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 | 137 | { |
fd1a30de | 138 | void *handle = zram->table[index].handle; |
306b0c95 | 139 | |
fd1a30de | 140 | if (unlikely(!handle)) { |
2e882281 NG |
141 | /* |
142 | * No memory is allocated for zero filled pages. | |
143 | * Simply clear zero page flag. | |
144 | */ | |
f1e3cfff NG |
145 | if (zram_test_flag(zram, index, ZRAM_ZERO)) { |
146 | zram_clear_flag(zram, index, ZRAM_ZERO); | |
147 | zram_stat_dec(&zram->stats.pages_zero); | |
306b0c95 NG |
148 | } |
149 | return; | |
150 | } | |
151 | ||
f1e3cfff | 152 | if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) { |
fd1a30de | 153 | __free_page(handle); |
f1e3cfff NG |
154 | zram_clear_flag(zram, index, ZRAM_UNCOMPRESSED); |
155 | zram_stat_dec(&zram->stats.pages_expand); | |
306b0c95 NG |
156 | goto out; |
157 | } | |
158 | ||
fd1a30de | 159 | zs_free(zram->mem_pool, handle); |
306b0c95 | 160 | |
fd1a30de | 161 | if (zram->table[index].size <= PAGE_SIZE / 2) |
f1e3cfff | 162 | zram_stat_dec(&zram->stats.good_compress); |
306b0c95 NG |
163 | |
164 | out: | |
fd1a30de NG |
165 | zram_stat64_sub(zram, &zram->stats.compr_size, |
166 | zram->table[index].size); | |
f1e3cfff | 167 | zram_stat_dec(&zram->stats.pages_stored); |
306b0c95 | 168 | |
fd1a30de NG |
169 | zram->table[index].handle = NULL; |
170 | zram->table[index].size = 0; | |
306b0c95 NG |
171 | } |
172 | ||
924bd88d | 173 | static void handle_zero_page(struct bio_vec *bvec) |
306b0c95 | 174 | { |
924bd88d | 175 | struct page *page = bvec->bv_page; |
306b0c95 | 176 | void *user_mem; |
306b0c95 NG |
177 | |
178 | user_mem = kmap_atomic(page, KM_USER0); | |
924bd88d | 179 | memset(user_mem + bvec->bv_offset, 0, bvec->bv_len); |
306b0c95 NG |
180 | kunmap_atomic(user_mem, KM_USER0); |
181 | ||
30fb8a71 | 182 | flush_dcache_page(page); |
306b0c95 NG |
183 | } |
184 | ||
924bd88d JM |
185 | static void handle_uncompressed_page(struct zram *zram, struct bio_vec *bvec, |
186 | u32 index, int offset) | |
306b0c95 | 187 | { |
924bd88d | 188 | struct page *page = bvec->bv_page; |
306b0c95 NG |
189 | unsigned char *user_mem, *cmem; |
190 | ||
306b0c95 | 191 | user_mem = kmap_atomic(page, KM_USER0); |
fd1a30de | 192 | cmem = kmap_atomic(zram->table[index].handle, KM_USER1); |
306b0c95 | 193 | |
924bd88d | 194 | memcpy(user_mem + bvec->bv_offset, cmem + offset, bvec->bv_len); |
306b0c95 | 195 | kunmap_atomic(cmem, KM_USER1); |
dffbb44d | 196 | kunmap_atomic(user_mem, KM_USER0); |
306b0c95 | 197 | |
30fb8a71 | 198 | flush_dcache_page(page); |
306b0c95 NG |
199 | } |
200 | ||
924bd88d JM |
201 | static inline int is_partial_io(struct bio_vec *bvec) |
202 | { | |
203 | return bvec->bv_len != PAGE_SIZE; | |
204 | } | |
205 | ||
8c921b2b | 206 | static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec, |
924bd88d | 207 | u32 index, int offset, struct bio *bio) |
306b0c95 | 208 | { |
8c921b2b JM |
209 | int ret; |
210 | size_t clen; | |
211 | struct page *page; | |
212 | struct zobj_header *zheader; | |
924bd88d | 213 | unsigned char *user_mem, *cmem, *uncmem = NULL; |
a1dd52af | 214 | |
8c921b2b | 215 | page = bvec->bv_page; |
306b0c95 | 216 | |
8c921b2b | 217 | if (zram_test_flag(zram, index, ZRAM_ZERO)) { |
924bd88d | 218 | handle_zero_page(bvec); |
8c921b2b JM |
219 | return 0; |
220 | } | |
306b0c95 | 221 | |
