projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge tag 'sound-4.2' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound
[linux-2.6-block.git] / drivers / block / zram / zram_drv.c
... / ...
CommitLineData
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/string.h>
29#include <linux/vmalloc.h>
30#include <linux/err.h>
31#include <linux/idr.h>
32#include <linux/sysfs.h>
33
34#include "zram_drv.h"
35
36static DEFINE_IDR(zram_index_idr);
37/* idr index must be protected */
38static DEFINE_MUTEX(zram_index_mutex);
39
40static int zram_major;
41static const char *default_compressor = "lzo";
42
43/* Module params (documentation at end) */
44static unsigned int num_devices = 1;
45
46static inline void deprecated_attr_warn(const char *name)
47{
48 pr_warn_once("%d (%s) Attribute %s (and others) will be removed. %s\n",
49 task_pid_nr(current),
50 current->comm,
51 name,
52 "See zram documentation.");
53}
54
55#define ZRAM_ATTR_RO(name) \
56static ssize_t name##_show(struct device *d, \
57 struct device_attribute *attr, char *b) \
58{ \
59 struct zram *zram = dev_to_zram(d); \
60 \
61 deprecated_attr_warn(__stringify(name)); \
62 return scnprintf(b, PAGE_SIZE, "%llu\n", \
63 (u64)atomic64_read(&zram->stats.name)); \
64} \
65static DEVICE_ATTR_RO(name);
66
67static inline bool init_done(struct zram *zram)
68{
69 return zram->disksize;
70}
71
72static inline struct zram *dev_to_zram(struct device *dev)
73{
74 return (struct zram *)dev_to_disk(dev)->private_data;
75}
76
77/* flag operations require table entry bit_spin_lock() being held */
78static int zram_test_flag(struct zram_meta *meta, u32 index,
79 enum zram_pageflags flag)
80{
81 return meta->table[index].value & BIT(flag);
82}
83
84static void zram_set_flag(struct zram_meta *meta, u32 index,
85 enum zram_pageflags flag)
86{
87 meta->table[index].value |= BIT(flag);
88}
89
90static void zram_clear_flag(struct zram_meta *meta, u32 index,
91 enum zram_pageflags flag)
92{
93 meta->table[index].value &= ~BIT(flag);
94}
95
96static size_t zram_get_obj_size(struct zram_meta *meta, u32 index)
97{
98 return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
99}
100
101static void zram_set_obj_size(struct zram_meta *meta,
102 u32 index, size_t size)
103{
104 unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT;
105
106 meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
107}
108
109static inline int is_partial_io(struct bio_vec *bvec)
110{
111 return bvec->bv_len != PAGE_SIZE;
112}
113
114/*
115 * Check if request is within bounds and aligned on zram logical blocks.
116 */
117static inline int valid_io_request(struct zram *zram,
118 sector_t start, unsigned int size)
119{
120 u64 end, bound;
121
122 /* unaligned request */
123 if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
124 return 0;
125 if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
126 return 0;
127
128 end = start + (size >> SECTOR_SHIFT);
129 bound = zram->disksize >> SECTOR_SHIFT;
130 /* out of range range */
131 if (unlikely(start >= bound || end > bound || start > end))
132 return 0;
133
134 /* I/O request is valid */
135 return 1;
136}
137
138static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
139{
140 if (*offset + bvec->bv_len >= PAGE_SIZE)
141 (*index)++;
142 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
143}
144
145static inline void update_used_max(struct zram *zram,
146 const unsigned long pages)
147{
148 unsigned long old_max, cur_max;
149
150 old_max = atomic_long_read(&zram->stats.max_used_pages);
151
152 do {
153 cur_max = old_max;
154 if (pages > cur_max)
155 old_max = atomic_long_cmpxchg(
156 &zram->stats.max_used_pages, cur_max, pages);
157 } while (old_max != cur_max);
158}
159
160static int page_zero_filled(void *ptr)
161{
162 unsigned int pos;
163 unsigned long *page;
164
165 page = (unsigned long *)ptr;
166
167 for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
168 if (page[pos])
169 return 0;
170 }
171
172 return 1;
173}
174
175static void handle_zero_page(struct bio_vec *bvec)
176{
177 struct page *page = bvec->bv_page;
178 void *user_mem;
179
180 user_mem = kmap_atomic(page);
181 if (is_partial_io(bvec))
182 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
183 else
184 clear_page(user_mem);
185 kunmap_atomic(user_mem);
186
187 flush_dcache_page(page);
188}
189
190static ssize_t initstate_show(struct device *dev,
191 struct device_attribute *attr, char *buf)
192{
193 u32 val;
194 struct zram *zram = dev_to_zram(dev);
195
196 down_read(&zram->init_lock);
197 val = init_done(zram);
