]>
git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - block/blk-map.c
1 // SPDX-License-Identifier: GPL-2.0
3 * Functions related to mapping data to requests
5 #include <linux/kernel.h>
6 #include <linux/sched/task_stack.h>
7 #include <linux/module.h>
9 #include <linux/blkdev.h>
10 #include <linux/uio.h>
20 static struct bio_map_data
*bio_alloc_map_data(struct iov_iter
*data
,
23 struct bio_map_data
*bmd
;
25 if (data
->nr_segs
> UIO_MAXIOV
)
28 bmd
= kmalloc(struct_size(bmd
, iov
, data
->nr_segs
), gfp_mask
);
31 memcpy(bmd
->iov
, data
->iov
, sizeof(struct iovec
) * data
->nr_segs
);
33 bmd
->iter
.iov
= bmd
->iov
;
38 * bio_copy_from_iter - copy all pages from iov_iter to bio
39 * @bio: The &struct bio which describes the I/O as destination
40 * @iter: iov_iter as source
42 * Copy all pages from iov_iter to bio.
43 * Returns 0 on success, or error on failure.
45 static int bio_copy_from_iter(struct bio
*bio
, struct iov_iter
*iter
)
48 struct bvec_iter_all iter_all
;
50 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
53 ret
= copy_page_from_iter(bvec
->bv_page
,
58 if (!iov_iter_count(iter
))
61 if (ret
< bvec
->bv_len
)
69 * bio_copy_to_iter - copy all pages from bio to iov_iter
70 * @bio: The &struct bio which describes the I/O as source
71 * @iter: iov_iter as destination
73 * Copy all pages from bio to iov_iter.
74 * Returns 0 on success, or error on failure.
76 static int bio_copy_to_iter(struct bio
*bio
, struct iov_iter iter
)
79 struct bvec_iter_all iter_all
;
81 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
84 ret
= copy_page_to_iter(bvec
->bv_page
,
89 if (!iov_iter_count(&iter
))
92 if (ret
< bvec
->bv_len
)
100 * bio_uncopy_user - finish previously mapped bio
101 * @bio: bio being terminated
103 * Free pages allocated from bio_copy_user_iov() and write back data
104 * to user space in case of a read.
106 static int bio_uncopy_user(struct bio
*bio
)
108 struct bio_map_data
*bmd
= bio
->bi_private
;
111 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
113 * if we're in a workqueue, the request is orphaned, so
114 * don't copy into a random user address space, just free
115 * and return -EINTR so user space doesn't expect any data.
119 else if (bio_data_dir(bio
) == READ
)
120 ret
= bio_copy_to_iter(bio
, bmd
->iter
);
121 if (bmd
->is_our_pages
)
130 * bio_copy_user_iov - copy user data to bio
131 * @q: destination block queue
132 * @map_data: pointer to the rq_map_data holding pages (if necessary)
133 * @iter: iovec iterator
134 * @gfp_mask: memory allocation flags
136 * Prepares and returns a bio for indirect user io, bouncing data
137 * to/from kernel pages as necessary. Must be paired with
138 * call bio_uncopy_user() on io completion.
140 static struct bio
*bio_copy_user_iov(struct request_queue
*q
,
141 struct rq_map_data
*map_data
, struct iov_iter
*iter
,
144 struct bio_map_data
*bmd
;
149 unsigned int len
= iter
->count
;
150 unsigned int offset
= map_data
? offset_in_page(map_data
->offset
) : 0;
152 bmd
= bio_alloc_map_data(iter
, gfp_mask
);
154 return ERR_PTR(-ENOMEM
);
157 * We need to do a deep copy of the iov_iter including the iovecs.
