Merge tag 'ceph-for-5.8-rc1' of git://github.com/ceph/ceph-client

Pull ceph updates from Ilya Dryomov:
 "The highlights are:

   - OSD/MDS latency and caps cache metrics infrastructure for the
     filesytem (Xiubo Li). Currently available through debugfs and will
     be periodically sent to the MDS in the future.

   - support for replica reads (balanced and localized reads) for rbd
     and the filesystem (myself). The default remains to always read
     from primary, users can opt-in with the new crush_location and
     read_from_replica options. Note that reading from replica is safe
     for general use only since Octopus.

   - support for RADOS allocation hint flags (myself). Currently used by
     rbd to propagate the compressible/incompressible hint given with
     the new compression_hint map option and ready for passing on more
     advanced hints, e.g. based on fadvise() from the filesystem.

   - support for efficient cross-quota-realm renames (Luis Henriques)

   - assorted cap handling improvements and cleanups, particularly
     untangling some of the locking (Jeff Layton)"

* tag 'ceph-for-5.8-rc1' of git://github.com/ceph/ceph-client: (29 commits)
  rbd: compression_hint option
  libceph: support for alloc hint flags
  libceph: read_from_replica option
  libceph: support for balanced and localized reads
  libceph: crush_location infrastructure
  libceph: decode CRUSH device/bucket types and names
  libceph: add non-asserting rbtree insertion helper
  ceph: skip checking caps when session reconnecting and releasing reqs
  ceph: make sure mdsc->mutex is nested in s->s_mutex to fix dead lock
  ceph: don't return -ESTALE if there's still an open file
  libceph, rbd: replace zero-length array with flexible-array
  ceph: allow rename operation under different quota realms
  ceph: normalize 'delta' parameter usage in check_quota_exceeded
  ceph: ceph_kick_flushing_caps needs the s_mutex
  ceph: request expedited service on session's last cap flush
  ceph: convert mdsc->cap_dirty to a per-session list
  ceph: reset i_requested_max_size if file write is not wanted
  ceph: throw a warning if we destroy session with mutex still locked
  ceph: fix potential race in ceph_check_caps
  ceph: document what protects i_dirty_item and i_flushing_item
  ...

29 files changed, 1405 insertions, 252 deletions

diff --git a/drivers/block/rbd.c b/drivers/block/rbd.c
index 67d65ac785e9..7420648a1de6 100644
--- a/drivers/block/rbd.c
+++ b/drivers/block/rbd.c
@@ -836,6 +836,7 @@ enum {
 	Opt_lock_timeout,
 	/* int args above */
 	Opt_pool_ns,
+	Opt_compression_hint,
 	/* string args above */
 	Opt_read_only,
 	Opt_read_write,
@@ -844,8 +845,23 @@ enum {
 	Opt_notrim,
 };
 
+enum {
+	Opt_compression_hint_none,
+	Opt_compression_hint_compressible,
+	Opt_compression_hint_incompressible,
+};
+
+static const struct constant_table rbd_param_compression_hint[] = {
+	{"none",		Opt_compression_hint_none},
+	{"compressible",	Opt_compression_hint_compressible},
+	{"incompressible",	Opt_compression_hint_incompressible},
+	{}
+};
+
 static const struct fs_parameter_spec rbd_parameters[] = {
 	fsparam_u32	("alloc_size",			Opt_alloc_size),
+	fsparam_enum	("compression_hint",		Opt_compression_hint,
+			 rbd_param_compression_hint),
 	fsparam_flag	("exclusive",			Opt_exclusive),
 	fsparam_flag	("lock_on_read",		Opt_lock_on_read),
 	fsparam_u32	("lock_timeout",		Opt_lock_timeout),
@@ -867,6 +883,8 @@ struct rbd_options {
 	bool	lock_on_read;
 	bool	exclusive;
 	bool	trim;
+
+	u32 alloc_hint_flags;  /* CEPH_OSD_OP_ALLOC_HINT_FLAG_* */
 };
 
