ceph: improve fscache revalidation

There are several issues in fscache revalidation code.
- In ceph_revalidate_work(), fscache_invalidate() is called when
  fscache_check_consistency() return 0. This is complete wrong
  because 0 means cache is valid.
- Handle_cap_grant() calls ceph_queue_revalidate() if client
  already has CAP_FILE_CACHE. This code is confusing. Client
  should revalidate the cache each time it got CAP_FILE_CACHE
  anew.
- In Handle_cap_grant(), fscache_invalidate() is called if MDS
  revokes CAP_FILE_CACHE. This is inconsistency with the case
  that inode get evicted. In the later case, the cache is not
  discarded. Client may use the cache when inode is reloaded.

This patch moves the fscache revalidation into ceph_get_caps().
Client revalidates the cache after it gets CAP_FILE_CACHE.
i_rdcache_gen should keep constance while CAP_FILE_CACHE is
used. If i_fscache_gen is not equal to i_rdcache_gen, client
needs to check cache's consistency.

Signed-off-by: Yan, Zheng <zyan@redhat.com>

4 files changed, 41 insertions, 83 deletions

diff --git a/fs/ceph/cache.c b/fs/ceph/cache.c
index c19db6afd0cc..5b3f4828f214 100644
--- a/fs/ceph/cache.c
+++ b/fs/ceph/cache.c
@@ -69,15 +69,8 @@ int ceph_fscache_register_fs(struct ceph_fs_client* fsc)
         fsc->fscache = fscache_acquire_cookie(ceph_cache_netfs.primary_index,
                                               &ceph_fscache_fsid_object_def,
                                               fsc, true);
-
-        if (fsc->fscache == NULL) {
+        if (!fsc->fscache)
                 pr_err("Unable to resgister fsid: %p fscache cookie", fsc);
-                return 0;
-        }
-
-        fsc->revalidate_wq = alloc_workqueue("ceph-revalidate", 0, 1);
-        if (fsc->revalidate_wq == NULL)
-                return -ENOMEM;
 
         return 0;
 }
@@ -260,8 +253,7 @@ static void ceph_vfs_readpage_complete_unlock(struct page *page, void *data, int
 
 static inline bool cache_valid(struct ceph_inode_info *ci)
 {
-        return ((ceph_caps_issued(ci) & CEPH_CAP_FILE_CACHE) &&
-                (ci->i_fscache_gen == ci->i_rdcache_gen));
+        return ci->i_fscache_gen == ci->i_rdcache_gen;
 }
 
 
@@ -354,69 +346,27 @@ void ceph_invalidate_fscache_page(struct inode* inode, struct page *page)
 
 void ceph_fscache_unregister_fs(struct ceph_fs_client* fsc)
 {
-        if (fsc->revalidate_wq)
-                destroy_workqueue(fsc->revalidate_wq);
-
         fscache_relinquish_cookie(fsc->fscache, 0);
         fsc->fscache = NULL;
 }
 
-static void ceph_revalidate_work(struct work_struct *work)
-{
-        int issued;
-        u32 orig_gen;
-        struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
-                                                  i_revalidate_work);
-        struct inode *inode = &ci->vfs_inode;
-
-        spin_lock(&ci->i_ceph_lock);
-        issued = __ceph_caps_issued(ci, NULL);
-        orig_gen = ci->i_rdcache_gen;
-        spin_unlock(&ci->i_ceph_lock);
-
-        if (!(issued & CEPH_CAP_FILE_CACHE)) {
-                dout("revalidate_work lost cache before validation %p\n",
-                     inode);
-                goto out;
-        }
-
-        if (!fscache_check_consistency(ci->fscache))
-                fscache_invalidate(ci->fscache);
-
-        spin_lock(&ci->i_ceph_lock);
-        /* Update the new valid generation (backwards sanity check too) */
-        if (orig_gen > ci->i_fscache_gen) {
-                ci->i_fscache_gen = orig_gen;
-        }
-        spin_unlock(&ci->i_ceph_lock);
-
-out:
-        iput(&ci->vfs_inode);
-}
-
-void ceph_queue_revalidate(struct inode *inode)
+/*
+ * caller should hold CEPH_CAP_FILE_{RD,CACHE}
+ */
+void ceph_fscache_revalidate_cookie(struct ceph_inode_info *ci)
 {
-        struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
-        struct ceph_inode_info *ci = ceph_inode(inode);
-
-        if (fsc->revalidate_wq == NULL || ci->fscache == NULL)
+        if (cache_valid(ci))
                 return;
 
