afs: Overhaul invalidation handling to better support RO volumes

Overhaul the third party-induced invalidation handling, making use of the
previously added volume-level event counters (cb_scrub and cb_ro_snapshot)
that are now being parsed out of the VolSync record returned by the
fileserver in many of its replies.

This allows better handling of RO (and Backup) volumes.  Since these are
snapshot of a RW volume that are updated atomically simultantanously across
all servers that host them, they only require a single callback promise for
the entire volume.  The currently upstream code assumes that RO volumes
operate in the same manner as RW volumes, and that each file has its own
individual callback - which means that it does a status fetch for *every*
file in a RO volume, whether or not the volume got "released" (volume
callback breaks can occur for other reasons too, such as the volumeserver
taking ownership of a volume from a fileserver).

To this end, make the following changes:

 (1) Change the meaning of the volume's cb_v_break counter so that it is
     now a hint that we need to issue a status fetch to work out the state
     of a volume.  cb_v_break is incremented by volume break callbacks and
     by server initialisation callbacks.

 (2) Add a second counter, cb_v_check, to the afs_volume struct such that
     if this differs from cb_v_break, we need to do a check.  When the
     check is complete, cb_v_check is advanced to what cb_v_break was at
     the start of the status fetch.

 (3) Move the list of mmap'd vnodes to the volume and trigger removal of
     PTEs that map to files on a volume break rather than on a server
     break.

 (4) When a server reinitialisation callback comes in, use the
     server-to-volume reverse mapping added in a preceding patch to iterate
     over all the volumes using that server and clear the volume callback
     promises for that server and the general volume promise as a whole to
     trigger reanalysis.

 (5) Replace the AFS_VNODE_CB_PROMISED flag with an AFS_NO_CB_PROMISE
     (TIME64_MIN) value in the cb_expires_at field, reducing the number of
     checks we need to make.

 (6) Change afs_check_validity() to quickly see if various event counters
     have been incremented or if the vnode or volume callback promise is
     due to expire/has expired without making any changes to the state.
     That is now left to afs_validate() as this may get more complicated in
     future as we may have to examine server records too.

 (7) Overhaul afs_validate() so that it does a single status fetch if we
     need to check the state of either the vnode or the volume - and do so
     under appropriate locking.  The function does the following steps:

     (A) If the vnode/volume is no longer seen as valid, then we take the
     vnode validation lock and, if the volume promise has expired, the
     volume check lock also.  The latter prevents redundant checks being
     made to find out if a new version of the volume got released.

     (B) If a previous RPC call found that the volsync changed unexpectedly
     or that a RO volume was updated, then we unmap all PTEs pointing to
     the file to stop mmap being used for access.

     (C) If the vnode is still seen to be of uncertain validity, then we
     perform an FS.FetchStatus RPC op to jointly update the volume status
     and the vnode status.  This assessment is done as part of parsing the
     reply:

	If the RO volume creation timestamp advances, cb_ro_snapshot is
	incremented; if either the creation or update timestamps changes in
	an unexpected way, the cb_scrub counter is incremented

	If the Data Version returned doesn't match the copy we have
	locally, then we ask for the pagecache to be zapped.  This takes
	care of handling RO update.

     (D) If cb_scrub differs between volume and vnode, the vnode's
     pagecache is zapped and the vnode's cb_scrub is updated unless the
     file is marked as having been deleted.

Signed-off-by: David Howells <dhowells@redhat.com>
cc: Marc Dionne <marc.dionne@auristor.com>
cc: linux-afs@lists.infradead.org

1 files changed, 200 insertions, 105 deletions

diff --git a/fs/afs/validation.c b/fs/afs/validation.c
index 6aadd5e075e4..46b37f2cce7d 100644
--- a/fs/afs/validation.c
+++ b/fs/afs/validation.c
@@ -11,6 +11,131 @@
 #include "internal.h"
 
