| Message ID | 520D3EA7.1010109@oracle.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, 15 Aug 2013 at 15:48, Dave Kleikamp wrote: > This patch replaces the one I posted yesterday. I like this better since > it doesn't require fixing existing on-disk cookies or skipping a > position in the in-inode index table. Thanks. Applied to 3.11-rc5 and tested, no more "readdir loop" messages and with unique inode numbers, great! Tested-by: Christian Kujau <lists@nerdbynature.de> Thanks for the fix! Christian.
On 08/15/2013 04:26 PM, Christian Kujau wrote: > On Thu, 15 Aug 2013 at 15:48, Dave Kleikamp wrote: >> This patch replaces the one I posted yesterday. I like this better since >> it doesn't require fixing existing on-disk cookies or skipping a >> position in the in-inode index table. > > Thanks. Applied to 3.11-rc5 and tested, no more "readdir loop" messages > and with unique inode numbers, great! > > Tested-by: Christian Kujau <lists@nerdbynature.de> > > Thanks for the fix! > Christian. Thanks for reporting, investigating and testing! Dave -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thu, 15 Aug 2013 at 15:48, Dave Kleikamp wrote: > This patch replaces the one I posted yesterday. I like this better since > it doesn't require fixing existing on-disk cookies or skipping a > position in the in-inode index table. > > NFSv4 reserves readdir cookie values 0-2 for special entries (. and ..), > but jfs allows a value of 2 for a non-special entry. This incompatibility > can result in the nfs client reporting a readdir loop. > > This patch doesn't change the value stored internally, but adds one to > the value exposed to the iterate method. Out of curiosity, will this land on 3.11? Christian.
On 08/24/2013 05:21 PM, Christian Kujau wrote: > On Thu, 15 Aug 2013 at 15:48, Dave Kleikamp wrote: >> This patch replaces the one I posted yesterday. I like this better since >> it doesn't require fixing existing on-disk cookies or skipping a >> position in the in-inode index table. >> >> NFSv4 reserves readdir cookie values 0-2 for special entries (. and ..), >> but jfs allows a value of 2 for a non-special entry. This incompatibility >> can result in the nfs client reporting a readdir loop. >> >> This patch doesn't change the value stored internally, but adds one to >> the value exposed to the iterate method. > > Out of curiosity, will this land on 3.11? Just sent a pull request to Linus. Once it's picked up, I'll submit a patch to the stable trees. -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/fs/jfs/jfs_dtree.c b/fs/jfs/jfs_dtree.c index 8743ba9..0ec767e 100644 --- a/fs/jfs/jfs_dtree.c +++ b/fs/jfs/jfs_dtree.c @@ -3047,6 +3047,14 @@ int jfs_readdir(struct file *file, struct dir_context *ctx) dir_index = (u32) ctx->pos; + /* + * NFSv4 reserves cookies 1 and 2 for . and .. so we add + * the value we return to the vfs is one greater than the + * one we use internally. + */ + if (dir_index) + dir_index--; + if (dir_index > 1) { struct dir_table_slot dirtab_slot; @@ -3086,7 +3094,7 @@ int jfs_readdir(struct file *file, struct dir_context *ctx) if (p->header.flag & BT_INTERNAL) { jfs_err("jfs_readdir: bad index table"); DT_PUTPAGE(mp); - ctx->pos = -1; + ctx->pos = DIREND; return 0; } } else { @@ -3094,14 +3102,14 @@ int jfs_readdir(struct file *file, struct dir_context *ctx) /* * self "." */ - ctx->pos = 0; + ctx->pos = 1; if (!dir_emit(ctx, ".", 1, ip->i_ino, DT_DIR)) return 0; } /* * parent ".." */ - ctx->pos = 1; + ctx->pos = 2; if (!dir_emit(ctx, "..", 2, PARENT(ip), DT_DIR)) return 0; @@ -3122,22 +3130,23 @@ int jfs_readdir(struct file *file, struct dir_context *ctx) /* * Legacy filesystem - OS/2 & Linux JFS < 0.3.6 * - * pn = index = 0: First entry "." - * pn = 0; index = 1: Second entry ".." + * pn = 0; index = 1: First entry "." + * pn = 0; index = 2: Second entry ".." * pn > 0: Real entries, pn=1 -> leftmost page * pn = index = -1: No more entries */ dtpos = ctx->pos; - if (dtpos == 0) { + if (dtpos < 2) { /* build "." entry */ + ctx->pos = 1; if (!dir_emit(ctx, ".", 1, ip->i_ino, DT_DIR)) return 0; - dtoffset->index = 1; + dtoffset->index = 2; ctx->pos = dtpos; } if (dtoffset->pn == 0) { - if (dtoffset->index == 1) { + if (dtoffset->index == 2) { /* build ".." entry */ if (!dir_emit(ctx, "..", 2, PARENT(ip), DT_DIR)) return 0; @@ -3228,6 +3237,12 @@ int jfs_readdir(struct file *file, struct dir_context *ctx) } jfs_dirent->position = unique_pos++; } + /* + * We add 1 to the index because we may + * use a value of 2 internally, and NFSv4 + * doesn't like that. + */ + jfs_dirent->position++; } else { jfs_dirent->position = dtpos; len = min(d_namleft, DTLHDRDATALEN_LEGACY);
This patch replaces the one I posted yesterday. I like this better since it doesn't require fixing existing on-disk cookies or skipping a position in the in-inode index table. NFSv4 reserves readdir cookie values 0-2 for special entries (. and ..), but jfs allows a value of 2 for a non-special entry. This incompatibility can result in the nfs client reporting a readdir loop. This patch doesn't change the value stored internally, but adds one to the value exposed to the iterate method. Signed-off-by: Dave Kleikamp <dave.kleikamp@oracle.com> --- fs/jfs/jfs_dtree.c | 31 +++++++++++++++++++++++-------- 1 file changed, 23 insertions(+), 8 deletions(-)