Merge branch 'for-linus' of git://git.infradead.org/users/eparis/notify

* 'for-linus' of git://git.infradead.org/users/eparis/notify: (132 commits)
  fanotify: use both marks when possible
  fsnotify: pass both the vfsmount mark and inode mark
  fsnotify: walk the inode and vfsmount lists simultaneously
  fsnotify: rework ignored mark flushing
  fsnotify: remove global fsnotify groups lists
  fsnotify: remove group->mask
  fsnotify: remove the global masks
  fsnotify: cleanup should_send_event
  fanotify: use the mark in handler functions
  audit: use the mark in handler functions
  dnotify: use the mark in handler functions
  inotify: use the mark in handler functions
  fsnotify: send fsnotify_mark to groups in event handling functions
  fsnotify: Exchange list heads instead of moving elements
  fsnotify: srcu to protect read side of inode and vfsmount locks
  fsnotify: use an explicit flag to indicate fsnotify_destroy_mark has been called
  fsnotify: use _rcu functions for mark list traversal
  fsnotify: place marks on object in order of group memory address
  vfs/fsnotify: fsnotify_close can delay the final work in fput
  fsnotify: store struct file not struct path
  ...

Fix up trivial delete/modify conflict in fs/notify/inotify/inotify.c.

Documentation/feature-removal-schedule.txt

@@ -360,14 +360,6 @@ When:	2.6.33
 Why:	Should be implemented in userspace, policy daemon.
 Who:	Johannes Berg <johannes@sipsolutions.net>
 
----------------------------
-
-What:	CONFIG_INOTIFY
-When:	2.6.33
-Why:	last user (audit) will be converted to the newer more generic
-	and more easily maintained fsnotify subsystem
-Who:	Eric Paris <eparis@redhat.com>
-
 ----------------------------
 
 What:	sound-slot/service-* module aliases and related clutters in

arch/x86/ia32/ia32entry.S

@@ -842,4 +842,6 @@ ia32_sys_call_table:
 	.quad compat_sys_rt_tgsigqueueinfo	/* 335 */
 	.quad sys_perf_event_open
 	.quad compat_sys_recvmmsg
+	.quad sys_fanotify_init
+	.quad sys32_fanotify_mark
 ia32_syscall_end:

arch/x86/ia32/sys_ia32.c

@@ -546,3 +546,12 @@ asmlinkage long sys32_fallocate(int fd, int mode, unsigned offset_lo,
 	return sys_fallocate(fd, mode, ((u64)offset_hi << 32) | offset_lo,
 			     ((u64)len_hi << 32) | len_lo);
 }
+
+asmlinkage long sys32_fanotify_mark(int fanotify_fd, unsigned int flags,
+				    u32 mask_lo, u32 mask_hi,
+				    int fd, const char  __user *pathname)
+{
+	return sys_fanotify_mark(fanotify_fd, flags,
+				 ((u64)mask_hi << 32) | mask_lo,
+				 fd, pathname);
+}

arch/x86/include/asm/sys_ia32.h

@@ -80,4 +80,7 @@ asmlinkage long sys32_rt_sigreturn(struct pt_regs *);
 
 /* ia32/ipc32.c */
 asmlinkage long sys32_ipc(u32, int, int, int, compat_uptr_t, u32);
+
+asmlinkage long sys32_fanotify_mark(int, unsigned int, u32, u32, int,
+				    const char __user *);
 #endif /* _ASM_X86_SYS_IA32_H */

arch/x86/include/asm/unistd_32.h

@@ -343,10 +343,12 @@
 #define __NR_rt_tgsigqueueinfo	335
 #define __NR_perf_event_open	336
 #define __NR_recvmmsg		337
+#define __NR_fanotify_init	338
+#define __NR_fanotify_mark	339
 
 #ifdef __KERNEL__
 
-#define NR_syscalls 338
+#define NR_syscalls 340
 
 #define __ARCH_WANT_IPC_PARSE_VERSION
 #define __ARCH_WANT_OLD_READDIR

arch/x86/include/asm/unistd_64.h

@@ -663,6 +663,10 @@ __SYSCALL(__NR_rt_tgsigqueueinfo, sys_rt_tgsigqueueinfo)
 __SYSCALL(__NR_perf_event_open, sys_perf_event_open)
 #define __NR_recvmmsg				299
 __SYSCALL(__NR_recvmmsg, sys_recvmmsg)
+#define __NR_fanotify_init			300
+__SYSCALL(__NR_fanotify_init, sys_fanotify_init)
+#define __NR_fanotify_mark			301
+__SYSCALL(__NR_fanotify_mark, sys_fanotify_mark)
 
 #ifndef __NO_STUBS
 #define __ARCH_WANT_OLD_READDIR

arch/x86/kernel/syscall_table_32.S

@@ -337,3 +337,5 @@ ENTRY(sys_call_table)
 	.long sys_rt_tgsigqueueinfo	/* 335 */
 	.long sys_perf_event_open
 	.long sys_recvmmsg
+	.long sys_fanotify_init
+	.long sys_fanotify_mark

fs/compat.c

@@ -1193,11 +1193,10 @@ static ssize_t compat_do_readv_writev(int type, struct file *file,
 	if (iov != iovstack)
 		kfree(iov);
 	if ((ret + (type == READ)) > 0) {
-		struct dentry *dentry = file->f_path.dentry;
 		if (type == READ)
-			fsnotify_access(dentry);
+			fsnotify_access(file);
 		else
-			fsnotify_modify(dentry);
+			fsnotify_modify(file);
 	}
 	return ret;
 }

fs/exec.c

@@ -128,7 +128,7 @@ SYSCALL_DEFINE1(uselib, const char __user *, library)
 	if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 		goto exit;
 
-	fsnotify_open(file->f_path.dentry);
+	fsnotify_open(file);
 
 	error = -ENOEXEC;
 	if(file->f_op) {
@@ -683,7 +683,7 @@ struct file *open_exec(const char *name)
 	if (file->f_path.mnt->mnt_flags & MNT_NOEXEC)
 		goto exit;
 
-	fsnotify_open(file->f_path.dentry);
+	fsnotify_open(file);
 
 	err = deny_write_access(file);
 	if (err)

fs/file_table.c

@@ -230,6 +230,15 @@ static void __fput(struct file *file)
 	might_sleep();
 
 	fsnotify_close(file);
+
+	/*
+	 * fsnotify_create_event may have taken one or more references on this
+	 * file.  If it did so it left one reference for us to drop to make sure
+	 * its calls to fput could not prematurely destroy the file.
+	 */
+	if (atomic_long_read(&file->f_count))
+		return fput(file);
+
 	/*
 	 * The function eventpoll_release() should be the first called
 	 * in the file cleanup chain.

fs/inode.c

@@ -20,7 +20,6 @@
 #include <linux/pagemap.h>
 #include <linux/cdev.h>
 #include <linux/bootmem.h>
-#include <linux/inotify.h>
 #include <linux/fsnotify.h>
 #include <linux/mount.h>
 #include <linux/async.h>
@@ -264,12 +263,8 @@ void inode_init_once(struct inode *inode)
 	INIT_RAW_PRIO_TREE_ROOT(&inode->i_data.i_mmap);
 	INIT_LIST_HEAD(&inode->i_data.i_mmap_nonlinear);
 	i_size_ordered_init(inode);
-#ifdef CONFIG_INOTIFY
-	INIT_LIST_HEAD(&inode->inotify_watches);
-	mutex_init(&inode->inotify_mutex);
-#endif
 #ifdef CONFIG_FSNOTIFY
-	INIT_HLIST_HEAD(&inode->i_fsnotify_mark_entries);
+	INIT_HLIST_HEAD(&inode->i_fsnotify_marks);
 #endif
 }
 EXPORT_SYMBOL(inode_init_once);
@@ -413,7 +408,6 @@ int invalidate_inodes(struct super_block *sb)
 
 	down_write(&iprune_sem);
 	spin_lock(&inode_lock);
-	inotify_unmount_inodes(&sb->s_inodes);
 	fsnotify_unmount_inodes(&sb->s_inodes);
 	busy = invalidate_list(&sb->s_inodes, &throw_away);
 	spin_unlock(&inode_lock);

fs/namei.c

@@ -2633,7 +2633,7 @@ int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
 {
 	int error;
 	int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
-	const char *old_name;
+	const unsigned char *old_name;
 
 	if (old_dentry->d_inode == new_dentry->d_inode)
  		return 0;

fs/namespace.c

@@ -29,6 +29,7 @@
 #include <linux/log2.h>
 #include <linux/idr.h>
 #include <linux/fs_struct.h>
+#include <linux/fsnotify.h>
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
 #include "pnode.h"
@@ -150,6 +151,9 @@ struct vfsmount *alloc_vfsmnt(const char *name)
 		INIT_LIST_HEAD(&mnt->mnt_share);
 		INIT_LIST_HEAD(&mnt->mnt_slave_list);
 		INIT_LIST_HEAD(&mnt->mnt_slave);
+#ifdef CONFIG_FSNOTIFY
+		INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
+#endif
 #ifdef CONFIG_SMP
 		mnt->mnt_writers = alloc_percpu(int);
 		if (!mnt->mnt_writers)
@@ -610,6 +614,7 @@ static inline void __mntput(struct vfsmount *mnt)
 	 * provides barriers, so count_mnt_writers() below is safe.  AV
 	 */
 	WARN_ON(count_mnt_writers(mnt));
+	fsnotify_vfsmount_delete(mnt);
 	dput(mnt->mnt_root);
 	free_vfsmnt(mnt);
 	deactivate_super(sb);

fs/nfsd/vfs.c

@@ -934,7 +934,7 @@ nfsd_vfs_read(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
 		nfsdstats.io_read += host_err;
 		*count = host_err;
 		err = 0;
-		fsnotify_access(file->f_path.dentry);
+		fsnotify_access(file);
 	} else 
 		err = nfserrno(host_err);
 out:
@@ -1045,7 +1045,7 @@ nfsd_vfs_write(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file *file,
 		goto out_nfserr;
 	*cnt = host_err;
 	nfsdstats.io_write += host_err;
-	fsnotify_modify(file->f_path.dentry);
+	fsnotify_modify(file);
 
 	/* clear setuid/setgid flag after write */
 	if (inode->i_mode & (S_ISUID | S_ISGID))

fs/notify/Kconfig

@@ -3,3 +3,4 @@ config FSNOTIFY
 
 source "fs/notify/dnotify/Kconfig"
 source "fs/notify/inotify/Kconfig"
+source "fs/notify/fanotify/Kconfig"

fs/notify/Makefile

-obj-$(CONFIG_FSNOTIFY)		+= fsnotify.o notification.o group.o inode_mark.o
+obj-$(CONFIG_FSNOTIFY)		+= fsnotify.o notification.o group.o inode_mark.o \
+				   mark.o vfsmount_mark.o
 
 obj-y			+= dnotify/
 obj-y			+= inotify/
+obj-y			+= fanotify/

fs/notify/dnotify/dnotify.c

@@ -29,17 +29,17 @@
 int dir_notify_enable __read_mostly = 1;
 
 static struct kmem_cache *dnotify_struct_cache __read_mostly;
-static struct kmem_cache *dnotify_mark_entry_cache __read_mostly;
+static struct kmem_cache *dnotify_mark_cache __read_mostly;
 static struct fsnotify_group *dnotify_group __read_mostly;
 static DEFINE_MUTEX(dnotify_mark_mutex);
 
