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 | */ | |
208 | if ((iov_iter_rw(iter) == WRITE && | |
209 | (!map_data || !map_data->null_mapped)) || | |
210 | (map_data && map_data->from_user)) { | |
211 | ret = bio_copy_from_iter(bio, iter); | |
212 | if (ret) | |
213 | goto cleanup; | |
214 | } else { | |
215 | if (bmd->is_our_pages) | |
216 | zero_fill_bio(bio); | |
217 | iov_iter_advance(iter, bio->bi_iter.bi_size); | |
218 | } | |
219 | ||
220 | bio->bi_private = bmd; | |
7589ad67 | 221 | |
393bb12e | 222 | ret = blk_rq_append_bio(rq, bio); |
7589ad67 CH |
223 | if (ret) |
224 | goto cleanup; | |
7589ad67 | 225 | return 0; |
130879f1 CH |
226 | cleanup: |
227 | if (!map_data) | |
228 | bio_free_pages(bio); | |
066ff571 CH |
229 | bio_uninit(bio); |
230 | kfree(bio); | |
130879f1 CH |
231 | out_bmd: |
232 | kfree(bmd); | |
7589ad67 | 233 | return ret; |
130879f1 CH |
234 | } |
235 | ||
32f1c71b | 236 | static void blk_mq_map_bio_put(struct bio *bio) |
8af870aa JA |
237 | { |
238 | if (bio->bi_opf & REQ_ALLOC_CACHE) { | |
239 | bio_put(bio); | |
240 | } else { | |
241 | bio_uninit(bio); | |
242 | kfree(bio); | |
243 | } | |
244 | } | |
245 | ||
ab89e8e7 KJ |
246 | static struct bio *blk_rq_map_bio_alloc(struct request *rq, |
247 | unsigned int nr_vecs, gfp_t gfp_mask) | |
130879f1 | 248 | { |
393bb12e | 249 | struct bio *bio; |
130879f1 | 250 | |
46930b7c | 251 | if (rq->cmd_flags & REQ_ALLOC_CACHE && (nr_vecs <= BIO_INLINE_VECS)) { |
7e2e355d | 252 | bio = bio_alloc_bioset(NULL, nr_vecs, rq->cmd_flags, gfp_mask, |
8af870aa JA |
253 | &fs_bio_set); |
254 | if (!bio) | |
ab89e8e7 | 255 | return NULL; |
8af870aa JA |
256 | } else { |
257 | bio = bio_kmalloc(nr_vecs, gfp_mask); | |
258 | if (!bio) | |
ab89e8e7 | 259 | return NULL; |
8af870aa JA |
260 | bio_init(bio, NULL, bio->bi_inline_vecs, nr_vecs, req_op(rq)); |
261 | } | |
ab89e8e7 KJ |
262 | return bio; |
263 | } | |
264 | ||
265 | static int bio_map_user_iov(struct request *rq, struct iov_iter *iter, | |
266 | gfp_t gfp_mask) | |
267 | { | |
f62e52d1 | 268 | iov_iter_extraction_t extraction_flags = 0; |
ab89e8e7 KJ |
269 | unsigned int max_sectors = queue_max_hw_sectors(rq->q); |
270 | unsigned int nr_vecs = iov_iter_npages(iter, BIO_MAX_VECS); | |
271 | struct bio *bio; | |
272 | int ret; | |
273 | int j; | |
274 | ||
275 | if (!iov_iter_count(iter)) | |
276 | return -EINVAL; | |
277 | ||
278 | bio = blk_rq_map_bio_alloc(rq, nr_vecs, gfp_mask); | |
279 | if (bio == NULL) | |
280 | return -ENOMEM; | |
130879f1 | 281 | |
7ee4ccf5 | 282 | if (blk_queue_pci_p2pdma(rq->q)) |
f62e52d1 | 283 | extraction_flags |= ITER_ALLOW_P2PDMA; |
403b6fb8 DH |
284 | if (iov_iter_extract_will_pin(iter)) |
285 | bio_set_flag(bio, BIO_PAGE_PINNED); | |
7ee4ccf5 | 286 | |
130879f1 | 287 | while (iov_iter_count(iter)) { |
403b6fb8 DH |
288 | struct page *stack_pages[UIO_FASTIOV]; |
289 | struct page **pages = stack_pages; | |
130879f1 | 290 | ssize_t bytes; |
91e5adda | 291 | size_t offs; |
130879f1 CH |
292 | int npages; |
