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>
36 #include <linux/part_stat.h>
40 static DEFINE_IDR(zram_index_idr);
41 /* idr index must be protected */
42 static DEFINE_MUTEX(zram_index_mutex);
44 static int zram_major;
45 static const char *default_compressor = CONFIG_ZRAM_DEF_COMP;
47 /* Module params (documentation at end) */
48 static unsigned int num_devices = 1;
50 * Pages that compress to sizes equals or greater than this are stored
51 * uncompressed in memory.
53 static size_t huge_class_size;
55 static const struct block_device_operations zram_devops;
56 static const struct block_device_operations zram_wb_devops;
58 static void zram_free_page(struct zram *zram, size_t index);
59 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
60 u32 index, int offset, struct bio *bio);
63 static int zram_slot_trylock(struct zram *zram, u32 index)
65 return bit_spin_trylock(ZRAM_LOCK, &zram->table[index].flags);
68 static void zram_slot_lock(struct zram *zram, u32 index)
70 bit_spin_lock(ZRAM_LOCK, &zram->table[index].flags);
73 static void zram_slot_unlock(struct zram *zram, u32 index)
75 bit_spin_unlock(ZRAM_LOCK, &zram->table[index].flags);
78 static inline bool init_done(struct zram *zram)
80 return zram->disksize;
83 static inline struct zram *dev_to_zram(struct device *dev)
85 return (struct zram *)dev_to_disk(dev)->private_data;
88 static unsigned long zram_get_handle(struct zram *zram, u32 index)
90 return zram->table[index].handle;
93 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
95 zram->table[index].handle = handle;
98 /* flag operations require table entry bit_spin_lock() being held */
99 static bool zram_test_flag(struct zram *zram, u32 index,
100 enum zram_pageflags flag)
102 return zram->table[index].flags & BIT(flag);
105 static void zram_set_flag(struct zram *zram, u32 index,
106 enum zram_pageflags flag)
108 zram->table[index].flags |= BIT(flag);
111 static void zram_clear_flag(struct zram *zram, u32 index,
112 enum zram_pageflags flag)
114 zram->table[index].flags &= ~BIT(flag);
117 static inline void zram_set_element(struct zram *zram, u32 index,
118 unsigned long element)
120 zram->table[index].element = element;
123 static unsigned long zram_get_element(struct zram *zram, u32 index)
125 return zram->table[index].element;
128 static size_t zram_get_obj_size(struct zram *zram, u32 index)
130 return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
133 static void zram_set_obj_size(struct zram *zram,
134 u32 index, size_t size)
136 unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
138 zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
141 static inline bool zram_allocated(struct zram *zram, u32 index)
143 return zram_get_obj_size(zram, index) ||
144 zram_test_flag(zram, index, ZRAM_SAME) ||
145 zram_test_flag(zram, index, ZRAM_WB);
148 #if PAGE_SIZE != 4096
149 static inline bool is_partial_io(struct bio_vec *bvec)
151 return bvec->bv_len != PAGE_SIZE;
154 static inline bool is_partial_io(struct bio_vec *bvec)
161 * Check if request is within bounds and aligned on zram logical blocks.
163 static inline bool valid_io_request(struct zram *zram,
164 sector_t start, unsigned int size)
168 /* unaligned request */
169 if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
171 if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
174 end = start + (size >> SECTOR_SHIFT);
175 bound = zram->disksize >> SECTOR_SHIFT;
176 /* out of range range */
177 if (unlikely(start >= bound || end > bound || start > end))
180 /* I/O request is valid */
184 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
186 *index += (*offset + bvec->bv_len) / PAGE_SIZE;
187 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
190 static inline void update_used_max(struct zram *zram,
191 const unsigned long pages)
193 unsigned long old_max, cur_max;
195 old_max = atomic_long_read(&zram->stats.max_used_pages);
200 old_max = atomic_long_cmpxchg(
201 &zram->stats.max_used_pages, cur_max, pages);
202 } while (old_max != cur_max);
205 static inline void zram_fill_page(void *ptr, unsigned long len,
208 WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
209 memset_l(ptr, value, len / sizeof(unsigned long));
212 static bool page_same_filled(void *ptr, unsigned long *element)
216 unsigned int pos, last_pos = PAGE_SIZE / sizeof(*page) - 1;
218 page = (unsigned long *)ptr;
221 if (val != page[last_pos])
224 for (pos = 1; pos < last_pos; pos++) {
225 if (val != page[pos])
234 static ssize_t initstate_show(struct device *dev,
235 struct device_attribute *attr, char *buf)
238 struct zram *zram = dev_to_zram(dev);
240 down_read(&zram->init_lock);
241 val = init_done(zram);
242 up_read(&zram->init_lock);
244 return scnprintf(buf, PAGE_SIZE, "%u\n", val);
247 static ssize_t disksize_show(struct device *dev,
248 struct device_attribute *attr, char *buf)
250 struct zram *zram = dev_to_zram(dev);
252 return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
255 static ssize_t mem_limit_store(struct device *dev,
256 struct device_attribute *attr, const char *buf, size_t len)
260 struct zram *zram = dev_to_zram(dev);
262 limit = memparse(buf, &tmp);
263 if (buf == tmp) /* no chars parsed, invalid input */
266 down_write(&zram->init_lock);
267 zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
268 up_write(&zram->init_lock);
273 static ssize_t mem_used_max_store(struct device *dev,
274 struct device_attribute *attr, const char *buf, size_t len)
278 struct zram *zram = dev_to_zram(dev);
280 err = kstrtoul(buf, 10, &val);
284 down_read(&zram->init_lock);
285 if (init_done(zram)) {
286 atomic_long_set(&zram->stats.max_used_pages,
287 zs_get_total_pages(zram->mem_pool));
289 up_read(&zram->init_lock);
295 * Mark all pages which are older than or equal to cutoff as IDLE.
296 * Callers should hold the zram init lock in read mode
298 static void mark_idle(struct zram *zram, ktime_t cutoff)
301 unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
304 for (index = 0; index < nr_pages; index++) {
306 * Do not mark ZRAM_UNDER_WB slot as ZRAM_IDLE to close race.
307 * See the comment in writeback_store.
309 zram_slot_lock(zram, index);
310 if (zram_allocated(zram, index) &&
311 !zram_test_flag(zram, index, ZRAM_UNDER_WB)) {
312 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
313 is_idle = !cutoff || ktime_after(cutoff, zram->table[index].ac_time);
316 zram_set_flag(zram, index, ZRAM_IDLE);
318 zram_slot_unlock(zram, index);
322 static ssize_t idle_store(struct device *dev,
323 struct device_attribute *attr, const char *buf, size_t len)
325 struct zram *zram = dev_to_zram(dev);
326 ktime_t cutoff_time = 0;
327 ssize_t rv = -EINVAL;
329 if (!sysfs_streq(buf, "all")) {
331 * If it did not parse as 'all' try to treat it as an integer when
332 * we have memory tracking enabled.