 /*
- * dnotify will attach one of these to each inode (i_fsnotify_mark_entries) which
+ * dnotify will attach one of these to each inode (i_fsnotify_marks) which
  * is being watched by dnotify.  If multiple userspace applications are watching
  * the same directory with dnotify their information is chained in dn
  */
-struct dnotify_mark_entry {
-	struct fsnotify_mark_entry fsn_entry;
+struct dnotify_mark {
+	struct fsnotify_mark fsn_mark;
 	struct dnotify_struct *dn;
 };
 
@@ -51,27 +51,27 @@ struct dnotify_mark_entry {
  * it calls the fsnotify function so it can update the set of all events relevant
  * to this inode.
  */
-static void dnotify_recalc_inode_mask(struct fsnotify_mark_entry *entry)
+static void dnotify_recalc_inode_mask(struct fsnotify_mark *fsn_mark)
 {
 	__u32 new_mask, old_mask;
 	struct dnotify_struct *dn;
-	struct dnotify_mark_entry *dnentry  = container_of(entry,
-							   struct dnotify_mark_entry,
-							   fsn_entry);
+	struct dnotify_mark *dn_mark  = container_of(fsn_mark,
+						     struct dnotify_mark,
+						     fsn_mark);
 
-	assert_spin_locked(&entry->lock);
+	assert_spin_locked(&fsn_mark->lock);
 
-	old_mask = entry->mask;
+	old_mask = fsn_mark->mask;
 	new_mask = 0;
-	for (dn = dnentry->dn; dn != NULL; dn = dn->dn_next)
+	for (dn = dn_mark->dn; dn != NULL; dn = dn->dn_next)
 		new_mask |= (dn->dn_mask & ~FS_DN_MULTISHOT);
-	entry->mask = new_mask;
+	fsnotify_set_mark_mask_locked(fsn_mark, new_mask);
 
 	if (old_mask == new_mask)
 		return;
 
-	if (entry->inode)
-		fsnotify_recalc_inode_mask(entry->inode);
+	if (fsn_mark->i.inode)
+		fsnotify_recalc_inode_mask(fsn_mark->i.inode);
 }
 
 /*
@@ -83,29 +83,25 @@ static void dnotify_recalc_inode_mask(struct fsnotify_mark_entry *entry)
  * events.
  */
 static int dnotify_handle_event(struct fsnotify_group *group,
+				struct fsnotify_mark *inode_mark,
+				struct fsnotify_mark *vfsmount_mark,
 				struct fsnotify_event *event)
 {
-	struct fsnotify_mark_entry *entry = NULL;
-	struct dnotify_mark_entry *dnentry;
+	struct dnotify_mark *dn_mark;
 	struct inode *to_tell;
 	struct dnotify_struct *dn;
 	struct dnotify_struct **prev;
 	struct fown_struct *fown;
 	__u32 test_mask = event->mask & ~FS_EVENT_ON_CHILD;
 
-	to_tell = event->to_tell;
+	BUG_ON(vfsmount_mark);
 
-	spin_lock(&to_tell->i_lock);
-	entry = fsnotify_find_mark_entry(group, to_tell);
-	spin_unlock(&to_tell->i_lock);
+	to_tell = event->to_tell;
 
-	/* unlikely since we alreay passed dnotify_should_send_event() */
-	if (unlikely(!entry))
-		return 0;
-	dnentry = container_of(entry, struct dnotify_mark_entry, fsn_entry);
+	dn_mark = container_of(inode_mark, struct dnotify_mark, fsn_mark);
 
-	spin_lock(&entry->lock);
-	prev = &dnentry->dn;
+	spin_lock(&inode_mark->lock);
+	prev = &dn_mark->dn;
 	while ((dn = *prev) != NULL) {
 		if ((dn->dn_mask & test_mask) == 0) {
 			prev = &dn->dn_next;
@@ -118,12 +114,11 @@ static int dnotify_handle_event(struct fsnotify_group *group,
 		else {
 			*prev = dn->dn_next;
 			kmem_cache_free(dnotify_struct_cache, dn);
-			dnotify_recalc_inode_mask(entry);
+			dnotify_recalc_inode_mask(inode_mark);
 		}
 	}
 
-	spin_unlock(&entry->lock);
-	fsnotify_put_mark(entry);
+	spin_unlock(&inode_mark->lock);
 
 	return 0;
 }
@@ -133,44 +128,27 @@ static int dnotify_handle_event(struct fsnotify_group *group,
  * userspace notification for that pair.
  */
 static bool dnotify_should_send_event(struct fsnotify_group *group,
-				      struct inode *inode, __u32 mask)
+				      struct inode *inode,
+				      struct fsnotify_mark *inode_mark,
+				      struct fsnotify_mark *vfsmount_mark,
+				      __u32 mask, void *data, int data_type)
 {
-	struct fsnotify_mark_entry *entry;
-	bool send;
-
-	/* !dir_notify_enable should never get here, don't waste time checking
-	if (!dir_notify_enable)
-		return 0; */
-
 	/* not a dir, dnotify doesn't care */
 	if (!S_ISDIR(inode->i_mode))
 		return false;
 
-	spin_lock(&inode->i_lock);
-	entry = fsnotify_find_mark_entry(group, inode);
-	spin_unlock(&inode->i_lock);
-
-	/* no mark means no dnotify watch */
-	if (!entry)
-		return false;
-
-	mask = (mask & ~FS_EVENT_ON_CHILD);
-	send = (mask & entry->mask);
-
-	fsnotify_put_mark(entry); /* matches fsnotify_find_mark_entry */
-
-	return send;
+	return true;
 }
 
-static void dnotify_free_mark(struct fsnotify_mark_entry *entry)
+static void dnotify_free_mark(struct fsnotify_mark *fsn_mark)
 {
-	struct dnotify_mark_entry *dnentry = container_of(entry,
-							  struct dnotify_mark_entry,
-							  fsn_entry);
+	struct dnotify_mark *dn_mark = container_of(fsn_mark,
+						    struct dnotify_mark,
+						    fsn_mark);
 
-	BUG_ON(dnentry->dn);
+	BUG_ON(dn_mark->dn);
 
-	kmem_cache_free(dnotify_mark_entry_cache, dnentry);
+	kmem_cache_free(dnotify_mark_cache, dn_mark);
 }
 
 static struct fsnotify_ops dnotify_fsnotify_ops = {
@@ -183,15 +161,15 @@ static struct fsnotify_ops dnotify_fsnotify_ops = {
 
 /*
  * Called every time a file is closed.  Looks first for a dnotify mark on the
- * inode.  If one is found run all of the ->dn entries attached to that
+ * inode.  If one is found run all of the ->dn structures attached to that
  * mark for one relevant to this process closing the file and remove that
  * dnotify_struct.  If that was the last dnotify_struct also remove the
- * fsnotify_mark_entry.
+ * fsnotify_mark.
  */
 void dnotify_flush(struct file *filp, fl_owner_t id)
 {
-	struct fsnotify_mark_entry *entry;
-	struct dnotify_mark_entry *dnentry;
+	struct fsnotify_mark *fsn_mark;
+	struct dnotify_mark *dn_mark;
 	struct dnotify_struct *dn;
 	struct dnotify_struct **prev;
 	struct inode *inode;
@@ -200,38 +178,34 @@ void dnotify_flush(struct file *filp, fl_owner_t id)
 	if (!S_ISDIR(inode->i_mode))
 		return;
 
-	spin_lock(&inode->i_lock);
-	entry = fsnotify_find_mark_entry(dnotify_group, inode);
-	spin_unlock(&inode->i_lock);
-	if (!entry)
+	fsn_mark = fsnotify_find_inode_mark(dnotify_group, inode);
+	if (!fsn_mark)
 		return;
-	dnentry = container_of(entry, struct dnotify_mark_entry, fsn_entry);
+	dn_mark = container_of(fsn_mark, struct dnotify_mark, fsn_mark);
 
 	mutex_lock(&dnotify_mark_mutex);
 
-	spin_lock(&entry->lock);
-	prev = &dnentry->dn;
+	spin_lock(&fsn_mark->lock);
+	prev = &dn_mark->dn;
 	while ((dn = *prev) != NULL) {
 		if ((dn->dn_owner == id) && (dn->dn_filp == filp)) {
 			*prev = dn->dn_next;
 			kmem_cache_free(dnotify_struct_cache, dn);
-			dnotify_recalc_inode_mask(entry);
+			dnotify_recalc_inode_mask(fsn_mark);
 			break;
 		}
 		prev = &dn->dn_next;
 	}
 
-	spin_unlock(&entry->lock);
+	spin_unlock(&fsn_mark->lock);
 
 	/* nothing else could have found us thanks to the dnotify_mark_mutex */
-	if (dnentry->dn == NULL)
-		fsnotify_destroy_mark_by_entry(entry);
-
-	fsnotify_recalc_group_mask(dnotify_group);
+	if (dn_mark->dn == NULL)
+		fsnotify_destroy_mark(fsn_mark);
 
 	mutex_unlock(&dnotify_mark_mutex);
 
-	fsnotify_put_mark(entry);
+	fsnotify_put_mark(fsn_mark);
 }
 
 /* this conversion is done only at watch creation */
@@ -259,16 +233,16 @@ static __u32 convert_arg(unsigned long arg)
 
 /*
  * If multiple processes watch the same inode with dnotify there is only one
- * dnotify mark in inode->i_fsnotify_mark_entries but we chain a dnotify_struct
+ * dnotify mark in inode->i_fsnotify_marks but we chain a dnotify_struct
  * onto that mark.  This function either attaches the new dnotify_struct onto
  * that list, or it |= the mask onto an existing dnofiy_struct.
  */
-static int attach_dn(struct dnotify_struct *dn, struct dnotify_mark_entry *dnentry,
+static int attach_dn(struct dnotify_struct *dn, struct dnotify_mark *dn_mark,
 		     fl_owner_t id, int fd, struct file *filp, __u32 mask)
 {
 	struct dnotify_struct *odn;
 
-	odn = dnentry->dn;
+	odn = dn_mark->dn;
 	while (odn != NULL) {
 		/* adding more events to existing dnofiy_struct? */
 		if ((odn->dn_owner == id) && (odn->dn_filp == filp)) {
@@ -283,8 +257,8 @@ static int attach_dn(struct dnotify_struct *dn, struct dnotify_mark_entry *dnent
 	dn->dn_fd = fd;
 	dn->dn_filp = filp;
 	dn->dn_owner = id;
-	dn->dn_next = dnentry->dn;
-	dnentry->dn = dn;
+	dn->dn_next = dn_mark->dn;
+	dn_mark->dn = dn;
 
 	return 0;
 }
@@ -296,8 +270,8 @@ static int attach_dn(struct dnotify_struct *dn, struct dnotify_mark_entry *dnent
  */
 int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
 {
-	struct dnotify_mark_entry *new_dnentry, *dnentry;
-	struct fsnotify_mark_entry *new_entry, *entry;
+	struct dnotify_mark *new_dn_mark, *dn_mark;
+	struct fsnotify_mark *new_fsn_mark, *fsn_mark;
 	struct dnotify_struct *dn;
 	struct inode *inode;
 	fl_owner_t id = current->files;
@@ -306,7 +280,7 @@ int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
 	__u32 mask;
 
 	/* we use these to tell if we need to kfree */
-	new_entry = NULL;
+	new_fsn_mark = NULL;
 	dn = NULL;
 
 	if (!dir_notify_enable) {
@@ -336,8 +310,8 @@ int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
 	}
 
