From b47d19de2c714020ba8f5545a6e7d4968f37eb45 Mon Sep 17 00:00:00 2001 From: Arun Bharadwaj Date: Thu, 18 Nov 2010 10:36:43 +0000 Subject: Pure nfs client performance using odirect. When an application opens a file with O_DIRECT flag, if the size of the data that is written is equal to wsize, the client sends a WRITE RPC with stable flag set to UNSTABLE followed by a single COMMIT RPC rather than sending a single WRITE RPC with the stable flag set to FILE_SYNC. This a bug. Patch to fix this. Signed-off-by: Arun R Bharadwaj Signed-off-by: Trond Myklebust --- fs/nfs/direct.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'fs') diff --git a/fs/nfs/direct.c b/fs/nfs/direct.c index 84d3c8b..e6ace0d 100644 --- a/fs/nfs/direct.c +++ b/fs/nfs/direct.c @@ -867,7 +867,7 @@ static ssize_t nfs_direct_write(struct kiocb *iocb, const struct iovec *iov, goto out; nfs_alloc_commit_data(dreq); - if (dreq->commit_data == NULL || count < wsize) + if (dreq->commit_data == NULL || count <= wsize) sync = NFS_FILE_SYNC; dreq->inode = inode; -- cgit v1.1 From 463a376eae1c92a66c912af539bfd4bbefa37673 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Sat, 20 Nov 2010 12:22:20 -0500 Subject: NFS: Buffer overflow in ->decode_dirent() should not be fatal Overflowing the buffer in the readdir ->decode_dirent() should not lead to a fatal error, but rather to an attempt to reread the record in question. Signed-off-by: Trond Myklebust --- fs/nfs/nfs2xdr.c | 2 +- fs/nfs/nfs3xdr.c | 2 +- fs/nfs/nfs4xdr.c | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) (limited to 'fs') diff --git a/fs/nfs/nfs2xdr.c b/fs/nfs/nfs2xdr.c index 2563f76..ab593773 100644 --- a/fs/nfs/nfs2xdr.c +++ b/fs/nfs/nfs2xdr.c @@ -495,7 +495,7 @@ nfs_decode_dirent(struct xdr_stream *xdr, struct nfs_entry *entry, struct nfs_se out_overflow: print_overflow_msg(__func__, xdr); - return ERR_PTR(-EIO); + return ERR_PTR(-EAGAIN); } /* diff --git a/fs/nfs/nfs3xdr.c b/fs/nfs/nfs3xdr.c index 748dc91..e79e4f5 100644 --- a/fs/nfs/nfs3xdr.c +++ b/fs/nfs/nfs3xdr.c @@ -656,7 +656,7 @@ nfs3_decode_dirent(struct xdr_stream *xdr, struct nfs_entry *entry, struct nfs_s out_overflow: print_overflow_msg(__func__, xdr); out_overflow_exit: - return ERR_PTR(-EIO); + return ERR_PTR(-EAGAIN); } /* diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c index b7a204f..a3b39cb 100644 --- a/fs/nfs/nfs4xdr.c +++ b/fs/nfs/nfs4xdr.c @@ -6221,7 +6221,7 @@ __be32 *nfs4_decode_dirent(struct xdr_stream *xdr, struct nfs_entry *entry, out_overflow: print_overflow_msg(__func__, xdr); - return ERR_PTR(-EIO); + return ERR_PTR(-EAGAIN); } /* -- cgit v1.1 From 5c346854d8ce6ca91931f8fc9177934257a667d0 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Sat, 20 Nov 2010 12:43:45 -0500 Subject: NFS: Assume eof if the server returns no readdir records Some servers are known to be buggy w.r.t. this. Deal with them... Signed-off-by: Trond Myklebust --- fs/nfs/dir.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) (limited to 'fs') diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c index 662df2a..2789cb3 100644 --- a/fs/nfs/dir.c +++ b/fs/nfs/dir.c @@ -466,8 +466,9 @@ int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *en struct xdr_stream stream; struct xdr_buf buf; __be32 *ptr = xdr_page; - int status; struct nfs_cache_array *array; + unsigned int count = 0; + int status; buf.head->iov_base = xdr_page; buf.head->iov_len = buflen; @@ -488,6 +489,8 @@ int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *en break; } + count++; + if (desc->plus == 1) nfs_prime_dcache(desc->file->f_path.dentry, entry); @@ -496,13 +499,14 @@ int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *en break; } while (!entry->eof); - if (status == -EBADCOOKIE && entry->eof) { + if (count == 0 || (status == -EBADCOOKIE && entry->eof == 1)) { array = nfs_readdir_get_array(page); if (!IS_ERR(array)) { array->eof_index = array->size; status = 0; nfs_readdir_release_array(page); - } + } else + status = PTR_ERR(array); } return status; } -- cgit v1.1 From e7c58e974a0318fcca5368e7b3570e10e9ae9028 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Sat, 20 Nov 2010 13:22:24 -0500 Subject: NFS: Fix a page leak in nfs_do_filldir() nfs_do_filldir() must always free desc->page when it is done, otherwise we end up leaking the page. Also remove unused variable 'dentry'. Signed-off-by: Trond Myklebust --- fs/nfs/dir.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) (limited to 'fs') diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c index 2789cb3..42e66e9 100644 --- a/fs/nfs/dir.c +++ b/fs/nfs/dir.c @@ -701,11 +701,12 @@ int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, int res = 0; struct nfs_cache_array *array = NULL; unsigned int d_type = DT_UNKNOWN; - struct dentry *dentry = NULL; array = nfs_readdir_get_array(desc->page); - if (IS_ERR(array)) - return PTR_ERR(array); + if (IS_ERR(array)) { + res = PTR_ERR(array); + goto out; + } for (i = desc->cache_entry_index; i < array->size; i++) { d_type = DT_UNKNOWN; @@ -726,9 +727,8 @@ int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, desc->eof = 1; nfs_readdir_release_array(desc->page); +out: cache_page_release(desc); - if (dentry != NULL) - dput(dentry); dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (unsigned long long)*desc->dir_cookie, res); return res; -- cgit v1.1 From 7a8e1dc34f52fd2927dbf7e520d7cd8eadc51336 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Sat, 20 Nov 2010 13:24:46 -0500 Subject: NFS: Fix a page leak in uncached_readdir() Signed-off-by: Trond Myklebust --- fs/nfs/dir.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'fs') diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c index 42e66e9..353f47c 100644 --- a/fs/nfs/dir.c +++ b/fs/nfs/dir.c @@ -763,13 +763,14 @@ int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, goto out; } + desc->page_index = 0; + desc->page = page; + if (nfs_readdir_xdr_to_array(desc, page, inode) == -1) { status = -EIO; goto out_release; } - desc->page_index = 0; - desc->page = page; status = nfs_do_filldir(desc, dirent, filldir); out: -- cgit v1.1 From 85f8607e163f8d281fb407357279cb4ac6df12e6 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Sat, 20 Nov 2010 13:24:49 -0500 Subject: NFS: Fix the error handling in "uncached_readdir()" Currently, uncached_readdir() is broken because if fails to handle the results from nfs_readdir_xdr_to_array() correctly. Signed-off-by: Trond Myklebust --- fs/nfs/dir.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) (limited to 'fs') diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c index 353f47c..2492bac 100644 --- a/fs/nfs/dir.c +++ b/fs/nfs/dir.c @@ -766,10 +766,9 @@ int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, desc->page_index = 0; desc->page = page; - if (nfs_readdir_xdr_to_array(desc, page, inode) == -1) { - status = -EIO; + status = nfs_readdir_xdr_to_array(desc, page, inode); + if (status < 0) goto out_release; - } status = nfs_do_filldir(desc, dirent, filldir); -- cgit v1.1 From ece0b4233b6b915d1f63add2bd9f2733aec6317a Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Sat, 20 Nov 2010 13:55:33 -0500 Subject: NFS: Don't ignore errors from nfs_do_filldir() We should ignore the errors from the filldir callback, and just interpret them as meaning we should exit, however we should definitely pass back ENOMEM errors. Signed-off-by: Trond Myklebust --- fs/nfs/dir.