]> Pileus Git - ~andy/linux/blob - drivers/staging/zram/zram_drv.c
Merge branch 'v4l_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab...
[~andy/linux] / drivers / staging / zram / zram_drv.c
1 /*
2  * Compressed RAM block device
3  *
4  * Copyright (C) 2008, 2009, 2010  Nitin Gupta
5  *
6  * This code is released using a dual license strategy: BSD/GPL
7  * You can choose the licence that better fits your requirements.
8  *
9  * Released under the terms of 3-clause BSD License
10  * Released under the terms of GNU General Public License Version 2.0
11  *
12  * Project home: http://compcache.googlecode.com
13  */
14
15 #define KMSG_COMPONENT "zram"
16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
17
18 #ifdef CONFIG_ZRAM_DEBUG
19 #define DEBUG
20 #endif
21
22 #include <linux/module.h>
23 #include <linux/kernel.h>
24 #include <linux/bio.h>
25 #include <linux/bitops.h>
26 #include <linux/blkdev.h>
27 #include <linux/buffer_head.h>
28 #include <linux/device.h>
29 #include <linux/genhd.h>
30 #include <linux/highmem.h>
31 #include <linux/slab.h>
32 #include <linux/lzo.h>
33 #include <linux/string.h>
34 #include <linux/vmalloc.h>
35
36 #include "zram_drv.h"
37
38 /* Globals */
39 static int zram_major;
40 struct zram *zram_devices;
41
42 /* Module params (documentation at end) */
43 static unsigned int num_devices = 1;
44
45 static void zram_stat64_add(struct zram *zram, u64 *v, u64 inc)
46 {
47         spin_lock(&zram->stat64_lock);
48         *v = *v + inc;
49         spin_unlock(&zram->stat64_lock);
50 }
51
52 static void zram_stat64_sub(struct zram *zram, u64 *v, u64 dec)
53 {
54         spin_lock(&zram->stat64_lock);
55         *v = *v - dec;
56         spin_unlock(&zram->stat64_lock);
57 }
58
59 static void zram_stat64_inc(struct zram *zram, u64 *v)
60 {
61         zram_stat64_add(zram, v, 1);
62 }
63
64 static int zram_test_flag(struct zram_meta *meta, u32 index,
65                         enum zram_pageflags flag)
66 {
67         return meta->table[index].flags & BIT(flag);
68 }
69
70 static void zram_set_flag(struct zram_meta *meta, u32 index,
71                         enum zram_pageflags flag)
72 {
73         meta->table[index].flags |= BIT(flag);
74 }
75
76 static void zram_clear_flag(struct zram_meta *meta, u32 index,
77                         enum zram_pageflags flag)
78 {
79         meta->table[index].flags &= ~BIT(flag);
80 }
81
82 static int page_zero_filled(void *ptr)
83 {
84         unsigned int pos;
85         unsigned long *page;
86
87         page = (unsigned long *)ptr;
88
89         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
90                 if (page[pos])
91                         return 0;
92         }
93
94         return 1;
95 }
96
97 static void zram_free_page(struct zram *zram, size_t index)
98 {
99         struct zram_meta *meta = zram->meta;
100         unsigned long handle = meta->table[index].handle;
101         u16 size = meta->table[index].size;
102
103         if (unlikely(!handle)) {
104                 /*
105                  * No memory is allocated for zero filled pages.
106                  * Simply clear zero page flag.
