1 // SPDX-License-Identifier: GPL-2.0
2 /* bounce buffer handling for block devices
3 *
4 * - Split from highmem.c
5 */
6
7 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9 #include <linux/mm.h>
10 #include <linux/export.h>
11 #include <linux/swap.h>
12 #include <linux/gfp.h>
13 #include <linux/bio.h>
14 #include <linux/pagemap.h>
15 #include <linux/mempool.h>
16 #include <linux/blkdev.h>
17 #include <linux/backing-dev.h>
18 #include <linux/init.h>
19 #include <linux/hash.h>
20 #include <linux/highmem.h>
21 #include <linux/memblock.h>
22 #include <linux/printk.h>
23 #include <asm/tlbflush.h>
24
25 #include <trace/events/block.h>
26 #include "blk.h"
27
28 #define POOL_SIZE 64
29 #define ISA_POOL_SIZE 16
30
31 static struct bio_set bounce_bio_set, bounce_bio_split;
32 static mempool_t page_pool, isa_page_pool;
33
init_bounce_bioset(void)34 static void init_bounce_bioset(void)
35 {
36 static bool bounce_bs_setup;
37 int ret;
38
39 if (bounce_bs_setup)
40 return;
41
42 ret = bioset_init(&bounce_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
43 BUG_ON(ret);
44 if (bioset_integrity_create(&bounce_bio_set, BIO_POOL_SIZE))
45 BUG_ON(1);
46
47 ret = bioset_init(&bounce_bio_split, BIO_POOL_SIZE, 0, 0);
48 BUG_ON(ret);
49 bounce_bs_setup = true;
50 }
51
52 #if defined(CONFIG_HIGHMEM)
init_emergency_pool(void)53 static __init int init_emergency_pool(void)
54 {
55 int ret;
56 #if defined(CONFIG_HIGHMEM) && !defined(CONFIG_MEMORY_HOTPLUG)
57 if (max_pfn <= max_low_pfn)
58 return 0;
59 #endif
60
61 ret = mempool_init_page_pool(&page_pool, POOL_SIZE, 0);
62 BUG_ON(ret);
63 pr_info("pool size: %d pages\n", POOL_SIZE);
64
65 init_bounce_bioset();
66 return 0;
67 }
68
69 __initcall(init_emergency_pool);
70 #endif
71
72 #ifdef CONFIG_HIGHMEM
73 /*
74 * highmem version, map in to vec
75 */
bounce_copy_vec(struct bio_vec * to,unsigned char * vfrom)76 static void bounce_copy_vec(struct bio_vec *to, unsigned char *vfrom)
77 {
78 unsigned char *vto;
79
80 vto = kmap_atomic(to->bv_page);
81 memcpy(vto + to->bv_offset, vfrom, to->bv_len);
82 kunmap_atomic(vto);
83 }
84
85 #else /* CONFIG_HIGHMEM */
86
87 #define bounce_copy_vec(to, vfrom) \
88 memcpy(page_address((to)->bv_page) + (to)->bv_offset, vfrom, (to)->bv_len)
89
90 #endif /* CONFIG_HIGHMEM */
91
92 /*
93 * allocate pages in the DMA region for the ISA pool
94 */
mempool_alloc_pages_isa(gfp_t gfp_mask,void * data)95 static void *mempool_alloc_pages_isa(gfp_t gfp_mask, void *data)
96 {
97 return mempool_alloc_pages(gfp_mask | GFP_DMA, data);
98 }
99
100 static DEFINE_MUTEX(isa_mutex);
101
102 /*
103 * gets called "every" time someone init's a queue with BLK_BOUNCE_ISA
104 * as the max address, so check if the pool has already been created.
105 */
init_emergency_isa_pool(void)106 int init_emergency_isa_pool(void)
107 {
108 int ret;
109
110 mutex_lock(&isa_mutex);
111
112 if (mempool_initialized(&isa_page_pool)) {
113 mutex_unlock(&isa_mutex);
114 return 0;
115 }
116
117 ret = mempool_init(&isa_page_pool, ISA_POOL_SIZE, mempool_alloc_pages_isa,
118 mempool_free_pages, (void *) 0);
119 BUG_ON(ret);
120
121 pr_info("isa pool size: %d pages\n", ISA_POOL_SIZE);
122 init_bounce_bioset();
123 mutex_unlock(&isa_mutex);
124 return 0;
125 }
126
127 /*
128 * Simple bounce buffer support for highmem pages. Depending on the
129 * queue gfp mask set, *to may or may not be a highmem page. kmap it
130 * always, it will do the Right Thing
131 */
copy_to_high_bio_irq(struct bio * to,struct bio * from)132 static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
133 {
134 unsigned char *vfrom;
135 struct bio_vec tovec, fromvec;
136 struct bvec_iter iter;
137 /*
138 * The bio of @from is created by bounce, so we can iterate
139 * its bvec from start to end, but the @from->bi_iter can't be
140 * trusted because it might be changed by splitting.
