Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0
2 : /*
3 : * Functions related to mapping data to requests
4 : */
5 : #include <linux/kernel.h>
6 : #include <linux/sched/task_stack.h>
7 : #include <linux/module.h>
8 : #include <linux/bio.h>
9 : #include <linux/blkdev.h>
10 : #include <linux/uio.h>
11 :
12 : #include "blk.h"
13 :
14 : struct bio_map_data {
15 : bool is_our_pages : 1;
16 : bool is_null_mapped : 1;
17 : struct iov_iter iter;
18 : struct iovec iov[];
19 : };
20 :
21 0 : 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 0 : if (data->nr_segs > UIO_MAXIOV)
27 : return NULL;
28 :
29 0 : bmd = kmalloc(struct_size(bmd, iov, data->nr_segs), gfp_mask);
30 0 : if (!bmd)
31 : return NULL;
32 0 : memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs);
33 0 : bmd->iter = *data;
34 0 : bmd->iter.iov = bmd->iov;
35 0 : return bmd;
36 : }
37 :
38 : /**
39 : * bio_copy_from_iter - copy all pages from iov_iter to bio
40 : * @bio: The &struct bio which describes the I/O as destination
41 : * @iter: iov_iter as source
42 : *
43 : * Copy all pages from iov_iter to bio.
44 : * Returns 0 on success, or error on failure.
45 : */
46 0 : static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
47 : {
48 : struct bio_vec *bvec;
49 : struct bvec_iter_all iter_all;
50 :
51 0 : bio_for_each_segment_all(bvec, bio, iter_all) {
52 : ssize_t ret;
53 :
54 0 : ret = copy_page_from_iter(bvec->bv_page,
55 0 : bvec->bv_offset,
56 0 : bvec->bv_len,
57 : iter);
58 :
59 0 : if (!iov_iter_count(iter))
60 : break;
61 :
62 0 : if (ret < bvec->bv_len)
63 : return -EFAULT;
64 : }
65 :
66 : return 0;
67 : }
68 :
69 : /**
70 : * bio_copy_to_iter - copy all pages from bio to iov_iter
71 : * @bio: The &struct bio which describes the I/O as source
72 : * @iter: iov_iter as destination
73 : *
74 : * Copy all pages from bio to iov_iter.
75 : * Returns 0 on success, or error on failure.
76 : */
77 0 : static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
78 : {
79 : struct bio_vec *bvec;
80 : struct bvec_iter_all iter_all;
81 :
82 0 : bio_for_each_segment_all(bvec, bio, iter_all) {
83 : ssize_t ret;
84 :
85 0 : ret = copy_page_to_iter(bvec->bv_page,
86 0 : bvec->bv_offset,
87 0 : bvec->bv_len,
88 : &iter);
89 :
90 0 : if (!iov_iter_count(&iter))
91 : break;
92 :
93 0 : if (ret < bvec->bv_len)
94 : return -EFAULT;
95 : }
96 :
97 : return 0;
98 : }
99 :
100 : /**
101 : * bio_uncopy_user - finish previously mapped bio
102 : * @bio: bio being terminated
103 : *
104 : * Free pages allocated from bio_copy_user_iov() and write back data
105 : * to user space in case of a read.
106 : */
107 0 : static int bio_uncopy_user(struct bio *bio)
108 : {
109 0 : struct bio_map_data *bmd = bio->bi_private;
110 0 : int ret = 0;
111 :
112 0 : if (!bmd->is_null_mapped) {
113 : /*
114 : * if we're in a workqueue, the request is orphaned, so
115 : * don't copy into a random user address space, just free
116 : * and return -EINTR so user space doesn't expect any data.
117 : */
118 0 : if (!current->mm)
119 : ret = -EINTR;
120 0 : else if (bio_data_dir(bio) == READ)
121 0 : ret = bio_copy_to_iter(bio, bmd->iter);
122 0 : if (bmd->is_our_pages)
123 0 : bio_free_pages(bio);
124 : }
125 0 : kfree(bmd);
126 0 : return ret;
127 : }
128 :
129 0 : static int bio_copy_user_iov(struct request *rq, struct rq_map_data *map_data,
130 : struct iov_iter *iter, gfp_t gfp_mask)
131 : {
132 : struct bio_map_data *bmd;
133 : struct page *page;
134 : struct bio *bio;
135 0 : int i = 0, ret;
136 : int nr_pages;
137 0 : unsigned int len = iter->count;
138 0 : unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
139 :
140 0 : bmd = bio_alloc_map_data(iter, gfp_mask);
141 0 : if (!bmd)
142 : return -ENOMEM;
143 :
144 : /*
145 : * We need to do a deep copy of the iov_iter including the iovecs.
