]>
Commit | Line | Data |
---|---|---|
306b0c95 | 1 | /* |
f1e3cfff | 2 | * Compressed RAM block device |
306b0c95 | 3 | * |
1130ebba | 4 | * Copyright (C) 2008, 2009, 2010 Nitin Gupta |
306b0c95 NG |
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 | ||
f1e3cfff | 15 | #define KMSG_COMPONENT "zram" |
306b0c95 NG |
16 | #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt |
17 | ||
b1f5b81e RJ |
18 | #ifdef CONFIG_ZRAM_DEBUG |
19 | #define DEBUG | |
20 | #endif | |
21 | ||
306b0c95 NG |
22 | #include <linux/module.h> |
23 | #include <linux/kernel.h> | |
8946a086 | 24 | #include <linux/bio.h> |
306b0c95 NG |
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> | |
5a0e3ad6 | 31 | #include <linux/slab.h> |
306b0c95 | 32 | #include <linux/lzo.h> |
306b0c95 | 33 | #include <linux/string.h> |
306b0c95 | 34 | #include <linux/vmalloc.h> |
306b0c95 | 35 | |
16a4bfb9 | 36 | #include "zram_drv.h" |
306b0c95 NG |
37 | |
38 | /* Globals */ | |
f1e3cfff | 39 | static int zram_major; |
43801f6e | 40 | struct zram *zram_devices; |
306b0c95 | 41 | |
306b0c95 | 42 | /* Module params (documentation at end) */ |
efd54f43 | 43 | unsigned int zram_num_devices; |
33863c21 NG |
44 | |
45 | static void zram_stat_inc(u32 *v) | |
46 | { | |
47 | *v = *v + 1; | |
48 | } | |
49 | ||
50 | static void zram_stat_dec(u32 *v) | |
51 | { | |
52 | *v = *v - 1; | |
53 | } | |
54 | ||
55 | static void zram_stat64_add(struct zram *zram, u64 *v, u64 inc) | |
56 | { | |
57 | spin_lock(&zram->stat64_lock); | |
58 | *v = *v + inc; | |
59 | spin_unlock(&zram->stat64_lock); | |
60 | } | |
61 | ||
62 | static void zram_stat64_sub(struct zram *zram, u64 *v, u64 dec) | |
63 | { | |
64 | spin_lock(&zram->stat64_lock); | |
65 | *v = *v - dec; | |
66 | spin_unlock(&zram->stat64_lock); | |
67 | } | |
68 | ||
69 | static void zram_stat64_inc(struct zram *zram, u64 *v) | |
70 | { | |
71 | zram_stat64_add(zram, v, 1); | |
72 | } | |
306b0c95 | 73 | |
f1e3cfff NG |
74 | static int zram_test_flag(struct zram *zram, u32 index, |
75 | enum zram_pageflags flag) | |
306b0c95 | 76 | { |
f1e3cfff | 77 | return zram->table[index].flags & BIT(flag); |
306b0c95 NG |
78 | } |
79 | ||
f1e3cfff NG |
80 | static void zram_set_flag(struct zram *zram, u32 index, |
81 | enum zram_pageflags flag) | |
306b0c95 | 82 | { |
f1e3cfff | 83 | zram->table[index].flags |= BIT(flag); |
306b0c95 NG |
84 | } |
85 | ||
f1e3cfff NG |
86 | static void zram_clear_flag(struct zram *zram, u32 index, |
87 | enum zram_pageflags flag) | |
306b0c95 | 88 | { |
f1e3cfff | 89 | zram->table[index].flags &= ~BIT(flag); |
306b0c95 NG |
90 | } |
91 | ||
92 | static int page_zero_filled(void *ptr) | |
93 | { | |
94 | unsigned int pos; | |
95 | unsigned long *page; | |
96 | ||
97 | page = (unsigned long *)ptr; | |
98 | ||
99 | for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) { | |
100 | if (page[pos]) | |
101 | return 0; | |
102 | } | |
103 | ||
104 | return 1; | |
105 | } | |
106 | ||
f1e3cfff | 107 | static void zram_set_disksize(struct zram *zram, size_t totalram_bytes) |
306b0c95 | 108 | { |
f1e3cfff | 109 | if (!zram->disksize) { |
306b0c95 NG |
110 | pr_info( |
111 | "disk size not provided. You can use disksize_kb module " | |
112 | "param to specify size.\nUsing default: (%u%% of RAM).\n", | |
113 | default_disksize_perc_ram | |
114 | ); | |
f1e3cfff | 115 | zram->disksize = default_disksize_perc_ram * |
306b0c95 NG |
116 | (totalram_bytes / 100); |
117 | } | |
118 | ||
f1e3cfff | 119 | if (zram->disksize > 2 * (totalram_bytes)) { |
306b0c95 | 120 | pr_info( |
f1e3cfff | 121 | "There is little point creating a zram of greater than " |
306b0c95 | 122 | "twice the size of memory since we expect a 2:1 compression " |
f1e3cfff NG |
123 | "ratio. Note that zram uses about 0.1%% of the size of " |
124 | "the disk when not in use so a huge zram is " | |
306b0c95 NG |
125 | "wasteful.\n" |
126 | "\tMemory Size: %zu kB\n" | |
