From 89aa593010135660991d05c92528c2c9163d5900 Mon Sep 17 00:00:00 2001 From: "Yan, Zheng" Date: Fri, 8 Sep 2017 15:23:18 +0800 Subject: ceph: keep auth cap when inode has flocks or posix locks file locks are tracked by inode's auth mds. dropping auth caps is equivalent to releasing all file locks. Signed-off-by: "Yan, Zheng" Acked-by: Jeff Layton Signed-off-by: Ilya Dryomov --- fs/ceph/mds_client.c | 5 +++++ 1 file changed, 5 insertions(+) (limited to 'fs/ceph/mds_client.c') diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c index 0687ab3c3267..c8a811db387a 100644 --- a/fs/ceph/mds_client.c +++ b/fs/ceph/mds_client.c @@ -1462,6 +1462,11 @@ static int trim_caps_cb(struct inode *inode, struct ceph_cap *cap, void *arg) goto out; if ((used | wanted) & CEPH_CAP_ANY_WR) goto out; + /* Note: it's possible that i_filelock_ref becomes non-zero + * after dropping auth caps. It doesn't hurt because reply + * of lock mds request will re-add auth caps. */ + if (atomic_read(&ci->i_filelock_ref) > 0) + goto out; } /* The inode has cached pages, but it's no longer used. * we can safely drop it */ -- cgit From 4deb14a2593dfade102dd94a803a63cf620cfd56 Mon Sep 17 00:00:00 2001 From: "Yan, Zheng" Date: Mon, 11 Sep 2017 10:36:28 +0800 Subject: ceph: optimize flock encoding during reconnect Don't malloc if there is no flock. Signed-off-by: "Yan, Zheng" Reviewed-by: Jeff Layton Signed-off-by: Ilya Dryomov --- fs/ceph/mds_client.c | 34 ++++++++++++++++++++-------------- 1 file changed, 20 insertions(+), 14 deletions(-) (limited to 'fs/ceph/mds_client.c') diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c index c8a811db387a..295cf5e42ea9 100644 --- a/fs/ceph/mds_client.c +++ b/fs/ceph/mds_client.c @@ -2899,26 +2899,32 @@ static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap, if (recon_state->msg_version >= 2) { int num_fcntl_locks, num_flock_locks; - struct ceph_filelock *flocks; + struct ceph_filelock *flocks = NULL; size_t struct_len, total_len = 0; u8 struct_v = 0; encode_again: ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks); - flocks = kmalloc((num_fcntl_locks+num_flock_locks) * - sizeof(struct ceph_filelock), GFP_NOFS); - if (!flocks) { - err = -ENOMEM; - goto out_free; - } - err = ceph_encode_locks_to_buffer(inode, flocks, - num_fcntl_locks, - num_flock_locks); - if (err) { + if (num_fcntl_locks + num_flock_locks > 0) { + flocks = kmalloc((num_fcntl_locks + num_flock_locks) * + sizeof(struct ceph_filelock), GFP_NOFS); + if (!flocks) { + err = -ENOMEM; + goto out_free; + } + err = ceph_encode_locks_to_buffer(inode, flocks, + num_fcntl_locks, + num_flock_locks); + if (err) { + kfree(flocks); + flocks = NULL; + if (err == -ENOSPC) + goto encode_again; + goto out_free; + } + } else { kfree(flocks); - if (err == -ENOSPC) - goto encode_again; - goto out_free; + flocks = NULL; } if (recon_state->msg_version >= 3) { -- cgit From b3f8d68f38a879daed1eab66c0e19bc293096d34 Mon Sep 17 00:00:00 2001 From: "Yan, Zheng" Date: Mon, 11 Sep 2017 10:58:55 +0800 Subject: ceph: handle 'session get evicted while there are file locks' When session get evicted, all file locks associated with the session get released remotely by mds. File locks tracked by kernel become stale. In this situation, set an error flag on inode. The flag makes further file locks return -EIO. Another option to handle this situation is cleanup file locks tracked kernel. I do not choose it because it is inconvenient to notify user program about the error. Signed-off-by: "Yan, Zheng" Acked-by: Jeff Layton Signed-off-by: Ilya Dryomov --- fs/ceph/mds_client.c | 21 ++++++++++++++++----- 1 file changed, 16 insertions(+), 5 deletions(-) (limited to 'fs/ceph/mds_client.c') diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c index 295cf5e42ea9..8c8361262ade 100644 --- a/fs/ceph/mds_client.c +++ b/fs/ceph/mds_client.c @@ -1215,6 +1215,13 @@ static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap, } spin_unlock(&mdsc->cap_dirty_lock); + if (atomic_read(&ci->i_filelock_ref) > 0) { + /* make further file lock syscall return -EIO */ + ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK; + pr_warn_ratelimited(" dropping file locks for %p %lld\n", + inode, ceph_ino(inode)); + } + if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) { list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove); ci->i_prealloc_cap_flush = NULL; @@ -2832,7 +2839,7 @@ static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap, struct ceph_mds_cap_reconnect v2; struct ceph_mds_cap_reconnect_v1 v1; } rec; - struct ceph_inode_info *ci; + struct ceph_inode_info *ci = cap->ci; struct ceph_reconnect_state *recon_state = arg; struct ceph_pagelist *pagelist = recon_state->pagelist; char *path; @@ -2841,8 +2848,6 @@ static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap, u64 snap_follows; struct dentry *dentry; - ci = cap->ci; - dout(" adding %p ino %llx.%llx cap %p %lld %s\n", inode, ceph_vinop(inode), cap, cap->cap_id, ceph_cap_string(cap->issued)); @@ -2875,7 +2880,8 @@ static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap, rec.v2.issued = cpu_to_le32(cap->issued); rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino); rec.v2.pathbase = cpu_to_le64(pathbase); - rec.v2.flock_len = 0; + rec.v2.flock_len = + (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1; } else { rec.v1.cap_id = cpu_to_le64(cap->cap_id); rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci)); @@ -2904,7 +2910,12 @@ static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap, u8 struct_v = 0; encode_again: - ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks); + if (rec.v2.flock_len) { + ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks); + } else { + num_fcntl_locks = 0; + num_flock_locks = 0; + } if (num_fcntl_locks + num_flock_locks > 0) { flocks = kmalloc((num_fcntl_locks + num_flock_locks) * sizeof(struct ceph_filelock), GFP_NOFS); -- cgit From 76bd6ec4988065d39983ba8e93bb738313f68050 Mon Sep 17 00:00:00 2001 From: Ilya Dryomov Date: Mon, 16 Oct 2017 10:32:50 +0200 Subject: ceph: -EINVAL on decoding failure in ceph_mdsc_handle_fsmap() Don't set ->mdsmap_err to -ENOENT unconditionally, and drop unneeded return statement while at it. Signed-off-by: Ilya Dryomov --- fs/ceph/mds_client.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'fs/ceph/mds_client.c') diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c index 8c8361262ade..b76506be4228 100644 --- a/fs/ceph/mds_client.c +++ b/fs/ceph/mds_client.c @@ -3879,14 +3879,14 @@ void ceph_mdsc_handle_fsmap(struct ceph_mds_client *mdsc, struct ceph_msg *msg) goto err_out; } return; + bad: pr_err("error decoding fsmap\n"); err_out: mutex_lock(&mdsc->mutex); - mdsc->mdsmap_err = -ENOENT; + mdsc->mdsmap_err = err; __wake_requests(mdsc, &mdsc->waiting_for_map); mutex_unlock(&mdsc->mutex); - return; } /* -- cgit From c8a96a31cb04c7664626ab6ada7f66c98c09efbd Mon Sep 17 00:00:00 2001 From: Jeff Layton Date: Thu, 19 Oct 2017 08:53:58 -0400 Subject: ceph: clean up spinlocking and list handling around cleanup_cap_releases() Functions that release a lock taken in a parent frame are notoriously hard to follow. Split cleanup_cap_releases into two functions, one to detach the cap releases from the session (which should be called with the spinlock held), and another to dispose of those caps. Signed-off-by: Jeff Layton Reviewed-by: "Yan, Zheng" Reviewed-by: Ilya Dryomov Signed-off-by: Ilya Dryomov --- fs/ceph/mds_client.c | 34 +++++++++++++++++++++------------- 1 file changed, 21 insertions(+), 13 deletions(-) (limited to 'fs/ceph/mds_client.c') diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c index b76506be4228..53cde84e698a 100644 --- a/fs/ceph/mds_client.c +++ b/fs/ceph/mds_client.c @@ -1039,22 +1039,23 @@ void