fsnotify: add marks to inodes so groups can interpret how to handle those inodes

This patch creates a way for fsnotify groups to attach marks to inodes.
These marks have little meaning to the generic fsnotify infrastructure
and thus their meaning should be interpreted by the group that attached
them to the inode's list.

dnotify and inotify  will make use of these markings to indicate which
inodes are of interest to their respective groups.  But this implementation
has the useful property that in the future other listeners could actually
use the marks for the exact opposite reason, aka to indicate which inodes
it had NO interest in.

Signed-off-by: Eric Paris <eparis@redhat.com>
Acked-by: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@lst.de>

5 files changed, 396 insertions, 2 deletions

diff --git a/fs/notify/Makefile b/fs/notify/Makefile
index db5467b5b58..0922cc826c4 100644
--- a/fs/notify/Makefile
+++ b/fs/notify/Makefile
@@ -1,4 +1,4 @@
-obj-$(CONFIG_FSNOTIFY)		+= fsnotify.o notification.o group.o
+obj-$(CONFIG_FSNOTIFY)		+= fsnotify.o notification.o group.o inode_mark.o
 
 obj-y			+= dnotify/
 obj-y			+= inotify/
diff --git a/fs/notify/fsnotify.c b/fs/notify/fsnotify.c
index 56bee0f10c3..d5654629c65 100644
--- a/fs/notify/fsnotify.c
+++ b/fs/notify/fsnotify.c
@@ -26,6 +26,15 @@
 #include "fsnotify.h"
 
 /*
+ * Clear all of the marks on an inode when it is being evicted from core
+ */
+void __fsnotify_inode_delete(struct inode *inode)
+{
+	fsnotify_clear_marks_by_inode(inode);
+}
+EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
+
+/*
  * This is the main call to fsnotify.  The VFS calls into hook specific functions
  * in linux/fsnotify.h.  Those functions then in turn call here.  Here will call
  * out to all of the registered fsnotify_group.  Those groups can then use the
@@ -43,6 +52,8 @@ void fsnotify(struct inode *to_tell, __u32 mask, void *data, int data_is)
 	if (!(mask & fsnotify_mask))
 		return;
 
+	if (!(mask & to_tell->i_fsnotify_mask))
+		return;
 	/*
 	 * SRCU!!  the groups list is very very much read only and the path is
 	 * very hot.  The VAST majority of events are not going to need to do
@@ -51,6 +62,8 @@ void fsnotify(struct inode *to_tell, __u32 mask, void *data, int data_is)
 	idx = srcu_read_lock(&fsnotify_grp_srcu);
 	list_for_each_entry_rcu(group, &fsnotify_groups, group_list) {
 		if (mask & group->mask) {
+			if (!group->ops->should_send_event(group, to_tell, mask))
+				continue;
 			if (!event) {
 				event = fsnotify_create_event(to_tell, mask, data, data_is);
 				/* shit, we OOM'd and now we can't tell, maybe
diff --git a/fs/notify/fsnotify.h b/fs/notify/fsnotify.h
index c6a8bd47657..8ebcbe893c9 100644
--- a/fs/notify/fsnotify.h
+++ b/fs/notify/fsnotify.h
@@ -12,4 +12,9 @@ extern struct srcu_struct fsnotify_grp_srcu;
 extern struct list_head fsnotify_groups;
 /* all bitwise OR of all event types (FS_*) for all fsnotify_groups */
 extern __u32 fsnotify_mask;
+
+/* final kfree of a group */
+extern void fsnotify_final_destroy_group(struct fsnotify_group *group);
+/* run the list of all marks associated with inode and flag them to be freed */
+extern void fsnotify_clear_marks_by_inode(struct inode *inode);
 #endif	/* __FS_NOTIFY_FSNOTIFY_H_ */
diff --git a/fs/notify/group.c b/fs/notify/group.c
index c6812953b96..a29d2fa6792 100644
--- a/fs/notify/group.c
+++ b/fs/notify/group.c
@@ -55,6 +55,29 @@ void fsnotify_recalc_global_mask(void)
 }
 
