_ATTR_RW(max_comp_streams);
1612 static DEVICE_ATTR_RW(comp_algorithm);
1613 #ifdef CONFIG_ZRAM_WRITEBACK
1614 static DEVICE_ATTR_RW(backing_dev);
1617 static struct attribute *zram_disk_attrs[] = {
1618 &dev_attr_disksize.attr,
1619 &dev_attr_initstate.attr,
1620 &dev_attr_reset.attr,
1621 &dev_attr_compact.attr,
1622 &dev_attr_mem_limit.attr,
1623 &dev_attr_mem_used_max.attr,
1624 &dev_attr_max_comp_streams.attr,
1625 &dev_attr_comp_algorithm.attr,
1626 #ifdef CONFIG_ZRAM_WRITEBACK
1627 &dev_attr_backing_dev.attr,
1629 &dev_attr_io_stat.attr,
1630 &dev_attr_mm_stat.attr,
1631 &dev_attr_debug_stat.attr,
1635 static const struct attribute_group zram_disk_attr_group = {
1636 .attrs = zram_disk_attrs,
1639 static const struct attribute_group *zram_disk_attr_groups[] = {
1640 &zram_disk_attr_group,
1645 * Allocate and initialize new zram device. the function returns
1646 * '>= 0' device_id upon success, and negative value otherwise.
1648 static int zram_add(void)
1651 struct request_queue *queue;
1654 zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1658 ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1663 init_rwsem(&zram->init_lock);
1665 queue = blk_alloc_queue(GFP_KERNEL);
1667 pr_err("Error allocating disk queue for device %d\n",
1673 blk_queue_make_request(queue, zram_make_request);
1675 /* gendisk structure */
1676 zram->disk = alloc_disk(1);
1678 pr_err("Error allocating disk structure for device %d\n",
1681 goto out_free_queue;
1684 zram->disk->major = zram_major;
1685 zram->disk->first_minor = device_id;
1686 zram->disk->fops = &zram_devops;
1687 zram->disk->queue = queue;
1688 zram->disk->queue->queuedata = zram;
1689 zram->disk->private_data = zram;
1690 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1692 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1693 set_capacity(zram->disk, 0);
1694 /* zram devices sort of resembles non-rotational disks */
1695 blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
1696 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1699 * To ensure that we always get PAGE_SIZE aligned
1700 * and n*PAGE_SIZED sized I/O requests.
1702 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1703 blk_queue_logical_block_size(zram->disk->queue,
1704 ZRAM_LOGICAL_BLOCK_SIZE);
1705 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1706 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1707 zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1708 blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1709 blk_queue_flag_set(QUEUE_FLAG_DISCARD, zram->disk->queue);
1712 * zram_bio_discard() will clear all logical blocks if logical block
1713 * size is identical with physical block size(PAGE_SIZE). But if it is
1714 * different, we will skip discarding some parts of logical blocks in
1715 * the part of the request range which isn't aligned to physical block
1716 * size. So we can't ensure that all discarded logical blocks are
1719 if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1720 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1722 zram->disk->queue->backing_dev_info->capabilities |=
1723 (BDI_CAP_STABLE_WRITES | BDI_CAP_SYNCHRONOUS_IO);
1724 device_add_disk(NULL, zram->disk, zram_disk_attr_groups);
1726 strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1728 zram_debugfs_register(zram);
1729 pr_info("Added device: %s\n", zram->disk->disk_name);
1733 blk_cleanup_queue(queue);
1735 idr_remove(&zram_index_idr, device_id);
1741 static int zram_remove(struct zram *zram)
1743 struct block_device *bdev;
1745 bdev = bdget_disk(zram->disk, 0);
1749 mutex_lock(&bdev->bd_mutex);
1750 if (bdev->bd_openers || zram->claim) {
1751 mutex_unlock(&bdev->bd_mutex);
1757 mutex_unlock(&bdev->bd_mutex);
1759 zram_debugfs_unregister(zram);
1761 /* Make sure all the pending I/O are finished */
1763 zram_reset_device(zram);
1766 pr_info("Removed device: %s\n", zram->disk->disk_name);
1768 del_gendisk(zram->disk);
1769 blk_cleanup_queue(zram->disk->queue);
1770 put_disk(zram->disk);
1775 /* zram-control sysfs attributes */
1778 * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
1779 * sense that reading from this file does alter the state of your system -- it
1780 * creates a new un-initialized zram device and returns back this device's
1781 * device_id (or an error code if it fails to create a new device).
1783 static ssize_t hot_add_show(struct class *class,
1784 struct class_attribute *attr,
1789 mutex_lock(&zram_index_mutex);
1791 mutex_unlock(&zram_index_mutex);
1795 return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1797 static CLASS_ATTR_RO(hot_add);
1799 static ssize_t hot_remove_store(struct class *class,
1800 struct class_attribute *attr,
1807 /* dev_id is gendisk->first_minor, which is `int' */
1808 ret = kstrtoint(buf, 10, &dev_id);
1814 mutex_lock(&zram_index_mutex);
1816 zram = idr_find(&zram_index_idr, dev_id);
1818 ret = zram_remove(zram);
1820 idr_remove(&zram_index_idr, dev_id);
1825 mutex_unlock(&zram_index_mutex);
1826 return ret ? ret : count;
1828 static CLASS_ATTR_WO(hot_remove);
1830 static struct attribute *zram_control_class_attrs[] = {
1831 &class_attr_hot_add.attr,
1832 &class_attr_hot_remove.attr,
1835 ATTRIBUTE_GROUPS(zram_control_class);
1837 static struct class zram_control_class = {
1838 .name = "zram-control",
1839 .owner = THIS_MODULE,
1840 .class_groups = zram_control_class_groups,
1843 static int zram_remove_cb(int id, void *ptr, void *data)
1849 static void destroy_devices(void)
1851 class_unregister(&zram_control_class);
1852 idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1853 zram_debugfs_destroy();
1854 idr_destroy(&zram_index_idr);
1855 unregister_blkdev(zram_major, "zram");
1856 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1859 static int __init zram_init(void)
1863 ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
1864 zcomp_cpu_up_prepare, zcomp_cpu_dead);
1868 ret = class_register(&zram_control_class);
1870 pr_err("Unable to register zram-control class\n");
1871 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1875 zram_debugfs_create();
1876 zram_major = register_blkdev(0, "zram");
1877 if (zram_major <= 0) {
1878 pr_err("Unable to get major number\n");
1879 class_unregister(&zram_control_class);
1880 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1884 while (num_devices != 0) {
1885 mutex_lock(&zram_index_mutex);
1887 mutex_unlock(&zram_index_mutex);
1900 static void __exit zram_exit(void)
1905 module_init(zram_init);
1906 module_exit(zram_exit);
1908 module_param(num_devices, uint, 0);
1909 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1911 MODULE_LICENSE("Dual BSD/GPL");
1912 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1913 MODULE_DESCRIPTION("Compressed RAM Block Device");