Loading...
Note: File does not exist in v3.1.
1/*
2 * Compressed RAM block device
3 *
4 * Copyright (C) 2008, 2009, 2010 Nitin Gupta
5 * 2012, 2013 Minchan Kim
6 *
7 * This code is released using a dual license strategy: BSD/GPL
8 * You can choose the licence that better fits your requirements.
9 *
10 * Released under the terms of 3-clause BSD License
11 * Released under the terms of GNU General Public License Version 2.0
12 *
13 */
14
15#define KMSG_COMPONENT "zram"
16#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
17
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>
37
38#include "zram_drv.h"
39
40static DEFINE_IDR(zram_index_idr);
41/* idr index must be protected */
42static DEFINE_MUTEX(zram_index_mutex);
43
44static int zram_major;
45static const char *default_compressor = CONFIG_ZRAM_DEF_COMP;
46
47/* Module params (documentation at end) */
48static unsigned int num_devices = 1;
49/*
50 * Pages that compress to sizes equals or greater than this are stored
51 * uncompressed in memory.
52 */
53static size_t huge_class_size;
54
55static const struct block_device_operations zram_devops;
56static const struct block_device_operations zram_wb_devops;
57
58static void zram_free_page(struct zram *zram, size_t index);
59static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
60 u32 index, int offset, struct bio *bio);
61
62
63static int zram_slot_trylock(struct zram *zram, u32 index)
64{
65 return bit_spin_trylock(ZRAM_LOCK, &zram->table[index].flags);
66}
67
68static void zram_slot_lock(struct zram *zram, u32 index)
69{
70 bit_spin_lock(ZRAM_LOCK, &zram->table[index].flags);
71}
72
73static void zram_slot_unlock(struct zram *zram, u32 index)
74{
75 bit_spin_unlock(ZRAM_LOCK, &zram->table[index].flags);
76}
77
78static inline bool init_done(struct zram *zram)
79{
80 return zram->disksize;
81}
82
83static inline struct zram *dev_to_zram(struct device *dev)
84{
85 return (struct zram *)dev_to_disk(dev)->private_data;
86}
87
88static unsigned long zram_get_handle(struct zram *zram, u32 index)
89{
90 return zram->table[index].handle;
91}
92
93static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
94{
95 zram->table[index].handle = handle;
96}
97
98/* flag operations require table entry bit_spin_lock() being held */
99static bool zram_test_flag(struct zram *zram, u32 index,
100 enum zram_pageflags flag)
101{
102 return zram->table[index].flags & BIT(flag);
103}
104
105static void zram_set_flag(struct zram *zram, u32 index,
106 enum zram_pageflags flag)
107{
108 zram->table[index].flags |= BIT(flag);
109}
110
111static void zram_clear_flag(struct zram *zram, u32 index,
112 enum zram_pageflags flag)
113{
114 zram->table[index].flags &= ~BIT(flag);
115}
116
117static inline void zram_set_element(struct zram *zram, u32 index,
118 unsigned long element)
119{
120 zram->table[index].element = element;
121}
122
123static unsigned long zram_get_element(struct zram *zram, u32 index)
124{
125 return zram->table[index].element;
126}
127
128static size_t zram_get_obj_size(struct zram *zram, u32 index)
129{
130 return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
131}
132
133static void zram_set_obj_size(struct zram *zram,
134 u32 index, size_t size)
135{
136 unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
137
138 zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
139}
140
141static inline bool zram_allocated(struct zram *zram, u32 index)
142{
143 return zram_get_obj_size(zram, index) ||
144 zram_test_flag(zram, index, ZRAM_SAME) ||
145 zram_test_flag(zram, index, ZRAM_WB);
146}
147
148#if PAGE_SIZE != 4096
149static inline bool is_partial_io(struct bio_vec *bvec)
150{
151 return bvec->bv_len != PAGE_SIZE;
152}
153#else
154static inline bool is_partial_io(struct bio_vec *bvec)
155{
156 return false;
157}
158#endif
159
160/*
161 * Check if request is within bounds and aligned on zram logical blocks.
162 */
163static inline bool valid_io_request(struct zram *zram,
164 sector_t start, unsigned int size)
165{
166 u64 end, bound;
167
168 /* unaligned request */
169 if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
170 return false;
171 if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
172 return false;
173
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))
178 return false;
179
180 /* I/O request is valid */
181 return true;
182}
183
184static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
185{
186 *index += (*offset + bvec->bv_len) / PAGE_SIZE;
187 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
188}
189
190static inline void update_used_max(struct zram *zram,
191 const unsigned long pages)
192{
193 unsigned long old_max, cur_max;
194
195 old_max = atomic_long_read(&zram->stats.max_used_pages);
196
197 do {
198 cur_max = old_max;
199 if (pages > cur_max)
200 old_max = atomic_long_cmpxchg(
201 &zram->stats.max_used_pages, cur_max, pages);
202 } while (old_max != cur_max);
203}
204
205static inline void zram_fill_page(void *ptr, unsigned long len,
206 unsigned long value)
207{
208 WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
209 memset_l(ptr, value, len / sizeof(unsigned long));
210}
211
212static bool page_same_filled(void *ptr, unsigned long *element)
213{
214 unsigned long *page;
215 unsigned long val;
216 unsigned int pos, last_pos = PAGE_SIZE / sizeof(*page) - 1;
217
218 page = (unsigned long *)ptr;
219 val = page[0];
220
221 if (val != page[last_pos])
222 return false;
223
224 for (pos = 1; pos < last_pos; pos++) {
225 if (val != page[pos])
226 return false;
227 }
228
229 *element = val;
230
231 return true;
232}
233
234static ssize_t initstate_show(struct device *dev,
235 struct device_attribute *attr, char *buf)
236{
237 u32 val;
238 struct zram *zram = dev_to_zram(dev);
239
240 down_read(&zram->init_lock);
241 val = init_done(zram);
242 up_read(&zram->init_lock);
243
244 return scnprintf(buf, PAGE_SIZE, "%u\n", val);
245}
246
247static ssize_t disksize_show(struct device *dev,
248 struct device_attribute *attr, char *buf)
249{
250 struct zram *zram = dev_to_zram(dev);
251
252 return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
253}
254
255static ssize_t mem_limit_store(struct device *dev,
256 struct device_attribute *attr, const char *buf, size_t len)
257{
258 u64 limit;
259 char *tmp;
260 struct zram *zram = dev_to_zram(dev);
261
262 limit = memparse(buf, &tmp);
263 if (buf == tmp) /* no chars parsed, invalid input */
264 return -EINVAL;
265
266 down_write(&zram->init_lock);
267 zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
268 up_write(&zram->init_lock);
269
270 return len;
271}
272
273static ssize_t mem_used_max_store(struct device *dev,
274 struct device_attribute *attr, const char *buf, size_t len)
275{
276 int err;
277 unsigned long val;
278 struct zram *zram = dev_to_zram(dev);
279
280 err = kstrtoul(buf, 10, &val);
281 if (err || val != 0)
282 return -EINVAL;
283
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));
288 }
289 up_read(&zram->init_lock);
290
291 return len;
292}
293
294static ssize_t idle_store(struct device *dev,
295 struct device_attribute *attr, const char *buf, size_t len)
296{
297 struct zram *zram = dev_to_zram(dev);
298 unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
299 int index;
300
301 if (!sysfs_streq(buf, "all"))
302 return -EINVAL;
303
304 down_read(&zram->init_lock);
305 if (!init_done(zram)) {
306 up_read(&zram->init_lock);
307 return -EINVAL;
308 }
309
310 for (index = 0; index < nr_pages; index++) {
311 /*
312 * Do not mark ZRAM_UNDER_WB slot as ZRAM_IDLE to close race.
313 * See the comment in writeback_store.
