Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
86db1e29 JA |
2 | /* |
3 | * Functions related to mapping data to requests | |
4 | */ | |
5 | #include <linux/kernel.h> | |
68db0cf1 | 6 | #include <linux/sched/task_stack.h> |
86db1e29 JA |
7 | #include <linux/module.h> |
8 | #include <linux/bio.h> | |
9 | #include <linux/blkdev.h> | |
26e49cfc | 10 | #include <linux/uio.h> |
86db1e29 JA |
11 | |
12 | #include "blk.h" | |
13 | ||
130879f1 | 14 | struct bio_map_data { |
f3256075 CH |
15 | bool is_our_pages : 1; |
16 | bool is_null_mapped : 1; | |
130879f1 CH |
17 | struct iov_iter iter; |
18 | struct iovec iov[]; | |
19 | }; | |
20 | ||
21 | static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data, | |
22 | gfp_t gfp_mask) | |
23 | { | |
24 | struct bio_map_data *bmd; | |
25 | ||
26 | if (data->nr_segs > UIO_MAXIOV) | |
27 | return NULL; | |
28 | ||
29 | bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask); | |
30 | if (!bmd) | |
31 | return NULL; | |
130879f1 | 32 | bmd->iter = *data; |
0a2481cd | 33 | if (iter_is_iovec(data)) { |
de4f5fed JA |
34 | memcpy(bmd->iov, iter_iov(data), sizeof(struct iovec) * data->nr_segs); |
35 | bmd->iter.__iov = bmd->iov; | |
0a2481cd | 36 | } |
130879f1 CH |
37 | return bmd; |
38 | } | |
39 | ||
40 | /** | |
41 | * bio_copy_from_iter - copy all pages from iov_iter to bio | |
42 | * @bio: The &struct bio which describes the I/O as destination | |
43 | * @iter: iov_iter as source | |
44 | * | |
45 | * Copy all pages from iov_iter to bio. | |
46 | * Returns 0 on success, or error on failure. | |
47 | */ | |
48 | static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter) | |
49 | { | |
50 | struct bio_vec *bvec; | |
51 | struct bvec_iter_all iter_all; | |
52 | ||
53 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
54 | ssize_t ret; | |
55 | ||
56 | ret = copy_page_from_iter(bvec->bv_page, | |
57 | bvec->bv_offset, | |
58 | bvec->bv_len, | |
59 | iter); | |
60 | ||
61 | if (!iov_iter_count(iter)) | |
62 | break; | |
63 | ||
64 | if (ret < bvec->bv_len) | |
65 | return -EFAULT; | |
66 | } | |
67 | ||
68 | return 0; | |
69 | } | |
70 | ||
71 | /** | |
72 | * bio_copy_to_iter - copy all pages from bio to iov_iter | |
73 | * @bio: The &struct bio which describes the I/O as source | |
74 | * @iter: iov_iter as destination | |
75 | * | |
76 | * Copy all pages from bio to iov_iter. | |
77 | * Returns 0 on success, or error on failure. | |
78 | */ | |
79 | static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter) | |
80 | { | |
81 | struct bio_vec *bvec; | |
82 | struct bvec_iter_all iter_all; | |
83 | ||
84 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
85 | ssize_t ret; | |
86 | ||
87 | ret = copy_page_to_iter(bvec->bv_page, | |
88 | bvec->bv_offset, | |
89 | bvec->bv_len, | |
90 | &iter); | |
91 | ||
92 | if (!iov_iter_count(&iter)) | |
93 | break; | |
94 | ||
95 | if (ret < bvec->bv_len) | |
96 | return -EFAULT; | |
97 | } | |
98 | ||
99 | return 0; | |
100 | } | |
101 | ||
102 | /** | |
103 | * bio_uncopy_user - finish previously mapped bio | |
104 | * @bio: bio being terminated | |
105 | * | |
106 | * Free pages allocated from bio_copy_user_iov() and write back data | |
107 | * to user space in case of a read. | |
108 | */ | |
109 | static int bio_uncopy_user(struct bio *bio) | |
110 | { | |
111 | struct bio_map_data *bmd = bio->bi_private; | |
112 | int ret = 0; | |
113 | ||
3310eeba | 114 | if (!bmd->is_null_mapped) { |
130879f1 CH |
115 | /* |
