zram: use zram->lock to protect zram_free_page() in swap free notify path
[deliverable/linux.git] / 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 goto out_cleanup;
212
213 if (is_partial_io(bvec))
214 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
215 bvec->bv_len);
216
217 flush_dcache_page(page);
218 ret = 0;
219 out_cleanup:
220 kunmap_atomic(user_mem);
221 if (is_partial_io(bvec))
222 kfree(uncmem);
223 return ret;
224 }
225
226 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
227 int offset)
228 {
229 int ret = 0;
230 size_t clen;
231 unsigned long handle;
232 struct page *page;
233 unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
234 struct zram_meta *meta = zram->meta;
235
236 page = bvec->bv_page;
237 src = meta->compress_buffer;
238
239 if (is_partial_io(bvec)) {
240 /*
241 * This is a partial IO. We need to read the full page
242 * before to write the changes.
243 */
244 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
245 if (!uncmem) {
246 ret = -ENOMEM;
247 goto out;
248 }
249 ret = zram_decompress_page(zram, uncmem, index);
250 if (ret)
251 goto out;
252 }
253
254 /*
255 * System overwrites unused sectors. Free memory associated
256 * with this sector now.
257 */
258 if (meta->table[index].handle ||
259 zram_test_flag(meta, index, ZRAM_ZERO))
260 zram_free_page(zram, index);
261
262 user_mem = kmap_atomic(page);
263
264 if (is_partial_io(bvec)) {
265 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
266 bvec->bv_len);
267 kunmap_atomic(user_mem);
268 user_mem = NULL;
269 } else {
270 uncmem = user_mem;
271 }
272
273 if (page_zero_filled(uncmem)) {
274 kunmap_atomic(user_mem);
275 if (is_partial_io(bvec))
276 kfree(uncmem);
277 zram->stats.pages_zero++;
278 zram_set_flag(meta, index, ZRAM_ZERO);
279 ret = 0;
280 goto out;
281 }
282
283 ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
284 meta->compress_workmem);
285
286 if (!is_partial_io(bvec)) {
287 kunmap_atomic(user_mem);
288 user_mem = NULL;
289 uncmem = NULL;
290 }
291
292 if (unlikely(ret != LZO_E_OK)) {
293 pr_err("Compression failed! err=%d\n", ret);
294 goto out;
295 }
296
297 if (unlikely(clen > max_zpage_size)) {
298 zram->stats.bad_compress++;
299 clen = PAGE_SIZE;
300 src = NULL;
301 if (is_partial_io(bvec))
302 src = uncmem;
303 }
304
305 handle = zs_malloc(meta->mem_pool, clen);
306 if (!handle) {
307 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
308 index, clen);
309 ret = -ENOMEM;
310 goto out;
311 }
312 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
313
314 if ((clen == PAGE_SIZE) && !is_partial_io(bvec))
315 src = kmap_atomic(page);
316 memcpy(cmem, src, clen);
317 if ((clen == PAGE_SIZE) && !is_partial_io(bvec))
318 kunmap_atomic(src);
319
320 zs_unmap_object(meta->mem_pool, handle);
321
322 meta->table[index].handle = handle;
323 meta->table[index].size = clen;
324
325 /* Update stats */
326 zram_stat64_add(zram, &zram->stats.compr_size, clen);
327 zram->stats.pages_stored++;
328 if (clen <= PAGE_SIZE / 2)
329 zram->stats.good_compress++;
330
331 out:
332 if (is_partial_io(bvec))
333 kfree(uncmem);
334
335 if (ret)
336 zram_stat64_inc(zram, &zram->stats.failed_writes);
337 return ret;
338 }
339
340 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
341 int offset, struct bio *bio, int rw)
342 {
343 int ret;
344
345 if (rw == READ) {
346 down_read(&zram->lock);
347 ret = zram_bvec_read(zram, bvec, index, offset, bio);
348 up_read(&zram->lock);
349 } else {
350 down_write(&zram->lock);
351 ret = zram_bvec_write(zram, bvec, index, offset);
352 up_write(&zram->lock);
353 }
354
355 return ret;
356 }
357
358 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
359 {
360 if (*offset + bvec->bv_len >= PAGE_SIZE)
361 (*index)++;
362 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
363 }
364
365 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
366 {
367 int i, offset;
368 u32 index;
369 struct bio_vec *bvec;
370
371 switch (rw) {
372 case READ:
373 zram_stat64_inc(zram, &zram->stats.num_reads);
374 break;
375 case WRITE:
376 zram_stat64_inc(zram, &zram->stats.num_writes);
377 break;
378 }
379
380 index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
381 offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
382
383 bio_for_each_segment(bvec, bio, i) {
384 int max_transfer_size = PAGE_SIZE - offset;
385
386 if (bvec->bv_len > max_transfer_size) {
387 /*
388 * zram_bvec_rw() can only make operation on a single
389 * zram page. Split the bio vector.
