I create the second version of the patch which uses QpathInfo and
open-close approach and posted to the list with "[PATCH] CIFS: Fix
symbolic links usage" subject (CC'ed Joao as well).
Joao, could you test it in your workload, please?
2013/10/30, Jeff Layton <jlayton@xxxxxxxxxx>:
> On Wed, 30 Oct 2013 14:48:38 +0400
> Pavel Shilovsky <piastry@xxxxxxxxxxx> wrote:
>
>> 2013/10/30 Jeff Layton <jlayton@xxxxxxxxxx>:
>> > On Wed, 30 Oct 2013 11:46:22 +0400
>> > Pavel Shilovsky <piastry@xxxxxxxxxxx> wrote:
>> >
>> >> 2013/10/29 Jeff Layton <jlayton@xxxxxxxxxx>:
>> >> > On Wed, 23 Oct 2013 18:42:54 +0400
>> >> > Pavel Shilovsky <piastry@xxxxxxxxxxx> wrote:
>> >> >
>> >> >> I created a slightly tested patch below but can't test with
>> >> >> deduplicated files now.
>> >> >>
>> >> >> Joao, can you check if it fixes your problem, please?
>> >> >>
>> >> >> --------------------------------------------------------------------------
>> >> >>
>> >> >> Distinguish reparse point types into two groups:
>> >> >> 1) that can be accessed directly through a reparse point
>> >> >> (junctions, deduplicated files, NFS symlinks);
>> >> >> 2) that need to be processed manually (Windows symbolic links,
>> >> >> DFS).
>> >> >>
>> >> >> In this case we map only Windows symbolic links to Unix ones.
>> >> >>
>> >> >> Signed-off-by: Pavel Shilovsky <piastry@xxxxxxxxxxx>
>> >> >> ---
>> >> >> fs/cifs/cifsglob.h | 2 +-
>> >> >> fs/cifs/inode.c | 23 +++++++++++++----------
>> >> >> fs/cifs/readdir.c | 40
>> >> >> ++++++++--------------------------------
>> >> >> fs/cifs/smb1ops.c | 27 ++++++++++++++++++++++-----
>> >> >> fs/cifs/smb2inode.c | 15 +++++++++++----
>> >> >> fs/cifs/smb2proto.h | 2 +-
>> >> >> 6 files changed, 56 insertions(+), 53 deletions(-)
>> >> >>
>> >> >> diff --git a/fs/cifs/cifsglob.h b/fs/cifs/cifsglob.h
>> >> >> index a67cf12..b688f2b 100644
>> >> >> --- a/fs/cifs/cifsglob.h
>> >> >> +++ b/fs/cifs/cifsglob.h
>> >> >> @@ -261,7 +261,7 @@ struct smb_version_operations {
>> >> >> /* query path data from the server */
>> >> >> int (*query_path_info)(const unsigned int, struct cifs_tcon
>> >> >> *,
>> >> >> struct cifs_sb_info *, const char *,
>> >> >> - FILE_ALL_INFO *, bool *);
>> >> >> + FILE_ALL_INFO *, bool *, bool *);
>> >> >> /* query file data from the server */
>> >> >> int (*query_file_info)(const unsigned int, struct cifs_tcon
>> >> >> *,
>> >> >> struct cifs_fid *, FILE_ALL_INFO *);
>> >> >> diff --git a/fs/cifs/inode.c b/fs/cifs/inode.c
>> >> >> index 867b7cd..36f9ebb 100644
>> >> >> --- a/fs/cifs/inode.c
>> >> >> +++ b/fs/cifs/inode.c
>> >> >> @@ -542,7 +542,8 @@ static int cifs_sfu_mode(struct cifs_fattr
>> >> >> *fattr, const unsigned char *path,
>> >> >> /* Fill a cifs_fattr struct with info from FILE_ALL_INFO */
>> >> >> static void
>> >> >> cifs_all_info_to_fattr(struct cifs_fattr *fattr, FILE_ALL_INFO
>> >> >> *info,
>> >> >> - struct cifs_sb_info *cifs_sb, bool adjust_tz)
>> >> >> + struct cifs_sb_info *cifs_sb, bool adjust_tz,
>> >> >> + bool symlink)
>> >> >> {