336 if (IS_ENABLED(CONFIG_ZRAM_MEMORY_TRACKING) && !kstrtoull(buf, 0, &age_sec))
337 cutoff_time = ktime_sub(ktime_get_boottime(),
338 ns_to_ktime(age_sec * NSEC_PER_SEC));
343 down_read(&zram->init_lock);
344 if (!init_done(zram))
347 /* A cutoff_time of 0 marks everything as idle, this is the "all" behavior */
348 mark_idle(zram, cutoff_time);
352 up_read(&zram->init_lock);
357 #ifdef CONFIG_ZRAM_WRITEBACK
358 static ssize_t writeback_limit_enable_store(struct device *dev,
359 struct device_attribute *attr, const char *buf, size_t len)
361 struct zram *zram = dev_to_zram(dev);
363 ssize_t ret = -EINVAL;
365 if (kstrtoull(buf, 10, &val))
368 down_read(&zram->init_lock);
369 spin_lock(&zram->wb_limit_lock);
370 zram->wb_limit_enable = val;
371 spin_unlock(&zram->wb_limit_lock);
372 up_read(&zram->init_lock);
378 static ssize_t writeback_limit_enable_show(struct device *dev,
379 struct device_attribute *attr, char *buf)
382 struct zram *zram = dev_to_zram(dev);
384 down_read(&zram->init_lock);
385 spin_lock(&zram->wb_limit_lock);
386 val = zram->wb_limit_enable;
387 spin_unlock(&zram->wb_limit_lock);
388 up_read(&zram->init_lock);
390 return scnprintf(buf, PAGE_SIZE, "%d\n", val);
393 static ssize_t writeback_limit_store(struct device *dev,
394 struct device_attribute *attr, const char *buf, size_t len)
396 struct zram *zram = dev_to_zram(dev);
398 ssize_t ret = -EINVAL;
400 if (kstrtoull(buf, 10, &val))
403 down_read(&zram->init_lock);
404 spin_lock(&zram->wb_limit_lock);
405 zram->bd_wb_limit = val;
406 spin_unlock(&zram->wb_limit_lock);
407 up_read(&zram->init_lock);
413 static ssize_t writeback_limit_show(struct device *dev,
414 struct device_attribute *attr, char *buf)
417 struct zram *zram = dev_to_zram(dev);
419 down_read(&zram->init_lock);
420 spin_lock(&zram->wb_limit_lock);
421 val = zram->bd_wb_limit;
422 spin_unlock(&zram->wb_limit_lock);
423 up_read(&zram->init_lock);
425 return scnprintf(buf, PAGE_SIZE, "%llu\n", val);
428 static void reset_bdev(struct zram *zram)
430 struct block_device *bdev;
432 if (!zram->backing_dev)
436 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
437 /* hope filp_close flush all of IO */
438 filp_close(zram->backing_dev, NULL);
439 zram->backing_dev = NULL;
441 zram->disk->fops = &zram_devops;
442 kvfree(zram->bitmap);
446 static ssize_t backing_dev_show(struct device *dev,
447 struct device_attribute *attr, char *buf)
450 struct zram *zram = dev_to_zram(dev);
454 down_read(&zram->init_lock);
455 file = zram->backing_dev;
457 memcpy(buf, "none\n", 5);
458 up_read(&zram->init_lock);
462 p = file_path(file, buf, PAGE_SIZE - 1);
469 memmove(buf, p, ret);
472 up_read(&zram->init_lock);
476 static ssize_t backing_dev_store(struct device *dev,
477 struct device_attribute *attr, const char *buf, size_t len)
481 struct file *backing_dev = NULL;
483 struct address_space *mapping;
484 unsigned int bitmap_sz;
485 unsigned long nr_pages, *bitmap = NULL;
486 struct block_device *bdev = NULL;
488 struct zram *zram = dev_to_zram(dev);
490 file_name = kmalloc(PATH_MAX, GFP_KERNEL);
494 down_write(&zram->init_lock);
495 if (init_done(zram)) {
496 pr_info("Can't setup backing device for initialized device\n");
501 strlcpy(file_name, buf, PATH_MAX);
502 /* ignore trailing newline */
503 sz = strlen(file_name);
504 if (sz > 0 && file_name[sz - 1] == '\n')
505 file_name[sz - 1] = 0x00;
507 backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
508 if (IS_ERR(backing_dev)) {
509 err = PTR_ERR(backing_dev);
514 mapping = backing_dev->f_mapping;
515 inode = mapping->host;
517 /* Support only block device in this moment */
518 if (!S_ISBLK(inode->i_mode)) {
523 bdev = blkdev_get_by_dev(inode->i_rdev,
524 FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
531 nr_pages = i_size_read(inode) >> PAGE_SHIFT;
532 bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
533 bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
542 zram->backing_dev = backing_dev;
543 zram->bitmap = bitmap;
544 zram->nr_pages = nr_pages;
546 * With writeback feature, zram does asynchronous IO so it's no longer
547 * synchronous device so let's remove synchronous io flag. Othewise,
548 * upper layer(e.g., swap) could wait IO completion rather than
549 * (submit and return), which will cause system sluggish.
550 * Furthermore, when the IO function returns(e.g., swap_readpage),
551 * upper layer expects IO was done so it could deallocate the page
552 * freely but in fact, IO is going on so finally could cause
553 * use-after-free when the IO is really done.
555 zram->disk->fops = &zram_wb_devops;
556 up_write(&zram->init_lock);
558 pr_info("setup backing device %s\n", file_name);
566 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
569 filp_close(backing_dev, NULL);
571 up_write(&zram->init_lock);
578 static unsigned long alloc_block_bdev(struct zram *zram)
580 unsigned long blk_idx = 1;
582 /* skip 0 bit to confuse zram.handle = 0 */
583 blk_idx = find_next_zero_bit(zram->bitmap, zram->nr_pages, blk_idx);
584 if (blk_idx == zram->nr_pages)
587 if (test_and_set_bit(blk_idx, zram->bitmap))
590 atomic64_inc(&zram->stats.bd_count);
594 static void free_block_bdev(struct zram *zram, unsigned long blk_idx)
598 was_set = test_and_clear_bit(blk_idx, zram->bitmap);
599 WARN_ON_ONCE(!was_set);
600 atomic64_dec(&zram->stats.bd_count);
603 static void zram_page_end_io(struct bio *bio)
605 struct page *page = bio_first_page_all(bio);
607 page_endio(page, op_is_write(bio_op(bio)),
608 blk_status_to_errno(bio->bi_status));
613 * Returns 1 if the submission is successful.