314 */
315 zram_slot_lock(zram, index);
316 if (zram_allocated(zram, index) &&
317 !zram_test_flag(zram, index, ZRAM_UNDER_WB))
318 zram_set_flag(zram, index, ZRAM_IDLE);
319 zram_slot_unlock(zram, index);
320 }
321
322 up_read(&zram->init_lock);
323
324 return len;
325}
326
327#ifdef CONFIG_ZRAM_WRITEBACK
328static ssize_t writeback_limit_enable_store(struct device *dev,
329 struct device_attribute *attr, const char *buf, size_t len)
330{
331 struct zram *zram = dev_to_zram(dev);
332 u64 val;
333 ssize_t ret = -EINVAL;
334
335 if (kstrtoull(buf, 10, &val))
336 return ret;
337
338 down_read(&zram->init_lock);
339 spin_lock(&zram->wb_limit_lock);
340 zram->wb_limit_enable = val;
341 spin_unlock(&zram->wb_limit_lock);
342 up_read(&zram->init_lock);
343 ret = len;
344
345 return ret;
346}
347
348static ssize_t writeback_limit_enable_show(struct device *dev,
349 struct device_attribute *attr, char *buf)
350{
351 bool val;
352 struct zram *zram = dev_to_zram(dev);
353
354 down_read(&zram->init_lock);
355 spin_lock(&zram->wb_limit_lock);
356 val = zram->wb_limit_enable;
357 spin_unlock(&zram->wb_limit_lock);
358 up_read(&zram->init_lock);
359
360 return scnprintf(buf, PAGE_SIZE, "%d\n", val);
361}
362
363static ssize_t writeback_limit_store(struct device *dev,
364 struct device_attribute *attr, const char *buf, size_t len)
365{
366 struct zram *zram = dev_to_zram(dev);
367 u64 val;
368 ssize_t ret = -EINVAL;
369
370 if (kstrtoull(buf, 10, &val))
371 return ret;
372
373 down_read(&zram->init_lock);
374 spin_lock(&zram->wb_limit_lock);
375 zram->bd_wb_limit = val;
376 spin_unlock(&zram->wb_limit_lock);
377 up_read(&zram->init_lock);
378 ret = len;
379
380 return ret;
381}
382
383static ssize_t writeback_limit_show(struct device *dev,
384 struct device_attribute *attr, char *buf)
385{
386 u64 val;
387 struct zram *zram = dev_to_zram(dev);
388
389 down_read(&zram->init_lock);
390 spin_lock(&zram->wb_limit_lock);
391 val = zram->bd_wb_limit;
392 spin_unlock(&zram->wb_limit_lock);
393 up_read(&zram->init_lock);
394
395 return scnprintf(buf, PAGE_SIZE, "%llu\n", val);
396}
397
398static void reset_bdev(struct zram *zram)
399{
400 struct block_device *bdev;
401
402 if (!zram->backing_dev)
403 return;
404
405 bdev = zram->bdev;
406 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
407 /* hope filp_close flush all of IO */
408 filp_close(zram->backing_dev, NULL);
409 zram->backing_dev = NULL;
410 zram->bdev = NULL;
411 zram->disk->fops = &zram_devops;
412 kvfree(zram->bitmap);
413 zram->bitmap = NULL;
414}
415
416static ssize_t backing_dev_show(struct device *dev,
417 struct device_attribute *attr, char *buf)
418{
419 struct file *file;
420 struct zram *zram = dev_to_zram(dev);
421 char *p;
422 ssize_t ret;
423
424 down_read(&zram->init_lock);
425 file = zram->backing_dev;
426 if (!file) {
427 memcpy(buf, "none\n", 5);
428 up_read(&zram->init_lock);
429 return 5;
430 }
431
432 p = file_path(file, buf, PAGE_SIZE - 1);
433 if (IS_ERR(p)) {
434 ret = PTR_ERR(p);
435 goto out;
436 }
437
438 ret = strlen(p);
439 memmove(buf, p, ret);
440 buf[ret++] = '\n';
441out:
442 up_read(&zram->init_lock);
443 return ret;
444}
445
446static ssize_t backing_dev_store(struct device *dev,
447 struct device_attribute *attr, const char *buf, size_t len)
448{
449 char *file_name;
450 size_t sz;
451 struct file *backing_dev = NULL;
452 struct inode *inode;
453 struct address_space *mapping;
454 unsigned int bitmap_sz;
455 unsigned long nr_pages, *bitmap = NULL;
456 struct block_device *bdev = NULL;
457 int err;
458 struct zram *zram = dev_to_zram(dev);
459
460 file_name = kmalloc(PATH_MAX, GFP_KERNEL);
461 if (!file_name)
462 return -ENOMEM;
463
464 down_write(&zram->init_lock);
465 if (init_done(zram)) {
466 pr_info("Can't setup backing device for initialized device\n");
467 err = -EBUSY;
468 goto out;
469 }
470
471 strlcpy(file_name, buf, PATH_MAX);
472 /* ignore trailing newline */
473 sz = strlen(file_name);
474 if (sz > 0 && file_name[sz - 1] == '\n')
475 file_name[sz - 1] = 0x00;
476
477 backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
478 if (IS_ERR(backing_dev)) {
479 err = PTR_ERR(backing_dev);
480 backing_dev = NULL;
481 goto out;
482 }
483
484 mapping = backing_dev->f_mapping;
485 inode = mapping->host;
486
487 /* Support only block device in this moment */
488 if (!S_ISBLK(inode->i_mode)) {
489 err = -ENOTBLK;
490 goto out;
491 }
492
493 bdev = blkdev_get_by_dev(inode->i_rdev,
494 FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
495 if (IS_ERR(bdev)) {
496 err = PTR_ERR(bdev);
497 bdev = NULL;
498 goto out;
499 }
500
501 nr_pages = i_size_read(inode) >> PAGE_SHIFT;
502 bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
503 bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
504 if (!bitmap) {
505 err = -ENOMEM;
506 goto out;
507 }
508
509 reset_bdev(zram);
510
511 zram->bdev = bdev;
512 zram->backing_dev = backing_dev;
513 zram->bitmap = bitmap;
514 zram->nr_pages = nr_pages;
515 /*
516 * With writeback feature, zram does asynchronous IO so it's no longer
517 * synchronous device so let's remove synchronous io flag. Othewise,
518 * upper layer(e.g., swap) could wait IO completion rather than
519 * (submit and return), which will cause system sluggish.
520 * Furthermore, when the IO function returns(e.g., swap_readpage),
521 * upper layer expects IO was done so it could deallocate the page
522 * freely but in fact, IO is going on so finally could cause
523 * use-after-free when the IO is really done.
524 */
525 zram->disk->fops = &zram_wb_devops;
526 up_write(&zram->init_lock);
527
528 pr_info("setup backing device %s\n", file_name);
529 kfree(file_name);
530
531 return len;
532out:
533 kvfree(bitmap);
534
535 if (bdev)
536 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
537
538 if (backing_dev)
539 filp_close(backing_dev, NULL);
540
541 up_write(&zram->init_lock);
542
543 kfree(file_name);
544
545 return err;
546}
547
548static unsigned long alloc_block_bdev(struct zram *zram)
549{
550 unsigned long blk_idx = 1;
551retry:
552 /* skip 0 bit to confuse zram.handle = 0 */
553 blk_idx = find_next_zero_bit(zram->bitmap, zram->nr_pages, blk_idx);
554 if (blk_idx == zram->nr_pages)
555 return 0;
556
557 if (test_and_set_bit(blk_idx, zram->bitmap))
558 goto retry;
559
560 atomic64_inc(&zram->stats.bd_count);
561 return blk_idx;
562}
563
564static void free_block_bdev(struct zram *zram, unsigned long blk_idx)
565{
566 int was_set;
567
568 was_set = test_and_clear_bit(blk_idx, zram->bitmap);
569 WARN_ON_ONCE(!was_set);
570 atomic64_dec(&zram->stats.bd_count);
571}
572
573static void zram_page_end_io(struct bio *bio)
574{
575 struct page *page = bio_first_page_all(bio);
576
577 page_endio(page, op_is_write(bio_op(bio)),
578 blk_status_to_errno(bio->bi_status));
579 bio_put(bio);
580}
581
582/*
583 * Returns 1 if the submission is successful.