 	/* new fsnotify mark, we expect most fcntl calls to add a new mark */
-	new_dnentry = kmem_cache_alloc(dnotify_mark_entry_cache, GFP_KERNEL);
-	if (!new_dnentry) {
+	new_dn_mark = kmem_cache_alloc(dnotify_mark_cache, GFP_KERNEL);
+	if (!new_dn_mark) {
 		error = -ENOMEM;
 		goto out_err;
 	}
@@ -345,29 +319,27 @@ int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
 	/* convert the userspace DN_* "arg" to the internal FS_* defines in fsnotify */
 	mask = convert_arg(arg);
 
-	/* set up the new_entry and new_dnentry */
-	new_entry = &new_dnentry->fsn_entry;
-	fsnotify_init_mark(new_entry, dnotify_free_mark);
-	new_entry->mask = mask;
-	new_dnentry->dn = NULL;
+	/* set up the new_fsn_mark and new_dn_mark */
+	new_fsn_mark = &new_dn_mark->fsn_mark;
+	fsnotify_init_mark(new_fsn_mark, dnotify_free_mark);
+	new_fsn_mark->mask = mask;
+	new_dn_mark->dn = NULL;
 
 	/* this is needed to prevent the fcntl/close race described below */
 	mutex_lock(&dnotify_mark_mutex);
 
-	/* add the new_entry or find an old one. */
-	spin_lock(&inode->i_lock);
-	entry = fsnotify_find_mark_entry(dnotify_group, inode);
-	spin_unlock(&inode->i_lock);
-	if (entry) {
-		dnentry = container_of(entry, struct dnotify_mark_entry, fsn_entry);
-		spin_lock(&entry->lock);
+	/* add the new_fsn_mark or find an old one. */
+	fsn_mark = fsnotify_find_inode_mark(dnotify_group, inode);
+	if (fsn_mark) {
+		dn_mark = container_of(fsn_mark, struct dnotify_mark, fsn_mark);
+		spin_lock(&fsn_mark->lock);
 	} else {
-		fsnotify_add_mark(new_entry, dnotify_group, inode);
-		spin_lock(&new_entry->lock);
-		entry = new_entry;
-		dnentry = new_dnentry;
-		/* we used new_entry, so don't free it */
-		new_entry = NULL;
+		fsnotify_add_mark(new_fsn_mark, dnotify_group, inode, NULL, 0);
+		spin_lock(&new_fsn_mark->lock);
+		fsn_mark = new_fsn_mark;
+		dn_mark = new_dn_mark;
+		/* we used new_fsn_mark, so don't free it */
+		new_fsn_mark = NULL;
 	}
 
 	rcu_read_lock();
@@ -376,17 +348,17 @@ int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
 
 	/* if (f != filp) means that we lost a race and another task/thread
 	 * actually closed the fd we are still playing with before we grabbed
-	 * the dnotify_mark_mutex and entry->lock.  Since closing the fd is the
-	 * only time we clean up the mark entries we need to get our mark off
+	 * the dnotify_mark_mutex and fsn_mark->lock.  Since closing the fd is the
+	 * only time we clean up the marks we need to get our mark off
 	 * the list. */
 	if (f != filp) {
 		/* if we added ourselves, shoot ourselves, it's possible that
-		 * the flush actually did shoot this entry.  That's fine too
+		 * the flush actually did shoot this fsn_mark.  That's fine too
 		 * since multiple calls to destroy_mark is perfectly safe, if
-		 * we found a dnentry already attached to the inode, just sod
+		 * we found a dn_mark already attached to the inode, just sod
 		 * off silently as the flush at close time dealt with it.
 		 */
-		if (dnentry == new_dnentry)
+		if (dn_mark == new_dn_mark)
 			destroy = 1;
 		goto out;
 	}
@@ -394,13 +366,13 @@ int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
 	error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
 	if (error) {
 		/* if we added, we must shoot */
-		if (dnentry == new_dnentry)
+		if (dn_mark == new_dn_mark)
 			destroy = 1;
 		goto out;
 	}
 
-	error = attach_dn(dn, dnentry, id, fd, filp, mask);
-	/* !error means that we attached the dn to the dnentry, so don't free it */
+	error = attach_dn(dn, dn_mark, id, fd, filp, mask);
+	/* !error means that we attached the dn to the dn_mark, so don't free it */
 	if (!error)
 		dn = NULL;
 	/* -EEXIST means that we didn't add this new dn and used an old one.
@@ -408,20 +380,18 @@ int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
 	else if (error == -EEXIST)
 		error = 0;
 
-	dnotify_recalc_inode_mask(entry);
+	dnotify_recalc_inode_mask(fsn_mark);
 out:
-	spin_unlock(&entry->lock);
+	spin_unlock(&fsn_mark->lock);
 
 	if (destroy)
-		fsnotify_destroy_mark_by_entry(entry);
-
-	fsnotify_recalc_group_mask(dnotify_group);
+		fsnotify_destroy_mark(fsn_mark);
 
 	mutex_unlock(&dnotify_mark_mutex);
-	fsnotify_put_mark(entry);
+	fsnotify_put_mark(fsn_mark);
 out_err:
-	if (new_entry)
-		fsnotify_put_mark(new_entry);
+	if (new_fsn_mark)
+		fsnotify_put_mark(new_fsn_mark);
 	if (dn)
 		kmem_cache_free(dnotify_struct_cache, dn);
 	return error;
@@ -430,10 +400,9 @@ int fcntl_dirnotify(int fd, struct file *filp, unsigned long arg)
 static int __init dnotify_init(void)
 {
 	dnotify_struct_cache = KMEM_CACHE(dnotify_struct, SLAB_PANIC);
-	dnotify_mark_entry_cache = KMEM_CACHE(dnotify_mark_entry, SLAB_PANIC);
+	dnotify_mark_cache = KMEM_CACHE(dnotify_mark, SLAB_PANIC);
 
-	dnotify_group = fsnotify_obtain_group(DNOTIFY_GROUP_NUM,
-					      0, &dnotify_fsnotify_ops);
+	dnotify_group = fsnotify_alloc_group(&dnotify_fsnotify_ops);
 	if (IS_ERR(dnotify_group))
 		panic("unable to allocate fsnotify group for dnotify\n");
 	return 0;

fs/notify/fanotify/Kconfig

+config FANOTIFY
+	bool "Filesystem wide access notification"
+	select FSNOTIFY
+	select ANON_INODES
+	default n
+	---help---
+	   Say Y here to enable fanotify suport.  fanotify is a file access
+	   notification system which differs from inotify in that it sends
+	   and open file descriptor to the userspace listener along with
+	   the event.
+
+	   If unsure, say Y.
+
+config FANOTIFY_ACCESS_PERMISSIONS
+	bool "fanotify permissions checking"
+	depends on FANOTIFY
+	depends on SECURITY
+	default n
+	---help---
+	   Say Y here is you want fanotify listeners to be able to make permissions
+	   decisions concerning filesystem events.  This is used by some fanotify
+	   listeners which need to scan files before allowing the system access to
+	   use those files.  This is used by some anti-malware vendors and by some
+	   hierarchical storage managent systems.
+
+	   If unsure, say N.

fs/notify/fanotify/Makefile

+obj-$(CONFIG_FANOTIFY)		+= fanotify.o fanotify_user.o

fs/notify/fanotify/fanotify.c

+#include <linux/fanotify.h>
+#include <linux/fdtable.h>
+#include <linux/fsnotify_backend.h>
+#include <linux/init.h>
+#include <linux/jiffies.h>
+#include <linux/kernel.h> /* UINT_MAX */
+#include <linux/mount.h>
+#include <linux/sched.h>
+#include <linux/types.h>
+#include <linux/wait.h>
+
+static bool should_merge(struct fsnotify_event *old, struct fsnotify_event *new)
+{
+	pr_debug("%s: old=%p new=%p\n", __func__, old, new);
+
+	if (old->to_tell == new->to_tell &&
+	    old->data_type == new->data_type &&
+	    old->tgid == new->tgid) {
+		switch (old->data_type) {
+		case (FSNOTIFY_EVENT_FILE):
+			if ((old->file->f_path.mnt == new->file->f_path.mnt) &&
+			    (old->file->f_path.dentry == new->file->f_path.dentry))
+				return true;
+		case (FSNOTIFY_EVENT_NONE):
+			return true;
+		default:
+			BUG();
+		};
+	}
+	return false;
+}
+
+/* and the list better be locked by something too! */
+static struct fsnotify_event *fanotify_merge(struct list_head *list,
+					     struct fsnotify_event *event)
+{
+	struct fsnotify_event_holder *test_holder;
+	struct fsnotify_event *test_event = NULL;
+	struct fsnotify_event *new_event;
+
+	pr_debug("%s: list=%p event=%p\n", __func__, list, event);
+
+
+	list_for_each_entry_reverse(test_holder, list, event_list) {
+		if (should_merge(test_holder->event, event)) {
+			test_event = test_holder->event;
+			break;
+		}
+	}
+
+	if (!test_event)
+		return NULL;
+
+	fsnotify_get_event(test_event);
+
+	/* if they are exactly the same we are done */
+	if (test_event->mask == event->mask)
+		return test_event;
+
+	/*
+	 * if the refcnt == 2 this is the only queue
+	 * for this event and so we can update the mask
+	 * in place.
+	 */
+	if (atomic_read(&test_event->refcnt) == 2) {
+		test_event->mask |= event->mask;
+		return test_event;
+	}
+
+	new_event = fsnotify_clone_event(test_event);
+
+	/* done with test_event */
+	fsnotify_put_event(test_event);
+
+	/* couldn't allocate memory, merge was not possible */
+	if (unlikely(!new_event))
+		return ERR_PTR(-ENOMEM);
+
+	/* build new event and replace it on the list */
+	new_event->mask = (test_event->mask | event->mask);
+	fsnotify_replace_event(test_holder, new_event);
+
+	/* we hold a reference on new_event from clone_event */
+	return new_event;
+}
+
+#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
+static int fanotify_get_response_from_access(struct fsnotify_group *group,
+					     struct fsnotify_event *event)
+{
+	int ret;
+
+	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
+
+	wait_event(group->fanotify_data.access_waitq, event->response);
+
+	/* userspace responded, convert to something usable */
+	spin_lock(&event->lock);
+	switch (event->response) {
+	case FAN_ALLOW:
+		ret = 0;
+		break;
+	case FAN_DENY:
+	default:
+		ret = -EPERM;
+	}
+	event->response = 0;
+	spin_unlock(&event->lock);
+
+	pr_debug("%s: group=%p event=%p about to return ret=%d\n", __func__,
+		 group, event, ret);
+	
+	return ret;
+}
+#endif
+
+static int fanotify_handle_event(struct fsnotify_group *group,
+				 struct fsnotify_mark *inode_mark,
+				 struct fsnotify_mark *fanotify_mark,
+				 struct fsnotify_event *event)
+{
+	int ret = 0;
+	struct fsnotify_event *notify_event = NULL;
+
+	BUILD_BUG_ON(FAN_ACCESS != FS_ACCESS);
+	BUILD_BUG_ON(FAN_MODIFY != FS_MODIFY);
+	BUILD_BUG_ON(FAN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
+	BUILD_BUG_ON(FAN_CLOSE_WRITE != FS_CLOSE_WRITE);
+	BUILD_BUG_ON(FAN_OPEN != FS_OPEN);
+	BUILD_BUG_ON(FAN_EVENT_ON_CHILD != FS_EVENT_ON_CHILD);
+	BUILD_BUG_ON(FAN_Q_OVERFLOW != FS_Q_OVERFLOW);
+	BUILD_BUG_ON(FAN_OPEN_PERM != FS_OPEN_PERM);
+	BUILD_BUG_ON(FAN_ACCESS_PERM != FS_ACCESS_PERM);
+
+	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
+
+	notify_event = fsnotify_add_notify_event(group, event, NULL, fanotify_merge);
+	if (IS_ERR(notify_event))
+		return PTR_ERR(notify_event);
+
+#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
+	if (event->mask & FAN_ALL_PERM_EVENTS) {
+		/* if we merged we need to wait on the new event */
+		if (notify_event)
+			event = notify_event;
+		ret = fanotify_get_response_from_access(group, event);
+	}
+#endif
+
+	if (notify_event)
+		fsnotify_put_event(notify_event);
+
+	return ret;
+}
+
+static bool fanotify_should_send_event(struct fsnotify_group *group,
+				       struct inode *to_tell,
+				       struct fsnotify_mark *inode_mark,
+				       struct fsnotify_mark *vfsmnt_mark,
+				       __u32 event_mask, void *data, int data_type)
+{
+	__u32 marks_mask, marks_ignored_mask;
+
+	pr_debug("%s: group=%p to_tell=%p inode_mark=%p vfsmnt_mark=%p "
+		 "mask=%x data=%p data_type=%d\n", __func__, group, to_tell,
+		 inode_mark, vfsmnt_mark, event_mask, data, data_type);
+
+	pr_debug("%s: group=%p vfsmount_mark=%p inode_mark=%p mask=%x\n",
+		 __func__, group, vfsmnt_mark, inode_mark, event_mask);
+
+	/* sorry, fanotify only gives a damn about files and dirs */
+	if (!S_ISREG(to_tell->i_mode) &&
+	    !S_ISDIR(to_tell->i_mode))
+		return false;
+
+	/* if we don't have enough info to send an event to userspace say no */
+	if (data_type != FSNOTIFY_EVENT_FILE)
+		return false;
+
+	if (inode_mark && vfsmnt_mark) {
+		marks_mask = (vfsmnt_mark->mask | inode_mark->mask);
+		marks_ignored_mask = (vfsmnt_mark->ignored_mask | inode_mark->ignored_mask);
+	} else if (inode_mark) {
+		/*
+		 * if the event is for a child and this inode doesn't care about
+		 * events on the child, don't send it!
+		 */
+		if ((event_mask & FS_EVENT_ON_CHILD) &&
+		    !(inode_mark->mask & FS_EVENT_ON_CHILD))
+			return false;
+		marks_mask = inode_mark->mask;
+		marks_ignored_mask = inode_mark->ignored_mask;
+	} else if (vfsmnt_mark) {
+		marks_mask = vfsmnt_mark->mask;
+		marks_ignored_mask = vfsmnt_mark->ignored_mask;
+	} else {
+		BUG();
+	}
+
+	if (event_mask & marks_mask & ~marks_ignored_mask)
+		return true;
+
+	return false;
+}
+
+const struct fsnotify_ops fanotify_fsnotify_ops = {
+	.handle_event = fanotify_handle_event,
+	.should_send_event = fanotify_should_send_event,
+	.free_group_priv = NULL,
+	.free_event_priv = NULL,
+	.freeing_mark = NULL,
+};