615 static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
616 unsigned long entry, struct bio *parent)
620 bio = bio_alloc(GFP_NOIO, 1);
624 bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
625 bio_set_dev(bio, zram->bdev);
626 if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
632 bio->bi_opf = REQ_OP_READ;
633 bio->bi_end_io = zram_page_end_io;
635 bio->bi_opf = parent->bi_opf;
636 bio_chain(bio, parent);
643 #define PAGE_WB_SIG "page_index="
645 #define PAGE_WRITEBACK 0
646 #define HUGE_WRITEBACK 1
647 #define IDLE_WRITEBACK 2
650 static ssize_t writeback_store(struct device *dev,
651 struct device_attribute *attr, const char *buf, size_t len)
653 struct zram *zram = dev_to_zram(dev);
654 unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
655 unsigned long index = 0;
657 struct bio_vec bio_vec;
661 unsigned long blk_idx = 0;
663 if (sysfs_streq(buf, "idle"))
664 mode = IDLE_WRITEBACK;
665 else if (sysfs_streq(buf, "huge"))
666 mode = HUGE_WRITEBACK;
668 if (strncmp(buf, PAGE_WB_SIG, sizeof(PAGE_WB_SIG) - 1))
671 if (kstrtol(buf + sizeof(PAGE_WB_SIG) - 1, 10, &index) ||
676 mode = PAGE_WRITEBACK;
679 down_read(&zram->init_lock);
680 if (!init_done(zram)) {
682 goto release_init_lock;
685 if (!zram->backing_dev) {
687 goto release_init_lock;
690 page = alloc_page(GFP_KERNEL);
693 goto release_init_lock;
696 for (; nr_pages != 0; index++, nr_pages--) {
700 bvec.bv_len = PAGE_SIZE;
703 spin_lock(&zram->wb_limit_lock);
704 if (zram->wb_limit_enable && !zram->bd_wb_limit) {
705 spin_unlock(&zram->wb_limit_lock);
709 spin_unlock(&zram->wb_limit_lock);
712 blk_idx = alloc_block_bdev(zram);
719 zram_slot_lock(zram, index);
720 if (!zram_allocated(zram, index))
723 if (zram_test_flag(zram, index, ZRAM_WB) ||
724 zram_test_flag(zram, index, ZRAM_SAME) ||
725 zram_test_flag(zram, index, ZRAM_UNDER_WB))
728 if (mode == IDLE_WRITEBACK &&
729 !zram_test_flag(zram, index, ZRAM_IDLE))
731 if (mode == HUGE_WRITEBACK &&
732 !zram_test_flag(zram, index, ZRAM_HUGE))
735 * Clearing ZRAM_UNDER_WB is duty of caller.
736 * IOW, zram_free_page never clear it.
738 zram_set_flag(zram, index, ZRAM_UNDER_WB);
739 /* Need for hugepage writeback racing */
740 zram_set_flag(zram, index, ZRAM_IDLE);
741 zram_slot_unlock(zram, index);
742 if (zram_bvec_read(zram, &bvec, index, 0, NULL)) {
743 zram_slot_lock(zram, index);
744 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
745 zram_clear_flag(zram, index, ZRAM_IDLE);
746 zram_slot_unlock(zram, index);
750 bio_init(&bio, &bio_vec, 1);
751 bio_set_dev(&bio, zram->bdev);
752 bio.bi_iter.bi_sector = blk_idx * (PAGE_SIZE >> 9);
753 bio.bi_opf = REQ_OP_WRITE | REQ_SYNC;
755 bio_add_page(&bio, bvec.bv_page, bvec.bv_len,
758 * XXX: A single page IO would be inefficient for write
759 * but it would be not bad as starter.
761 err = submit_bio_wait(&bio);
763 zram_slot_lock(zram, index);
764 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
765 zram_clear_flag(zram, index, ZRAM_IDLE);
766 zram_slot_unlock(zram, index);
768 * Return last IO error unless every IO were
775 atomic64_inc(&zram->stats.bd_writes);
777 * We released zram_slot_lock so need to check if the slot was
778 * changed. If there is freeing for the slot, we can catch it
779 * easily by zram_allocated.
780 * A subtle case is the slot is freed/reallocated/marked as
781 * ZRAM_IDLE again. To close the race, idle_store doesn't
782 * mark ZRAM_IDLE once it found the slot was ZRAM_UNDER_WB.
783 * Thus, we could close the race by checking ZRAM_IDLE bit.
785 zram_slot_lock(zram, index);
786 if (!zram_allocated(zram, index) ||
787 !zram_test_flag(zram, index, ZRAM_IDLE)) {
788 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
789 zram_clear_flag(zram, index, ZRAM_IDLE);
793 zram_free_page(zram, index);
794 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
795 zram_set_flag(zram, index, ZRAM_WB);
796 zram_set_element(zram, index, blk_idx);
798 atomic64_inc(&zram->stats.pages_stored);
799 spin_lock(&zram->wb_limit_lock);
800 if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
801 zram->bd_wb_limit -= 1UL << (PAGE_SHIFT - 12);
802 spin_unlock(&zram->wb_limit_lock);
804 zram_slot_unlock(zram, index);
808 free_block_bdev(zram, blk_idx);
811 up_read(&zram->init_lock);
817 struct work_struct work;
824 #if PAGE_SIZE != 4096
825 static void zram_sync_read(struct work_struct *work)
827 struct zram_work *zw = container_of(work, struct zram_work, work);
828 struct zram *zram = zw->zram;
829 unsigned long entry = zw->entry;
830 struct bio *bio = zw->bio;
832 read_from_bdev_async(zram, &zw->bvec, entry, bio);
836 * Block layer want one ->submit_bio to be active at a time, so if we use
837 * chained IO with parent IO in same context, it's a deadlock. To avoid that,
838 * use a worker thread context.