293 | ||
403b6fb8 DH |
294 | if (nr_vecs > ARRAY_SIZE(stack_pages)) |
295 | pages = NULL; | |
296 | ||
297 | bytes = iov_iter_extract_pages(iter, &pages, LONG_MAX, | |
298 | nr_vecs, extraction_flags, &offs); | |
130879f1 CH |
299 | if (unlikely(bytes <= 0)) { |
300 | ret = bytes ? bytes : -EFAULT; | |
301 | goto out_unmap; | |
302 | } | |
303 | ||
304 | npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE); | |
305 | ||
7ab89db9 | 306 | if (unlikely(offs & queue_dma_alignment(rq->q))) |
130879f1 | 307 | j = 0; |
7ab89db9 | 308 | else { |
130879f1 CH |
309 | for (j = 0; j < npages; j++) { |
310 | struct page *page = pages[j]; | |
311 | unsigned int n = PAGE_SIZE - offs; | |
312 | bool same_page = false; | |
313 | ||
314 | if (n > bytes) | |
315 | n = bytes; | |
316 | ||
7589ad67 | 317 | if (!bio_add_hw_page(rq->q, bio, page, n, offs, |
5905afc2 | 318 | max_sectors, &same_page)) |
130879f1 | 319 | break; |
130879f1 | 320 | |
5905afc2 CH |
321 | if (same_page) |
322 | bio_release_page(bio, page); | |
130879f1 CH |
323 | bytes -= n; |
324 | offs = 0; | |
325 | } | |
130879f1 CH |
326 | } |
327 | /* | |
328 | * release the pages we didn't map into the bio, if any | |
329 | */ | |
330 | while (j < npages) | |
403b6fb8 | 331 | bio_release_page(bio, pages[j++]); |
e88811bc JA |
332 | if (pages != stack_pages) |
333 | kvfree(pages); | |
130879f1 | 334 | /* couldn't stuff something into bio? */ |
480cb846 AV |
335 | if (bytes) { |
336 | iov_iter_revert(iter, bytes); | |
130879f1 | 337 | break; |
480cb846 | 338 | } |
130879f1 CH |
339 | } |
340 | ||
393bb12e | 341 | ret = blk_rq_append_bio(rq, bio); |
7589ad67 | 342 | if (ret) |
393bb12e | 343 | goto out_unmap; |
7589ad67 CH |
344 | return 0; |
345 | ||
130879f1 CH |
346 | out_unmap: |
347 | bio_release_pages(bio, false); | |
32f1c71b | 348 | blk_mq_map_bio_put(bio); |
7589ad67 | 349 | return ret; |
130879f1 CH |
350 | } |
351 | ||
130879f1 CH |
352 | static void bio_invalidate_vmalloc_pages(struct bio *bio) |
353 | { | |
f358afc5 | 354 | #ifdef ARCH_IMPLEMENTS_FLUSH_KERNEL_VMAP_RANGE |
130879f1 CH |
355 | if (bio->bi_private && !op_is_write(bio_op(bio))) { |
356 | unsigned long i, len = 0; | |
357 | ||
358 | for (i = 0; i < bio->bi_vcnt; i++) | |
359 | len += bio->bi_io_vec[i].bv_len; | |
360 | invalidate_kernel_vmap_range(bio->bi_private, len); | |
361 | } | |
362 | #endif | |
363 | } | |
364 | ||
365 | static void bio_map_kern_endio(struct bio *bio) | |
366 | { | |
367 | bio_invalidate_vmalloc_pages(bio); | |
066ff571 CH |
368 | bio_uninit(bio); |
369 | kfree(bio); | |
130879f1 CH |
370 | } |
371 | ||
372 | /** | |
373 | * bio_map_kern - map kernel address into bio | |
374 | * @q: the struct request_queue for the bio | |
375 | * @data: pointer to buffer to map | |
376 | * @len: length in bytes | |
377 | * @gfp_mask: allocation flags for bio allocation | |
378 | * | |
379 | * Map the kernel address into a bio suitable for io to a block | |
380 | * device. Returns an error pointer in case of error. | |