584 */
585static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
586 unsigned long entry, struct bio *parent)
587{
588 struct bio *bio;
589
590 bio = bio_alloc(GFP_ATOMIC, 1);
591 if (!bio)
592 return -ENOMEM;
593
594 bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
595 bio_set_dev(bio, zram->bdev);
596 if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
597 bio_put(bio);
598 return -EIO;
599 }
600
601 if (!parent) {
602 bio->bi_opf = REQ_OP_READ;
603 bio->bi_end_io = zram_page_end_io;
604 } else {
605 bio->bi_opf = parent->bi_opf;
606 bio_chain(bio, parent);
607 }
608
609 submit_bio(bio);
610 return 1;
611}
612
613#define PAGE_WB_SIG "page_index="
614
615#define PAGE_WRITEBACK 0
616#define HUGE_WRITEBACK 1
617#define IDLE_WRITEBACK 2
618
619
620static ssize_t writeback_store(struct device *dev,
621 struct device_attribute *attr, const char *buf, size_t len)
622{
623 struct zram *zram = dev_to_zram(dev);
624 unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
625 unsigned long index = 0;
626 struct bio bio;
627 struct bio_vec bio_vec;
628 struct page *page;
629 ssize_t ret = len;
630 int mode, err;
631 unsigned long blk_idx = 0;
632
633 if (sysfs_streq(buf, "idle"))
634 mode = IDLE_WRITEBACK;
635 else if (sysfs_streq(buf, "huge"))
636 mode = HUGE_WRITEBACK;
637 else {
638 if (strncmp(buf, PAGE_WB_SIG, sizeof(PAGE_WB_SIG) - 1))
639 return -EINVAL;
640
641 if (kstrtol(buf + sizeof(PAGE_WB_SIG) - 1, 10, &index) ||
642 index >= nr_pages)
643 return -EINVAL;
644
645 nr_pages = 1;
646 mode = PAGE_WRITEBACK;
647 }
648
649 down_read(&zram->init_lock);
650 if (!init_done(zram)) {
651 ret = -EINVAL;
652 goto release_init_lock;
653 }
654
655 if (!zram->backing_dev) {
656 ret = -ENODEV;
657 goto release_init_lock;
658 }
659
660 page = alloc_page(GFP_KERNEL);
661 if (!page) {
662 ret = -ENOMEM;
663 goto release_init_lock;
664 }
665
666 for (; nr_pages != 0; index++, nr_pages--) {
667 struct bio_vec bvec;
668
669 bvec.bv_page = page;
670 bvec.bv_len = PAGE_SIZE;
671 bvec.bv_offset = 0;
672
673 spin_lock(&zram->wb_limit_lock);
674 if (zram->wb_limit_enable && !zram->bd_wb_limit) {
675 spin_unlock(&zram->wb_limit_lock);
676 ret = -EIO;
677 break;
678 }
679 spin_unlock(&zram->wb_limit_lock);
680
681 if (!blk_idx) {
682 blk_idx = alloc_block_bdev(zram);
683 if (!blk_idx) {
684 ret = -ENOSPC;
685 break;
686 }
687 }
688
689 zram_slot_lock(zram, index);
690 if (!zram_allocated(zram, index))
691 goto next;
692
693 if (zram_test_flag(zram, index, ZRAM_WB) ||
694 zram_test_flag(zram, index, ZRAM_SAME) ||
695 zram_test_flag(zram, index, ZRAM_UNDER_WB))
696 goto next;
697
698 if (mode == IDLE_WRITEBACK &&
699 !zram_test_flag(zram, index, ZRAM_IDLE))
700 goto next;
701 if (mode == HUGE_WRITEBACK &&
702 !zram_test_flag(zram, index, ZRAM_HUGE))
703 goto next;
704 /*
705 * Clearing ZRAM_UNDER_WB is duty of caller.
706 * IOW, zram_free_page never clear it.
707 */
708 zram_set_flag(zram, index, ZRAM_UNDER_WB);
709 /* Need for hugepage writeback racing */
710 zram_set_flag(zram, index, ZRAM_IDLE);
711 zram_slot_unlock(zram, index);
712 if (zram_bvec_read(zram, &bvec, index, 0, NULL)) {
713 zram_slot_lock(zram, index);
714 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
715 zram_clear_flag(zram, index, ZRAM_IDLE);
716 zram_slot_unlock(zram, index);
717 continue;
718 }
719
720 bio_init(&bio, &bio_vec, 1);
721 bio_set_dev(&bio, zram->bdev);
722 bio.bi_iter.bi_sector = blk_idx * (PAGE_SIZE >> 9);
723 bio.bi_opf = REQ_OP_WRITE | REQ_SYNC;
724
725 bio_add_page(&bio, bvec.bv_page, bvec.bv_len,
726 bvec.bv_offset);
727 /*
728 * XXX: A single page IO would be inefficient for write
729 * but it would be not bad as starter.
730 */
731 err = submit_bio_wait(&bio);
732 if (err) {
733 zram_slot_lock(zram, index);
734 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
735 zram_clear_flag(zram, index, ZRAM_IDLE);
736 zram_slot_unlock(zram, index);
737 /*
738 * Return last IO error unless every IO were
739 * not suceeded.
740 */
741 ret = err;
742 continue;
743 }
744
745 atomic64_inc(&zram->stats.bd_writes);
746 /*
747 * We released zram_slot_lock so need to check if the slot was
748 * changed. If there is freeing for the slot, we can catch it
749 * easily by zram_allocated.
750 * A subtle case is the slot is freed/reallocated/marked as
751 * ZRAM_IDLE again. To close the race, idle_store doesn't
752 * mark ZRAM_IDLE once it found the slot was ZRAM_UNDER_WB.
753 * Thus, we could close the race by checking ZRAM_IDLE bit.
754 */
755 zram_slot_lock(zram, index);
756 if (!zram_allocated(zram, index) ||
757 !zram_test_flag(zram, index, ZRAM_IDLE)) {
758 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
759 zram_clear_flag(zram, index, ZRAM_IDLE);
760 goto next;
761 }
762
763 zram_free_page(zram, index);
764 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
765 zram_set_flag(zram, index, ZRAM_WB);
766 zram_set_element(zram, index, blk_idx);
767 blk_idx = 0;
768 atomic64_inc(&zram->stats.pages_stored);
769 spin_lock(&zram->wb_limit_lock);
770 if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
771 zram->bd_wb_limit -= 1UL << (PAGE_SHIFT - 12);
772 spin_unlock(&zram->wb_limit_lock);
773next:
774 zram_slot_unlock(zram, index);
775 }
776
777 if (blk_idx)
778 free_block_bdev(zram, blk_idx);
779 __free_page(page);
780release_init_lock:
781 up_read(&zram->init_lock);
782
783 return ret;
784}
785
786struct zram_work {
787 struct work_struct work;
788 struct zram *zram;
789 unsigned long entry;
790 struct bio *bio;
791 struct bio_vec bvec;
792};
793
794#if PAGE_SIZE != 4096
795static void zram_sync_read(struct work_struct *work)
796{
797 struct zram_work *zw = container_of(work, struct zram_work, work);
798 struct zram *zram = zw->zram;
799 unsigned long entry = zw->entry;
800 struct bio *bio = zw->bio;
801
802 read_from_bdev_async(zram, &zw->bvec, entry, bio);
803}
804
805/*
806 * Block layer want one ->submit_bio to be active at a time, so if we use
807 * chained IO with parent IO in same context, it's a deadlock. To avoid that,
808 * use a worker thread context.