 #define RBD_QUEUE_DEPTH_DEFAULT	BLKDEV_MAX_RQ
@@ -2253,7 +2271,8 @@ static void __rbd_osd_setup_write_ops(struct ceph_osd_request *osd_req,
 	    !(obj_req->flags & RBD_OBJ_FLAG_MAY_EXIST)) {
 		osd_req_op_alloc_hint_init(osd_req, which++,
 					   rbd_dev->layout.object_size,
-					   rbd_dev->layout.object_size);
+					   rbd_dev->layout.object_size,
+					   rbd_dev->opts->alloc_hint_flags);
 	}
 
 	if (rbd_obj_is_entire(obj_req))
@@ -6331,6 +6350,29 @@ static int rbd_parse_param(struct fs_parameter *param,
 		pctx->spec->pool_ns = param->string;
 		param->string = NULL;
 		break;
+	case Opt_compression_hint:
+		switch (result.uint_32) {
+		case Opt_compression_hint_none:
+			opt->alloc_hint_flags &=
+			    ~(CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE |
+			      CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE);
+			break;
+		case Opt_compression_hint_compressible:
+			opt->alloc_hint_flags |=
+			    CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE;
+			opt->alloc_hint_flags &=
+			    ~CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE;
+			break;
+		case Opt_compression_hint_incompressible:
+			opt->alloc_hint_flags |=
+			    CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE;
+			opt->alloc_hint_flags &=
+			    ~CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE;
+			break;
+		default:
+			BUG();
+		}
+		break;
 	case Opt_read_only:
 		opt->read_only = true;
 		break;
diff --git a/drivers/block/rbd_types.h b/drivers/block/rbd_types.h
index ac98ab6ccd3b..a600e0eb6b6f 100644
--- a/drivers/block/rbd_types.h
+++ b/drivers/block/rbd_types.h
@@ -93,7 +93,7 @@ struct rbd_image_header_ondisk {
 	__le32 snap_count;
 	__le32 reserved;
 	__le64 snap_names_len;
-	struct rbd_image_snap_ondisk snaps[0];
+	struct rbd_image_snap_ondisk snaps[];
 } __attribute__((packed));
 
 
diff --git a/fs/ceph/Makefile b/fs/ceph/Makefile
index 0a0823d378db..50c635dc7f71 100644
--- a/fs/ceph/Makefile
+++ b/fs/ceph/Makefile
@@ -8,7 +8,7 @@ obj-$(CONFIG_CEPH_FS) += ceph.o
 ceph-y := super.o inode.o dir.o file.o locks.o addr.o ioctl.o \
 	export.o caps.o snap.o xattr.o quota.o io.o \
 	mds_client.o mdsmap.o strings.o ceph_frag.o \
-	debugfs.o util.o
+	debugfs.o util.o metric.o
 
 ceph-$(CONFIG_CEPH_FSCACHE) += cache.o
 ceph-$(CONFIG_CEPH_FS_POSIX_ACL) += acl.o
diff --git a/fs/ceph/acl.c b/fs/ceph/acl.c
index 26be6520d3fb..e0465741c591 100644
--- a/fs/ceph/acl.c
+++ b/fs/ceph/acl.c
@@ -22,7 +22,7 @@ static inline void ceph_set_cached_acl(struct inode *inode,
 	struct ceph_inode_info *ci = ceph_inode(inode);
 
 	spin_lock(&ci->i_ceph_lock);
-	if (__ceph_caps_issued_mask(ci, CEPH_CAP_XATTR_SHARED, 0))
+	if (__ceph_caps_issued_mask_metric(ci, CEPH_CAP_XATTR_SHARED, 0))
 		set_cached_acl(inode, type, acl);
 	else
 		forget_cached_acl(inode, type);
diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
index 6f4678d98df7..01ad09733ac7 100644
--- a/fs/ceph/addr.c
+++ b/fs/ceph/addr.c
@@ -11,10 +11,12 @@
 #include <linux/task_io_accounting_ops.h>
 #include <linux/signal.h>
 #include <linux/iversion.h>
+#include <linux/ktime.h>
 
 #include "super.h"
 #include "mds_client.h"
 #include "cache.h"
+#include "metric.h"
 #include <linux/ceph/osd_client.h>
 #include <linux/ceph/striper.h>
 
@@ -216,6 +218,9 @@ static int ceph_sync_readpages(struct ceph_fs_client *fsc,
 	if (!rc)
 		rc = ceph_osdc_wait_request(osdc, req);
 