158 * The caller provided iov might point to an on-stack or otherwise
161 bmd
->is_our_pages
= map_data
? 0 : 1;
163 nr_pages
= DIV_ROUND_UP(offset
+ len
, PAGE_SIZE
);
164 if (nr_pages
> BIO_MAX_PAGES
)
165 nr_pages
= BIO_MAX_PAGES
;
168 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
175 nr_pages
= 1 << map_data
->page_order
;
176 i
= map_data
->offset
/ PAGE_SIZE
;
179 unsigned int bytes
= PAGE_SIZE
;
187 if (i
== map_data
->nr_entries
* nr_pages
) {
192 page
= map_data
->pages
[i
/ nr_pages
];
193 page
+= (i
% nr_pages
);
197 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
204 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
) {
218 map_data
->offset
+= bio
->bi_iter
.bi_size
;
223 if ((iov_iter_rw(iter
) == WRITE
&&
224 (!map_data
|| !map_data
->null_mapped
)) ||
225 (map_data
&& map_data
->from_user
)) {
226 ret
= bio_copy_from_iter(bio
, iter
);
230 if (bmd
->is_our_pages
)
232 iov_iter_advance(iter
, bio
->bi_iter
.bi_size
);
235 bio
->bi_private
= bmd
;
236 if (map_data
&& map_data
->null_mapped
)
237 bio_set_flag(bio
, BIO_NULL_MAPPED
);
249 * bio_map_user_iov - map user iovec into bio
250 * @q: the struct request_queue for the bio
251 * @iter: iovec iterator
252 * @gfp_mask: memory allocation flags
254 * Map the user space address into a bio suitable for io to a block
255 * device. Returns an error pointer in case of error.
257 static struct bio
*bio_map_user_iov(struct request_queue
*q
,
258 struct iov_iter
*iter
, gfp_t gfp_mask
)
264 if (!iov_iter_count(iter
))
265 return ERR_PTR(-EINVAL
);
267 bio
= bio_kmalloc(gfp_mask
, iov_iter_npages(iter
, BIO_MAX_PAGES
));
269 return ERR_PTR(-ENOMEM
);
271 while (iov_iter_count(iter
)) {
274 size_t offs
, added
= 0;
277 bytes
= iov_iter_get_pages_alloc(iter
, &pages
, LONG_MAX
, &offs
);
278 if (unlikely(bytes
<= 0)) {
279 ret
= bytes
? bytes
: -EFAULT
;
283 npages
= DIV_ROUND_UP(offs
+ bytes
, PAGE_SIZE
);
285 if (unlikely(offs
& queue_dma_alignment(q
))) {
289 for (j
= 0; j
< npages
; j
++) {
290 struct page
*page
= pages
[j
];
291 unsigned int n
= PAGE_SIZE
- offs
;
292 bool same_page
= false;
297 if (!__bio_add_pc_page(q
, bio
, page
, n
, offs
,
308 iov_iter_advance(iter
, added
);
311 * release the pages we didn't map into the bio, if any
314 put_page(pages
[j
++]);
316 /* couldn't stuff something into bio? */
321 bio_set_flag(bio
, BIO_USER_MAPPED
);
324 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
325 * it would normally disappear when its bi_end_io is run.
326 * however, we need it for the unmap, so grab an extra
333 bio_release_pages(bio
, false);
339 * bio_unmap_user - unmap a bio
340 * @bio: the bio being unmapped
342 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
345 * bio_unmap_user() may sleep.
347 static void bio_unmap_user(struct bio
*bio
)
349 bio_release_pages(bio
, bio_data_dir(bio
) == READ
);
354 static void bio_invalidate_vmalloc_pages(struct bio
*bio
)
356 #ifdef ARCH_HAS_FLUSH_KERNEL_DCACHE_PAGE
357 if (bio
->bi_private
&& !op_is_write(bio_op(bio
))) {
358 unsigned long i
, len
= 0;
360 for (i
= 0; i
< bio
->bi_vcnt
; i
++)
361 len
+= bio
->bi_io_vec
[i
].bv_len
;
362 invalidate_kernel_vmap_range(bio
->bi_private
, len
);
367 static void bio_map_kern_endio(struct bio
*bio
)
369 bio_invalidate_vmalloc_pages(bio
);
374 * bio_map_kern - map kernel address into bio
375 * @q: the struct request_queue for the bio
376 * @data: pointer to buffer to map
377 * @len: length in bytes
378 * @gfp_mask: allocation flags for bio allocation
380 * Map the kernel address into a bio suitable for io to a block
381 * device. Returns an error pointer in case of error.