107                  */
108                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
109                         zram_clear_flag(meta, index, ZRAM_ZERO);
110                         zram->stats.pages_zero--;
111                 }
112                 return;
113         }
114
115         if (unlikely(size > max_zpage_size))
116                 zram->stats.bad_compress--;
117
118         zs_free(meta->mem_pool, handle);
119
120         if (size <= PAGE_SIZE / 2)
121                 zram->stats.good_compress--;
122
123         zram_stat64_sub(zram, &zram->stats.compr_size,
124                         meta->table[index].size);
125         zram->stats.pages_stored--;
126
127         meta->table[index].handle = 0;
128         meta->table[index].size = 0;
129 }
130
131 static void handle_zero_page(struct bio_vec *bvec)
132 {
133         struct page *page = bvec->bv_page;
134         void *user_mem;
135
136         user_mem = kmap_atomic(page);
137         memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
138         kunmap_atomic(user_mem);
139
140         flush_dcache_page(page);
141 }
142
143 static inline int is_partial_io(struct bio_vec *bvec)
144 {
145         return bvec->bv_len != PAGE_SIZE;
146 }
147
148 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
149 {
150         int ret = LZO_E_OK;
151         size_t clen = PAGE_SIZE;
152         unsigned char *cmem;
153         struct zram_meta *meta = zram->meta;
154         unsigned long handle = meta->table[index].handle;
155
156         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
157                 memset(mem, 0, PAGE_SIZE);
158                 return 0;
159         }
160
161         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
162         if (meta->table[index].size == PAGE_SIZE)
163                 memcpy(mem, cmem, PAGE_SIZE);
164         else
165                 ret = lzo1x_decompress_safe(cmem, meta->table[index].size,
166                                                 mem, &clen);
167         zs_unmap_object(meta->mem_pool, handle);
168
169         /* Should NEVER happen. Return bio error if it does. */
170         if (unlikely(ret != LZO_E_OK)) {
171                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
172                 zram_stat64_inc(zram, &zram->stats.failed_reads);
173                 return ret;
174         }
175
176         return 0;
177 }
178
179 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
180                           u32 index, int offset, struct bio *bio)
181 {
182         int ret;
183         struct page *page;
184         unsigned char *user_mem, *uncmem = NULL;
185         struct zram_meta *meta = zram->meta;
186         page = bvec->bv_page;
187
188         if (unlikely(!meta->table[index].handle) ||
189                         zram_test_flag(meta, index, ZRAM_ZERO)) {
190                 handle_zero_page(bvec);
191                 return 0;
192         }
193
194         if (is_partial_io(bvec))
195                 /* Use  a temporary buffer to decompress the page */
196                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
197
198         user_mem = kmap_atomic(page);
199         if (!is_partial_io(bvec))
200                 uncmem = user_mem;
201
202         if (!uncmem) {
203                 pr_info("Unable to allocate temp memory\n");
204                 ret = -ENOMEM;
205                 goto out_cleanup;
206         }
207
208         ret = zram_decompress_page(zram, uncmem, index);
209         /* Should NEVER happen. Return bio error if it does. */
210         if (unlikely(ret != LZO_E_OK)) {
211                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
212                 zram_stat64_inc(zram, &zram->stats.failed_reads);
213                 goto out_cleanup;
214         }
215
216         if (is_partial_io(bvec))
217                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
218                                 bvec->bv_len);
219
220         flush_dcache_page(page);
221         ret = 0;
222 out_cleanup:
223         kunmap_atomic(user_mem);
224         if (is_partial_io(bvec))
225                 kfree(uncmem);
226         return ret;
227 }
228
229 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
230                            int offset)
231 {
232         int ret = 0;
233         size_t clen;
234         unsigned long handle;
235         struct page *page;
236         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
237         struct zram_meta *meta = zram->meta;
238
239         page = bvec->bv_page;
240         src = meta->compress_buffer;
241
242         if (is_partial_io(bvec)) {
243                 /*
244                  * This is a partial IO. We need to read the full page
245                  * before to write the changes.
246                  */
247                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
248                 if (!uncmem) {
249                         ret = -ENOMEM;
250                         goto out;
251                 }
252                 ret = zram_decompress_page(zram, uncmem, index);
253                 if (ret)
254                         goto out;
255         }
256
257         /*
258          * System overwrites unused sectors. Free memory associated
259          * with this sector now.