116 | * if we're in a workqueue, the request is orphaned, so | |
117 | * don't copy into a random user address space, just free | |
118 | * and return -EINTR so user space doesn't expect any data. | |
119 | */ | |
120 | if (!current->mm) | |
121 | ret = -EINTR; | |
122 | else if (bio_data_dir(bio) == READ) | |
123 | ret = bio_copy_to_iter(bio, bmd->iter); | |
124 | if (bmd->is_our_pages) | |
125 | bio_free_pages(bio); | |
126 | } | |
127 | kfree(bmd); | |
130879f1 CH |
128 | return ret; |
129 | } | |
130 | ||
7589ad67 CH |
131 | static int bio_copy_user_iov(struct request *rq, struct rq_map_data *map_data, |
132 | struct iov_iter *iter, gfp_t gfp_mask) | |
130879f1 CH |
133 | { |
134 | struct bio_map_data *bmd; | |
135 | struct page *page; | |
393bb12e | 136 | struct bio *bio; |
130879f1 CH |
137 | int i = 0, ret; |
138 | int nr_pages; | |
139 | unsigned int len = iter->count; | |
140 | unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0; | |
141 | ||
142 | bmd = bio_alloc_map_data(iter, gfp_mask); | |
143 | if (!bmd) | |
7589ad67 | 144 | return -ENOMEM; |
130879f1 CH |
145 | |
146 | /* | |
147 | * We need to do a deep copy of the iov_iter including the iovecs. | |
148 | * The caller provided iov might point to an on-stack or otherwise | |
149 | * shortlived one. | |
150 | */ | |
f3256075 | 151 | bmd->is_our_pages = !map_data; |
03859717 | 152 | bmd->is_null_mapped = (map_data && map_data->null_mapped); |
130879f1 | 153 | |
5f7136db | 154 | nr_pages = bio_max_segs(DIV_ROUND_UP(offset + len, PAGE_SIZE)); |
130879f1 CH |
155 | |
156 | ret = -ENOMEM; | |
066ff571 | 157 | bio = bio_kmalloc(nr_pages, gfp_mask); |
130879f1 CH |
158 | if (!bio) |
159 | goto out_bmd; | |
066ff571 | 160 | bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, req_op(rq)); |
130879f1 CH |
161 | |
162 | if (map_data) { | |
f5d632d1 | 163 | nr_pages = 1U << map_data->page_order; |
130879f1 CH |
164 | i = map_data->offset / PAGE_SIZE; |
165 | } | |
166 | while (len) { | |
167 | unsigned int bytes = PAGE_SIZE; | |
168 | ||
169 | bytes -= offset; | |
170 | ||
171 | if (bytes > len) | |
172 | bytes = len; | |
173 | ||
174 | if (map_data) { | |
175 | if (i == map_data->nr_entries * nr_pages) { | |
176 | ret = -ENOMEM; | |
7589ad67 | 177 | goto cleanup; |
130879f1 CH |
178 | } |
179 | ||
180 | page = map_data->pages[i / nr_pages]; | |
181 | page += (i % nr_pages); | |
182 | ||
183 | i++; | |
184 | } else { | |
ce288e05 | 185 | page = alloc_page(GFP_NOIO | gfp_mask); |
130879f1 CH |
186 | if (!page) { |
187 | ret = -ENOMEM; | |
7589ad67 | 188 | goto cleanup; |
130879f1 CH |
189 | } |
190 | } | |
191 | ||
7589ad67 | 192 | if (bio_add_pc_page(rq->q, bio, page, bytes, offset) < bytes) { |
130879f1 CH |
193 | if (!map_data) |
194 | __free_page(page); | |
195 | break; | |
196 | } | |
197 | ||
198 | len -= bytes; | |
199 | offset = 0; | |
200 | } | |
201 | ||
130879f1 CH |
202 | if (map_data) |
203 | map_data->offset += bio->bi_iter.bi_size; | |
204 | ||
205 | /* | |
206 | * success | |
207 | */ | |
13f3956e CE |
208 | if (iov_iter_rw(iter) == WRITE && |
209 | (!map_data || !map_data->null_mapped)) { | |
130879f1 CH |
210 | ret = bio_copy_from_iter(bio, iter); |
211 | if (ret) | |
212 | goto cleanup; | |
13f3956e CE |
213 | } else if (map_data && map_data->from_user) { |
214 | struct iov_iter iter2 = *iter; | |
215 | ||
216 | /* This is the copy-in part of SG_DXFER_TO_FROM_DEV. */ | |
217 | iter2.data_source = ITER_SOURCE; | |