840 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
841 unsigned long entry, struct bio *bio)
843 struct zram_work work;
850 INIT_WORK_ONSTACK(&work.work, zram_sync_read);
851 queue_work(system_unbound_wq, &work.work);
852 flush_work(&work.work);
853 destroy_work_on_stack(&work.work);
858 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
859 unsigned long entry, struct bio *bio)
866 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
867 unsigned long entry, struct bio *parent, bool sync)
869 atomic64_inc(&zram->stats.bd_reads);
871 return read_from_bdev_sync(zram, bvec, entry, parent);
873 return read_from_bdev_async(zram, bvec, entry, parent);
876 static inline void reset_bdev(struct zram *zram) {};
877 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
878 unsigned long entry, struct bio *parent, bool sync)
883 static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
886 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
888 static struct dentry *zram_debugfs_root;
890 static void zram_debugfs_create(void)
892 zram_debugfs_root = debugfs_create_dir("zram", NULL);
895 static void zram_debugfs_destroy(void)
897 debugfs_remove_recursive(zram_debugfs_root);
900 static void zram_accessed(struct zram *zram, u32 index)
902 zram_clear_flag(zram, index, ZRAM_IDLE);
903 zram->table[index].ac_time = ktime_get_boottime();
906 static ssize_t read_block_state(struct file *file, char __user *buf,
907 size_t count, loff_t *ppos)
910 ssize_t index, written = 0;
911 struct zram *zram = file->private_data;
912 unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
913 struct timespec64 ts;
915 kbuf = kvmalloc(count, GFP_KERNEL);
919 down_read(&zram->init_lock);
920 if (!init_done(zram)) {
921 up_read(&zram->init_lock);
926 for (index = *ppos; index < nr_pages; index++) {
929 zram_slot_lock(zram, index);
930 if (!zram_allocated(zram, index))
933 ts = ktime_to_timespec64(zram->table[index].ac_time);
934 copied = snprintf(kbuf + written, count,
935 "%12zd %12lld.%06lu %c%c%c%c\n",
936 index, (s64)ts.tv_sec,
937 ts.tv_nsec / NSEC_PER_USEC,
938 zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
939 zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
940 zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.',
941 zram_test_flag(zram, index, ZRAM_IDLE) ? 'i' : '.');
943 if (count <= copied) {
944 zram_slot_unlock(zram, index);
950 zram_slot_unlock(zram, index);
954 up_read(&zram->init_lock);
955 if (copy_to_user(buf, kbuf, written))
962 static const struct file_operations proc_zram_block_state_op = {
964 .read = read_block_state,
965 .llseek = default_llseek,
968 static void zram_debugfs_register(struct zram *zram)
970 if (!zram_debugfs_root)
973 zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
975 debugfs_create_file("block_state", 0400, zram->debugfs_dir,
976 zram, &proc_zram_block_state_op);
979 static void zram_debugfs_unregister(struct zram *zram)
981 debugfs_remove_recursive(zram->debugfs_dir);
984 static void zram_debugfs_create(void) {};
985 static void zram_debugfs_destroy(void) {};
986 static void zram_accessed(struct zram *zram, u32 index)
988 zram_clear_flag(zram, index, ZRAM_IDLE);
990 static void zram_debugfs_register(struct zram *zram) {};
991 static void zram_debugfs_unregister(struct zram *zram) {};
995 * We switched to per-cpu streams and this attr is not needed anymore.
996 * However, we will keep it around for some time, because:
997 * a) we may revert per-cpu streams in the future
998 * b) it's visible to user space and we need to follow our 2 years
999 * retirement rule; but we already have a number of 'soon to be
1000 * altered' attrs, so max_comp_streams need to wait for the next
1003 static ssize_t max_comp_streams_show(struct device *dev,
1004 struct device_attribute *attr, char *buf)
1006 return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
1009 static ssize_t max_comp_streams_store(struct device *dev,
1010 struct device_attribute *attr, const char *buf, size_t len)
1015 static ssize_t comp_algorithm_show(struct device *dev,
1016 struct device_attribute *attr, char *buf)
1019 struct zram *zram = dev_to_zram(dev);
1021 down_read(&zram->init_lock);
1022 sz = zcomp_available_show(zram->compressor, buf);
1023 up_read(&zram->init_lock);
1028 static ssize_t comp_algorithm_store(struct device *dev,
1029 struct device_attribute *attr, const char *buf, size_t len)
1031 struct zram *zram = dev_to_zram(dev);
1032 char compressor[ARRAY_SIZE(zram->compressor)];
1035 strlcpy(compressor, buf, sizeof(compressor));
1036 /* ignore trailing newline */
1037 sz = strlen(compressor);
1038 if (sz > 0 && compressor[sz - 1] == '\n')
1039 compressor[sz - 1] = 0x00;
1041 if (!zcomp_available_algorithm(compressor))
1044 down_write(&zram->init_lock);
1045 if (init_done(zram)) {
1046 up_write(&zram->init_lock);
1047 pr_info("Can't change algorithm for initialized device\n");
1051 strcpy(zram->compressor, compressor);
1052 up_write(&zram->init_lock);
1056 static ssize_t compact_store(struct device *dev,
1057 struct device_attribute *attr, const char *buf, size_t len)
1059 struct zram *zram = dev_to_zram(dev);
1061 down_read(&zram->init_lock);
1062 if (!init_done(zram)) {
1063 up_read(&zram->init_lock);
1067 zs_compact(zram->mem_pool);
1068 up_read(&zram->init_lock);
1073 static ssize_t io_stat_show(struct device *dev,
1074 struct device_attribute *attr, char *buf)
1076 struct zram *zram = dev_to_zram(dev);
1079 down_read(&zram->init_lock);
1080 ret = scnprintf(buf, PAGE_SIZE,
1081 "%8llu %8llu %8llu %8llu\n",
1082 (u64)atomic64_read(&zram->stats.failed_reads),
1083 (u64)atomic64_read(&zram->stats.failed_writes),
1084 (u64)atomic64_read(&zram->stats.invalid_io),
1085 (u64)atomic64_read(&zram->stats.notify_free));
1086 up_read(&zram->init_lock);
1091 static ssize_t mm_stat_show(struct device *dev,
1092 struct device_attribute *attr, char *buf)
1094 struct zram *zram = dev_to_zram(dev);
1095 struct zs_pool_stats pool_stats;
1096 u64 orig_size, mem_used = 0;
1100 memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
1102 down_read(&zram->init_lock);
1103 if (init_done(zram)) {
1104 mem_used = zs_get_total_pages(zram->mem_pool);
1105 zs_pool_stats(zram->mem_pool, &pool_stats);
1108 orig_size = atomic64_read(&zram->stats.pages_stored);
1109 max_used = atomic_long_read(&zram->stats.max_used_pages);
1111 ret = scnprintf(buf, PAGE_SIZE,
1112 "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu %8llu\n",
1113 orig_size << PAGE_SHIFT,
1114 (u64)atomic64_read(&zram->stats.compr_data_size),
1115 mem_used << PAGE_SHIFT,