198 up_read(&zram->init_lock);
199
200 return scnprintf(buf, PAGE_SIZE, "%u\n", val);
201}
202
203static ssize_t disksize_show(struct device *dev,
204 struct device_attribute *attr, char *buf)
205{
206 struct zram *zram = dev_to_zram(dev);
207
208 return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
209}
210
211static ssize_t orig_data_size_show(struct device *dev,
212 struct device_attribute *attr, char *buf)
213{
214 struct zram *zram = dev_to_zram(dev);
215
216 deprecated_attr_warn("orig_data_size");
217 return scnprintf(buf, PAGE_SIZE, "%llu\n",
218 (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
219}
220
221static ssize_t mem_used_total_show(struct device *dev,
222 struct device_attribute *attr, char *buf)
223{
224 u64 val = 0;
225 struct zram *zram = dev_to_zram(dev);
226
227 deprecated_attr_warn("mem_used_total");
228 down_read(&zram->init_lock);
229 if (init_done(zram)) {
230 struct zram_meta *meta = zram->meta;
231 val = zs_get_total_pages(meta->mem_pool);
232 }
233 up_read(&zram->init_lock);
234
235 return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
236}
237
238static ssize_t mem_limit_show(struct device *dev,
239 struct device_attribute *attr, char *buf)
240{
241 u64 val;
242 struct zram *zram = dev_to_zram(dev);
243
244 deprecated_attr_warn("mem_limit");
245 down_read(&zram->init_lock);
246 val = zram->limit_pages;
247 up_read(&zram->init_lock);
248
249 return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
250}
251
252static ssize_t mem_limit_store(struct device *dev,
253 struct device_attribute *attr, const char *buf, size_t len)
254{
255 u64 limit;
256 char *tmp;
257 struct zram *zram = dev_to_zram(dev);
258
259 limit = memparse(buf, &tmp);
260 if (buf == tmp) /* no chars parsed, invalid input */
261 return -EINVAL;
262
263 down_write(&zram->init_lock);
264 zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
265 up_write(&zram->init_lock);
266
267 return len;
268}
269
270static ssize_t mem_used_max_show(struct device *dev,
271 struct device_attribute *attr, char *buf)
272{
273 u64 val = 0;
274 struct zram *zram = dev_to_zram(dev);
275
276 deprecated_attr_warn("mem_used_max");
277 down_read(&zram->init_lock);
278 if (init_done(zram))
279 val = atomic_long_read(&zram->stats.max_used_pages);
280 up_read(&zram->init_lock);
281
282 return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
283}
284
285static ssize_t mem_used_max_store(struct device *dev,
286 struct device_attribute *attr, const char *buf, size_t len)
287{
288 int err;
289 unsigned long val;
290 struct zram *zram = dev_to_zram(dev);
291
292 err = kstrtoul(buf, 10, &val);
293 if (err || val != 0)
294 return -EINVAL;
295
296 down_read(&zram->init_lock);
297 if (init_done(zram)) {
298 struct zram_meta *meta = zram->meta;
299 atomic_long_set(&zram->stats.max_used_pages,
300 zs_get_total_pages(meta->mem_pool));
301 }
302 up_read(&zram->init_lock);
303
304 return len;
305}
306
307static ssize_t max_comp_streams_show(struct device *dev,
308 struct device_attribute *attr, char *buf)
309{
310 int val;
311 struct zram *zram = dev_to_zram(dev);
312
313 down_read(&zram->init_lock);
314 val = zram->max_comp_streams;
315 up_read(&zram->init_lock);
316
317 return scnprintf(buf, PAGE_SIZE, "%d\n", val);
318}
319
320static ssize_t max_comp_streams_store(struct device *dev,
321 struct device_attribute *attr, const char *buf, size_t len)
322{
323 int num;
324 struct zram *zram = dev_to_zram(dev);
325 int ret;
326
327 ret = kstrtoint(buf, 0, &num);
328 if (ret < 0)
329 return ret;
330 if (num < 1)
331 return -EINVAL;
332
333 down_write(&zram->init_lock);
334 if (init_done(zram)) {
335 if (!zcomp_set_max_streams(zram->comp, num)) {
336 pr_info("Cannot change max compression streams\n");
337 ret = -EINVAL;
338 goto out;
339 }
340 }
341
342 zram->max_comp_streams = num;
343 ret = len;
344out:
345 up_write(&zram->init_lock);
346 return ret;
347}
348
349static ssize_t comp_algorithm_show(struct device *dev,
350 struct device_attribute *attr, char *buf)
351{
352 size_t sz;
353 struct zram *zram = dev_to_zram(dev);
354
355 down_read(&zram->init_lock);
356 sz = zcomp_available_show(zram->compressor, buf);
357 up_read(&zram->init_lock);
358
359 return sz;
360}
361
362static ssize_t comp_algorithm_store(struct device *dev,
363 struct device_attribute *attr, const char *buf, size_t len)
364{
365 struct zram *zram = dev_to_zram(dev);
366 size_t sz;
367
368 down_write(&zram->init_lock);
369 if (init_done(zram)) {
370 up_write(&zram->init_lock);
371 pr_info("Can't change algorithm for initialized device\n");
