ceph: rework dcache readdir
Previously our dcache readdir code relies on that child dentries in
directory dentry's d_subdir list are sorted by dentry's offset in
descending order. When adding dentries to the dcache, if a dentry
already exists, our readdir code moves it to head of directory
dentry's d_subdir list. This design relies on dcache internals.
Al Viro suggests using ncpfs's approach: keeping array of pointers
to dentries in page cache of directory inode. the validity of those
pointers are presented by directory inode's complete and ordered
flags. When a dentry gets pruned, we clear directory inode's complete
flag in the d_prune() callback. Before moving a dentry to other
directory, we clear the ordered flag for both old and new directory.
Signed-off-by: Yan, Zheng <zyan@redhat.com>
diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
index e86d1a4..2a6d93b 100644
--- a/fs/ceph/inode.c
+++ b/fs/ceph/inode.c
@@ -390,9 +390,10 @@
ci->i_inline_version = 0;
ci->i_time_warp_seq = 0;
ci->i_ceph_flags = 0;
- ci->i_ordered_count = 0;
- atomic_set(&ci->i_release_count, 1);
- atomic_set(&ci->i_complete_count, 0);
+ atomic64_set(&ci->i_ordered_count, 1);
+ atomic64_set(&ci->i_release_count, 1);
+ atomic64_set(&ci->i_complete_seq[0], 0);
+ atomic64_set(&ci->i_complete_seq[1], 0);
ci->i_symlink = NULL;
memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
@@ -860,9 +861,10 @@
(issued & CEPH_CAP_FILE_EXCL) == 0 &&
!__ceph_dir_is_complete(ci)) {
dout(" marking %p complete (empty)\n", inode);
+ i_size_write(inode, 0);
__ceph_dir_set_complete(ci,
- atomic_read(&ci->i_release_count),
- ci->i_ordered_count);
+ atomic64_read(&ci->i_release_count),
+ atomic64_read(&ci->i_ordered_count));
}
wake = true;
@@ -1214,6 +1216,10 @@
dout("fill_trace doing d_move %p -> %p\n",
req->r_old_dentry, dn);
+ /* d_move screws up sibling dentries' offsets */
+ ceph_dir_clear_ordered(dir);
+ ceph_dir_clear_ordered(olddir);
+
d_move(req->r_old_dentry, dn);
dout(" src %p '%pd' dst %p '%pd'\n",
req->r_old_dentry,
@@ -1224,10 +1230,6 @@
rehashing bug in vfs_rename_dir */
ceph_invalidate_dentry_lease(dn);
- /* d_move screws up sibling dentries' offsets */
- ceph_dir_clear_ordered(dir);
- ceph_dir_clear_ordered(olddir);
-
dout("dn %p gets new offset %lld\n", req->r_old_dentry,
ceph_dentry(req->r_old_dentry)->offset);
@@ -1335,6 +1337,49 @@
return err;
}
+void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
+{
+ if (ctl->page) {
+ kunmap(ctl->page);
+ page_cache_release(ctl->page);
+ ctl->page = NULL;
+ }
+}
+
+static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
+ struct ceph_readdir_cache_control *ctl,
+ struct ceph_mds_request *req)
+{
+ struct ceph_inode_info *ci = ceph_inode(dir);
+ unsigned nsize = PAGE_CACHE_SIZE / sizeof(struct dentry*);
+ unsigned idx = ctl->index % nsize;
+ pgoff_t pgoff = ctl->index / nsize;
+
+ if (!ctl->page || pgoff != page_index(ctl->page)) {
+ ceph_readdir_cache_release(ctl);
+ ctl->page = grab_cache_page(&dir->i_data, pgoff);
+ if (!ctl->page) {
+ ctl->index = -1;
+ return -ENOMEM;
+ }
+ /* reading/filling the cache are serialized by
+ * i_mutex, no need to use page lock */
+ unlock_page(ctl->page);
+ ctl->dentries = kmap(ctl->page);
+ }
+
+ if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
+ req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
+ dout("readdir cache dn %p idx %d\n", dn, ctl->index);
+ ctl->dentries[idx] = dn;
+ ctl->index++;
+ } else {
+ dout("disable readdir cache\n");
+ ctl->index = -1;
+ }
+ return 0;
+}
+
int ceph_readdir_prepopulate(struct ceph_mds_request *req,
struct ceph_mds_session *session)
{
@@ -1347,8 +1392,11 @@
struct inode *snapdir = NULL;
struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
struct ceph_dentry_info *di;
- u64 r_readdir_offset = req->r_readdir_offset;
u32 frag = le32_to_cpu(rhead->args.readdir.frag);
+ struct ceph_readdir_cache_control cache_ctl = {};
+
+ if (req->r_aborted)
+ return readdir_prepopulate_inodes_only(req, session);
if (rinfo->dir_dir &&
le32_to_cpu(rinfo->dir_dir->frag) != frag) {
@@ -1356,14 +1404,11 @@
frag, le32_to_cpu(rinfo->dir_dir->frag));
frag = le32_to_cpu(rinfo->dir_dir->frag);
if (ceph_frag_is_leftmost(frag))
- r_readdir_offset = 2;
+ req->r_readdir_offset = 2;
else
- r_readdir_offset = 0;
+ req->r_readdir_offset = 0;
}
- if (req->r_aborted)
- return readdir_prepopulate_inodes_only(req, session);
-
if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
snapdir = ceph_get_snapdir(d_inode(parent));
parent = d_find_alias(snapdir);
@@ -1376,6 +1421,17 @@
ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
}
+ if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
+ /* note dir version at start of readdir so we can tell
+ * if any dentries get dropped */
+ struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
+ req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
+ req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
+ req->r_readdir_cache_idx = 0;
+ }
+
+ cache_ctl.index = req->r_readdir_cache_idx;
+
/* FIXME: release caps/leases if error occurs */
for (i = 0; i < rinfo->dir_nr; i++) {
struct ceph_vino vino;
@@ -1415,13 +1471,6 @@
d_delete(dn);
dput(dn);
goto retry_lookup;
- } else {
- /* reorder parent's d_subdirs */
- spin_lock(&parent->d_lock);
- spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
- list_move(&dn->d_child, &parent->d_subdirs);
- spin_unlock(&dn->d_lock);
- spin_unlock(&parent->d_lock);
}
/* inode */
@@ -1438,13 +1487,15 @@
}
}
- if (fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
- req->r_request_started, -1,
- &req->r_caps_reservation) < 0) {
+ ret = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
+ req->r_request_started, -1,
+ &req->r_caps_reservation);
+ if (ret < 0) {
pr_err("fill_inode badness on %p\n", in);
if (d_really_is_negative(dn))
iput(in);
d_drop(dn);
+ err = ret;
goto next_item;
}
@@ -1460,19 +1511,28 @@
}
di = dn->d_fsdata;
- di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
+ di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
update_dentry_lease(dn, rinfo->dir_dlease[i],
req->r_session,
req->r_request_started);
+
+ if (err == 0 && cache_ctl.index >= 0) {
+ ret = fill_readdir_cache(d_inode(parent), dn,
+ &cache_ctl, req);
+ if (ret < 0)
+ err = ret;
+ }
next_item:
if (dn)
dput(dn);
}
- if (err == 0)
- req->r_did_prepopulate = true;
-
out:
+ if (err == 0) {
+ req->r_did_prepopulate = true;
+ req->r_readdir_cache_idx = cache_ctl.index;
+ }
+ ceph_readdir_cache_release(&cache_ctl);
if (snapdir) {
iput(snapdir);
dput(parent);