1116 zram->limit_pages << PAGE_SHIFT,
1117 max_used << PAGE_SHIFT,
1118 (u64)atomic64_read(&zram->stats.same_pages),
1119 atomic_long_read(&pool_stats.pages_compacted),
1120 (u64)atomic64_read(&zram->stats.huge_pages),
1121 (u64)atomic64_read(&zram->stats.huge_pages_since));
1122 up_read(&zram->init_lock);
1127 #ifdef CONFIG_ZRAM_WRITEBACK
1128 #define FOUR_K(x) ((x) * (1 << (PAGE_SHIFT - 12)))
1129 static ssize_t bd_stat_show(struct device *dev,
1130 struct device_attribute *attr, char *buf)
1132 struct zram *zram = dev_to_zram(dev);
1135 down_read(&zram->init_lock);
1136 ret = scnprintf(buf, PAGE_SIZE,
1137 "%8llu %8llu %8llu\n",
1138 FOUR_K((u64)atomic64_read(&zram->stats.bd_count)),
1139 FOUR_K((u64)atomic64_read(&zram->stats.bd_reads)),
1140 FOUR_K((u64)atomic64_read(&zram->stats.bd_writes)));
1141 up_read(&zram->init_lock);
1147 static ssize_t debug_stat_show(struct device *dev,
1148 struct device_attribute *attr, char *buf)
1151 struct zram *zram = dev_to_zram(dev);
1154 down_read(&zram->init_lock);
1155 ret = scnprintf(buf, PAGE_SIZE,
1156 "version: %d\n%8llu %8llu\n",
1158 (u64)atomic64_read(&zram->stats.writestall),
1159 (u64)atomic64_read(&zram->stats.miss_free));
1160 up_read(&zram->init_lock);
1165 static DEVICE_ATTR_RO(io_stat);
1166 static DEVICE_ATTR_RO(mm_stat);
1167 #ifdef CONFIG_ZRAM_WRITEBACK
1168 static DEVICE_ATTR_RO(bd_stat);
1170 static DEVICE_ATTR_RO(debug_stat);
1172 static void zram_meta_free(struct zram *zram, u64 disksize)
1174 size_t num_pages = disksize >> PAGE_SHIFT;
1177 /* Free all pages that are still in this zram device */
1178 for (index = 0; index < num_pages; index++)
1179 zram_free_page(zram, index);
1181 zs_destroy_pool(zram->mem_pool);
1185 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
1189 num_pages = disksize >> PAGE_SHIFT;
1190 zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
1194 zram->mem_pool = zs_create_pool(zram->disk->disk_name);
1195 if (!zram->mem_pool) {
1200 if (!huge_class_size)
1201 huge_class_size = zs_huge_class_size(zram->mem_pool);
1206 * To protect concurrent access to the same index entry,
1207 * caller should hold this table index entry's bit_spinlock to
1208 * indicate this index entry is accessing.
1210 static void zram_free_page(struct zram *zram, size_t index)
1212 unsigned long handle;
1214 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
1215 zram->table[index].ac_time = 0;
1217 if (zram_test_flag(zram, index, ZRAM_IDLE))
1218 zram_clear_flag(zram, index, ZRAM_IDLE);
1220 if (zram_test_flag(zram, index, ZRAM_HUGE)) {
1221 zram_clear_flag(zram, index, ZRAM_HUGE);
1222 atomic64_dec(&zram->stats.huge_pages);
1225 if (zram_test_flag(zram, index, ZRAM_WB)) {
1226 zram_clear_flag(zram, index, ZRAM_WB);
1227 free_block_bdev(zram, zram_get_element(zram, index));
1232 * No memory is allocated for same element filled pages.
1233 * Simply clear same page flag.
1235 if (zram_test_flag(zram, index, ZRAM_SAME)) {
1236 zram_clear_flag(zram, index, ZRAM_SAME);
1237 atomic64_dec(&zram->stats.same_pages);
1241 handle = zram_get_handle(zram, index);
1245 zs_free(zram->mem_pool, handle);
1247 atomic64_sub(zram_get_obj_size(zram, index),
1248 &zram->stats.compr_data_size);
1250 atomic64_dec(&zram->stats.pages_stored);
1251 zram_set_handle(zram, index, 0);
1252 zram_set_obj_size(zram, index, 0);
1253 WARN_ON_ONCE(zram->table[index].flags &
1254 ~(1UL << ZRAM_LOCK | 1UL << ZRAM_UNDER_WB));
1257 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
1258 struct bio *bio, bool partial_io)
1260 struct zcomp_strm *zstrm;
1261 unsigned long handle;
1266 zram_slot_lock(zram, index);
1267 if (zram_test_flag(zram, index, ZRAM_WB)) {
1268 struct bio_vec bvec;
1270 zram_slot_unlock(zram, index);
1272 bvec.bv_page = page;
1273 bvec.bv_len = PAGE_SIZE;
1275 return read_from_bdev(zram, &bvec,
1276 zram_get_element(zram, index),
1280 handle = zram_get_handle(zram, index);
1281 if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1282 unsigned long value;
1285 value = handle ? zram_get_element(zram, index) : 0;
1286 mem = kmap_atomic(page);
1287 zram_fill_page(mem, PAGE_SIZE, value);
1289 zram_slot_unlock(zram, index);
1293 size = zram_get_obj_size(zram, index);
1295 if (size != PAGE_SIZE)
1296 zstrm = zcomp_stream_get(zram->comp);
1298 src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1299 if (size == PAGE_SIZE) {
1300 dst = kmap_atomic(page);
1301 memcpy(dst, src, PAGE_SIZE);
1305 dst = kmap_atomic(page);
1306 ret = zcomp_decompress(zstrm, src, size, dst);
1308 zcomp_stream_put(zram->comp);
1310 zs_unmap_object(zram->mem_pool, handle);
1311 zram_slot_unlock(zram, index);
1313 /* Should NEVER happen. Return bio error if it does. */
1315 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1320 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1321 u32 index, int offset, struct bio *bio)
1326 page = bvec->bv_page;
1327 if (is_partial_io(bvec)) {
1328 /* Use a temporary buffer to decompress the page */
1329 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1334 ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
1338 if (is_partial_io(bvec)) {
1339 void *dst = kmap_atomic(bvec->bv_page);
1340 void *src = kmap_atomic(page);
1342 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
1347 if (is_partial_io(bvec))
1353 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1354 u32 index, struct bio *bio)
1357 unsigned long alloced_pages;
1358 unsigned long handle = 0;
1359 unsigned int comp_len = 0;
1360 void *src, *dst, *mem;
1361 struct zcomp_strm *zstrm;
1362 struct page *page = bvec->bv_page;
1363 unsigned long element = 0;
1364 enum zram_pageflags flags = 0;
1366 mem = kmap_atomic(page);
1367 if (page_same_filled(mem, &element)) {
1369 /* Free memory associated with this sector now. */
1371 atomic64_inc(&zram->stats.same_pages);
1377 zstrm = zcomp_stream_get(zram->comp);
1378 src = kmap_atomic(page);
1379 ret = zcomp_compress(zstrm, src, &comp_len);
1382 if (unlikely(ret)) {
1383 zcomp_stream_put(zram->comp);
1384 pr_err("Compression failed! err=%d\n", ret);
1385 zs_free(zram->mem_pool, handle);
1389 if (comp_len >= huge_class_size)
1390 comp_len = PAGE_SIZE;
1392 * handle allocation has 2 paths:
1393 * a) fast path is executed with preemption disabled (for
1394 * per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1395 * since we can't sleep;
1396 * b) slow path enables preemption and attempts to allocate
1397 * the page with __GFP_DIRECT_RECLAIM bit set. we have to
1398 * put per-cpu compression stream and, thus, to re-do
1399 * the compression once handle is allocated.