809 */
810static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
811 unsigned long entry, struct bio *bio)
812{
813 struct zram_work work;
814
815 work.bvec = *bvec;
816 work.zram = zram;
817 work.entry = entry;
818 work.bio = bio;
819
820 INIT_WORK_ONSTACK(&work.work, zram_sync_read);
821 queue_work(system_unbound_wq, &work.work);
822 flush_work(&work.work);
823 destroy_work_on_stack(&work.work);
824
825 return 1;
826}
827#else
828static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
829 unsigned long entry, struct bio *bio)
830{
831 WARN_ON(1);
832 return -EIO;
833}
834#endif
835
836static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
837 unsigned long entry, struct bio *parent, bool sync)
838{
839 atomic64_inc(&zram->stats.bd_reads);
840 if (sync)
841 return read_from_bdev_sync(zram, bvec, entry, parent);
842 else
843 return read_from_bdev_async(zram, bvec, entry, parent);
844}
845#else
846static inline void reset_bdev(struct zram *zram) {};
847static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
848 unsigned long entry, struct bio *parent, bool sync)
849{
850 return -EIO;
851}
852
853static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
854#endif
855
856#ifdef CONFIG_ZRAM_MEMORY_TRACKING
857
858static struct dentry *zram_debugfs_root;
859
860static void zram_debugfs_create(void)
861{
862 zram_debugfs_root = debugfs_create_dir("zram", NULL);
863}
864
865static void zram_debugfs_destroy(void)
866{
867 debugfs_remove_recursive(zram_debugfs_root);
868}
869
870static void zram_accessed(struct zram *zram, u32 index)
871{
872 zram_clear_flag(zram, index, ZRAM_IDLE);
873 zram->table[index].ac_time = ktime_get_boottime();
874}
875
876static ssize_t read_block_state(struct file *file, char __user *buf,
877 size_t count, loff_t *ppos)
878{
879 char *kbuf;
880 ssize_t index, written = 0;
881 struct zram *zram = file->private_data;
882 unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
883 struct timespec64 ts;
884
885 kbuf = kvmalloc(count, GFP_KERNEL);
886 if (!kbuf)
887 return -ENOMEM;
888
889 down_read(&zram->init_lock);
890 if (!init_done(zram)) {
891 up_read(&zram->init_lock);
892 kvfree(kbuf);
893 return -EINVAL;
894 }
895
896 for (index = *ppos; index < nr_pages; index++) {
897 int copied;
898
899 zram_slot_lock(zram, index);
900 if (!zram_allocated(zram, index))
901 goto next;
902
903 ts = ktime_to_timespec64(zram->table[index].ac_time);
904 copied = snprintf(kbuf + written, count,
905 "%12zd %12lld.%06lu %c%c%c%c\n",
906 index, (s64)ts.tv_sec,
907 ts.tv_nsec / NSEC_PER_USEC,
908 zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
909 zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
910 zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.',
911 zram_test_flag(zram, index, ZRAM_IDLE) ? 'i' : '.');
912
913 if (count < copied) {
914 zram_slot_unlock(zram, index);
915 break;
916 }
917 written += copied;
918 count -= copied;
919next:
920 zram_slot_unlock(zram, index);
921 *ppos += 1;
922 }
923
924 up_read(&zram->init_lock);
925 if (copy_to_user(buf, kbuf, written))
926 written = -EFAULT;
927 kvfree(kbuf);
928
929 return written;
930}
931
932static const struct file_operations proc_zram_block_state_op = {
933 .open = simple_open,
934 .read = read_block_state,
935 .llseek = default_llseek,
936};
937
938static void zram_debugfs_register(struct zram *zram)
939{
940 if (!zram_debugfs_root)
941 return;
942
943 zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
944 zram_debugfs_root);
945 debugfs_create_file("block_state", 0400, zram->debugfs_dir,
946 zram, &proc_zram_block_state_op);
947}
948
949static void zram_debugfs_unregister(struct zram *zram)
950{
951 debugfs_remove_recursive(zram->debugfs_dir);
952}
953#else
954static void zram_debugfs_create(void) {};
955static void zram_debugfs_destroy(void) {};
956static void zram_accessed(struct zram *zram, u32 index)
957{
958 zram_clear_flag(zram, index, ZRAM_IDLE);
959};
960static void zram_debugfs_register(struct zram *zram) {};
961static void zram_debugfs_unregister(struct zram *zram) {};
962#endif
963
964/*
965 * We switched to per-cpu streams and this attr is not needed anymore.
966 * However, we will keep it around for some time, because:
967 * a) we may revert per-cpu streams in the future
968 * b) it's visible to user space and we need to follow our 2 years
969 * retirement rule; but we already have a number of 'soon to be
970 * altered' attrs, so max_comp_streams need to wait for the next
971 * layoff cycle.
972 */
973static ssize_t max_comp_streams_show(struct device *dev,
974 struct device_attribute *attr, char *buf)
975{
976 return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
977}
978
979static ssize_t max_comp_streams_store(struct device *dev,
980 struct device_attribute *attr, const char *buf, size_t len)
981{
982 return len;
983}
984
985static ssize_t comp_algorithm_show(struct device *dev,
986 struct device_attribute *attr, char *buf)
987{
988 size_t sz;
989 struct zram *zram = dev_to_zram(dev);
990
991 down_read(&zram->init_lock);
992 sz = zcomp_available_show(zram->compressor, buf);
993 up_read(&zram->init_lock);
994
995 return sz;
996}
997
998static ssize_t comp_algorithm_store(struct device *dev,
999 struct device_attribute *attr, const char *buf, size_t len)
1000{
1001 struct zram *zram = dev_to_zram(dev);
1002 char compressor[ARRAY_SIZE(zram->compressor)];
1003 size_t sz;
1004
1005 strlcpy(compressor, buf, sizeof(compressor));
1006 /* ignore trailing newline */
1007 sz = strlen(compressor);
1008 if (sz > 0 && compressor[sz - 1] == '\n')
1009 compressor[sz - 1] = 0x00;
1010
1011 if (!zcomp_available_algorithm(compressor))
1012 return -EINVAL;
1013
1014 down_write(&zram->init_lock);
1015 if (init_done(zram)) {
1016 up_write(&zram->init_lock);
1017 pr_info("Can't change algorithm for initialized device\n");
1018 return -EBUSY;
1019 }
1020
1021 strcpy(zram->compressor, compressor);
1022 up_write(&zram->init_lock);
1023 return len;
1024}
1025
1026static ssize_t compact_store(struct device *dev,
1027 struct device_attribute *attr, const char *buf, size_t len)
1028{
1029 struct zram *zram = dev_to_zram(dev);
1030
1031 down_read(&zram->init_lock);
1032 if (!init_done(zram)) {
1033 up_read(&zram->init_lock);
1034 return -EINVAL;
1035 }
1036
1037 zs_compact(zram->mem_pool);
1038 up_read(&zram->init_lock);
1039
1040 return len;
1041}
1042
1043static ssize_t io_stat_show(struct device *dev,
1044 struct device_attribute *attr, char *buf)
1045{
1046 struct zram *zram = dev_to_zram(dev);
1047 ssize_t ret;
1048
1049 down_read(&zram->init_lock);
1050 ret = scnprintf(buf, PAGE_SIZE,
1051 "%8llu %8llu %8llu %8llu\n",
1052 (u64)atomic64_read(&zram->stats.failed_reads),
1053 (u64)atomic64_read(&zram->stats.failed_writes),
1054 (u64)atomic64_read(&zram->stats.invalid_io),
1055 (u64)atomic64_read(&zram->stats.notify_free));
1056 up_read(&zram->init_lock);
1057
1058 return ret;
1059}
1060
1061static ssize_t mm_stat_show(struct device *dev,
1062 struct device_attribute *attr, char *buf)
1063{
1064 struct zram *zram = dev_to_zram(dev);
1065 struct zs_pool_stats pool_stats;
1066 u64 orig_size, mem_used = 0;
1067 long max_used;
1068 ssize_t ret;
1069
1070 memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
1071
1072 down_read(&zram->init_lock);
1073 if (init_done(zram)) {
1074 mem_used = zs_get_total_pages(zram->mem_pool);
1075 zs_pool_stats(zram->mem_pool, &pool_stats);
1076 }
1077
1078 orig_size = atomic64_read(&zram->stats.pages_stored);
1079 max_used = atomic_long_read(&zram->stats.max_used_pages);
1080
1081 ret = scnprintf(buf, PAGE_SIZE,
1082 "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu %8llu\n",
1083 orig_size << PAGE_SHIFT,
1084 (u64)atomic64_read(&zram->stats.compr_data_size),
1085 mem_used << PAGE_SHIFT,
1086 zram->limit_pages << PAGE_SHIFT,
1087 max_used << PAGE_SHIFT,
1088 (u64)atomic64_read(&zram->stats.same_pages),
1089 atomic_long_read(&pool_stats.pages_compacted),
1090 (u64)atomic64_read(&zram->stats.huge_pages),
1091 (u64)atomic64_read(&zram->stats.huge_pages_since));
1092 up_read(&zram->init_lock);
1093
1094 return ret;
1095}
1096
1097#ifdef CONFIG_ZRAM_WRITEBACK
1098#define FOUR_K(x) ((x) * (1 << (PAGE_SHIFT - 12)))
1099static ssize_t bd_stat_show(struct device *dev,
1100 struct device_attribute *attr, char *buf)
1101{
1102 struct zram *zram = dev_to_zram(dev);
1103 ssize_t ret;
1104
1105 down_read(&zram->init_lock);
1106 ret = scnprintf(buf, PAGE_SIZE,
1107 "%8llu %8llu %8llu\n",
1108 FOUR_K((u64)atomic64_read(&zram->stats.bd_count)),
1109 FOUR_K((u64)atomic64_read(&zram->stats.bd_reads)),
1110 FOUR_K((u64)atomic64_read(&zram->stats.bd_writes)));
1111 up_read(&zram->init_lock);
1112
1113 return ret;
1114}
1115#endif
1116
1117static ssize_t debug_stat_show(struct device *dev,
1118 struct device_attribute *attr, char *buf)
1119{
1120 int version = 1;
1121 struct zram *zram = dev_to_zram(dev);
1122 ssize_t ret;
1123
1124 down_read(&zram->init_lock);
1125 ret = scnprintf(buf, PAGE_SIZE,
1126 "version: %d\n%8llu %8llu\n",
1127 version,
1128 (u64)atomic64_read(&zram->stats.writestall),
1129 (u64)atomic64_read(&zram->stats.miss_free));
1130 up_read(&zram->init_lock);
1131
1132 return ret;
1133}
1134
1135static DEVICE_ATTR_RO(io_stat);
1136static DEVICE_ATTR_RO(mm_stat);
1137#ifdef CONFIG_ZRAM_WRITEBACK
1138static DEVICE_ATTR_RO(bd_stat);
1139#endif
1140static DEVICE_ATTR_RO(debug_stat);
1141
1142static void zram_meta_free(struct zram *zram, u64 disksize)
1143{
1144 size_t num_pages = disksize >> PAGE_SHIFT;
1145 size_t index;
1146
1147 /* Free all pages that are still in this zram device */
1148 for (index = 0; index < num_pages; index++)
1149 zram_free_page(zram, index);
1150
1151 zs_destroy_pool(zram->mem_pool);
1152 vfree(zram->table);
1153}
1154
1155static bool zram_meta_alloc(struct zram *zram, u64 disksize)
1156{
1157 size_t num_pages;
1158
1159 num_pages = disksize >> PAGE_SHIFT;
1160 zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
1161 if (!zram->table)
1162 return false;
1163
1164 zram->mem_pool = zs_create_pool(zram->disk->disk_name);
1165 if (!zram->mem_pool) {
1166 vfree(zram->table);
1167 return false;
1168 }
1169
1170 if (!huge_class_size)
1171 huge_class_size = zs_huge_class_size(zram->mem_pool);
1172 return true;
1173}
1174
1175/*
1176 * To protect concurrent access to the same index entry,
1177 * caller should hold this table index entry's bit_spinlock to
1178 * indicate this index entry is accessing.