33863c21 | 127 | "\tSize you selected: %llu kB\n" |
306b0c95 | 128 | "Continuing anyway ...\n", |
f1e3cfff | 129 | totalram_bytes >> 10, zram->disksize |
306b0c95 NG |
130 | ); |
131 | } | |
132 | ||
f1e3cfff | 133 | zram->disksize &= PAGE_MASK; |
306b0c95 NG |
134 | } |
135 | ||
f1e3cfff | 136 | static void zram_free_page(struct zram *zram, size_t index) |
306b0c95 NG |
137 | { |
138 | u32 clen; | |
139 | void *obj; | |
140 | ||
f1e3cfff NG |
141 | struct page *page = zram->table[index].page; |
142 | u32 offset = zram->table[index].offset; | |
306b0c95 NG |
143 | |
144 | if (unlikely(!page)) { | |
2e882281 NG |
145 | /* |
146 | * No memory is allocated for zero filled pages. | |
147 | * Simply clear zero page flag. | |
148 | */ | |
f1e3cfff NG |
149 | if (zram_test_flag(zram, index, ZRAM_ZERO)) { |
150 | zram_clear_flag(zram, index, ZRAM_ZERO); | |
151 | zram_stat_dec(&zram->stats.pages_zero); | |
306b0c95 NG |
152 | } |
153 | return; | |
154 | } | |
155 | ||
f1e3cfff | 156 | if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) { |
306b0c95 NG |
157 | clen = PAGE_SIZE; |
158 | __free_page(page); | |
f1e3cfff NG |
159 | zram_clear_flag(zram, index, ZRAM_UNCOMPRESSED); |
160 | zram_stat_dec(&zram->stats.pages_expand); | |
306b0c95 NG |
161 | goto out; |
162 | } | |
163 | ||
164 | obj = kmap_atomic(page, KM_USER0) + offset; | |
165 | clen = xv_get_object_size(obj) - sizeof(struct zobj_header); | |
166 | kunmap_atomic(obj, KM_USER0); | |
167 | ||
f1e3cfff | 168 | xv_free(zram->mem_pool, page, offset); |
306b0c95 | 169 | if (clen <= PAGE_SIZE / 2) |
f1e3cfff | 170 | zram_stat_dec(&zram->stats.good_compress); |
306b0c95 NG |
171 | |
172 | out: | |
33863c21 | 173 | zram_stat64_sub(zram, &zram->stats.compr_size, clen); |
f1e3cfff | 174 | zram_stat_dec(&zram->stats.pages_stored); |
306b0c95 | 175 | |
f1e3cfff NG |
176 | zram->table[index].page = NULL; |
177 | zram->table[index].offset = 0; | |
306b0c95 NG |
178 | } |
179 | ||
924bd88d | 180 | static void handle_zero_page(struct bio_vec *bvec) |
306b0c95 | 181 | { |
924bd88d | 182 | struct page *page = bvec->bv_page; |
306b0c95 | 183 | void *user_mem; |
306b0c95 NG |
184 | |
185 | user_mem = kmap_atomic(page, KM_USER0); | |
924bd88d | 186 | memset(user_mem + bvec->bv_offset, 0, bvec->bv_len); |
306b0c95 NG |
187 | kunmap_atomic(user_mem, KM_USER0); |
188 | ||
30fb8a71 | 189 | flush_dcache_page(page); |
306b0c95 NG |
190 | } |
191 | ||
924bd88d JM |
192 | static void handle_uncompressed_page(struct zram *zram, struct bio_vec *bvec, |
193 | u32 index, int offset) | |
306b0c95 | 194 | { |
924bd88d | 195 | struct page *page = bvec->bv_page; |
306b0c95 NG |
196 | unsigned char *user_mem, *cmem; |
197 | ||
306b0c95 | 198 | user_mem = kmap_atomic(page, KM_USER0); |
6a587e83 | 199 | cmem = kmap_atomic(zram->table[index].page, KM_USER1); |
306b0c95 | 200 | |
924bd88d | 201 | memcpy(user_mem + bvec->bv_offset, cmem + offset, bvec->bv_len); |
306b0c95 | 202 | kunmap_atomic(cmem, KM_USER1); |
dffbb44d | 203 | kunmap_atomic(user_mem, KM_USER0); |
306b0c95 | 204 | |
30fb8a71 | 205 | flush_dcache_page(page); |
306b0c95 NG |
206 | } |
207 | ||
924bd88d JM |
208 | static inline int is_partial_io(struct bio_vec *bvec) |
209 | { | |
210 | return bvec->bv_len != PAGE_SIZE; | |
211 | } | |
212 | ||
8c921b2b | 213 | static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec, |
924bd88d | 214 | u32 index, int offset, struct bio *bio) |
306b0c95 | 215 | { |
8c921b2b JM |
216 | int ret; |
217 | size_t clen; | |
218 | struct page *page; | |
219 | struct zobj_header *zheader; | |
924bd88d | 220 | unsigned char *user_mem, *cmem, *uncmem = NULL; |
a1dd52af | 221 | |
8c921b2b | 222 | page = bvec->bv_page; |
306b0c95 | 223 | |
8c921b2b | 224 | if (zram_test_flag(zram, index, ZRAM_ZERO)) { |
924bd88d | 225 | handle_zero_page(bvec); |
8c921b2b JM |
226 | return 0; |
227 | } | |
306b0c95 | 228 | |
8c921b2b JM |