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) (limited to 'fs') diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c index 2492bac..ddc2e43 100644 --- a/fs/nfs/dir.c +++ b/fs/nfs/dir.c @@ -709,13 +709,15 @@ int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, } for (i = desc->cache_entry_index; i < array->size; i++) { + struct nfs_cache_array_entry *ent; d_type = DT_UNKNOWN; - res = filldir(dirent, array->array[i].string.name, - array->array[i].string.len, file->f_pos, - nfs_compat_user_ino64(array->array[i].ino), d_type); - if (res < 0) + ent = &array->array[i]; + if (filldir(dirent, ent->string.name, ent->string.len, + file->f_pos, nfs_compat_user_ino64(ent->ino), d_type) < 0) { + desc->eof = 1; break; + } file->f_pos++; desc->cache_entry_index = i; if (i < (array->size-1)) @@ -820,14 +822,14 @@ static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) res = readdir_search_pagecache(desc); if (res == -EBADCOOKIE) { + res = 0; /* This means either end of directory */ if (*desc->dir_cookie && desc->eof == 0) { /* Or that the server has 'lost' a cookie */ res = uncached_readdir(desc, dirent, filldir); - if (res >= 0) + if (res == 0) continue; } - res = 0; break; } if (res == -ETOOSMALL && desc->plus) { @@ -842,10 +844,8 @@ static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) break; res = nfs_do_filldir(desc, dirent, filldir); - if (res < 0) { - res = 0; + if (res < 0) break; - } } out: nfs_unblock_sillyrename(dentry); -- cgit v1.1 From 3020093f578fb6c9acc6914dfd887a1ebd1db659 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Sat, 20 Nov 2010 15:18:22 -0500 Subject: NFS: Correct the array bound calculation in nfs_readdir_add_to_array It looks as if the array size calculation in MAX_READDIR_ARRAY does not take the alignment of struct nfs_cache_array_entry into account. Signed-off-by: Trond Myklebust --- fs/nfs/dir.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) (limited to 'fs') diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c index ddc2e43..ced7291 100644 --- a/fs/nfs/dir.c +++ b/fs/nfs/dir.c @@ -171,8 +171,6 @@ struct nfs_cache_array { struct nfs_cache_array_entry array[0]; }; -#define MAX_READDIR_ARRAY ((PAGE_SIZE - sizeof(struct nfs_cache_array)) / sizeof(struct nfs_cache_array_entry)) - typedef __be32 * (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, struct nfs_server *, int); typedef struct { struct file *file; @@ -257,11 +255,14 @@ int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) if (IS_ERR(array)) return PTR_ERR(array); + + cache_entry = &array->array[array->size]; + + /* Check that this entry lies within the page bounds */ ret = -ENOSPC; - if (array->size >= MAX_READDIR_ARRAY) + if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) goto out; - cache_entry = &array->array[array->size]; cache_entry->cookie = entry->prev_cookie; cache_entry->ino = entry->ino; ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); -- cgit v1.1 From 0b26a0bf6ff398185546432420bb772bcfdf8d94 Mon Sep 17 00:00:00 2001 From: Trond Myklebust Date: Sat, 20 Nov 2010 14:26:44 -0500 Subject: NFS: Ensure we return the dirent->d_type when it is known Store the dirent->d_type in the