390 */
391 struct bio_vec bv;
392
393 bv.bv_page = bvec->bv_page;
394 bv.bv_len = max_transfer_size;
395 bv.bv_offset = bvec->bv_offset;
396
397 if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
398 goto out;
399
400 bv.bv_len = bvec->bv_len - max_transfer_size;
401 bv.bv_offset += max_transfer_size;
402 if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
403 goto out;
404 } else
405 if (zram_bvec_rw(zram, bvec, index, offset, bio, rw)
406 < 0)
407 goto out;
408
409 update_position(&index, &offset, bvec);
410 }
411
412 set_bit(BIO_UPTODATE, &bio->bi_flags);
413 bio_endio(bio, 0);
414 return;
415
416 out:
417 bio_io_error(bio);
418 }
419
420 /*
421 * Check if request is within bounds and aligned on zram logical blocks.
422 */
423 static inline int valid_io_request(struct zram *zram, struct bio *bio)
424 {
425 if (unlikely(
426 (bio->bi_sector >= (zram->disksize >> SECTOR_SHIFT)) ||
427 (bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)) ||
428 (bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))) {
429
430 return 0;
431 }
432
433 /* I/O request is valid */
434 return 1;
435 }
436
437 /*
438 * Handler function for all zram I/O requests.
439 */
440 static void zram_make_request(struct request_queue *queue, struct bio *bio)
441 {
442 struct zram *zram = queue->queuedata;
443
444 down_read(&zram->init_lock);
445 if (unlikely(!zram->init_done))
446 goto error;
447
448 if (!valid_io_request(zram, bio)) {
449 zram_stat64_inc(zram, &zram->stats.invalid_io);
450 goto error;
451 }
452
453 __zram_make_request(zram, bio, bio_data_dir(bio));
454 up_read(&zram->init_lock);
455
456 return;
457
458 error:
459 up_read(&zram->init_lock);
460 bio_io_error(bio);
461 }
462
463 static void __zram_reset_device(struct zram *zram)
464 {
465 size_t index;
466 struct zram_meta *meta;
467
468 if (!zram->init_done)
469 return;
470
471 meta = zram->meta;
472 zram->init_done = 0;
473
474 /* Free all pages that are still in this zram device */
475 for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
476 unsigned long handle = meta->table[index].handle;
477 if (!handle)
478 continue;
479
480 zs_free(meta->mem_pool, handle);
481 }
482
483 zram_meta_free(zram->meta);
484 zram->meta = NULL;
485 /* Reset stats */
486 memset(&zram->stats, 0, sizeof(zram->stats));
487
488 zram->disksize = 0;
489 set_capacity(zram->disk, 0);
490 }
491
492 void zram_reset_device(struct zram *zram)
493 {
494 down_write(&zram->init_lock);
495 __zram_reset_device(zram);
496 up_write(&zram->init_lock);
497 }
498
499 void zram_meta_free(struct zram_meta *meta)
500 {
501 zs_destroy_pool(meta->mem_pool);
502 kfree(meta->compress_workmem);
503 free_pages((unsigned long)meta->compress_buffer, 1);
504 vfree(meta->table);
505 kfree(meta);
506 }
507
508 struct zram_meta *zram_meta_alloc(u64 disksize)
509 {
510 size_t num_pages;
511 struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
512 if (!meta)
513 goto out;
514
515 meta->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
516 if (!meta->compress_workmem)
517 goto free_meta;
518
519 meta->compress_buffer =
520 (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
521 if (!meta->compress_buffer) {
522 pr_err("Error allocating compressor buffer space\n");
523 goto free_workmem;
524 }
525
526 num_pages = disksize >> PAGE_SHIFT;
527 meta->table = vzalloc(num_pages * sizeof(*meta->table));
528 if (!meta->table) {
529 pr_err("Error allocating zram address table\n");
530 goto free_buffer;
531 }
532
533 meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
534 if (!meta->mem_pool) {
535 pr_err("Error creating memory pool\n");
536 goto free_table;
537 }
538
539 return meta;
540
541 free_table:
542 vfree(meta->table);
543 free_buffer:
544 free_pages((unsigned long)meta->compress_buffer, 1);
545 free_workmem:
546 kfree(meta->compress_workmem);
547 free_meta:
548 kfree(meta);
549 meta = NULL;
550 out:
551 return meta;
552 }
553
554 void zram_init_device(struct zram *zram, struct zram_meta *meta)
555 {
556 if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) {
557 pr_info(
558 "There is little point creating a zram of greater than "
559 "twice the size of memory since we expect a 2:1 compression "
560 "ratio. Note that zram uses about 0.1%% of the size of "
561 "the disk when not in use so a huge zram is "
562 "wasteful.\n"
563 "\tMemory Size: %lu kB\n"
564 "\tSize you selected: %llu kB\n"
565 "Continuing anyway ...\n",
566 (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10
567 );
568 }
569
570 /* zram devices sort of resembles non-rotational disks */