218 | ret = bio_copy_from_iter(bio, &iter2); | |
219 | if (ret) | |
220 | goto cleanup; | |
130879f1 CH |
221 | } else { |
222 | if (bmd->is_our_pages) | |
223 | zero_fill_bio(bio); | |
224 | iov_iter_advance(iter, bio->bi_iter.bi_size); | |
225 | } | |
226 | ||
227 | bio->bi_private = bmd; | |
7589ad67 | 228 | |
393bb12e | 229 | ret = blk_rq_append_bio(rq, bio); |
7589ad67 CH |
230 | if (ret) |
231 | goto cleanup; | |
7589ad67 | 232 | return 0; |
130879f1 CH |
233 | cleanup: |
234 | if (!map_data) | |
235 | bio_free_pages(bio); | |
066ff571 CH |
236 | bio_uninit(bio); |
237 | kfree(bio); | |
130879f1 CH |
238 | out_bmd: |
239 | kfree(bmd); | |
7589ad67 | 240 | return ret; |
130879f1 CH |
241 | } |
242 | ||
32f1c71b | 243 | static void blk_mq_map_bio_put(struct bio *bio) |
8af870aa JA |
244 | { |
245 | if (bio->bi_opf & REQ_ALLOC_CACHE) { | |
246 | bio_put(bio); | |
247 | } else { | |
248 | bio_uninit(bio); | |
249 | kfree(bio); | |
250 | } | |
251 | } | |
252 | ||
ab89e8e7 KJ |
253 | static struct bio *blk_rq_map_bio_alloc(struct request *rq, |
254 | unsigned int nr_vecs, gfp_t gfp_mask) | |
130879f1 | 255 | { |
393bb12e | 256 | struct bio *bio; |
130879f1 | 257 | |
46930b7c | 258 | if (rq->cmd_flags & REQ_ALLOC_CACHE && (nr_vecs <= BIO_INLINE_VECS)) { |
7e2e355d | 259 | bio = bio_alloc_bioset(NULL, nr_vecs, rq->cmd_flags, gfp_mask, |
8af870aa JA |
260 | &fs_bio_set); |
261 | if (!bio) | |
ab89e8e7 | 262 | return NULL; |
8af870aa JA |
263 | } else { |
264 | bio = bio_kmalloc(nr_vecs, gfp_mask); | |
265 | if (!bio) | |
ab89e8e7 | 266 | return NULL; |
8af870aa JA |
267 | bio_init(bio, NULL, bio->bi_inline_vecs, nr_vecs, req_op(rq)); |
268 | } | |
ab89e8e7 KJ |
269 | return bio; |
270 | } | |
271 | ||
272 | static int bio_map_user_iov(struct request *rq, struct iov_iter *iter, | |
273 | gfp_t gfp_mask) | |
274 | { | |
f62e52d1 | 275 | iov_iter_extraction_t extraction_flags = 0; |
ab89e8e7 KJ |
276 | unsigned int max_sectors = queue_max_hw_sectors(rq->q); |
277 | unsigned int nr_vecs = iov_iter_npages(iter, BIO_MAX_VECS); | |
278 | struct bio *bio; | |
279 | int ret; | |
280 | int j; | |
281 | ||
282 | if (!iov_iter_count(iter)) | |
283 | return -EINVAL; | |
284 | ||
285 | bio = blk_rq_map_bio_alloc(rq, nr_vecs, gfp_mask); | |
286 | if (bio == NULL) | |
287 | return -ENOMEM; | |
130879f1 | 288 | |
7ee4ccf5 | 289 | if (blk_queue_pci_p2pdma(rq->q)) |
f62e52d1 | 290 | extraction_flags |= ITER_ALLOW_P2PDMA; |
403b6fb8 DH |
291 | if (iov_iter_extract_will_pin(iter)) |
292 | bio_set_flag(bio, BIO_PAGE_PINNED); | |
7ee4ccf5 | 293 | |
130879f1 | 294 | while (iov_iter_count(iter)) { |
403b6fb8 DH |
295 | struct page *stack_pages[UIO_FASTIOV]; |
296 | struct page **pages = stack_pages; | |
130879f1 | 297 | ssize_t bytes; |
91e5adda | 298 | size_t offs; |
130879f1 CH |
299 | int npages; |
300 | ||
403b6fb8 DH |
301 | if (nr_vecs > ARRAY_SIZE(stack_pages)) |
302 | pages = NULL; | |
303 | ||
304 | bytes = iov_iter_extract_pages(iter, &pages, LONG_MAX, | |
305 | nr_vecs, extraction_flags, &offs); | |
130879f1 CH |
306 | if (unlikely(bytes <= 0)) { |
307 | ret = bytes ? bytes : -EFAULT; | |
308 | goto out_unmap; | |
309 | } | |
310 | ||
311 | npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); | |
312 | ||
7ab89db9 | 313 | if (unlikely(offs & queue_dma_alignment(rq->q))) |
130879f1 | 314 | j = 0; |
7ab89db9 | 315 | else { |
130879f1 CH |
316 | for (j = 0; j < npages; j++) { |
317 | struct page *page = pages[j]; | |
318 | unsigned int n = PAGE_SIZE - offs; | |