8c921b2b | 222 | /* Requested page is not present in compressed area */ |
fd1a30de | 223 | if (unlikely(!zram->table[index].handle)) { |
8c921b2b JM |
224 | pr_debug("Read before write: sector=%lu, size=%u", |
225 | (ulong)(bio->bi_sector), bio->bi_size); | |
924bd88d | 226 | handle_zero_page(bvec); |
8c921b2b JM |
227 | return 0; |
228 | } | |
306b0c95 | 229 | |
8c921b2b JM |
230 | /* Page is stored uncompressed since it's incompressible */ |
231 | if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) { | |
924bd88d | 232 | handle_uncompressed_page(zram, bvec, index, offset); |
8c921b2b JM |
233 | return 0; |
234 | } | |
306b0c95 | 235 | |
924bd88d JM |
236 | if (is_partial_io(bvec)) { |
237 | /* Use a temporary buffer to decompress the page */ | |
238 | uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
239 | if (!uncmem) { | |
240 | pr_info("Error allocating temp memory!\n"); | |
241 | return -ENOMEM; | |
242 | } | |
243 | } | |
244 | ||
8c921b2b | 245 | user_mem = kmap_atomic(page, KM_USER0); |
924bd88d JM |
246 | if (!is_partial_io(bvec)) |
247 | uncmem = user_mem; | |
8c921b2b | 248 | clen = PAGE_SIZE; |
306b0c95 | 249 | |
fd1a30de | 250 | cmem = zs_map_object(zram->mem_pool, zram->table[index].handle); |
306b0c95 | 251 | |
8c921b2b | 252 | ret = lzo1x_decompress_safe(cmem + sizeof(*zheader), |
fd1a30de | 253 | zram->table[index].size, |
924bd88d JM |
254 | uncmem, &clen); |
255 | ||
256 | if (is_partial_io(bvec)) { | |
257 | memcpy(user_mem + bvec->bv_offset, uncmem + offset, | |
258 | bvec->bv_len); | |
259 | kfree(uncmem); | |
260 | } | |
306b0c95 | 261 | |
fd1a30de | 262 | zs_unmap_object(zram->mem_pool, zram->table[index].handle); |
dffbb44d | 263 | kunmap_atomic(user_mem, KM_USER0); |
a1dd52af | 264 | |
8c921b2b JM |
265 | /* Should NEVER happen. Return bio error if it does. */ |
266 | if (unlikely(ret != LZO_E_OK)) { | |
267 | pr_err("Decompression failed! err=%d, page=%u\n", ret, index); | |
268 | zram_stat64_inc(zram, &zram->stats.failed_reads); | |
269 | return ret; | |
a1dd52af | 270 | } |
306b0c95 | 271 | |
8c921b2b | 272 | flush_dcache_page(page); |
306b0c95 | 273 | |
8c921b2b | 274 | return 0; |
306b0c95 NG |
275 | } |
276 | ||
924bd88d JM |
277 | static int zram_read_before_write(struct zram *zram, char *mem, u32 index) |
278 | { | |
279 | int ret; | |
280 | size_t clen = PAGE_SIZE; | |
281 | struct zobj_header *zheader; | |
282 | unsigned char *cmem; | |
283 | ||
284 | if (zram_test_flag(zram, index, ZRAM_ZERO) || | |
fd1a30de | 285 | !zram->table[index].handle) { |
924bd88d JM |
286 | memset(mem, 0, PAGE_SIZE); |
287 | return 0; | |
288 | } | |
289 | ||
fd1a30de | 290 | cmem = zs_map_object(zram->mem_pool, zram->table[index].handle); |
924bd88d JM |
291 | |
292 | /* Page is stored uncompressed since it's incompressible */ | |
293 | if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) { | |
294 | memcpy(mem, cmem, PAGE_SIZE); | |
295 | kunmap_atomic(cmem, KM_USER0); | |
296 | return 0; | |
297 | } | |
298 | ||
299 | ret = lzo1x_decompress_safe(cmem + sizeof(*zheader), | |
fd1a30de | 300 | zram->table[index].size, |
924bd88d | 301 | mem, &clen); |
fd1a30de | 302 | zs_unmap_object(zram->mem_pool, zram->table[index].handle); |
924bd88d JM |
303 | |
304 | /* Should NEVER happen. Return bio error if it does. */ | |
305 | if (unlikely(ret != LZO_E_OK)) { | |
306 | pr_err("Decompression failed! err=%d, page=%u\n", ret, index); | |