1179 */
1180static void zram_free_page(struct zram *zram, size_t index)
1181{
1182 unsigned long handle;
1183
1184#ifdef CONFIG_ZRAM_MEMORY_TRACKING
1185 zram->table[index].ac_time = 0;
1186#endif
1187 if (zram_test_flag(zram, index, ZRAM_IDLE))
1188 zram_clear_flag(zram, index, ZRAM_IDLE);
1189
1190 if (zram_test_flag(zram, index, ZRAM_HUGE)) {
1191 zram_clear_flag(zram, index, ZRAM_HUGE);
1192 atomic64_dec(&zram->stats.huge_pages);
1193 }
1194
1195 if (zram_test_flag(zram, index, ZRAM_WB)) {
1196 zram_clear_flag(zram, index, ZRAM_WB);
1197 free_block_bdev(zram, zram_get_element(zram, index));
1198 goto out;
1199 }
1200
1201 /*
1202 * No memory is allocated for same element filled pages.
1203 * Simply clear same page flag.
1204 */
1205 if (zram_test_flag(zram, index, ZRAM_SAME)) {
1206 zram_clear_flag(zram, index, ZRAM_SAME);
1207 atomic64_dec(&zram->stats.same_pages);
1208 goto out;
1209 }
1210
1211 handle = zram_get_handle(zram, index);
1212 if (!handle)
1213 return;
1214
1215 zs_free(zram->mem_pool, handle);
1216
1217 atomic64_sub(zram_get_obj_size(zram, index),
1218 &zram->stats.compr_data_size);
1219out:
1220 atomic64_dec(&zram->stats.pages_stored);
1221 zram_set_handle(zram, index, 0);
1222 zram_set_obj_size(zram, index, 0);
1223 WARN_ON_ONCE(zram->table[index].flags &
1224 ~(1UL << ZRAM_LOCK | 1UL << ZRAM_UNDER_WB));
1225}
1226
1227static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
1228 struct bio *bio, bool partial_io)
1229{
1230 struct zcomp_strm *zstrm;
1231 unsigned long handle;
1232 unsigned int size;
1233 void *src, *dst;
1234 int ret;
1235
1236 zram_slot_lock(zram, index);
1237 if (zram_test_flag(zram, index, ZRAM_WB)) {
1238 struct bio_vec bvec;
1239
1240 zram_slot_unlock(zram, index);
1241
1242 bvec.bv_page = page;
1243 bvec.bv_len = PAGE_SIZE;
1244 bvec.bv_offset = 0;
1245 return read_from_bdev(zram, &bvec,
1246 zram_get_element(zram, index),
1247 bio, partial_io);
1248 }
1249
1250 handle = zram_get_handle(zram, index);
1251 if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1252 unsigned long value;
1253 void *mem;
1254
1255 value = handle ? zram_get_element(zram, index) : 0;
1256 mem = kmap_atomic(page);
1257 zram_fill_page(mem, PAGE_SIZE, value);
1258 kunmap_atomic(mem);
1259 zram_slot_unlock(zram, index);
1260 return 0;
1261 }
1262
1263 size = zram_get_obj_size(zram, index);
1264
1265 if (size != PAGE_SIZE)
1266 zstrm = zcomp_stream_get(zram->comp);
1267
1268 src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1269 if (size == PAGE_SIZE) {
1270 dst = kmap_atomic(page);
1271 memcpy(dst, src, PAGE_SIZE);
1272 kunmap_atomic(dst);
1273 ret = 0;
1274 } else {
1275 dst = kmap_atomic(page);
1276 ret = zcomp_decompress(zstrm, src, size, dst);
1277 kunmap_atomic(dst);
1278 zcomp_stream_put(zram->comp);
1279 }
1280 zs_unmap_object(zram->mem_pool, handle);
1281 zram_slot_unlock(zram, index);
1282
1283 /* Should NEVER happen. Return bio error if it does. */
1284 if (WARN_ON(ret))
1285 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1286
1287 return ret;
1288}
1289
1290static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1291 u32 index, int offset, struct bio *bio)
1292{
1293 int ret;
1294 struct page *page;
1295
1296 page = bvec->bv_page;
1297 if (is_partial_io(bvec)) {
1298 /* Use a temporary buffer to decompress the page */
1299 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1300 if (!page)
1301 return -ENOMEM;
1302 }
1303
1304 ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
1305 if (unlikely(ret))
1306 goto out;
1307
1308 if (is_partial_io(bvec)) {
1309 void *dst = kmap_atomic(bvec->bv_page);
1310 void *src = kmap_atomic(page);
1311
1312 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
1313 kunmap_atomic(src);
1314 kunmap_atomic(dst);
1315 }
1316out:
1317 if (is_partial_io(bvec))
1318 __free_page(page);
1319
1320 return ret;
1321}
1322
1323static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1324 u32 index, struct bio *bio)
1325{
1326 int ret = 0;
1327 unsigned long alloced_pages;
1328 unsigned long handle = 0;
1329 unsigned int comp_len = 0;
1330 void *src, *dst, *mem;
1331 struct zcomp_strm *zstrm;
1332 struct page *page = bvec->bv_page;
1333 unsigned long element = 0;
1334 enum zram_pageflags flags = 0;
1335
1336 mem = kmap_atomic(page);
1337 if (page_same_filled(mem, &element)) {
1338 kunmap_atomic(mem);
1339 /* Free memory associated with this sector now. */
1340 flags = ZRAM_SAME;
1341 atomic64_inc(&zram->stats.same_pages);
1342 goto out;
1343 }
1344 kunmap_atomic(mem);
1345
1346compress_again:
1347 zstrm = zcomp_stream_get(zram->comp);
1348 src = kmap_atomic(page);
1349 ret = zcomp_compress(zstrm, src, &comp_len);
1350 kunmap_atomic(src);
1351
1352 if (unlikely(ret)) {
1353 zcomp_stream_put(zram->comp);
1354 pr_err("Compression failed! err=%d\n", ret);
1355 zs_free(zram->mem_pool, handle);
1356 return ret;
1357 }
1358
1359 if (comp_len >= huge_class_size)
1360 comp_len = PAGE_SIZE;
1361 /*
1362 * handle allocation has 2 paths:
1363 * a) fast path is executed with preemption disabled (for
1364 * per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1365 * since we can't sleep;
1366 * b) slow path enables preemption and attempts to allocate
1367 * the page with __GFP_DIRECT_RECLAIM bit set. we have to
1368 * put per-cpu compression stream and, thus, to re-do
1369 * the compression once handle is allocated.