372 return -EBUSY;
373 }
374 strlcpy(zram->compressor, buf, sizeof(zram->compressor));
375
376 /* ignore trailing newline */
377 sz = strlen(zram->compressor);
378 if (sz > 0 && zram->compressor[sz - 1] == '\n')
379 zram->compressor[sz - 1] = 0x00;
380
381 if (!zcomp_available_algorithm(zram->compressor))
382 len = -EINVAL;
383
384 up_write(&zram->init_lock);
385 return len;
386}
387
388static ssize_t compact_store(struct device *dev,
389 struct device_attribute *attr, const char *buf, size_t len)
390{
391 unsigned long nr_migrated;
392 struct zram *zram = dev_to_zram(dev);
393 struct zram_meta *meta;
394
395 down_read(&zram->init_lock);
396 if (!init_done(zram)) {
397 up_read(&zram->init_lock);
398 return -EINVAL;
399 }
400
401 meta = zram->meta;
402 nr_migrated = zs_compact(meta->mem_pool);
403 atomic64_add(nr_migrated, &zram->stats.num_migrated);
404 up_read(&zram->init_lock);
405
406 return len;
407}
408
409static ssize_t io_stat_show(struct device *dev,
410 struct device_attribute *attr, char *buf)
411{
412 struct zram *zram = dev_to_zram(dev);
413 ssize_t ret;
414
415 down_read(&zram->init_lock);
416 ret = scnprintf(buf, PAGE_SIZE,
417 "%8llu %8llu %8llu %8llu\n",
418 (u64)atomic64_read(&zram->stats.failed_reads),
419 (u64)atomic64_read(&zram->stats.failed_writes),
420 (u64)atomic64_read(&zram->stats.invalid_io),
421 (u64)atomic64_read(&zram->stats.notify_free));
422 up_read(&zram->init_lock);
423
424 return ret;
425}
426
427static ssize_t mm_stat_show(struct device *dev,
428 struct device_attribute *attr, char *buf)
429{
430 struct zram *zram = dev_to_zram(dev);
431 u64 orig_size, mem_used = 0;
432 long max_used;
433 ssize_t ret;
434
435 down_read(&zram->init_lock);
436 if (init_done(zram))
437 mem_used = zs_get_total_pages(zram->meta->mem_pool);
438
439 orig_size = atomic64_read(&zram->stats.pages_stored);
440 max_used = atomic_long_read(&zram->stats.max_used_pages);
441
442 ret = scnprintf(buf, PAGE_SIZE,
443 "%8llu %8llu %8llu %8lu %8ld %8llu %8llu\n",
444 orig_size << PAGE_SHIFT,
445 (u64)atomic64_read(&zram->stats.compr_data_size),
446 mem_used << PAGE_SHIFT,
447 zram->limit_pages << PAGE_SHIFT,
448 max_used << PAGE_SHIFT,
449 (u64)atomic64_read(&zram->stats.zero_pages),
450 (u64)atomic64_read(&zram->stats.num_migrated));
451 up_read(&zram->init_lock);
452
453 return ret;
454}
455
456static DEVICE_ATTR_RO(io_stat);
457static DEVICE_ATTR_RO(mm_stat);
458ZRAM_ATTR_RO(num_reads);
459ZRAM_ATTR_RO(num_writes);
460ZRAM_ATTR_RO(failed_reads);
461ZRAM_ATTR_RO(failed_writes);
462ZRAM_ATTR_RO(invalid_io);
463ZRAM_ATTR_RO(notify_free);
464ZRAM_ATTR_RO(zero_pages);
465ZRAM_ATTR_RO(compr_data_size);
466
467static inline bool zram_meta_get(struct zram *zram)
468{
469 if (atomic_inc_not_zero(&zram->refcount))
470 return true;
471 return false;
472}
473
474static inline void zram_meta_put(struct zram *zram)
475{
476 atomic_dec(&zram->refcount);
477}
478
479static void zram_meta_free(struct zram_meta *meta, u64 disksize)
480{
481 size_t num_pages = disksize >> PAGE_SHIFT;
482 size_t index;
483
484 /* Free all pages that are still in this zram device */
485 for (index = 0; index < num_pages; index++) {
486 unsigned long handle = meta->table[index].handle;
487
488 if (!handle)
489 continue;
490
491 zs_free(meta->mem_pool, handle);
492 }
493
494 zs_destroy_pool(meta->mem_pool);
495 vfree(meta->table);
496 kfree(meta);
497}
498
499static struct zram_meta *zram_meta_alloc(char *pool_name, u64 disksize)
500{
501 size_t num_pages;
502 struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
503
504 if (!meta)
505 return NULL;
506
507 num_pages = disksize >> PAGE_SHIFT;
508 meta->table = vzalloc(num_pages * sizeof(*meta->table));
509 if (!meta->table) {
510 pr_err("Error allocating zram address table\n");
511 goto out_error;
512 }
513
514 meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM);
515 if (!meta->mem_pool) {
516 pr_err("Error creating memory pool\n");
517 goto out_error;
518 }
519
520 return meta;
521
522out_error:
523 vfree(meta->table);
524 kfree(meta);
525 return NULL;
526}
527
528/*
529 * To protect concurrent access to the same index entry,
530 * caller should hold this table index entry's bit_spinlock to
531 * indicate this index entry is accessing.
532 */
533static void zram_free_page(struct zram *zram, size_t index)
534{
535 struct zram_meta *meta = zram->meta;
536 unsigned long handle = meta->table[index].handle;
537
538 if (unlikely(!handle)) {
539 /*
540 * No memory is allocated for zero filled pages.
541 * Simply clear zero page flag.