>> >> >> struct cifs_tcon *tcon = cifs_sb_master_tcon(cifs_sb);
>> >> >>
>> >> >> @@ -569,7 +570,11 @@ cifs_all_info_to_fattr(struct cifs_fattr
>> >> >> *fattr, FILE_ALL_INFO *info,
>> >> >> fattr->cf_createtime = le64_to_cpu(info->CreationTime);
>> >> >>
>> >> >> fattr->cf_nlink = le32_to_cpu(info->NumberOfLinks);
>> >> >> - if (fattr->cf_cifsattrs & ATTR_DIRECTORY) {
>> >> >> +
>> >> >> + if (symlink) {
>> >> >> + fattr->cf_mode = S_IFLNK;
>> >> >> + fattr->cf_dtype = DT_LNK;
>> >> >> + } else if (fattr->cf_cifsattrs & ATTR_DIRECTORY) {
>> >> >> fattr->cf_mode = S_IFDIR | cifs_sb->mnt_dir_mode;
>> >> >> fattr->cf_dtype = DT_DIR;
>> >> >> /*
>> >> >> @@ -578,10 +583,6 @@ cifs_all_info_to_fattr(struct cifs_fattr
>> >> >> *fattr, FILE_ALL_INFO *info,
>> >> >> */
>> >> >> if (!tcon->unix_ext)
>> >> >> fattr->cf_flags |= CIFS_FATTR_UNKNOWN_NLINK;
>> >> >> - } else if (fattr->cf_cifsattrs & ATTR_REPARSE) {
>> >> >> - fattr->cf_mode = S_IFLNK;
>> >> >> - fattr->cf_dtype = DT_LNK;
>> >> >> - fattr->cf_nlink = le32_to_cpu(info->NumberOfLinks);
>> >> >> } else {
>> >> >> fattr->cf_mode = S_IFREG | cifs_sb->mnt_file_mode;
>> >> >> fattr->cf_dtype = DT_REG;
>> >> >> @@ -626,7 +627,8 @@ cifs_get_file_info(struct file *filp)
>> >> >> rc = server->ops->query_file_info(xid, tcon, &cfile->fid,
>> >> >> &find_data);
>> >> >> switch (rc) {
>> >> >> case 0:
>> >> >> - cifs_all_info_to_fattr(&fattr, &find_data, cifs_sb,
>> >> >> false);
>> >> >> + cifs_all_info_to_fattr(&fattr, &find_data, cifs_sb,
>> >> >> false,
>> >> >> + false);
>> >> >> break;
>> >> >> case -EREMOTE:
>> >> >> cifs_create_dfs_fattr(&fattr, inode->i_sb);
>> >> >> @@ -673,6 +675,7 @@ cifs_get_inode_info(struct inode **inode, const
>> >> >> char *full_path,
>> >> >> bool adjust_tz = false;
>> >> >> struct cifs_fattr fattr;
>> >> >> struct cifs_search_info *srchinf = NULL;
>> >> >> + bool symlink = false;
>> >> >>
>> >> >> tlink = cifs_sb_tlink(cifs_sb);
>> >> >> if (IS_ERR(tlink))
>> >> >> @@ -702,12 +705,12 @@ cifs_get_inode_info(struct inode **inode,
>> >> >> const char *full_path,
>> >> >> }
>> >> >> data = (FILE_ALL_INFO *)buf;
>> >> >> rc = server->ops->query_path_info(xid, tcon, cifs_sb,
>> >> >> full_path,
>> >> >> - data, &adjust_tz);
>> >> >> + data, &adjust_tz,
>> >> >> &symlink);
>> >> >> }
>> >> >>
>> >> >> if (!rc) {
>> >> >> - cifs_all_info_to_fattr(&fattr, (FILE_ALL_INFO *)data,
>> >> >> cifs_sb,
>> >> >> - adjust_tz);
>> >> >> + cifs_all_info_to_fattr(&fattr, data, cifs_sb,
>> >> >> adjust_tz,
>> >> >> + symlink);
>> >> >> } else if (rc == -EREMOTE) {
>> >> >> cifs_create_dfs_fattr(&fattr, sb);
>> >> >> rc = 0;
>> >> >> diff --git a/fs/cifs/readdir.c b/fs/cifs/readdir.c
>> >> >> index 53a75f3..5940eca 100644
>> >> >> --- a/fs/cifs/readdir.c
>> >> >> +++ b/fs/cifs/readdir.c
>> >> >> @@ -134,22 +134,6 @@ out:
>> >> >> dput(dentry);
>> >> >> }
>> >> >>
>> >> >> -/*
>> >> >> - * Is it possible that this directory might turn out to be a DFS
>> >> >> referral
>> >> >> - * once we go to try and use it?