 /*
+ * Data validation is managed through a number of mechanisms from the server:
+ *
+ *  (1) On first contact with a server (such as if it has just been rebooted),
+ *      the server sends us a CB.InitCallBackState* request.
+ *
+ *  (2) On a RW volume, in response to certain vnode (inode)-accessing RPC
+ *      calls, the server maintains a time-limited per-vnode promise that it
+ *      will send us a CB.CallBack request if a third party alters the vnodes
+ *      accessed.
+ *
+ *      Note that a vnode-level callbacks may also be sent for other reasons,
+ *      such as filelock release.
+ *
+ *  (3) On a RO (or Backup) volume, in response to certain vnode-accessing RPC
+ *      calls, each server maintains a time-limited per-volume promise that it
+ *      will send us a CB.CallBack request if the RO volume is updated to a
+ *      snapshot of the RW volume ("vos release").  This is an atomic event
+ *      that cuts over all instances of the RO volume across multiple servers
+ *      simultaneously.
+ *
+ *	Note that a volume-level callbacks may also be sent for other reasons,
+ *	such as the volumeserver taking over control of the volume from the
+ *	fileserver.
+ *
+ *	Note also that each server maintains an independent time limit on an
+ *	independent callback.
+ *
+ *  (4) Certain RPC calls include a volume information record "VolSync" in
+ *      their reply.  This contains a creation date for the volume that should
+ *      remain unchanged for a RW volume (but will be changed if the volume is
+ *      restored from backup) or will be bumped to the time of snapshotting
+ *      when a RO volume is released.
+ *
+ * In order to track this events, the following are provided:
+ *
+ *	->cb_v_break.  A counter of events that might mean that the contents of
+ *	a volume have been altered since we last checked a vnode.
+ *
+ *	->cb_v_check.  A counter of the number of events that we've sent a
+ *	query to the server for.  Everything's up to date if this equals
+ *	cb_v_break.
+ *
+ *	->cb_scrub.  A counter of the number of regression events for which we
+ *	have to completely wipe the cache.
+ *
+ *	->cb_ro_snapshot.  A counter of the number of times that we've
+ *      recognised that a RO volume has been updated.
+ *
+ *	->cb_break.  A counter of events that might mean that the contents of a
+ *      vnode have been altered.
+ *
+ *	->cb_expires_at.  The time at which the callback promise expires or
+ *      AFS_NO_CB_PROMISE if we have no promise.
+ *
+ * The way we manage things is:
+ *
+ *  (1) When a volume-level CB.CallBack occurs, we increment ->cb_v_break on
+ *      the volume and reset ->cb_expires_at (ie. set AFS_NO_CB_PROMISE) on the
+ *      volume and volume's server record.
+ *
+ *  (2) When a CB.InitCallBackState occurs, we treat this as a volume-level
+ *	callback break on all the volumes that have been using that volume
+ *	(ie. increment ->cb_v_break and reset ->cb_expires_at).
+ *
+ *  (3) When a vnode-level CB.CallBack occurs, we increment ->cb_break on the
+ *	vnode and reset its ->cb_expires_at.  If the vnode is mmapped, we also
+ *	dispatch a work item to unmap all PTEs to the vnode's pagecache to
+ *	force reentry to the filesystem for revalidation.
+ *
+ *  (4) When entering the filesystem, we call afs_validate() to check the
+ *	validity of a vnode.  This first checks to see if ->cb_v_check and
+ *	->cb_v_break match, and if they don't, we lock volume->cb_check_lock
+ *	exclusively and perform an FS.FetchStatus on the vnode.
+ *
+ *	After checking the volume, we check the vnode.  If there's a mismatch
+ *	between the volume counters and the vnode's mirrors of those counters,
+ *	we lock vnode->validate_lock and issue an FS.FetchStatus on the vnode.
+ *
+ *  (5) When the reply from FS.FetchStatus arrives, the VolSync record is
+ *      parsed:
+ *
+ *	(A) If the Creation timestamp has changed on a RW volume or regressed
+ *	    on a RO volume, we try to increment ->cb_scrub; if it advances on a
+ *	    RO volume, we assume "vos release" happened and try to increment
+ *	    ->cb_ro_snapshot.
+ *
+ *      (B) If the Update timestamp has regressed, we try to increment
+ *	    ->cb_scrub.
+ *
+ *      Note that in both of these cases, we only do the increment if we can
+ *      cmpxchg the value of the timestamp from the value we noted before the
+ *      op.  This tries to prevent parallel ops from fighting one another.
+ *
+ *	volume->cb_v_check is then set to ->cb_v_break.
+ *
+ *  (6) The AFSCallBack record included in the FS.FetchStatus reply is also
+ *	parsed and used to set the promise in ->cb_expires_at for the vnode,
+ *	the volume and the volume's server record.
+ *
+ *  (7) If ->cb_scrub is seen to have advanced, we invalidate the pagecache for
+ *      the vnode.
+ */
+
+/*
+ * Check the validity of a vnode/inode and its parent volume.
+ */
+bool afs_check_validity(const struct afs_vnode *vnode)
+{
+	const struct afs_volume *volume = vnode->volume;
+	time64_t deadline = ktime_get_real_seconds() + 10;
+
+	if (atomic_read(&volume->cb_v_check) != atomic_read(&volume->cb_v_break) ||
+	    atomic64_read(&vnode->cb_expires_at)  <= deadline ||
+	    volume->cb_expires_at <= deadline ||
+	    vnode->cb_ro_snapshot != atomic_read(&volume->cb_ro_snapshot) ||
+	    vnode->cb_scrub	  != atomic_read(&volume->cb_scrub) ||
+	    test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags)) {
+		_debug("inval");
+		return false;
+	}
+
+	return true;
+}
+
+/*
  * See if the server we've just talked to is currently excluded.
  */
 static bool __afs_is_server_excluded(struct afs_operation *op, struct afs_volume *volume)
@@ -185,11 +310,17 @@ out:
 }
 