fs/notify/fanotify/fanotify_user.c

+#include <linux/fanotify.h>
+#include <linux/fcntl.h>
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/anon_inodes.h>
+#include <linux/fsnotify_backend.h>
+#include <linux/init.h>
+#include <linux/mount.h>
+#include <linux/namei.h>
+#include <linux/poll.h>
+#include <linux/security.h>
+#include <linux/syscalls.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+#include <linux/uaccess.h>
+
+#include <asm/ioctls.h>
+
+extern const struct fsnotify_ops fanotify_fsnotify_ops;
+
+static struct kmem_cache *fanotify_mark_cache __read_mostly;
+static struct kmem_cache *fanotify_response_event_cache __read_mostly;
+
+struct fanotify_response_event {
+	struct list_head list;
+	__s32 fd;
+	struct fsnotify_event *event;
+};
+
+/*
+ * Get an fsnotify notification event if one exists and is small
+ * enough to fit in "count". Return an error pointer if the count
+ * is not large enough.
+ *
+ * Called with the group->notification_mutex held.
+ */
+static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
+					    size_t count)
+{
+	BUG_ON(!mutex_is_locked(&group->notification_mutex));
+
+	pr_debug("%s: group=%p count=%zd\n", __func__, group, count);
+
+	if (fsnotify_notify_queue_is_empty(group))
+		return NULL;
+
+	if (FAN_EVENT_METADATA_LEN > count)
+		return ERR_PTR(-EINVAL);
+
+	/* held the notification_mutex the whole time, so this is the
+	 * same event we peeked above */
+	return fsnotify_remove_notify_event(group);
+}
+
+static int create_fd(struct fsnotify_group *group, struct fsnotify_event *event)
+{
+	int client_fd;
+	struct dentry *dentry;
+	struct vfsmount *mnt;
+	struct file *new_file;
+
+	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
+
+	client_fd = get_unused_fd();
+	if (client_fd < 0)
+		return client_fd;
+
+	if (event->data_type != FSNOTIFY_EVENT_FILE) {
+		WARN_ON(1);
+		put_unused_fd(client_fd);
+		return -EINVAL;
+	}
+
+	/*
+	 * we need a new file handle for the userspace program so it can read even if it was
+	 * originally opened O_WRONLY.
+	 */
+	dentry = dget(event->file->f_path.dentry);
+	mnt = mntget(event->file->f_path.mnt);
+	/* it's possible this event was an overflow event.  in that case dentry and mnt
+	 * are NULL;  That's fine, just don't call dentry open */
+	if (dentry && mnt)
+		new_file = dentry_open(dentry, mnt,
+				       group->fanotify_data.f_flags | FMODE_NONOTIFY,
+				       current_cred());
+	else
+		new_file = ERR_PTR(-EOVERFLOW);
+	if (IS_ERR(new_file)) {
+		/*
+		 * we still send an event even if we can't open the file.  this
+		 * can happen when say tasks are gone and we try to open their
+		 * /proc files or we try to open a WRONLY file like in sysfs
+		 * we just send the errno to userspace since there isn't much
+		 * else we can do.
+		 */
+		put_unused_fd(client_fd);
+		client_fd = PTR_ERR(new_file);
+	} else {
+		fd_install(client_fd, new_file);
+	}
+
+	return client_fd;
+}
+
+static ssize_t fill_event_metadata(struct fsnotify_group *group,
+				   struct fanotify_event_metadata *metadata,
+				   struct fsnotify_event *event)
+{
+	pr_debug("%s: group=%p metadata=%p event=%p\n", __func__,
+		 group, metadata, event);
+
+	metadata->event_len = FAN_EVENT_METADATA_LEN;
+	metadata->vers = FANOTIFY_METADATA_VERSION;
+	metadata->mask = event->mask & FAN_ALL_OUTGOING_EVENTS;
+	metadata->pid = pid_vnr(event->tgid);
+	metadata->fd = create_fd(group, event);
+
+	return metadata->fd;
+}
+
+#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
+static struct fanotify_response_event *dequeue_re(struct fsnotify_group *group,
+						  __s32 fd)
+{
+	struct fanotify_response_event *re, *return_re = NULL;
+
+	mutex_lock(&group->fanotify_data.access_mutex);
+	list_for_each_entry(re, &group->fanotify_data.access_list, list) {
+		if (re->fd != fd)
+			continue;
+
+		list_del_init(&re->list);
+		return_re = re;
+		break;
+	}
+	mutex_unlock(&group->fanotify_data.access_mutex);
+
+	pr_debug("%s: found return_re=%p\n", __func__, return_re);
+
+	return return_re;
+}
+
+static int process_access_response(struct fsnotify_group *group,
+				   struct fanotify_response *response_struct)
+{
+	struct fanotify_response_event *re;
+	__s32 fd = response_struct->fd;
+	__u32 response = response_struct->response;
+
+	pr_debug("%s: group=%p fd=%d response=%d\n", __func__, group,
+		 fd, response);
+	/*
+	 * make sure the response is valid, if invalid we do nothing and either
+	 * userspace can send a valid responce or we will clean it up after the
+	 * timeout
+	 */
+	switch (response) {
+	case FAN_ALLOW:
+	case FAN_DENY:
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	if (fd < 0)
+		return -EINVAL;
+
+	re = dequeue_re(group, fd);
+	if (!re)
+		return -ENOENT;
+
+	re->event->response = response;
+
+	wake_up(&group->fanotify_data.access_waitq);
+
+	kmem_cache_free(fanotify_response_event_cache, re);
+
+	return 0;
+}
+
+static int prepare_for_access_response(struct fsnotify_group *group,
+				       struct fsnotify_event *event,
+				       __s32 fd)
+{
+	struct fanotify_response_event *re;
+
+	if (!(event->mask & FAN_ALL_PERM_EVENTS))
+		return 0;
+
+	re = kmem_cache_alloc(fanotify_response_event_cache, GFP_KERNEL);
+	if (!re)
+		return -ENOMEM;
+
+	re->event = event;
+	re->fd = fd;
+
+	mutex_lock(&group->fanotify_data.access_mutex);
+	list_add_tail(&re->list, &group->fanotify_data.access_list);
+	mutex_unlock(&group->fanotify_data.access_mutex);
+
+	return 0;
+}
+
+static void remove_access_response(struct fsnotify_group *group,
+				   struct fsnotify_event *event,
+				   __s32 fd)
+{
+	struct fanotify_response_event *re;
+
+	if (!(event->mask & FAN_ALL_PERM_EVENTS))
+		return;
+
+	re = dequeue_re(group, fd);
+	if (!re)
+		return;
+
+	BUG_ON(re->event != event);
+
+	kmem_cache_free(fanotify_response_event_cache, re);
+
+	return;
+}
+#else
+static int prepare_for_access_response(struct fsnotify_group *group,
+				       struct fsnotify_event *event,
+				       __s32 fd)
+{
+	return 0;
+}
+
+static void remove_access_response(struct fsnotify_group *group,
+				   struct fsnotify_event *event,
+				   __s32 fd)
+{
+	return;
+}
+#endif
+
+static ssize_t copy_event_to_user(struct fsnotify_group *group,
+				  struct fsnotify_event *event,
+				  char __user *buf)
+{
+	struct fanotify_event_metadata fanotify_event_metadata;
+	int fd, ret;
+
+	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
+
+	fd = fill_event_metadata(group, &fanotify_event_metadata, event);
+	if (fd < 0)
+		return fd;
+
+	ret = prepare_for_access_response(group, event, fd);
+	if (ret)
+		goto out_close_fd;
+
+	ret = -EFAULT;
+	if (copy_to_user(buf, &fanotify_event_metadata, FAN_EVENT_METADATA_LEN))
+		goto out_kill_access_response;
+
+	return FAN_EVENT_METADATA_LEN;
+
+out_kill_access_response:
+	remove_access_response(group, event, fd);
+out_close_fd:
+	sys_close(fd);
+	return ret;
+}
+
+/* intofiy userspace file descriptor functions */
+static unsigned int fanotify_poll(struct file *file, poll_table *wait)
+{
+	struct fsnotify_group *group = file->private_data;
+	int ret = 0;
+
+	poll_wait(file, &group->notification_waitq, wait);
+	mutex_lock(&group->notification_mutex);
+	if (!fsnotify_notify_queue_is_empty(group))
+		ret = POLLIN | POLLRDNORM;
+	mutex_unlock(&group->notification_mutex);
+
+	return ret;
+}
+
+static ssize_t fanotify_read(struct file *file, char __user *buf,
+			     size_t count, loff_t *pos)
+{
+	struct fsnotify_group *group;
+	struct fsnotify_event *kevent;
+	char __user *start;
+	int ret;
+	DEFINE_WAIT(wait);
+
+	start = buf;
+	group = file->private_data;
+
+	pr_debug("%s: group=%p\n", __func__, group);
+
+	while (1) {
+		prepare_to_wait(&group->notification_waitq, &wait, TASK_INTERRUPTIBLE);
+
+		mutex_lock(&group->notification_mutex);
+		kevent = get_one_event(group, count);
+		mutex_unlock(&group->notification_mutex);
+
+		if (kevent) {
+			ret = PTR_ERR(kevent);
+			if (IS_ERR(kevent))
+				break;
+			ret = copy_event_to_user(group, kevent, buf);
+			fsnotify_put_event(kevent);
+			if (ret < 0)
+				break;
+			buf += ret;
+			count -= ret;
+			continue;
+		}
+
+		ret = -EAGAIN;
+		if (file->f_flags & O_NONBLOCK)
+			break;
+		ret = -EINTR;
+		if (signal_pending(current))
+			break;
+
+		if (start != buf)
+			break;
+
+		schedule();
+	}
+
+	finish_wait(&group->notification_waitq, &wait);
+	if (start != buf && ret != -EFAULT)
+		ret = buf - start;
+	return ret;
+}
+
+static ssize_t fanotify_write(struct file *file, const char __user *buf, size_t count, loff_t *pos)
+{
+#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
+	struct fanotify_response response = { .fd = -1, .response = -1 };
+	struct fsnotify_group *group;
+	int ret;
+
+	group = file->private_data;
+
+	if (count > sizeof(response))
+		count = sizeof(response);
+
+	pr_debug("%s: group=%p count=%zu\n", __func__, group, count);
+
+	if (copy_from_user(&response, buf, count))
+		return -EFAULT;
+
+	ret = process_access_response(group, &response);
+	if (ret < 0)
+		count = ret;
+
+	return count;
+#else
+	return -EINVAL;
+#endif
+}
+
+static int fanotify_release(struct inode *ignored, struct file *file)
+{
+	struct fsnotify_group *group = file->private_data;
+
+	pr_debug("%s: file=%p group=%p\n", __func__, file, group);
+
+	/* matches the fanotify_init->fsnotify_alloc_group */
+	fsnotify_put_group(group);
+
+	return 0;
+}
+
+static long fanotify_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+	struct fsnotify_group *group;
+	struct fsnotify_event_holder *holder;
+	void __user *p;
+	int ret = -ENOTTY;
+	size_t send_len = 0;
+
+	group = file->private_data;
+
+	p = (void __user *) arg;
+
+	switch (cmd) {
+	case FIONREAD:
+		mutex_lock(&group->notification_mutex);
+		list_for_each_entry(holder, &group->notification_list, event_list)