1370 *
1371 * if we have a 'non-null' handle here then we are coming
1372 * from the slow path and handle has already been allocated.
1373 */
1374 if (!handle)
1375 handle = zs_malloc(zram->mem_pool, comp_len,
1376 __GFP_KSWAPD_RECLAIM |
1377 __GFP_NOWARN |
1378 __GFP_HIGHMEM |
1379 __GFP_MOVABLE);
1380 if (!handle) {
1381 zcomp_stream_put(zram->comp);
1382 atomic64_inc(&zram->stats.writestall);
1383 handle = zs_malloc(zram->mem_pool, comp_len,
1384 GFP_NOIO | __GFP_HIGHMEM |
1385 __GFP_MOVABLE);
1386 if (handle)
1387 goto compress_again;
1388 return -ENOMEM;
1389 }
1390
1391 alloced_pages = zs_get_total_pages(zram->mem_pool);
1392 update_used_max(zram, alloced_pages);
1393
1394 if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1395 zcomp_stream_put(zram->comp);
1396 zs_free(zram->mem_pool, handle);
1397 return -ENOMEM;
1398 }
1399
1400 dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1401
1402 src = zstrm->buffer;
1403 if (comp_len == PAGE_SIZE)
1404 src = kmap_atomic(page);
1405 memcpy(dst, src, comp_len);
1406 if (comp_len == PAGE_SIZE)
1407 kunmap_atomic(src);
1408
1409 zcomp_stream_put(zram->comp);
1410 zs_unmap_object(zram->mem_pool, handle);
1411 atomic64_add(comp_len, &zram->stats.compr_data_size);
1412out:
1413 /*
1414 * Free memory associated with this sector
1415 * before overwriting unused sectors.
1416 */
1417 zram_slot_lock(zram, index);
1418 zram_free_page(zram, index);
1419
1420 if (comp_len == PAGE_SIZE) {
1421 zram_set_flag(zram, index, ZRAM_HUGE);
1422 atomic64_inc(&zram->stats.huge_pages);
1423 atomic64_inc(&zram->stats.huge_pages_since);
1424 }
1425
1426 if (flags) {
1427 zram_set_flag(zram, index, flags);
1428 zram_set_element(zram, index, element);
1429 } else {
1430 zram_set_handle(zram, index, handle);
1431 zram_set_obj_size(zram, index, comp_len);
1432 }
1433 zram_slot_unlock(zram, index);
1434
1435 /* Update stats */
1436 atomic64_inc(&zram->stats.pages_stored);
1437 return ret;
1438}
1439
1440static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1441 u32 index, int offset, struct bio *bio)
1442{
1443 int ret;
1444 struct page *page = NULL;
1445 void *src;
1446 struct bio_vec vec;
1447
1448 vec = *bvec;
1449 if (is_partial_io(bvec)) {
1450 void *dst;
1451 /*
1452 * This is a partial IO. We need to read the full page
1453 * before to write the changes.
1454 */
1455 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1456 if (!page)
1457 return -ENOMEM;
1458
1459 ret = __zram_bvec_read(zram, page, index, bio, true);
1460 if (ret)
1461 goto out;
1462
1463 src = kmap_atomic(bvec->bv_page);
1464 dst = kmap_atomic(page);
1465 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1466 kunmap_atomic(dst);
1467 kunmap_atomic(src);
1468
1469 vec.bv_page = page;
1470 vec.bv_len = PAGE_SIZE;
1471 vec.bv_offset = 0;
1472 }
1473
1474 ret = __zram_bvec_write(zram, &vec, index, bio);
1475out:
1476 if (is_partial_io(bvec))
1477 __free_page(page);
1478 return ret;
1479}
1480
1481/*
1482 * zram_bio_discard - handler on discard request
1483 * @index: physical block index in PAGE_SIZE units
1484 * @offset: byte offset within physical block
1485 */
1486static void zram_bio_discard(struct zram *zram, u32 index,
1487 int offset, struct bio *bio)
1488{
1489 size_t n = bio->bi_iter.bi_size;
1490
1491 /*
1492 * zram manages data in physical block size units. Because logical block
1493 * size isn't identical with physical block size on some arch, we
1494 * could get a discard request pointing to a specific offset within a
1495 * certain physical block. Although we can handle this request by
1496 * reading that physiclal block and decompressing and partially zeroing
1497 * and re-compressing and then re-storing it, this isn't reasonable
1498 * because our intent with a discard request is to save memory. So
1499 * skipping this logical block is appropriate here.
1500 */
1501 if (offset) {
1502 if (n <= (PAGE_SIZE - offset))
1503 return;
1504
1505 n -= (PAGE_SIZE - offset);
1506 index++;
1507 }
1508
1509 while (n >= PAGE_SIZE) {
1510 zram_slot_lock(zram, index);
1511 zram_free_page(zram, index);
1512 zram_slot_unlock(zram, index);
1513 atomic64_inc(&zram->stats.notify_free);
1514 index++;
1515 n -= PAGE_SIZE;
1516 }
1517}
1518
1519/*
1520 * Returns errno if it has some problem. Otherwise return 0 or 1.
1521 * Returns 0 if IO request was done synchronously
1522 * Returns 1 if IO request was successfully submitted.
1523 */
1524static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1525 int offset, unsigned int op, struct bio *bio)
1526{
1527 int ret;
1528
1529 if (!op_is_write(op)) {
1530 atomic64_inc(&zram->stats.num_reads);
1531 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1532 flush_dcache_page(bvec->bv_page);
1533 } else {
1534 atomic64_inc(&zram->stats.num_writes);
1535 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1536 }
1537
1538 zram_slot_lock(zram, index);
1539 zram_accessed(zram, index);
1540 zram_slot_unlock(zram, index);
1541
1542 if (unlikely(ret < 0)) {
1543 if (!op_is_write(op))
1544 atomic64_inc(&zram->stats.failed_reads);
1545 else
1546 atomic64_inc(&zram->stats.failed_writes);
1547 }
1548
1549 return ret;
1550}
1551
1552static void __zram_make_request(struct zram *zram, struct bio *bio)
1553{
1554 int offset;
1555 u32 index;
1556 struct bio_vec bvec;
1557 struct bvec_iter iter;
1558 unsigned long start_time;
1559
1560 index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1561 offset = (bio->bi_iter.bi_sector &
1562 (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1563
1564 switch (bio_op(bio)) {
1565 case REQ_OP_DISCARD:
1566 case REQ_OP_WRITE_ZEROES:
1567 zram_bio_discard(zram, index, offset, bio);
1568 bio_endio(bio);
1569 return;
1570 default:
1571 break;
1572 }
1573
1574 start_time = bio_start_io_acct(bio);
1575 bio_for_each_segment(bvec, bio, iter) {
1576 struct bio_vec bv = bvec;
1577 unsigned int unwritten = bvec.bv_len;
1578
1579 do {
1580 bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1581 unwritten);
1582 if (zram_bvec_rw(zram, &bv, index, offset,
1583 bio_op(bio), bio) < 0) {
1584 bio->bi_status = BLK_STS_IOERR;
1585 break;
1586 }
1587
1588 bv.bv_offset += bv.bv_len;
1589 unwritten -= bv.bv_len;
1590
1591 update_position(&index, &offset, &bv);
1592 } while (unwritten);
1593 }
1594 bio_end_io_acct(bio, start_time);
1595 bio_endio(bio);
1596}
1597
1598/*
1599 * Handler function for all zram I/O requests.