1401 * if we have a 'non-null' handle here then we are coming
1402 * from the slow path and handle has already been allocated.
1405 handle = zs_malloc(zram->mem_pool, comp_len,
1406 __GFP_KSWAPD_RECLAIM |
1411 zcomp_stream_put(zram->comp);
1412 atomic64_inc(&zram->stats.writestall);
1413 handle = zs_malloc(zram->mem_pool, comp_len,
1414 GFP_NOIO | __GFP_HIGHMEM |
1417 goto compress_again;
1421 alloced_pages = zs_get_total_pages(zram->mem_pool);
1422 update_used_max(zram, alloced_pages);
1424 if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1425 zcomp_stream_put(zram->comp);
1426 zs_free(zram->mem_pool, handle);
1430 dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1432 src = zstrm->buffer;
1433 if (comp_len == PAGE_SIZE)
1434 src = kmap_atomic(page);
1435 memcpy(dst, src, comp_len);
1436 if (comp_len == PAGE_SIZE)
1439 zcomp_stream_put(zram->comp);
1440 zs_unmap_object(zram->mem_pool, handle);
1441 atomic64_add(comp_len, &zram->stats.compr_data_size);
1444 * Free memory associated with this sector
1445 * before overwriting unused sectors.
1447 zram_slot_lock(zram, index);
1448 zram_free_page(zram, index);
1450 if (comp_len == PAGE_SIZE) {
1451 zram_set_flag(zram, index, ZRAM_HUGE);
1452 atomic64_inc(&zram->stats.huge_pages);
1453 atomic64_inc(&zram->stats.huge_pages_since);
1457 zram_set_flag(zram, index, flags);
1458 zram_set_element(zram, index, element);
1460 zram_set_handle(zram, index, handle);
1461 zram_set_obj_size(zram, index, comp_len);
1463 zram_slot_unlock(zram, index);
1466 atomic64_inc(&zram->stats.pages_stored);
1470 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1471 u32 index, int offset, struct bio *bio)
1474 struct page *page = NULL;
1479 if (is_partial_io(bvec)) {
1482 * This is a partial IO. We need to read the full page
1483 * before to write the changes.
1485 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1489 ret = __zram_bvec_read(zram, page, index, bio, true);
1493 src = kmap_atomic(bvec->bv_page);
1494 dst = kmap_atomic(page);
1495 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1500 vec.bv_len = PAGE_SIZE;
1504 ret = __zram_bvec_write(zram, &vec, index, bio);
1506 if (is_partial_io(bvec))
1512 * zram_bio_discard - handler on discard request
1513 * @index: physical block index in PAGE_SIZE units
1514 * @offset: byte offset within physical block
1516 static void zram_bio_discard(struct zram *zram, u32 index,
1517 int offset, struct bio *bio)
1519 size_t n = bio->bi_iter.bi_size;
1522 * zram manages data in physical block size units. Because logical block
1523 * size isn't identical with physical block size on some arch, we
1524 * could get a discard request pointing to a specific offset within a
1525 * certain physical block. Although we can handle this request by
1526 * reading that physiclal block and decompressing and partially zeroing
1527 * and re-compressing and then re-storing it, this isn't reasonable
1528 * because our intent with a discard request is to save memory. So
1529 * skipping this logical block is appropriate here.
1532 if (n <= (PAGE_SIZE - offset))
1535 n -= (PAGE_SIZE - offset);
1539 while (n >= PAGE_SIZE) {
1540 zram_slot_lock(zram, index);
1541 zram_free_page(zram, index);
1542 zram_slot_unlock(zram, index);
1543 atomic64_inc(&zram->stats.notify_free);
1550 * Returns errno if it has some problem. Otherwise return 0 or 1.
1551 * Returns 0 if IO request was done synchronously
1552 * Returns 1 if IO request was successfully submitted.
1554 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1555 int offset, unsigned int op, struct bio *bio)
1559 if (!op_is_write(op)) {
1560 atomic64_inc(&zram->stats.num_reads);
1561 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1562 flush_dcache_page(bvec->bv_page);
1564 atomic64_inc(&zram->stats.num_writes);
1565 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1568 zram_slot_lock(zram, index);
1569 zram_accessed(zram, index);
1570 zram_slot_unlock(zram, index);
1572 if (unlikely(ret < 0)) {
1573 if (!op_is_write(op))
1574 atomic64_inc(&zram->stats.failed_reads);
1576 atomic64_inc(&zram->stats.failed_writes);
1582 static void __zram_make_request(struct zram *zram, struct bio *bio)
1586 struct bio_vec bvec;
1587 struct bvec_iter iter;
1588 unsigned long start_time;
1590 index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1591 offset = (bio->bi_iter.bi_sector &
1592 (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1594 switch (bio_op(bio)) {
1595 case REQ_OP_DISCARD:
1596 case REQ_OP_WRITE_ZEROES:
1597 zram_bio_discard(zram, index, offset, bio);
1604 start_time = bio_start_io_acct(bio);
1605 bio_for_each_segment(bvec, bio, iter) {
1606 struct bio_vec bv = bvec;
1607 unsigned int unwritten = bvec.bv_len;
1610 bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1612 if (zram_bvec_rw(zram, &bv, index, offset,
1613 bio_op(bio), bio) < 0) {
1614 bio->bi_status = BLK_STS_IOERR;
1618 bv.bv_offset += bv.bv_len;
1619 unwritten -= bv.bv_len;
1621 update_position(&index, &offset, &bv);
1622 } while (unwritten);
1624 bio_end_io_acct(bio, start_time);
1629 * Handler function for all zram I/O requests.
