zram: avoid access beyond the zram device
[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 zram->stats.pages_zero++;
276 zram_set_flag(meta, index, ZRAM_ZERO);
277 ret = 0;
278 goto out;
279 }
280
281 ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
282 meta->compress_workmem);
283
284 if (!is_partial_io(bvec)) {
285 kunmap_atomic(user_mem);
286 user_mem = NULL;
287 uncmem = NULL;
288 }
289
290 if (unlikely(ret != LZO_E_OK)) {
291 pr_err("Compression failed! err=%d\n", ret);
292 goto out;
293 }
294
295 if (unlikely(clen > max_zpage_size)) {
296 zram->stats.bad_compress++;
297 clen = PAGE_SIZE;
298 src = NULL;
299 if (is_partial_io(bvec))
300 src = uncmem;
301 }
302
303 handle = zs_malloc(meta->mem_pool, clen);
304 if (!handle) {
305 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
306 index, clen);
307 ret = -ENOMEM;
308 goto out;
309 }
310 cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
311
312 if ((clen == PAGE_SIZE) && !is_partial_io(bvec))
313 src = kmap_atomic(page);
314 memcpy(cmem, src, clen);
315 if ((clen == PAGE_SIZE) && !is_partial_io(bvec))
316 kunmap_atomic(src);
317
318 zs_unmap_object(meta->mem_pool, handle);
319
320 meta->table[index].handle = handle;
321 meta->table[index].size = clen;
322
323 /* Update stats */
324 zram_stat64_add(zram, &zram->stats.compr_size, clen);
325 zram->stats.pages_stored++;
326 if (clen <= PAGE_SIZE / 2)
327 zram->stats.good_compress++;
328
329 out:
330 if (is_partial_io(bvec))
331 kfree(uncmem);
332
333 if (ret)
334 zram_stat64_inc(zram, &zram->stats.failed_writes);
335 return ret;
336 }
337
338 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
339 int offset, struct bio *bio, int rw)
340 {
341 int ret;
342
343 if (rw == READ) {
344 down_read(&zram->lock);
345 ret = zram_bvec_read(zram, bvec, index, offset, bio);
346 up_read(&zram->lock);
347 } else {
348 down_write(&zram->lock);
349 ret = zram_bvec_write(zram, bvec, index, offset);
350 up_write(&zram->lock);
351 }
352
353 return ret;
354 }
355
356 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
357 {
358 if (*offset + bvec->bv_len >= PAGE_SIZE)
359 (*index)++;
360 *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
361 }
362
363 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
364 {
365 int i, offset;
366 u32 index;
367 struct bio_vec *bvec;
368
369 switch (rw) {
370 case READ:
371 zram_stat64_inc(zram, &zram->stats.num_reads);
372 break;
373 case WRITE:
374 zram_stat64_inc(zram, &zram->stats.num_writes);
375 break;
376 }
377
378 index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
379 offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
380
381 bio_for_each_segment(bvec, bio, i) {
382 int max_transfer_size = PAGE_SIZE - offset;
383
384 if (bvec->bv_len > max_transfer_size) {
385 /*
386 * zram_bvec_rw() can only make operation on a single
387 * zram page. Split the bio vector.
388 */
389 struct bio_vec bv;
390
391 bv.bv_page = bvec->bv_page;
392 bv.bv_len = max_transfer_size;
393 bv.bv_offset = bvec->bv_offset;
394
395 if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
396 goto out;
397
398 bv.bv_len = bvec->bv_len - max_transfer_size;
399 bv.bv_offset += max_transfer_size;
400 if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
401 goto out;
402 } else
403 if (zram_bvec_rw(zram, bvec, index, offset, bio, rw)
404 < 0)
405 goto out;
406
407 update_position(&index, &offset, bvec);
408 }
409
410 set_bit(BIO_UPTODATE, &bio->bi_flags);
411 bio_endio(bio, 0);
412 return;
413
414 out:
415 bio_io_error(bio);
416 }
417
418 /*
419 * Check if request is within bounds and aligned on zram logical blocks.
420 */
421 static inline int valid_io_request(struct zram *zram, struct bio *bio)
422 {
423 u64 start, end, bound;
424
425 /* unaligned request */
426 if (unlikely(bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
427 return 0;
428 if (unlikely(bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
429 return 0;
430
431 start = bio->bi_sector;
432 end = start + (bio->bi_size >> SECTOR_SHIFT);
433 bound = zram->disksize >> SECTOR_SHIFT;
434 /* out of range range */
435 if (unlikely(start >= bound || end >= bound || start > end))
436 return 0;
437
438 /* I/O request is valid */
439 return 1;
440 }
441
442 /*
443 * Handler function for all zram I/O requests.