260          */
261         if (meta->table[index].handle ||
262             zram_test_flag(meta, index, ZRAM_ZERO))
263                 zram_free_page(zram, index);
264
265         user_mem = kmap_atomic(page);
266
267         if (is_partial_io(bvec)) {
268                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
269                        bvec->bv_len);
270                 kunmap_atomic(user_mem);
271                 user_mem = NULL;
272         } else {
273                 uncmem = user_mem;
274         }
275
276         if (page_zero_filled(uncmem)) {
277                 kunmap_atomic(user_mem);
278                 if (is_partial_io(bvec))
279                         kfree(uncmem);
280                 zram->stats.pages_zero++;
281                 zram_set_flag(meta, index, ZRAM_ZERO);
282                 ret = 0;
283                 goto out;
284         }
285
286         ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
287                                meta->compress_workmem);
288
289         if (!is_partial_io(bvec)) {
290                 kunmap_atomic(user_mem);
291                 user_mem = NULL;
292                 uncmem = NULL;
293         }
294
295         if (unlikely(ret != LZO_E_OK)) {
296                 pr_err("Compression failed! err=%d\n", ret);
297                 goto out;
298         }
299
300         if (unlikely(clen > max_zpage_size)) {
301                 zram->stats.bad_compress++;
302                 clen = PAGE_SIZE;
303                 src = NULL;
304                 if (is_partial_io(bvec))
305                         src = uncmem;
306         }
307
308         handle = zs_malloc(meta->mem_pool, clen);
309         if (!handle) {
310                 pr_info("Error allocating memory for compressed "
311                         "page: %u, size=%zu\n", index, clen);
312                 ret = -ENOMEM;
313                 goto out;
314         }
315         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
316
317         if ((clen == PAGE_SIZE) && !is_partial_io(bvec))
318                 src = kmap_atomic(page);
319         memcpy(cmem, src, clen);
320         if ((clen == PAGE_SIZE) && !is_partial_io(bvec))
321                 kunmap_atomic(src);
322
323         zs_unmap_object(meta->mem_pool, handle);
324
325         meta->table[index].handle = handle;
326         meta->table[index].size = clen;
327
328         /* Update stats */
329         zram_stat64_add(zram, &zram->stats.compr_size, clen);
330         zram->stats.pages_stored++;
331         if (clen <= PAGE_SIZE / 2)
332                 zram->stats.good_compress++;
333
334 out:
335         if (is_partial_io(bvec))
336                 kfree(uncmem);
337
338         if (ret)
339                 zram_stat64_inc(zram, &zram->stats.failed_writes);
340         return ret;
341 }
342
343 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
344                         int offset, struct bio *bio, int rw)
345 {
346         int ret;
347
348         if (rw == READ) {
349                 down_read(&zram->lock);
350                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
351                 up_read(&zram->lock);
352         } else {
353                 down_write(&zram->lock);
354                 ret = zram_bvec_write(zram, bvec, index, offset);
355                 up_write(&zram->lock);
356         }
357
358         return ret;
359 }
360
361 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
362 {
363         if (*offset + bvec->bv_len >= PAGE_SIZE)
364                 (*index)++;
365         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
366 }
367
368 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
369 {
370         int i, offset;
371         u32 index;
372         struct bio_vec *bvec;
373
374         switch (rw) {
375         case READ:
376                 zram_stat64_inc(zram, &zram->stats.num_reads);
377                 break;
378         case WRITE:
379                 zram_stat64_inc(zram, &zram->stats.num_writes);
380                 break;
381         }
382
383         index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
384         offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
385
386         bio_for_each_segment(bvec, bio, i) {
387                 int max_transfer_size = PAGE_SIZE - offset;
388
389                 if (bvec->bv_len > max_transfer_size) {
390                         /*
391                          * zram_bvec_rw() can only make operation on a single
392                          * zram page. Split the bio vector.
393                          */
394                         struct bio_vec bv;
395
396                         bv.bv_page = bvec->bv_page;
397                         bv.bv_len = max_transfer_size;
398                         bv.bv_offset = bvec->bv_offset;
399
400                         if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
401                                 goto out;
402
403                         bv.bv_len = bvec->bv_len - max_transfer_size;
404                         bv.bv_offset += max_transfer_size;
405                         if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
406                                 goto out;
407                 } else
408                         if (zram_bvec_rw(zram, bvec, index, offset, bio, rw)
409                             < 0)
410                                 goto out;
411
412                 update_position(&index, &offset, bvec);
413         }
414
415         set_bit(BIO_UPTODATE, &bio->bi_flags);
416         bio_endio(bio, 0);
417         return;
418
419 out:
420         bio_io_error(bio);
421 }
422
423 /*
424  * Check if request is within bounds and aligned on zram logical blocks.
425  */
426 static inline int valid_io_request(struct zram *zram, struct bio *bio)
427 {
428         if (unlikely(
429                 (bio->bi_sector >= (zram->disksize >> SECTOR_SHIFT)) ||
430                 (bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)) ||
431                 (bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))) {
432
433                 return 0;
434         }
435
436         /* I/O request is valid */
437         return 1;
438 }
439
440 /*
441  * Handler function for all zram I/O requests.