+	ceph_update_read_latency(&fsc->mdsc->metric, req->r_start_latency,
+				 req->r_end_latency, rc);
+
 	ceph_osdc_put_request(req);
 	dout("readpages result %d\n", rc);
 	return rc;
@@ -299,6 +304,7 @@ static int ceph_readpage(struct file *filp, struct page *page)
 static void finish_read(struct ceph_osd_request *req)
 {
 	struct inode *inode = req->r_inode;
+	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
 	struct ceph_osd_data *osd_data;
 	int rc = req->r_result <= 0 ? req->r_result : 0;
 	int bytes = req->r_result >= 0 ? req->r_result : 0;
@@ -336,6 +342,10 @@ unlock:
 		put_page(page);
 		bytes -= PAGE_SIZE;
 	}
+
+	ceph_update_read_latency(&fsc->mdsc->metric, req->r_start_latency,
+				 req->r_end_latency, rc);
+
 	kfree(osd_data->pages);
 }
 
@@ -643,6 +653,9 @@ static int ceph_sync_writepages(struct ceph_fs_client *fsc,
 	if (!rc)
 		rc = ceph_osdc_wait_request(osdc, req);
 
+	ceph_update_write_latency(&fsc->mdsc->metric, req->r_start_latency,
+				  req->r_end_latency, rc);
+
 	ceph_osdc_put_request(req);
 	if (rc == 0)
 		rc = len;
@@ -794,6 +807,9 @@ static void writepages_finish(struct ceph_osd_request *req)
 		ceph_clear_error_write(ci);
 	}
 
+	ceph_update_write_latency(&fsc->mdsc->metric, req->r_start_latency,
+				  req->r_end_latency, rc);
+
 	/*
 	 * We lost the cache cap, need to truncate the page before
 	 * it is unlocked, otherwise we'd truncate it later in the
@@ -1852,6 +1868,10 @@ int ceph_uninline_data(struct file *filp, struct page *locked_page)
 	err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
 	if (!err)
 		err = ceph_osdc_wait_request(&fsc->client->osdc, req);
+
+	ceph_update_write_latency(&fsc->mdsc->metric, req->r_start_latency,
+				  req->r_end_latency, err);
+
 out_put:
 	ceph_osdc_put_request(req);
 	if (err == -ECANCELED)
diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index f1acde6fb9a6..972c13aa4225 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -597,6 +597,27 @@ static void __check_cap_issue(struct ceph_inode_info *ci, struct ceph_cap *cap,
 	}
 }
 
+/**
+ * change_auth_cap_ses - move inode to appropriate lists when auth caps change
+ * @ci: inode to be moved
+ * @session: new auth caps session
+ */
+static void change_auth_cap_ses(struct ceph_inode_info *ci,
+				struct ceph_mds_session *session)
+{
+	lockdep_assert_held(&ci->i_ceph_lock);
+
+	if (list_empty(&ci->i_dirty_item) && list_empty(&ci->i_flushing_item))
+		return;
+
+	spin_lock(&session->s_mdsc->cap_dirty_lock);
+	if (!list_empty(&ci->i_dirty_item))
+		list_move(&ci->i_dirty_item, &session->s_cap_dirty);
+	if (!list_empty(&ci->i_flushing_item))
+		list_move_tail(&ci->i_flushing_item, &session->s_cap_flushing);
+	spin_unlock(&session->s_mdsc->cap_dirty_lock);
+}
+
 /*
  * Add a capability under the given MDS session.
  *
@@ -727,6 +748,9 @@ void ceph_add_cap(struct inode *inode,
 	if (flags & CEPH_CAP_FLAG_AUTH) {
 		if (!ci->i_auth_cap ||
 		    ceph_seq_cmp(ci->i_auth_cap->mseq, mseq) < 0) {
+			if (ci->i_auth_cap &&
+			    ci->i_auth_cap->session != cap->session)
+				change_auth_cap_ses(ci, cap->session);
 			ci->i_auth_cap = cap;
 			cap->mds_wanted = wanted;
 		}
@@ -912,6 +936,20 @@ int __ceph_caps_issued_mask(struct ceph_inode_info *ci, int mask, int touch)
 	return 0;
 }
 
+int __ceph_caps_issued_mask_metric(struct ceph_inode_info *ci, int mask,
+				   int touch)
+{
+	struct ceph_fs_client *fsc = ceph_sb_to_client(ci->vfs_inode.i_sb);
+	int r;
+
+	r = __ceph_caps_issued_mask(ci, mask, touch);
+	if (r)
+		ceph_update_cap_hit(&fsc->mdsc->metric);
+	else
+		ceph_update_cap_mis(&fsc->mdsc->metric);
+	return r;
+}
+
 /*
  * Return true if mask caps are currently being revoked by an MDS.
  */
@@ -1109,8 +1147,10 @@ void __ceph_remove_cap(struct ceph_cap *cap, bool queue_release)
 
