2 * Compressed RAM block device
4 * Copyright (C) 2008, 2009, 2010 Nitin Gupta
6 * This code is released using a dual license strategy: BSD/GPL
7 * You can choose the licence that better fits your requirements.
9 * Released under the terms of 3-clause BSD License
10 * Released under the terms of GNU General Public License Version 2.0
12 * Project home: http://compcache.googlecode.com
15 #define KMSG_COMPONENT "zram"
16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
18 #ifdef CONFIG_ZRAM_DEBUG
22 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/bio.h>
25 #include <linux/bitops.h>
26 #include <linux/blkdev.h>
27 #include <linux/buffer_head.h>
28 #include <linux/device.h>
29 #include <linux/genhd.h>
30 #include <linux/highmem.h>
31 #include <linux/slab.h>
32 #include <linux/lzo.h>
33 #include <linux/string.h>
34 #include <linux/vmalloc.h>
39 static int zram_major;
40 struct zram *zram_devices;
42 /* Module params (documentation at end) */
43 static unsigned int num_devices = 1;
45 static void zram_stat64_add(struct zram *zram, u64 *v, u64 inc)
47 spin_lock(&zram->stat64_lock);
49 spin_unlock(&zram->stat64_lock);
52 static void zram_stat64_sub(struct zram *zram, u64 *v, u64 dec)
54 spin_lock(&zram->stat64_lock);
56 spin_unlock(&zram->stat64_lock);
59 static void zram_stat64_inc(struct zram *zram, u64 *v)
61 zram_stat64_add(zram, v, 1);
64 static int zram_test_flag(struct zram *zram, u32 index,
65 enum zram_pageflags flag)
67 return zram->table[index].flags & BIT(flag);
70 static void zram_set_flag(struct zram *zram, u32 index,
71 enum zram_pageflags flag)
73 zram->table[index].flags |= BIT(flag);
76 static void zram_clear_flag(struct zram *zram, u32 index,
77 enum zram_pageflags flag)
79 zram->table[index].flags &= ~BIT(flag);
82 static int page_zero_filled(void *ptr)
87 page = (unsigned long *)ptr;
89 for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
97 static void zram_set_disksize(struct zram *zram, size_t totalram_bytes)
99 if (!zram->disksize) {
101 "disk size not provided. You can use disksize_kb module "
102 "param to specify size.\nUsing default: (%u%% of RAM).\n",
103 default_disksize_perc_ram
105 zram->disksize = default_disksize_perc_ram *
106 (totalram_bytes / 100);
109 if (zram->disksize > 2 * (totalram_bytes)) {
111 "There is little point creating a zram of greater than "
112 "twice the size of memory since we expect a 2:1 compression "
113 "ratio. Note that zram uses about 0.1%% of the size of "
114 "the disk when not in use so a huge zram is "
116 "\tMemory Size: %zu kB\n"
117 "\tSize you selected: %llu kB\n"
118 "Continuing anyway ...\n",
119 totalram_bytes >> 10, zram->disksize >> 10);
122 zram->disksize &= PAGE_MASK;
125 static void zram_free_page(struct zram *zram, size_t index)
127 unsigned long handle = zram->table[index].handle;
128 u16 size = zram->table[index].size;
130 if (unlikely(!handle)) {
132 * No memory is allocated for zero filled pages.
133 * Simply clear zero page flag.
