[PATCH 1/7] block: Add blk_rq_map_kern_sg()

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



This patch implements function blk_rq_map_kern_sg(), which allows to
map a kernel-originated SG vector to a block request. It is necessary
to execute SCSI commands with from the kernel going SG buffer. SCST needs
this functionality, because its target drivers, which are, basically,
SCSI drivers, can deal only with SGs, not with BIOs.

Highlights of this implementations:

 - It uses BIOs chaining instead of kmalloc()'ing the whole bio.

 - It uses SGs chaining instead of kmalloc()'ing one big SG in case if
direct mapping failed (e.g. because of DMA alignment or padding).

 - When needed, copy_page() is used instead of memcpy() to copy the
whole pages faster.

Also this patch adds and exports function sg_copy(), which copies
one SG vector to another.

This patch has already been reviewed in LKML/linux-scsi in
http://lkml.org/lkml/2009/8/12/304. Since then it was only updated for
2.6.37.

Signed-off-by: Vladislav Bolkhovitin <vst@xxxxxxxx>
Cc: Jens Axboe <jaxboe@xxxxxxxxxxxx>
---
 block/blk-map.c             |  333 +++++++++++++++++++++++++++++++++++++++++++
 include/linux/blkdev.h      |    3 +
 include/linux/scatterlist.h |    5 +
 lib/scatterlist.c           |  129 +++++++++++++++++
 4 files changed, 470 insertions(+), 0 deletions(-)

diff --git a/block/blk-map.c b/block/blk-map.c
index 5d5dbe4..bc036bc 100644
--- a/block/blk-map.c
+++ b/block/blk-map.c
@@ -5,6 +5,8 @@
 #include <linux/module.h>
 #include <linux/bio.h>
 #include <linux/blkdev.h>
+#include <linux/scatterlist.h>
+#include <linux/slab.h>
 #include <scsi/sg.h>		/* for struct sg_iovec */
 
 #include "blk.h"
@@ -273,6 +275,337 @@ int blk_rq_unmap_user(struct bio *bio)
 }
 EXPORT_SYMBOL(blk_rq_unmap_user);
 