1631 static void zram_submit_bio(struct bio *bio)
1633 struct zram *zram = bio->bi_bdev->bd_disk->private_data;
1635 if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1636 bio->bi_iter.bi_size)) {
1637 atomic64_inc(&zram->stats.invalid_io);
1642 __zram_make_request(zram, bio);
1645 static void zram_slot_free_notify(struct block_device *bdev,
1646 unsigned long index)
1650 zram = bdev->bd_disk->private_data;
1652 atomic64_inc(&zram->stats.notify_free);
1653 if (!zram_slot_trylock(zram, index)) {
1654 atomic64_inc(&zram->stats.miss_free);
1658 zram_free_page(zram, index);
1659 zram_slot_unlock(zram, index);
1662 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1663 struct page *page, unsigned int op)
1669 unsigned long start_time;
1671 if (PageTransHuge(page))
1673 zram = bdev->bd_disk->private_data;
1675 if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1676 atomic64_inc(&zram->stats.invalid_io);
1681 index = sector >> SECTORS_PER_PAGE_SHIFT;
1682 offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1685 bv.bv_len = PAGE_SIZE;
1688 start_time = disk_start_io_acct(bdev->bd_disk, SECTORS_PER_PAGE, op);
1689 ret = zram_bvec_rw(zram, &bv, index, offset, op, NULL);
1690 disk_end_io_acct(bdev->bd_disk, op, start_time);
1693 * If I/O fails, just return error(ie, non-zero) without
1694 * calling page_endio.
1695 * It causes resubmit the I/O with bio request by upper functions
1696 * of rw_page(e.g., swap_readpage, __swap_writepage) and
1697 * bio->bi_end_io does things to handle the error
1698 * (e.g., SetPageError, set_page_dirty and extra works).
1700 if (unlikely(ret < 0))
1705 page_endio(page, op_is_write(op), 0);
1716 static void zram_reset_device(struct zram *zram)
1721 down_write(&zram->init_lock);
1723 zram->limit_pages = 0;
1725 if (!init_done(zram)) {
1726 up_write(&zram->init_lock);
1731 disksize = zram->disksize;
1734 set_capacity_and_notify(zram->disk, 0);
1735 part_stat_set_all(zram->disk->part0, 0);
1737 /* I/O operation under all of CPU are done so let's free */
1738 zram_meta_free(zram, disksize);
1739 memset(&zram->stats, 0, sizeof(zram->stats));
1740 zcomp_destroy(comp);
1743 up_write(&zram->init_lock);
1746 static ssize_t disksize_store(struct device *dev,
1747 struct device_attribute *attr, const char *buf, size_t len)
1751 struct zram *zram = dev_to_zram(dev);
1754 disksize = memparse(buf, NULL);
1758 down_write(&zram->init_lock);
1759 if (init_done(zram)) {
1760 pr_info("Cannot change disksize for initialized device\n");
1765 disksize = PAGE_ALIGN(disksize);
1766 if (!zram_meta_alloc(zram, disksize)) {
1771 comp = zcomp_create(zram->compressor);
1773 pr_err("Cannot initialise %s compressing backend\n",
1775 err = PTR_ERR(comp);
1780 zram->disksize = disksize;
1781 set_capacity_and_notify(zram->disk, zram->disksize >> SECTOR_SHIFT);
1782 up_write(&zram->init_lock);
1787 zram_meta_free(zram, disksize);
1789 up_write(&zram->init_lock);
1793 static ssize_t reset_store(struct device *dev,
1794 struct device_attribute *attr, const char *buf, size_t len)
1797 unsigned short do_reset;
1799 struct block_device *bdev;
1801 ret = kstrtou16(buf, 10, &do_reset);
1808 zram = dev_to_zram(dev);
1809 bdev = zram->disk->part0;
1811 mutex_lock(&bdev->bd_disk->open_mutex);
1812 /* Do not reset an active device or claimed device */
1813 if (bdev->bd_openers || zram->claim) {
1814 mutex_unlock(&bdev->bd_disk->open_mutex);
1818 /* From now on, anyone can't open /dev/zram[0-9] */
1820 mutex_unlock(&bdev->bd_disk->open_mutex);
1822 /* Make sure all the pending I/O are finished */
1823 sync_blockdev(bdev);
1824 zram_reset_device(zram);
1826 mutex_lock(&bdev->bd_disk->open_mutex);
1827 zram->claim = false;
1828 mutex_unlock(&bdev->bd_disk->open_mutex);
1833 static int zram_open(struct block_device *bdev, fmode_t mode)
1838 WARN_ON(!mutex_is_locked(&bdev->bd_disk->open_mutex));
1840 zram = bdev->bd_disk->private_data;
1841 /* zram was claimed to reset so open request fails */
1848 static const struct block_device_operations zram_devops = {
1850 .submit_bio = zram_submit_bio,
1851 .swap_slot_free_notify = zram_slot_free_notify,
1852 .rw_page = zram_rw_page,
1853 .owner = THIS_MODULE
1856 static const struct block_device_operations zram_wb_devops = {
1858 .submit_bio = zram_submit_bio,
1859 .swap_slot_free_notify = zram_slot_free_notify,
1860 .owner = THIS_MODULE
1863 static DEVICE_ATTR_WO(compact);
1864 static DEVICE_ATTR_RW(disksize);
1865 static DEVICE_ATTR_RO(initstate);
1866 static DEVICE_ATTR_WO(reset);
1867 static DEVICE_ATTR_WO(mem_limit);
1868 static DEVICE_ATTR_WO(mem_used_max);
1869 static DEVICE_ATTR_WO(idle);
1870 static DEVICE_ATTR_RW(max_comp_streams);
1871 static DEVICE_ATTR_RW(comp_algorithm);
1872 #ifdef CONFIG_ZRAM_WRITEBACK
1873 static DEVICE_ATTR_RW(backing_dev);
1874 static DEVICE_ATTR_WO(writeback);
1875 static DEVICE_ATTR_RW(writeback_limit);
1876 static DEVICE_ATTR_RW(writeback_limit_enable);
1879 static struct attribute *zram_disk_attrs[] = {
1880 &dev_attr_disksize.attr,
1881 &dev_attr_initstate.attr,
1882 &dev_attr_reset.attr,
1883 &dev_attr_compact.attr,
1884 &dev_attr_mem_limit.attr,
1885 &dev_attr_mem_used_max.attr,
1886 &dev_attr_idle.attr,
1887 &dev_attr_max_comp_streams.attr,
1888 &dev_attr_comp_algorithm.attr,
1889 #ifdef CONFIG_ZRAM_WRITEBACK
1890 &dev_attr_backing_dev.attr,
1891 &dev_attr_writeback.attr,
1892 &dev_attr_writeback_limit.attr,
1893 &dev_attr_writeback_limit_enable.attr,
1895 &dev_attr_io_stat.attr,
1896 &dev_attr_mm_stat.attr,
1897 #ifdef CONFIG_ZRAM_WRITEBACK
1898 &dev_attr_bd_stat.attr,
1900 &dev_attr_debug_stat.attr,
1904 static const struct attribute_group zram_disk_attr_group = {
1905 .attrs = zram_disk_attrs,
1908 static const struct attribute_group *zram_disk_attr_groups[] = {
1909 &zram_disk_attr_group,
1914 * Allocate and initialize new zram device. the function returns
1915 * '>= 0' device_id upon success, and negative value otherwise.