319 | bool same_page = false; | |
320 | ||
321 | if (n > bytes) | |
322 | n = bytes; | |
323 | ||
7589ad67 | 324 | if (!bio_add_hw_page(rq->q, bio, page, n, offs, |
5905afc2 | 325 | max_sectors, &same_page)) |
130879f1 | 326 | break; |
130879f1 | 327 | |
5905afc2 CH |
328 | if (same_page) |
329 | bio_release_page(bio, page); | |
130879f1 CH |
330 | bytes -= n; |
331 | offs = 0; | |
332 | } | |
130879f1 CH |
333 | } |
334 | /* | |
335 | * release the pages we didn't map into the bio, if any | |
336 | */ | |
337 | while (j < npages) | |
403b6fb8 | 338 | bio_release_page(bio, pages[j++]); |
e88811bc JA |
339 | if (pages != stack_pages) |
340 | kvfree(pages); | |
130879f1 | 341 | /* couldn't stuff something into bio? */ |
480cb846 AV |
342 | if (bytes) { |
343 | iov_iter_revert(iter, bytes); | |
130879f1 | 344 | break; |
480cb846 | 345 | } |
130879f1 CH |
346 | } |
347 | ||
393bb12e | 348 | ret = blk_rq_append_bio(rq, bio); |
7589ad67 | 349 | if (ret) |
393bb12e | 350 | goto out_unmap; |
7589ad67 CH |
351 | return 0; |
352 | ||
130879f1 CH |
353 | out_unmap: |
354 | bio_release_pages(bio, false); | |
32f1c71b | 355 | blk_mq_map_bio_put(bio); |
7589ad67 | 356 | return ret; |
130879f1 CH |
357 | } |
358 | ||
130879f1 CH |
359 | static void bio_invalidate_vmalloc_pages(struct bio *bio) |
360 | { | |
f358afc5 | 361 | #ifdef ARCH_IMPLEMENTS_FLUSH_KERNEL_VMAP_RANGE |
130879f1 CH |
362 | if (bio->bi_private && !op_is_write(bio_op(bio))) { |
363 | unsigned long i, len = 0; | |
364 | ||
365 | for (i = 0; i < bio->bi_vcnt; i++) | |
366 | len += bio->bi_io_vec[i].bv_len; | |
367 | invalidate_kernel_vmap_range(bio->bi_private, len); | |
368 | } | |
369 | #endif | |
370 | } | |
371 | ||
372 | static void bio_map_kern_endio(struct bio *bio) | |
373 | { | |
374 | bio_invalidate_vmalloc_pages(bio); | |
066ff571 CH |
375 | bio_uninit(bio); |
376 | kfree(bio); | |
130879f1 CH |
377 | } |
378 | ||
379 | /** | |
380 | * bio_map_kern - map kernel address into bio | |
381 | * @q: the struct request_queue for the bio | |
382 | * @data: pointer to buffer to map | |
383 | * @len: length in bytes | |
384 | * @gfp_mask: allocation flags for bio allocation | |
385 | * | |
386 | * Map the kernel address into a bio suitable for io to a block | |
387 | * device. Returns an error pointer in case of error. | |
388 | */ | |
389 | static struct bio *bio_map_kern(struct request_queue *q, void *data, | |
390 | unsigned int len, gfp_t gfp_mask) | |
391 | { | |
392 | unsigned long kaddr = (unsigned long)data; | |
393 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
394 | unsigned long start = kaddr >> PAGE_SHIFT; | |
395 | const int nr_pages = end - start; | |
396 | bool is_vmalloc = is_vmalloc_addr(data); | |
397 | struct page *page; | |
398 | int offset, i; | |
399 | struct bio *bio; | |
400 | ||
066ff571 | 401 | bio = bio_kmalloc(nr_pages, gfp_mask); |
130879f1 CH |
402 | if (!bio) |
403 | return ERR_PTR(-ENOMEM); | |
066ff571 | 404 | bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0); |
130879f1 CH |
405 | |
406 | if (is_vmalloc) { | |
407 | flush_kernel_vmap_range(data, len); | |
408 | bio->bi_private = data; | |
409 | } | |
410 | ||
411 | offset = offset_in_page(kaddr); | |
412 | for (i = 0; i < nr_pages; i++) { | |
413 | unsigned int bytes = PAGE_SIZE - offset; | |
414 | ||
415 | if (len <= 0) | |
416 | break; | |
417 | ||
418 | if (bytes > len) | |
419 | bytes = len; | |
420 | ||
421 | if (!is_vmalloc) | |