 /*
- * Update the state of a volume.  Returns 1 to redo the operation from the start.
+ * Update the state of a volume, including recording the expiration time of the
+ * callback promise.  Returns 1 to redo the operation from the start.
  */
 int afs_update_volume_state(struct afs_operation *op)
 {
+	struct afs_server_list *slist = op->server_list;
+	struct afs_server_entry *se = &slist->servers[op->server_index];
+	struct afs_callback *cb = &op->file[0].scb.callback;
 	struct afs_volume *volume = op->volume;
+	unsigned int cb_v_break = atomic_read(&volume->cb_v_break);
+	unsigned int cb_v_check = atomic_read(&volume->cb_v_check);
 	int ret;
 
 	_enter("%llx", op->volume->vid);
@@ -202,6 +333,18 @@ int afs_update_volume_state(struct afs_operation *op)
 		}
 	}
 
+	if (op->cb_v_break == cb_v_break &&
+	    (op->file[0].scb.have_cb || op->file[1].scb.have_cb)) {
+		time64_t expires_at = cb->expires_at;
+
+		if (!op->file[0].scb.have_cb)
+			expires_at = op->file[1].scb.callback.expires_at;
+
+		se->cb_expires_at = expires_at;
+		volume->cb_expires_at = expires_at;
+	}
+	if (cb_v_check < op->cb_v_break)
+		atomic_cmpxchg(&volume->cb_v_check, cb_v_check, op->cb_v_break);
 	return 0;
 }
 
@@ -225,99 +368,6 @@ static void afs_zap_data(struct afs_vnode *vnode)
 }
 
 /*
- * Check to see if we have a server currently serving this volume and that it
- * hasn't been reinitialised or dropped from the list.
- */
-static bool afs_check_server_good(struct afs_vnode *vnode)
-{
-	struct afs_server_list *slist;
-	struct afs_server *server;
-	bool good;
-	int i;
-
-	if (vnode->cb_fs_s_break == atomic_read(&vnode->volume->cell->fs_s_break))
-		return true;
-
-	rcu_read_lock();
-
-	slist = rcu_dereference(vnode->volume->servers);
-	for (i = 0; i < slist->nr_servers; i++) {
-		server = slist->servers[i].server;
-		if (server == vnode->cb_server) {
-			good = (vnode->cb_s_break == server->cb_s_break);
-			rcu_read_unlock();
-			return good;
-		}
-	}
-
-	rcu_read_unlock();
-	return false;
-}
-
-/*
- * Check the validity of a vnode/inode.
- */
-bool afs_check_validity(struct afs_vnode *vnode)
-{
-	enum afs_cb_break_reason need_clear = afs_cb_break_no_break;
-	time64_t now = ktime_get_real_seconds();
-	unsigned int cb_break;
-	int seq;
-
-	do {
-		seq = read_seqbegin(&vnode->cb_lock);
-		cb_break = vnode->cb_break;
-
-		if (test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
-			if (vnode->cb_v_break != atomic_read(&vnode->volume->cb_v_break))
-				need_clear = afs_cb_break_for_v_break;
-			else if (!afs_check_server_good(vnode))
-				need_clear = afs_cb_break_for_s_reinit;
-			else if (test_bit(AFS_VNODE_ZAP_DATA, &vnode->flags))
-				need_clear = afs_cb_break_for_zap;
-			else if (vnode->cb_expires_at - 10 <= now)
-				need_clear = afs_cb_break_for_lapsed;
-		} else if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
-			;
-		} else {
-			need_clear = afs_cb_break_no_promise;
-		}
-
-	} while (read_seqretry(&vnode->cb_lock, seq));
-
-	if (need_clear == afs_cb_break_no_break)
-		return true;
-
-	write_seqlock(&vnode->cb_lock);
-	if (need_clear == afs_cb_break_no_promise)
-		vnode->cb_v_break = atomic_read(&vnode->volume->cb_v_break);
-	else if (cb_break == vnode->cb_break)
-		__afs_break_callback(vnode, need_clear);
-	else
-		trace_afs_cb_miss(&vnode->fid, need_clear);
-	write_sequnlock(&vnode->cb_lock);
-	return false;
-}
-
-/*
- * Returns true if the pagecache is still valid.  Does not sleep.
- */
-bool afs_pagecache_valid(struct afs_vnode *vnode)
-{
-	if (unlikely(test_bit(AFS_VNODE_DELETED, &vnode->flags))) {
-		if (vnode->netfs.inode.i_nlink)
-			clear_nlink(&vnode->netfs.inode);
-		return true;
-	}
-
-	if (test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags) &&
-	    afs_check_validity(vnode))
-		return true;
-
-	return false;
-}
-
-/*
  * validate a vnode/inode
  * - there are several things we need to check
  *   - parent dir data changes (rm, rmdir, rename, mkdir, create, link,
@@ -328,23 +378,48 @@ bool afs_pagecache_valid(struct afs_vnode *vnode)
  */
 int afs_validate(struct afs_vnode *vnode, struct key *key)
 {
+	struct afs_volume *volume = vnode->volume;
+	unsigned int cb_ro_snapshot, cb_scrub;
+	time64_t deadline = ktime_get_real_seconds() + 10;
+	bool zap = false, locked_vol = false;
 	int ret;
 