229 | /* Requested page is not present in compressed area */ |
230 | if (unlikely(!zram->table[index].page)) { | |
231 | pr_debug("Read before write: sector=%lu, size=%u", | |
232 | (ulong)(bio->bi_sector), bio->bi_size); | |
924bd88d | 233 | handle_zero_page(bvec); |
8c921b2b JM |
234 | return 0; |
235 | } | |
306b0c95 | 236 | |
8c921b2b JM |
237 | /* Page is stored uncompressed since it's incompressible */ |
238 | if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) { | |
924bd88d | 239 | handle_uncompressed_page(zram, bvec, index, offset); |
8c921b2b JM |
240 | return 0; |
241 | } | |
306b0c95 | 242 | |
924bd88d JM |
243 | if (is_partial_io(bvec)) { |
244 | /* Use a temporary buffer to decompress the page */ | |
245 | uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
246 | if (!uncmem) { | |
247 | pr_info("Error allocating temp memory!\n"); | |
248 | return -ENOMEM; | |
249 | } | |
250 | } | |
251 | ||
8c921b2b | 252 | user_mem = kmap_atomic(page, KM_USER0); |
924bd88d JM |
253 | if (!is_partial_io(bvec)) |
254 | uncmem = user_mem; | |
8c921b2b | 255 | clen = PAGE_SIZE; |
306b0c95 | 256 | |
8c921b2b JM |
257 | cmem = kmap_atomic(zram->table[index].page, KM_USER1) + |
258 | zram->table[index].offset; | |
306b0c95 | 259 | |
8c921b2b JM |
260 | ret = lzo1x_decompress_safe(cmem + sizeof(*zheader), |
261 | xv_get_object_size(cmem) - sizeof(*zheader), | |
924bd88d JM |
262 | uncmem, &clen); |
263 | ||
264 | if (is_partial_io(bvec)) { | |
265 | memcpy(user_mem + bvec->bv_offset, uncmem + offset, | |
266 | bvec->bv_len); | |
267 | kfree(uncmem); | |
268 | } | |
306b0c95 | 269 | |
8c921b2b | 270 | kunmap_atomic(cmem, KM_USER1); |
dffbb44d | 271 | kunmap_atomic(user_mem, KM_USER0); |
a1dd52af | 272 | |
8c921b2b JM |
273 | /* Should NEVER happen. Return bio error if it does. */ |
274 | if (unlikely(ret != LZO_E_OK)) { | |
275 | pr_err("Decompression failed! err=%d, page=%u\n", ret, index); | |
276 | zram_stat64_inc(zram, &zram->stats.failed_reads); | |
277 | return ret; | |
a1dd52af | 278 | } |
306b0c95 | 279 | |
8c921b2b | 280 | flush_dcache_page(page); |
306b0c95 | 281 | |
8c921b2b | 282 | return 0; |
306b0c95 NG |
283 | } |
284 | ||
924bd88d JM |
285 | static int zram_read_before_write(struct zram *zram, char *mem, u32 index) |
286 | { | |
287 | int ret; | |
288 | size_t clen = PAGE_SIZE; | |
289 | struct zobj_header *zheader; | |
290 | unsigned char *cmem; | |
291 | ||
292 | if (zram_test_flag(zram, index, ZRAM_ZERO) || | |
293 | !zram->table[index].page) { | |
294 | memset(mem, 0, PAGE_SIZE); | |
295 | return 0; | |
296 | } | |
297 | ||
298 | cmem = kmap_atomic(zram->table[index].page, KM_USER0) + | |
299 | zram->table[index].offset; | |
300 | ||
301 | /* Page is stored uncompressed since it's incompressible */ | |
302 | if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) { | |
303 | memcpy(mem, cmem, PAGE_SIZE); | |
304 | kunmap_atomic(cmem, KM_USER0); | |
305 | return 0; | |
306 | } | |
307 | ||
308 | ret = lzo1x_decompress_safe(cmem + sizeof(*zheader), | |
309 | xv_get_object_size(cmem) - sizeof(*zheader), | |
310 | mem, &clen); | |
311 | kunmap_atomic(cmem, KM_USER0); | |
312 | ||
313 | /* Should NEVER happen. Return bio error if it does. */ | |
314 | if (unlikely(ret != LZO_E_OK)) { | |
315 | pr_err("Decompression failed! err=%d, page=%u\n", ret, index); | |
316 | zram_stat64_inc(zram, &zram->stats.failed_reads); | |
317 | return ret; | |
318 | } | |
319 | ||
320 | return 0; | |
321 | } | |
322 | ||
323 | static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index, | |
324 | int offset) | |
306b0c95 | 325 | { |
8c921b2b | 326 | int ret; |
924bd88d | 327 | u32 store_offset; |
8c921b2b JM |
328 | size_t clen; |
329 | struct zobj_header *zheader; | |
330 | struct page *page, *page_store; | |
924bd88d | 331 | unsigned char *user_mem, *cmem, *src, *uncmem = NULL; |
306b0c95 | 332 | |
8c921b2b JM |
333 | page = bvec->bv_page; |