542 */
543 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
544 zram_clear_flag(meta, index, ZRAM_ZERO);
545 atomic64_dec(&zram->stats.zero_pages);
546 }
547 return;
548 }
549
550 zs_free(meta->mem_pool, handle);
551
552 atomic64_sub(zram_get_obj_size(meta, index),
553 &zram->stats.compr_data_size);
554 atomic64_dec(&zram->stats.pages_stored);
555
556 meta->table[index].handle = 0;
557 zram_set_obj_size(meta, index, 0);
558}
559
560static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
561{
562 int ret = 0;
563 unsigned char *cmem;
564 struct zram_meta *meta = zram->meta;
565 unsigned long handle;
566 size_t size;
567
568 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
569 handle = meta->table[index].handle;
570 size = zram_get_obj_size(meta, index);
571
572 if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
573 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
574 clear_page(mem);
575 return 0;
576 }
577
578 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
579 if (size == PAGE_SIZE)
580 copy_page(mem, cmem);
581 else
582 ret = zcomp_decompress(zram->comp, cmem, size, mem);
583 zs_unmap_object(meta->mem_pool, handle);
584 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
585
586 /* Should NEVER happen. Return bio error if it does. */
587 if (unlikely(ret)) {
588 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
589 return ret;
590 }
591
592 return 0;
593}
594
595static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
596 u32 index, int offset)
597{
598 int ret;
599 struct page *page;
600 unsigned char *user_mem, *uncmem = NULL;
601 struct zram_meta *meta = zram->meta;
602 page = bvec->bv_page;
603
604 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
605 if (unlikely(!meta->table[index].handle) ||
606 zram_test_flag(meta, index, ZRAM_ZERO)) {
607 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
608 handle_zero_page(bvec);
609 return 0;
610 }
611 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
612
613 if (is_partial_io(bvec))
614 /* Use a temporary buffer to decompress the page */
615 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
616
617 user_mem = kmap_atomic(page);
618 if (!is_partial_io(bvec))
619 uncmem = user_mem;
620
621 if (!uncmem) {
622 pr_info("Unable to allocate temp memory\n");
623 ret = -ENOMEM;
624 goto out_cleanup;
625 }
626
627 ret = zram_decompress_page(zram, uncmem, index);
628 /* Should NEVER happen. Return bio error if it does. */
629 if (unlikely(ret))
630 goto out_cleanup;
631
632 if (is_partial_io(bvec))
633 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
634 bvec->bv_len);
635
636 flush_dcache_page(page);
637 ret = 0;
638out_cleanup:
639 kunmap_atomic(user_mem);
640 if (is_partial_io(bvec))
641 kfree(uncmem);
642 return ret;
643}
644
645static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
646 int offset)
647{
648 int ret = 0;
649 size_t clen;
650 unsigned long handle;
651 struct page *page;
652 unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
653 struct zram_meta *meta = zram->meta;
654 struct zcomp_strm *zstrm = NULL;
655 unsigned long alloced_pages;
656
657 page = bvec->bv_page;
658 if (is_partial_io(bvec)) {
659 /*
660 * This is a partial IO. We need to read the full page
661 * before to write the changes.
662 */
663 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
664 if (!uncmem) {
665 ret = -ENOMEM;
666 goto out;
667 }
668 ret = zram_decompress_page(zram, uncmem, index);
669 if (ret)
670 goto out;
671 }
672
673 zstrm = zcomp_strm_find(zram->comp);
674 user_mem = kmap_atomic(page);
675
676 if (is_partial_io(bvec)) {
677 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
678 bvec->bv_len);
679 kunmap_atomic(user_mem);
680 user_mem = NULL;
681 } else {
682 uncmem = user_mem;
683 }
684
685 if (page_zero_filled(uncmem)) {
686 if (user_mem)
687 kunmap_atomic(user_mem);
688 /* Free memory associated with this sector now. */
689 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
690 zram_free_page(zram, index);
691 zram_set_flag(meta, index, ZRAM_ZERO);
692 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
693
694 atomic64_inc(&zram->stats.zero_pages);
695 ret = 0;
696 goto out;
697 }
698
699 ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
700 if (!is_partial_io(bvec)) {
701 kunmap_atomic(user_mem);
702 user_mem = NULL;
703 uncmem = NULL;
704 }
705
706 if (unlikely(ret)) {
707 pr_err("Compression failed! err=%d\n", ret);
708 goto out;
709 }
710 src = zstrm->buffer;
711 if (unlikely(clen > max_zpage_size)) {
712 clen = PAGE_SIZE;
713 if (is_partial_io(bvec))
714 src = uncmem;
715 }
716
717 handle = zs_malloc(meta->mem_pool, clen);
718 if (!handle) {
719 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
720 index, clen);
721 ret = -ENOMEM;
722 goto out;
723 }
724
725 alloced_pages = zs_get_total_pages(meta->mem_pool);
726 if (zram->limit_pages && alloced_pages > zram->limit_pages) {
727 zs_free(meta->mem_pool, handle);
728 ret = -ENOMEM;
729 goto out;
730 }
731
732 update_used_max(zram, alloced_pages);
733
734 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
735
736 if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
737 src = kmap_atomic(page);
738 copy_page(cmem, src);
739 kunmap_atomic(src);
740 } else {
741 memcpy(cmem, src, clen);
742 }
743
744 zcomp_strm_release(zram->comp, zstrm);
745 zstrm = NULL;
746 zs_unmap_object(meta->mem_pool, handle);
747
748 /*
749 * Free memory associated with this sector
750 * before overwriting unused sectors.