571 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
572
573 zram->meta = meta;
574 zram->init_done = 1;
575
576 pr_debug("Initialization done!\n");
577 }
578
579 static void zram_slot_free_notify(struct block_device *bdev,
580 unsigned long index)
581 {
582 struct zram *zram;
583
584 zram = bdev->bd_disk->private_data;
585 down_write(&zram->lock);
586 zram_free_page(zram, index);
587 up_write(&zram->lock);
588 zram_stat64_inc(zram, &zram->stats.notify_free);
589 }
590
591 static const struct block_device_operations zram_devops = {
592 .swap_slot_free_notify = zram_slot_free_notify,
593 .owner = THIS_MODULE
594 };
595
596 static int create_device(struct zram *zram, int device_id)
597 {
598 int ret = 0;
599
600 init_rwsem(&zram->lock);
601 init_rwsem(&zram->init_lock);
602 spin_lock_init(&zram->stat64_lock);
603
604 zram->queue = blk_alloc_queue(GFP_KERNEL);
605 if (!zram->queue) {
606 pr_err("Error allocating disk queue for device %d\n",
607 device_id);
608 ret = -ENOMEM;
609 goto out;
610 }
611
612 blk_queue_make_request(zram->queue, zram_make_request);
613 zram->queue->queuedata = zram;
614
615 /* gendisk structure */
616 zram->disk = alloc_disk(1);
617 if (!zram->disk) {
618 blk_cleanup_queue(zram->queue);
619 pr_warn("Error allocating disk structure for device %d\n",
620 device_id);
621 ret = -ENOMEM;
622 goto out;
623 }
624
625 zram->disk->major = zram_major;
626 zram->disk->first_minor = device_id;
627 zram->disk->fops = &zram_devops;
628 zram->disk->queue = zram->queue;
629 zram->disk->private_data = zram;
630 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
631
632 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
633 set_capacity(zram->disk, 0);
634
635 /*
636 * To ensure that we always get PAGE_SIZE aligned
637 * and n*PAGE_SIZED sized I/O requests.
638 */
639 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
640 blk_queue_logical_block_size(zram->disk->queue,
641 ZRAM_LOGICAL_BLOCK_SIZE);
642 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
643 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
644
645 add_disk(zram->disk);
646
647 ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
648 &zram_disk_attr_group);
649 if (ret < 0) {
650 pr_warn("Error creating sysfs group");
651 goto out;
652 }
653
654 zram->init_done = 0;
655
656 out:
657 return ret;
658 }
659
660 static void destroy_device(struct zram *zram)
661 {
662 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
663 &zram_disk_attr_group);
664
665 if (zram->disk) {
666 del_gendisk(zram->disk);
667 put_disk(zram->disk);
668 }
669
670 if (zram->queue)
671 blk_cleanup_queue(zram->queue);
672 }
673
674 unsigned int zram_get_num_devices(void)
675 {
676 return num_devices;
677 }
678
679 static int __init zram_init(void)
680 {
681 int ret, dev_id;
682
683 if (num_devices > max_num_devices) {
684 pr_warn("Invalid value for num_devices: %u\n",
685 num_devices);
686 ret = -EINVAL;
687 goto out;
688 }
689
690 zram_major = register_blkdev(0, "zram");
691 if (zram_major <= 0) {
692 pr_warn("Unable to get major number\n");
693 ret = -EBUSY;
694 goto out;
695 }
696
697 /* Allocate the device array and initialize each one */
698 zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
699 if (!zram_devices) {
700 ret = -ENOMEM;
701 goto unregister;
702 }
703
704 for (dev_id = 0; dev_id < num_devices; dev_id++) {
705 ret = create_device(&zram_devices[dev_id], dev_id);
706 if (ret)
707 goto free_devices;
708 }
709
710 pr_info("Created %u device(s) ...\n", num_devices);
711
712 return 0;
713
714 free_devices:
715 while (dev_id)
716 destroy_device(&zram_devices[--dev_id]);
717 kfree(zram_devices);
718 unregister:
719 unregister_blkdev(zram_major, "zram");
720 out:
721 return ret;
722 }
723
724 static void __exit zram_exit(void)
725 {
726 int i;
727 struct zram *zram;
728
729 for (i = 0; i < num_devices; i++) {
730 zram = &zram_devices[i];
731
732 get_disk(zram->disk);
733 destroy_device(zram);
734 zram_reset_device(zram);
735 put_disk(zram->disk);
736 }
737
738 unregister_blkdev(zram_major, "zram");
739
740 kfree(zram_devices);
741 pr_debug("Cleanup done!\n");
742 }
743
744 module_param(num_devices, uint, 0);
745 MODULE_PARM_DESC(num_devices, "Number of zram devices");
746
747 module_init(zram_init);
748 module_exit(zram_exit);
749
750 MODULE_LICENSE("Dual BSD/GPL");
751 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
752 MODULE_DESCRIPTION("Compressed RAM Block Device");
This page took 0.045856 seconds and 6 git commands to generate.