146 : * The caller provided iov might point to an on-stack or otherwise
147 : * shortlived one.
148 : */
149 0 : bmd->is_our_pages = !map_data;
150 0 : bmd->is_null_mapped = (map_data && map_data->null_mapped);
151 :
152 0 : nr_pages = bio_max_segs(DIV_ROUND_UP(offset + len, PAGE_SIZE));
153 :
154 0 : ret = -ENOMEM;
155 0 : bio = bio_kmalloc(gfp_mask, nr_pages);
156 0 : if (!bio)
157 : goto out_bmd;
158 0 : bio->bi_opf |= req_op(rq);
159 :
160 0 : if (map_data) {
161 0 : nr_pages = 1 << map_data->page_order;
162 0 : i = map_data->offset / PAGE_SIZE;
163 : }
164 0 : while (len) {
165 0 : unsigned int bytes = PAGE_SIZE;
166 :
167 0 : bytes -= offset;
168 :
169 0 : if (bytes > len)
170 0 : bytes = len;
171 :
172 0 : if (map_data) {
173 0 : if (i == map_data->nr_entries * nr_pages) {
174 : ret = -ENOMEM;
175 : goto cleanup;
176 : }
177 :
178 0 : page = map_data->pages[i / nr_pages];
179 0 : page += (i % nr_pages);
180 :
181 0 : i++;
182 : } else {
183 0 : page = alloc_page(GFP_NOIO | gfp_mask);
184 0 : if (!page) {
185 : ret = -ENOMEM;
186 : goto cleanup;
187 : }
188 : }
189 :
190 0 : if (bio_add_pc_page(rq->q, bio, page, bytes, offset) < bytes) {
191 0 : if (!map_data)
192 0 : __free_page(page);
193 : break;
194 : }
195 :
196 0 : len -= bytes;
197 0 : offset = 0;
198 : }
199 :
200 0 : if (map_data)
201 0 : map_data->offset += bio->bi_iter.bi_size;
202 :
203 : /*
204 : * success
205 : */
206 0 : if ((iov_iter_rw(iter) == WRITE &&
207 0 : (!map_data || !map_data->null_mapped)) ||
208 0 : (map_data && map_data->from_user)) {
209 0 : ret = bio_copy_from_iter(bio, iter);
210 0 : if (ret)
211 : goto cleanup;
212 : } else {
213 0 : if (bmd->is_our_pages)
214 0 : zero_fill_bio(bio);
215 0 : iov_iter_advance(iter, bio->bi_iter.bi_size);
216 : }
217 :
218 0 : bio->bi_private = bmd;
219 :
220 0 : ret = blk_rq_append_bio(rq, bio);
221 0 : if (ret)
222 : goto cleanup;
223 : return 0;
224 : cleanup:
225 0 : if (!map_data)
226 0 : bio_free_pages(bio);
227 0 : bio_put(bio);
228 : out_bmd:
229 0 : kfree(bmd);
230 0 : return ret;
231 : }
232 :
233 0 : static int bio_map_user_iov(struct request *rq, struct iov_iter *iter,
234 : gfp_t gfp_mask)
235 : {
236 0 : unsigned int max_sectors = queue_max_hw_sectors(rq->q);
237 : struct bio *bio;
238 : int ret;
239 : int j;
240 :
241 0 : if (!iov_iter_count(iter))
242 : return -EINVAL;
243 :
244 0 : bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_VECS));
245 0 : if (!bio)
246 : return -ENOMEM;
247 0 : bio->bi_opf |= req_op(rq);
248 :
249 0 : while (iov_iter_count(iter)) {
250 : struct page **pages;
251 : ssize_t bytes;
252 0 : size_t offs, added = 0;
253 : int npages;
254 :
255 0 : bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs);
256 0 : if (unlikely(bytes <= 0)) {
257 0 : ret = bytes ? bytes : -EFAULT;
258 0 : goto out_unmap;
259 : }
260 :
261 0 : npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
262 :
263 0 : if (unlikely(offs & queue_dma_alignment(rq->q))) {
264 : ret = -EINVAL;
265 : j = 0;
266 : } else {
267 0 : for (j = 0; j < npages; j++) {
268 0 : struct page *page = pages[j];
269 0 : unsigned int n = PAGE_SIZE - offs;
270 0 : bool same_page = false;
271 :
272 0 : if (n > bytes)
273 0 : n = bytes;
274 :
275 0 : if (!bio_add_hw_page(rq->q, bio, page, n, offs,
276 : max_sectors, &same_page)) {
277 0 : if (same_page)
278 0 : put_page(page);
279 0 : break;
280 : }
281 :
282 0 : added += n;
283 0 : bytes -= n;
284 0 : offs = 0;
285 : }
286 0 : iov_iter_advance(iter, added);
287 : }
288 : /*
289 : * release the pages we didn't map into the bio, if any
290 : */