422 | page = virt_to_page(data); | |
423 | else | |
424 | page = vmalloc_to_page(data); | |
425 | if (bio_add_pc_page(q, bio, page, bytes, | |
426 | offset) < bytes) { | |
427 | /* we don't support partial mappings */ | |
066ff571 CH |
428 | bio_uninit(bio); |
429 | kfree(bio); | |
130879f1 CH |
430 | return ERR_PTR(-EINVAL); |
431 | } | |
432 | ||
433 | data += bytes; | |
434 | len -= bytes; | |
435 | offset = 0; | |
436 | } | |
437 | ||
438 | bio->bi_end_io = bio_map_kern_endio; | |
439 | return bio; | |
440 | } | |
441 | ||
442 | static void bio_copy_kern_endio(struct bio *bio) | |
443 | { | |
444 | bio_free_pages(bio); | |
066ff571 CH |
445 | bio_uninit(bio); |
446 | kfree(bio); | |
130879f1 CH |
447 | } |
448 | ||
449 | static void bio_copy_kern_endio_read(struct bio *bio) | |
450 | { | |
451 | char *p = bio->bi_private; | |
452 | struct bio_vec *bvec; | |
453 | struct bvec_iter_all iter_all; | |
454 | ||
455 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
d24920e2 | 456 | memcpy_from_bvec(p, bvec); |
130879f1 CH |
457 | p += bvec->bv_len; |
458 | } | |
459 | ||
460 | bio_copy_kern_endio(bio); | |
461 | } | |
462 | ||
463 | /** | |
464 | * bio_copy_kern - copy kernel address into bio | |
465 | * @q: the struct request_queue for the bio | |
466 | * @data: pointer to buffer to copy | |
467 | * @len: length in bytes | |
468 | * @gfp_mask: allocation flags for bio and page allocation | |
469 | * @reading: data direction is READ | |
470 | * | |
471 | * copy the kernel address into a bio suitable for io to a block | |
472 | * device. Returns an error pointer in case of error. | |
473 | */ | |
474 | static struct bio *bio_copy_kern(struct request_queue *q, void *data, | |
475 | unsigned int len, gfp_t gfp_mask, int reading) | |
476 | { | |
477 | unsigned long kaddr = (unsigned long)data; | |
478 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
479 | unsigned long start = kaddr >> PAGE_SHIFT; | |
480 | struct bio *bio; | |
481 | void *p = data; | |
482 | int nr_pages = 0; | |
483 | ||
484 | /* | |
485 | * Overflow, abort | |
486 | */ | |
487 | if (end < start) | |
488 | return ERR_PTR(-EINVAL); | |
489 | ||
490 | nr_pages = end - start; | |
066ff571 | 491 | bio = bio_kmalloc(nr_pages, gfp_mask); |
130879f1 CH |
492 | if (!bio) |
493 | return ERR_PTR(-ENOMEM); | |
066ff571 | 494 | bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0); |
130879f1 CH |
495 | |
496 | while (len) { | |
497 | struct page *page; | |
498 | unsigned int bytes = PAGE_SIZE; | |
499 | ||
500 | if (bytes > len) | |
501 | bytes = len; | |
502 | ||
cc8f7fe1 | 503 | page = alloc_page(GFP_NOIO | __GFP_ZERO | gfp_mask); |
130879f1 CH |
504 | if (!page) |
505 | goto cleanup; | |
506 | ||
507 | if (!reading) | |
508 | memcpy(page_address(page), p, bytes); | |
509 | ||
510 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) | |
511 | break; | |
512 | ||
513 | len -= bytes; | |
514 | p += bytes; | |
515 | } | |
516 | ||
517 | if (reading) { | |
518 | bio->bi_end_io = bio_copy_kern_endio_read; | |
519 | bio->bi_private = data; | |
520 | } else { | |
521 | bio->bi_end_io = bio_copy_kern_endio; | |
522 | } | |
523 | ||
524 | return bio; | |
525 | ||
526 | cleanup: | |
527 | bio_free_pages(bio); | |
066ff571 CH |
528 | bio_uninit(bio); |
529 | kfree(bio); | |
130879f1 CH |
530 | return ERR_PTR(-ENOMEM); |
531 | } | |
532 | ||
98d61d5b | 533 | /* |
0abc2a10 JA |
534 | * Append a bio to a passthrough request. Only works if the bio can be merged |
