| Message ID | 20180523205427.17678-1-lsahlber@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
merged into cifs-2.6.git for-next On Wed, May 23, 2018 at 3:54 PM, Ronnie Sahlberg <lsahlber@redhat.com> wrote: > RHBZ: 1566345 > > When truncating a file we always do this synchronously to the server. > Thus we need to make sure that the cached inode metadata is > marked as stale so that on next getattr we will refresh the metadata. > In this particular bug we want to ensure that both ctime and mtime > are updated and become visible to the application after a truncate. > > Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com> > --- > fs/cifs/inode.c | 13 +++++++++---- > 1 file changed, 9 insertions(+), 4 deletions(-) > > diff --git a/fs/cifs/inode.c b/fs/cifs/inode.c > index 3c371f7f5963..745fd7fe8d0e 100644 > --- a/fs/cifs/inode.c > +++ b/fs/cifs/inode.c > @@ -746,7 +746,8 @@ cifs_get_inode_info(struct inode **inode, const char *full_path, > cifs_dbg(FYI, "Getting info on %s\n", full_path); > > if ((data == NULL) && (*inode != NULL)) { > - if (CIFS_CACHE_READ(CIFS_I(*inode))) { > + if (CIFS_CACHE_READ(CIFS_I(*inode)) && > + CIFS_I(*inode)->time != 0) { > cifs_dbg(FYI, "No need to revalidate cached inode sizes\n"); > goto cgii_exit; > } > @@ -1857,15 +1858,15 @@ cifs_inode_needs_reval(struct inode *inode) > struct cifsInodeInfo *cifs_i = CIFS_I(inode); > struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb); > > + if (cifs_i->time == 0) > + return true; > + > if (CIFS_CACHE_READ(cifs_i)) > return false; > > if (!lookupCacheEnabled) > return true; > > - if (cifs_i->time == 0) > - return true; > - > if (!cifs_sb->actimeo) > return true; > > @@ -2104,10 +2105,14 @@ static int cifs_truncate_page(struct address_space *mapping, loff_t from) > > static void cifs_setsize(struct inode *inode, loff_t offset) > { > + struct cifsInodeInfo *cifs_i = CIFS_I(inode); > + > spin_lock(&inode->i_lock); > i_size_write(inode, offset); > spin_unlock(&inode->i_lock); > > + /* Cached inode must be refreshed on truncate */ > + cifs_i->time = 0; > truncate_pagecache(inode, offset); > } > > -- > 2.13.3 >
diff --git a/fs/cifs/inode.c b/fs/cifs/inode.c index 3c371f7f5963..745fd7fe8d0e 100644 --- a/fs/cifs/inode.c +++ b/fs/cifs/inode.c @@ -746,7 +746,8 @@ cifs_get_inode_info(struct inode **inode, const char *full_path, cifs_dbg(FYI, "Getting info on %s\n", full_path); if ((data == NULL) && (*inode != NULL)) { - if (CIFS_CACHE_READ(CIFS_I(*inode))) { + if (CIFS_CACHE_READ(CIFS_I(*inode)) && + CIFS_I(*inode)->time != 0) { cifs_dbg(FYI, "No need to revalidate cached inode sizes\n"); goto cgii_exit; } @@ -1857,15 +1858,15 @@ cifs_inode_needs_reval(struct inode *inode) struct cifsInodeInfo *cifs_i = CIFS_I(inode); struct cifs_sb_info *cifs_sb = CIFS_SB(inode->i_sb); + if (cifs_i->time == 0) + return true; + if (CIFS_CACHE_READ(cifs_i)) return false; if (!lookupCacheEnabled) return true; - if (cifs_i->time == 0) - return true; - if (!cifs_sb->actimeo) return true; @@ -2104,10 +2105,14 @@ static int cifs_truncate_page(struct address_space *mapping, loff_t from) static void cifs_setsize(struct inode *inode, loff_t offset) { + struct cifsInodeInfo *cifs_i = CIFS_I(inode); + spin_lock(&inode->i_lock); i_size_write(inode, offset); spin_unlock(&inode->i_lock); + /* Cached inode must be refreshed on truncate */ + cifs_i->time = 0; truncate_pagecache(inode, offset); }
RHBZ: 1566345 When truncating a file we always do this synchronously to the server. Thus we need to make sure that the cached inode metadata is marked as stale so that on next getattr we will refresh the metadata. In this particular bug we want to ensure that both ctime and mtime are updated and become visible to the application after a truncate. Signed-off-by: Ronnie Sahlberg <lsahlber@redhat.com> --- fs/cifs/inode.c | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-)