751 */
752 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
753 zram_free_page(zram, index);
754
755 meta->table[index].handle = handle;
756 zram_set_obj_size(meta, index, clen);
757 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
758
759 /* Update stats */
760 atomic64_add(clen, &zram->stats.compr_data_size);
761 atomic64_inc(&zram->stats.pages_stored);
762out:
763 if (zstrm)
764 zcomp_strm_release(zram->comp, zstrm);
765 if (is_partial_io(bvec))
766 kfree(uncmem);
767 return ret;
768}
769
770/*
771 * zram_bio_discard - handler on discard request
772 * @index: physical block index in PAGE_SIZE units
773 * @offset: byte offset within physical block
774 */
775static void zram_bio_discard(struct zram *zram, u32 index,
776 int offset, struct bio *bio)
777{
778 size_t n = bio->bi_iter.bi_size;
779 struct zram_meta *meta = zram->meta;
780
781 /*
782 * zram manages data in physical block size units. Because logical block
783 * size isn't identical with physical block size on some arch, we
784 * could get a discard request pointing to a specific offset within a
785 * certain physical block. Although we can handle this request by
786 * reading that physiclal block and decompressing and partially zeroing
787 * and re-compressing and then re-storing it, this isn't reasonable
788 * because our intent with a discard request is to save memory. So
789 * skipping this logical block is appropriate here.
790 */
791 if (offset) {
792 if (n <= (PAGE_SIZE - offset))
793 return;
794
795 n -= (PAGE_SIZE - offset);
796 index++;
797 }
798
799 while (n >= PAGE_SIZE) {
800 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
801 zram_free_page(zram, index);
802 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
803 atomic64_inc(&zram->stats.notify_free);
804 index++;
805 n -= PAGE_SIZE;
806 }
807}
808
809static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
810 int offset, int rw)
811{
812 unsigned long start_time = jiffies;
813 int ret;
814
815 generic_start_io_acct(rw, bvec->bv_len >> SECTOR_SHIFT,
816 &zram->disk->part0);
817
818 if (rw == READ) {
819 atomic64_inc(&zram->stats.num_reads);
820 ret = zram_bvec_read(zram, bvec, index, offset);
821 } else {
822 atomic64_inc(&zram->stats.num_writes);
823 ret = zram_bvec_write(zram, bvec, index, offset);
824 }
825
826 generic_end_io_acct(rw, &zram->disk->part0, start_time);
827
828 if (unlikely(ret)) {
829 if (rw == READ)
830 atomic64_inc(&zram->stats.failed_reads);
831 else
832 atomic64_inc(&zram->stats.failed_writes);
833 }
834
835 return ret;
836}
837
838static void __zram_make_request(struct zram *zram, struct bio *bio)
839{
840 int offset, rw;
841 u32 index;
842 struct bio_vec bvec;
843 struct bvec_iter iter;
844
845 index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
846 offset = (bio->bi_iter.bi_sector &
847 (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
848
849 if (unlikely(bio->bi_rw & REQ_DISCARD)) {
850 zram_bio_discard(zram, index, offset, bio);
851 bio_endio(bio, 0);
852 return;
853 }
854
855 rw = bio_data_dir(bio);
856 bio_for_each_segment(bvec, bio, iter) {
857 int max_transfer_size = PAGE_SIZE - offset;
858
859 if (bvec.bv_len > max_transfer_size) {
860 /*
861 * zram_bvec_rw() can only make operation on a single
862 * zram page. Split the bio vector.
863 */
864 struct bio_vec bv;
865
866 bv.bv_page = bvec.bv_page;
867 bv.bv_len = max_transfer_size;
868 bv.bv_offset = bvec.bv_offset;
869
870 if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0)
871 goto out;
872
873 bv.bv_len = bvec.bv_len - max_transfer_size;
874 bv.bv_offset += max_transfer_size;
875 if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0)
876 goto out;
877 } else
878 if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0)
879 goto out;
880
881 update_position(&index, &offset, &bvec);
882 }
883
884 set_bit(BIO_UPTODATE, &bio->bi_flags);
885 bio_endio(bio, 0);
886 return;
887
888out:
889 bio_io_error(bio);
890}
891
892/*
893 * Handler function for all zram I/O requests.
894 */
895static void zram_make_request(struct request_queue *queue, struct bio *bio)
896{
897 struct zram *zram = queue->queuedata;
898
899 if (unlikely(!zram_meta_get(zram)))
900 goto error;
901
902 if (!valid_io_request(zram, bio->bi_iter.bi_sector,
903 bio->bi_iter.bi_size)) {
904 atomic64_inc(&zram->stats.invalid_io);
905 goto put_zram;
906 }
907
908 __zram_make_request(zram, bio);
909 zram_meta_put(zram);
910 return;
911put_zram:
912 zram_meta_put(zram);
913error:
914 bio_io_error(bio);
915}
916
917static void zram_slot_free_notify(struct block_device *bdev,
918 unsigned long index)
919{
920 struct zram *zram;
921 struct zram_meta *meta;
922
923 zram = bdev->bd_disk->private_data;
924 meta = zram->meta;
925
926 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
927 zram_free_page(zram, index);
928 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
929 atomic64_inc(&zram->stats.notify_free);
930}
931
932static int zram_rw_page(struct block_device *bdev, sector_t sector,
933 struct page *page, int rw)
934{
935 int offset, err = -EIO;
936 u32 index;
937 struct zram *zram;
938 struct bio_vec bv;
939
940 zram = bdev->bd_disk->private_data;
941 if (unlikely(!zram_meta_get(zram)))
942 goto out;
943
944 if (!valid_io_request(zram, sector, PAGE_SIZE)) {
945 atomic64_inc(&zram->stats.invalid_io);
946 err = -EINVAL;
947 goto put_zram;
948 }
949
950 index = sector >> SECTORS_PER_PAGE_SHIFT;
951 offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
952
953 bv.bv_page = page;
954 bv.bv_len = PAGE_SIZE;
955 bv.bv_offset = 0;
956
957 err = zram_bvec_rw(zram, &bv, index, offset, rw);
958put_zram:
959 zram_meta_put(zram);
960out:
961 /*
962 * If I/O fails, just return error(ie, non-zero) without
963 * calling page_endio.