+struct blk_kern_sg_work {
+	atomic_t bios_inflight;
+	struct sg_table sg_table;
+	struct scatterlist *src_sgl;
+};
+
+static void blk_free_kern_sg_work(struct blk_kern_sg_work *bw)
+{
+	sg_free_table(&bw->sg_table);
+	kfree(bw);
+	return;
+}
+
+static void blk_bio_map_kern_endio(struct bio *bio, int err)
+{
+	struct blk_kern_sg_work *bw = bio->bi_private;
+
+	if (bw != NULL) {
+		/* Decrement the bios in processing and, if zero, free */
+		BUG_ON(atomic_read(&bw->bios_inflight) <= 0);
+		if (atomic_dec_and_test(&bw->bios_inflight)) {
+			if ((bio_data_dir(bio) == READ) && (err == 0)) {
+				unsigned long flags;
+
+				local_irq_save(flags);	/* to protect KMs */
+				sg_copy(bw->src_sgl, bw->sg_table.sgl, 0, 0,
+					KM_BIO_DST_IRQ, KM_BIO_SRC_IRQ);
+				local_irq_restore(flags);
+			}
+			blk_free_kern_sg_work(bw);
+		}
+	}
+
+	bio_put(bio);
+	return;
+}
+
+static int blk_rq_copy_kern_sg(struct request *rq, struct scatterlist *sgl,
+			       int nents, struct blk_kern_sg_work **pbw,
+			       gfp_t gfp, gfp_t page_gfp)
+{
+	int res = 0, i;
+	struct scatterlist *sg;
+	struct scatterlist *new_sgl;
+	int new_sgl_nents;
+	size_t len = 0, to_copy;
+	struct blk_kern_sg_work *bw;
+
+	bw = kzalloc(sizeof(*bw), gfp);
+	if (bw == NULL)
+		goto out;
+
+	bw->src_sgl = sgl;
+
+	for_each_sg(sgl, sg, nents, i)
+		len += sg->length;
+	to_copy = len;
+
+	new_sgl_nents = PFN_UP(len);
+
+	res = sg_alloc_table(&bw->sg_table, new_sgl_nents, gfp);
+	if (res != 0)
+		goto out_free_bw;
+
+	new_sgl = bw->sg_table.sgl;
+
+	for_each_sg(new_sgl, sg, new_sgl_nents, i) {
+		struct page *pg;
+
+		pg = alloc_page(page_gfp);
+		if (pg == NULL)
+			goto err_free_new_sgl;
+
+		sg_assign_page(sg, pg);
+		sg->length = min_t(size_t, PAGE_SIZE, len);
+
+		len -= PAGE_SIZE;
+	}
+
+	if (rq_data_dir(rq) == WRITE) {
+		/*
+		 * We need to limit amount of copied data to to_copy, because
+		 * sgl might have the last element in sgl not marked as last in
+		 * SG chaining.
+		 */
+		sg_copy(new_sgl, sgl, 0, to_copy,
+			KM_USER0, KM_USER1);
+	}
+
+	*pbw = bw;
+	/*
+	 * REQ_COPY_USER name is misleading. It should be something like
+	 * REQ_HAS_TAIL_SPACE_FOR_PADDING.
+	 */
+	rq->cmd_flags |= REQ_COPY_USER;
+
+out:
+	return res;
+
+err_free_new_sgl:
+	for_each_sg(new_sgl, sg, new_sgl_nents, i) {
+		struct page *pg = sg_page(sg);
+		if (pg == NULL)
+			break;
+		__free_page(pg);
+	}
+	sg_free_table(&bw->sg_table);
+
+out_free_bw:
+	kfree(bw);
+	res = -ENOMEM;
+	goto out;
+}
+
+static int __blk_rq_map_kern_sg(struct request *rq, struct scatterlist *sgl,
+	int nents, struct blk_kern_sg_work *bw, gfp_t gfp)
+{
+	int res;
+	struct request_queue *q = rq->q;
+	int rw = rq_data_dir(rq);
+	int max_nr_vecs, i;
+	size_t tot_len;
+	bool need_new_bio;