135 if (zram_test_flag(zram, index, ZRAM_ZERO)) {
136 zram_clear_flag(zram, index, ZRAM_ZERO);
137 zram->stats.pages_zero--;
142 if (unlikely(size > max_zpage_size))
143 zram->stats.bad_compress--;
145 zs_free(zram->mem_pool, handle);
147 if (size <= PAGE_SIZE / 2)
148 zram->stats.good_compress--;
150 zram_stat64_sub(zram, &zram->stats.compr_size,
151 zram->table[index].size);
152 zram->stats.pages_stored--;
154 zram->table[index].handle = 0;
155 zram->table[index].size = 0;
158 static void handle_zero_page(struct bio_vec *bvec)
160 struct page *page = bvec->bv_page;
163 user_mem = kmap_atomic(page);
164 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
165 kunmap_atomic(user_mem);
167 flush_dcache_page(page);
170 static inline int is_partial_io(struct bio_vec *bvec)
172 return bvec->bv_len != PAGE_SIZE;
175 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
178 size_t clen = PAGE_SIZE;
180 unsigned long handle = zram->table[index].handle;
182 if (!handle || zram_test_flag(zram, index, ZRAM_ZERO)) {
183 memset(mem, 0, PAGE_SIZE);
187 cmem = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
188 if (zram->table[index].size == PAGE_SIZE)
189 memcpy(mem, cmem, PAGE_SIZE);
191 ret = lzo1x_decompress_safe(cmem, zram->table[index].size,
193 zs_unmap_object(zram->mem_pool, handle);
195 /* Should NEVER happen. Return bio error if it does. */
196 if (unlikely(ret != LZO_E_OK)) {
197 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
198 zram_stat64_inc(zram, &zram->stats.failed_reads);
205 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
206 u32 index, int offset, struct bio *bio)
210 unsigned char *user_mem, *uncmem = NULL;
212 page = bvec->bv_page;
214 if (unlikely(!zram->table[index].handle) ||
215 zram_test_flag(zram, index, ZRAM_ZERO)) {
216 handle_zero_page(bvec);
220 user_mem = kmap_atomic(page);
221 if (is_partial_io(bvec))
222 /* Use a temporary buffer to decompress the page */
223 uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL);
228 pr_info("Unable to allocate temp memory\n");
233 ret = zram_decompress_page(zram, uncmem, index);
234 /* Should NEVER happen. Return bio error if it does. */
235 if (unlikely(ret != LZO_E_OK)) {
236 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
237 zram_stat64_inc(zram, &zram->stats.failed_reads);
241 if (is_partial_io(bvec))
242 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
245 flush_dcache_page(page);
248 kunmap_atomic(user_mem);
249 if (is_partial_io(bvec))
254 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
259 unsigned long handle;
261 unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
263 page = bvec->bv_page;
264 src = zram->compress_buffer;
266 if (is_partial_io(bvec)) {
268 * This is a partial IO. We need to read the full page
269 * before to write the changes.
271 uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL);
273 pr_info("Error allocating temp memory!\n");
277 ret = zram_decompress_page(zram, uncmem, index);
285 * System overwrites unused sectors. Free memory associated
286 * with this sector now.
288 if (zram->table[index].handle ||
289 zram_test_flag(zram, index, ZRAM_ZERO))
290 zram_free_page(zram, index);
292 user_mem = kmap_atomic(page);
294 if (is_partial_io(bvec))
295 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
300 if (page_zero_filled(uncmem)) {
301 kunmap_atomic(user_mem);
302 if (is_partial_io(bvec))
304 zram->stats.pages_zero++;
305 zram_set_flag(zram, index, ZRAM_ZERO);
310 ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
311 zram->compress_workmem);
313 kunmap_atomic(user_mem);
314 if (is_partial_io(bvec))
317 if (unlikely(ret != LZO_E_OK)) {
318 pr_err("Compression failed! err=%d\n", ret);
322 if (unlikely(clen > max_zpage_size)) {
323 zram->stats.bad_compress++;
328 handle = zs_malloc(zram->mem_pool, clen);
330 pr_info("Error allocating memory for compressed "
331 "page: %u, size=%zu\n", index, clen);
335 cmem = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
337 memcpy(cmem, src, clen);
339 zs_unmap_object(zram->mem_pool, handle);
341 zram->table[index].handle = handle;
342 zram->table[index].size = clen;
345 zram_stat64_add(zram, &zram->stats.compr_size, clen);
346 zram->stats.pages_stored++;
347 if (clen <= PAGE_SIZE / 2)
348 zram->stats.good_compress++;
354 zram_stat64_inc(zram, &zram->stats.failed_writes);
358 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
359 int offset, struct bio *bio, int rw)
364 down_read(&zram->lock);
365 ret = zram_bvec_read(zram, bvec, index, offset, bio);
366 up_read(&zram->lock);
368 down_write(&zram->lock);
369 ret = zram_bvec_write(zram, bvec, index, offset);
370 up_write(&zram->lock);
376 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
378 if (*offset + bvec->bv_len >= PAGE_SIZE)
380 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
383 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
387 struct bio_vec *bvec;
391 zram_stat64_inc(zram, &zram->stats.num_reads);
394 zram_stat64_inc(zram, &zram->stats.num_writes);
398 index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
399 offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
401 bio_for_each_segment(bvec, bio, i) {
402 int max_transfer_size = PAGE_SIZE - offset;
404 if (bvec->bv_len > max_transfer_size) {
406 * zram_bvec_rw() can only make operation on a single
407 * zram page. Split the bio vector.