334 | src = zram->compress_buffer; | |
306b0c95 | 335 | |
924bd88d JM |
336 | if (is_partial_io(bvec)) { |
337 | /* | |
338 | * This is a partial IO. We need to read the full page | |
339 | * before to write the changes. | |
340 | */ | |
341 | uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
342 | if (!uncmem) { | |
343 | pr_info("Error allocating temp memory!\n"); | |
344 | ret = -ENOMEM; | |
345 | goto out; | |
346 | } | |
347 | ret = zram_read_before_write(zram, uncmem, index); | |
348 | if (ret) { | |
349 | kfree(uncmem); | |
350 | goto out; | |
351 | } | |
352 | } | |
353 | ||
8c921b2b JM |
354 | /* |
355 | * System overwrites unused sectors. Free memory associated | |
356 | * with this sector now. | |
357 | */ | |
358 | if (zram->table[index].page || | |
359 | zram_test_flag(zram, index, ZRAM_ZERO)) | |
360 | zram_free_page(zram, index); | |
306b0c95 | 361 | |
8c921b2b | 362 | user_mem = kmap_atomic(page, KM_USER0); |
924bd88d JM |
363 | |
364 | if (is_partial_io(bvec)) | |
365 | memcpy(uncmem + offset, user_mem + bvec->bv_offset, | |
366 | bvec->bv_len); | |
367 | else | |
368 | uncmem = user_mem; | |
369 | ||
370 | if (page_zero_filled(uncmem)) { | |
8c921b2b | 371 | kunmap_atomic(user_mem, KM_USER0); |
924bd88d JM |
372 | if (is_partial_io(bvec)) |
373 | kfree(uncmem); | |
8c921b2b JM |
374 | zram_stat_inc(&zram->stats.pages_zero); |
375 | zram_set_flag(zram, index, ZRAM_ZERO); | |
924bd88d JM |
376 | ret = 0; |
377 | goto out; | |
8c921b2b | 378 | } |
306b0c95 | 379 | |
924bd88d | 380 | ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen, |
8c921b2b | 381 | zram->compress_workmem); |
306b0c95 | 382 | |
8c921b2b | 383 | kunmap_atomic(user_mem, KM_USER0); |
924bd88d JM |
384 | if (is_partial_io(bvec)) |
385 | kfree(uncmem); | |
306b0c95 | 386 | |
8c921b2b | 387 | if (unlikely(ret != LZO_E_OK)) { |
8c921b2b | 388 | pr_err("Compression failed! err=%d\n", ret); |
924bd88d | 389 | goto out; |
8c921b2b | 390 | } |
306b0c95 | 391 | |
8c921b2b JM |
392 | /* |
393 | * Page is incompressible. Store it as-is (uncompressed) | |
394 | * since we do not want to return too many disk write | |
395 | * errors which has side effect of hanging the system. | |
396 | */ | |
397 | if (unlikely(clen > max_zpage_size)) { | |
398 | clen = PAGE_SIZE; | |
399 | page_store = alloc_page(GFP_NOIO | __GFP_HIGHMEM); | |
400 | if (unlikely(!page_store)) { | |
8c921b2b JM |
401 | pr_info("Error allocating memory for " |
402 | "incompressible page: %u\n", index); | |
924bd88d JM |
403 | ret = -ENOMEM; |
404 | goto out; | |
405 | } | |
a1dd52af | 406 | |
924bd88d | 407 | store_offset = 0; |
8c921b2b JM |
408 | zram_set_flag(zram, index, ZRAM_UNCOMPRESSED); |
409 | zram_stat_inc(&zram->stats.pages_expand); | |
410 | zram->table[index].page = page_store; | |
411 | src = kmap_atomic(page, KM_USER0); | |
412 | goto memstore; | |
413 | } | |
306b0c95 | 414 | |
8c921b2b | 415 | if (xv_malloc(zram->mem_pool, clen + sizeof(*zheader), |
924bd88d | 416 | &zram->table[index].page, &store_offset, |
8c921b2b | 417 | GFP_NOIO | __GFP_HIGHMEM)) { |
8c921b2b JM |
418 | pr_info("Error allocating memory for compressed " |
419 | "page: %u, size=%zu\n", index, clen); | |
924bd88d JM |
420 | ret = -ENOMEM; |
421 | goto out; | |
8c921b2b | 422 | } |
306b0c95 NG |
423 | |
424 | memstore: | |
924bd88d | 425 | zram->table[index].offset = store_offset; |
306b0c95 | 426 | |
8c921b2b JM |
427 | cmem = kmap_atomic(zram->table[index].page, KM_USER1) + |
428 | zram->table[index].offset; | |
306b0c95 NG |
429 | |
430 | #if 0 | |
8c921b2b JM |
431 | /* Back-reference needed for memory defragmentation */ |
432 | if (!zram_test_flag(zram, index, ZRAM_UNCOMPRESSED)) { | |
433 | zheader = (struct zobj_header *)cmem; | |
434 | zheader->table_idx = index; | |
435 | cmem += sizeof(*zheader); | |
436 | } | |
306b0c95 NG |
437 | #endif |
438 | ||
8c921b2b | 439 | memcpy(cmem, src, clen); |