+	struct scatterlist *sg, *prev_sg = NULL;
+	struct bio *bio = NULL, *hbio = NULL, *tbio = NULL;
+	int bios;
+
+	if (unlikely((sgl == NULL) || (sgl->length == 0) || (nents <= 0))) {
+		WARN_ON(1);
+		res = -EINVAL;
+		goto out;
+	}
+
+	/*
+	 * Let's keep each bio allocation inside a single page to decrease
+	 * probability of failure.
+	 */
+	max_nr_vecs =  min_t(size_t,
+		((PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec)),
+		BIO_MAX_PAGES);
+
+	need_new_bio = true;
+	tot_len = 0;
+	bios = 0;
+	for_each_sg(sgl, sg, nents, i) {
+		struct page *page = sg_page(sg);
+		void *page_addr = page_address(page);
+		size_t len = sg->length, l;
+		size_t offset = sg->offset;
+
+		tot_len += len;
+		prev_sg = sg;
+
+		/*
+		 * Each segment must be aligned on DMA boundary and
+		 * not on stack. The last one may have unaligned
+		 * length as long as the total length is aligned to
+		 * DMA padding alignment.
+		 */
+		if (i == nents - 1)
+			l = 0;
+		else
+			l = len;
+		if (((sg->offset | l) & queue_dma_alignment(q)) ||
+		    (page_addr && object_is_on_stack(page_addr + sg->offset))) {
+			res = -EINVAL;
+			goto out_free_bios;
+		}
+
+		while (len > 0) {
+			size_t bytes;
+			int rc;
+
+			if (need_new_bio) {
+				bio = bio_kmalloc(gfp, max_nr_vecs);
+				if (bio == NULL) {
+					res = -ENOMEM;
+					goto out_free_bios;
+				}
+
+				if (rw == WRITE)
+					bio->bi_rw |= REQ_WRITE;
+
+				bios++;
+				bio->bi_private = bw;
+				bio->bi_end_io = blk_bio_map_kern_endio;
+
+				if (hbio == NULL)
+					hbio = tbio = bio;
+				else
+					tbio = tbio->bi_next = bio;
+			}
+
+			bytes = min_t(size_t, len, PAGE_SIZE - offset);
+
+			rc = bio_add_pc_page(q, bio, page, bytes, offset);
+			if (rc < bytes) {
+				if (unlikely(need_new_bio || (rc < 0))) {
+					if (rc < 0)
+						res = rc;
+					else
+						res = -EIO;
+					goto out_free_bios;
+				} else {
+					need_new_bio = true;
+					len -= rc;
+					offset += rc;
+					continue;
+				}
+			}
+
+			need_new_bio = false;
+			offset = 0;
+			len -= bytes;
+			page = nth_page(page, 1);
+		}
+	}
+
+	if (hbio == NULL) {
+		res = -EINVAL;
+		goto out_free_bios;
+	}
+
+	/* Total length must be aligned on DMA padding alignment */
+	if ((tot_len & q->dma_pad_mask) &&
+	    !(rq->cmd_flags & REQ_COPY_USER)) {
+		res = -EINVAL;
+		goto out_free_bios;
+	}
+
+	if (bw != NULL)
+		atomic_set(&bw->bios_inflight, bios);
+
+	while (hbio != NULL) {
+		bio = hbio;
+		hbio = hbio->bi_next;
+		bio->bi_next = NULL;
+
+		blk_queue_bounce(q, &bio);
+
+		res = blk_rq_append_bio(q, rq, bio);
+		if (unlikely(res != 0)) {
+			bio->bi_next = hbio;
+			hbio = bio;
+			/* We can have one or more bios bounced */
+			goto out_unmap_bios;
+		}
+	}
+
+	res = 0;
+