307 | zram_stat64_inc(zram, &zram->stats.failed_reads); | |
308 | return ret; | |
309 | } | |
310 | ||
311 | return 0; | |
312 | } | |
313 | ||
314 | static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index, | |
315 | int offset) | |
306b0c95 | 316 | { |
8c921b2b | 317 | int ret; |
924bd88d | 318 | u32 store_offset; |
8c921b2b | 319 | size_t clen; |
fd1a30de | 320 | void *handle; |
8c921b2b JM |
321 | struct zobj_header *zheader; |
322 | struct page *page, *page_store; | |
924bd88d | 323 | unsigned char *user_mem, *cmem, *src, *uncmem = NULL; |
306b0c95 | 324 | |
8c921b2b JM |
325 | page = bvec->bv_page; |
326 | src = zram->compress_buffer; | |
306b0c95 | 327 | |
924bd88d JM |
328 | if (is_partial_io(bvec)) { |
329 | /* | |
330 | * This is a partial IO. We need to read the full page | |
331 | * before to write the changes. | |
332 | */ | |
333 | uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
334 | if (!uncmem) { | |
335 | pr_info("Error allocating temp memory!\n"); | |
336 | ret = -ENOMEM; | |
337 | goto out; | |
338 | } | |
339 | ret = zram_read_before_write(zram, uncmem, index); | |
340 | if (ret) { | |
341 | kfree(uncmem); | |
342 | goto out; | |
343 | } | |
344 | } | |
345 | ||
8c921b2b JM |
346 | /* |
347 | * System overwrites unused sectors. Free memory associated | |
348 | * with this sector now. | |
349 | */ | |
fd1a30de | 350 | if (zram->table[index].handle || |
8c921b2b JM |
351 | zram_test_flag(zram, index, ZRAM_ZERO)) |
352 | zram_free_page(zram, index); | |
306b0c95 | 353 | |
8c921b2b | 354 | user_mem = kmap_atomic(page, KM_USER0); |
924bd88d JM |
355 | |
356 | if (is_partial_io(bvec)) | |
357 | memcpy(uncmem + offset, user_mem + bvec->bv_offset, | |
358 | bvec->bv_len); | |
359 | else | |
360 | uncmem = user_mem; | |
361 | ||
362 | if (page_zero_filled(uncmem)) { | |
8c921b2b | 363 | kunmap_atomic(user_mem, KM_USER0); |
924bd88d JM |
364 | if (is_partial_io(bvec)) |
365 | kfree(uncmem); | |
8c921b2b JM |
366 | zram_stat_inc(&zram->stats.pages_zero); |
367 | zram_set_flag(zram, index, ZRAM_ZERO); | |
924bd88d JM |
368 | ret = 0; |
369 | goto out; | |
8c921b2b | 370 | } |
306b0c95 | 371 | |
924bd88d | 372 | ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen, |
8c921b2b | 373 | zram->compress_workmem); |
306b0c95 | 374 | |
8c921b2b | 375 | kunmap_atomic(user_mem, KM_USER0); |
924bd88d JM |
376 | if (is_partial_io(bvec)) |
377 | kfree(uncmem); | |
306b0c95 | 378 | |
8c921b2b | 379 | if (unlikely(ret != LZO_E_OK)) { |
8c921b2b | 380 | pr_err("Compression failed! err=%d\n", ret); |
924bd88d | 381 | goto out; |
8c921b2b | 382 | } |
306b0c95 | 383 | |
8c921b2b JM |
384 | /* |
385 | * Page is incompressible. Store it as-is (uncompressed) | |
386 | * since we do not want to return too many disk write | |
387 | * errors which has side effect of hanging the system. | |
388 | */ | |
389 | if (unlikely(clen > max_zpage_size)) { | |
390 | clen = PAGE_SIZE; | |
391 | page_store = alloc_page(GFP_NOIO | __GFP_HIGHMEM); | |
392 | if (unlikely(!page_store)) { | |
8c921b2b JM |
393 | pr_info("Error allocating memory for " |
394 | "incompressible page: %u\n", index); | |
924bd88d JM |
395 | ret = -ENOMEM; |
396 | goto out; | |
397 | } | |
a1dd52af | 398 | |
924bd88d | 399 | store_offset = 0; |
8c921b2b JM |
400 | zram_set_flag(zram, index, ZRAM_UNCOMPRESSED); |