1917 static int zram_add(void)
1922 zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1926 ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1931 init_rwsem(&zram->init_lock);
1932 #ifdef CONFIG_ZRAM_WRITEBACK
1933 spin_lock_init(&zram->wb_limit_lock);
1936 /* gendisk structure */
1937 zram->disk = blk_alloc_disk(NUMA_NO_NODE);
1939 pr_err("Error allocating disk structure for device %d\n",
1945 zram->disk->major = zram_major;
1946 zram->disk->first_minor = device_id;
1947 zram->disk->minors = 1;
1948 zram->disk->fops = &zram_devops;
1949 zram->disk->private_data = zram;
1950 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1952 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1953 set_capacity(zram->disk, 0);
1954 /* zram devices sort of resembles non-rotational disks */
1955 blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
1956 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1959 * To ensure that we always get PAGE_SIZE aligned
1960 * and n*PAGE_SIZED sized I/O requests.
1962 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1963 blk_queue_logical_block_size(zram->disk->queue,
1964 ZRAM_LOGICAL_BLOCK_SIZE);
1965 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1966 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1967 zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1968 blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1969 blk_queue_flag_set(QUEUE_FLAG_DISCARD, zram->disk->queue);
1972 * zram_bio_discard() will clear all logical blocks if logical block
1973 * size is identical with physical block size(PAGE_SIZE). But if it is
1974 * different, we will skip discarding some parts of logical blocks in
1975 * the part of the request range which isn't aligned to physical block
1976 * size. So we can't ensure that all discarded logical blocks are
1979 if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1980 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1982 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, zram->disk->queue);
1983 ret = device_add_disk(NULL, zram->disk, zram_disk_attr_groups);
1985 goto out_cleanup_disk;
1987 strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1989 zram_debugfs_register(zram);
1990 pr_info("Added device: %s\n", zram->disk->disk_name);
1994 blk_cleanup_disk(zram->disk);
1996 idr_remove(&zram_index_idr, device_id);
2002 static int zram_remove(struct zram *zram)
2004 struct block_device *bdev = zram->disk->part0;
2007 mutex_lock(&bdev->bd_disk->open_mutex);
2008 if (bdev->bd_openers) {
2009 mutex_unlock(&bdev->bd_disk->open_mutex);
2013 claimed = zram->claim;
2016 mutex_unlock(&bdev->bd_disk->open_mutex);
2018 zram_debugfs_unregister(zram);
2022 * If we were claimed by reset_store(), del_gendisk() will
2023 * wait until reset_store() is done, so nothing need to do.
2027 /* Make sure all the pending I/O are finished */
2028 sync_blockdev(bdev);
2029 zram_reset_device(zram);
2032 pr_info("Removed device: %s\n", zram->disk->disk_name);
2034 del_gendisk(zram->disk);
2036 /* del_gendisk drains pending reset_store */
2037 WARN_ON_ONCE(claimed && zram->claim);
2040 * disksize_store() may be called in between zram_reset_device()
2041 * and del_gendisk(), so run the last reset to avoid leaking
2042 * anything allocated with disksize_store()
2044 zram_reset_device(zram);
2046 blk_cleanup_disk(zram->disk);
2051 /* zram-control sysfs attributes */
2054 * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
2055 * sense that reading from this file does alter the state of your system -- it
2056 * creates a new un-initialized zram device and returns back this device's
2057 * device_id (or an error code if it fails to create a new device).
2059 static ssize_t hot_add_show(struct class *class,
2060 struct class_attribute *attr,
2065 mutex_lock(&zram_index_mutex);
2067 mutex_unlock(&zram_index_mutex);
2071 return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
2073 static struct class_attribute class_attr_hot_add =
2074 __ATTR(hot_add, 0400, hot_add_show, NULL);
2076 static ssize_t hot_remove_store(struct class *class,
2077 struct class_attribute *attr,
2084 /* dev_id is gendisk->first_minor, which is `int' */
2085 ret = kstrtoint(buf, 10, &dev_id);
2091 mutex_lock(&zram_index_mutex);
2093 zram = idr_find(&zram_index_idr, dev_id);
2095 ret = zram_remove(zram);
2097 idr_remove(&zram_index_idr, dev_id);
2102 mutex_unlock(&zram_index_mutex);
2103 return ret ? ret : count;
2105 static CLASS_ATTR_WO(hot_remove);
2107 static struct attribute *zram_control_class_attrs[] = {
2108 &class_attr_hot_add.attr,
2109 &class_attr_hot_remove.attr,
2112 ATTRIBUTE_GROUPS(zram_control_class);
2114 static struct class zram_control_class = {
2115 .name = "zram-control",
2116 .owner = THIS_MODULE,
2117 .class_groups = zram_control_class_groups,
2120 static int zram_remove_cb(int id, void *ptr, void *data)
2122 WARN_ON_ONCE(zram_remove(ptr));
2126 static void destroy_devices(void)
2128 class_unregister(&zram_control_class);
2129 idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
2130 zram_debugfs_destroy();
2131 idr_destroy(&zram_index_idr);
2132 unregister_blkdev(zram_major, "zram");
2133 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2136 static int __init zram_init(void)
2140 ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
2141 zcomp_cpu_up_prepare, zcomp_cpu_dead);
2145 ret = class_register(&zram_control_class);
2147 pr_err("Unable to register zram-control class\n");
2148 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2152 zram_debugfs_create();
2153 zram_major = register_blkdev(0, "zram");
2154 if (zram_major <= 0) {
2155 pr_err("Unable to get major number\n");
2156 class_unregister(&zram_control_class);
2157 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2161 while (num_devices != 0) {
2162 mutex_lock(&zram_index_mutex);
2164 mutex_unlock(&zram_index_mutex);
2177 static void __exit zram_exit(void)
2182 module_init(zram_init);
2183 module_exit(zram_exit);
2185 module_param(num_devices, uint, 0);
2186 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
2188 MODULE_LICENSE("Dual BSD/GPL");
2189 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
2190 MODULE_DESCRIPTION("Compressed RAM Block Device");