+	rq->buffer = NULL;
+out:
+	return res;
+
+out_unmap_bios:
+	blk_rq_unmap_kern_sg(rq, res);
+
+out_free_bios:
+	while (hbio != NULL) {
+		bio = hbio;
+		hbio = hbio->bi_next;
+		bio_put(bio);
+	}
+	goto out;
+}
+
+/**
+ * blk_rq_map_kern_sg - map kernel data to a request, for REQ_TYPE_BLOCK_PC
+ * @rq:		request to fill
+ * @sgl:	area to map
+ * @nents:	number of elements in @sgl
+ * @gfp:	memory allocation flags
+ *
+ * Description:
+ *    Data will be mapped directly if possible. Otherwise a bounce
+ *    buffer will be used.
+ */
+int blk_rq_map_kern_sg(struct request *rq, struct scatterlist *sgl,
+		       int nents, gfp_t gfp)
+{
+	int res;
+
+	res = __blk_rq_map_kern_sg(rq, sgl, nents, NULL, gfp);
+	if (unlikely(res != 0)) {
+		struct blk_kern_sg_work *bw = NULL;
+
+		res = blk_rq_copy_kern_sg(rq, sgl, nents, &bw,
+				gfp, rq->q->bounce_gfp | gfp);
+		if (unlikely(res != 0))
+			goto out;
+
+		res = __blk_rq_map_kern_sg(rq, bw->sg_table.sgl,
+				bw->sg_table.nents, bw, gfp);
+		if (res != 0) {
+			blk_free_kern_sg_work(bw);
+			goto out;
+		}
+	}
+
+	rq->buffer = NULL;
+
+out:
+	return res;
+}
+EXPORT_SYMBOL(blk_rq_map_kern_sg);
+
+/**
+ * blk_rq_unmap_kern_sg - unmap a request with kernel sg
+ * @rq:		request to unmap
+ * @err:	non-zero error code
+ *
+ * Description:
+ *    Unmap a rq previously mapped by blk_rq_map_kern_sg(). Must be called
+ *    only in case of an error!
+ */
+void blk_rq_unmap_kern_sg(struct request *rq, int err)
+{
+	struct bio *bio = rq->bio;
+
+	while (bio) {
+		struct bio *b = bio;
+		bio = bio->bi_next;
+		b->bi_end_io(b, err);
+	}
+	rq->bio = NULL;
+
+	return;
+}
+EXPORT_SYMBOL(blk_rq_unmap_kern_sg);
+
 /**
  * blk_rq_map_kern - map kernel data to a request, for REQ_TYPE_BLOCK_PC usage
  * @q:		request queue where request should be inserted
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index aae86fd..9965320 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -706,6 +706,9 @@ extern int blk_rq_map_kern(struct request_queue *, struct request *, void *, uns
 extern int blk_rq_map_user_iov(struct request_queue *, struct request *,
 			       struct rq_map_data *, struct sg_iovec *, int,
 			       unsigned int, gfp_t);
+extern int blk_rq_map_kern_sg(struct request *rq, struct scatterlist *sgl,
+			      int nents, gfp_t gfp);
+extern void blk_rq_unmap_kern_sg(struct request *rq, int err);
 extern int blk_execute_rq(struct request_queue *, struct gendisk *,
 			  struct request *, int);
 extern void blk_execute_rq_nowait(struct request_queue *, struct gendisk *,
diff --git a/include/linux/scatterlist.h b/include/linux/scatterlist.h
index 9aaf5bf..0177996 100644
--- a/include/linux/scatterlist.h
+++ b/include/linux/scatterlist.h
@@ -3,6 +3,7 @@
 
