From 0ae9b536ab4d44793251987a5c540feae252b40d Mon Sep 17 00:00:00 2001 Message-Id: <0ae9b536ab4d44793251987a5c540feae252b40d.1601675152.git.zanussi@kernel.org> In-Reply-To: <5b5a156f9808b1acf1205606e03da117214549ea.1601675151.git.zanussi@kernel.org> References: <5b5a156f9808b1acf1205606e03da117214549ea.1601675151.git.zanussi@kernel.org> From: Sebastian Andrzej Siewior Date: Fri, 20 Oct 2017 11:29:53 +0200 Subject: [PATCH 172/333] fs/dcache: disable preemption on i_dir_seq's write side Origin: https://www.kernel.org/pub/linux/kernel/projects/rt/4.19/older/patches-4.19.148-rt64.tar.xz i_dir_seq is an opencoded seqcounter. Based on the code it looks like we could have two writers in parallel despite the fact that the d_lock is held. The problem is that during the write process on RT the preemption is still enabled and if this process is interrupted by a reader with RT priority then we lock up. To avoid that lock up I am disabling the preemption during the update. The rename of i_dir_seq is here to ensure to catch new write sides in future. Cc: stable-rt@vger.kernel.org Reported-by: Oleg.Karfich@wago.com Signed-off-by: Sebastian Andrzej Siewior --- fs/dcache.c | 12 +++++++----- fs/inode.c | 2 +- include/linux/fs.h | 2 +- 3 files changed, 9 insertions(+), 7 deletions(-) diff --git a/fs/dcache.c b/fs/dcache.c index 10225a9135fb..dcde8ffe384c 100644 --- a/fs/dcache.c +++ b/fs/dcache.c @@ -2404,9 +2404,10 @@ EXPORT_SYMBOL(d_rehash); static inline unsigned start_dir_add(struct inode *dir) { + preempt_disable_rt(); for (;;) { - unsigned n = dir->i_dir_seq; - if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n) + unsigned n = dir->__i_dir_seq; + if (!(n & 1) && cmpxchg(&dir->__i_dir_seq, n, n + 1) == n) return n; cpu_relax(); } @@ -2414,7 +2415,8 @@ static inline unsigned start_dir_add(struct inode *dir) static inline void end_dir_add(struct inode *dir, unsigned n) { - smp_store_release(&dir->i_dir_seq, n + 2); + smp_store_release(&dir->__i_dir_seq, n + 2); + preempt_enable_rt(); } static void d_wait_lookup(struct dentry *dentry) @@ -2447,7 +2449,7 @@ struct dentry *d_alloc_parallel(struct dentry *parent, retry: rcu_read_lock(); - seq = smp_load_acquire(&parent->d_inode->i_dir_seq); + seq = smp_load_acquire(&parent->d_inode->__i_dir_seq); r_seq = read_seqbegin(&rename_lock); dentry = __d_lookup_rcu(parent, name, &d_seq); if (unlikely(dentry)) { @@ -2475,7 +2477,7 @@ struct dentry *d_alloc_parallel(struct dentry *parent, } hlist_bl_lock(b); - if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) { + if (unlikely(READ_ONCE(parent->d_inode->__i_dir_seq) != seq)) { hlist_bl_unlock(b); rcu_read_unlock(); goto retry; diff --git a/fs/inode.c b/fs/inode.c index c9eb5041ffae..e1fb0b743c96 100644 --- a/fs/inode.c +++ b/fs/inode.c @@ -156,7 +156,7 @@ int inode_init_always(struct super_block *sb, struct inode *inode) inode->i_bdev = NULL; inode->i_cdev = NULL; inode->i_link = NULL; - inode->i_dir_seq = 0; + inode->__i_dir_seq = 0; inode->i_rdev = 0; inode->dirtied_when = 0; diff --git a/include/linux/fs.h b/include/linux/fs.h index 876bfb6df06a..0b102f962518 100644 --- a/include/linux/fs.h +++ b/include/linux/fs.h @@ -679,7 +679,7 @@ struct inode { struct block_device *i_bdev; struct cdev *i_cdev; char *i_link; - unsigned i_dir_seq; + unsigned __i_dir_seq; }; __u32 i_generation; -- 2.17.1