 /*
+ * Update the group->mask by running all of the marks associated with this
+ * group and finding the bitwise | of all of the mark->mask.  If we change
+ * the group->mask we need to update the global mask of events interesting
+ * to the system.
+ */
+void fsnotify_recalc_group_mask(struct fsnotify_group *group)
+{
+	__u32 mask = 0;
+	__u32 old_mask = group->mask;
+	struct fsnotify_mark_entry *entry;
+
+	spin_lock(&group->mark_lock);
+	list_for_each_entry(entry, &group->mark_entries, g_list)
+		mask |= entry->mask;
+	spin_unlock(&group->mark_lock);
+
+	group->mask = mask;
+
+	if (old_mask != mask)
+		fsnotify_recalc_global_mask();
+}
+
+/*
  * Take a reference to a group so things found under the fsnotify_grp_mutex
  * can't get freed under us
  */
@@ -66,7 +89,7 @@ static void fsnotify_get_group(struct fsnotify_group *group)
 /*
  * Final freeing of a group
  */
-static void fsnotify_destroy_group(struct fsnotify_group *group)
+void fsnotify_final_destroy_group(struct fsnotify_group *group)
 {
 	if (group->ops->free_group_priv)
 		group->ops->free_group_priv(group);
@@ -75,6 +98,24 @@ static void fsnotify_destroy_group(struct fsnotify_group *group)
 }
 
 /*
+ * Trying to get rid of a group.  We need to first get rid of any outstanding
+ * allocations and then free the group.  Remember that fsnotify_clear_marks_by_group
+ * could miss marks that are being freed by inode and those marks could still
+ * hold a reference to this group (via group->num_marks)  If we get into that
+ * situtation, the fsnotify_final_destroy_group will get called when that final
+ * mark is freed.
+ */
+static void fsnotify_destroy_group(struct fsnotify_group *group)
+{
+	/* clear all inode mark entries for this group */
+	fsnotify_clear_marks_by_group(group);
+
+	/* past the point of no return, matches the initial value of 1 */
+	if (atomic_dec_and_test(&group->num_marks))
+		fsnotify_final_destroy_group(group);
+}
+
+/*
  * Remove this group from the global list of groups that will get events
  * this can be done even if there are still references and things still using
  * this group.  This just stops the group from getting new events.
@@ -173,6 +214,10 @@ struct fsnotify_group *fsnotify_obtain_group(unsigned int group_num, __u32 mask,
 	group->group_num = group_num;
 	group->mask = mask;
 
+	spin_lock_init(&group->mark_lock);
+	atomic_set(&group->num_marks, 0);
+	INIT_LIST_HEAD(&group->mark_entries);
+
 	group->ops = ops;
 
 	mutex_lock(&fsnotify_grp_mutex);
@@ -188,6 +233,8 @@ struct fsnotify_group *fsnotify_obtain_group(unsigned int group_num, __u32 mask,
 	/* group not found, add a new one */
 	list_add_rcu(&group->group_list, &fsnotify_groups);
 	group->on_group_list = 1;
+	/* being on the fsnotify_groups list holds one num_marks */
+	atomic_inc(&group->num_marks);
 
 	mutex_unlock(&fsnotify_grp_mutex);
 
