crypto: scatterwalk - Backport memcpy_sglist()

This backports the current implementation of memcpy_sglist() from
upstream commit 4dffc9bbffb9ccfcda730d899c97c553599e7ca8.

This function was rewritten twice.  The earlier implementations had many
prerequisite commits, while the latest implementation is standalone.
It's much easier to just backport the latest code directly.

Signed-off-by: Eric Biggers <ebiggers@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/crypto/scatterwalk.c b/crypto/scatterwalk.c
index 16f6ba896fb63bdc3cc9ad1ab56881270fd2c18d..9f0b27005166ca05c70fff0f24ecaa665fa3730b 100644 (file)
--- a/crypto/scatterwalk.c
+++ b/crypto/scatterwalk.c
@@ -69,6 +69,100 @@ void scatterwalk_map_and_copy(void *buf, struct scatterlist *sg,
 }
 EXPORT_SYMBOL_GPL(scatterwalk_map_and_copy);
 
+/**
+ * memcpy_sglist() - Copy data from one scatterlist to another
+ * @dst: The destination scatterlist.  Can be NULL if @nbytes == 0.
+ * @src: The source scatterlist.  Can be NULL if @nbytes == 0.
+ * @nbytes: Number of bytes to copy
+ *
+ * The scatterlists can describe exactly the same memory, in which case this
+ * function is a no-op.  No other overlaps are supported.
+ *
+ * Context: Any context
+ */
+void memcpy_sglist(struct scatterlist *dst, struct scatterlist *src,
+                  unsigned int nbytes)
+{
+       unsigned int src_offset, dst_offset;
+
+       if (unlikely(nbytes == 0)) /* in case src and/or dst is NULL */
+               return;
+
+       src_offset = src->offset;
+       dst_offset = dst->offset;
+       for (;;) {
+               /* Compute the length to copy this step. */
+               unsigned int len = min3(src->offset + src->length - src_offset,
+                                       dst->offset + dst->length - dst_offset,
+                                       nbytes);
+               struct page *src_page = sg_page(src);
+               struct page *dst_page = sg_page(dst);
+               const void *src_virt;
+               void *dst_virt;
+
+               if (IS_ENABLED(CONFIG_HIGHMEM)) {
+                       /* HIGHMEM: we may have to actually map the pages. */
+                       const unsigned int src_oip = offset_in_page(src_offset);
+                       const unsigned int dst_oip = offset_in_page(dst_offset);
+                       const unsigned int limit = PAGE_SIZE;
+
+                       /* Further limit len to not cross a page boundary. */
+                       len = min3(len, limit - src_oip, limit - dst_oip);
+
+                       /* Compute the source and destination pages. */
+                       src_page += src_offset / PAGE_SIZE;
+                       dst_page += dst_offset / PAGE_SIZE;
+
+                       if (src_page != dst_page) {
+                               /* Copy between different pages. */
+                               memcpy_page(dst_page, dst_oip,
+                                           src_page, src_oip, len);
+                               flush_dcache_page(dst_page);
+                       } else if (src_oip != dst_oip) {
+                               /* Copy between different parts of same page. */
+                               dst_virt = kmap_local_page(dst_page);
+                               memcpy(dst_virt + dst_oip, dst_virt + src_oip,
+                                      len);
+                               kunmap_local(dst_virt);
+                               flush_dcache_page(dst_page);
+                       } /* Else, it's the same memory.  No action needed. */
+               } else {
+                       /*
+                        * !HIGHMEM: no mapping needed.  Just work in the linear
+                        * buffer of each sg entry.  Note that we can cross page
+                        * boundaries, as they are not significant in this case.
+                        */
+                       src_virt = page_address(src_page) + src_offset;
+                       dst_virt = page_address(dst_page) + dst_offset;
+                       if (src_virt != dst_virt) {
+                               memcpy(dst_virt, src_virt, len);
+                               if (ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE)
+                                       __scatterwalk_flush_dcache_pages(
+                                               dst_page, dst_offset, len);
+                       } /* Else, it's the same memory.  No action needed. */
+               }
+               nbytes -= len;
+               if (nbytes == 0) /* No more to copy? */
+                       break;
+
+               /*
+                * There's more to copy.  Advance the offsets by the length
+                * copied this step, and advance the sg entries as needed.
+                */
+               src_offset += len;
+               if (src_offset >= src->offset + src->length) {
+                       src = sg_next(src);
+                       src_offset = src->offset;
+               }
+               dst_offset += len;
+               if (dst_offset >= dst->offset + dst->length) {
+                       dst = sg_next(dst);
+                       dst_offset = dst->offset;
+               }
+       }
+}
+EXPORT_SYMBOL_GPL(memcpy_sglist);
+
 struct scatterlist *scatterwalk_ffwd(struct scatterlist dst[2],
                                     struct scatterlist *src,
                                     unsigned int len)

diff --git a/include/crypto/scatterwalk.h b/include/crypto/scatterwalk.h
index ccdb05f68a75c3a2e6465ccf21390ac94e886c7f..92ecd4ffbd07775163815266fc29d216257e0675 100644 (file)
--- a/include/crypto/scatterwalk.h
+++ b/include/crypto/scatterwalk.h
@@ -83,6 +83,34 @@ static inline void scatterwalk_pagedone(struct scatter_walk *walk, int out,
                scatterwalk_start(walk, sg_next(walk->sg));
 }
 
+/*
+ * Flush the dcache of any pages that overlap the region
+ * [offset, offset + nbytes) relative to base_page.
+ *
+ * This should be called only when ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE, to ensure
+ * that all relevant code (including the call to sg_page() in the caller, if
+ * applicable) gets fully optimized out when !ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE.
+ */
+static inline void __scatterwalk_flush_dcache_pages(struct page *base_page,
+                                                   unsigned int offset,
+                                                   unsigned int nbytes)
+{
+       unsigned int num_pages;
+
+       base_page += offset / PAGE_SIZE;
+       offset %= PAGE_SIZE;
+
+       /*
+        * This is an overflow-safe version of
+        * num_pages = DIV_ROUND_UP(offset + nbytes, PAGE_SIZE).
+        */
+       num_pages = nbytes / PAGE_SIZE;
+       num_pages += DIV_ROUND_UP(offset + (nbytes % PAGE_SIZE), PAGE_SIZE);
+
+       for (unsigned int i = 0; i < num_pages; i++)
+               flush_dcache_page(base_page + i);
+}
+
 static inline void scatterwalk_done(struct scatter_walk *walk, int out,
                                    int more)
 {
@@ -95,6 +123,9 @@ void scatterwalk_copychunks(void *buf, struct scatter_walk *walk,
                            size_t nbytes, int out);
 void *scatterwalk_map(struct scatter_walk *walk);
 
+void memcpy_sglist(struct scatterlist *dst, struct scatterlist *src,
+                  unsigned int nbytes);
+
 void scatterwalk_map_and_copy(void *buf, struct scatterlist *sg,
                              unsigned int start, unsigned int nbytes, int out);