 	/* remove from inode's cap rbtree, and clear auth cap */
 	rb_erase(&cap->ci_node, &ci->i_caps);
-	if (ci->i_auth_cap == cap)
+	if (ci->i_auth_cap == cap) {
+		WARN_ON_ONCE(!list_empty(&ci->i_dirty_item));
 		ci->i_auth_cap = NULL;
+	}
 
 	/* remove from session list */
 	spin_lock(&session->s_cap_lock);
@@ -1167,6 +1207,7 @@ struct cap_msg_args {
 	u64			xattr_version;
 	u64			change_attr;
 	struct ceph_buffer	*xattr_buf;
+	struct ceph_buffer	*old_xattr_buf;
 	struct timespec64	atime, mtime, ctime, btime;
 	int			op, caps, wanted, dirty;
 	u32			seq, issue_seq, mseq, time_warp_seq;
@@ -1175,6 +1216,7 @@ struct cap_msg_args {
 	kgid_t			gid;
 	umode_t			mode;
 	bool			inline_data;
+	bool			wake;
 };
 
 /*
@@ -1304,44 +1346,29 @@ void __ceph_remove_caps(struct ceph_inode_info *ci)
 }
 
 /*
- * Send a cap msg on the given inode.  Update our caps state, then
- * drop i_ceph_lock and send the message.
+ * Prepare to send a cap message to an MDS. Update the cap state, and populate
+ * the arg struct with the parameters that will need to be sent. This should
+ * be done under the i_ceph_lock to guard against changes to cap state.
  *
  * Make note of max_size reported/requested from mds, revoked caps
  * that have now been implemented.
- *
- * Return non-zero if delayed release, or we experienced an error
- * such that the caller should requeue + retry later.
- *
- * called with i_ceph_lock, then drops it.
- * caller should hold snap_rwsem (read), s_mutex.
  */
-static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
-		      int op, int flags, int used, int want, int retain,
-		      int flushing, u64 flush_tid, u64 oldest_flush_tid)
-	__releases(cap->ci->i_ceph_lock)
+static void __prep_cap(struct cap_msg_args *arg, struct ceph_cap *cap,
+		       int op, int flags, int used, int want, int retain,
+		       int flushing, u64 flush_tid, u64 oldest_flush_tid)
 {
 	struct ceph_inode_info *ci = cap->ci;
 	struct inode *inode = &ci->vfs_inode;
-	struct ceph_buffer *old_blob = NULL;
-	struct cap_msg_args arg;
 	int held, revoking;
-	int wake = 0;
-	int ret;
 
-	/* Don't send anything if it's still being created. Return delayed */
-	if (ci->i_ceph_flags & CEPH_I_ASYNC_CREATE) {
-		spin_unlock(&ci->i_ceph_lock);
-		dout("%s async create in flight for %p\n", __func__, inode);
-		return 1;
-	}
+	lockdep_assert_held(&ci->i_ceph_lock);
 
 	held = cap->issued | cap->implemented;
 	revoking = cap->implemented & ~cap->issued;
 	retain &= ~revoking;
 
-	dout("__send_cap %p cap %p session %p %s -> %s (revoking %s)\n",
-	     inode, cap, cap->session,
+	dout("%s %p cap %p session %p %s -> %s (revoking %s)\n",
+	     __func__, inode, cap, cap->session,
 	     ceph_cap_string(held), ceph_cap_string(held & retain),
 	     ceph_cap_string(revoking));
 	BUG_ON((retain & CEPH_CAP_PIN) == 0);
@@ -1349,60 +1376,62 @@ static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
 	ci->i_ceph_flags &= ~CEPH_I_FLUSH;
 
 	cap->issued &= retain;  /* drop bits we don't want */
-	if (cap->implemented & ~cap->issued) {
-		/*
-		 * Wake up any waiters on wanted -> needed transition.
-		 * This is due to the weird transition from buffered
-		 * to sync IO... we need to flush dirty pages _before_
-		 * allowing sync writes to avoid reordering.
-		 */
-		wake = 1;
-	}
+	/*
+	 * Wake up any waiters on wanted -> needed transition. This is due to
+	 * the weird transition from buffered to sync IO... we need to flush
+	 * dirty pages _before_ allowing sync writes to avoid reordering.
+	 */
+	arg->wake = cap->implemented & ~cap->issued;
 	cap->implemented &= cap->issued | used;
 	cap->mds_wanted = want;
 