291 0 : while (j < npages)
292 0 : put_page(pages[j++]);
293 0 : kvfree(pages);
294 : /* couldn't stuff something into bio? */
295 0 : if (bytes)
296 : break;
297 : }
298 :
299 0 : ret = blk_rq_append_bio(rq, bio);
300 0 : if (ret)
301 : goto out_unmap;
302 : return 0;
303 :
304 : out_unmap:
305 0 : bio_release_pages(bio, false);
306 0 : bio_put(bio);
307 0 : return ret;
308 : }
309 :
310 : static void bio_invalidate_vmalloc_pages(struct bio *bio)
311 : {
312 : #ifdef ARCH_IMPLEMENTS_FLUSH_KERNEL_VMAP_RANGE
313 : if (bio->bi_private && !op_is_write(bio_op(bio))) {
314 : unsigned long i, len = 0;
315 :
316 : for (i = 0; i < bio->bi_vcnt; i++)
317 : len += bio->bi_io_vec[i].bv_len;
318 : invalidate_kernel_vmap_range(bio->bi_private, len);
319 : }
320 : #endif
321 : }
322 :
323 0 : static void bio_map_kern_endio(struct bio *bio)
324 : {
325 0 : bio_invalidate_vmalloc_pages(bio);
326 0 : bio_put(bio);
327 0 : }
328 :
329 : /**
330 : * bio_map_kern - map kernel address into bio
331 : * @q: the struct request_queue for the bio
332 : * @data: pointer to buffer to map
333 : * @len: length in bytes
334 : * @gfp_mask: allocation flags for bio allocation
335 : *
336 : * Map the kernel address into a bio suitable for io to a block
337 : * device. Returns an error pointer in case of error.
338 : */
339 0 : static struct bio *bio_map_kern(struct request_queue *q, void *data,
340 : unsigned int len, gfp_t gfp_mask)
341 : {
342 0 : unsigned long kaddr = (unsigned long)data;
343 0 : unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
344 0 : unsigned long start = kaddr >> PAGE_SHIFT;
345 0 : const int nr_pages = end - start;
346 0 : bool is_vmalloc = is_vmalloc_addr(data);
347 : struct page *page;
348 : int offset, i;
349 : struct bio *bio;
350 :
351 0 : bio = bio_kmalloc(gfp_mask, nr_pages);
352 0 : if (!bio)
353 : return ERR_PTR(-ENOMEM);
354 :
355 0 : if (is_vmalloc) {
356 0 : flush_kernel_vmap_range(data, len);
357 0 : bio->bi_private = data;
358 : }
359 :
360 0 : offset = offset_in_page(kaddr);
361 0 : for (i = 0; i < nr_pages; i++) {
362 0 : unsigned int bytes = PAGE_SIZE - offset;
363 :
364 0 : if (len <= 0)
365 : break;
366 :
367 0 : if (bytes > len)
368 0 : bytes = len;
369 :
370 0 : if (!is_vmalloc)
371 0 : page = virt_to_page(data);
372 : else
373 0 : page = vmalloc_to_page(data);
374 0 : if (bio_add_pc_page(q, bio, page, bytes,
375 : offset) < bytes) {
376 : /* we don't support partial mappings */
377 0 : bio_put(bio);
378 0 : return ERR_PTR(-EINVAL);
379 : }
380 :
381 0 : data += bytes;
382 0 : len -= bytes;
383 0 : offset = 0;
384 : }
385 :
386 0 : bio->bi_end_io = bio_map_kern_endio;
387 0 : return bio;
388 : }
389 :
390 0 : static void bio_copy_kern_endio(struct bio *bio)
391 : {
392 0 : bio_free_pages(bio);
393 0 : bio_put(bio);
394 0 : }
395 :
396 0 : static void bio_copy_kern_endio_read(struct bio *bio)
397 : {
398 0 : char *p = bio->bi_private;
399 : struct bio_vec *bvec;
400 : struct bvec_iter_all iter_all;
401 :
402 0 : bio_for_each_segment_all(bvec, bio, iter_all) {
403 0 : memcpy_from_bvec(p, bvec);
404 0 : p += bvec->bv_len;
405 : }
406 :
407 0 : bio_copy_kern_endio(bio);
408 0 : }
409 :
410 : /**
411 : * bio_copy_kern - copy kernel address into bio
412 : * @q: the struct request_queue for the bio
413 : * @data: pointer to buffer to copy
414 : * @len: length in bytes
415 : * @gfp_mask: allocation flags for bio and page allocation
416 : * @reading: data direction is READ
417 : *
418 : * copy the kernel address into a bio suitable for io to a block
419 : * device. Returns an error pointer in case of error.