381 | */ | |
382 | static struct bio *bio_map_kern(struct request_queue *q, void *data, | |
383 | unsigned int len, gfp_t gfp_mask) | |
384 | { | |
385 | unsigned long kaddr = (unsigned long)data; | |
386 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
387 | unsigned long start = kaddr >> PAGE_SHIFT; | |
388 | const int nr_pages = end - start; | |
389 | bool is_vmalloc = is_vmalloc_addr(data); | |
390 | struct page *page; | |
391 | int offset, i; | |
392 | struct bio *bio; | |
393 | ||
066ff571 | 394 | bio = bio_kmalloc(nr_pages, gfp_mask); |
130879f1 CH |
395 | if (!bio) |
396 | return ERR_PTR(-ENOMEM); | |
066ff571 | 397 | bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0); |
130879f1 CH |
398 | |
399 | if (is_vmalloc) { | |
400 | flush_kernel_vmap_range(data, len); | |
401 | bio->bi_private = data; | |
402 | } | |
403 | ||
404 | offset = offset_in_page(kaddr); | |
405 | for (i = 0; i < nr_pages; i++) { | |
406 | unsigned int bytes = PAGE_SIZE - offset; | |
407 | ||
408 | if (len <= 0) | |
409 | break; | |
410 | ||
411 | if (bytes > len) | |
412 | bytes = len; | |
413 | ||
414 | if (!is_vmalloc) | |
415 | page = virt_to_page(data); | |
416 | else | |
417 | page = vmalloc_to_page(data); | |
418 | if (bio_add_pc_page(q, bio, page, bytes, | |
419 | offset) < bytes) { | |
420 | /* we don't support partial mappings */ | |
066ff571 CH |
421 | bio_uninit(bio); |
422 | kfree(bio); | |
130879f1 CH |
423 | return ERR_PTR(-EINVAL); |
424 | } | |
425 | ||
426 | data += bytes; | |
427 | len -= bytes; | |
428 | offset = 0; | |
429 | } | |
430 | ||
431 | bio->bi_end_io = bio_map_kern_endio; | |
432 | return bio; | |
433 | } | |
434 | ||
435 | static void bio_copy_kern_endio(struct bio *bio) | |
436 | { | |
437 | bio_free_pages(bio); | |
066ff571 CH |
438 | bio_uninit(bio); |
439 | kfree(bio); | |
130879f1 CH |
440 | } |
441 | ||
442 | static void bio_copy_kern_endio_read(struct bio *bio) | |
443 | { | |
444 | char *p = bio->bi_private; | |
445 | struct bio_vec *bvec; | |
446 | struct bvec_iter_all iter_all; | |
447 | ||
448 | bio_for_each_segment_all(bvec, bio, iter_all) { | |
d24920e2 | 449 | memcpy_from_bvec(p, bvec); |
130879f1 CH |
450 | p += bvec->bv_len; |
451 | } | |
452 | ||
453 | bio_copy_kern_endio(bio); | |
454 | } | |
455 | ||
456 | /** | |
457 | * bio_copy_kern - copy kernel address into bio | |
458 | * @q: the struct request_queue for the bio | |
459 | * @data: pointer to buffer to copy | |
460 | * @len: length in bytes | |
461 | * @gfp_mask: allocation flags for bio and page allocation | |
462 | * @reading: data direction is READ | |
463 | * | |
464 | * copy the kernel address into a bio suitable for io to a block | |
465 | * device. Returns an error pointer in case of error. | |
466 | */ | |
467 | static struct bio *bio_copy_kern(struct request_queue *q, void *data, | |
468 | unsigned int len, gfp_t gfp_mask, int reading) | |
469 | { | |
470 | unsigned long kaddr = (unsigned long)data; | |
471 | unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
