[PATCH] use kmap_atomic instaed of kmap in NFS client

Chuck Lever (cel@citi.umich.edu)
Wed, 18 Dec 2002 18:19:43 -0500 (EST)

Messages sorted by: [ date ][ thread ][ subject ][ author ]
Next message: Chuck Lever: "[PATCH] give NFS client a "set_page_dirty" address space op."
Previous message: Greg KH: "Re: converting cap_set_pg() to for_each_task_pid()"

Description:
andrew morton suggested there are places in the NFS client that could
make use of kmap_atomic instead of vanilla kmap in order to improve
scalability on 8-way and higher SMP systems.

Apply Against:
2.5.52

Test status:
Passes all Connectathon '02 tests with v2 and v3, UDP and TCP; passes
NFS torture tests on a UP HIGHMEM x86 system.

diff -Naurp 00-stock/fs/nfs/dir.c 01-kmap_atomic/fs/nfs/dir.c
--- 00-stock/fs/nfs/dir.c Sun Dec 15 21:07:57 2002
+++ 01-kmap_atomic/fs/nfs/dir.c Wed Dec 18 17:22:06 2002
@@ -208,7 +208,7 @@ int find_dirent_page(nfs_readdir_descrip

/* NOTE: Someone else may have changed the READDIRPLUS flag */
desc->page = page;
- desc->ptr = kmap(page);
+ desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
status = find_dirent(desc, page);
if (status < 0)
dir_page_release(desc);
@@ -345,7 +345,7 @@ int uncached_readdir(nfs_readdir_descrip
NFS_SERVER(inode)->dtsize,
desc->plus);
desc->page = page;
- desc->ptr = kmap(page);
+ desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
if (desc->error >= 0) {
if ((status = dir_decode(desc)) == 0)
desc->entry->prev_cookie = desc->target;
@@ -717,9 +717,9 @@ int nfs_cached_lookup(struct inode *dir,

res = -EIO;
if (PageUptodate(page)) {
- desc.ptr = kmap(page);
+ desc.ptr = kmap_atomic(page, KM_USER0);
res = find_dirent_name(&desc, page, dentry);
- kunmap(page);
+ kunmap_atomic(desc.ptr, KM_USER0);
}
page_cache_release(page);

diff -Naurp 00-stock/fs/nfs/nfs2xdr.c 01-kmap_atomic/fs/nfs/nfs2xdr.c
--- 00-stock/fs/nfs/nfs2xdr.c Sun Dec 15 21:07:53 2002
+++ 01-kmap_atomic/fs/nfs/nfs2xdr.c Wed Dec 18 17:22:06 2002
@@ -378,7 +378,7 @@ nfs_xdr_readdirres(struct rpc_rqst *req,
int hdrlen, recvd;
int status, nr;
unsigned int len, pglen;
- u32 *end, *entry;
+ u32 *end, *entry, *kaddr;

if ((status = ntohl(*p++)))
return -nfs_stat_to_errno(status);
@@ -398,7 +398,7 @@ nfs_xdr_readdirres(struct rpc_rqst *req,
if (pglen > recvd)
pglen = recvd;
page = rcvbuf->pages;
- p = kmap(*page);
+ kaddr = p = (u32 *)kmap_atomic(*page, KM_USER0);
end = (u32 *)((char *)p + pglen);
entry = p;
for (nr = 0; *p++; nr++) {
@@ -419,7 +419,7 @@ nfs_xdr_readdirres(struct rpc_rqst *req,
if (!nr && (entry[0] != 0 || entry[1] == 0))
goto short_pkt;
out:
- kunmap(*page);
+ kunmap_atomic(kaddr, KM_USER0);
return nr;
short_pkt:
entry[0] = entry[1] = 0;
@@ -430,8 +430,8 @@ nfs_xdr_readdirres(struct rpc_rqst *req,
}
goto out;
err_unmap:
- kunmap(*page);
- return -errno_NFSERR_IO;
+ nr = -errno_NFSERR_IO;
+ goto out;
}

u32 *
@@ -542,7 +542,7 @@ nfs_xdr_readlinkres(struct rpc_rqst *req
xdr_shift_buf(rcvbuf, iov->iov_len - hdrlen);
}

