From: Chuck Lever <cel@kernel.org>
To: NeilBrown <neil@brown.name>, Jeff Layton <jlayton@kernel.org>,
Olga Kornievskaia <okorniev@redhat.com>,
Dai Ngo <dai.ngo@oracle.com>, Tom Talpey <tom@talpey.com>
Cc: <linux-nfs@vger.kernel.org>, Christoph Hellwig <hch@lst.de>,
Chuck Lever <chuck.lever@oracle.com>
Subject: [PATCH v8 09/12] NFSD: Handle both offset and memory alignment for direct I/O
Date: Mon, 27 Oct 2025 11:46:27 -0400 [thread overview]
Message-ID: <20251027154630.1774-10-cel@kernel.org> (raw)
In-Reply-To: <20251027154630.1774-1-cel@kernel.org>
From: Chuck Lever <chuck.lever@oracle.com>
Currently, nfsd_is_write_dio_possible() only considers file offset
alignment (nf_dio_offset_align) when splitting an NFS WRITE request
into segments. This leaves accounting for memory buffer alignment
(nf_dio_mem_align) until nfsd_setup_write_dio_iters(). If this
second check fails, the code falls back to cached I/O entirely,
wasting the opportunity to use direct I/O for the bulk of the
request.
Enhance the logic to find a beginning segment size that satisfies
both alignment constraints simultaneously. The search algorithm
starts with the file offset alignment requirement and steps through
multiples of offset_align, checking memory alignment at each step.
The search is bounded by lcm(offset_align, mem_align) to ensure that
it always terminates.
Signed-off-by: Chuck Lever <cel@kernel.org>
---
fs/nfsd/filecache.c | 5 ++
fs/nfsd/filecache.h | 1 +
fs/nfsd/vfs.c | 119 +++++++++++++++++++++++++++++---------------
3 files changed, 86 insertions(+), 39 deletions(-)
diff --git a/fs/nfsd/filecache.c b/fs/nfsd/filecache.c
index a238b6725008..89adc4ab5b24 100644
--- a/fs/nfsd/filecache.c
+++ b/fs/nfsd/filecache.c
@@ -40,6 +40,7 @@
#include <linux/seq_file.h>
#include <linux/rhashtable.h>
#include <linux/nfslocalio.h>
+#include <linux/lcm.h>
#include "vfs.h"
#include "nfsd.h"
@@ -234,6 +235,7 @@ nfsd_file_alloc(struct net *net, struct inode *inode, unsigned char need,
nf->nf_dio_mem_align = 0;
nf->nf_dio_offset_align = 0;
nf->nf_dio_read_offset_align = 0;
+ nf->nf_dio_align_lcm = 0;
return nf;
}
@@ -1071,6 +1073,9 @@ nfsd_file_get_dio_attrs(const struct svc_fh *fhp, struct nfsd_file *nf)
if (stat.result_mask & STATX_DIOALIGN) {
nf->nf_dio_mem_align = stat.dio_mem_align;
nf->nf_dio_offset_align = stat.dio_offset_align;
+ if (stat.dio_mem_align && stat.dio_offset_align)
+ nf->nf_dio_align_lcm = lcm(stat.dio_mem_align,
+ stat.dio_offset_align);
}
if (stat.result_mask & STATX_DIO_READ_ALIGN)
nf->nf_dio_read_offset_align = stat.dio_read_offset_align;
diff --git a/fs/nfsd/filecache.h b/fs/nfsd/filecache.h
index e3d6ca2b6030..2648aaab5a1b 100644
--- a/fs/nfsd/filecache.h
+++ b/fs/nfsd/filecache.h
@@ -58,6 +58,7 @@ struct nfsd_file {
u32 nf_dio_mem_align;
u32 nf_dio_offset_align;
u32 nf_dio_read_offset_align;
+ unsigned long nf_dio_align_lcm;
};
int nfsd_file_cache_init(void);
diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
index 37353fb48d58..a872be300c9f 100644
--- a/fs/nfsd/vfs.c
+++ b/fs/nfsd/vfs.c
@@ -1261,49 +1261,73 @@ struct nfsd_write_dio_seg {
struct nfsd_write_dio_args {
struct nfsd_file *nf;
- size_t first, middle, last;
unsigned int nsegs;
struct nfsd_write_dio_seg segment[3];
};
+/*
+ * Find the minimum offset within the write request that aligns both
+ * the file offset and memory buffer for direct I/O.
+ *
+ * Returns the size of the unaligned prefix, or SIZE_MAX if no alignment
+ * is possible within reasonable bounds.