-	arg.session = cap->session;
-	arg.ino = ceph_vino(inode).ino;
-	arg.cid = cap->cap_id;
-	arg.follows = flushing ? ci->i_head_snapc->seq : 0;
-	arg.flush_tid = flush_tid;
-	arg.oldest_flush_tid = oldest_flush_tid;
-
-	arg.size = inode->i_size;
-	ci->i_reported_size = arg.size;
-	arg.max_size = ci->i_wanted_max_size;
-	if (cap == ci->i_auth_cap)
-		ci->i_requested_max_size = arg.max_size;
+	arg->session = cap->session;
+	arg->ino = ceph_vino(inode).ino;
+	arg->cid = cap->cap_id;
+	arg->follows = flushing ? ci->i_head_snapc->seq : 0;
+	arg->flush_tid = flush_tid;
+	arg->oldest_flush_tid = oldest_flush_tid;
+
+	arg->size = inode->i_size;
+	ci->i_reported_size = arg->size;
+	arg->max_size = ci->i_wanted_max_size;
+	if (cap == ci->i_auth_cap) {
+		if (want & CEPH_CAP_ANY_FILE_WR)
+			ci->i_requested_max_size = arg->max_size;
+		else
+			ci->i_requested_max_size = 0;
+	}
 
 	if (flushing & CEPH_CAP_XATTR_EXCL) {
-		old_blob = __ceph_build_xattrs_blob(ci);
-		arg.xattr_version = ci->i_xattrs.version;
-		arg.xattr_buf = ci->i_xattrs.blob;
+		arg->old_xattr_buf = __ceph_build_xattrs_blob(ci);
+		arg->xattr_version = ci->i_xattrs.version;
+		arg->xattr_buf = ci->i_xattrs.blob;
 	} else {
-		arg.xattr_buf = NULL;
+		arg->xattr_buf = NULL;
+		arg->old_xattr_buf = NULL;
 	}
 
-	arg.mtime = inode->i_mtime;
-	arg.atime = inode->i_atime;
-	arg.ctime = inode->i_ctime;
-	arg.btime = ci->i_btime;
-	arg.change_attr = inode_peek_iversion_raw(inode);
+	arg->mtime = inode->i_mtime;
+	arg->atime = inode->i_atime;
+	arg->ctime = inode->i_ctime;
+	arg->btime = ci->i_btime;
+	arg->change_attr = inode_peek_iversion_raw(inode);
 
-	arg.op = op;
-	arg.caps = cap->implemented;
-	arg.wanted = want;
-	arg.dirty = flushing;
+	arg->op = op;
+	arg->caps = cap->implemented;
+	arg->wanted = want;
+	arg->dirty = flushing;
 
-	arg.seq = cap->seq;
-	arg.issue_seq = cap->issue_seq;
-	arg.mseq = cap->mseq;
-	arg.time_warp_seq = ci->i_time_warp_seq;
+	arg->seq = cap->seq;
+	arg->issue_seq = cap->issue_seq;
+	arg->mseq = cap->mseq;
+	arg->time_warp_seq = ci->i_time_warp_seq;
 
-	arg.uid = inode->i_uid;
-	arg.gid = inode->i_gid;
-	arg.mode = inode->i_mode;
+	arg->uid = inode->i_uid;
+	arg->gid = inode->i_gid;
+	arg->mode = inode->i_mode;
 
-	arg.inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
+	arg->inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
 	if (!(flags & CEPH_CLIENT_CAPS_PENDING_CAPSNAP) &&
 	    !list_empty(&ci->i_cap_snaps)) {
 		struct ceph_cap_snap *capsnap;
@@ -1415,27 +1444,35 @@ static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
 			}
 		}
 	}
-	arg.flags = flags;
-
-	spin_unlock(&ci->i_ceph_lock);
+	arg->flags = flags;
+}
 
-	ceph_buffer_put(old_blob);
+/*
+ * Send a cap msg on the given inode.
+ *
+ * Caller should hold snap_rwsem (read), s_mutex.
+ */
+static void __send_cap(struct ceph_mds_client *mdsc, struct cap_msg_args *arg,
+		       struct ceph_inode_info *ci)
+{
+	struct inode *inode = &ci->vfs_inode;
+	int ret;
 