141 */
142 struct bvec_iter from_iter = BVEC_ITER_ALL_INIT;
143
144 bio_for_each_segment(tovec, to, iter) {
145 fromvec = bio_iter_iovec(from, from_iter);
146 if (tovec.bv_page != fromvec.bv_page) {
147 /*
148 * fromvec->bv_offset and fromvec->bv_len might have
149 * been modified by the block layer, so use the original
150 * copy, bounce_copy_vec already uses tovec->bv_len
151 */
152 vfrom = page_address(fromvec.bv_page) +
153 tovec.bv_offset;
154
155 bounce_copy_vec(&tovec, vfrom);
156 flush_dcache_page(tovec.bv_page);
157 }
158 bio_advance_iter(from, &from_iter, tovec.bv_len);
159 }
160 }
161
bounce_end_io(struct bio * bio,mempool_t * pool)162 static void bounce_end_io(struct bio *bio, mempool_t *pool)
163 {
164 struct bio *bio_orig = bio->bi_private;
165 struct bio_vec *bvec, orig_vec;
166 struct bvec_iter orig_iter = bio_orig->bi_iter;
167 struct bvec_iter_all iter_all;
168
169 /*
170 * free up bounce indirect pages used
171 */
172 bio_for_each_segment_all(bvec, bio, iter_all) {
173 orig_vec = bio_iter_iovec(bio_orig, orig_iter);
174 if (bvec->bv_page != orig_vec.bv_page) {
175 dec_zone_page_state(bvec->bv_page, NR_BOUNCE);
176 mempool_free(bvec->bv_page, pool);
177 }
178 bio_advance_iter(bio_orig, &orig_iter, orig_vec.bv_len);
179 }
180
181 bio_orig->bi_status = bio->bi_status;
182 bio_endio(bio_orig);
183 bio_put(bio);
184 }
185
bounce_end_io_write(struct bio * bio)186 static void bounce_end_io_write(struct bio *bio)
187 {
188 bounce_end_io(bio, &page_pool);
189 }
190
bounce_end_io_write_isa(struct bio * bio)191 static void bounce_end_io_write_isa(struct bio *bio)
192 {
193
194 bounce_end_io(bio, &isa_page_pool);
195 }
196
__bounce_end_io_read(struct bio * bio,mempool_t * pool)197 static void __bounce_end_io_read(struct bio *bio, mempool_t *pool)
198 {
199 struct bio *bio_orig = bio->bi_private;
200
201 if (!bio->bi_status)
202 copy_to_high_bio_irq(bio_orig, bio);
203
204 bounce_end_io(bio, pool);
205 }
206
bounce_end_io_read(struct bio * bio)207 static void bounce_end_io_read(struct bio *bio)
208 {
209 __bounce_end_io_read(bio, &page_pool);
210 }
211
bounce_end_io_read_isa(struct bio * bio)212 static void bounce_end_io_read_isa(struct bio *bio)
213 {
214 __bounce_end_io_read(bio, &isa_page_pool);
215 }
216
bounce_clone_bio(struct bio * bio_src,gfp_t gfp_mask,struct bio_set * bs)217 static struct bio *bounce_clone_bio(struct bio *bio_src, gfp_t gfp_mask,
218 struct bio_set *bs)
219 {
220 struct bvec_iter iter;
221 struct bio_vec bv;
222 struct bio *bio;
223
224 /*
225 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
226 * bio_src->bi_io_vec to bio->bi_io_vec.
227 *
228 * We can't do that anymore, because:
229 *
230 * - The point of cloning the biovec is to produce a bio with a biovec
231 * the caller can modify: bi_idx and bi_bvec_done should be 0.
232 *
233 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
234 * we tried to clone the whole thing bio_alloc_bioset() would fail.
235 * But the clone should succeed as long as the number of biovecs we
236 * actually need to allocate is fewer than BIO_MAX_PAGES.
237 *
238 * - Lastly, bi_vcnt should not be looked at or relied upon by code
239 * that does not own the bio - reason being drivers don't use it for
240 * iterating over the biovec anymore, so expecting it to be kept up
241 * to date (i.e. for clones that share the parent biovec) is just
242 * asking for trouble and would force extra work on
243 * __bio_clone_fast() anyways.