442  */
443 static void zram_make_request(struct request_queue *queue, struct bio *bio)
444 {
445         struct zram *zram = queue->queuedata;
446
447         down_read(&zram->init_lock);
448         if (unlikely(!zram->init_done))
449                 goto error;
450
451         if (!valid_io_request(zram, bio)) {
452                 zram_stat64_inc(zram, &zram->stats.invalid_io);
453                 goto error;
454         }
455
456         __zram_make_request(zram, bio, bio_data_dir(bio));
457         up_read(&zram->init_lock);
458
459         return;
460
461 error:
462         up_read(&zram->init_lock);
463         bio_io_error(bio);
464 }
465
466 static void __zram_reset_device(struct zram *zram)
467 {
468         size_t index;
469         struct zram_meta *meta;
470
471         if (!zram->init_done)
472                 return;
473
474         meta = zram->meta;
475         zram->init_done = 0;
476
477         /* Free all pages that are still in this zram device */
478         for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
479                 unsigned long handle = meta->table[index].handle;
480                 if (!handle)
481                         continue;
482
483                 zs_free(meta->mem_pool, handle);
484         }
485
486         zram_meta_free(zram->meta);
487         zram->meta = NULL;
488         /* Reset stats */
489         memset(&zram->stats, 0, sizeof(zram->stats));
490
491         zram->disksize = 0;
492         set_capacity(zram->disk, 0);
493 }
494
495 void zram_reset_device(struct zram *zram)
496 {
497         down_write(&zram->init_lock);
498         __zram_reset_device(zram);
499         up_write(&zram->init_lock);
500 }
501
502 void zram_meta_free(struct zram_meta *meta)
503 {
504         zs_destroy_pool(meta->mem_pool);
505         kfree(meta->compress_workmem);
506         free_pages((unsigned long)meta->compress_buffer, 1);
507         vfree(meta->table);
508         kfree(meta);
509 }
510
511 struct zram_meta *zram_meta_alloc(u64 disksize)
512 {
513         size_t num_pages;
514         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
515         if (!meta)
516                 goto out;
517
518         meta->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
519         if (!meta->compress_workmem)
520                 goto free_meta;
521
522         meta->compress_buffer =
523                 (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
524         if (!meta->compress_buffer) {
525                 pr_err("Error allocating compressor buffer space\n");
526                 goto free_workmem;
527         }
528
529         num_pages = disksize >> PAGE_SHIFT;
530         meta->table = vzalloc(num_pages * sizeof(*meta->table));
531         if (!meta->table) {
532                 pr_err("Error allocating zram address table\n");
533                 goto free_buffer;
534         }
535
536         meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
537         if (!meta->mem_pool) {
538                 pr_err("Error creating memory pool\n");
539                 goto free_table;
540         }
541
542         return meta;
543
544 free_table:
545         vfree(meta->table);
546 free_buffer:
547         free_pages((unsigned long)meta->compress_buffer, 1);
548 free_workmem:
549         kfree(meta->compress_workmem);
550 free_meta:
551         kfree(meta);
552         meta = NULL;
553 out:
554         return meta;
555 }
556
557 void zram_init_device(struct zram *zram, struct zram_meta *meta)
558 {
559         if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) {
560                 pr_info(
561                 "There is little point creating a zram of greater than "
562                 "twice the size of memory since we expect a 2:1 compression "
563                 "ratio. Note that zram uses about 0.1%% of the size of "
564                 "the disk when not in use so a huge zram is "
565                 "wasteful.\n"
566                 "\tMemory Size: %lu kB\n"
567                 "\tSize you selected: %llu kB\n"
568                 "Continuing anyway ...\n",
569                 (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10
570                 );
571         }
572
573         /* zram devices sort of resembles non-rotational disks */
574         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
575
576         zram->meta = meta;
577         zram->init_done = 1;
578
579         pr_debug("Initialization done!\n");
580 }
581
582 static void zram_slot_free_notify(struct block_device *bdev,
583                                 unsigned long index)
584 {
585         struct zram *zram;
586
587         zram = bdev->bd_disk->private_data;
588         zram_free_page(zram, index);
589         zram_stat64_inc(zram, &zram->stats.notify_free);
590 }
591
592 static const struct block_device_operations zram_devops = {
593         .swap_slot_free_notify = zram_slot_free_notify,