306b0c95 | 440 | |
8c921b2b JM |
441 | kunmap_atomic(cmem, KM_USER1); |
442 | if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) | |
443 | kunmap_atomic(src, KM_USER0); | |
306b0c95 | 444 | |
8c921b2b JM |
445 | /* Update stats */ |
446 | zram_stat64_add(zram, &zram->stats.compr_size, clen); | |
447 | zram_stat_inc(&zram->stats.pages_stored); | |
448 | if (clen <= PAGE_SIZE / 2) | |
449 | zram_stat_inc(&zram->stats.good_compress); | |
306b0c95 | 450 | |
8c921b2b | 451 | return 0; |
924bd88d JM |
452 | |
453 | out: | |
454 | if (ret) | |
455 | zram_stat64_inc(zram, &zram->stats.failed_writes); | |
456 | return ret; | |
8c921b2b JM |
457 | } |
458 | ||
459 | static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index, | |
924bd88d | 460 | int offset, struct bio *bio, int rw) |
8c921b2b | 461 | { |
c5bde238 | 462 | int ret; |
8c921b2b | 463 | |
c5bde238 JM |
464 | if (rw == READ) { |
465 | down_read(&zram->lock); | |
466 | ret = zram_bvec_read(zram, bvec, index, offset, bio); | |
467 | up_read(&zram->lock); | |
468 | } else { | |
469 | down_write(&zram->lock); | |
470 | ret = zram_bvec_write(zram, bvec, index, offset); | |
471 | up_write(&zram->lock); | |
472 | } | |
473 | ||
474 | return ret; | |
924bd88d JM |
475 | } |
476 | ||
477 | static void update_position(u32 *index, int *offset, struct bio_vec *bvec) | |
478 | { | |
479 | if (*offset + bvec->bv_len >= PAGE_SIZE) | |
480 | (*index)++; | |
481 | *offset = (*offset + bvec->bv_len) % PAGE_SIZE; | |
8c921b2b JM |
482 | } |
483 | ||
484 | static void __zram_make_request(struct zram *zram, struct bio *bio, int rw) | |
485 | { | |
924bd88d | 486 | int i, offset; |
8c921b2b JM |
487 | u32 index; |
488 | struct bio_vec *bvec; | |
489 | ||
490 | switch (rw) { | |
491 | case READ: | |
492 | zram_stat64_inc(zram, &zram->stats.num_reads); | |
493 | break; | |
494 | case WRITE: | |
495 | zram_stat64_inc(zram, &zram->stats.num_writes); | |
496 | break; | |
497 | } | |
498 | ||
499 | index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT; | |
924bd88d | 500 | offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT; |
8c921b2b JM |
501 | |
502 | bio_for_each_segment(bvec, bio, i) { | |
924bd88d JM |
503 | int max_transfer_size = PAGE_SIZE - offset; |
504 | ||
505 | if (bvec->bv_len > max_transfer_size) { | |
506 | /* | |
507 | * zram_bvec_rw() can only make operation on a single | |
508 | * zram page. Split the bio vector. | |
509 | */ | |
510 | struct bio_vec bv; | |
511 | ||
512 | bv.bv_page = bvec->bv_page; | |
513 | bv.bv_len = max_transfer_size; | |
514 | bv.bv_offset = bvec->bv_offset; | |
515 | ||
516 | if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0) | |
517 | goto out; | |
518 | ||
519 | bv.bv_len = bvec->bv_len - max_transfer_size; | |
520 | bv.bv_offset += max_transfer_size; | |
521 | if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0) | |
522 | goto out; | |
523 | } else | |
524 | if (zram_bvec_rw(zram, bvec, index, offset, bio, rw) | |
525 | < 0) | |
526 | goto out; | |
527 | ||
528 | update_position(&index, &offset, bvec); | |
a1dd52af | 529 | } |
306b0c95 NG |
530 | |
531 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
532 | bio_endio(bio, 0); | |
7d7854b4 | 533 | return; |
306b0c95 NG |
534 | |
535 | out: | |
306b0c95 | 536 | bio_io_error(bio); |
306b0c95 NG |
537 | } |
538 | ||
306b0c95 | 539 | /* |
924bd88d | 540 | * Check if request is within bounds and aligned on zram logical blocks. |
306b0c95 | 541 | */ |
f1e3cfff | 542 | static inline int valid_io_request(struct zram *zram, struct bio *bio) |
306b0c95 NG |
543 | { |
544 | if (unlikely( | |
f1e3cfff | 545 | (bio->bi_sector >= (zram->disksize >> SECTOR_SHIFT)) || |
924bd88d JM |
546 | (bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)) || |
547 | (bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))) { | |
306b0c95 NG |
548 | |
549 | return 0; | |
550 | } | |