383 static struct bio
*bio_map_kern(struct request_queue
*q
, void *data
,
384 unsigned int len
, gfp_t gfp_mask
)
386 unsigned long kaddr
= (unsigned long)data
;
387 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
388 unsigned long start
= kaddr
>> PAGE_SHIFT
;
389 const int nr_pages
= end
- start
;
390 bool is_vmalloc
= is_vmalloc_addr(data
);
395 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
397 return ERR_PTR(-ENOMEM
);
400 flush_kernel_vmap_range(data
, len
);
401 bio
->bi_private
= data
;
404 offset
= offset_in_page(kaddr
);
405 for (i
= 0; i
< nr_pages
; i
++) {
406 unsigned int bytes
= PAGE_SIZE
- offset
;
415 page
= virt_to_page(data
);
417 page
= vmalloc_to_page(data
);
418 if (bio_add_pc_page(q
, bio
, page
, bytes
,
420 /* we don't support partial mappings */
422 return ERR_PTR(-EINVAL
);
430 bio
->bi_end_io
= bio_map_kern_endio
;
434 static void bio_copy_kern_endio(struct bio
*bio
)
440 static void bio_copy_kern_endio_read(struct bio
*bio
)
442 char *p
= bio
->bi_private
;
443 struct bio_vec
*bvec
;
444 struct bvec_iter_all iter_all
;
446 bio_for_each_segment_all(bvec
, bio
, iter_all
) {
447 memcpy(p
, page_address(bvec
->bv_page
), bvec
->bv_len
);
451 bio_copy_kern_endio(bio
);
455 * bio_copy_kern - copy kernel address into bio
456 * @q: the struct request_queue for the bio
457 * @data: pointer to buffer to copy
458 * @len: length in bytes
459 * @gfp_mask: allocation flags for bio and page allocation
460 * @reading: data direction is READ
462 * copy the kernel address into a bio suitable for io to a block
463 * device. Returns an error pointer in case of error.
465 static struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
,
466 unsigned int len
, gfp_t gfp_mask
, int reading
)
468 unsigned long kaddr
= (unsigned long)data
;
469 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
470 unsigned long start
= kaddr
>> PAGE_SHIFT
;
479 return ERR_PTR(-EINVAL
);
481 nr_pages
= end
- start
;
482 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
484 return ERR_PTR(-ENOMEM
);
488 unsigned int bytes
= PAGE_SIZE
;
493 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
498 memcpy(page_address(page
), p
, bytes
);
500 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
508 bio
->bi_end_io
= bio_copy_kern_endio_read
;
509 bio
->bi_private
= data
;
511 bio
->bi_end_io
= bio_copy_kern_endio
;
519 return ERR_PTR(-ENOMEM
);
523 * Append a bio to a passthrough request. Only works if the bio can be merged
524 * into the request based on the driver constraints.
526 int blk_rq_append_bio(struct request
*rq
, struct bio
**bio
)
528 struct bio
*orig_bio
= *bio
;
529 struct bvec_iter iter
;
531 unsigned int nr_segs
= 0;
533 blk_queue_bounce(rq
->q
, bio
);
535 bio_for_each_bvec(bv
, *bio
, iter
)
539 blk_rq_bio_prep(rq
, *bio
, nr_segs
);
541 if (!ll_back_merge_fn(rq
, *bio
, nr_segs
)) {
542 if (orig_bio
!= *bio
) {
549 rq
->biotail
->bi_next
= *bio
;
551 rq
->__data_len
+= (*bio
)->bi_iter
.bi_size
;
556 EXPORT_SYMBOL(blk_rq_append_bio
);
558 static int __blk_rq_unmap_user(struct bio
*bio
)
563 if (bio_flagged(bio
, BIO_USER_MAPPED
))
566 ret
= bio_uncopy_user(bio
);
572 static int __blk_rq_map_user_iov(struct request
*rq
,
573 struct rq_map_data
*map_data
, struct iov_iter
*iter
,
574 gfp_t gfp_mask
, bool copy
)
576 struct request_queue
*q
= rq
->q
;
577 struct bio
*bio
, *orig_bio
;
581 bio
= bio_copy_user_iov(q
, map_data
, iter
, gfp_mask
);
583 bio
= bio_map_user_iov(q
, iter
, gfp_mask
);
588 bio
->bi_opf
&= ~REQ_OP_MASK
;
589 bio
->bi_opf
|= req_op(rq
);
594 * We link the bounce buffer in and could have to traverse it
595 * later so we have to get a ref to prevent it from being freed
597 ret
= blk_rq_append_bio(rq
, &bio
);
599 __blk_rq_unmap_user(orig_bio
);
608 * blk_rq_map_user_iov - map user data to a request, for passthrough requests
609 * @q: request queue where request should be inserted
610 * @rq: request to map data to
611 * @map_data: pointer to the rq_map_data holding pages (if necessary)
612 * @iter: iovec iterator
613 * @gfp_mask: memory allocation flags
616 * Data will be mapped directly for zero copy I/O, if possible. Otherwise
617 * a kernel bounce buffer is used.