420 : */
421 0 : static struct bio *bio_copy_kern(struct request_queue *q, void *data,
422 : unsigned int len, gfp_t gfp_mask, int reading)
423 : {
424 0 : unsigned long kaddr = (unsigned long)data;
425 0 : unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
426 0 : unsigned long start = kaddr >> PAGE_SHIFT;
427 : struct bio *bio;
428 0 : void *p = data;
429 0 : int nr_pages = 0;
430 :
431 : /*
432 : * Overflow, abort
433 : */
434 0 : if (end < start)
435 : return ERR_PTR(-EINVAL);
436 :
437 0 : nr_pages = end - start;
438 0 : bio = bio_kmalloc(gfp_mask, nr_pages);
439 0 : if (!bio)
440 : return ERR_PTR(-ENOMEM);
441 :
442 0 : while (len) {
443 : struct page *page;
444 0 : unsigned int bytes = PAGE_SIZE;
445 :
446 0 : if (bytes > len)
447 0 : bytes = len;
448 :
449 0 : page = alloc_page(GFP_NOIO | __GFP_ZERO | gfp_mask);
450 0 : if (!page)
451 : goto cleanup;
452 :
453 0 : if (!reading)
454 0 : memcpy(page_address(page), p, bytes);
455 :
456 0 : if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
457 : break;
458 :
459 0 : len -= bytes;
460 0 : p += bytes;
461 : }
462 :
463 0 : if (reading) {
464 0 : bio->bi_end_io = bio_copy_kern_endio_read;
465 0 : bio->bi_private = data;
466 : } else {
467 0 : bio->bi_end_io = bio_copy_kern_endio;
468 : }
469 :
470 : return bio;
471 :
472 : cleanup:
473 0 : bio_free_pages(bio);
474 0 : bio_put(bio);
475 0 : return ERR_PTR(-ENOMEM);
476 : }
477 :
478 : /*
479 : * Append a bio to a passthrough request. Only works if the bio can be merged
480 : * into the request based on the driver constraints.
481 : */
482 0 : int blk_rq_append_bio(struct request *rq, struct bio *bio)
483 : {
484 : struct bvec_iter iter;
485 : struct bio_vec bv;
486 0 : unsigned int nr_segs = 0;
487 :
488 0 : bio_for_each_bvec(bv, bio, iter)
489 0 : nr_segs++;
490 :
491 0 : if (!rq->bio) {
492 : blk_rq_bio_prep(rq, bio, nr_segs);
493 : } else {
494 0 : if (!ll_back_merge_fn(rq, bio, nr_segs))
495 : return -EINVAL;
496 0 : rq->biotail->bi_next = bio;
497 0 : rq->biotail = bio;
498 0 : rq->__data_len += (bio)->bi_iter.bi_size;
499 0 : bio_crypt_free_ctx(bio);
500 : }
501 :
502 : return 0;
503 : }
504 : EXPORT_SYMBOL(blk_rq_append_bio);
505 :
506 : /**
507 : * blk_rq_map_user_iov - map user data to a request, for passthrough requests
508 : * @q: request queue where request should be inserted
509 : * @rq: request to map data to
510 : * @map_data: pointer to the rq_map_data holding pages (if necessary)
511 : * @iter: iovec iterator
512 : * @gfp_mask: memory allocation flags
513 : *
514 : * Description:
515 : * Data will be mapped directly for zero copy I/O, if possible. Otherwise
516 : * a kernel bounce buffer is used.
517 : *
518 : * A matching blk_rq_unmap_user() must be issued at the end of I/O, while
519 : * still in process context.