472 | unsigned long start = kaddr >> PAGE_SHIFT; | |
473 | struct bio *bio; | |
474 | void *p = data; | |
475 | int nr_pages = 0; | |
476 | ||
477 | /* | |
478 | * Overflow, abort | |
479 | */ | |
480 | if (end < start) | |
481 | return ERR_PTR(-EINVAL); | |
482 | ||
483 | nr_pages = end - start; | |
066ff571 | 484 | bio = bio_kmalloc(nr_pages, gfp_mask); |
130879f1 CH |
485 | if (!bio) |
486 | return ERR_PTR(-ENOMEM); | |
066ff571 | 487 | bio_init(bio, NULL, bio->bi_inline_vecs, nr_pages, 0); |
130879f1 CH |
488 | |
489 | while (len) { | |
490 | struct page *page; | |
491 | unsigned int bytes = PAGE_SIZE; | |
492 | ||
493 | if (bytes > len) | |
494 | bytes = len; | |
495 | ||
cc8f7fe1 | 496 | page = alloc_page(GFP_NOIO | __GFP_ZERO | gfp_mask); |
130879f1 CH |
497 | if (!page) |
498 | goto cleanup; | |
499 | ||
500 | if (!reading) | |
501 | memcpy(page_address(page), p, bytes); | |
502 | ||
503 | if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) | |
504 | break; | |
505 | ||
506 | len -= bytes; | |
507 | p += bytes; | |
508 | } | |
509 | ||
510 | if (reading) { | |
511 | bio->bi_end_io = bio_copy_kern_endio_read; | |
512 | bio->bi_private = data; | |
513 | } else { | |
514 | bio->bi_end_io = bio_copy_kern_endio; | |
515 | } | |
516 | ||
517 | return bio; | |
518 | ||
519 | cleanup: | |
520 | bio_free_pages(bio); | |
066ff571 CH |
521 | bio_uninit(bio); |
522 | kfree(bio); | |
130879f1 CH |
523 | return ERR_PTR(-ENOMEM); |
524 | } | |
525 | ||
98d61d5b | 526 | /* |
0abc2a10 JA |
527 | * Append a bio to a passthrough request. Only works if the bio can be merged |
528 | * into the request based on the driver constraints. | |
98d61d5b | 529 | */ |
393bb12e | 530 | int blk_rq_append_bio(struct request *rq, struct bio *bio) |
86db1e29 | 531 | { |
14ccb66b CH |
532 | struct bvec_iter iter; |
533 | struct bio_vec bv; | |
534 | unsigned int nr_segs = 0; | |
0abc2a10 | 535 | |
393bb12e | 536 | bio_for_each_bvec(bv, bio, iter) |
14ccb66b CH |
537 | nr_segs++; |
538 | ||
98d61d5b | 539 | if (!rq->bio) { |
393bb12e | 540 | blk_rq_bio_prep(rq, bio, nr_segs); |
98d61d5b | 541 | } else { |
393bb12e | 542 | if (!ll_back_merge_fn(rq, bio, nr_segs)) |
98d61d5b | 543 | return -EINVAL; |
393bb12e CH |
544 | rq->biotail->bi_next = bio; |
545 | rq->biotail = bio; | |
546 | rq->__data_len += (bio)->bi_iter.bi_size; | |
547 | bio_crypt_free_ctx(bio); | |
86db1e29 | 548 | } |
98d61d5b | 549 | |
86db1e29 JA |
550 | return 0; |
551 | } | |
98d61d5b | 552 | EXPORT_SYMBOL(blk_rq_append_bio); |
86db1e29 | 553 | |
37987547 KJ |
554 | /* Prepare bio for passthrough IO given ITER_BVEC iter */ |
555 | static int blk_rq_map_user_bvec(struct request *rq, const struct iov_iter *iter) | |
556 | { | |
557 | struct request_queue *q = rq->q; | |
558 | size_t nr_iter = iov_iter_count(iter); | |
559 | size_t nr_segs = iter->nr_segs; | |
560 | struct bio_vec *bvecs, *bvprvp = NULL; | |
aa261f20 | 561 | const struct queue_limits *lim = &q->limits; |
37987547 KJ |
562 | unsigned int nsegs = 0, bytes = 0; |
