If the symlink has already be been read-in, then page_follow_link_light can succeed in RCU-walk mode. page_getlink_rcu() is added to support this. With this many filesystems can follow links in RCU-walk mode when everything is cached. This includes ext?fs and others. If the page is a HighMem page we do *not* try to kmap_atomic, but simply give up - only page_address() is used. This is because we need to be able to sleep while holding the address of the page, particularly over calls to do_last() which can be quite slow and in particular takes a mutex. If this were a problem, then copying into a GFP_ATOMIC allocation might be a workable solution. This selective calling of kmap requires us to know, in page_put_link, whether or not kunmap() need to be called. Pass this information in the lsb of the cookie. Signed-off-by: NeilBrown <neilb@xxxxxxx> --- fs/namei.c | 35 ++++++++++++++++++++++++++++++----- 1 file changed, 30 insertions(+), 5 deletions(-) diff --git a/fs/namei.c b/fs/namei.c index c9c58cd1af2a..2602d31ecc99 100644 --- a/fs/namei.c +++ b/fs/namei.c @@ -4499,6 +4499,28 @@ static char *page_getlink(struct dentry * dentry, struct page **ppage) return kaddr; } +/* get the link contents from pagecache under RCU */ +static char *page_getlink_rcu(struct dentry * dentry, struct page **ppage) +{ + char *kaddr; + struct page *page; + struct address_space *mapping = dentry->d_inode->i_mapping; + page = find_get_page(mapping, 0); + if (page && + (!PageUptodate(page) || PageHighMem(page))) { + put_page(page); + page = NULL; + } + if (!page) { + *ppage = ERR_PTR(-ECHILD); + return NULL; + } + *ppage = page; + kaddr = page_address(page); + nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); + return kaddr; +} + int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) { struct page *page = NULL; @@ -4514,19 +4536,22 @@ EXPORT_SYMBOL(page_readlink); void *page_follow_link_light(struct dentry *dentry, int flags) { struct page *page = NULL; - if (flags & LOOKUP_RCU) - return ERR_PTR(-ECHILD); - nd_set_link(page_getlink(dentry, &page)); + if (flags & LOOKUP_RCU) { + nd_set_link(page_getlink_rcu(dentry, &page)); + page = (void*)((unsigned long)page | 1); + } else + nd_set_link(page_getlink(dentry, &page)); return page; } EXPORT_SYMBOL(page_follow_link_light); void page_put_link(struct dentry *dentry, void *cookie) { - struct page *page = cookie; + struct page *page = (void*)((unsigned long)cookie & ~1UL) ; if (page) { - kunmap(page); + if (page == cookie) + kunmap(page); page_cache_release(page); } } -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html