411 bv.bv_page = bvec->bv_page;
412 bv.bv_len = max_transfer_size;
413 bv.bv_offset = bvec->bv_offset;
415 if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
418 bv.bv_len = bvec->bv_len - max_transfer_size;
419 bv.bv_offset += max_transfer_size;
420 if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
423 if (zram_bvec_rw(zram, bvec, index, offset, bio, rw)
427 update_position(&index, &offset, bvec);
430 set_bit(BIO_UPTODATE, &bio->bi_flags);
439 * Check if request is within bounds and aligned on zram logical blocks.
441 static inline int valid_io_request(struct zram *zram, struct bio *bio)
444 (bio->bi_sector >= (zram->disksize >> SECTOR_SHIFT)) ||
445 (bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)) ||
446 (bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))) {
451 /* I/O request is valid */
456 * Handler function for all zram I/O requests.
458 static void zram_make_request(struct request_queue *queue, struct bio *bio)
460 struct zram *zram = queue->queuedata;
462 if (unlikely(!zram->init_done) && zram_init_device(zram))
465 down_read(&zram->init_lock);
466 if (unlikely(!zram->init_done))
469 if (!valid_io_request(zram, bio)) {
470 zram_stat64_inc(zram, &zram->stats.invalid_io);
474 __zram_make_request(zram, bio, bio_data_dir(bio));
475 up_read(&zram->init_lock);
480 up_read(&zram->init_lock);
485 void __zram_reset_device(struct zram *zram)
491 /* Free various per-device buffers */
492 kfree(zram->compress_workmem);
493 free_pages((unsigned long)zram->compress_buffer, 1);
495 zram->compress_workmem = NULL;
496 zram->compress_buffer = NULL;
498 /* Free all pages that are still in this zram device */
499 for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
500 unsigned long handle = zram->table[index].handle;
504 zs_free(zram->mem_pool, handle);
510 zs_destroy_pool(zram->mem_pool);
511 zram->mem_pool = NULL;
514 memset(&zram->stats, 0, sizeof(zram->stats));
519 void zram_reset_device(struct zram *zram)
521 down_write(&zram->init_lock);
522 __zram_reset_device(zram);
523 up_write(&zram->init_lock);
526 int zram_init_device(struct zram *zram)
531 down_write(&zram->init_lock);
533 if (zram->init_done) {
534 up_write(&zram->init_lock);
538 zram_set_disksize(zram, totalram_pages << PAGE_SHIFT);
540 zram->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
541 if (!zram->compress_workmem) {
542 pr_err("Error allocating compressor working memory!\n");
547 zram->compress_buffer =
548 (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
549 if (!zram->compress_buffer) {
550 pr_err("Error allocating compressor buffer space\n");
555 num_pages = zram->disksize >> PAGE_SHIFT;
556 zram->table = vzalloc(num_pages * sizeof(*zram->table));
558 pr_err("Error allocating zram address table\n");
563 set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
565 /* zram devices sort of resembles non-rotational disks */
566 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
568 zram->mem_pool = zs_create_pool("zram", GFP_NOIO | __GFP_HIGHMEM);
569 if (!zram->mem_pool) {
570 pr_err("Error creating memory pool\n");
576 up_write(&zram->init_lock);
578 pr_debug("Initialization done!\n");
582 /* To prevent accessing table entries during cleanup */