563 | struct bio *bio; | |
564 | size_t i; | |
565 | ||
566 | if (!nr_iter || (nr_iter >> SECTOR_SHIFT) > queue_max_hw_sectors(q)) | |
567 | return -EINVAL; | |
568 | if (nr_segs > queue_max_segments(q)) | |
569 | return -EINVAL; | |
570 | ||
571 | /* no iovecs to alloc, as we already have a BVEC iterator */ | |
572 | bio = blk_rq_map_bio_alloc(rq, 0, GFP_KERNEL); | |
573 | if (bio == NULL) | |
574 | return -ENOMEM; | |
575 | ||
576 | bio_iov_bvec_set(bio, (struct iov_iter *)iter); | |
577 | blk_rq_bio_prep(rq, bio, nr_segs); | |
578 | ||
579 | /* loop to perform a bunch of sanity checks */ | |
580 | bvecs = (struct bio_vec *)iter->bvec; | |
581 | for (i = 0; i < nr_segs; i++) { | |
582 | struct bio_vec *bv = &bvecs[i]; | |
583 | ||
584 | /* | |
585 | * If the queue doesn't support SG gaps and adding this | |
586 | * offset would create a gap, fallback to copy. | |
587 | */ | |
588 | if (bvprvp && bvec_gap_to_prev(lim, bvprvp, bv->bv_offset)) { | |
589 | blk_mq_map_bio_put(bio); | |
590 | return -EREMOTEIO; | |
591 | } | |
592 | /* check full condition */ | |
593 | if (nsegs >= nr_segs || bytes > UINT_MAX - bv->bv_len) | |
594 | goto put_bio; | |
595 | if (bytes + bv->bv_len > nr_iter) | |
596 | goto put_bio; | |
597 | if (bv->bv_offset + bv->bv_len > PAGE_SIZE) | |
598 | goto put_bio; | |
599 | ||
600 | nsegs++; | |
601 | bytes += bv->bv_len; | |
602 | bvprvp = bv; | |
603 | } | |
604 | return 0; | |
605 | put_bio: | |
606 | blk_mq_map_bio_put(bio); | |
607 | return -EINVAL; | |
608 | } | |
609 | ||
86db1e29 | 610 | /** |
aebf526b | 611 | * blk_rq_map_user_iov - map user data to a request, for passthrough requests |
86db1e29 JA |
612 | * @q: request queue where request should be inserted |
613 | * @rq: request to map data to | |
152e283f | 614 | * @map_data: pointer to the rq_map_data holding pages (if necessary) |
26e49cfc | 615 | * @iter: iovec iterator |
a3bce90e | 616 | * @gfp_mask: memory allocation flags |
86db1e29 JA |
617 | * |
618 | * Description: | |
710027a4 | 619 | * Data will be mapped directly for zero copy I/O, if possible. Otherwise |
86db1e29 JA |
620 | * a kernel bounce buffer is used. |
621 | * | |
710027a4 | 622 | * A matching blk_rq_unmap_user() must be issued at the end of I/O, while |
86db1e29 | 623 | * still in process context. |
86db1e29 JA |
624 | */ |
625 | int blk_rq_map_user_iov(struct request_queue *q, struct request *rq, | |
26e49cfc KO |
626 | struct rq_map_data *map_data, |
627 | const struct iov_iter *iter, gfp_t gfp_mask) | |
86db1e29 | 628 | { |
37987547 | 629 | bool copy = false, map_bvec = false; |
357f435d | 630 | unsigned long align = q->dma_pad_mask | queue_dma_alignment(q); |
4d6af73d CH |
631 | struct bio *bio = NULL; |
632 | struct iov_iter i; | |
69e0927b | 633 | int ret = -EINVAL; |
86db1e29 | 634 | |
357f435d AV |
635 | if (map_data) |
636 | copy = true; | |
393bb12e CH |
637 | else if (blk_queue_may_bounce(q)) |
638 | copy = true; | |
357f435d AV |
639 | else if (iov_iter_alignment(iter) & align) |