594         .owner = THIS_MODULE
595 };
596
597 static int create_device(struct zram *zram, int device_id)
598 {
599         int ret = 0;
600
601         init_rwsem(&zram->lock);
602         init_rwsem(&zram->init_lock);
603         spin_lock_init(&zram->stat64_lock);
604
605         zram->queue = blk_alloc_queue(GFP_KERNEL);
606         if (!zram->queue) {
607                 pr_err("Error allocating disk queue for device %d\n",
608                         device_id);
609                 ret = -ENOMEM;
610                 goto out;
611         }
612
613         blk_queue_make_request(zram->queue, zram_make_request);
614         zram->queue->queuedata = zram;
615
616          /* gendisk structure */
617         zram->disk = alloc_disk(1);
618         if (!zram->disk) {
619                 blk_cleanup_queue(zram->queue);
620                 pr_warn("Error allocating disk structure for device %d\n",
621                         device_id);
622                 ret = -ENOMEM;
623                 goto out;
624         }
625
626         zram->disk->major = zram_major;
627         zram->disk->first_minor = device_id;
628         zram->disk->fops = &zram_devops;
629         zram->disk->queue = zram->queue;
630         zram->disk->private_data = zram;
631         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
632
633         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
634         set_capacity(zram->disk, 0);
635
636         /*
637          * To ensure that we always get PAGE_SIZE aligned
638          * and n*PAGE_SIZED sized I/O requests.
639          */
640         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
641         blk_queue_logical_block_size(zram->disk->queue,
642                                         ZRAM_LOGICAL_BLOCK_SIZE);
643         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
644         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
645
646         add_disk(zram->disk);
647
648         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
649                                 &zram_disk_attr_group);
650         if (ret < 0) {
651                 pr_warn("Error creating sysfs group");
652                 goto out;
653         }
654
655         zram->init_done = 0;
656
657 out:
658         return ret;
659 }
660
661 static void destroy_device(struct zram *zram)
662 {
663         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
664                         &zram_disk_attr_group);
665
666         if (zram->disk) {
667                 del_gendisk(zram->disk);
668                 put_disk(zram->disk);
669         }
670
671         if (zram->queue)
672                 blk_cleanup_queue(zram->queue);
673 }
674
675 unsigned int zram_get_num_devices(void)
676 {
677         return num_devices;
678 }
679
680 static int __init zram_init(void)
681 {
682         int ret, dev_id;
683
684         if (num_devices > max_num_devices) {
685                 pr_warn("Invalid value for num_devices: %u\n",
686                                 num_devices);
687                 ret = -EINVAL;
688                 goto out;
689         }
690
691         zram_major = register_blkdev(0, "zram");
692         if (zram_major <= 0) {
693                 pr_warn("Unable to get major number\n");
694                 ret = -EBUSY;
695                 goto out;
696         }
697
698         /* Allocate the device array and initialize each one */
699         zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
700         if (!zram_devices) {
701                 ret = -ENOMEM;
702                 goto unregister;
703         }
704
705         for (dev_id = 0; dev_id < num_devices; dev_id++) {
706                 ret = create_device(&zram_devices[dev_id], dev_id);
707                 if (ret)
708                         goto free_devices;
709         }
710
711         pr_info("Created %u device(s) ...\n", num_devices);
712
713         return 0;
714
715 free_devices:
716         while (dev_id)
717                 destroy_device(&zram_devices[--dev_id]);
718         kfree(zram_devices);
719 unregister:
720         unregister_blkdev(zram_major, "zram");
721 out:
722         return ret;
723 }
724
725 static void __exit zram_exit(void)
726 {
727         int i;
728         struct zram *zram;
729
730         for (i = 0; i < num_devices; i++) {
731                 zram = &zram_devices[i];
732
733                 destroy_device(zram);
734                 zram_reset_device(zram);
735         }
736
737         unregister_blkdev(zram_major, "zram");
738
739         kfree(zram_devices);
740         pr_debug("Cleanup done!\n");
741 }
742
743 module_param(num_devices, uint, 0);
744 MODULE_PARM_DESC(num_devices, "Number of zram devices");
745
746 module_init(zram_init);
747 module_exit(zram_exit);
748
749 MODULE_LICENSE("Dual BSD/GPL");
750 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
751 MODULE_DESCRIPTION("Compressed RAM Block Device");