964 * It causes resubmit the I/O with bio request by upper functions
965 * of rw_page(e.g., swap_readpage, __swap_writepage) and
966 * bio->bi_end_io does things to handle the error
967 * (e.g., SetPageError, set_page_dirty and extra works).
968 */
969 if (err == 0)
970 page_endio(page, rw, 0);
971 return err;
972}
973
974static void zram_reset_device(struct zram *zram)
975{
976 struct zram_meta *meta;
977 struct zcomp *comp;
978 u64 disksize;
979
980 down_write(&zram->init_lock);
981
982 zram->limit_pages = 0;
983
984 if (!init_done(zram)) {
985 up_write(&zram->init_lock);
986 return;
987 }
988
989 meta = zram->meta;
990 comp = zram->comp;
991 disksize = zram->disksize;
992 /*
993 * Refcount will go down to 0 eventually and r/w handler
994 * cannot handle further I/O so it will bail out by
995 * check zram_meta_get.
996 */
997 zram_meta_put(zram);
998 /*
999 * We want to free zram_meta in process context to avoid
1000 * deadlock between reclaim path and any other locks.
1001 */
1002 wait_event(zram->io_done, atomic_read(&zram->refcount) == 0);
1003
1004 /* Reset stats */
1005 memset(&zram->stats, 0, sizeof(zram->stats));
1006 zram->disksize = 0;
1007 zram->max_comp_streams = 1;
1008
1009 set_capacity(zram->disk, 0);
1010 part_stat_set_all(&zram->disk->part0, 0);
1011
1012 up_write(&zram->init_lock);
1013 /* I/O operation under all of CPU are done so let's free */
1014 zram_meta_free(meta, disksize);
1015 zcomp_destroy(comp);
1016}
1017
1018static ssize_t disksize_store(struct device *dev,
1019 struct device_attribute *attr, const char *buf, size_t len)
1020{
1021 u64 disksize;
1022 struct zcomp *comp;
1023 struct zram_meta *meta;
1024 struct zram *zram = dev_to_zram(dev);
1025 int err;
1026
1027 disksize = memparse(buf, NULL);
1028 if (!disksize)
1029 return -EINVAL;
1030
1031 disksize = PAGE_ALIGN(disksize);
1032 meta = zram_meta_alloc(zram->disk->disk_name, disksize);
1033 if (!meta)
1034 return -ENOMEM;
1035
1036 comp = zcomp_create(zram->compressor, zram->max_comp_streams);
1037 if (IS_ERR(comp)) {
1038 pr_info("Cannot initialise %s compressing backend\n",
1039 zram->compressor);
1040 err = PTR_ERR(comp);
1041 goto out_free_meta;
1042 }
1043
1044 down_write(&zram->init_lock);
1045 if (init_done(zram)) {
1046 pr_info("Cannot change disksize for initialized device\n");
1047 err = -EBUSY;
1048 goto out_destroy_comp;
1049 }
1050
1051 init_waitqueue_head(&zram->io_done);
1052 atomic_set(&zram->refcount, 1);
1053 zram->meta = meta;
1054 zram->comp = comp;
1055 zram->disksize = disksize;
1056 set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1057 up_write(&zram->init_lock);
1058
1059 /*
1060 * Revalidate disk out of the init_lock to avoid lockdep splat.
1061 * It's okay because disk's capacity is protected by init_lock
1062 * so that revalidate_disk always sees up-to-date capacity.