551 | ||
a1dd52af | 552 | /* I/O request is valid */ |
306b0c95 NG |
553 | return 1; |
554 | } | |
555 | ||
556 | /* | |
f1e3cfff | 557 | * Handler function for all zram I/O requests. |
306b0c95 | 558 | */ |
5a7bbad2 | 559 | static void zram_make_request(struct request_queue *queue, struct bio *bio) |
306b0c95 | 560 | { |
f1e3cfff | 561 | struct zram *zram = queue->queuedata; |
306b0c95 | 562 | |
0900beae JM |
563 | if (unlikely(!zram->init_done) && zram_init_device(zram)) |
564 | goto error; | |
565 | ||
566 | down_read(&zram->init_lock); | |
567 | if (unlikely(!zram->init_done)) | |
568 | goto error_unlock; | |
569 | ||
f1e3cfff NG |
570 | if (!valid_io_request(zram, bio)) { |
571 | zram_stat64_inc(zram, &zram->stats.invalid_io); | |
0900beae | 572 | goto error_unlock; |
6642a67c JM |
573 | } |
574 | ||
8c921b2b | 575 | __zram_make_request(zram, bio, bio_data_dir(bio)); |
0900beae | 576 | up_read(&zram->init_lock); |
306b0c95 | 577 | |
b4fdcb02 | 578 | return; |
0900beae JM |
579 | |
580 | error_unlock: | |
581 | up_read(&zram->init_lock); | |
582 | error: | |
583 | bio_io_error(bio); | |
306b0c95 NG |
584 | } |
585 | ||
0900beae | 586 | void __zram_reset_device(struct zram *zram) |
306b0c95 | 587 | { |
97a06382 | 588 | size_t index; |
306b0c95 | 589 | |
f1e3cfff | 590 | zram->init_done = 0; |
7eef7533 | 591 | |
306b0c95 | 592 | /* Free various per-device buffers */ |
f1e3cfff NG |
593 | kfree(zram->compress_workmem); |
594 | free_pages((unsigned long)zram->compress_buffer, 1); | |
306b0c95 | 595 | |
f1e3cfff NG |
596 | zram->compress_workmem = NULL; |
597 | zram->compress_buffer = NULL; | |
306b0c95 | 598 | |
f1e3cfff NG |
599 | /* Free all pages that are still in this zram device */ |
600 | for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) { | |
306b0c95 NG |
601 | struct page *page; |
602 | u16 offset; | |
603 | ||
f1e3cfff NG |
604 | page = zram->table[index].page; |
605 | offset = zram->table[index].offset; | |
306b0c95 NG |
606 | |
607 | if (!page) | |
608 | continue; | |
609 | ||
f1e3cfff | 610 | if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) |
306b0c95 NG |
611 | __free_page(page); |
612 | else | |
f1e3cfff | 613 | xv_free(zram->mem_pool, page, offset); |
306b0c95 NG |
614 | } |
615 | ||
f1e3cfff NG |
616 | vfree(zram->table); |
617 | zram->table = NULL; | |
306b0c95 | 618 | |
f1e3cfff NG |
619 | xv_destroy_pool(zram->mem_pool); |
620 | zram->mem_pool = NULL; | |
306b0c95 | 621 | |
306b0c95 | 622 | /* Reset stats */ |
f1e3cfff | 623 | memset(&zram->stats, 0, sizeof(zram->stats)); |
306b0c95 | 624 | |
f1e3cfff | 625 | zram->disksize = 0; |
0900beae JM |
626 | } |
627 | ||
628 | void zram_reset_device(struct zram *zram) | |
629 | { | |
630 | down_write(&zram->init_lock); | |
631 | __zram_reset_device(zram); | |
632 | up_write(&zram->init_lock); | |
306b0c95 NG |
633 | } |
634 | ||
33863c21 | 635 | int zram_init_device(struct zram *zram) |
306b0c95 NG |
636 | { |
637 | int ret; | |
638 | size_t num_pages; | |
306b0c95 | 639 | |
0900beae | 640 | down_write(&zram->init_lock); |
484875ad | 641 | |
f1e3cfff | 642 | if (zram->init_done) { |
0900beae | 643 | up_write(&zram->init_lock); |
484875ad | 644 | return 0; |
306b0c95 NG |
645 | } |
646 | ||
f1e3cfff | 647 | zram_set_disksize(zram, totalram_pages << PAGE_SHIFT); |
306b0c95 | 648 | |
f1e3cfff NG |
649 | zram->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL); |
650 | if (!zram->compress_workmem) { | |
306b0c95 NG |
651 | pr_err("Error allocating compressor working memory!\n"); |
652 | ret = -ENOMEM; | |
5a18c531 | 653 | goto fail_no_table; |
306b0c95 NG |
654 | } |
655 | ||
fb927284 JM |
656 | zram->compress_buffer = |
657 | (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1); | |
f1e3cfff | 658 | if (!zram->compress_buffer) { |
306b0c95 NG |
659 | pr_err("Error allocating compressor buffer space\n"); |
660 | ret = -ENOMEM; | |
5a18c531 | 661 | goto fail_no_table; |