-        ihold(inode);
-
-        if (queue_work(ceph_sb_to_client(inode->i_sb)->revalidate_wq,
-                       &ci->i_revalidate_work)) {
-                dout("ceph_queue_revalidate %p\n", inode);
-        } else {
-                dout("ceph_queue_revalidate %p failed\n)", inode);
-                iput(inode);
+        /* resue i_truncate_mutex. There should be no pending
+         * truncate while the caller holds CEPH_CAP_FILE_RD */
+        mutex_lock(&ci->i_truncate_mutex);
+        if (!cache_valid(ci)) {
+                if (fscache_check_consistency(ci->fscache))
+                        fscache_invalidate(ci->fscache);
+                spin_lock(&ci->i_ceph_lock);
+                ci->i_fscache_gen = ci->i_rdcache_gen;
+                spin_unlock(&ci->i_ceph_lock);
         }
-}
-
-void ceph_fscache_inode_init(struct ceph_inode_info *ci)
-{
-        ci->fscache = NULL;
-        /* The first load is verifed cookie open time */
-        ci->i_fscache_gen = 1;
-        INIT_WORK(&ci->i_revalidate_work, ceph_revalidate_work);
+        mutex_unlock(&ci->i_truncate_mutex);
 }
diff --git a/fs/ceph/cache.h b/fs/ceph/cache.h
index dfe9b5823bd3..7e72c7594f0c 100644
--- a/fs/ceph/cache.h
+++ b/fs/ceph/cache.h
@@ -34,10 +34,10 @@ void ceph_fscache_unregister(void);
 int ceph_fscache_register_fs(struct ceph_fs_client* fsc);
 void ceph_fscache_unregister_fs(struct ceph_fs_client* fsc);
 
-void ceph_fscache_inode_init(struct ceph_inode_info *ci);
 void ceph_fscache_register_inode_cookie(struct inode *inode);
 void ceph_fscache_unregister_inode_cookie(struct ceph_inode_info* ci);
 void ceph_fscache_file_set_cookie(struct inode *inode, struct file *filp);
+void ceph_fscache_revalidate_cookie(struct ceph_inode_info *ci);
 
 int ceph_readpage_from_fscache(struct inode *inode, struct page *page);
 int ceph_readpages_from_fscache(struct inode *inode,
@@ -46,7 +46,12 @@ int ceph_readpages_from_fscache(struct inode *inode,
                                 unsigned *nr_pages);
 void ceph_readpage_to_fscache(struct inode *inode, struct page *page);
 void ceph_invalidate_fscache_page(struct inode* inode, struct page *page);
-void ceph_queue_revalidate(struct inode *inode);
+
+static inline void ceph_fscache_inode_init(struct ceph_inode_info *ci)
+{
+        ci->fscache = NULL;
+        ci->i_fscache_gen = 0;
+}
 
 static inline void ceph_fscache_invalidate(struct inode *inode)
 {
@@ -82,6 +87,11 @@ static inline void ceph_fscache_readpages_cancel(struct inode *inode,
         return fscache_readpages_cancel(ci->fscache, pages);
 }
 
+static inline void ceph_disable_fscache_readpage(struct ceph_inode_info *ci)
+{
+        ci->i_fscache_gen = ci->i_rdcache_gen - 1;
+}
+
 #else
 
 static inline int ceph_fscache_register(void)
@@ -119,6 +129,10 @@ static inline void ceph_fscache_file_set_cookie(struct inode *inode,
 {
 }
 