- strlen = (u32*)kmap(rcvbuf->pages[0]);
+ strlen = (u32*)kmap_atomic(rcvbuf->pages[0], KM_USER0);
/* Convert length of symlink */
len = ntohl(*strlen);
if (len > rcvbuf->page_len)
@@ -551,7 +551,7 @@ nfs_xdr_readlinkres(struct rpc_rqst *req
/* NULL terminate the string we got */
string = (char *)(strlen + 1);
string[len] = 0;
- kunmap(rcvbuf->pages[0]);
+ kunmap_atomic(strlen, KM_USER0);
return 0;
}

diff -Naurp 00-stock/fs/nfs/nfs3xdr.c 01-kmap_atomic/fs/nfs/nfs3xdr.c
--- 00-stock/fs/nfs/nfs3xdr.c Sun Dec 15 21:08:24 2002
+++ 01-kmap_atomic/fs/nfs/nfs3xdr.c Wed Dec 18 17:22:06 2002
@@ -488,7 +488,7 @@ nfs3_xdr_readdirres(struct rpc_rqst *req
int hdrlen, recvd;
int status, nr;
unsigned int len, pglen;
- u32 *entry, *end;
+ u32 *entry, *end, *kaddr;

status = ntohl(*p++);
/* Decode post_op_attrs */
@@ -518,7 +518,7 @@ nfs3_xdr_readdirres(struct rpc_rqst *req
if (pglen > recvd)
pglen = recvd;
page = rcvbuf->pages;
- p = kmap(*page);
+ kaddr = p = (u32 *)kmap_atomic(*page, KM_USER0);
end = (u32 *)((char *)p + pglen);
entry = p;
for (nr = 0; *p++; nr++) {
@@ -563,7 +563,7 @@ nfs3_xdr_readdirres(struct rpc_rqst *req
if (!nr && (entry[0] != 0 || entry[1] == 0))
goto short_pkt;
out:
- kunmap(*page);
+ kunmap_atomic(kaddr, KM_USER0);
return nr;
short_pkt:
entry[0] = entry[1] = 0;
@@ -574,8 +574,8 @@ nfs3_xdr_readdirres(struct rpc_rqst *req
}
goto out;
err_unmap:
- kunmap(*page);
- return -errno_NFSERR_IO;
+ nr = -errno_NFSERR_IO;
+ goto out;
}

u32 *
@@ -738,7 +738,7 @@ nfs3_xdr_readlinkres(struct rpc_rqst *re
xdr_shift_buf(rcvbuf, iov->iov_len - hdrlen);
}

- strlen = (u32*)kmap(rcvbuf->pages[0]);
+ strlen = (u32*)kmap_atomic(rcvbuf->pages[0], KM_USER0);
/* Convert length of symlink */
len = ntohl(*strlen);
if (len > rcvbuf->page_len)
@@ -747,7 +747,7 @@ nfs3_xdr_readlinkres(struct rpc_rqst *re
/* NULL terminate the string we got */
string = (char *)(strlen + 1);
string[len] = 0;
- kunmap(rcvbuf->pages[0]);
+ kunmap_atomic(strlen, KM_USER0);
return 0;
}

diff -Naurp 00-stock/fs/nfs/nfs4proc.c 01-kmap_atomic/fs/nfs/nfs4proc.c
--- 00-stock/fs/nfs/nfs4proc.c Sun Dec 15 21:08:14 2002
+++ 01-kmap_atomic/fs/nfs/nfs4proc.c Wed Dec 18 17:22:06 2002
@@ -447,7 +447,7 @@ nfs4_setup_readdir(struct nfs4_compound
* when talking to the server, we always send cookie 0
* instead of 1 or 2.
*/
- start = p = (u32 *)kmap(*pages);
+ start = p = (u32 *)kmap_atomic(*pages, KM_USER0);

if (cookie == 0) {
*p++ = xdr_one; /* next */
@@ -475,7 +475,7 @@ nfs4_setup_readdir(struct nfs4_compound

readdir->rd_pgbase = (char *)p - (char *)start;
readdir->rd_count -= readdir->rd_pgbase;
- kunmap(*pages);
+ kunmap_atomic(start, KM_USER0);
}

static void
diff -Naurp 00-stock/fs/nfs/nfs4xdr.c 01-kmap_atomic/fs/nfs/nfs4xdr.c
--- 00-stock/fs/nfs/nfs4xdr.c Sun Dec 15 21:08:24 2002
+++ 01-kmap_atomic/fs/nfs/nfs4xdr.c Wed Dec 18 17:31:28 2002
@@ -1410,7 +1410,7 @@ decode_readdir(struct xdr_stream *xdr, s
struct page *page = *rcvbuf->pages;
struct iovec *iov = rcvbuf->head;
unsigned int nr, pglen = rcvbuf->page_len;
- uint32_t *end, *entry, *p;
+ uint32_t *end, *entry, *p, *kaddr;
uint32_t len, attrlen, word;
int i, hdrlen, recvd, status;