+ */
+static size_t
+nfsd_find_dio_aligned_offset(struct nfsd_file *nf, loff_t file_offset,
+ unsigned long mem_offset, size_t total_len)
+{
+ u32 offset_align = nf->nf_dio_offset_align;
+ u32 mem_align = nf->nf_dio_mem_align;
+ unsigned long search_limit;
+ size_t first;
+
+ /* Start with the file offset alignment requirement */
+ first = round_up(file_offset, offset_align) - file_offset;
+
+ /* Quick check: does this also satisfy memory alignment? */
+ if (((mem_offset + first) & (mem_align - 1)) == 0)
+ return first;
+
+ /*
+ * Search for a value that satisfies both constraints by stepping
+ * through multiples of offset_align. Limit search to one period
+ * of the LCM. We need to check up through the search_limit to
+ * cover all possible alignments within the LCM period.
+ */
+ search_limit = min_t(unsigned long, nf->nf_dio_align_lcm, total_len);
+
+ for (; first <= search_limit && first < total_len; first += offset_align) {
+ if (((mem_offset + first) & (mem_align - 1)) == 0)
+ return first;
+ }
+
+ return SIZE_MAX; /* No alignment found */
+}
+
+/*
+ * Check if the underlying file system implements direct I/O.
+ */
static bool
nfsd_is_write_dio_possible(loff_t offset, unsigned long len,
struct nfsd_write_dio_args *args)
{
- u32 dio_blocksize = args->nf->nf_dio_offset_align;
- loff_t first_end, orig_end, middle_end;
+ u32 offset_align = args->nf->nf_dio_offset_align;
+ u32 mem_align = args->nf->nf_dio_mem_align;
- if (unlikely(!args->nf->nf_dio_mem_align || !dio_blocksize))
- return false;
- if (unlikely(len < dio_blocksize))
+ if (unlikely(!mem_align || !offset_align))
return false;
- first_end = round_up(offset, dio_blocksize);
- orig_end = offset + len;
- middle_end = round_down(orig_end, dio_blocksize);
+ /*
+ * Need enough data to potentially find an aligned segment.
+ * In the worst case, we might need up to
+ * lcm(offset_align, mem_align) bytes for the prefix.
+ */
+ if (unlikely(len < max(offset_align, mem_align)))
+ return false;
- args->first = first_end - offset;
- args->middle = middle_end - first_end;
- args->last = orig_end - middle_end;
return true;
}
-/*
- * Check if the bvec iterator is aligned for direct I/O.
- *
- * bvecs generated from RPC receive buffers are contiguous: After the first
- * bvec, all subsequent bvecs start at bv_offset zero (page-aligned).
- * Therefore, only the first bvec is checked.
- */
-static bool
-nfsd_iov_iter_aligned_bvec(const struct nfsd_file *nf, const struct iov_iter *i)
-{
- unsigned int addr_mask = nf->nf_dio_mem_align - 1;
- const struct bio_vec *bvec = i->bvec;
-
- return !((unsigned long)(bvec->bv_offset + i->iov_offset) & addr_mask);
-}
-
static void
nfsd_write_dio_seg_init(struct nfsd_write_dio_seg *segment,
struct bio_vec *bvec, unsigned int nvecs,
@@ -1318,29 +1342,45 @@ nfsd_write_dio_seg_init(struct nfsd_write_dio_seg *segment,
static bool
nfsd_setup_write_dio_iters(struct bio_vec *bvec, unsigned int nvecs,
- unsigned long total,
+ loff_t offset, unsigned long total,
struct nfsd_write_dio_args *args)
{
+ u32 offset_align = args->nf->nf_dio_offset_align;
+ unsigned long mem_offset = bvec->bv_offset;
+ loff_t prefix_end, orig_end, middle_end;
+ size_t prefix, middle, suffix;
+
args->nsegs = 0;
- if (args->first) {
+ prefix = nfsd_find_dio_aligned_offset(args->nf, offset, mem_offset,
+ total);
+ if (prefix == SIZE_MAX)
+ return false; /* No alignment possible */
+
+ prefix_end = offset + prefix;
+ orig_end = offset + total;
+ middle_end = round_down(orig_end, offset_align);