>> >> >> - */
>> >> >> -static bool
>> >> >> -cifs_dfs_is_possible(struct cifs_sb_info *cifs_sb)
>> >> >> -{
>> >> >> -#ifdef CONFIG_CIFS_DFS_UPCALL
>> >> >> - struct cifs_tcon *tcon = cifs_sb_master_tcon(cifs_sb);
>> >> >> -
>> >> >> - if (tcon->Flags & SMB_SHARE_IS_IN_DFS)
>> >> >> - return true;
>> >> >> -#endif
>> >> >> - return false;
>> >> >> -}
>> >> >> -
>> >> >> static void
>> >> >> cifs_fill_common_info(struct cifs_fattr *fattr, struct cifs_sb_info
>> >> >> *cifs_sb)
>> >> >> {
>> >> >> @@ -159,27 +143,19 @@ cifs_fill_common_info(struct cifs_fattr
>> >> >> *fattr, struct cifs_sb_info *cifs_sb)
>> >> >> if (fattr->cf_cifsattrs & ATTR_DIRECTORY) {
>> >> >> fattr->cf_mode = S_IFDIR | cifs_sb->mnt_dir_mode;
>> >> >> fattr->cf_dtype = DT_DIR;
>> >> >> - /*
>> >> >> - * Windows CIFS servers generally make DFS referrals
>> >> >> look
>> >> >> - * like directories in FIND_* responses with the
>> >> >> reparse
>> >> >> - * attribute flag also set (since DFS junctions are
>> >> >> - * reparse points). We must revalidate at least these
>> >> >> - * directory inodes before trying to use them (if
>> >> >> - * they are DFS we will get PATH_NOT_COVERED back
>> >> >> - * when queried directly and can then try to connect
>> >> >> - * to the DFS target)
>> >> >> - */
>> >> >> - if (cifs_dfs_is_possible(cifs_sb) &&
>> >> >> - (fattr->cf_cifsattrs & ATTR_REPARSE))
>> >> >> - fattr->cf_flags |= CIFS_FATTR_NEED_REVAL;
>> >> >> - } else if (fattr->cf_cifsattrs & ATTR_REPARSE) {
>> >> >> - fattr->cf_mode = S_IFLNK;
>> >> >> - fattr->cf_dtype = DT_LNK;
>> >> >> } else {
>> >> >> fattr->cf_mode = S_IFREG | cifs_sb->mnt_file_mode;
>> >> >> fattr->cf_dtype = DT_REG;
>> >> >> }
>> >> >>
>> >> >> + /*
>> >> >> + * We need to revalidate it further to make a decision about
>> >> >> whether it
>> >> >> + * is a symbolic link, DFS referral or a reparse point with a
>> >> >> direct
>> >> >> + * access like junctions, deduplicated files, NFS symlinks.