444 */
445 static void zram_make_request(struct request_queue *queue, struct bio *bio)
446 {
447 struct zram *zram = queue->queuedata;
448
449 down_read(&zram->init_lock);
450 if (unlikely(!zram->init_done))
451 goto error;
452
453 if (!valid_io_request(zram, bio)) {
454 zram_stat64_inc(zram, &zram->stats.invalid_io);
455 goto error;
456 }
457
458 __zram_make_request(zram, bio, bio_data_dir(bio));
459 up_read(&zram->init_lock);
460
461 return;
462
463 error:
464 up_read(&zram->init_lock);
465 bio_io_error(bio);
466 }
467
468 static void __zram_reset_device(struct zram *zram)
469 {
470 size_t index;
471 struct zram_meta *meta;
472
473 if (!zram->init_done)
474 return;
475
476 meta = zram->meta;
477 zram->init_done = 0;
478
479 /* Free all pages that are still in this zram device */
480 for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
481 unsigned long handle = meta->table[index].handle;
482 if (!handle)
483 continue;
484
485 zs_free(meta->mem_pool, handle);
486 }
487
488 zram_meta_free(zram->meta);
489 zram->meta = NULL;
490 /* Reset stats */
491 memset(&zram->stats, 0, sizeof(zram->stats));
492
493 zram->disksize = 0;
494 set_capacity(zram->disk, 0);
495 }
496
497 void zram_reset_device(struct zram *zram)
498 {
499 down_write(&zram->init_lock);
500 __zram_reset_device(zram);
501 up_write(&zram->init_lock);
502 }
503
504 void zram_meta_free(struct zram_meta *meta)
505 {
506 zs_destroy_pool(meta->mem_pool);
507 kfree(meta->compress_workmem);
508 free_pages((unsigned long)meta->compress_buffer, 1);
509 vfree(meta->table);
510 kfree(meta);
511 }
512
513 struct zram_meta *zram_meta_alloc(u64 disksize)
514 {
515 size_t num_pages;
516 struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
517 if (!meta)
518 goto out;
519
520 meta->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
521 if (!meta->compress_workmem)
522 goto free_meta;
523
524 meta->compress_buffer =
525 (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
526 if (!meta->compress_buffer) {
527 pr_err("Error allocating compressor buffer space\n");
528 goto free_workmem;
529 }
530
531 num_pages = disksize >> PAGE_SHIFT;
532 meta->table = vzalloc(num_pages * sizeof(*meta->table));
533 if (!meta->table) {
534 pr_err("Error allocating zram address table\n");
535 goto free_buffer;
536 }
537
538 meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
539 if (!meta->mem_pool) {
540 pr_err("Error creating memory pool\n");
541 goto free_table;
542 }
543
544 return meta;
545
546 free_table:
547 vfree(meta->table);
548 free_buffer:
549 free_pages((unsigned long)meta->compress_buffer, 1);
550 free_workmem:
551 kfree(meta->compress_workmem);
552 free_meta:
553 kfree(meta);
554 meta = NULL;
555 out:
556 return meta;
557 }
558
559 void zram_init_device(struct zram *zram, struct zram_meta *meta)
560 {
561 if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) {
562 pr_info(
563 "There is little point creating a zram of greater than "
564 "twice the size of memory since we expect a 2:1 compression "
565 "ratio. Note that zram uses about 0.1%% of the size of "
566 "the disk when not in use so a huge zram is "
567 "wasteful.\n"
568 "\tMemory Size: %lu kB\n"
569 "\tSize you selected: %llu kB\n"
570 "Continuing anyway ...\n",
571 (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10
572 );
573 }
574
575 /* zram devices sort of resembles non-rotational disks */
576 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
577
578 zram->meta = meta;
579 zram->init_done = 1;
580
581 pr_debug("Initialization done!\n");
582 }
583
584 static void zram_slot_free_notify(struct block_device *bdev,
585 unsigned long index)
586 {
587 struct zram *zram;
588
589 zram = bdev->bd_disk->private_data;
590 down_write(&zram->lock);
591 zram_free_page(zram, index);
592 up_write(&zram->lock);
593 zram_stat64_inc(zram, &zram->stats.notify_free);
594 }
595
596 static const struct block_device_operations zram_devops = {
597 .swap_slot_free_notify = zram_slot_free_notify,
598 .owner = THIS_MODULE
599 };
600
601 static int create_device(struct zram *zram, int device_id)
602 {
603 int ret = -ENOMEM;
604
605 init_rwsem(&zram->lock);
606 init_rwsem(&zram->init_lock);
607 spin_lock_init(&zram->stat64_lock);
608
609 zram->queue = blk_alloc_queue(GFP_KERNEL);
610 if (!zram->queue) {
611 pr_err("Error allocating disk queue for device %d\n",
612 device_id);
613 goto out;
614 }
615
616 blk_queue_make_request(zram->queue, zram_make_request);
617 zram->queue->queuedata = zram;
618
619 /* gendisk structure */
620 zram->disk = alloc_disk(1);
621 if (!zram->disk) {
622 pr_warn("Error allocating disk structure for device %d\n",
623 device_id);
624 goto out_free_queue;
625 }
626
627 zram->disk->major = zram_major;
628 zram->disk->first_minor = device_id;
629 zram->disk->fops = &zram_devops;
630 zram->disk->queue = zram->queue;
631 zram->disk->private_data = zram;
632 snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
633
634 /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
635 set_capacity(zram->disk, 0);
636
637 /*
638 * To ensure that we always get PAGE_SIZE aligned
639 * and n*PAGE_SIZED sized I/O requests.