1600 */
1601static blk_qc_t zram_submit_bio(struct bio *bio)
1602{
1603 struct zram *zram = bio->bi_bdev->bd_disk->private_data;
1604
1605 if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1606 bio->bi_iter.bi_size)) {
1607 atomic64_inc(&zram->stats.invalid_io);
1608 goto error;
1609 }
1610
1611 __zram_make_request(zram, bio);
1612 return BLK_QC_T_NONE;
1613
1614error:
1615 bio_io_error(bio);
1616 return BLK_QC_T_NONE;
1617}
1618
1619static void zram_slot_free_notify(struct block_device *bdev,
1620 unsigned long index)
1621{
1622 struct zram *zram;
1623
1624 zram = bdev->bd_disk->private_data;
1625
1626 atomic64_inc(&zram->stats.notify_free);
1627 if (!zram_slot_trylock(zram, index)) {
1628 atomic64_inc(&zram->stats.miss_free);
1629 return;
1630 }
1631
1632 zram_free_page(zram, index);
1633 zram_slot_unlock(zram, index);
1634}
1635
1636static int zram_rw_page(struct block_device *bdev, sector_t sector,
1637 struct page *page, unsigned int op)
1638{
1639 int offset, ret;
1640 u32 index;
1641 struct zram *zram;
1642 struct bio_vec bv;
1643 unsigned long start_time;
1644
1645 if (PageTransHuge(page))
1646 return -ENOTSUPP;
1647 zram = bdev->bd_disk->private_data;
1648
1649 if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1650 atomic64_inc(&zram->stats.invalid_io);
1651 ret = -EINVAL;
1652 goto out;
1653 }
1654
1655 index = sector >> SECTORS_PER_PAGE_SHIFT;
1656 offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1657
1658 bv.bv_page = page;
1659 bv.bv_len = PAGE_SIZE;
1660 bv.bv_offset = 0;
1661
1662 start_time = disk_start_io_acct(bdev->bd_disk, SECTORS_PER_PAGE, op);
1663 ret = zram_bvec_rw(zram, &bv, index, offset, op, NULL);
1664 disk_end_io_acct(bdev->bd_disk, op, start_time);
1665out:
1666 /*
1667 * If I/O fails, just return error(ie, non-zero) without
1668 * calling page_endio.
1669 * It causes resubmit the I/O with bio request by upper functions
1670 * of rw_page(e.g., swap_readpage, __swap_writepage) and
1671 * bio->bi_end_io does things to handle the error
1672 * (e.g., SetPageError, set_page_dirty and extra works).
1673 */
1674 if (unlikely(ret < 0))
1675 return ret;
1676
1677 switch (ret) {
1678 case 0:
1679 page_endio(page, op_is_write(op), 0);
1680 break;
1681 case 1:
1682 ret = 0;
1683 break;
1684 default:
1685 WARN_ON(1);
1686 }
1687 return ret;
1688}
1689
1690static void zram_reset_device(struct zram *zram)
1691{
1692 struct zcomp *comp;
1693 u64 disksize;
1694
1695 down_write(&zram->init_lock);
1696
1697 zram->limit_pages = 0;
1698
1699 if (!init_done(zram)) {
1700 up_write(&zram->init_lock);
1701 return;
1702 }
1703
1704 comp = zram->comp;
1705 disksize = zram->disksize;
1706 zram->disksize = 0;
1707
1708 set_capacity_and_notify(zram->disk, 0);
1709 part_stat_set_all(zram->disk->part0, 0);
1710
1711 up_write(&zram->init_lock);
1712 /* I/O operation under all of CPU are done so let's free */
1713 zram_meta_free(zram, disksize);
1714 memset(&zram->stats, 0, sizeof(zram->stats));
1715 zcomp_destroy(comp);
1716 reset_bdev(zram);
1717}
1718
1719static ssize_t disksize_store(struct device *dev,
1720 struct device_attribute *attr, const char *buf, size_t len)
1721{
1722 u64 disksize;
1723 struct zcomp *comp;
1724 struct zram *zram = dev_to_zram(dev);
1725 int err;
1726
1727 disksize = memparse(buf, NULL);
1728 if (!disksize)
1729 return -EINVAL;
1730
1731 down_write(&zram->init_lock);
1732 if (init_done(zram)) {
1733 pr_info("Cannot change disksize for initialized device\n");
1734 err = -EBUSY;
1735 goto out_unlock;
1736 }
1737
1738 disksize = PAGE_ALIGN(disksize);
1739 if (!zram_meta_alloc(zram, disksize)) {
1740 err = -ENOMEM;
1741 goto out_unlock;
1742 }
1743
1744 comp = zcomp_create(zram->compressor);
1745 if (IS_ERR(comp)) {
1746 pr_err("Cannot initialise %s compressing backend\n",
1747 zram->compressor);
1748 err = PTR_ERR(comp);
1749 goto out_free_meta;
1750 }
1751
1752 zram->comp = comp;
1753 zram->disksize = disksize;
1754 set_capacity_and_notify(zram->disk, zram->disksize >> SECTOR_SHIFT);
1755 up_write(&zram->init_lock);
1756
1757 return len;
1758
1759out_free_meta:
1760 zram_meta_free(zram, disksize);
1761out_unlock:
1762 up_write(&zram->init_lock);
1763 return err;
1764}
1765
1766static ssize_t reset_store(struct device *dev,
1767 struct device_attribute *attr, const char *buf, size_t len)
1768{
1769 int ret;
1770 unsigned short do_reset;
1771 struct zram *zram;
1772 struct block_device *bdev;
1773
1774 ret = kstrtou16(buf, 10, &do_reset);
1775 if (ret)
1776 return ret;
1777
1778 if (!do_reset)
1779 return -EINVAL;
1780
1781 zram = dev_to_zram(dev);
1782 bdev = zram->disk->part0;
1783
1784 mutex_lock(&bdev->bd_disk->open_mutex);
1785 /* Do not reset an active device or claimed device */
1786 if (bdev->bd_openers || zram->claim) {
1787 mutex_unlock(&bdev->bd_disk->open_mutex);
1788 return -EBUSY;
1789 }
1790
1791 /* From now on, anyone can't open /dev/zram[0-9] */
1792 zram->claim = true;
1793 mutex_unlock(&bdev->bd_disk->open_mutex);
1794
1795 /* Make sure all the pending I/O are finished */
1796 fsync_bdev(bdev);
1797 zram_reset_device(zram);
1798
1799 mutex_lock(&bdev->bd_disk->open_mutex);
1800 zram->claim = false;
1801 mutex_unlock(&bdev->bd_disk->open_mutex);
1802
1803 return len;
1804}
1805
1806static int zram_open(struct block_device *bdev, fmode_t mode)
1807{
1808 int ret = 0;
1809 struct zram *zram;
1810
1811 WARN_ON(!mutex_is_locked(&bdev->bd_disk->open_mutex));
1812
1813 zram = bdev->bd_disk->private_data;
1814 /* zram was claimed to reset so open request fails */
1815 if (zram->claim)
1816 ret = -EBUSY;
1817
1818 return ret;
1819}
1820
1821static const struct block_device_operations zram_devops = {
1822 .open = zram_open,
1823 .submit_bio = zram_submit_bio,
1824 .swap_slot_free_notify = zram_slot_free_notify,
1825 .rw_page = zram_rw_page,
1826 .owner = THIS_MODULE
1827};
1828
1829static const struct block_device_operations zram_wb_devops = {
1830 .open = zram_open,
1831 .submit_bio = zram_submit_bio,
1832 .swap_slot_free_notify = zram_slot_free_notify,
1833 .owner = THIS_MODULE
1834};
1835
1836static DEVICE_ATTR_WO(compact);
1837static DEVICE_ATTR_RW(disksize);
1838static DEVICE_ATTR_RO(initstate);
1839static DEVICE_ATTR_WO(reset);
1840static DEVICE_ATTR_WO(mem_limit);
1841static DEVICE_ATTR_WO(mem_used_max);
1842static DEVICE_ATTR_WO(idle);
1843static DEVICE_ATTR_RW(max_comp_streams);
1844static DEVICE_ATTR_RW(comp_algorithm);
1845#ifdef CONFIG_ZRAM_WRITEBACK
1846static DEVICE_ATTR_RW(backing_dev);
1847static DEVICE_ATTR_WO(writeback);
1848static DEVICE_ATTR_RW(writeback_limit);
1849static DEVICE_ATTR_RW(writeback_limit_enable);
1850#endif
1851
1852static struct attribute *zram_disk_attrs[] = {
1853 &dev_attr_disksize.attr,
1854 &dev_attr_initstate.attr,
1855 &dev_attr_reset.attr,
1856 &dev_attr_compact.attr,
1857 &dev_attr_mem_limit.attr,
1858 &dev_attr_mem_used_max.attr,
1859 &dev_attr_idle.attr,
1860 &dev_attr_max_comp_streams.attr,
1861 &dev_attr_comp_algorithm.attr,
1862#ifdef CONFIG_ZRAM_WRITEBACK
1863 &dev_attr_backing_dev.attr,
1864 &dev_attr_writeback.attr,
1865 &dev_attr_writeback_limit.attr,
1866 &dev_attr_writeback_limit_enable.attr,
1867#endif
1868 &dev_attr_io_stat.attr,
1869 &dev_attr_mm_stat.attr,
1870#ifdef CONFIG_ZRAM_WRITEBACK
1871 &dev_attr_bd_stat.attr,
1872#endif
1873 &dev_attr_debug_stat.attr,
1874 NULL,
1875};
1876
1877static const struct attribute_group zram_disk_attr_group = {
1878 .attrs = zram_disk_attrs,
1879};
1880
1881static const struct attribute_group *zram_disk_attr_groups[] = {
1882 &zram_disk_attr_group,
1883 NULL,
1884};
1885
1886/*
1887 * Allocate and initialize new zram device. the function returns
1888 * '>= 0' device_id upon success, and negative value otherwise.