520 : */
521 0 : int blk_rq_map_user_iov(struct request_queue *q, struct request *rq,
522 : struct rq_map_data *map_data,
523 : const struct iov_iter *iter, gfp_t gfp_mask)
524 : {
525 0 : bool copy = false;
526 0 : unsigned long align = q->dma_pad_mask | queue_dma_alignment(q);
527 0 : struct bio *bio = NULL;
528 : struct iov_iter i;
529 0 : int ret = -EINVAL;
530 :
531 0 : if (!iter_is_iovec(iter))
532 : goto fail;
533 :
534 0 : if (map_data)
535 : copy = true;
536 0 : else if (blk_queue_may_bounce(q))
537 : copy = true;
538 0 : else if (iov_iter_alignment(iter) & align)
539 : copy = true;
540 0 : else if (queue_virt_boundary(q))
541 0 : copy = queue_virt_boundary(q) & iov_iter_gap_alignment(iter);
542 :
543 0 : i = *iter;
544 : do {
545 0 : if (copy)
546 0 : ret = bio_copy_user_iov(rq, map_data, &i, gfp_mask);
547 : else
548 0 : ret = bio_map_user_iov(rq, &i, gfp_mask);
549 0 : if (ret)
550 : goto unmap_rq;
551 0 : if (!bio)
552 0 : bio = rq->bio;
553 0 : } while (iov_iter_count(&i));
554 :
555 : return 0;
556 :
557 : unmap_rq:
558 0 : blk_rq_unmap_user(bio);
559 : fail:
560 0 : rq->bio = NULL;
561 0 : return ret;
562 : }
563 : EXPORT_SYMBOL(blk_rq_map_user_iov);
564 :
565 0 : int blk_rq_map_user(struct request_queue *q, struct request *rq,
566 : struct rq_map_data *map_data, void __user *ubuf,
567 : unsigned long len, gfp_t gfp_mask)
568 : {
569 : struct iovec iov;
570 : struct iov_iter i;
571 0 : int ret = import_single_range(rq_data_dir(rq), ubuf, len, &iov, &i);
572 :
573 0 : if (unlikely(ret < 0))
574 : return ret;
575 :
576 0 : return blk_rq_map_user_iov(q, rq, map_data, &i, gfp_mask);
577 : }
578 : EXPORT_SYMBOL(blk_rq_map_user);
579 :
580 : /**
581 : * blk_rq_unmap_user - unmap a request with user data
582 : * @bio: start of bio list
583 : *
584 : * Description:
585 : * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
586 : * supply the original rq->bio from the blk_rq_map_user() return, since
587 : * the I/O completion may have changed rq->bio.
588 : */
589 0 : int blk_rq_unmap_user(struct bio *bio)
590 : {
591 : struct bio *next_bio;
592 0 : int ret = 0, ret2;
593 :
594 0 : while (bio) {
595 0 : if (bio->bi_private) {
596 0 : ret2 = bio_uncopy_user(bio);
597 0 : if (ret2 && !ret)
598 0 : ret = ret2;
599 : } else {
600 0 : bio_release_pages(bio, bio_data_dir(bio) == READ);
601 : }
602 :
603 0 : next_bio = bio;
604 0 : bio = bio->bi_next;
605 0 : bio_put(next_bio);
606 : }
607 :
608 0 : return ret;
609 : }
610 : EXPORT_SYMBOL(blk_rq_unmap_user);
611 :
612 : /**
613 : * blk_rq_map_kern - map kernel data to a request, for passthrough requests
614 : * @q: request queue where request should be inserted
615 : * @rq: request to fill
616 : * @kbuf: the kernel buffer
617 : * @len: length of user data
618 : * @gfp_mask: memory allocation flags
619 : *
620 : * Description:
621 : * Data will be mapped directly if possible. Otherwise a bounce
622 : * buffer is used. Can be called multiple times to append multiple
623 : * buffers.
624 : */
625 0 : int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
626 : unsigned int len, gfp_t gfp_mask)
627 : {
628 0 : int reading = rq_data_dir(rq) == READ;
629 0 : unsigned long addr = (unsigned long) kbuf;
630 : struct bio *bio;
631 : int ret;
632 :
633 0 : if (len > (queue_max_hw_sectors(q) << 9))
634 : return -EINVAL;
635 0 : if (!len || !kbuf)
636 : return -EINVAL;
637 :
638 0 : if (!blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf) ||
639 0 : blk_queue_may_bounce(q))
640 0 : bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);
641 : else
642 0 : bio = bio_map_kern(q, kbuf, len, gfp_mask);
643 :
644 0 : if (IS_ERR(bio))
645 0 : return PTR_ERR(bio);
646 :
647 0 : bio->bi_opf &= ~REQ_OP_MASK;
648 0 : bio->bi_opf |= req_op(rq);
649 :
650 0 : ret = blk_rq_append_bio(rq, bio);
651 0 : if (unlikely(ret))
652 0 : bio_put(bio);
653 : return ret;
654 : }
655 : EXPORT_SYMBOL(blk_rq_map_kern);
|