>> >> >> + */
>> >> >> + if (fattr->cf_cifsattrs & ATTR_REPARSE)
>> >> >> + fattr->cf_flags |= CIFS_FATTR_NEED_REVAL;
>> >> >> +
>> >> >> /* non-unix readdir doesn't provide nlink */
>> >> >> fattr->cf_flags |= CIFS_FATTR_UNKNOWN_NLINK;
>> >> >>
>> >> >> diff --git a/fs/cifs/smb1ops.c b/fs/cifs/smb1ops.c
>> >> >> index ea99efe..3bb41cf 100644
>> >> >> --- a/fs/cifs/smb1ops.c
>> >> >> +++ b/fs/cifs/smb1ops.c
>> >> >> @@ -534,26 +534,43 @@ cifs_is_path_accessible(const unsigned int
>> >> >> xid, struct cifs_tcon *tcon,
>> >> >> static int
>> >> >> cifs_query_path_info(const unsigned int xid, struct cifs_tcon
>> >> >> *tcon,
>> >> >> struct cifs_sb_info *cifs_sb, const char
>> >> >> *full_path,
>> >> >> - FILE_ALL_INFO *data, bool *adjustTZ)
>> >> >> + FILE_ALL_INFO *data, bool *adjustTZ, bool
>> >> >> *symlink)
>> >> >> {
>> >> >> int rc;
>> >> >> + int oplock = 0;
>> >> >> + __u16 netfid;
>> >> >> +
>> >> >> + rc = CIFSSMBOpen(xid, tcon, full_path, FILE_OPEN,
>> >> >> FILE_READ_ATTRIBUTES,
>> >> >> + 0, &netfid, &oplock, NULL,
>> >> >> cifs_sb->local_nls,
>> >> >> + cifs_sb->mnt_cifs_flags &
>> >> >> CIFS_MOUNT_MAP_SPECIAL_CHR);
>> >> >> + if (rc == -EOPNOTSUPP) {
>> >> >> + *symlink = true;
>> >> >> + /* Failed on a symbolic link - query a reparse point
>> >> >> info */
>> >> >> + rc = CIFSSMBOpen(xid, tcon, full_path, FILE_OPEN,
>> >> >> + FILE_READ_ATTRIBUTES,
>> >> >> OPEN_REPARSE_POINT,
>> >> >> + &netfid, &oplock, NULL,
>> >> >> cifs_sb->local_nls,
>> >> >> + cifs_sb->mnt_cifs_flags &
>> >> >> CIFS_MOUNT_MAP_SPECIAL_CHR);
>> >> >> + }
>> >> >> +
>> >> >> + if (!rc) {
>> >> >> + rc = CIFSSMBQFileInfo(xid, tcon, netfid, data);
>> >> >> + CIFSSMBClose(xid, tcon, netfid);
>> >> >> + }
>> >> >
>> >> > Blech...
>> >> >
>> >> > So basically for any normal, non-symlink file you'll end up breaking
>> >> > any oplocks that might be held just by stat()'ing it? I sincerely
>> >> > hope
>> >> > there's a better way to solve this problem...
>> >> >
>> >>
>> >> No, we don't break any oplocks/leases here. Open with READ_ATTRIBUTES
>> >> access doesn't force the server to break oplocks. Note, that SMB2.0
>> >> protocol has been already doing stat with open-query-close mechanism
>> >> that cause no influence to oplocks/leases held by other clients.
>> >>
>> >> Citing MSDN
>> >> (http://msdn.microsoft.com/en-us/library/windows/hardware/ff548639(v=vs.85).aspx):
>> >> "Note When processing IRP_MJ_CREATE for any oplock, if the desired
>> >> access contains nothing other than FILE_READ_ATTRIBUTES,
>> >> FILE_WRITE_ATTRIBUTES, or SYNCHRONIZE, the oplock does not break
>> >> unless FILE_RESERVE_OPFILTER is specified."
>> >>
>> >
>> > That's good at least, but it is still ugly. Plus, with this change will
>> > triple the number of round trips to the server, which will cause
>> > performance issues. I think you really want to avoid doing this unless
>> > you properly redesign this to use a compound.
>>
>> While I agree with merging these three commands into a compound
>> request, I don't think that we should add it during the stable phase.
>> The approach proposed by this patch will bring a performance penalty,
>> of course, but it will be only on non-Unix shares with an application
>> doing 'stat' on many files.
>>
>
> Which is a common workload on our most common deployment scenario.
>
>> Another possibility is to query file attributes with QpathInfo and
>> check if it's a reparse point. If no - return as usual. If yes - do
>> open-close to determine if the reparse point is symbolic link or not.
>> In this case we will not get a performance penalty for normal files
>> but will use 3 commands for every reparse points.
>>
>
> That sounds like a better option even if it isn't as elegant as it
> could be. Reparse points aren't terribly common and we shouldn't
> optimize the code for them.
>
> --
> Jeff Layton <jlayton@xxxxxxxxxx>
>
--
Best regards,
Pavel Shilovsky.
--
To unsubscribe from this list: send the line "unsubscribe linux-cifs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