619 * A matching blk_rq_unmap_user() must be issued at the end of I/O, while
620 * still in process context.
622 * Note: The mapped bio may need to be bounced through blk_queue_bounce()
623 * before being submitted to the device, as pages mapped may be out of
624 * reach. It's the callers responsibility to make sure this happens. The
625 * original bio must be passed back in to blk_rq_unmap_user() for proper
628 int blk_rq_map_user_iov(struct request_queue
*q
, struct request
*rq
,
629 struct rq_map_data
*map_data
,
630 const struct iov_iter
*iter
, gfp_t gfp_mask
)
633 unsigned long align
= q
->dma_pad_mask
| queue_dma_alignment(q
);
634 struct bio
*bio
= NULL
;
638 if (!iter_is_iovec(iter
))
643 else if (iov_iter_alignment(iter
) & align
)
645 else if (queue_virt_boundary(q
))
646 copy
= queue_virt_boundary(q
) & iov_iter_gap_alignment(iter
);
650 ret
=__blk_rq_map_user_iov(rq
, map_data
, &i
, gfp_mask
, copy
);
655 } while (iov_iter_count(&i
));
657 if (!bio_flagged(bio
, BIO_USER_MAPPED
))
658 rq
->rq_flags
|= RQF_COPY_USER
;
662 blk_rq_unmap_user(bio
);
667 EXPORT_SYMBOL(blk_rq_map_user_iov
);
669 int blk_rq_map_user(struct request_queue
*q
, struct request
*rq
,
670 struct rq_map_data
*map_data
, void __user
*ubuf
,
671 unsigned long len
, gfp_t gfp_mask
)
675 int ret
= import_single_range(rq_data_dir(rq
), ubuf
, len
, &iov
, &i
);
677 if (unlikely(ret
< 0))
680 return blk_rq_map_user_iov(q
, rq
, map_data
, &i
, gfp_mask
);
682 EXPORT_SYMBOL(blk_rq_map_user
);
685 * blk_rq_unmap_user - unmap a request with user data
686 * @bio: start of bio list
689 * Unmap a rq previously mapped by blk_rq_map_user(). The caller must
690 * supply the original rq->bio from the blk_rq_map_user() return, since
691 * the I/O completion may have changed rq->bio.
693 int blk_rq_unmap_user(struct bio
*bio
)
695 struct bio
*mapped_bio
;
700 if (unlikely(bio_flagged(bio
, BIO_BOUNCED
)))
701 mapped_bio
= bio
->bi_private
;
703 ret2
= __blk_rq_unmap_user(mapped_bio
);
714 EXPORT_SYMBOL(blk_rq_unmap_user
);
717 * blk_rq_map_kern - map kernel data to a request, for passthrough requests
718 * @q: request queue where request should be inserted
719 * @rq: request to fill
720 * @kbuf: the kernel buffer
721 * @len: length of user data
722 * @gfp_mask: memory allocation flags
725 * Data will be mapped directly if possible. Otherwise a bounce
726 * buffer is used. Can be called multiple times to append multiple
729 int blk_rq_map_kern(struct request_queue
*q
, struct request
*rq
, void *kbuf
,
730 unsigned int len
, gfp_t gfp_mask
)
732 int reading
= rq_data_dir(rq
) == READ
;
733 unsigned long addr
= (unsigned long) kbuf
;
735 struct bio
*bio
, *orig_bio
;
738 if (len
> (queue_max_hw_sectors(q
) << 9))
743 do_copy
= !blk_rq_aligned(q
, addr
, len
) || object_is_on_stack(kbuf
);
745 bio
= bio_copy_kern(q
, kbuf
, len
, gfp_mask
, reading
);
747 bio
= bio_map_kern(q
, kbuf
, len
, gfp_mask
);
752 bio
->bi_opf
&= ~REQ_OP_MASK
;
753 bio
->bi_opf
|= req_op(rq
);
756 rq
->rq_flags
|= RQF_COPY_USER
;
759 ret
= blk_rq_append_bio(rq
, &bio
);
761 /* request is too big */
768 EXPORT_SYMBOL(blk_rq_map_kern
);