244 */
245
246 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
247 if (!bio)
248 return NULL;
249 bio->bi_disk = bio_src->bi_disk;
250 bio->bi_opf = bio_src->bi_opf;
251 bio->bi_ioprio = bio_src->bi_ioprio;
252 bio->bi_write_hint = bio_src->bi_write_hint;
253 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
254 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
255
256 switch (bio_op(bio)) {
257 case REQ_OP_DISCARD:
258 case REQ_OP_SECURE_ERASE:
259 case REQ_OP_WRITE_ZEROES:
260 break;
261 case REQ_OP_WRITE_SAME:
262 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
263 break;
264 default:
265 bio_for_each_segment(bv, bio_src, iter)
266 bio->bi_io_vec[bio->bi_vcnt++] = bv;
267 break;
268 }
269
270 if (bio_crypt_clone(bio, bio_src, gfp_mask) < 0)
271 goto err_put;
272
273 if (bio_integrity(bio_src) &&
274 bio_integrity_clone(bio, bio_src, gfp_mask) < 0)
275 goto err_put;
276
277 bio_clone_blkg_association(bio, bio_src);
278 blkcg_bio_issue_init(bio);
279
280 return bio;
281
282 err_put:
283 bio_put(bio);
284 return NULL;
285 }
286
__blk_queue_bounce(struct request_queue * q,struct bio ** bio_orig,mempool_t * pool)287 static void __blk_queue_bounce(struct request_queue *q, struct bio **bio_orig,
288 mempool_t *pool)
289 {
290 struct bio *bio;
291 int rw = bio_data_dir(*bio_orig);
292 struct bio_vec *to, from;
293 struct bvec_iter iter;
294 unsigned i = 0;
295 bool bounce = false;
296 int sectors = 0;
297 bool passthrough = bio_is_passthrough(*bio_orig);
298
299 bio_for_each_segment(from, *bio_orig, iter) {
300 if (i++ < BIO_MAX_PAGES)
301 sectors += from.bv_len >> 9;
302 if (page_to_pfn(from.bv_page) > q->limits.bounce_pfn)
303 bounce = true;
304 }
305 if (!bounce)
306 return;
307
308 if (!passthrough && sectors < bio_sectors(*bio_orig)) {
309 bio = bio_split(*bio_orig, sectors, GFP_NOIO, &bounce_bio_split);
310 bio_chain(bio, *bio_orig);
311 submit_bio_noacct(*bio_orig);
312 *bio_orig = bio;
313 }
314 bio = bounce_clone_bio(*bio_orig, GFP_NOIO, passthrough ? NULL :
315 &bounce_bio_set);
316
317 /*
318 * Bvec table can't be updated by bio_for_each_segment_all(),
319 * so retrieve bvec from the table directly. This way is safe
320 * because the 'bio' is single-page bvec.
321 */
322 for (i = 0, to = bio->bi_io_vec; i < bio->bi_vcnt; to++, i++) {
323 struct page *page = to->bv_page;
324
325 if (page_to_pfn(page) <= q->limits.bounce_pfn)
326 continue;
327
328 to->bv_page = mempool_alloc(pool, q->bounce_gfp);
329 inc_zone_page_state(to->bv_page, NR_BOUNCE);
330
331 if (rw == WRITE) {
332 char *vto, *vfrom;
333
334 flush_dcache_page(page);
335
336 vto = page_address(to->bv_page) + to->bv_offset;
337 vfrom = kmap_atomic(page) + to->bv_offset;
338 memcpy(vto, vfrom, to->bv_len);
339 kunmap_atomic(vfrom);
340 }
341 }
342
343 trace_block_bio_bounce(q, *bio_orig);
344
345 bio->bi_flags |= (1 << BIO_BOUNCED);
346
347 if (pool == &page_pool) {
348 bio->bi_end_io = bounce_end_io_write;
349 if (rw == READ)
350 bio->bi_end_io = bounce_end_io_read;
351 } else {
352 bio->bi_end_io = bounce_end_io_write_isa;
353 if (rw == READ)
354 bio->bi_end_io = bounce_end_io_read_isa;
355 }
356
357 bio->bi_private = *bio_orig;
358 *bio_orig = bio;
359 }
360
blk_queue_bounce(struct request_queue * q,struct bio ** bio_orig)361 void blk_queue_bounce(struct request_queue *q, struct bio **bio_orig)
362 {
363 mempool_t *pool;
364
365 /*
366 * Data-less bio, nothing to bounce
367 */
368 if (!bio_has_data(*bio_orig))
369 return;
370
371 /*
372 * for non-isa bounce case, just check if the bounce pfn is equal
373 * to or bigger than the highest pfn in the system -- in that case,
374 * don't waste time iterating over bio segments
375 */
376 if (!(q->bounce_gfp & GFP_DMA)) {
377 if (q->limits.bounce_pfn >= blk_max_pfn)
378 return;
379 pool = &page_pool;
380 } else {
381 BUG_ON(!mempool_initialized(&isa_page_pool));
382 pool = &isa_page_pool;
383 }
384
385 /*
386 * slow path
387 */
388 __blk_queue_bounce(q, bio_orig, pool);
389 }
390