535 | * into the request based on the driver constraints. | |
98d61d5b | 536 | */ |
393bb12e | 537 | int blk_rq_append_bio(struct request *rq, struct bio *bio) |
86db1e29 | 538 | { |
14ccb66b CH |
539 | struct bvec_iter iter; |
540 | struct bio_vec bv; | |
541 | unsigned int nr_segs = 0; | |
0abc2a10 | 542 | |
393bb12e | 543 | bio_for_each_bvec(bv, bio, iter) |
14ccb66b CH |
544 | nr_segs++; |
545 | ||
98d61d5b | 546 | if (!rq->bio) { |
393bb12e | 547 | blk_rq_bio_prep(rq, bio, nr_segs); |
98d61d5b | 548 | } else { |
393bb12e | 549 | if (!ll_back_merge_fn(rq, bio, nr_segs)) |
98d61d5b | 550 | return -EINVAL; |
393bb12e CH |
551 | rq->biotail->bi_next = bio; |
552 | rq->biotail = bio; | |
553 | rq->__data_len += (bio)->bi_iter.bi_size; | |
554 | bio_crypt_free_ctx(bio); | |
86db1e29 | 555 | } |
98d61d5b | 556 | |
86db1e29 JA |
557 | return 0; |
558 | } | |
98d61d5b | 559 | EXPORT_SYMBOL(blk_rq_append_bio); |
86db1e29 | 560 | |
37987547 KJ |
561 | /* Prepare bio for passthrough IO given ITER_BVEC iter */ |
562 | static int blk_rq_map_user_bvec(struct request *rq, const struct iov_iter *iter) | |
563 | { | |
564 | struct request_queue *q = rq->q; | |
565 | size_t nr_iter = iov_iter_count(iter); | |
566 | size_t nr_segs = iter->nr_segs; | |
567 | struct bio_vec *bvecs, *bvprvp = NULL; | |
aa261f20 | 568 | const struct queue_limits *lim = &q->limits; |
37987547 KJ |
569 | unsigned int nsegs = 0, bytes = 0; |
570 | struct bio *bio; | |
571 | size_t i; | |
572 | ||
573 | if (!nr_iter || (nr_iter >> SECTOR_SHIFT) > queue_max_hw_sectors(q)) | |
574 | return -EINVAL; | |
575 | if (nr_segs > queue_max_segments(q)) | |
576 | return -EINVAL; | |
577 | ||
578 | /* no iovecs to alloc, as we already have a BVEC iterator */ | |
579 | bio = blk_rq_map_bio_alloc(rq, 0, GFP_KERNEL); | |
580 | if (bio == NULL) | |
581 | return -ENOMEM; | |
582 | ||
583 | bio_iov_bvec_set(bio, (struct iov_iter *)iter); | |
584 | blk_rq_bio_prep(rq, bio, nr_segs); | |
585 | ||
586 | /* loop to perform a bunch of sanity checks */ | |
587 | bvecs = (struct bio_vec *)iter->bvec; | |
588 | for (i = 0; i < nr_segs; i++) { | |
589 | struct bio_vec *bv = &bvecs[i]; | |
590 | ||
591 | /* | |
592 | * If the queue doesn't support SG gaps and adding this | |
593 | * offset would create a gap, fallback to copy. | |
594 | */ | |
595 | if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv->bv_offset)) { | |
596 | blk_mq_map_bio_put(bio); | |
597 | return -EREMOTEIO; | |
598 | } | |
599 | /* check full condition */ | |
600 | if (nsegs >= nr_segs || bytes > UINT_MAX - bv->bv_len) | |
601 | goto put_bio; | |
602 | if (bytes + bv->bv_len > nr_iter) | |
603 | goto put_bio; | |
604 | if (bv->bv_offset + bv->bv_len > PAGE_SIZE) | |
605 | goto put_bio; | |
606 | ||
607 | nsegs++; | |
608 | bytes += bv->bv_len; | |
609 | bvprvp = bv; | |
610 | } | |
611 | return 0; | |
612 | put_bio: | |
613 | blk_mq_map_bio_put(bio); | |
614 | return -EINVAL; | |
615 | } | |
616 | ||
86db1e29 | 617 | /** |
aebf526b | 618 | * blk_rq_map_user_iov - map user data to a request, for passthrough requests |
86db1e29 JA |
619 | * @q: request queue where request should be inserted |
620 | * @rq: request to map data to | |
152e283f | 621 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
26e49cfc | 622 | * @iter: iovec iterator |
a3bce90e | 623 | * @gfp_mask: memory allocation flags |
86db1e29 JA |
624 | * |
625 | * Description: | |
710027a4 | 626 | * Data will be mapped directly for zero copy I/O, if possible. Otherwise |