401 | zram_stat_inc(&zram->stats.pages_expand); | |
fd1a30de | 402 | handle = page_store; |
8c921b2b | 403 | src = kmap_atomic(page, KM_USER0); |
fd1a30de | 404 | cmem = kmap_atomic(page_store, KM_USER1); |
8c921b2b JM |
405 | goto memstore; |
406 | } | |
306b0c95 | 407 | |
fd1a30de NG |
408 | handle = zs_malloc(zram->mem_pool, clen + sizeof(*zheader)); |
409 | if (!handle) { | |
8c921b2b JM |
410 | pr_info("Error allocating memory for compressed " |
411 | "page: %u, size=%zu\n", index, clen); | |
924bd88d JM |
412 | ret = -ENOMEM; |
413 | goto out; | |
8c921b2b | 414 | } |
fd1a30de | 415 | cmem = zs_map_object(zram->mem_pool, handle); |
306b0c95 NG |
416 | |
417 | memstore: | |
306b0c95 | 418 | #if 0 |
8c921b2b JM |
419 | /* Back-reference needed for memory defragmentation */ |
420 | if (!zram_test_flag(zram, index, ZRAM_UNCOMPRESSED)) { | |
421 | zheader = (struct zobj_header *)cmem; | |
422 | zheader->table_idx = index; | |
423 | cmem += sizeof(*zheader); | |
424 | } | |
306b0c95 NG |
425 | #endif |
426 | ||
8c921b2b | 427 | memcpy(cmem, src, clen); |
306b0c95 | 428 | |
fd1a30de NG |
429 | if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) { |
430 | kunmap_atomic(cmem, KM_USER1); | |
8c921b2b | 431 | kunmap_atomic(src, KM_USER0); |
fd1a30de NG |
432 | } else { |
433 | zs_unmap_object(zram->mem_pool, handle); | |
434 | } | |
435 | ||
436 | zram->table[index].handle = handle; | |
437 | zram->table[index].size = clen; | |
306b0c95 | 438 | |
8c921b2b JM |
439 | /* Update stats */ |
440 | zram_stat64_add(zram, &zram->stats.compr_size, clen); | |
441 | zram_stat_inc(&zram->stats.pages_stored); | |
442 | if (clen <= PAGE_SIZE / 2) | |
443 | zram_stat_inc(&zram->stats.good_compress); | |
306b0c95 | 444 | |
8c921b2b | 445 | return 0; |
924bd88d JM |
446 | |
447 | out: | |
448 | if (ret) | |
449 | zram_stat64_inc(zram, &zram->stats.failed_writes); | |
450 | return ret; | |
8c921b2b JM |
451 | } |
452 | ||
453 | static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index, | |
924bd88d | 454 | int offset, struct bio *bio, int rw) |
8c921b2b | 455 | { |
c5bde238 | 456 | int ret; |
8c921b2b | 457 | |
c5bde238 JM |
458 | if (rw == READ) { |
459 | down_read(&zram->lock); | |
460 | ret = zram_bvec_read(zram, bvec, index, offset, bio); | |
461 | up_read(&zram->lock); | |
462 | } else { | |
463 | down_write(&zram->lock); | |
464 | ret = zram_bvec_write(zram, bvec, index, offset); | |
465 | up_write(&zram->lock); | |
466 | } | |
467 | ||
468 | return ret; | |
924bd88d JM |
469 | } |
470 | ||
471 | static void update_position(u32 *index, int *offset, struct bio_vec *bvec) | |
472 | { | |
473 | if (*offset + bvec->bv_len >= PAGE_SIZE) | |
474 | (*index)++; | |
475 | *offset = (*offset + bvec->bv_len) % PAGE_SIZE; | |
8c921b2b JM |
476 | } |
477 | ||
478 | static void __zram_make_request(struct zram *zram, struct bio *bio, int rw) | |
479 | { | |
924bd88d | 480 | int i, offset; |
8c921b2b JM |
481 | u32 index; |
482 | struct bio_vec *bvec; | |
483 | ||
484 | switch (rw) { | |
485 | case READ: | |
486 | zram_stat64_inc(zram, &zram->stats.num_reads); | |
487 | break; | |
488 | case WRITE: | |
489 | zram_stat64_inc(zram, &zram->stats.num_writes); | |
490 | break; | |
491 | } | |
492 | ||
493 | index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT; | |
924bd88d | 494 | offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT; |