640 */
641 blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
642 blk_queue_logical_block_size(zram->disk->queue,
643 ZRAM_LOGICAL_BLOCK_SIZE);
644 blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
645 blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
646
647 add_disk(zram->disk);
648
649 ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
650 &zram_disk_attr_group);
651 if (ret < 0) {
652 pr_warn("Error creating sysfs group");
653 goto out_free_disk;
654 }
655
656 zram->init_done = 0;
657 return 0;
658
659 out_free_disk:
660 del_gendisk(zram->disk);
661 put_disk(zram->disk);
662 out_free_queue:
663 blk_cleanup_queue(zram->queue);
664 out:
665 return ret;
666 }
667
668 static void destroy_device(struct zram *zram)
669 {
670 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
671 &zram_disk_attr_group);
672
673 if (zram->disk) {
674 del_gendisk(zram->disk);
675 put_disk(zram->disk);
676 }
677
678 if (zram->queue)
679 blk_cleanup_queue(zram->queue);
680 }
681
682 unsigned int zram_get_num_devices(void)
683 {
684 return num_devices;
685 }
686
687 static int __init zram_init(void)
688 {
689 int ret, dev_id;
690
691 if (num_devices > max_num_devices) {
692 pr_warn("Invalid value for num_devices: %u\n",
693 num_devices);
694 ret = -EINVAL;
695 goto out;
696 }
697
698 zram_major = register_blkdev(0, "zram");
699 if (zram_major <= 0) {
700 pr_warn("Unable to get major number\n");
701 ret = -EBUSY;
702 goto out;
703 }
704
705 /* Allocate the device array and initialize each one */
706 zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
707 if (!zram_devices) {
708 ret = -ENOMEM;
709 goto unregister;
710 }
711
712 for (dev_id = 0; dev_id < num_devices; dev_id++) {
713 ret = create_device(&zram_devices[dev_id], dev_id);
714 if (ret)
715 goto free_devices;
716 }
717
718 pr_info("Created %u device(s) ...\n", num_devices);
719
720 return 0;
721
722 free_devices:
723 while (dev_id)
724 destroy_device(&zram_devices[--dev_id]);
725 kfree(zram_devices);
726 unregister:
727 unregister_blkdev(zram_major, "zram");
728 out:
729 return ret;
730 }
731
732 static void __exit zram_exit(void)
733 {
734 int i;
735 struct zram *zram;
736
737 for (i = 0; i < num_devices; i++) {
738 zram = &zram_devices[i];
739
740 get_disk(zram->disk);
741 destroy_device(zram);
742 zram_reset_device(zram);
743 put_disk(zram->disk);
744 }
745
746 unregister_blkdev(zram_major, "zram");
747
748 kfree(zram_devices);
749 pr_debug("Cleanup done!\n");
750 }
751
752 module_param(num_devices, uint, 0);
753 MODULE_PARM_DESC(num_devices, "Number of zram devices");
754
755 module_init(zram_init);
756 module_exit(zram_exit);
757
758 MODULE_LICENSE("Dual BSD/GPL");
759 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
760 MODULE_DESCRIPTION("Compressed RAM Block Device");
This page took 0.046771 seconds and 6 git commands to generate.