struct nfs_cache_array_entry so that we can use it in getdents() calls. This fixes a regression with the new readdir code. Signed-off-by: Trond Myklebust --- fs/nfs/dir.c | 7 ++++--- fs/nfs/internal.h | 9 +++++++++ fs/nfs/nfs2xdr.c | 2 ++ fs/nfs/nfs3xdr.c | 2 ++ fs/nfs/nfs4xdr.c | 4 ++++ 5 files changed, 21 insertions(+), 3 deletions(-) (limited to 'fs') diff --git a/fs/nfs/dir.c b/fs/nfs/dir.c index ced7291..8ea4a41 100644 --- a/fs/nfs/dir.c +++ b/fs/nfs/dir.c @@ -162,6 +162,7 @@ struct nfs_cache_array_entry { u64 cookie; u64 ino; struct qstr string; + unsigned char d_type; }; struct nfs_cache_array { @@ -265,6 +266,7 @@ int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) cache_entry->cookie = entry->prev_cookie; cache_entry->ino = entry->ino; + cache_entry->d_type = entry->d_type; ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); if (ret) goto out; @@ -701,7 +703,6 @@ int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, int i = 0; int res = 0; struct nfs_cache_array *array = NULL; - unsigned int d_type = DT_UNKNOWN; array = nfs_readdir_get_array(desc->page); if (IS_ERR(array)) { @@ -711,11 +712,11 @@ int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, for (i = desc->cache_entry_index; i < array->size; i++) { struct nfs_cache_array_entry *ent; - d_type = DT_UNKNOWN; ent = &array->array[i]; if (filldir(dirent, ent->string.name, ent->string.len, - file->f_pos, nfs_compat_user_ino64(ent->ino), d_type) < 0) { + file->f_pos, nfs_compat_user_ino64(ent->ino), + ent->d_type) < 0) { desc->eof = 1; break; } diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h index db08ff3..e6356b7 100644 --- a/fs/nfs/internal.h +++ b/fs/nfs/internal.h @@ -362,6 +362,15 @@ unsigned int nfs_page_length(struct page *page) } /* + * Convert a umode to a dirent->d_type + */ +static inline +unsigned char nfs_umode_to_dtype(umode_t mode) +{ + return (mode >> 12) & 15; +} + +/* * Determine the number of pages in an array of length 'len' and * with a base offset of 'base' */ diff --git a/fs/nfs/nfs2xdr.c b/fs/nfs/nfs2xdr.c index ab593773..5914a19 100644 --- a/fs/nfs/nfs2xdr.c +++ b/fs/nfs/nfs2xdr.c @@ -485,6 +485,8 @@ nfs_decode_dirent(struct xdr_stream *xdr, struct nfs_entry *entry, struct nfs_se entry->prev_cookie = entry->cookie; entry->cookie = ntohl(*p++); + entry->d_type = DT_UNKNOWN; + p = xdr_inline_peek(xdr, 8); if (p != NULL) entry->eof = !p[0] && p[1]; diff --git a/fs/nfs/nfs3xdr.c b/fs/nfs/nfs3xdr.c index e79e4f5..f6cc60f 100644 --- a/fs/nfs/nfs3xdr.c +++ b/fs/nfs/nfs3xdr.c @@ -622,11 +622,13 @@ nfs3_decode_dirent(struct xdr_stream *xdr, struct nfs_entry *entry, struct nfs_s entry->prev_cookie = entry->cookie; p = xdr_decode_hyper(p, &entry->cookie); + entry->d_type = DT_UNKNOWN; if (plus) { entry->fattr->valid = 0; p = xdr_decode_post_op_attr_stream(xdr, entry->fattr); if (IS_ERR(p)) goto out_overflow_exit; + entry->d_type = nfs_umode_to_dtype(entry->fattr->mode); /* In fact, a post_op_fh3: */ p = xdr_inline_decode(xdr, 4); if (unlikely(!p)) diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c index a3b39cb..9f1826b 100644 --- a/fs/nfs/nfs4xdr.c +++ b/fs/nfs/nfs4xdr.c @@ -6208,6 +6208,10 @@ __be32 *nfs4_decode_dirent(struct xdr_stream *xdr, struct nfs_entry *entry, if (entry->fattr->valid & NFS_ATTR_FATTR_FILEID) entry->ino = entry->fattr->fileid; + entry->d_type = DT_UNKNOWN; + if (entry->fattr->valid & NFS_ATTR_FATTR_TYPE) + entry->d_type = nfs_umode_to_dtype(entry->fattr->mode); + if (verify_attr_len(xdr, p, len) < 0) goto out_overflow; -- cgit v1.1