585 __zram_reset_device(zram);
586 up_write(&zram->init_lock);
587 pr_err("Initialization failed: err=%d\n", ret);
591 static void zram_slot_free_notify(struct block_device *bdev,
596 zram = bdev->bd_disk->private_data;
597 zram_free_page(zram, index);
598 zram_stat64_inc(zram, &zram->stats.notify_free);
601 static const struct block_device_operations zram_devops = {
602 .swap_slot_free_notify = zram_slot_free_notify,
606 static int create_device(struct zram *zram, int device_id)
610 init_rwsem(&zram->lock);
611 init_rwsem(&zram->init_lock);
612 spin_lock_init(&zram->stat64_lock);
614 zram->queue = blk_alloc_queue(GFP_KERNEL);
616 pr_err("Error allocating disk queue for device %d\n",
622 blk_queue_make_request(zram->queue, zram_make_request);
623 zram->queue->queuedata = zram;
625 /* gendisk structure */
626 zram->disk = alloc_disk(1);
628 blk_cleanup_queue(zram->queue);
629 pr_warn("Error allocating disk structure for device %d\n",
635 zram->disk->major = zram_major;
636 zram->disk->first_minor = device_id;
637 zram->disk->fops = &zram_devops;
638 zram->disk->queue = zram->queue;
639 zram->disk->private_data = zram;
640 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
642 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
643 set_capacity(zram->disk, 0);
646 * To ensure that we always get PAGE_SIZE aligned
647 * and n*PAGE_SIZED sized I/O requests.
649 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
650 blk_queue_logical_block_size(zram->disk->queue,
651 ZRAM_LOGICAL_BLOCK_SIZE);
652 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
653 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
655 add_disk(zram->disk);
657 ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
658 &zram_disk_attr_group);
660 pr_warn("Error creating sysfs group");
670 static void destroy_device(struct zram *zram)
672 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
673 &zram_disk_attr_group);
676 del_gendisk(zram->disk);
677 put_disk(zram->disk);
681 blk_cleanup_queue(zram->queue);
684 unsigned int zram_get_num_devices(void)
689 static int __init zram_init(void)
693 if (num_devices > max_num_devices) {
694 pr_warn("Invalid value for num_devices: %u\n",
700 zram_major = register_blkdev(0, "zram");
701 if (zram_major <= 0) {
702 pr_warn("Unable to get major number\n");
707 /* Allocate the device array and initialize each one */
708 zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
714 for (dev_id = 0; dev_id < num_devices; dev_id++) {
715 ret = create_device(&zram_devices[dev_id], dev_id);
720 pr_info("Created %u device(s) ...\n", num_devices);
726 destroy_device(&zram_devices[--dev_id]);
729 unregister_blkdev(zram_major, "zram");
734 static void __exit zram_exit(void)
739 for (i = 0; i < num_devices; i++) {
740 zram = &zram_devices[i];
742 destroy_device(zram);
744 zram_reset_device(zram);
747 unregister_blkdev(zram_major, "zram");
750 pr_debug("Cleanup done!\n");
753 module_param(num_devices, uint, 0);
754 MODULE_PARM_DESC(num_devices, "Number of zram devices");
756 module_init(zram_init);
757 module_exit(zram_exit);
759 MODULE_LICENSE("Dual BSD/GPL");
760 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
761 MODULE_DESCRIPTION("Compressed RAM Block Device");