+static inline void ceph_fscache_revalidate_cookie(struct ceph_inode_info *ci)
+{
+}
+
 static inline void ceph_fscache_uncache_page(struct inode *inode,
                                              struct page *pages)
 {
@@ -167,7 +181,7 @@ static inline void ceph_fscache_readpages_cancel(struct inode *inode,
 {
 }
 
-static inline void ceph_queue_revalidate(struct inode *inode)
+static inline void ceph_disable_fscache_readpage(struct ceph_inode_info *ci)
 {
 }
 
diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index 7bdf7d59a36d..6f60d0a3d0f9 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -2393,6 +2393,9 @@ again:
                                 snap_rwsem_locked = true;
                         }
                         *got = need | (have & want);
+                        if ((need & CEPH_CAP_FILE_RD) &&
+                            !(*got & CEPH_CAP_FILE_CACHE))
+                                ceph_disable_fscache_readpage(ci);
                         __take_cap_refs(ci, *got, true);
                         ret = 1;
                 }
@@ -2554,6 +2557,9 @@ int ceph_get_caps(struct ceph_inode_info *ci, int need, int want,
                 break;
         }
 
+        if ((_got & CEPH_CAP_FILE_RD) && (_got & CEPH_CAP_FILE_CACHE))
+                ceph_fscache_revalidate_cookie(ci);
+
         *got = _got;
         return 0;
 }
@@ -2795,7 +2801,6 @@ static void handle_cap_grant(struct ceph_mds_client *mdsc,
         bool writeback = false;
         bool queue_trunc = false;
         bool queue_invalidate = false;
-        bool queue_revalidate = false;
         bool deleted_inode = false;
         bool fill_inline = false;
 
@@ -2837,8 +2842,6 @@ static void handle_cap_grant(struct ceph_mds_client *mdsc,
                                 ci->i_rdcache_revoking = ci->i_rdcache_gen;
                         }
                 }
-
-                ceph_fscache_invalidate(inode);
         }
 
         /* side effects now are allowed */
@@ -2880,11 +2883,6 @@ static void handle_cap_grant(struct ceph_mds_client *mdsc,
                 }
         }
 
-        /* Do we need to revalidate our fscache cookie. Don't bother on the
-         * first cache cap as we already validate at cookie creation time. */
-        if ((issued & CEPH_CAP_FILE_CACHE) && ci->i_rdcache_gen > 1)
-                queue_revalidate = true;
-
         if (newcaps & CEPH_CAP_ANY_RD) {
                 /* ctime/mtime/atime? */
                 ceph_decode_timespec(&mtime, &grant->mtime);
@@ -2995,8 +2993,6 @@ static void handle_cap_grant(struct ceph_mds_client *mdsc,
 
         if (queue_trunc)
                 ceph_queue_vmtruncate(inode);
-        else if (queue_revalidate)
-                ceph_queue_revalidate(inode);
 
         if (writeback)
                 /*
diff --git a/fs/ceph/super.h b/fs/ceph/super.h
index 0130a8592191..0168b49fb6ad 100644
--- a/fs/ceph/super.h
+++ b/fs/ceph/super.h
@@ -103,7 +103,6 @@ struct ceph_fs_client {
 
 #ifdef CONFIG_CEPH_FSCACHE
         struct fscache_cookie *fscache;
-        struct workqueue_struct *revalidate_wq;
 #endif
 };
 
@@ -360,8 +359,7 @@ struct ceph_inode_info {
 
 #ifdef CONFIG_CEPH_FSCACHE
         struct fscache_cookie *fscache;
-        u32 i_fscache_gen; /* sequence, for delayed fscache validate */
-        struct work_struct i_revalidate_work;
+        u32 i_fscache_gen;
 #endif
         struct inode vfs_inode; /* at end */
 };