-	ret = send_cap_msg(&arg);
+	ret = send_cap_msg(arg);
 	if (ret < 0) {
 		pr_err("error sending cap msg, ino (%llx.%llx) "
 		       "flushing %s tid %llu, requeue\n",
-		       ceph_vinop(inode), ceph_cap_string(flushing),
-		       flush_tid);
+		       ceph_vinop(inode), ceph_cap_string(arg->dirty),
+		       arg->flush_tid);
 		spin_lock(&ci->i_ceph_lock);
 		__cap_delay_requeue(mdsc, ci);
 		spin_unlock(&ci->i_ceph_lock);
 	}
 
-	if (wake)
-		wake_up_all(&ci->i_cap_wq);
+	ceph_buffer_put(arg->old_xattr_buf);
 
-	return ret;
+	if (arg->wake)
+		wake_up_all(&ci->i_cap_wq);
 }
 
 static inline int __send_flush_snap(struct inode *inode,
@@ -1456,6 +1493,7 @@ static inline int __send_flush_snap(struct inode *inode,
 	arg.max_size = 0;
 	arg.xattr_version = capsnap->xattr_version;
 	arg.xattr_buf = capsnap->xattr_blob;
+	arg.old_xattr_buf = NULL;
 
 	arg.atime = capsnap->atime;
 	arg.mtime = capsnap->mtime;
@@ -1479,6 +1517,7 @@ static inline int __send_flush_snap(struct inode *inode,
 
 	arg.inline_data = capsnap->inline_data;
 	arg.flags = 0;
+	arg.wake = false;
 
 	return send_cap_msg(&arg);
 }
@@ -1676,6 +1715,8 @@ int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask,
 	     ceph_cap_string(was | mask));
 	ci->i_dirty_caps |= mask;
 	if (was == 0) {
+		struct ceph_mds_session *session = ci->i_auth_cap->session;
+
 		WARN_ON_ONCE(ci->i_prealloc_cap_flush);
 		swap(ci->i_prealloc_cap_flush, *pcf);
 
@@ -1688,7 +1729,7 @@ int __ceph_mark_dirty_caps(struct ceph_inode_info *ci, int mask,
 		     &ci->vfs_inode, ci->i_head_snapc, ci->i_auth_cap);
 		BUG_ON(!list_empty(&ci->i_dirty_item));
 		spin_lock(&mdsc->cap_dirty_lock);
-		list_add(&ci->i_dirty_item, &mdsc->cap_dirty);
+		list_add(&ci->i_dirty_item, &session->s_cap_dirty);
 		spin_unlock(&mdsc->cap_dirty_lock);
 		if (ci->i_flushing_caps == 0) {
 			ihold(inode);
@@ -1731,30 +1772,33 @@ static u64 __get_oldest_flush_tid(struct ceph_mds_client *mdsc)
  * Remove cap_flush from the mdsc's or inode's flushing cap list.
  * Return true if caller needs to wake up flush waiters.
  */
-static bool __finish_cap_flush(struct ceph_mds_client *mdsc,
-			       struct ceph_inode_info *ci,
-			       struct ceph_cap_flush *cf)
+static bool __detach_cap_flush_from_mdsc(struct ceph_mds_client *mdsc,
+					 struct ceph_cap_flush *cf)
 {
 	struct ceph_cap_flush *prev;
 	bool wake = cf->wake;
-	if (mdsc) {
-		/* are there older pending cap flushes? */
-		if (wake && cf->g_list.prev != &mdsc->cap_flush_list) {
-			prev = list_prev_entry(cf, g_list);
-			prev->wake = true;
-			wake = false;
-		}
-		list_del(&cf->g_list);
-	} else if (ci) {
-		if (wake && cf->i_list.prev != &ci->i_cap_flush_list) {
-			prev = list_prev_entry(cf, i_list);
-			prev->wake = true;
-			wake = false;
-		}
-		list_del(&cf->i_list);
-	} else {
-		BUG_ON(1);
+
+	if (wake && cf->g_list.prev != &mdsc->cap_flush_list) {
+		prev = list_prev_entry(cf, g_list);
+		prev->wake = true;
+		wake = false;
 	}
+	list_del(&cf->g_list);
+	return wake;
+}
+
+static bool __detach_cap_flush_from_ci(struct ceph_inode_info *ci,
+				       struct ceph_cap_flush *cf)
+{
+	struct ceph_cap_flush *prev;
+	bool wake = cf->wake;
+
+	if (wake && cf->i_list.prev != &ci->i_cap_flush_list) {
+		prev = list_prev_entry(cf, i_list);
+		prev->wake = true;
+		wake = false;
+	}
+	list_del(&cf->i_list);
 	return wake;
 }
 