+
+ middle = middle_end - prefix_end;
+ suffix = orig_end - middle_end;
+
+ if (prefix) {
nfsd_write_dio_seg_init(&args->segment[args->nsegs], bvec,
- nvecs, total, 0, args->first);
+ nvecs, total, 0, prefix);
++args->nsegs;
}
+ if (!middle)
+ return false; /* No aligned region for DIO */
+
nfsd_write_dio_seg_init(&args->segment[args->nsegs], bvec, nvecs,
- total, args->first, args->middle);
- if (!nfsd_iov_iter_aligned_bvec(args->nf,
- &args->segment[args->nsegs].iter))
- return false; /* no DIO-aligned IO possible */
+ total, prefix, middle);
args->segment[args->nsegs].use_dio = true;
++args->nsegs;
- if (args->last) {
+ if (suffix) {
nfsd_write_dio_seg_init(&args->segment[args->nsegs], bvec,
- nvecs, total, args->first +
- args->middle, args->last);
+ nvecs, total, prefix + middle, suffix);
++args->nsegs;
}
@@ -1373,7 +1413,8 @@ nfsd_issue_write_dio(struct svc_rqst *rqstp, struct svc_fh *fhp, u32 *stable_how
ssize_t host_err;
unsigned int i;
- if (!nfsd_setup_write_dio_iters(rqstp->rq_bvec, nvecs, *cnt, args))
+ if (!nfsd_setup_write_dio_iters(rqstp->rq_bvec, nvecs, kiocb->ki_pos,
+ *cnt, args))
return nfsd_buffered_write(rqstp, file, nvecs, cnt, kiocb);
/*
--
2.51.0
next prev parent reply other threads:[~2025-10-27 15:46 UTC|newest]
Thread overview: 30+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-10-27 15:46 [PATCH v8 00/12] NFSD: Implement NFSD_IO_DIRECT for NFS WRITE Chuck Lever
2025-10-27 15:46 ` [PATCH v8 01/12] NFSD: Make FILE_SYNC WRITEs comply with spec Chuck Lever
2025-10-27 15:46 ` [PATCH v8 02/12] NFSD: Enable return of an updated stable_how to NFS clients Chuck Lever
2025-10-27 15:46 ` [PATCH v8 03/12] NFSD: Implement NFSD_IO_DIRECT for NFS WRITE Chuck Lever
2025-10-28 22:56 ` kernel test robot
2025-10-29 14:10 ` Chuck Lever
2025-10-30 1:29 ` Philip Li
2025-10-27 15:46 ` [PATCH v8 04/12] NFSD: Remove specific error handling Chuck Lever
2025-10-27 15:46 ` [PATCH v8 05/12] NFSD: Remove alignment size checking Chuck Lever
2025-10-27 15:46 ` [PATCH v8 06/12] NFSD: Clean up struct nfsd_write_dio Chuck Lever
2025-10-27 15:46 ` [PATCH v8 07/12] NFSD: Introduce struct nfsd_write_dio_seg Chuck Lever
2025-10-27 15:46 ` [PATCH v8 08/12] NFSD: Simplify nfsd_iov_iter_aligned_bvec() Chuck Lever
2025-10-30 15:00 ` Jeff Layton
2025-10-31 13:16 ` Christoph Hellwig
2025-10-27 15:46 ` Chuck Lever [this message]
2025-10-30 19:52 ` [PATCH v8 09/12] NFSD: Handle both offset and memory alignment for direct I/O Jeff Layton
2025-10-30 19:55 ` Chuck Lever
2025-10-31 9:13 ` Christoph Hellwig
2025-10-31 13:19 ` Christoph Hellwig
2025-10-31 13:21 ` Chuck Lever
2025-10-31 13:23 ` Christoph Hellwig
2025-10-31 16:07 ` Mike Snitzer
2025-10-27 15:46 ` [PATCH v8 10/12] NFSD: Combine direct I/O feasibility check with iterator setup Chuck Lever
2025-10-30 19:59 ` Jeff Layton
2025-10-31 13:20 ` Christoph Hellwig
2025-10-27 15:46 ` [PATCH v8 11/12] NFSD: Handle kiocb->ki_flags correctly Chuck Lever
2025-10-30 20:01 ` Jeff Layton
2025-10-31 13:21 ` Christoph Hellwig
2025-10-27 15:46 ` [PATCH v8 12/12] NFSD: Refactor nfsd_vfs_write Chuck Lever
2025-10-30 20:02 ` Jeff Layton
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20251027154630.1774-10-cel@kernel.org \
--to=cel@kernel.org \
--cc=chuck.lever@oracle.com \
--cc=dai.ngo@oracle.com \
--cc=hch@lst.de \
--cc=jlayton@kernel.org \
--cc=linux-nfs@vger.kernel.org \
--cc=neil@brown.name \
--cc=okorniev@redhat.com \
--cc=tom@talpey.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.