+			send_len += FAN_EVENT_METADATA_LEN;
+		mutex_unlock(&group->notification_mutex);
+		ret = put_user(send_len, (int __user *) p);
+		break;
+	}
+
+	return ret;
+}
+
+static const struct file_operations fanotify_fops = {
+	.poll		= fanotify_poll,
+	.read		= fanotify_read,
+	.write		= fanotify_write,
+	.fasync		= NULL,
+	.release	= fanotify_release,
+	.unlocked_ioctl	= fanotify_ioctl,
+	.compat_ioctl	= fanotify_ioctl,
+};
+
+static void fanotify_free_mark(struct fsnotify_mark *fsn_mark)
+{
+	kmem_cache_free(fanotify_mark_cache, fsn_mark);
+}
+
+static int fanotify_find_path(int dfd, const char __user *filename,
+			      struct path *path, unsigned int flags)
+{
+	int ret;
+
+	pr_debug("%s: dfd=%d filename=%p flags=%x\n", __func__,
+		 dfd, filename, flags);
+
+	if (filename == NULL) {
+		struct file *file;
+		int fput_needed;
+
+		ret = -EBADF;
+		file = fget_light(dfd, &fput_needed);
+		if (!file)
+			goto out;
+
+		ret = -ENOTDIR;
+		if ((flags & FAN_MARK_ONLYDIR) &&
+		    !(S_ISDIR(file->f_path.dentry->d_inode->i_mode))) {
+			fput_light(file, fput_needed);
+			goto out;
+		}
+
+		*path = file->f_path;
+		path_get(path);
+		fput_light(file, fput_needed);
+	} else {
+		unsigned int lookup_flags = 0;
+
+		if (!(flags & FAN_MARK_DONT_FOLLOW))
+			lookup_flags |= LOOKUP_FOLLOW;
+		if (flags & FAN_MARK_ONLYDIR)
+			lookup_flags |= LOOKUP_DIRECTORY;
+
+		ret = user_path_at(dfd, filename, lookup_flags, path);
+		if (ret)
+			goto out;
+	}
+
+	/* you can only watch an inode if you have read permissions on it */
+	ret = inode_permission(path->dentry->d_inode, MAY_READ);
+	if (ret)
+		path_put(path);
+out:
+	return ret;
+}
+
+static __u32 fanotify_mark_remove_from_mask(struct fsnotify_mark *fsn_mark,
+					    __u32 mask,
+					    unsigned int flags)
+{
+	__u32 oldmask;
+
+	spin_lock(&fsn_mark->lock);
+	if (!(flags & FAN_MARK_IGNORED_MASK)) {
+		oldmask = fsn_mark->mask;
+		fsnotify_set_mark_mask_locked(fsn_mark, (oldmask & ~mask));
+	} else {
+		oldmask = fsn_mark->ignored_mask;
+		fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask & ~mask));
+	}
+	spin_unlock(&fsn_mark->lock);
+
+	if (!(oldmask & ~mask))
+		fsnotify_destroy_mark(fsn_mark);
+
+	return mask & oldmask;
+}
+
+static int fanotify_remove_vfsmount_mark(struct fsnotify_group *group,
+					 struct vfsmount *mnt, __u32 mask,
+					 unsigned int flags)
+{
+	struct fsnotify_mark *fsn_mark = NULL;
+	__u32 removed;
+
+	fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
+	if (!fsn_mark)
+		return -ENOENT;
+
+	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
+	fsnotify_put_mark(fsn_mark);
+	if (removed & mnt->mnt_fsnotify_mask)
+		fsnotify_recalc_vfsmount_mask(mnt);
+
+	return 0;
+}
+
+static int fanotify_remove_inode_mark(struct fsnotify_group *group,
+				      struct inode *inode, __u32 mask,
+				      unsigned int flags)
+{
+	struct fsnotify_mark *fsn_mark = NULL;
+	__u32 removed;
+
+	fsn_mark = fsnotify_find_inode_mark(group, inode);
+	if (!fsn_mark)
+		return -ENOENT;
+
+	removed = fanotify_mark_remove_from_mask(fsn_mark, mask, flags);
+	/* matches the fsnotify_find_inode_mark() */
+	fsnotify_put_mark(fsn_mark);
+	if (removed & inode->i_fsnotify_mask)
+		fsnotify_recalc_inode_mask(inode);
+
+	return 0;
+}
+
+static __u32 fanotify_mark_add_to_mask(struct fsnotify_mark *fsn_mark,
+				       __u32 mask,
+				       unsigned int flags)
+{
+	__u32 oldmask;
+
+	spin_lock(&fsn_mark->lock);
+	if (!(flags & FAN_MARK_IGNORED_MASK)) {
+		oldmask = fsn_mark->mask;
+		fsnotify_set_mark_mask_locked(fsn_mark, (oldmask | mask));
+	} else {
+		oldmask = fsn_mark->ignored_mask;
+		fsnotify_set_mark_ignored_mask_locked(fsn_mark, (oldmask | mask));
+		if (flags & FAN_MARK_IGNORED_SURV_MODIFY)
+			fsn_mark->flags |= FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY;
+	}
+	spin_unlock(&fsn_mark->lock);
+
+	return mask & ~oldmask;
+}
+
+static int fanotify_add_vfsmount_mark(struct fsnotify_group *group,
+				      struct vfsmount *mnt, __u32 mask,
+				      unsigned int flags)
+{
+	struct fsnotify_mark *fsn_mark;
+	__u32 added;
+
+	fsn_mark = fsnotify_find_vfsmount_mark(group, mnt);
+	if (!fsn_mark) {
+		int ret;
+
+		fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
+		if (!fsn_mark)
+			return -ENOMEM;
+
+		fsnotify_init_mark(fsn_mark, fanotify_free_mark);
+		ret = fsnotify_add_mark(fsn_mark, group, NULL, mnt, 0);
+		if (ret) {
+			fanotify_free_mark(fsn_mark);
+			return ret;
+		}
+	}
+	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
+	fsnotify_put_mark(fsn_mark);
+	if (added & ~mnt->mnt_fsnotify_mask)
+		fsnotify_recalc_vfsmount_mask(mnt);
+
+	return 0;
+}
+
+static int fanotify_add_inode_mark(struct fsnotify_group *group,
+				   struct inode *inode, __u32 mask,
+				   unsigned int flags)
+{
+	struct fsnotify_mark *fsn_mark;
+	__u32 added;
+
+	pr_debug("%s: group=%p inode=%p\n", __func__, group, inode);
+
+	fsn_mark = fsnotify_find_inode_mark(group, inode);
+	if (!fsn_mark) {
+		int ret;
+
+		fsn_mark = kmem_cache_alloc(fanotify_mark_cache, GFP_KERNEL);
+		if (!fsn_mark)
+			return -ENOMEM;
+
+		fsnotify_init_mark(fsn_mark, fanotify_free_mark);
+		ret = fsnotify_add_mark(fsn_mark, group, inode, NULL, 0);
+		if (ret) {
+			fanotify_free_mark(fsn_mark);
+			return ret;
+		}
+	}
+	added = fanotify_mark_add_to_mask(fsn_mark, mask, flags);
+	fsnotify_put_mark(fsn_mark);
+	if (added & ~inode->i_fsnotify_mask)
+		fsnotify_recalc_inode_mask(inode);
+	return 0;
+}
+
+/* fanotify syscalls */
+SYSCALL_DEFINE2(fanotify_init, unsigned int, flags, unsigned int, event_f_flags)
+{
+	struct fsnotify_group *group;
+	int f_flags, fd;
+
+	pr_debug("%s: flags=%d event_f_flags=%d\n",
+		__func__, flags, event_f_flags);
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EACCES;
+
+	if (flags & ~FAN_ALL_INIT_FLAGS)
+		return -EINVAL;
+
+	f_flags = O_RDWR | FMODE_NONOTIFY;
+	if (flags & FAN_CLOEXEC)
+		f_flags |= O_CLOEXEC;
+	if (flags & FAN_NONBLOCK)
+		f_flags |= O_NONBLOCK;
+
+	/* fsnotify_alloc_group takes a ref.  Dropped in fanotify_release */
+	group = fsnotify_alloc_group(&fanotify_fsnotify_ops);
+	if (IS_ERR(group))
+		return PTR_ERR(group);
+
+	group->fanotify_data.f_flags = event_f_flags;
+#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
+	mutex_init(&group->fanotify_data.access_mutex);
+	init_waitqueue_head(&group->fanotify_data.access_waitq);
+	INIT_LIST_HEAD(&group->fanotify_data.access_list);
+#endif
+
+	fd = anon_inode_getfd("[fanotify]", &fanotify_fops, group, f_flags);
+	if (fd < 0)
+		goto out_put_group;
+
+	return fd;
+
+out_put_group:
+	fsnotify_put_group(group);
+	return fd;
+}
+
+SYSCALL_DEFINE(fanotify_mark)(int fanotify_fd, unsigned int flags,
+			      __u64 mask, int dfd,
+			      const char  __user * pathname)
+{
+	struct inode *inode = NULL;
+	struct vfsmount *mnt = NULL;
+	struct fsnotify_group *group;
+	struct file *filp;
+	struct path path;
+	int ret, fput_needed;
+
+	pr_debug("%s: fanotify_fd=%d flags=%x dfd=%d pathname=%p mask=%llx\n",
+		 __func__, fanotify_fd, flags, dfd, pathname, mask);
+
+	/* we only use the lower 32 bits as of right now. */
+	if (mask & ((__u64)0xffffffff << 32))
+		return -EINVAL;
+
+	if (flags & ~FAN_ALL_MARK_FLAGS)
+		return -EINVAL;
+	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
+	case FAN_MARK_ADD:
+	case FAN_MARK_REMOVE:
+	case FAN_MARK_FLUSH:
+		break;
+	default:
+		return -EINVAL;
+	}
+#ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
+	if (mask & ~(FAN_ALL_EVENTS | FAN_ALL_PERM_EVENTS | FAN_EVENT_ON_CHILD))
+#else
+	if (mask & ~(FAN_ALL_EVENTS | FAN_EVENT_ON_CHILD))
+#endif
+		return -EINVAL;
+
+	filp = fget_light(fanotify_fd, &fput_needed);
+	if (unlikely(!filp))
+		return -EBADF;
+
+	/* verify that this is indeed an fanotify instance */
+	ret = -EINVAL;
+	if (unlikely(filp->f_op != &fanotify_fops))
+		goto fput_and_out;
+
+	ret = fanotify_find_path(dfd, pathname, &path, flags);
+	if (ret)
+		goto fput_and_out;
+
+	/* inode held in place by reference to path; group by fget on fd */
+	if (!(flags & FAN_MARK_MOUNT))
+		inode = path.dentry->d_inode;
+	else
+		mnt = path.mnt;
+	group = filp->private_data;
+
+	/* create/update an inode mark */
+	switch (flags & (FAN_MARK_ADD | FAN_MARK_REMOVE | FAN_MARK_FLUSH)) {
+	case FAN_MARK_ADD:
+		if (flags & FAN_MARK_MOUNT)
+			ret = fanotify_add_vfsmount_mark(group, mnt, mask, flags);
+		else
+			ret = fanotify_add_inode_mark(group, inode, mask, flags);
+		break;
+	case FAN_MARK_REMOVE:
+		if (flags & FAN_MARK_MOUNT)
+			ret = fanotify_remove_vfsmount_mark(group, mnt, mask, flags);
+		else
+			ret = fanotify_remove_inode_mark(group, inode, mask, flags);
+		break;
+	case FAN_MARK_FLUSH:
+		if (flags & FAN_MARK_MOUNT)
+			fsnotify_clear_vfsmount_marks_by_group(group);
+		else
+			fsnotify_clear_inode_marks_by_group(group);
+		break;
+	default:
+		ret = -EINVAL;
+	}
+
+	path_put(&path);
+fput_and_out:
+	fput_light(filp, fput_needed);
+	return ret;
+}
+
+#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
+asmlinkage long SyS_fanotify_mark(long fanotify_fd, long flags, __u64 mask,
+				  long dfd, long pathname)
+{
+	return SYSC_fanotify_mark((int) fanotify_fd, (unsigned int) flags,
+				  mask, (int) dfd,
+				  (const char  __user *) pathname);
+}
+SYSCALL_ALIAS(sys_fanotify_mark, SyS_fanotify_mark);
+#endif
+
+/*
+ * fanotify_user_setup - Our initialization function.  Note that we cannnot return
+ * error because we have compiled-in VFS hooks.  So an (unlikely) failure here
+ * must result in panic().
+ */
+static int __init fanotify_user_setup(void)
+{
+	fanotify_mark_cache = KMEM_CACHE(fsnotify_mark, SLAB_PANIC);
+	fanotify_response_event_cache = KMEM_CACHE(fanotify_response_event,
+						   SLAB_PANIC);
+
+	return 0;
+}
+device_initcall(fanotify_user_setup);