 	_enter("{v={%llx:%llu} fl=%lx},%x",
 	       vnode->fid.vid, vnode->fid.vnode, vnode->flags,
 	       key_serial(key));
 
-	if (afs_pagecache_valid(vnode))
-		goto valid;
+	if (afs_check_validity(vnode))
+		return 0;
 
-	down_write(&vnode->validate_lock);
+	ret = down_write_killable(&vnode->validate_lock);
+	if (ret < 0)
+		goto error;
+
+	/* Validate a volume after the v_break has changed or the volume
+	 * callback expired.  We only want to do this once per volume per
+	 * v_break change.  The actual work will be done when parsing the
+	 * status fetch reply.
+	 */
+	if (volume->cb_expires_at <= deadline ||
+	    atomic_read(&volume->cb_v_check) != atomic_read(&volume->cb_v_break)) {
+		ret = mutex_lock_interruptible(&volume->cb_check_lock);
+		if (ret < 0)
+			goto error_unlock;
+		locked_vol = true;
+	}
 
-	/* if the promise has expired, we need to check the server again to get
-	 * a new promise - note that if the (parent) directory's metadata was
-	 * changed then the security may be different and we may no longer have
-	 * access */
-	if (!test_bit(AFS_VNODE_CB_PROMISED, &vnode->flags)) {
-		_debug("not promised");
+	cb_ro_snapshot = atomic_read(&volume->cb_ro_snapshot);
+	cb_scrub = atomic_read(&volume->cb_scrub);
+	if (vnode->cb_ro_snapshot != cb_ro_snapshot ||
+	    vnode->cb_scrub	  != cb_scrub)
+		unmap_mapping_pages(vnode->netfs.inode.i_mapping, 0, 0, false);
+
+	if (vnode->cb_ro_snapshot != cb_ro_snapshot ||
+	    vnode->cb_scrub	  != cb_scrub ||
+	    volume->cb_expires_at <= deadline ||
+	    atomic_read(&volume->cb_v_check) != atomic_read(&volume->cb_v_break) ||
+	    atomic64_read(&vnode->cb_expires_at) <= deadline
+	    ) {
 		ret = afs_fetch_status(vnode, key, false, NULL);
 		if (ret < 0) {
 			if (ret == -ENOENT) {
@@ -353,9 +428,26 @@ int afs_validate(struct afs_vnode *vnode, struct key *key)
 			}
 			goto error_unlock;
 		}
+
 		_debug("new promise [fl=%lx]", vnode->flags);
 	}
 
+	/* We can drop the volume lock now as. */
+	if (locked_vol) {
+		mutex_unlock(&volume->cb_check_lock);
+		locked_vol = false;
+	}
+
+	cb_ro_snapshot = atomic_read(&volume->cb_ro_snapshot);
+	cb_scrub = atomic_read(&volume->cb_scrub);
+	_debug("vnode inval %x==%x %x==%x",
+	       vnode->cb_ro_snapshot, cb_ro_snapshot,
+	       vnode->cb_scrub, cb_scrub);
+	if (vnode->cb_scrub != cb_scrub)
+		zap = true;
+	vnode->cb_ro_snapshot = cb_ro_snapshot;
+	vnode->cb_scrub = cb_scrub;
+
 	if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) {
 		_debug("file already deleted");
 		ret = -ESTALE;
@@ -364,15 +456,18 @@ int afs_validate(struct afs_vnode *vnode, struct key *key)
 
 	/* if the vnode's data version number changed then its contents are
 	 * different */
-	if (test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags))
+	zap |= test_and_clear_bit(AFS_VNODE_ZAP_DATA, &vnode->flags);
+	if (zap)
 		afs_zap_data(vnode);
 	up_write(&vnode->validate_lock);
-valid:
 	_leave(" = 0");
 	return 0;
 
 error_unlock:
+	if (locked_vol)
+		mutex_unlock(&volume->cb_check_lock);
 	up_write(&vnode->validate_lock);
+error:
 	_leave(" = %d", ret);
 	return ret;
 }