8c921b2b JM |
495 | |
496 | bio_for_each_segment(bvec, bio, i) { | |
924bd88d JM |
497 | int max_transfer_size = PAGE_SIZE - offset; |
498 | ||
499 | if (bvec->bv_len > max_transfer_size) { | |
500 | /* | |
501 | * zram_bvec_rw() can only make operation on a single | |
502 | * zram page. Split the bio vector. | |
503 | */ | |
504 | struct bio_vec bv; | |
505 | ||
506 | bv.bv_page = bvec->bv_page; | |
507 | bv.bv_len = max_transfer_size; | |
508 | bv.bv_offset = bvec->bv_offset; | |
509 | ||
510 | if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0) | |
511 | goto out; | |
512 | ||
513 | bv.bv_len = bvec->bv_len - max_transfer_size; | |
514 | bv.bv_offset += max_transfer_size; | |
515 | if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0) | |
516 | goto out; | |
517 | } else | |
518 | if (zram_bvec_rw(zram, bvec, index, offset, bio, rw) | |
519 | < 0) | |
520 | goto out; | |
521 | ||
522 | update_position(&index, &offset, bvec); | |
a1dd52af | 523 | } |
306b0c95 NG |
524 | |
525 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
526 | bio_endio(bio, 0); | |
7d7854b4 | 527 | return; |
306b0c95 NG |
528 | |
529 | out: | |
306b0c95 | 530 | bio_io_error(bio); |
306b0c95 NG |
531 | } |
532 | ||
306b0c95 | 533 | /* |
924bd88d | 534 | * Check if request is within bounds and aligned on zram logical blocks. |
306b0c95 | 535 | */ |
f1e3cfff | 536 | static inline int valid_io_request(struct zram *zram, struct bio *bio) |
306b0c95 NG |
537 | { |
538 | if (unlikely( | |
f1e3cfff | 539 | (bio->bi_sector >= (zram->disksize >> SECTOR_SHIFT)) || |
924bd88d JM |
540 | (bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)) || |
541 | (bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))) { | |
306b0c95 NG |
542 | |
543 | return 0; | |
544 | } | |
545 | ||
a1dd52af | 546 | /* I/O request is valid */ |
306b0c95 NG |
547 | return 1; |
548 | } | |
549 | ||
550 | /* | |
f1e3cfff | 551 | * Handler function for all zram I/O requests. |
306b0c95 | 552 | */ |
5a7bbad2 | 553 | static void zram_make_request(struct request_queue *queue, struct bio *bio) |
306b0c95 | 554 | { |
f1e3cfff | 555 | struct zram *zram = queue->queuedata; |
306b0c95 | 556 | |
0900beae JM |
557 | if (unlikely(!zram->init_done) && zram_init_device(zram)) |
558 | goto error; | |
559 | ||
560 | down_read(&zram->init_lock); | |
561 | if (unlikely(!zram->init_done)) | |
562 | goto error_unlock; | |
563 | ||
f1e3cfff NG |
564 | if (!valid_io_request(zram, bio)) { |
565 | zram_stat64_inc(zram, &zram->stats.invalid_io); | |
0900beae | 566 | goto error_unlock; |
6642a67c JM |
567 | } |
568 | ||
8c921b2b | 569 | __zram_make_request(zram, bio, bio_data_dir(bio)); |
0900beae | 570 | up_read(&zram->init_lock); |
306b0c95 | 571 | |
b4fdcb02 | 572 | return; |
0900beae JM |
573 | |
574 | error_unlock: | |
575 | up_read(&zram->init_lock); | |
576 | error: | |
577 | bio_io_error(bio); | |
306b0c95 NG |
578 | } |
579 | ||
0900beae | 580 | void __zram_reset_device(struct zram *zram) |
306b0c95 | 581 | { |
97a06382 | 582 | size_t index; |
306b0c95 | 583 | |
f1e3cfff | 584 | zram->init_done = 0; |
7eef7533 | 585 | |
306b0c95 | 586 | /* Free various per-device buffers */ |
f1e3cfff NG |
587 | kfree(zram->compress_workmem); |
588 | free_pages((unsigned long)zram->compress_buffer, 1); | |
306b0c95 | 589 | |
f1e3cfff NG |
590 | zram->compress_workmem = NULL; |