diff --git a/fs/notify/inode_mark.c b/fs/notify/inode_mark.c
new file mode 100644
index 00000000000..cdc15414697
--- /dev/null
+++ b/fs/notify/inode_mark.c
@@ -0,0 +1,329 @@
+/*
+ *  Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2, or (at your option)
+ *  any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; see the file COPYING.  If not, write to
+ *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+/*
+ * fsnotify inode mark locking/lifetime/and refcnting
+ *
+ * REFCNT:
+ * The mark->refcnt tells how many "things" in the kernel currently are
+ * referencing this object.  The object typically will live inside the kernel
+ * with a refcnt of 2, one for each list it is on (i_list, g_list).  Any task
+ * which can find this object holding the appropriete locks, can take a reference
+ * and the object itself is guarenteed to survive until the reference is dropped.
+ *
+ * LOCKING:
+ * There are 3 spinlocks involved with fsnotify inode marks and they MUST
+ * be taken in order as follows:
+ *
+ * entry->lock
+ * group->mark_lock
+ * inode->i_lock
+ *
+ * entry->lock protects 2 things, entry->group and entry->inode.  You must hold
+ * that lock to dereference either of these things (they could be NULL even with
+ * the lock)
+ *
+ * group->mark_lock protects the mark_entries list anchored inside a given group
+ * and each entry is hooked via the g_list.  It also sorta protects the
+ * free_g_list, which when used is anchored by a private list on the stack of the
+ * task which held the group->mark_lock.
+ *
+ * inode->i_lock protects the i_fsnotify_mark_entries list anchored inside a
+ * given inode and each entry is hooked via the i_list. (and sorta the
+ * free_i_list)
+ *
+ *
+ * LIFETIME:
+ * Inode marks survive between when they are added to an inode and when their
+ * refcnt==0.
+ *
+ * The inode mark can be cleared for a number of different reasons including:
+ * - The inode is unlinked for the last time.  (fsnotify_inode_remove)
+ * - The inode is being evicted from cache. (fsnotify_inode_delete)
+ * - The fs the inode is on is unmounted.  (fsnotify_inode_delete/fsnotify_unmount_inodes)
+ * - Something explicitly requests that it be removed.  (fsnotify_destroy_mark_by_entry)
+ * - The fsnotify_group associated with the mark is going away and all such marks
+ *   need to be cleaned up. (fsnotify_clear_marks_by_group)
+ *
+ * Worst case we are given an inode and need to clean up all the marks on that
+ * inode.  We take i_lock and walk the i_fsnotify_mark_entries safely.  For each
+ * mark on the list we take a reference (so the mark can't disappear under us).
+ * We remove that mark form the inode's list of marks and we add this mark to a
+ * private list anchored on the stack using i_free_list;  At this point we no
+ * longer fear anything finding the mark using the inode's list of marks.
+ *
+ * We can safely and locklessly run the private list on the stack of everything
+ * we just unattached from the original inode.  For each mark on the private list
+ * we grab the mark-> and can thus dereference mark->group and mark->inode.  If
+ * we see the group and inode are not NULL we take those locks.  Now holding all
+ * 3 locks we can completely remove the mark from other tasks finding it in the
+ * future.  Remember, 10 things might already be referencing this mark, but they
+ * better be holding a ref.  We drop our reference we took before we unhooked it
+ * from the inode.  When the ref hits 0 we can free the mark.
+ *
+ * Very similarly for freeing by group, except we use free_g_list.
+ *
+ * This has the very interesting property of being able to run concurrently with
+ * any (or all) other directions.
+ */
+
+#include <linux/fs.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+
+#include <asm/atomic.h>
+
+#include <linux/fsnotify_backend.h>
+#include "fsnotify.h"
+
+void fsnotify_get_mark(struct fsnotify_mark_entry *entry)
+{
+	atomic_inc(&entry->refcnt);
+}
+
+void fsnotify_put_mark(struct fsnotify_mark_entry *entry)
+{
+	if (atomic_dec_and_test(&entry->refcnt))
+		entry->free_mark(entry);
+}
+
+/*
+ * Recalculate the mask of events relevant to a given inode locked.
+ */
+static void fsnotify_recalc_inode_mask_locked(struct inode *inode)
+{
+	struct fsnotify_mark_entry *entry;
+	struct hlist_node *pos;
+	__u32 new_mask = 0;
+
+	assert_spin_locked(&inode->i_lock);
+
+	hlist_for_each_entry(entry, pos, &inode->i_fsnotify_mark_entries, i_list)
+		new_mask |= entry->mask;
+	inode->i_fsnotify_mask = new_mask;
+}
+
+/*
+ * Recalculate the inode->i_fsnotify_mask, or the mask of all FS_* event types
+ * any notifier is interested in hearing for this inode.
+ */
+void fsnotify_recalc_inode_mask(struct inode *inode)
+{
+	spin_lock(&inode->i_lock);
+	fsnotify_recalc_inode_mask_locked(inode);
+	spin_unlock(&inode->i_lock);
+}
+
+/*
+ * Any time a mark is getting freed we end up here.
+ * The caller had better be holding a reference to this mark so we don't actually