ceph_mdsc_open_export_target_sessions(struct ceph_mds_client *mdsc, * session caps */ -/* caller holds s_cap_lock, we drop it */ -static void cleanup_cap_releases(struct ceph_mds_client *mdsc, - struct ceph_mds_session *session) - __releases(session->s_cap_lock) +static void detach_cap_releases(struct ceph_mds_session *session, + struct list_head *target) { - LIST_HEAD(tmp_list); - list_splice_init(&session->s_cap_releases, &tmp_list); + lockdep_assert_held(&session->s_cap_lock); + + list_splice_init(&session->s_cap_releases, target); session->s_num_cap_releases = 0; - spin_unlock(&session->s_cap_lock); + dout("dispose_cap_releases mds%d\n", session->s_mds); +} - dout("cleanup_cap_releases mds%d\n", session->s_mds); - while (!list_empty(&tmp_list)) { +static void dispose_cap_releases(struct ceph_mds_client *mdsc, + struct list_head *dispose) +{ + while (!list_empty(dispose)) { struct ceph_cap *cap; /* zero out the in-progress message */ - cap = list_first_entry(&tmp_list, - struct ceph_cap, session_caps); + cap = list_first_entry(dispose, struct ceph_cap, session_caps); list_del(&cap->session_caps); ceph_put_cap(mdsc, cap); } @@ -1251,6 +1252,8 @@ static void remove_session_caps(struct ceph_mds_session *session) { struct ceph_fs_client *fsc = session->s_mdsc->fsc; struct super_block *sb = fsc->sb; + LIST_HEAD(dispose); + dout("remove_session_caps on %p\n", session); iterate_session_caps(session, remove_session_caps_cb, fsc); @@ -1285,10 +1288,12 @@ static void remove_session_caps(struct ceph_mds_session *session) } // drop cap expires and unlock s_cap_lock - cleanup_cap_releases(session->s_mdsc, session); + detach_cap_releases(session, &dispose); BUG_ON(session->s_nr_caps > 0); BUG_ON(!list_empty(&session->s_cap_flushing)); + spin_unlock(&session->s_cap_lock); + dispose_cap_releases(session->s_mdsc, &dispose); } /* @@ -3015,6 +3020,7 @@ static void send_mds_reconnect(struct ceph_mds_client *mdsc, int s_nr_caps; struct ceph_pagelist *pagelist; struct ceph_reconnect_state recon_state; + LIST_HEAD(dispose); pr_info("mds%d reconnect start\n", mds); @@ -3048,7 +3054,9 @@ static void send_mds_reconnect(struct ceph_mds_client *mdsc, */ session->s_cap_reconnect = 1; /* drop old cap expires; we're about to reestablish that state */ - cleanup_cap_releases(mdsc, session); + detach_cap_releases(session, &dispose); + spin_unlock(&session->s_cap_lock); + dispose_cap_releases(mdsc, &dispose); /* trim unused caps to reduce MDS's cache rejoin time */ if (mdsc->fsc->sb->s_root) -- cgit From ec1dff25b0a012711e10290039fbc8e1c5dd69fb Mon Sep 17 00:00:00 2001 From: Jeff Layton Date: Tue, 31 Oct 2017 15:51:14 -0400 Subject: ceph: silence sparse endianness warning in encode_caps_cb sparse warns: fs/ceph/mds_client.c:2887:34: warning: incorrect type in assignment (different base types) fs/ceph/mds_client.c:2887:34: expected restricted __le32 [assigned] [usertype] flock_len fs/ceph/mds_client.c:2887:34: got int At this point, it's just being used as a flag. It gets overwritten later if the rest of the encoding succeeds. Signed-off-by: Jeff Layton Reviewed-by: "Yan, Zheng" Signed-off-by: Ilya Dryomov --- fs/ceph/mds_client.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'fs/ceph/mds_client.c') diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c index 53cde84e698a..ab69dcb70e8a 100644 --- a/fs/ceph/mds_client.c +++ b/fs/ceph/mds_client.c @@ -2885,8 +2885,8 @@ static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap, rec.v2.issued = cpu_to_le32(cap->issued); rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino); rec.v2.pathbase = cpu_to_le64(pathbase); - rec.v2.flock_len = - (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1; + rec.v2.flock_len = (__force __le32) + ((ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1); } else { rec.v1.cap_id = cpu_to_le64(cap->cap_id); rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci)); -- cgit