fs/notify/fsnotify.c

@@ -21,6 +21,7 @@
 #include <linux/gfp.h>
 #include <linux/init.h>
 #include <linux/module.h>
+#include <linux/mount.h>
 #include <linux/srcu.h>
 
 #include <linux/fsnotify_backend.h>
@@ -35,6 +36,11 @@ void __fsnotify_inode_delete(struct inode *inode)
 }
 EXPORT_SYMBOL_GPL(__fsnotify_inode_delete);
 
+void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
+{
+	fsnotify_clear_marks_by_mount(mnt);
+}
+
 /*
  * Given an inode, first check if we care what happens to our children.  Inotify
  * and dnotify both tell their parents about events.  If we care about any event
@@ -78,13 +84,16 @@ void __fsnotify_update_child_dentry_flags(struct inode *inode)
 }
 
 /* Notify this dentry's parent about a child's events. */
-void __fsnotify_parent(struct dentry *dentry, __u32 mask)
+void __fsnotify_parent(struct file *file, struct dentry *dentry, __u32 mask)
 {
 	struct dentry *parent;
 	struct inode *p_inode;
 	bool send = false;
 	bool should_update_children = false;
 
+	if (!dentry)
+		dentry = file->f_path.dentry;
+
 	if (!(dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED))
 		return;
 
@@ -115,8 +124,12 @@ void __fsnotify_parent(struct dentry *dentry, __u32 mask)
 		 * specifies these are events which came from a child. */
 		mask |= FS_EVENT_ON_CHILD;
 
-		fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE,
-			 dentry->d_name.name, 0);
+		if (file)
+			fsnotify(p_inode, mask, file, FSNOTIFY_EVENT_FILE,
+				 dentry->d_name.name, 0);
+		else
+			fsnotify(p_inode, mask, dentry->d_inode, FSNOTIFY_EVENT_INODE,
+				 dentry->d_name.name, 0);
 		dput(parent);
 	}
 
@@ -127,63 +140,181 @@ void __fsnotify_parent(struct dentry *dentry, __u32 mask)
 }
 EXPORT_SYMBOL_GPL(__fsnotify_parent);
 
+static int send_to_group(struct inode *to_tell, struct vfsmount *mnt,
+			 struct fsnotify_mark *inode_mark,
+			 struct fsnotify_mark *vfsmount_mark,
+			 __u32 mask, void *data,
+			 int data_is, u32 cookie,
+			 const unsigned char *file_name,
+			 struct fsnotify_event **event)
+{
+	struct fsnotify_group *group = inode_mark->group;
+	__u32 inode_test_mask = (mask & ~FS_EVENT_ON_CHILD);
+	__u32 vfsmount_test_mask = (mask & ~FS_EVENT_ON_CHILD);
+
+	pr_debug("%s: group=%p to_tell=%p mnt=%p mark=%p mask=%x data=%p"
+		 " data_is=%d cookie=%d event=%p\n", __func__, group, to_tell,
+		 mnt, inode_mark, mask, data, data_is, cookie, *event);
+
+	/* clear ignored on inode modification */
+	if (mask & FS_MODIFY) {
+		if (inode_mark &&
+		    !(inode_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
+			inode_mark->ignored_mask = 0;
+		if (vfsmount_mark &&
+		    !(vfsmount_mark->flags & FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY))
+			vfsmount_mark->ignored_mask = 0;
+	}
+
+	/* does the inode mark tell us to do something? */
+	if (inode_mark) {
+		inode_test_mask &= inode_mark->mask;
+		inode_test_mask &= ~inode_mark->ignored_mask;
+	}
+
+	/* does the vfsmount_mark tell us to do something? */
+	if (vfsmount_mark) {
+		vfsmount_test_mask &= vfsmount_mark->mask;
+		vfsmount_test_mask &= ~vfsmount_mark->ignored_mask;
+		if (inode_mark)
+			vfsmount_test_mask &= ~inode_mark->ignored_mask;
+	}
+
+	if (!inode_test_mask && !vfsmount_test_mask)
+		return 0;
+
+	if (group->ops->should_send_event(group, to_tell, inode_mark,
+					  vfsmount_mark, mask, data,
+					  data_is) == false)
+		return 0;
+
+	if (!*event) {
+		*event = fsnotify_create_event(to_tell, mask, data,
+						data_is, file_name,
+						cookie, GFP_KERNEL);
+		if (!*event)
+			return -ENOMEM;
+	}
+	return group->ops->handle_event(group, inode_mark, vfsmount_mark, *event);
+}
+
 /*
  * 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
  * notification event in whatever means they feel necessary.
  */
-void fsnotify(struct inode *to_tell, __u32 mask, void *data, int data_is, const char *file_name, u32 cookie)
+int fsnotify(struct inode *to_tell, __u32 mask, void *data, int data_is,
+	     const unsigned char *file_name, u32 cookie)
 {
-	struct fsnotify_group *group;
+	struct hlist_node *inode_node, *vfsmount_node;
+	struct fsnotify_mark *inode_mark = NULL, *vfsmount_mark = NULL;
+	struct fsnotify_group *inode_group, *vfsmount_group;
 	struct fsnotify_event *event = NULL;
-	int idx;
+	struct vfsmount *mnt;
+	int idx, ret = 0;
+	bool used_inode = false, used_vfsmount = false;
 	/* global tests shouldn't care about events on child only the specific event */
 	__u32 test_mask = (mask & ~FS_EVENT_ON_CHILD);
 
-	if (list_empty(&fsnotify_groups))
-		return;
+	if (data_is == FSNOTIFY_EVENT_FILE)
+		mnt = ((struct file *)data)->f_path.mnt;
+	else
+		mnt = NULL;
 