@@ -1953,6 +1997,9 @@ retry_locked:
 	}
 
 	for (p = rb_first(&ci->i_caps); p; p = rb_next(p)) {
+		int mflags = 0;
+		struct cap_msg_args arg;
+
 		cap = rb_entry(p, struct ceph_cap, ci_node);
 
 		/* avoid looping forever */
@@ -2030,12 +2077,24 @@ ack:
 			if (mutex_trylock(&session->s_mutex) == 0) {
 				dout("inverting session/ino locks on %p\n",
 				     session);
+				session = ceph_get_mds_session(session);
 				spin_unlock(&ci->i_ceph_lock);
 				if (took_snap_rwsem) {
 					up_read(&mdsc->snap_rwsem);
 					took_snap_rwsem = 0;
 				}
-				mutex_lock(&session->s_mutex);
+				if (session) {
+					mutex_lock(&session->s_mutex);
+					ceph_put_mds_session(session);
+				} else {
+					/*
+					 * Because we take the reference while
+					 * holding the i_ceph_lock, it should
+					 * never be NULL. Throw a warning if it
+					 * ever is.
+					 */
+					WARN_ON_ONCE(true);
+				}
 				goto retry;
 			}
 		}
@@ -2070,6 +2129,9 @@ ack:
 			flushing = ci->i_dirty_caps;
 			flush_tid = __mark_caps_flushing(inode, session, false,
 							 &oldest_flush_tid);
+			if (flags & CHECK_CAPS_FLUSH &&
+			    list_empty(&session->s_cap_dirty))
+				mflags |= CEPH_CLIENT_CAPS_SYNC;
 		} else {
 			flushing = 0;
 			flush_tid = 0;
@@ -2080,9 +2142,12 @@ ack:
 
 		mds = cap->mds;  /* remember mds, so we don't repeat */
 
-		/* __send_cap drops i_ceph_lock */
-		__send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, 0, cap_used, want,
-			   retain, flushing, flush_tid, oldest_flush_tid);
+		__prep_cap(&arg, cap, CEPH_CAP_OP_UPDATE, mflags, cap_used,
+			   want, retain, flushing, flush_tid, oldest_flush_tid);
+		spin_unlock(&ci->i_ceph_lock);
+
+		__send_cap(mdsc, &arg, ci);
+
 		goto retry; /* retake i_ceph_lock and restart our cap scan. */
 	}
 
@@ -2121,6 +2186,7 @@ retry:
 retry_locked:
 	if (ci->i_dirty_caps && ci->i_auth_cap) {
 		struct ceph_cap *cap = ci->i_auth_cap;
+		struct cap_msg_args arg;
 
 		if (session != cap->session) {
 			spin_unlock(&ci->i_ceph_lock);
@@ -2148,11 +2214,13 @@ retry_locked:
 		flush_tid = __mark_caps_flushing(inode, session, true,
 						 &oldest_flush_tid);
 
-		/* __send_cap drops i_ceph_lock */
-		__send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, CEPH_CLIENT_CAPS_SYNC,
+		__prep_cap(&arg, cap, CEPH_CAP_OP_FLUSH, CEPH_CLIENT_CAPS_SYNC,
 			   __ceph_caps_used(ci), __ceph_caps_wanted(ci),
 			   (cap->issued | cap->implemented),
 			   flushing, flush_tid, oldest_flush_tid);
+		spin_unlock(&ci->i_ceph_lock);
+
+		__send_cap(mdsc, &arg, ci);
 	} else {
 		if (!list_empty(&ci->i_cap_flush_list)) {
 			struct ceph_cap_flush *cf =
@@ -2354,15 +2422,19 @@ static void __kick_flushing_caps(struct ceph_mds_client *mdsc,
 		first_tid = cf->tid + 1;
 
 		if (cf->caps) {
+			struct cap_msg_args arg;
+
 			dout("kick_flushing_caps %p cap %p tid %llu %s\n",
 			     inode, cap, cf->tid, ceph_cap_string(cf->caps));
-			__send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
+			__prep_cap(&arg, cap, CEPH_CAP_OP_FLUSH,
 					 (cf->tid < last_snap_flush ?
 					  CEPH_CLIENT_CAPS_PENDING_CAPSNAP : 0),
 					  __ceph_caps_used(ci),
 					  __ceph_caps_wanted(ci),
 					  (cap->issued | cap->implemented),
 					  cf->caps, cf->tid, oldest_flush_tid);
+			spin_unlock(&ci->i_ceph_lock);
+			__send_cap(mdsc, &arg, ci);
 		} else {
 			struct ceph_cap_snap *capsnap =
 					container_of(cf, struct ceph_cap_snap,
@@ -2446,6 +2518,8 @@ void ceph_kick_flushing_caps(struct ceph_mds_client *mdsc,
 	struct ceph_cap *cap;
 	u64 oldest_flush_tid;
 