306b0c95 NG |
662 | } |
663 | ||
f1e3cfff | 664 | num_pages = zram->disksize >> PAGE_SHIFT; |
5b84cc78 | 665 | zram->table = vzalloc(num_pages * sizeof(*zram->table)); |
f1e3cfff NG |
666 | if (!zram->table) { |
667 | pr_err("Error allocating zram address table\n"); | |
306b0c95 | 668 | ret = -ENOMEM; |
5a18c531 | 669 | goto fail_no_table; |
306b0c95 | 670 | } |
306b0c95 | 671 | |
f1e3cfff | 672 | set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT); |
306b0c95 | 673 | |
f1e3cfff NG |
674 | /* zram devices sort of resembles non-rotational disks */ |
675 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue); | |
306b0c95 | 676 | |
f1e3cfff NG |
677 | zram->mem_pool = xv_create_pool(); |
678 | if (!zram->mem_pool) { | |
306b0c95 NG |
679 | pr_err("Error creating memory pool\n"); |
680 | ret = -ENOMEM; | |
681 | goto fail; | |
682 | } | |
683 | ||
f1e3cfff | 684 | zram->init_done = 1; |
0900beae | 685 | up_write(&zram->init_lock); |
306b0c95 NG |
686 | |
687 | pr_debug("Initialization done!\n"); | |
688 | return 0; | |
689 | ||
5a18c531 JM |
690 | fail_no_table: |
691 | /* To prevent accessing table entries during cleanup */ | |
692 | zram->disksize = 0; | |
306b0c95 | 693 | fail: |
0900beae JM |
694 | __zram_reset_device(zram); |
695 | up_write(&zram->init_lock); | |
306b0c95 NG |
696 | pr_err("Initialization failed: err=%d\n", ret); |
697 | return ret; | |
698 | } | |
699 | ||
2ccbec05 NG |
700 | static void zram_slot_free_notify(struct block_device *bdev, |
701 | unsigned long index) | |
107c161b | 702 | { |
f1e3cfff | 703 | struct zram *zram; |
107c161b | 704 | |
f1e3cfff NG |
705 | zram = bdev->bd_disk->private_data; |
706 | zram_free_page(zram, index); | |
707 | zram_stat64_inc(zram, &zram->stats.notify_free); | |
107c161b NG |
708 | } |
709 | ||
f1e3cfff | 710 | static const struct block_device_operations zram_devops = { |
f1e3cfff | 711 | .swap_slot_free_notify = zram_slot_free_notify, |
107c161b | 712 | .owner = THIS_MODULE |
306b0c95 NG |
713 | }; |
714 | ||
f1e3cfff | 715 | static int create_device(struct zram *zram, int device_id) |
306b0c95 | 716 | { |
de1a21a0 NG |
717 | int ret = 0; |
718 | ||
c5bde238 | 719 | init_rwsem(&zram->lock); |
0900beae | 720 | init_rwsem(&zram->init_lock); |
f1e3cfff | 721 | spin_lock_init(&zram->stat64_lock); |
306b0c95 | 722 | |
f1e3cfff NG |
723 | zram->queue = blk_alloc_queue(GFP_KERNEL); |
724 | if (!zram->queue) { | |
306b0c95 NG |
725 | pr_err("Error allocating disk queue for device %d\n", |
726 | device_id); | |
de1a21a0 NG |
727 | ret = -ENOMEM; |
728 | goto out; | |
306b0c95 NG |
729 | } |
730 | ||
f1e3cfff NG |
731 | blk_queue_make_request(zram->queue, zram_make_request); |
732 | zram->queue->queuedata = zram; | |
306b0c95 NG |
733 | |
734 | /* gendisk structure */ | |
f1e3cfff NG |
735 | zram->disk = alloc_disk(1); |
736 | if (!zram->disk) { | |
737 | blk_cleanup_queue(zram->queue); | |
306b0c95 NG |
738 | pr_warning("Error allocating disk structure for device %d\n", |
739 | device_id); | |
de1a21a0 NG |
740 | ret = -ENOMEM; |
741 | goto out; | |
306b0c95 NG |
742 | } |
743 | ||
f1e3cfff NG |
744 | zram->disk->major = zram_major; |
745 | zram->disk->first_minor = device_id; | |
746 | zram->disk->fops = &zram_devops; | |
747 | zram->disk->queue = zram->queue; | |
748 | zram->disk->private_data = zram; | |
749 | snprintf(zram->disk->disk_name, 16, "zram%d", device_id); | |
306b0c95 | 750 | |
33863c21 | 751 | /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */ |
f1e3cfff | 752 | set_capacity(zram->disk, 0); |
5d83d5a0 | 753 | |
a1dd52af NG |
754 | /* |
755 | * To ensure that we always get PAGE_SIZE aligned | |
756 | * and n*PAGE_SIZED sized I/O requests. | |
757 | */ | |
f1e3cfff | 758 | blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE); |
7b19b8d4 RJ |
759 | blk_queue_logical_block_size(zram->disk->queue, |
760 | ZRAM_LOGICAL_BLOCK_SIZE); | |