-	if (!(test_mask & fsnotify_mask))
-		return;
-
-	if (!(test_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
-	 * anything other than walk the list so it's crazy to pre-allocate.
+	 * if this is a modify event we may need to clear the ignored masks
+	 * otherwise return if neither the inode nor the vfsmount care about
+	 * this type of event.
 	 */
-	idx = srcu_read_lock(&fsnotify_grp_srcu);
-	list_for_each_entry_rcu(group, &fsnotify_groups, group_list) {
-		if (test_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, file_name, cookie,
-							      GFP_KERNEL);
-				/* shit, we OOM'd and now we can't tell, maybe
-				 * someday someone else will want to do something
-				 * here */
-				if (!event)
-					break;
-			}
-			group->ops->handle_event(group, event);
+	if (!(mask & FS_MODIFY) &&
+	    !(test_mask & to_tell->i_fsnotify_mask) &&
+	    !(mnt && test_mask & mnt->mnt_fsnotify_mask))
+		return 0;
+
+	idx = srcu_read_lock(&fsnotify_mark_srcu);
+
+	if ((mask & FS_MODIFY) ||
+	    (test_mask & to_tell->i_fsnotify_mask))
+		inode_node = srcu_dereference(to_tell->i_fsnotify_marks.first,
+					      &fsnotify_mark_srcu);
+	else
+		inode_node = NULL;
+
+	if (mnt) {
+		if ((mask & FS_MODIFY) ||
+		    (test_mask & mnt->mnt_fsnotify_mask))
+			vfsmount_node = srcu_dereference(mnt->mnt_fsnotify_marks.first,
+							 &fsnotify_mark_srcu);
+		else
+			vfsmount_node = NULL;
+	} else {
+		mnt = NULL;
+		vfsmount_node = NULL;
+	}
+
+	while (inode_node || vfsmount_node) {
+		if (inode_node) {
+			inode_mark = hlist_entry(srcu_dereference(inode_node, &fsnotify_mark_srcu),
+						 struct fsnotify_mark, i.i_list);
+			inode_group = inode_mark->group;
+		} else
+			inode_group = (void *)-1;
+
+		if (vfsmount_node) {
+			vfsmount_mark = hlist_entry(srcu_dereference(vfsmount_node, &fsnotify_mark_srcu),
+							struct fsnotify_mark, m.m_list);
+			vfsmount_group = vfsmount_mark->group;
+		} else
+			vfsmount_group = (void *)-1;
+
+		if (inode_group < vfsmount_group) {
+			/* handle inode */
+			send_to_group(to_tell, NULL, inode_mark, NULL, mask, data,
+				      data_is, cookie, file_name, &event);
+			used_inode = true;
+		} else if (vfsmount_group < inode_group) {
+			send_to_group(to_tell, mnt, NULL, vfsmount_mark, mask, data,
+				      data_is, cookie, file_name, &event);
+			used_vfsmount = true;
+		} else {
+			send_to_group(to_tell, mnt, inode_mark, vfsmount_mark,
+				      mask, data, data_is, cookie, file_name,
+				      &event);
+			used_vfsmount = true;
+			used_inode = true;
 		}
+
+		if (used_inode)
+			inode_node = srcu_dereference(inode_node->next,
+						      &fsnotify_mark_srcu);
+		if (used_vfsmount)
+			vfsmount_node = srcu_dereference(vfsmount_node->next,
+							 &fsnotify_mark_srcu);
 	}
-	srcu_read_unlock(&fsnotify_grp_srcu, idx);
+
+	srcu_read_unlock(&fsnotify_mark_srcu, idx);
 	/*
 	 * fsnotify_create_event() took a reference so the event can't be cleaned
 	 * up while we are still trying to add it to lists, drop that one.
 	 */
 	if (event)
 		fsnotify_put_event(event);
+
+	return ret;
 }
 EXPORT_SYMBOL_GPL(fsnotify);
 
 static __init int fsnotify_init(void)
 {
-	return init_srcu_struct(&fsnotify_grp_srcu);
+	int ret;
+
+	BUG_ON(hweight32(ALL_FSNOTIFY_EVENTS) != 23);
+
+	ret = init_srcu_struct(&fsnotify_mark_srcu);
+	if (ret)
+		panic("initializing fsnotify_mark_srcu");
+
+	return 0;
 }
-subsys_initcall(fsnotify_init);
+core_initcall(fsnotify_init);

fs/notify/fsnotify.h

@@ -6,21 +6,34 @@
 #include <linux/srcu.h>
 #include <linux/types.h>
 
-/* protects reads of fsnotify_groups */
-extern struct srcu_struct fsnotify_grp_srcu;
-/* all groups which receive fsnotify events */
-extern struct list_head fsnotify_groups;
-/* all bitwise OR of all event types (FS_*) for all fsnotify_groups */
-extern __u32 fsnotify_mask;
-
 /* destroy all events sitting in this groups notification queue */
 extern void fsnotify_flush_notify(struct fsnotify_group *group);
 
+/* protects reads of inode and vfsmount marks list */
+extern struct srcu_struct fsnotify_mark_srcu;
+
+extern void fsnotify_set_inode_mark_mask_locked(struct fsnotify_mark *fsn_mark,
+						__u32 mask);
+/* add a mark to an inode */
+extern int fsnotify_add_inode_mark(struct fsnotify_mark *mark,
+				   struct fsnotify_group *group, struct inode *inode,
+				   int allow_dups);
+/* add a mark to a vfsmount */
+extern int fsnotify_add_vfsmount_mark(struct fsnotify_mark *mark,
+				      struct fsnotify_group *group, struct vfsmount *mnt,
+				      int allow_dups);
+
 /* final kfree of a group */
 extern void fsnotify_final_destroy_group(struct fsnotify_group *group);
 
+/* vfsmount specific destruction of a mark */
+extern void fsnotify_destroy_vfsmount_mark(struct fsnotify_mark *mark);
+/* inode specific destruction of a mark */
+extern void fsnotify_destroy_inode_mark(struct fsnotify_mark *mark);
 /* 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);
+/* run the list of all marks associated with vfsmount and flag them to be freed */
+extern void fsnotify_clear_marks_by_mount(struct vfsmount *mnt);
 /*
  * update the dentry->d_flags of all of inode's children to indicate if inode cares
  * about events that happen to its children.

fs/notify/group.c

@@ -28,64 +28,6 @@
 
 #include <asm/atomic.h>
 
-/* protects writes to fsnotify_groups and fsnotify_mask */
-static DEFINE_MUTEX(fsnotify_grp_mutex);
-/* protects reads while running the fsnotify_groups list */
-struct srcu_struct fsnotify_grp_srcu;
-/* all groups registered to receive filesystem notifications */
-LIST_HEAD(fsnotify_groups);
-/* bitwise OR of all events (FS_*) interesting to some group on this system */
-__u32 fsnotify_mask;
-
-/*
- * When a new group registers or changes it's set of interesting events
- * this function updates the fsnotify_mask to contain all interesting events
- */
-void fsnotify_recalc_global_mask(void)
-{
-	struct fsnotify_group *group;
-	__u32 mask = 0;
-	int idx;
-
-	idx = srcu_read_lock(&fsnotify_grp_srcu);
-	list_for_each_entry_rcu(group, &fsnotify_groups, group_list)
-		mask |= group->mask;
-	srcu_read_unlock(&fsnotify_grp_srcu, idx);
-	fsnotify_mask = mask;
-}
-
-/*
- * 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
- */
-static void fsnotify_get_group(struct fsnotify_group *group)
-{
-	atomic_inc(&group->refcnt);
-}
-
 /*
  * Final freeing of a group
  */
@@ -110,145 +52,53 @@ void fsnotify_final_destroy_group(struct fsnotify_group *group)
  */
 static void fsnotify_destroy_group(struct fsnotify_group *group)
 {
-	/* clear all inode mark entries for this group */
+	/* clear all inode marks for this group */
 	fsnotify_clear_marks_by_group(group);
 
+	synchronize_srcu(&fsnotify_mark_srcu);
+
 	/* 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.
- */
-static void __fsnotify_evict_group(struct fsnotify_group *group)
-{
-	BUG_ON(!mutex_is_locked(&fsnotify_grp_mutex));
-
-	if (group->on_group_list)
-		list_del_rcu(&group->group_list);
-	group->on_group_list = 0;
-}
-
-/*
- * Called when a group is no longer interested in getting events.  This can be
- * used if a group is misbehaving or if for some reason a group should no longer
- * get any filesystem events.
- */
-void fsnotify_evict_group(struct fsnotify_group *group)
-{
-	mutex_lock(&fsnotify_grp_mutex);
-	__fsnotify_evict_group(group);
-	mutex_unlock(&fsnotify_grp_mutex);
-}
-
 /*
  * Drop a reference to a group.  Free it if it's through.
  */
 void fsnotify_put_group(struct fsnotify_group *group)
 {
-	if (!atomic_dec_and_mutex_lock(&group->refcnt, &fsnotify_grp_mutex))
-		return;
-
-	/*
-	 * OK, now we know that there's no other users *and* we hold mutex,
-	 * so no new references will appear
-	 */
-	__fsnotify_evict_group(group);
-
-	/*
-	 * now it's off the list, so the only thing we might care about is
-	 * srcu access....
-	 */
-	mutex_unlock(&fsnotify_grp_mutex);
-	synchronize_srcu(&fsnotify_grp_srcu);
-
-	/* and now it is really dead. _Nothing_ could be seeing it */
-	fsnotify_recalc_global_mask();
-	fsnotify_destroy_group(group);
-}
-
-/*
- * Simply run the fsnotify_groups list and find a group which matches
- * the given parameters.  If a group is found we take a reference to that
- * group.
- */
-static struct fsnotify_group *fsnotify_find_group(unsigned int group_num, __u32 mask,
-						  const struct fsnotify_ops *ops)
-{
-	struct fsnotify_group *group_iter;
-	struct fsnotify_group *group = NULL;
-
-	BUG_ON(!mutex_is_locked(&fsnotify_grp_mutex));
-
-	list_for_each_entry_rcu(group_iter, &fsnotify_groups, group_list) {
-		if (group_iter->group_num == group_num) {
-			if ((group_iter->mask == mask) &&
-			    (group_iter->ops == ops)) {
-				fsnotify_get_group(group_iter);
-				group = group_iter;
-			} else
-				group = ERR_PTR(-EEXIST);
-		}
-	}
-	return group;
+	if (atomic_dec_and_test(&group->refcnt))
+		fsnotify_destroy_group(group);
 }
 
 /*
- * Either finds an existing group which matches the group_num, mask, and ops or
- * creates a new group and adds it to the global group list.  In either case we
- * take a reference for the group returned.
+ * Create a new fsnotify_group and hold a reference for the group returned.
  */
-struct fsnotify_group *fsnotify_obtain_group(unsigned int group_num, __u32 mask,
-					     const struct fsnotify_ops *ops)
+struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops)
 {
-	struct fsnotify_group *group, *tgroup;
+	struct fsnotify_group *group;
 
-	/* very low use, simpler locking if we just always alloc */
-	group = kmalloc(sizeof(struct fsnotify_group), GFP_KERNEL);
+	group = kzalloc(sizeof(struct fsnotify_group), GFP_KERNEL);
 	if (!group)
 		return ERR_PTR(-ENOMEM);
 
+	/* set to 0 when there a no external references to this group */
 	atomic_set(&group->refcnt, 1);
-
-	group->on_group_list = 0;
-	group->group_num = group_num;
-	group->mask = mask;
+	/*
+	 * hits 0 when there are no external references AND no marks for
+	 * this group
+	 */
+	atomic_set(&group->num_marks, 1);
 
 	mutex_init(&group->notification_mutex);
 	INIT_LIST_HEAD(&group->notification_list);
 	init_waitqueue_head(&group->notification_waitq);
-	group->q_len = 0;
 	group->max_events = UINT_MAX;
 
 	spin_lock_init(&group->mark_lock);
-	atomic_set(&group->num_marks, 0);
-	INIT_LIST_HEAD(&group->mark_entries);
+	INIT_LIST_HEAD(&group->marks_list);
 
 	group->ops = ops;
 
-	mutex_lock(&fsnotify_grp_mutex);
-	tgroup = fsnotify_find_group(group_num, mask, ops);
-	if (tgroup) {
-		/* group already exists */
-		mutex_unlock(&fsnotify_grp_mutex);
-		/* destroy the new one we made */
-		fsnotify_put_group(group);
-		return tgroup;
-	}
-
-	/* 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);
-
-	if (mask)
-		fsnotify_recalc_global_mask();
-
 	return group;
 }

fs/notify/inode_mark.c

@@ -16,72 +16,6 @@
  *  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>
@@ -95,30 +29,19 @@
 #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 fsnotify_mark *mark;
 	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;
+	hlist_for_each_entry(mark, pos, &inode->i_fsnotify_marks, i.i_list)
+		new_mask |= mark->mask;
 	inode->i_fsnotify_mask = new_mask;
 }
 