86db1e29 JA |
627 | * a kernel bounce buffer is used. |
628 | * | |
710027a4 | 629 | * A matching blk_rq_unmap_user() must be issued at the end of I/O, while |
86db1e29 | 630 | * still in process context. |
86db1e29 JA |
631 | */ |
632 | int blk_rq_map_user_iov(struct request_queue *q, struct request *rq, | |
26e49cfc KO |
633 | struct rq_map_data *map_data, |
634 | const struct iov_iter *iter, gfp_t gfp_mask) | |
86db1e29 | 635 | { |
37987547 | 636 | bool copy = false, map_bvec = false; |
357f435d | 637 | unsigned long align = q->dma_pad_mask | queue_dma_alignment(q); |
4d6af73d CH |
638 | struct bio *bio = NULL; |
639 | struct iov_iter i; | |
69e0927b | 640 | int ret = -EINVAL; |
86db1e29 | 641 | |
357f435d AV |
642 | if (map_data) |
643 | copy = true; | |
393bb12e CH |
644 | else if (blk_queue_may_bounce(q)) |
645 | copy = true; | |
357f435d AV |
646 | else if (iov_iter_alignment(iter) & align) |
647 | copy = true; | |
37987547 KJ |
648 | else if (iov_iter_is_bvec(iter)) |
649 | map_bvec = true; | |
d46aa786 | 650 | else if (!user_backed_iter(iter)) |
37987547 | 651 | copy = true; |
357f435d AV |
652 | else if (queue_virt_boundary(q)) |
653 | copy = queue_virt_boundary(q) & iov_iter_gap_alignment(iter); | |
afdc1a78 | 654 | |
37987547 KJ |
655 | if (map_bvec) { |
656 | ret = blk_rq_map_user_bvec(rq, iter); | |
657 | if (!ret) | |
658 | return 0; | |
659 | if (ret != -EREMOTEIO) | |
660 | goto fail; | |
661 | /* fall back to copying the data on limits mismatches */ | |
662 | copy = true; | |
663 | } | |
664 | ||
4d6af73d CH |
665 | i = *iter; |
666 | do { | |
7589ad67 CH |
667 | if (copy) |
668 | ret = bio_copy_user_iov(rq, map_data, &i, gfp_mask); | |
669 | else | |
670 | ret = bio_map_user_iov(rq, &i, gfp_mask); | |
4d6af73d CH |
671 | if (ret) |
672 | goto unmap_rq; | |
673 | if (!bio) | |
674 | bio = rq->bio; | |
675 | } while (iov_iter_count(&i)); | |
86db1e29 | 676 | |
86db1e29 | 677 | return 0; |
4d6af73d CH |
678 | |
679 | unmap_rq: | |
3b7995a9 | 680 | blk_rq_unmap_user(bio); |
a0ac402c | 681 | fail: |
4d6af73d | 682 | rq->bio = NULL; |
69e0927b | 683 | return ret; |
86db1e29 | 684 | } |
152e283f | 685 | EXPORT_SYMBOL(blk_rq_map_user_iov); |
86db1e29 | 686 | |
ddad8dd0 CH |
687 | int blk_rq_map_user(struct request_queue *q, struct request *rq, |
688 | struct rq_map_data *map_data, void __user *ubuf, | |
689 | unsigned long len, gfp_t gfp_mask) | |
690 | { | |
26e49cfc | 691 | struct iov_iter i; |
d46aa786 | 692 | int ret = import_ubuf(rq_data_dir(rq), ubuf, len, &i); |
ddad8dd0 | 693 | |
8f7e885a AV |
694 | if (unlikely(ret < 0)) |
695 | return ret; | |
ddad8dd0 | 696 | |
26e49cfc | 697 | return blk_rq_map_user_iov(q, rq, map_data, &i, gfp_mask); |
ddad8dd0 CH |
698 | } |
699 | EXPORT_SYMBOL(blk_rq_map_user); | |
700 | ||
55765402 AG |
701 | int blk_rq_map_user_io(struct request *req, struct rq_map_data *map_data, |
702 | void __user *ubuf, unsigned long buf_len, gfp_t gfp_mask, | |
703 | bool vec, int iov_count, bool check_iter_count, int rw) | |
704 | { | |
705 | int ret = 0; | |
706 | ||
707 | if (vec) { | |
708 | struct iovec fast_iov[UIO_FASTIOV]; | |
709 | struct iovec *iov = fast_iov; | |
710 | struct iov_iter iter; | |
711 | ||
712 | ret = import_iovec(rw, ubuf, iov_count ? iov_count : buf_len, | |
713 | UIO_FASTIOV, &iov, &iter); | |
714 | if (ret < 0) | |
715 | return ret; | |
716 | ||
717 | if (iov_count) { | |
718 | /* SG_IO howto says that the shorter of the two wins */ | |