1889 */
1890static int zram_add(void)
1891{
1892 struct zram *zram;
1893 int ret, device_id;
1894
1895 zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1896 if (!zram)
1897 return -ENOMEM;
1898
1899 ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1900 if (ret < 0)
1901 goto out_free_dev;
1902 device_id = ret;
1903
1904 init_rwsem(&zram->init_lock);
1905#ifdef CONFIG_ZRAM_WRITEBACK
1906 spin_lock_init(&zram->wb_limit_lock);
1907#endif
1908
1909 /* gendisk structure */
1910 zram->disk = blk_alloc_disk(NUMA_NO_NODE);
1911 if (!zram->disk) {
1912 pr_err("Error allocating disk structure for device %d\n",
1913 device_id);
1914 ret = -ENOMEM;
1915 goto out_free_idr;
1916 }
1917
1918 zram->disk->major = zram_major;
1919 zram->disk->first_minor = device_id;
1920 zram->disk->minors = 1;
1921 zram->disk->fops = &zram_devops;
1922 zram->disk->private_data = zram;
1923 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1924
1925 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1926 set_capacity(zram->disk, 0);
1927 /* zram devices sort of resembles non-rotational disks */
1928 blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
1929 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1930
1931 /*
1932 * To ensure that we always get PAGE_SIZE aligned
1933 * and n*PAGE_SIZED sized I/O requests.
1934 */
1935 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1936 blk_queue_logical_block_size(zram->disk->queue,
1937 ZRAM_LOGICAL_BLOCK_SIZE);
1938 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1939 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1940 zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1941 blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1942 blk_queue_flag_set(QUEUE_FLAG_DISCARD, zram->disk->queue);
1943
1944 /*
1945 * zram_bio_discard() will clear all logical blocks if logical block
1946 * size is identical with physical block size(PAGE_SIZE). But if it is
1947 * different, we will skip discarding some parts of logical blocks in
1948 * the part of the request range which isn't aligned to physical block
1949 * size. So we can't ensure that all discarded logical blocks are
1950 * zeroed.
1951 */
1952 if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1953 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1954
1955 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, zram->disk->queue);
1956 device_add_disk(NULL, zram->disk, zram_disk_attr_groups);
1957
1958 strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1959
1960 zram_debugfs_register(zram);
1961 pr_info("Added device: %s\n", zram->disk->disk_name);
1962 return device_id;
1963
1964out_free_idr:
1965 idr_remove(&zram_index_idr, device_id);
1966out_free_dev:
1967 kfree(zram);
1968 return ret;
1969}
1970
1971static int zram_remove(struct zram *zram)
1972{
1973 struct block_device *bdev = zram->disk->part0;
1974
1975 mutex_lock(&bdev->bd_disk->open_mutex);
1976 if (bdev->bd_openers || zram->claim) {
1977 mutex_unlock(&bdev->bd_disk->open_mutex);
1978 return -EBUSY;
1979 }
1980
1981 zram->claim = true;
1982 mutex_unlock(&bdev->bd_disk->open_mutex);
1983
1984 zram_debugfs_unregister(zram);
1985
1986 /* Make sure all the pending I/O are finished */
1987 fsync_bdev(bdev);
1988 zram_reset_device(zram);
1989
1990 pr_info("Removed device: %s\n", zram->disk->disk_name);
1991
1992 del_gendisk(zram->disk);
1993 blk_cleanup_disk(zram->disk);
1994 kfree(zram);
1995 return 0;
1996}
1997
1998/* zram-control sysfs attributes */
1999
2000/*
2001 * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
2002 * sense that reading from this file does alter the state of your system -- it
2003 * creates a new un-initialized zram device and returns back this device's
2004 * device_id (or an error code if it fails to create a new device).
2005 */
2006static ssize_t hot_add_show(struct class *class,
2007 struct class_attribute *attr,
2008 char *buf)
2009{
2010 int ret;
2011
2012 mutex_lock(&zram_index_mutex);
2013 ret = zram_add();
2014 mutex_unlock(&zram_index_mutex);
2015
2016 if (ret < 0)
2017 return ret;
2018 return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
2019}
2020static struct class_attribute class_attr_hot_add =
2021 __ATTR(hot_add, 0400, hot_add_show, NULL);
2022
2023static ssize_t hot_remove_store(struct class *class,
2024 struct class_attribute *attr,
2025 const char *buf,
2026 size_t count)
2027{
2028 struct zram *zram;
2029 int ret, dev_id;
2030
2031 /* dev_id is gendisk->first_minor, which is `int' */
2032 ret = kstrtoint(buf, 10, &dev_id);
2033 if (ret)
2034 return ret;
2035 if (dev_id < 0)
2036 return -EINVAL;
2037
2038 mutex_lock(&zram_index_mutex);
2039
2040 zram = idr_find(&zram_index_idr, dev_id);
2041 if (zram) {
2042 ret = zram_remove(zram);
2043 if (!ret)
2044 idr_remove(&zram_index_idr, dev_id);
2045 } else {
2046 ret = -ENODEV;
2047 }
2048
2049 mutex_unlock(&zram_index_mutex);
2050 return ret ? ret : count;
2051}
2052static CLASS_ATTR_WO(hot_remove);
2053
2054static struct attribute *zram_control_class_attrs[] = {
2055 &class_attr_hot_add.attr,
2056 &class_attr_hot_remove.attr,
2057 NULL,
2058};
2059ATTRIBUTE_GROUPS(zram_control_class);
2060
2061static struct class zram_control_class = {
2062 .name = "zram-control",
2063 .owner = THIS_MODULE,
2064 .class_groups = zram_control_class_groups,
2065};
2066
2067static int zram_remove_cb(int id, void *ptr, void *data)
2068{
2069 zram_remove(ptr);
2070 return 0;
2071}
2072
2073static void destroy_devices(void)
2074{
2075 class_unregister(&zram_control_class);
2076 idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
2077 zram_debugfs_destroy();
2078 idr_destroy(&zram_index_idr);
2079 unregister_blkdev(zram_major, "zram");
2080 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2081}
2082
2083static int __init zram_init(void)
2084{
2085 int ret;
2086
2087 ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
2088 zcomp_cpu_up_prepare, zcomp_cpu_dead);
2089 if (ret < 0)
2090 return ret;
2091
2092 ret = class_register(&zram_control_class);
2093 if (ret) {
2094 pr_err("Unable to register zram-control class\n");
2095 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2096 return ret;
2097 }
2098
2099 zram_debugfs_create();
2100 zram_major = register_blkdev(0, "zram");
2101 if (zram_major <= 0) {
2102 pr_err("Unable to get major number\n");
2103 class_unregister(&zram_control_class);
2104 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2105 return -EBUSY;
2106 }
2107
2108 while (num_devices != 0) {
2109 mutex_lock(&zram_index_mutex);
2110 ret = zram_add();
2111 mutex_unlock(&zram_index_mutex);
2112 if (ret < 0)
2113 goto out_error;
2114 num_devices--;
2115 }
2116
2117 return 0;
2118
2119out_error:
2120 destroy_devices();
2121 return ret;
2122}
2123
2124static void __exit zram_exit(void)
2125{
2126 destroy_devices();
2127}
2128
2129module_init(zram_init);
2130module_exit(zram_exit);
2131
2132module_param(num_devices, uint, 0);
2133MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
2134
2135MODULE_LICENSE("Dual BSD/GPL");
2136MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
2137MODULE_DESCRIPTION("Compressed RAM Block Device");