1063 */
1064 revalidate_disk(zram->disk);
1065
1066 return len;
1067
1068out_destroy_comp:
1069 up_write(&zram->init_lock);
1070 zcomp_destroy(comp);
1071out_free_meta:
1072 zram_meta_free(meta, disksize);
1073 return err;
1074}
1075
1076static ssize_t reset_store(struct device *dev,
1077 struct device_attribute *attr, const char *buf, size_t len)
1078{
1079 int ret;
1080 unsigned short do_reset;
1081 struct zram *zram;
1082 struct block_device *bdev;
1083
1084 ret = kstrtou16(buf, 10, &do_reset);
1085 if (ret)
1086 return ret;
1087
1088 if (!do_reset)
1089 return -EINVAL;
1090
1091 zram = dev_to_zram(dev);
1092 bdev = bdget_disk(zram->disk, 0);
1093 if (!bdev)
1094 return -ENOMEM;
1095
1096 mutex_lock(&bdev->bd_mutex);
1097 /* Do not reset an active device or claimed device */
1098 if (bdev->bd_openers || zram->claim) {
1099 mutex_unlock(&bdev->bd_mutex);
1100 bdput(bdev);
1101 return -EBUSY;
1102 }
1103
1104 /* From now on, anyone can't open /dev/zram[0-9] */
1105 zram->claim = true;
1106 mutex_unlock(&bdev->bd_mutex);
1107
1108 /* Make sure all the pending I/O are finished */
1109 fsync_bdev(bdev);
1110 zram_reset_device(zram);
1111 revalidate_disk(zram->disk);
1112 bdput(bdev);
1113
1114 mutex_lock(&bdev->bd_mutex);
1115 zram->claim = false;
1116 mutex_unlock(&bdev->bd_mutex);
1117
1118 return len;
1119}
1120
1121static int zram_open(struct block_device *bdev, fmode_t mode)
1122{
1123 int ret = 0;
1124 struct zram *zram;
1125
1126 WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1127
1128 zram = bdev->bd_disk->private_data;
1129 /* zram was claimed to reset so open request fails */
1130 if (zram->claim)
1131 ret = -EBUSY;
1132
1133 return ret;
1134}
1135
1136static const struct block_device_operations zram_devops = {
1137 .open = zram_open,
1138 .swap_slot_free_notify = zram_slot_free_notify,
1139 .rw_page = zram_rw_page,
1140 .owner = THIS_MODULE
1141};
1142
1143static DEVICE_ATTR_WO(compact);
1144static DEVICE_ATTR_RW(disksize);
1145static DEVICE_ATTR_RO(initstate);
1146static DEVICE_ATTR_WO(reset);
1147static DEVICE_ATTR_RO(orig_data_size);
1148static DEVICE_ATTR_RO(mem_used_total);
1149static DEVICE_ATTR_RW(mem_limit);
1150static DEVICE_ATTR_RW(mem_used_max);
1151static DEVICE_ATTR_RW(max_comp_streams);
1152static DEVICE_ATTR_RW(comp_algorithm);
1153
1154static struct attribute *zram_disk_attrs[] = {
1155 &dev_attr_disksize.attr,
1156 &dev_attr_initstate.attr,
1157 &dev_attr_reset.attr,
1158 &dev_attr_num_reads.attr,
1159 &dev_attr_num_writes.attr,
1160 &dev_attr_failed_reads.attr,
1161 &dev_attr_failed_writes.attr,
1162 &dev_attr_compact.attr,
1163 &dev_attr_invalid_io.attr,
1164 &dev_attr_notify_free.attr,
1165 &dev_attr_zero_pages.attr,
1166 &dev_attr_orig_data_size.attr,
1167 &dev_attr_compr_data_size.attr,
1168 &dev_attr_mem_used_total.attr,
1169 &dev_attr_mem_limit.attr,
1170 &dev_attr_mem_used_max.attr,
1171 &dev_attr_max_comp_streams.attr,
1172 &dev_attr_comp_algorithm.attr,
1173 &dev_attr_io_stat.attr,
1174 &dev_attr_mm_stat.attr,
1175 NULL,
1176};
1177
1178static struct attribute_group zram_disk_attr_group = {
1179 .attrs = zram_disk_attrs,
1180};
1181
1182/*
1183 * Allocate and initialize new zram device. the function returns
1184 * '>= 0' device_id upon success, and negative value otherwise.
1185 */
1186static int zram_add(void)
1187{
1188 struct zram *zram;
1189 struct request_queue *queue;
1190 int ret, device_id;
1191
1192 zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1193 if (!zram)
1194 return -ENOMEM;
1195
1196 ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1197 if (ret < 0)
1198 goto out_free_dev;
1199 device_id = ret;
1200
1201 init_rwsem(&zram->init_lock);
1202
1203 queue = blk_alloc_queue(GFP_KERNEL);
1204 if (!queue) {
1205 pr_err("Error allocating disk queue for device %d\n",
1206 device_id);
1207 ret = -ENOMEM;
1208 goto out_free_idr;
1209 }
1210
1211 blk_queue_make_request(queue, zram_make_request);
1212
1213 /* gendisk structure */
1214 zram->disk = alloc_disk(1);
1215 if (!zram->disk) {
1216 pr_warn("Error allocating disk structure for device %d\n",
1217 device_id);
1218 ret = -ENOMEM;
1219 goto out_free_queue;
1220 }
1221
1222 zram->disk->major = zram_major;
1223 zram->disk->first_minor = device_id;
1224 zram->disk->fops = &zram_devops;
1225 zram->disk->queue = queue;
1226 zram->disk->queue->queuedata = zram;
1227 zram->disk->private_data = zram;
1228 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1229
1230 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1231 set_capacity(zram->disk, 0);
1232 /* zram devices sort of resembles non-rotational disks */
1233 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1234 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1235 /*
1236 * To ensure that we always get PAGE_SIZE aligned
1237 * and n*PAGE_SIZED sized I/O requests.
1238 */
1239 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1240 blk_queue_logical_block_size(zram->disk->queue,
1241 ZRAM_LOGICAL_BLOCK_SIZE);
1242 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1243 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1244 zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1245 zram->disk->queue->limits.max_discard_sectors = UINT_MAX;
1246 /*
1247 * zram_bio_discard() will clear all logical blocks if logical block
1248 * size is identical with physical block size(PAGE_SIZE). But if it is
1249 * different, we will skip discarding some parts of logical blocks in
1250 * the part of the request range which isn't aligned to physical block
1251 * size. So we can't ensure that all discarded logical blocks are
1252 * zeroed.