@@ -1434,7 +1434,7 @@ decode_readdir(struct xdr_stream *xdr, s
pglen = recvd;

BUG_ON(pglen + readdir->rd_pgbase > PAGE_CACHE_SIZE);
- p = (uint32_t *) kmap(page);
+ kaddr = p = (uint32_t *) kmap_atomic(page, KM_USER0);
end = (uint32_t *) ((char *)p + pglen + readdir->rd_pgbase);
entry = p;
for (nr = 0; *p++; nr++) {
@@ -1480,7 +1480,7 @@ decode_readdir(struct xdr_stream *xdr, s
if (!nr && (entry[0] != 0 || entry[1] == 0))
goto short_pkt;
out:
- kunmap(page);
+ kunmap_atomic(kaddr, KM_USER0);
return 0;
short_pkt:
entry[0] = entry[1] = 0;
@@ -1491,7 +1491,7 @@ short_pkt:
}
goto out;
err_unmap:
- kunmap(page);
+ kunmap_atomic(kaddr, KM_USER0);
return -errno_NFSERR_IO;
}

@@ -1522,18 +1522,18 @@ decode_readlink(struct xdr_stream *xdr,
* and and null-terminate the text (the VFS expects
* null-termination).
*/
- strlen = (uint32_t *) kmap(rcvbuf->pages[0]);
+ strlen = (uint32_t *) kmap_atomic(rcvbuf->pages[0], KM_USER0);
len = ntohl(*strlen);
if (len > PAGE_CACHE_SIZE - 5) {
printk(KERN_WARNING "nfs: server returned giant symlink!\n");
- kunmap(rcvbuf->pages[0]);
+ kunmap_atomic(strlen, KM_USER0);
return -EIO;
}
*strlen = len;

string = (char *)(strlen + 1);
string[len] = '\0';
- kunmap(rcvbuf->pages[0]);
+ kunmap_atomic(strlen, KM_USER0);
return 0;
}

diff -Naurp 00-stock/fs/nfs/read.c 01-kmap_atomic/fs/nfs/read.c
--- 00-stock/fs/nfs/read.c Sun Dec 15 21:07:42 2002
+++ 01-kmap_atomic/fs/nfs/read.c Wed Dec 18 17:22:06 2002
@@ -118,12 +118,8 @@ nfs_readpage_sync(struct file *file, str
break;
} while (count);

- if (count) {
- char *kaddr = kmap(page);
- memset(kaddr + offset, 0, count);
- kunmap(page);
- }
- flush_dcache_page(page);
+ if (count)
+ memclear_highpage_flush(page, offset, count);
SetPageUptodate(page);
if (PageError(page))
ClearPageError(page);
@@ -272,12 +268,9 @@ nfs_readpage_result(struct rpc_task *tas

if (task->tk_status >= 0) {
if (count < PAGE_CACHE_SIZE) {
- char *p = kmap(page);
-
- if (count < req->wb_bytes)
- memset(p + req->wb_offset + count, 0,
+ memclear_highpage_flush(page,
+ req->wb_offset + count,
req->wb_bytes - count);
- kunmap(page);

if (eof ||
((fattr->valid & NFS_ATTR_FATTR) &&
@@ -293,7 +286,6 @@ nfs_readpage_result(struct rpc_task *tas
}
} else
SetPageError(page);
- flush_dcache_page(page);
unlock_page(page);

dprintk("NFS: read (%s/%Ld %d@%Ld)\n",
diff -Naurp 00-stock/net/sunrpc/xdr.c 01-kmap_atomic/net/sunrpc/xdr.c
--- 00-stock/net/sunrpc/xdr.c Sun Dec 15 21:08:13 2002
+++ 01-kmap_atomic/net/sunrpc/xdr.c Wed Dec 18 17:22:06 2002
@@ -306,6 +306,7 @@ xdr_partial_copy_from_skb(struct xdr_buf
len = pglen;
ret = copy_actor(desc, kaddr, len);
}
+ flush_dcache_page(*ppage);
kunmap_atomic(kaddr, KM_SKB_SUNRPC_DATA);
if (ret != len || !desc->count)
return;

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Next message: Chuck Lever: "[PATCH] give NFS client a "set_page_dirty" address space op."
Previous message: Greg KH: "Re: converting cap_set_pg() to for_each_task_pid()"