f1e3cfff NG |
761 | blk_queue_io_min(zram->disk->queue, PAGE_SIZE); |
762 | blk_queue_io_opt(zram->disk->queue, PAGE_SIZE); | |
5d83d5a0 | 763 | |
f1e3cfff | 764 | add_disk(zram->disk); |
306b0c95 | 765 | |
33863c21 NG |
766 | ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj, |
767 | &zram_disk_attr_group); | |
768 | if (ret < 0) { | |
769 | pr_warning("Error creating sysfs group"); | |
770 | goto out; | |
771 | } | |
33863c21 | 772 | |
f1e3cfff | 773 | zram->init_done = 0; |
de1a21a0 NG |
774 | |
775 | out: | |
776 | return ret; | |
306b0c95 NG |
777 | } |
778 | ||
f1e3cfff | 779 | static void destroy_device(struct zram *zram) |
306b0c95 | 780 | { |
33863c21 NG |
781 | sysfs_remove_group(&disk_to_dev(zram->disk)->kobj, |
782 | &zram_disk_attr_group); | |
33863c21 | 783 | |
f1e3cfff NG |
784 | if (zram->disk) { |
785 | del_gendisk(zram->disk); | |
786 | put_disk(zram->disk); | |
306b0c95 NG |
787 | } |
788 | ||
f1e3cfff NG |
789 | if (zram->queue) |
790 | blk_cleanup_queue(zram->queue); | |
306b0c95 NG |
791 | } |
792 | ||
f1e3cfff | 793 | static int __init zram_init(void) |
306b0c95 | 794 | { |
de1a21a0 | 795 | int ret, dev_id; |
306b0c95 | 796 | |
efd54f43 | 797 | if (zram_num_devices > max_num_devices) { |
306b0c95 | 798 | pr_warning("Invalid value for num_devices: %u\n", |
efd54f43 | 799 | zram_num_devices); |
de1a21a0 NG |
800 | ret = -EINVAL; |
801 | goto out; | |
306b0c95 NG |
802 | } |
803 | ||
f1e3cfff NG |
804 | zram_major = register_blkdev(0, "zram"); |
805 | if (zram_major <= 0) { | |
306b0c95 | 806 | pr_warning("Unable to get major number\n"); |
de1a21a0 NG |
807 | ret = -EBUSY; |
808 | goto out; | |
306b0c95 NG |
809 | } |
810 | ||
efd54f43 | 811 | if (!zram_num_devices) { |
306b0c95 | 812 | pr_info("num_devices not specified. Using default: 1\n"); |
efd54f43 | 813 | zram_num_devices = 1; |
306b0c95 NG |
814 | } |
815 | ||
816 | /* Allocate the device array and initialize each one */ | |
efd54f43 NW |
817 | pr_info("Creating %u devices ...\n", zram_num_devices); |
818 | zram_devices = kzalloc(zram_num_devices * sizeof(struct zram), GFP_KERNEL); | |
43801f6e | 819 | if (!zram_devices) { |
de1a21a0 NG |
820 | ret = -ENOMEM; |
821 | goto unregister; | |
822 | } | |
306b0c95 | 823 | |
efd54f43 | 824 | for (dev_id = 0; dev_id < zram_num_devices; dev_id++) { |
43801f6e | 825 | ret = create_device(&zram_devices[dev_id], dev_id); |
de1a21a0 | 826 | if (ret) |
3bf040c7 | 827 | goto free_devices; |
de1a21a0 NG |
828 | } |
829 | ||
306b0c95 | 830 | return 0; |
de1a21a0 | 831 | |
3bf040c7 | 832 | free_devices: |
de1a21a0 | 833 | while (dev_id) |
43801f6e NW |
834 | destroy_device(&zram_devices[--dev_id]); |
835 | kfree(zram_devices); | |
de1a21a0 | 836 | unregister: |
f1e3cfff | 837 | unregister_blkdev(zram_major, "zram"); |
de1a21a0 | 838 | out: |
306b0c95 NG |
839 | return ret; |
840 | } | |
841 | ||
f1e3cfff | 842 | static void __exit zram_exit(void) |
306b0c95 NG |
843 | { |
844 | int i; | |
f1e3cfff | 845 | struct zram *zram; |
306b0c95 | 846 | |
efd54f43 | 847 | for (i = 0; i < zram_num_devices; i++) { |
43801f6e | 848 | zram = &zram_devices[i]; |
306b0c95 | 849 | |
f1e3cfff NG |
850 | destroy_device(zram); |
851 | if (zram->init_done) | |
33863c21 | 852 | zram_reset_device(zram); |
306b0c95 NG |
853 | } |
854 | ||
f1e3cfff | 855 | unregister_blkdev(zram_major, "zram"); |
306b0c95 | 856 | |
43801f6e | 857 | kfree(zram_devices); |
306b0c95 NG |
858 | pr_debug("Cleanup done!\n"); |
859 | } | |
860 | ||
efd54f43 NW |
861 | module_param(zram_num_devices, uint, 0); |
862 | MODULE_PARM_DESC(zram_num_devices, "Number of zram devices"); | |
306b0c95 | 863 | |
f1e3cfff NG |
864 | module_init(zram_init); |
865 | module_exit(zram_exit); | |
306b0c95 NG |
866 | |
867 | MODULE_LICENSE("Dual BSD/GPL"); | |
868 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); | |
f1e3cfff | 869 | MODULE_DESCRIPTION("Compressed RAM Block Device"); |