591 | zram->compress_buffer = NULL; | |
306b0c95 | 592 | |
f1e3cfff NG |
593 | /* Free all pages that are still in this zram device */ |
594 | for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) { | |
fd1a30de NG |
595 | void *handle = zram->table[index].handle; |
596 | if (!handle) | |
306b0c95 NG |
597 | continue; |
598 | ||
f1e3cfff | 599 | if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) |
fd1a30de | 600 | __free_page(handle); |
306b0c95 | 601 | else |
fd1a30de | 602 | zs_free(zram->mem_pool, handle); |
306b0c95 NG |
603 | } |
604 | ||
f1e3cfff NG |
605 | vfree(zram->table); |
606 | zram->table = NULL; | |
306b0c95 | 607 | |
fd1a30de | 608 | zs_destroy_pool(zram->mem_pool); |
f1e3cfff | 609 | zram->mem_pool = NULL; |
306b0c95 | 610 | |
306b0c95 | 611 | /* Reset stats */ |
f1e3cfff | 612 | memset(&zram->stats, 0, sizeof(zram->stats)); |
306b0c95 | 613 | |
f1e3cfff | 614 | zram->disksize = 0; |
0900beae JM |
615 | } |
616 | ||
617 | void zram_reset_device(struct zram *zram) | |
618 | { | |
619 | down_write(&zram->init_lock); | |
620 | __zram_reset_device(zram); | |
621 | up_write(&zram->init_lock); | |
306b0c95 NG |
622 | } |
623 | ||
33863c21 | 624 | int zram_init_device(struct zram *zram) |
306b0c95 NG |
625 | { |
626 | int ret; | |
627 | size_t num_pages; | |
306b0c95 | 628 | |
0900beae | 629 | down_write(&zram->init_lock); |
484875ad | 630 | |
f1e3cfff | 631 | if (zram->init_done) { |
0900beae | 632 | up_write(&zram->init_lock); |
484875ad | 633 | return 0; |
306b0c95 NG |
634 | } |
635 | ||
f1e3cfff | 636 | zram_set_disksize(zram, totalram_pages << PAGE_SHIFT); |
306b0c95 | 637 | |
f1e3cfff NG |
638 | zram->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL); |
639 | if (!zram->compress_workmem) { | |
306b0c95 NG |
640 | pr_err("Error allocating compressor working memory!\n"); |
641 | ret = -ENOMEM; | |
5a18c531 | 642 | goto fail_no_table; |
306b0c95 NG |
643 | } |
644 | ||
fb927284 JM |
645 | zram->compress_buffer = |
646 | (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1); | |
f1e3cfff | 647 | if (!zram->compress_buffer) { |
306b0c95 NG |
648 | pr_err("Error allocating compressor buffer space\n"); |
649 | ret = -ENOMEM; | |
5a18c531 | 650 | goto fail_no_table; |
306b0c95 NG |
651 | } |
652 | ||
f1e3cfff | 653 | num_pages = zram->disksize >> PAGE_SHIFT; |
5b84cc78 | 654 | zram->table = vzalloc(num_pages * sizeof(*zram->table)); |
f1e3cfff NG |
655 | if (!zram->table) { |
656 | pr_err("Error allocating zram address table\n"); | |
306b0c95 | 657 | ret = -ENOMEM; |
5a18c531 | 658 | goto fail_no_table; |
306b0c95 | 659 | } |
306b0c95 | 660 | |
f1e3cfff | 661 | set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT); |
306b0c95 | 662 | |
f1e3cfff NG |
663 | /* zram devices sort of resembles non-rotational disks */ |
664 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue); | |
306b0c95 | 665 | |
fd1a30de | 666 | zram->mem_pool = zs_create_pool("zram", GFP_NOIO | __GFP_HIGHMEM); |
f1e3cfff | 667 | if (!zram->mem_pool) { |
306b0c95 NG |
668 | pr_err("Error creating memory pool\n"); |
669 | ret = -ENOMEM; | |
670 | goto fail; | |
671 | } | |
672 | ||
f1e3cfff | 673 | zram->init_done = 1; |
0900beae | 674 | up_write(&zram->init_lock); |
306b0c95 NG |
675 | |
676 | pr_debug("Initialization done!\n"); | |
677 | return 0; | |
678 | ||
5a18c531 JM |