640 | copy = true; | |
37987547 KJ |
641 | else if (iov_iter_is_bvec(iter)) |
642 | map_bvec = true; | |
d46aa786 | 643 | else if (!user_backed_iter(iter)) |
37987547 | 644 | copy = true; |
357f435d AV |
645 | else if (queue_virt_boundary(q)) |
646 | copy = queue_virt_boundary(q) & iov_iter_gap_alignment(iter); | |
afdc1a78 | 647 | |
37987547 KJ |
648 | if (map_bvec) { |
649 | ret = blk_rq_map_user_bvec(rq, iter); | |
650 | if (!ret) | |
651 | return 0; | |
652 | if (ret != -EREMOTEIO) | |
653 | goto fail; | |
654 | /* fall back to copying the data on limits mismatches */ | |
655 | copy = true; | |
656 | } | |
657 | ||
4d6af73d CH |
658 | i = *iter; |
659 | do { | |
7589ad67 CH |
660 | if (copy) |
661 | ret = bio_copy_user_iov(rq, map_data, &i, gfp_mask); | |
662 | else | |
663 | ret = bio_map_user_iov(rq, &i, gfp_mask); | |
4d6af73d CH |
664 | if (ret) |
665 | goto unmap_rq; | |
666 | if (!bio) | |
667 | bio = rq->bio; | |
668 | } while (iov_iter_count(&i)); | |
86db1e29 | 669 | |
86db1e29 | 670 | return 0; |
4d6af73d CH |
671 | |
672 | unmap_rq: | |
3b7995a9 | 673 | blk_rq_unmap_user(bio); |
a0ac402c | 674 | fail: |
4d6af73d | 675 | rq->bio = NULL; |
69e0927b | 676 | return ret; |
86db1e29 | 677 | } |
152e283f | 678 | EXPORT_SYMBOL(blk_rq_map_user_iov); |
86db1e29 | 679 | |
ddad8dd0 CH |
680 | int blk_rq_map_user(struct request_queue *q, struct request *rq, |
681 | struct rq_map_data *map_data, void __user *ubuf, | |
682 | unsigned long len, gfp_t gfp_mask) | |
683 | { | |
26e49cfc | 684 | struct iov_iter i; |
d46aa786 | 685 | int ret = import_ubuf(rq_data_dir(rq), ubuf, len, &i); |
ddad8dd0 | 686 | |
8f7e885a AV |
687 | if (unlikely(ret < 0)) |
688 | return ret; | |
ddad8dd0 | 689 | |
26e49cfc | 690 | return blk_rq_map_user_iov(q, rq, map_data, &i, gfp_mask); |
ddad8dd0 CH |
691 | } |
692 | EXPORT_SYMBOL(blk_rq_map_user); | |
693 | ||
55765402 AG |
694 | int blk_rq_map_user_io(struct request *req, struct rq_map_data *map_data, |
695 | void __user *ubuf, unsigned long buf_len, gfp_t gfp_mask, | |
696 | bool vec, int iov_count, bool check_iter_count, int rw) | |
697 | { | |
698 | int ret = 0; | |
699 | ||
700 | if (vec) { | |
701 | struct iovec fast_iov[UIO_FASTIOV]; | |
702 | struct iovec *iov = fast_iov; | |
703 | struct iov_iter iter; | |
704 | ||
705 | ret = import_iovec(rw, ubuf, iov_count ? iov_count : buf_len, | |
706 | UIO_FASTIOV, &iov, &iter); | |
707 | if (ret < 0) | |
708 | return ret; | |
709 | ||
710 | if (iov_count) { | |
711 | /* SG_IO howto says that the shorter of the two wins */ | |
712 | iov_iter_truncate(&iter, buf_len); | |
713 | if (check_iter_count && !iov_iter_count(&iter)) { | |
714 | kfree(iov); | |
715 | return -EINVAL; | |
716 | } | |
717 | } | |
718 | ||
719 | ret = blk_rq_map_user_iov(req->q, req, map_data, &iter, | |
720 | gfp_mask); | |
721 | kfree(iov); | |
722 | } else if (buf_len) { | |
723 | ret = blk_rq_map_user(req->q, req, map_data, ubuf, buf_len, | |
724 | gfp_mask); | |
725 | } | |
726 | return ret; | |