719 | iov_iter_truncate(&iter, buf_len); | |
720 | if (check_iter_count && !iov_iter_count(&iter)) { | |
721 | kfree(iov); | |
722 | return -EINVAL; | |
723 | } | |
724 | } | |
725 | ||
726 | ret = blk_rq_map_user_iov(req->q, req, map_data, &iter, | |
727 | gfp_mask); | |
728 | kfree(iov); | |
729 | } else if (buf_len) { | |
730 | ret = blk_rq_map_user(req->q, req, map_data, ubuf, buf_len, | |
731 | gfp_mask); | |
732 | } | |
733 | return ret; | |
734 | } | |
735 | EXPORT_SYMBOL(blk_rq_map_user_io); | |
736 | ||
86db1e29 JA |
737 | /** |
738 | * blk_rq_unmap_user - unmap a request with user data | |
739 | * @bio: start of bio list | |
740 | * | |
741 | * Description: | |
742 | * Unmap a rq previously mapped by blk_rq_map_user(). The caller must | |
743 | * supply the original rq->bio from the blk_rq_map_user() return, since | |
710027a4 | 744 | * the I/O completion may have changed rq->bio. |
86db1e29 JA |
745 | */ |
746 | int blk_rq_unmap_user(struct bio *bio) | |
747 | { | |
393bb12e | 748 | struct bio *next_bio; |
86db1e29 JA |
749 | int ret = 0, ret2; |
750 | ||
751 | while (bio) { | |
3310eeba | 752 | if (bio->bi_private) { |
393bb12e | 753 | ret2 = bio_uncopy_user(bio); |
7b63c052 CH |
754 | if (ret2 && !ret) |
755 | ret = ret2; | |
3310eeba | 756 | } else { |
393bb12e | 757 | bio_release_pages(bio, bio_data_dir(bio) == READ); |
7b63c052 | 758 | } |
86db1e29 | 759 | |
393bb12e | 760 | next_bio = bio; |
86db1e29 | 761 | bio = bio->bi_next; |
32f1c71b | 762 | blk_mq_map_bio_put(next_bio); |
86db1e29 JA |
763 | } |
764 | ||
765 | return ret; | |
766 | } | |
86db1e29 JA |
767 | EXPORT_SYMBOL(blk_rq_unmap_user); |
768 | ||
769 | /** | |
aebf526b | 770 | * blk_rq_map_kern - map kernel data to a request, for passthrough requests |
86db1e29 JA |
771 | * @q: request queue where request should be inserted |
772 | * @rq: request to fill | |
773 | * @kbuf: the kernel buffer | |
774 | * @len: length of user data | |
775 | * @gfp_mask: memory allocation flags | |
68154e90 FT |
776 | * |
777 | * Description: | |
778 | * Data will be mapped directly if possible. Otherwise a bounce | |
e227867f | 779 | * buffer is used. Can be called multiple times to append multiple |
3a5a3927 | 780 | * buffers. |
86db1e29 JA |
781 | */ |
782 | int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf, | |
783 | unsigned int len, gfp_t gfp_mask) | |
784 | { | |
68154e90 | 785 | int reading = rq_data_dir(rq) == READ; |
14417799 | 786 | unsigned long addr = (unsigned long) kbuf; |
393bb12e | 787 | struct bio *bio; |
3a5a3927 | 788 | int ret; |
86db1e29 | 789 | |
ae03bf63 | 790 | if (len > (queue_max_hw_sectors(q) << 9)) |
86db1e29 JA |
791 | return -EINVAL; |
792 | if (!len || !kbuf) | |
793 | return -EINVAL; | |
794 | ||
393bb12e CH |
795 | if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf) || |
796 | blk_queue_may_bounce(q)) | |
68154e90 FT |
797 | bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading); |
798 | else | |
799 | bio = bio_map_kern(q, kbuf, len, gfp_mask); | |
800 | ||
86db1e29 JA |
801 | if (IS_ERR(bio)) |
802 | return PTR_ERR(bio); | |
803 | ||
aebf526b CH |
804 | bio->bi_opf &= ~REQ_OP_MASK; |
805 | bio->bi_opf |= req_op(rq); | |
86db1e29 | 806 | |
393bb12e | 807 | ret = blk_rq_append_bio(rq, bio); |
066ff571 CH |
808 | if (unlikely(ret)) { |
809 | bio_uninit(bio); | |
810 | kfree(bio); | |
811 | } | |
393bb12e | 812 | return ret; |
86db1e29 | 813 | } |
86db1e29 | 814 | EXPORT_SYMBOL(blk_rq_map_kern); |