679 | fail_no_table: |
680 | /* To prevent accessing table entries during cleanup */ | |
681 | zram->disksize = 0; | |
306b0c95 | 682 | fail: |
0900beae JM |
683 | __zram_reset_device(zram); |
684 | up_write(&zram->init_lock); | |
306b0c95 NG |
685 | pr_err("Initialization failed: err=%d\n", ret); |
686 | return ret; | |
687 | } | |
688 | ||
2ccbec05 NG |
689 | static void zram_slot_free_notify(struct block_device *bdev, |
690 | unsigned long index) | |
107c161b | 691 | { |
f1e3cfff | 692 | struct zram *zram; |
107c161b | 693 | |
f1e3cfff NG |
694 | zram = bdev->bd_disk->private_data; |
695 | zram_free_page(zram, index); | |
696 | zram_stat64_inc(zram, &zram->stats.notify_free); | |
107c161b NG |
697 | } |
698 | ||
f1e3cfff | 699 | static const struct block_device_operations zram_devops = { |
f1e3cfff | 700 | .swap_slot_free_notify = zram_slot_free_notify, |
107c161b | 701 | .owner = THIS_MODULE |
306b0c95 NG |
702 | }; |
703 | ||
f1e3cfff | 704 | static int create_device(struct zram *zram, int device_id) |
306b0c95 | 705 | { |
de1a21a0 NG |
706 | int ret = 0; |
707 | ||
c5bde238 | 708 | init_rwsem(&zram->lock); |
0900beae | 709 | init_rwsem(&zram->init_lock); |
f1e3cfff | 710 | spin_lock_init(&zram->stat64_lock); |
306b0c95 | 711 | |
f1e3cfff NG |
712 | zram->queue = blk_alloc_queue(GFP_KERNEL); |
713 | if (!zram->queue) { | |
306b0c95 NG |
714 | pr_err("Error allocating disk queue for device %d\n", |
715 | device_id); | |
de1a21a0 NG |
716 | ret = -ENOMEM; |
717 | goto out; | |
306b0c95 NG |
718 | } |
719 | ||
f1e3cfff NG |
720 | blk_queue_make_request(zram->queue, zram_make_request); |
721 | zram->queue->queuedata = zram; | |
306b0c95 NG |
722 | |
723 | /* gendisk structure */ | |
f1e3cfff NG |
724 | zram->disk = alloc_disk(1); |
725 | if (!zram->disk) { | |
726 | blk_cleanup_queue(zram->queue); | |
306b0c95 NG |
727 | pr_warning("Error allocating disk structure for device %d\n", |
728 | device_id); | |
de1a21a0 NG |
729 | ret = -ENOMEM; |
730 | goto out; | |
306b0c95 NG |
731 | } |
732 | ||
f1e3cfff NG |
733 | zram->disk->major = zram_major; |
734 | zram->disk->first_minor = device_id; | |
735 | zram->disk->fops = &zram_devops; | |
736 | zram->disk->queue = zram->queue; | |
737 | zram->disk->private_data = zram; | |
738 | snprintf(zram->disk->disk_name, 16, "zram%d", device_id); | |
306b0c95 | 739 | |
33863c21 | 740 | /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */ |
f1e3cfff | 741 | set_capacity(zram->disk, 0); |
5d83d5a0 | 742 | |
a1dd52af NG |
743 | /* |
744 | * To ensure that we always get PAGE_SIZE aligned | |
745 | * and n*PAGE_SIZED sized I/O requests. | |
746 | */ | |
f1e3cfff | 747 | blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE); |
7b19b8d4 RJ |
748 | blk_queue_logical_block_size(zram->disk->queue, |
749 | ZRAM_LOGICAL_BLOCK_SIZE); | |
f1e3cfff NG |
750 | blk_queue_io_min(zram->disk->queue, PAGE_SIZE); |
751 | blk_queue_io_opt(zram->disk->queue, PAGE_SIZE); | |
5d83d5a0 | 752 | |
f1e3cfff | 753 | add_disk(zram->disk); |
306b0c95 | 754 | |
33863c21 NG |
755 | ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj, |
756 | &zram_disk_attr_group); | |
757 | if (ret < 0) { | |
758 | pr_warning("Error creating sysfs group"); | |
759 | goto out; | |
760 | } | |
33863c21 | 761 | |
f1e3cfff | 762 | zram->init_done = 0; |
de1a21a0 NG |
763 | |
764 | out: | |
765 | return ret; | |