+ * do the final put under the entry->lock
+ */
+void fsnotify_destroy_mark_by_entry(struct fsnotify_mark_entry *entry)
+{
+	struct fsnotify_group *group;
+	struct inode *inode;
+
+	spin_lock(&entry->lock);
+
+	group = entry->group;
+	inode = entry->inode;
+
+	BUG_ON(group && !inode);
+	BUG_ON(!group && inode);
+
+	/* if !group something else already marked this to die */
+	if (!group) {
+		spin_unlock(&entry->lock);
+		return;
+	}
+
+	/* 1 from caller and 1 for being on i_list/g_list */
+	BUG_ON(atomic_read(&entry->refcnt) < 2);
+
+	spin_lock(&group->mark_lock);
+	spin_lock(&inode->i_lock);
+
+	hlist_del_init(&entry->i_list);
+	entry->inode = NULL;
+
+	list_del_init(&entry->g_list);
+	entry->group = NULL;
+
+	fsnotify_put_mark(entry); /* for i_list and g_list */
+
+	/*
+	 * this mark is now off the inode->i_fsnotify_mark_entries list and we
+	 * hold the inode->i_lock, so this is the perfect time to update the
+	 * inode->i_fsnotify_mask
+	 */
+	fsnotify_recalc_inode_mask_locked(inode);
+
+	spin_unlock(&inode->i_lock);
+	spin_unlock(&group->mark_lock);
+	spin_unlock(&entry->lock);
+
+	/*
+	 * Some groups like to know that marks are being freed.  This is a
+	 * callback to the group function to let it know that this entry
+	 * is being freed.
+	 */
+	group->ops->freeing_mark(entry, group);
+
+	/*
+	 * it's possible that this group tried to destroy itself, but this
+	 * this mark was simultaneously being freed by inode.  If that's the
+	 * case, we finish freeing the group here.
+	 */
+	if (unlikely(atomic_dec_and_test(&group->num_marks)))
+		fsnotify_final_destroy_group(group);
+}
+
+/*
+ * Given a group, destroy all of the marks associated with that group.
+ */
+void fsnotify_clear_marks_by_group(struct fsnotify_group *group)
+{
+	struct fsnotify_mark_entry *lentry, *entry;
+	LIST_HEAD(free_list);
+
+	spin_lock(&group->mark_lock);
+	list_for_each_entry_safe(entry, lentry, &group->mark_entries, g_list) {
+		list_add(&entry->free_g_list, &free_list);
+		list_del_init(&entry->g_list);
+		fsnotify_get_mark(entry);
+	}
+	spin_unlock(&group->mark_lock);
+
+	list_for_each_entry_safe(entry, lentry, &free_list, free_g_list) {
+		fsnotify_destroy_mark_by_entry(entry);
+		fsnotify_put_mark(entry);
+	}
+}
+
+/*
+ * Given an inode, destroy all of the marks associated with that inode.
+ */
+void fsnotify_clear_marks_by_inode(struct inode *inode)
+{
+	struct fsnotify_mark_entry *entry, *lentry;
+	struct hlist_node *pos, *n;
+	LIST_HEAD(free_list);
+
+	spin_lock(&inode->i_lock);
+	hlist_for_each_entry_safe(entry, pos, n, &inode->i_fsnotify_mark_entries, i_list) {
+		list_add(&entry->free_i_list, &free_list);
+		hlist_del_init(&entry->i_list);
+		fsnotify_get_mark(entry);
+	}
+	spin_unlock(&inode->i_lock);
+
+	list_for_each_entry_safe(entry, lentry, &free_list, free_i_list) {
+		fsnotify_destroy_mark_by_entry(entry);
+		fsnotify_put_mark(entry);
+	}
+}
+
+/*
+ * given a group and inode, find the mark associated with that combination.
+ * if found take a reference to that mark and return it, else return NULL
+ */
+struct fsnotify_mark_entry *fsnotify_find_mark_entry(struct fsnotify_group *group,
+						     struct inode *inode)
+{
+	struct fsnotify_mark_entry *entry;
+	struct hlist_node *pos;
+
+	assert_spin_locked(&inode->i_lock);
+
+	hlist_for_each_entry(entry, pos, &inode->i_fsnotify_mark_entries, i_list) {
+		if (entry->group == group) {
+			fsnotify_get_mark(entry);
+			return entry;
+		}
+	}
+	return NULL;
+}
+
+/*
+ * Nothing fancy, just initialize lists and locks and counters.
+ */
+void fsnotify_init_mark(struct fsnotify_mark_entry *entry,
+			void (*free_mark)(struct fsnotify_mark_entry *entry))
+
+{
+	spin_lock_init(&entry->lock);
+	atomic_set(&entry->refcnt, 1);
+	INIT_HLIST_NODE(&entry->i_list);
+	entry->group = NULL;
+	entry->mask = 0;
+	entry->inode = NULL;
+	entry->free_mark = free_mark;
+}
+
+/*
+ * Attach an initialized mark entry to a given group and inode.
+ * These marks may be used for the fsnotify backend to determine which
+ * event types should be delivered to which group and for which inodes.
+ */
+int fsnotify_add_mark(struct fsnotify_mark_entry *entry,
+		      struct fsnotify_group *group, struct inode *inode)
+{
+	struct fsnotify_mark_entry *lentry;
+	int ret = 0;
+
+	/*
+	 * LOCKING ORDER!!!!
+	 * entry->lock
+	 * group->mark_lock
+	 * inode->i_lock
+	 */
+	spin_lock(&entry->lock);
+	spin_lock(&group->mark_lock);
+	spin_lock(&inode->i_lock);
+
+	entry->group = group;
+	entry->inode = inode;
+
+	lentry = fsnotify_find_mark_entry(group, inode);
+	if (!lentry) {
+		hlist_add_head(&entry->i_list, &inode->i_fsnotify_mark_entries);
+		list_add(&entry->g_list, &group->mark_entries);
+
+		fsnotify_get_mark(entry); /* for i_list and g_list */
+
+		atomic_inc(&group->num_marks);
+
+		fsnotify_recalc_inode_mask_locked(inode);
+	}
+
+	spin_unlock(&inode->i_lock);
+	spin_unlock(&group->mark_lock);
+	spin_unlock(&entry->lock);
+
+	if (lentry) {
+		ret = -EEXIST;
+		fsnotify_put_mark(lentry);
+	}
+
+	return ret;
+}