+	lockdep_assert_held(&session->s_mutex);
+
 	dout("kick_flushing_caps mds%d\n", session->s_mds);
 
 	spin_lock(&mdsc->cap_dirty_lock);
@@ -2685,6 +2759,11 @@ out_unlock:
 	if (snap_rwsem_locked)
 		up_read(&mdsc->snap_rwsem);
 
+	if (!ret)
+		ceph_update_cap_mis(&mdsc->metric);
+	else if (ret == 1)
+		ceph_update_cap_hit(&mdsc->metric);
+
 	dout("get_cap_refs %p ret %d got %s\n", inode,
 	     ret, ceph_cap_string(*got));
 	return ret;
@@ -2937,7 +3016,8 @@ static int ceph_try_drop_cap_snap(struct ceph_inode_info *ci,
  * If we are releasing a WR cap (from a sync write), finalize any affected
  * cap_snap, and wake up any waiters.
  */
-void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
+static void __ceph_put_cap_refs(struct ceph_inode_info *ci, int had,
+				bool skip_checking_caps)
 {
 	struct inode *inode = &ci->vfs_inode;
 	int last = 0, put = 0, flushsnaps = 0, wake = 0;
@@ -2993,7 +3073,7 @@ void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
 	dout("put_cap_refs %p had %s%s%s\n", inode, ceph_cap_string(had),
 	     last ? " last" : "", put ? " put" : "");
 
-	if (last)
+	if (last && !skip_checking_caps)
 		ceph_check_caps(ci, 0, NULL);
 	else if (flushsnaps)
 		ceph_flush_snaps(ci, NULL);
@@ -3003,6 +3083,16 @@ void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
 		iput(inode);
 }
 
+void ceph_put_cap_refs(struct ceph_inode_info *ci, int had)
+{
+	__ceph_put_cap_refs(ci, had, false);
+}
+
+void ceph_put_cap_refs_no_check_caps(struct ceph_inode_info *ci, int had)
+{
+	__ceph_put_cap_refs(ci, had, true);
+}
+
 /*
  * Release @nr WRBUFFER refs on dirty pages for the given @snapc snap
  * context.  Adjust per-snap dirty page accounting as appropriate.
@@ -3301,10 +3391,6 @@ static void handle_cap_grant(struct inode *inode,
 				ci->i_requested_max_size = 0;
 			}
 			wake = true;
-		} else if (ci->i_wanted_max_size > ci->i_max_size &&
-			   ci->i_wanted_max_size > ci->i_requested_max_size) {
-			/* CEPH_CAP_OP_IMPORT */
-			wake = true;
 		}
 	}
 
@@ -3380,9 +3466,18 @@ static void handle_cap_grant(struct inode *inode,
 			fill_inline = true;
 	}
 
-	if (le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) {
+	if (ci->i_auth_cap == cap &&
+	    le32_to_cpu(grant->op) == CEPH_CAP_OP_IMPORT) {
 		if (newcaps & ~extra_info->issued)
 			wake = true;
+
+		if (ci->i_requested_max_size > max_size ||
+		    !(le32_to_cpu(grant->wanted) & CEPH_CAP_ANY_FILE_WR)) {
+			/* re-request max_size if necessary */
+			ci->i_requested_max_size = 0;
+			wake = true;
+		}
+
 		ceph_kick_flushing_inode_caps(session, ci);
 		spin_unlock(&ci->i_ceph_lock);
 		up_read(&session->s_mdsc->snap_rwsem);
@@ -3442,15 +3537,26 @@ static void handle_cap_flush_ack(struct inode *inode, u64 flush_tid,
 	bool wake_mdsc = false;
 
 	list_for_each_entry_safe(cf, tmp_cf, &ci->i_cap_flush_list, i_list) {
+		/* Is this the one that was flushed? */
 		if (cf->tid == flush_tid)
 			cleaned = cf->caps;
-		if (cf->caps == 0) /* capsnap */
+
+		/* Is this a capsnap? */