1253 */
1254 if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1255 zram->disk->queue->limits.discard_zeroes_data = 1;
1256 else
1257 zram->disk->queue->limits.discard_zeroes_data = 0;
1258 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1259
1260 add_disk(zram->disk);
1261
1262 ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1263 &zram_disk_attr_group);
1264 if (ret < 0) {
1265 pr_warn("Error creating sysfs group");
1266 goto out_free_disk;
1267 }
1268 strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1269 zram->meta = NULL;
1270 zram->max_comp_streams = 1;
1271
1272 pr_info("Added device: %s\n", zram->disk->disk_name);
1273 return device_id;
1274
1275out_free_disk:
1276 del_gendisk(zram->disk);
1277 put_disk(zram->disk);
1278out_free_queue:
1279 blk_cleanup_queue(queue);
1280out_free_idr:
1281 idr_remove(&zram_index_idr, device_id);
1282out_free_dev:
1283 kfree(zram);
1284 return ret;
1285}
1286
1287static int zram_remove(struct zram *zram)
1288{
1289 struct block_device *bdev;
1290
1291 bdev = bdget_disk(zram->disk, 0);
1292 if (!bdev)
1293 return -ENOMEM;
1294
1295 mutex_lock(&bdev->bd_mutex);
1296 if (bdev->bd_openers || zram->claim) {
1297 mutex_unlock(&bdev->bd_mutex);
1298 bdput(bdev);
1299 return -EBUSY;
1300 }
1301
1302 zram->claim = true;
1303 mutex_unlock(&bdev->bd_mutex);
1304
1305 /*
1306 * Remove sysfs first, so no one will perform a disksize
1307 * store while we destroy the devices. This also helps during
1308 * hot_remove -- zram_reset_device() is the last holder of
1309 * ->init_lock, no later/concurrent disksize_store() or any
1310 * other sysfs handlers are possible.
1311 */
1312 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1313 &zram_disk_attr_group);
1314
1315 /* Make sure all the pending I/O are finished */
1316 fsync_bdev(bdev);
1317 zram_reset_device(zram);
1318 bdput(bdev);
1319
1320 pr_info("Removed device: %s\n", zram->disk->disk_name);
1321
1322 idr_remove(&zram_index_idr, zram->disk->first_minor);
1323 blk_cleanup_queue(zram->disk->queue);
1324 del_gendisk(zram->disk);
1325 put_disk(zram->disk);
1326 kfree(zram);
1327 return 0;
1328}
1329
1330/* zram-control sysfs attributes */
1331static ssize_t hot_add_show(struct class *class,
1332 struct class_attribute *attr,
1333 char *buf)
1334{
1335 int ret;
1336
1337 mutex_lock(&zram_index_mutex);
1338 ret = zram_add();
1339 mutex_unlock(&zram_index_mutex);
1340
1341 if (ret < 0)
1342 return ret;
1343 return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1344}
1345
1346static ssize_t hot_remove_store(struct class *class,
1347 struct class_attribute *attr,
1348 const char *buf,
1349 size_t count)
1350{
1351 struct zram *zram;
1352 int ret, dev_id;
1353
1354 /* dev_id is gendisk->first_minor, which is `int' */
1355 ret = kstrtoint(buf, 10, &dev_id);
1356 if (ret)
1357 return ret;
1358 if (dev_id < 0)
1359 return -EINVAL;
1360
1361 mutex_lock(&zram_index_mutex);
1362
1363 zram = idr_find(&zram_index_idr, dev_id);
1364 if (zram)
1365 ret = zram_remove(zram);
1366 else
1367 ret = -ENODEV;
1368
1369 mutex_unlock(&zram_index_mutex);
1370 return ret ? ret : count;
1371}
1372
1373static struct class_attribute zram_control_class_attrs[] = {
1374 __ATTR_RO(hot_add),
1375 __ATTR_WO(hot_remove),
1376 __ATTR_NULL,
1377};
1378
1379static struct class zram_control_class = {
1380 .name = "zram-control",
1381 .owner = THIS_MODULE,
1382 .class_attrs = zram_control_class_attrs,
1383};
1384
1385static int zram_remove_cb(int id, void *ptr, void *data)
1386{
1387 zram_remove(ptr);
1388 return 0;
1389}
1390
1391static void destroy_devices(void)
1392{
1393 class_unregister(&zram_control_class);
1394 idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1395 idr_destroy(&zram_index_idr);
1396 unregister_blkdev(zram_major, "zram");
1397}
1398
1399static int __init zram_init(void)
1400{
1401 int ret;
1402
1403 ret = class_register(&zram_control_class);
1404 if (ret) {
1405 pr_warn("Unable to register zram-control class\n");
1406 return ret;
1407 }
1408
1409 zram_major = register_blkdev(0, "zram");
1410 if (zram_major <= 0) {
1411 pr_warn("Unable to get major number\n");
1412 class_unregister(&zram_control_class);
1413 return -EBUSY;
1414 }
1415
1416 while (num_devices != 0) {
1417 mutex_lock(&zram_index_mutex);
1418 ret = zram_add();
1419 mutex_unlock(&zram_index_mutex);
1420 if (ret < 0)
1421 goto out_error;
1422 num_devices--;
1423 }
1424
1425 return 0;
1426
1427out_error:
1428 destroy_devices();
1429 return ret;
1430}
1431
1432static void __exit zram_exit(void)
1433{
1434 destroy_devices();
1435}
1436
1437module_init(zram_init);
1438module_exit(zram_exit);
1439
1440module_param(num_devices, uint, 0);
1441MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1442
1443MODULE_LICENSE("Dual BSD/GPL");
1444MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1445MODULE_DESCRIPTION("Compressed RAM Block Device");
Linux 6.x block layer and io_uring tree(s)