| Message ID | 20220920170021.1391560-1-bcodding@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | NFS: Trigger the "ls -l" readdir heuristic sooner | expand |
On 20 Sep 2022, at 13:00, Benjamin Coddington wrote: > Since commit 1a34c8c9a49e ("NFS: Support larger readdir buffers") has > updated dtsize, and with recent improvements to the READDIRPLUS helper > heuristic, the heuristic may not trigger until many dentries are emitted > to userspace. This will cause many thousands of GETATTR calls for "ls > -l" when the directory's pagecache has already been populated. This > manifests as poor performance for long directory listings after an > initially fast "ls -l". > > Fix this by emitting only 17 entries for any first pass through the NFS > directory's ->iterate_shared(), which allows userpace to prime the > counters for the heuristic. > > Signed-off-by: Benjamin Coddington <bcodding@redhat.com> > --- Hello Trond and Anna, please consider this patch. We have folks that are sad and confused that "ls -l" becomes much slower after running it once. Olga also tested this fix and reported performance gains here: https://lore.kernel.org/linux-nfs/CAN-5tyFsJy-D9cEfk8zCH8-0KErWteVW8Zb7BMqjL7==2qtT_A@mail.gmail.com/ Olga, would you be willing to contribute your "Tested-by" for this patch? Any comments from anyone would also be welcome. Ben
diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c index 5d6c2ddc7ea6..3b24b8af5514 100644 --- a/fs/nfs/dir.c +++ b/fs/nfs/dir.c @@ -1074,6 +1074,8 @@ static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc) return res; } +#define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL) + /* * Once we've found the start of the dirent within a page: fill 'er up... */ @@ -1083,6 +1085,7 @@ static void nfs_do_filldir(struct nfs_readdir_descriptor *desc, struct file *file = desc->file; struct nfs_cache_array *array; unsigned int i; + bool first_emit = !desc->dir_cookie; array = kmap_local_page(desc->page); for (i = desc->cache_entry_index; i < array->size; i++) { @@ -1106,6 +1109,10 @@ static void nfs_do_filldir(struct nfs_readdir_descriptor *desc, desc->ctx->pos = desc->dir_cookie; else desc->ctx->pos++; + if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 1) { + desc->eob = true; + break; + } } if (array->page_is_eof) desc->eof = !desc->eob; @@ -1187,8 +1194,6 @@ static int uncached_readdir(struct nfs_readdir_descriptor *desc) return status; } -#define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL) - static bool nfs_readdir_handle_cache_misses(struct inode *inode, struct nfs_readdir_descriptor *desc, unsigned int cache_misses,
Since commit 1a34c8c9a49e ("NFS: Support larger readdir buffers") has updated dtsize, and with recent improvements to the READDIRPLUS helper heuristic, the heuristic may not trigger until many dentries are emitted to userspace. This will cause many thousands of GETATTR calls for "ls -l" when the directory's pagecache has already been populated. This manifests as poor performance for long directory listings after an initially fast "ls -l". Fix this by emitting only 17 entries for any first pass through the NFS directory's ->iterate_shared(), which allows userpace to prime the counters for the heuristic. Signed-off-by: Benjamin Coddington <bcodding@redhat.com> --- fs/nfs/dir.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)