306b0c95 NG |
766 | } |
767 | ||
f1e3cfff | 768 | static void destroy_device(struct zram *zram) |
306b0c95 | 769 | { |
33863c21 NG |
770 | sysfs_remove_group(&disk_to_dev(zram->disk)->kobj, |
771 | &zram_disk_attr_group); | |
33863c21 | 772 | |
f1e3cfff NG |
773 | if (zram->disk) { |
774 | del_gendisk(zram->disk); | |
775 | put_disk(zram->disk); | |
306b0c95 NG |
776 | } |
777 | ||
f1e3cfff NG |
778 | if (zram->queue) |
779 | blk_cleanup_queue(zram->queue); | |
306b0c95 NG |
780 | } |
781 | ||
f1e3cfff | 782 | static int __init zram_init(void) |
306b0c95 | 783 | { |
de1a21a0 | 784 | int ret, dev_id; |
306b0c95 | 785 | |
efd54f43 | 786 | if (zram_num_devices > max_num_devices) { |
306b0c95 | 787 | pr_warning("Invalid value for num_devices: %u\n", |
efd54f43 | 788 | zram_num_devices); |
de1a21a0 NG |
789 | ret = -EINVAL; |
790 | goto out; | |
306b0c95 NG |
791 | } |
792 | ||
f1e3cfff NG |
793 | zram_major = register_blkdev(0, "zram"); |
794 | if (zram_major <= 0) { | |
306b0c95 | 795 | pr_warning("Unable to get major number\n"); |
de1a21a0 NG |
796 | ret = -EBUSY; |
797 | goto out; | |
306b0c95 NG |
798 | } |
799 | ||
efd54f43 | 800 | if (!zram_num_devices) { |
306b0c95 | 801 | pr_info("num_devices not specified. Using default: 1\n"); |
efd54f43 | 802 | zram_num_devices = 1; |
306b0c95 NG |
803 | } |
804 | ||
805 | /* Allocate the device array and initialize each one */ | |
efd54f43 NW |
806 | pr_info("Creating %u devices ...\n", zram_num_devices); |
807 | zram_devices = kzalloc(zram_num_devices * sizeof(struct zram), GFP_KERNEL); | |
43801f6e | 808 | if (!zram_devices) { |
de1a21a0 NG |
809 | ret = -ENOMEM; |
810 | goto unregister; | |
811 | } | |
306b0c95 | 812 | |
efd54f43 | 813 | for (dev_id = 0; dev_id < zram_num_devices; dev_id++) { |
43801f6e | 814 | ret = create_device(&zram_devices[dev_id], dev_id); |
de1a21a0 | 815 | if (ret) |
3bf040c7 | 816 | goto free_devices; |
de1a21a0 NG |
817 | } |
818 | ||
306b0c95 | 819 | return 0; |
de1a21a0 | 820 | |
3bf040c7 | 821 | free_devices: |
de1a21a0 | 822 | while (dev_id) |
43801f6e NW |
823 | destroy_device(&zram_devices[--dev_id]); |
824 | kfree(zram_devices); | |
de1a21a0 | 825 | unregister: |
f1e3cfff | 826 | unregister_blkdev(zram_major, "zram"); |
de1a21a0 | 827 | out: |
306b0c95 NG |
828 | return ret; |
829 | } | |
830 | ||
f1e3cfff | 831 | static void __exit zram_exit(void) |
306b0c95 NG |
832 | { |
833 | int i; | |
f1e3cfff | 834 | struct zram *zram; |
306b0c95 | 835 | |
efd54f43 | 836 | for (i = 0; i < zram_num_devices; i++) { |
43801f6e | 837 | zram = &zram_devices[i]; |
306b0c95 | 838 | |
f1e3cfff NG |
839 | destroy_device(zram); |
840 | if (zram->init_done) | |
33863c21 | 841 | zram_reset_device(zram); |
306b0c95 NG |
842 | } |
843 | ||
f1e3cfff | 844 | unregister_blkdev(zram_major, "zram"); |
306b0c95 | 845 | |
43801f6e | 846 | kfree(zram_devices); |
306b0c95 NG |
847 | pr_debug("Cleanup done!\n"); |
848 | } | |
849 | ||
efd54f43 NW |
850 | module_param(zram_num_devices, uint, 0); |
851 | MODULE_PARM_DESC(zram_num_devices, "Number of zram devices"); | |
306b0c95 | 852 | |
f1e3cfff NG |
853 | module_init(zram_init); |
854 | module_exit(zram_exit); | |
306b0c95 NG |
855 | |
856 | MODULE_LICENSE("Dual BSD/GPL"); | |
857 | MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>"); | |
f1e3cfff | 858 | MODULE_DESCRIPTION("Compressed RAM Block Device"); |