727 | } | |
728 | EXPORT_SYMBOL(blk_rq_map_user_io); | |
729 | ||
86db1e29 JA |
730 | /** |
731 | * blk_rq_unmap_user - unmap a request with user data | |
732 | * @bio: start of bio list | |
733 | * | |
734 | * Description: | |
735 | * Unmap a rq previously mapped by blk_rq_map_user(). The caller must | |
736 | * supply the original rq->bio from the blk_rq_map_user() return, since | |
710027a4 | 737 | * the I/O completion may have changed rq->bio. |
86db1e29 JA |
738 | */ |
739 | int blk_rq_unmap_user(struct bio *bio) | |
740 | { | |
393bb12e | 741 | struct bio *next_bio; |
86db1e29 JA |
742 | int ret = 0, ret2; |
743 | ||
744 | while (bio) { | |
3310eeba | 745 | if (bio->bi_private) { |
393bb12e | 746 | ret2 = bio_uncopy_user(bio); |
7b63c052 CH |
747 | if (ret2 && !ret) |
748 | ret = ret2; | |
3310eeba | 749 | } else { |
393bb12e | 750 | bio_release_pages(bio, bio_data_dir(bio) == READ); |
7b63c052 | 751 | } |
86db1e29 | 752 | |
393bb12e | 753 | next_bio = bio; |
86db1e29 | 754 | bio = bio->bi_next; |
32f1c71b | 755 | blk_mq_map_bio_put(next_bio); |
86db1e29 JA |
756 | } |
757 | ||
758 | return ret; | |
759 | } | |
86db1e29 JA |
760 | EXPORT_SYMBOL(blk_rq_unmap_user); |
761 | ||
762 | /** | |
aebf526b | 763 | * blk_rq_map_kern - map kernel data to a request, for passthrough requests |
86db1e29 JA |
764 | * @q: request queue where request should be inserted |
765 | * @rq: request to fill | |
766 | * @kbuf: the kernel buffer | |
767 | * @len: length of user data | |
768 | * @gfp_mask: memory allocation flags | |
68154e90 FT |
769 | * |
770 | * Description: | |
771 | * Data will be mapped directly if possible. Otherwise a bounce | |
e227867f | 772 | * buffer is used. Can be called multiple times to append multiple |
3a5a3927 | 773 | * buffers. |
86db1e29 JA |
774 | */ |
775 | int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf, | |
776 | unsigned int len, gfp_t gfp_mask) | |
777 | { | |
68154e90 | 778 | int reading = rq_data_dir(rq) == READ; |
14417799 | 779 | unsigned long addr = (unsigned long) kbuf; |
393bb12e | 780 | struct bio *bio; |
3a5a3927 | 781 | int ret; |
86db1e29 | 782 | |
ae03bf63 | 783 | if (len > (queue_max_hw_sectors(q) << 9)) |
86db1e29 JA |
784 | return -EINVAL; |
785 | if (!len || !kbuf) | |
786 | return -EINVAL; | |
787 | ||
393bb12e CH |
788 | if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf) || |
789 | blk_queue_may_bounce(q)) | |
68154e90 FT |
790 | bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading); |
791 | else | |
792 | bio = bio_map_kern(q, kbuf, len, gfp_mask); | |
793 | ||
86db1e29 JA |
794 | if (IS_ERR(bio)) |
795 | return PTR_ERR(bio); | |
796 | ||
aebf526b CH |
797 | bio->bi_opf &= ~REQ_OP_MASK; |
798 | bio->bi_opf |= req_op(rq); | |
86db1e29 | 799 | |
393bb12e | 800 | ret = blk_rq_append_bio(rq, bio); |
066ff571 CH |
801 | if (unlikely(ret)) { |
802 | bio_uninit(bio); | |
803 | kfree(bio); | |
804 | } | |
393bb12e | 805 | return ret; |
86db1e29 | 806 | } |
86db1e29 | 807 | EXPORT_SYMBOL(blk_rq_map_kern); |