 #include <asm/types.h>
 #include <asm/scatterlist.h>
+#include <asm/kmap_types.h>
 #include <linux/mm.h>
 #include <linux/string.h>
 #include <asm/io.h>
@@ -218,6 +219,10 @@ size_t sg_copy_from_buffer(struct scatterlist *sgl, unsigned int nents,
 size_t sg_copy_to_buffer(struct scatterlist *sgl, unsigned int nents,
 			 void *buf, size_t buflen);
 
+int sg_copy(struct scatterlist *dst_sg, struct scatterlist *src_sg,
+	    int nents_to_copy, size_t copy_len,
+	    enum km_type d_km_type, enum km_type s_km_type);
+
 /*
  * Maximum number of entries that will be allocated in one piece, if
  * a list larger than this is required then chaining will be utilized.
diff --git a/lib/scatterlist.c b/lib/scatterlist.c
index 4ceb05d..caf58c4 100644
--- a/lib/scatterlist.c
+++ b/lib/scatterlist.c
@@ -517,3 +517,132 @@ size_t sg_copy_to_buffer(struct scatterlist *sgl, unsigned int nents,
 	return sg_copy_buffer(sgl, nents, buf, buflen, 1);
 }
 EXPORT_SYMBOL(sg_copy_to_buffer);
+
+/*
+ * Can switch to the next dst_sg element, so, to copy to strictly only
+ * one dst_sg element, it must be either last in the chain, or
+ * copy_len == dst_sg->length.
+ */
+static int sg_copy_elem(struct scatterlist **pdst_sg, size_t *pdst_len,
+			size_t *pdst_offs, struct scatterlist *src_sg,
+			size_t copy_len,
+			enum km_type d_km_type, enum km_type s_km_type)
+{
+	int res = 0;
+	struct scatterlist *dst_sg;
+	size_t src_len, dst_len, src_offs, dst_offs;
+	struct page *src_page, *dst_page;
+
+	dst_sg = *pdst_sg;
+	dst_len = *pdst_len;
+	dst_offs = *pdst_offs;
+	dst_page = sg_page(dst_sg);
+
+	src_page = sg_page(src_sg);
+	src_len = src_sg->length;
+	src_offs = src_sg->offset;
+
+	do {
+		void *saddr, *daddr;
+		size_t n;
+
+		saddr = kmap_atomic(src_page +
+					 (src_offs >> PAGE_SHIFT), s_km_type) +
+				    (src_offs & ~PAGE_MASK);
+		daddr = kmap_atomic(dst_page +
+					(dst_offs >> PAGE_SHIFT), d_km_type) +
+				    (dst_offs & ~PAGE_MASK);
+
+		if (((src_offs & ~PAGE_MASK) == 0) &&
+		    ((dst_offs & ~PAGE_MASK) == 0) &&
+		    (src_len >= PAGE_SIZE) && (dst_len >= PAGE_SIZE) &&
+		    (copy_len >= PAGE_SIZE)) {
+			copy_page(daddr, saddr);
+			n = PAGE_SIZE;
+		} else {
+			n = min_t(size_t, PAGE_SIZE - (dst_offs & ~PAGE_MASK),
+					  PAGE_SIZE - (src_offs & ~PAGE_MASK));
+			n = min(n, src_len);
+			n = min(n, dst_len);
+			n = min_t(size_t, n, copy_len);
+			memcpy(daddr, saddr, n);
+		}
+		dst_offs += n;
+		src_offs += n;
+
+		kunmap_atomic(saddr, s_km_type);
+		kunmap_atomic(daddr, d_km_type);
+
+		res += n;
+		copy_len -= n;
+		if (copy_len == 0)
+			goto out;
+
+		src_len -= n;
+		dst_len -= n;
+		if (dst_len == 0) {
+			dst_sg = sg_next(dst_sg);
+			if (dst_sg == NULL)
+				goto out;
+			dst_page = sg_page(dst_sg);
+			dst_len = dst_sg->length;
+			dst_offs = dst_sg->offset;
+		}
+	} while (src_len > 0);
+
+out:
+	*pdst_sg = dst_sg;
+	*pdst_len = dst_len;
+	*pdst_offs = dst_offs;
+	return res;
+}
+
+/**
+ * sg_copy - copy one SG vector to another
+ * @dst_sg:	destination SG
+ * @src_sg:	source SG
+ * @nents_to_copy: maximum number of entries to copy
+ * @copy_len:	maximum amount of data to copy. If 0, then copy all.
+ * @d_km_type:	kmap_atomic type for the destination SG
+ * @s_km_type:	kmap_atomic type for the source SG
+ *
+ * Description:
+ *    Data from the source SG vector will be copied to the destination SG
+ *    vector. End of the vectors will be determined by sg_next() returning
+ *    NULL. Returns number of bytes copied.
+ */
+int sg_copy(struct scatterlist *dst_sg, struct scatterlist *src_sg,
+	    int nents_to_copy, size_t copy_len,
+	    enum km_type d_km_type, enum km_type s_km_type)
+{
+	int res = 0;
+	size_t dst_len, dst_offs;
+
+	if (copy_len == 0)
+		copy_len = 0x7FFFFFFF; /* copy all */
+
+	if (nents_to_copy == 0)
+		nents_to_copy = 0x7FFFFFFF; /* copy all */
+
+	dst_len = dst_sg->length;
+	dst_offs = dst_sg->offset;
+
+	do {
+		int copied = sg_copy_elem(&dst_sg, &dst_len, &dst_offs,
+				src_sg, copy_len, d_km_type, s_km_type);
+		copy_len -= copied;
+		res += copied;
+		if ((copy_len == 0) || (dst_sg == NULL))
+			goto out;
+
+		nents_to_copy--;
+		if (nents_to_copy == 0)
+			goto out;
+
+		src_sg = sg_next(src_sg);
+	} while (src_sg != NULL);
+
+out:
+	return res;
+}
+EXPORT_SYMBOL(sg_copy);
-- 
1.7.1

--
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Index of Archives]     [SCSI Target Devel]     [Linux SCSI Target Infrastructure]     [Kernel Newbies]     [IDE]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux ATA RAID]     [Linux IIO]     [Samba]     [Device Mapper]
  Powered by Linux