@@ -135,107 +58,26 @@ void fsnotify_recalc_inode_mask(struct inode *inode)
 	__fsnotify_update_child_dentry_flags(inode);
 }
 
-/*
- * 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)
+void fsnotify_destroy_inode_mark(struct fsnotify_mark *mark)
 {
-	struct fsnotify_group *group;
-	struct inode *inode;
+	struct inode *inode = mark->i.inode;
 
-	spin_lock(&entry->lock);
+	assert_spin_locked(&mark->lock);
+	assert_spin_locked(&mark->group->mark_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 */
+	hlist_del_init_rcu(&mark->i.i_list);
+	mark->i.inode = NULL;
 
 	/*
-	 * this mark is now off the inode->i_fsnotify_mark_entries list and we
+	 * this mark is now off the inode->i_fsnotify_marks 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.
-	 */
-	if (group->ops->freeing_mark)
-		group->ops->freeing_mark(entry, group);
-
-	/*
-	 * __fsnotify_update_child_dentry_flags(inode);
-	 *
-	 * I really want to call that, but we can't, we have no idea if the inode
-	 * still exists the second we drop the entry->lock.
-	 *
-	 * The next time an event arrive to this inode from one of it's children
-	 * __fsnotify_parent will see that the inode doesn't care about it's
-	 * children and will update all of these flags then.  So really this
-	 * is just a lazy update (and could be a perf win...)
-	 */
-
-
-	iput(inode);
-
-	/*
-	 * 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);
-	}
 }
 
 /*
@@ -243,112 +85,145 @@ void fsnotify_clear_marks_by_group(struct fsnotify_group *group)
  */
 void fsnotify_clear_marks_by_inode(struct inode *inode)
 {
-	struct fsnotify_mark_entry *entry, *lentry;
+	struct fsnotify_mark *mark, *lmark;
 	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);
+	hlist_for_each_entry_safe(mark, pos, n, &inode->i_fsnotify_marks, i.i_list) {
+		list_add(&mark->i.free_i_list, &free_list);
+		hlist_del_init_rcu(&mark->i.i_list);
+		fsnotify_get_mark(mark);
 	}
 	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);
+	list_for_each_entry_safe(mark, lmark, &free_list, i.free_i_list) {
+		fsnotify_destroy_mark(mark);
+		fsnotify_put_mark(mark);
 	}
 }
 
+/*
+ * Given a group clear all of the inode marks associated with that group.
+ */
+void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group)
+{
+	fsnotify_clear_marks_by_group_flags(group, FSNOTIFY_MARK_FLAG_INODE);
+}
+
 /*
  * 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 *fsnotify_find_inode_mark_locked(struct fsnotify_group *group,
+						      struct inode *inode)
 {
-	struct fsnotify_mark_entry *entry;
+	struct fsnotify_mark *mark;
 	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;
+	hlist_for_each_entry(mark, pos, &inode->i_fsnotify_marks, i.i_list) {
+		if (mark->group == group) {
+			fsnotify_get_mark(mark);
+			return mark;
 		}
 	}
 	return NULL;
 }
 
 /*
- * Nothing fancy, just initialize lists and locks and counters.
+ * 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
  */
-void fsnotify_init_mark(struct fsnotify_mark_entry *entry,
-			void (*free_mark)(struct fsnotify_mark_entry *entry))
+struct fsnotify_mark *fsnotify_find_inode_mark(struct fsnotify_group *group,
+					       struct inode *inode)
+{
+	struct fsnotify_mark *mark;
+
+	spin_lock(&inode->i_lock);
+	mark = fsnotify_find_inode_mark_locked(group, inode);
+	spin_unlock(&inode->i_lock);
 
+	return mark;
+}
+
+/*
+ * If we are setting a mark mask on an inode mark we should pin the inode
+ * in memory.
+ */
+void fsnotify_set_inode_mark_mask_locked(struct fsnotify_mark *mark,
+					 __u32 mask)
 {
-	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;
+	struct inode *inode;
+
+	assert_spin_locked(&mark->lock);
+
+	if (mask &&
+	    mark->i.inode &&
+	    !(mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED)) {
+		mark->flags |= FSNOTIFY_MARK_FLAG_OBJECT_PINNED;
+		inode = igrab(mark->i.inode);
+		/*
+		 * we shouldn't be able to get here if the inode wasn't
+		 * already safely held in memory.  But bug in case it
+		 * ever is wrong.
+		 */
+		BUG_ON(!inode);
+	}
 }
 
 /*
- * Attach an initialized mark entry to a given group and inode.
+ * Attach an initialized mark to a given 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.
+ * event types should be delivered to which group and for which inodes.  These
+ * marks are ordered according to the group's location in memory.
  */
-int fsnotify_add_mark(struct fsnotify_mark_entry *entry,
-		      struct fsnotify_group *group, struct inode *inode)
+int fsnotify_add_inode_mark(struct fsnotify_mark *mark,
+			    struct fsnotify_group *group, struct inode *inode,
+			    int allow_dups)
 {
-	struct fsnotify_mark_entry *lentry;
+	struct fsnotify_mark *lmark;
+	struct hlist_node *node, *last = NULL;
 	int ret = 0;
 
-	inode = igrab(inode);
-	if (unlikely(!inode))
-		return -EINVAL;
+	mark->flags |= FSNOTIFY_MARK_FLAG_INODE;
+
+	assert_spin_locked(&mark->lock);
+	assert_spin_locked(&group->mark_lock);
 
-	/*
-	 * LOCKING ORDER!!!!
-	 * entry->lock
-	 * group->mark_lock
-	 * inode->i_lock
-	 */
-	spin_lock(&entry->lock);
-	spin_lock(&group->mark_lock);
 	spin_lock(&inode->i_lock);
 
-	lentry = fsnotify_find_mark_entry(group, inode);
-	if (!lentry) {
-		entry->group = group;
-		entry->inode = inode;
+	mark->i.inode = inode;
 
-		hlist_add_head(&entry->i_list, &inode->i_fsnotify_mark_entries);
-		list_add(&entry->g_list, &group->mark_entries);
+	/* is mark the first mark? */
+	if (hlist_empty(&inode->i_fsnotify_marks)) {
+		hlist_add_head_rcu(&mark->i.i_list, &inode->i_fsnotify_marks);
+		goto out;
+	}
 
-		fsnotify_get_mark(entry); /* for i_list and g_list */
+	/* should mark be in the middle of the current list? */
+	hlist_for_each_entry(lmark, node, &inode->i_fsnotify_marks, i.i_list) {
+		last = node;
+
+		if ((lmark->group == group) && !allow_dups) {
+			ret = -EEXIST;
+			goto out;
+		}
 
-		atomic_inc(&group->num_marks);
+		if (mark->group < lmark->group)
+			continue;
 
-		fsnotify_recalc_inode_mask_locked(inode);
+		hlist_add_before_rcu(&mark->i.i_list, &lmark->i.i_list);
+		goto out;
 	}
 
+	BUG_ON(last == NULL);
+	/* mark should be the last entry.  last is the current last entry */
+	hlist_add_after_rcu(last, &mark->i.i_list);
+out:
+	fsnotify_recalc_inode_mask_locked(inode);
 	spin_unlock(&inode->i_lock);
-	spin_unlock(&group->mark_lock);
-	spin_unlock(&entry->lock);
-
-	if (lentry) {
-		ret = -EEXIST;
-		iput(inode);
-		fsnotify_put_mark(lentry);
-	} else {
-		__fsnotify_update_child_dentry_flags(inode);
-	}
 
 	return ret;
 }

fs/notify/inotify/Kconfig

-config INOTIFY
-	bool "Inotify file change notification support"
-	default n
-	---help---
-	  Say Y here to enable legacy in kernel inotify support.  Inotify is a
-	  file change notification system.  It is a replacement for dnotify.
-	  This option only provides the legacy inotify in kernel API.  There
-	  are no in tree kernel users of this interface since it is deprecated.
-	  You only need this if you are loading an out of tree kernel module
-	  that uses inotify.
-
-	  For more information, see <file:Documentation/filesystems/inotify.txt>
-
-	  If unsure, say N.
-
 config INOTIFY_USER
 	bool "Inotify support for userspace"
 	select ANON_INODES

fs/notify/inotify/Makefile

-obj-$(CONFIG_INOTIFY)		+= inotify.o
 obj-$(CONFIG_INOTIFY_USER)	+= inotify_fsnotify.o inotify_user.o

fs/notify/inotify/inotify.h

@@ -9,13 +9,12 @@ struct inotify_event_private_data {
 	int wd;
 };
 
-struct inotify_inode_mark_entry {
-	/* fsnotify_mark_entry MUST be the first thing */
-	struct fsnotify_mark_entry fsn_entry;
+struct inotify_inode_mark {
+	struct fsnotify_mark fsn_mark;
 	int wd;
 };
 
-extern void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
+extern void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
 					   struct fsnotify_group *group);
 extern void inotify_free_event_priv(struct fsnotify_event_private_data *event_priv);
 

fs/notify/inotify/inotify_fsnotify.c

@@ -22,6 +22,7 @@
  * General Public License for more details.
  */
 
+#include <linux/dcache.h> /* d_unlinked */
 #include <linux/fs.h> /* struct inode */
 #include <linux/fsnotify_backend.h>
 #include <linux/inotify.h>
@@ -32,26 +33,84 @@
 
 #include "inotify.h"
 
-static int inotify_handle_event(struct fsnotify_group *group, struct fsnotify_event *event)
+/*
+ * Check if 2 events contain the same information.  We do not compare private data
+ * but at this moment that isn't a problem for any know fsnotify listeners.
+ */
+static bool event_compare(struct fsnotify_event *old, struct fsnotify_event *new)
+{
+	if ((old->mask == new->mask) &&
+	    (old->to_tell == new->to_tell) &&
+	    (old->data_type == new->data_type) &&
+	    (old->name_len == new->name_len)) {
+		switch (old->data_type) {
+		case (FSNOTIFY_EVENT_INODE):
+			/* remember, after old was put on the wait_q we aren't
+			 * allowed to look at the inode any more, only thing
+			 * left to check was if the file_name is the same */
+			if (!old->name_len ||
+			    !strcmp(old->file_name, new->file_name))
+				return true;
+			break;
+		case (FSNOTIFY_EVENT_FILE):
+			if ((old->file->f_path.mnt == new->file->f_path.mnt) &&
+			    (old->file->f_path.dentry == new->file->f_path.dentry))
+				return true;
+			break;
+		case (FSNOTIFY_EVENT_NONE):
+			if (old->mask & FS_Q_OVERFLOW)
+				return true;
+			else if (old->mask & FS_IN_IGNORED)
+				return false;
+			return true;
+		};
+	}
+	return false;
+}
+
+static struct fsnotify_event *inotify_merge(struct list_head *list,
+					    struct fsnotify_event *event)
 {
-	struct fsnotify_mark_entry *entry;
-	struct inotify_inode_mark_entry *ientry;
+	struct fsnotify_event_holder *last_holder;
+	struct fsnotify_event *last_event;
+
+	/* and the list better be locked by something too */
+	spin_lock(&event->lock);
+
+	last_holder = list_entry(list->prev, struct fsnotify_event_holder, event_list);
+	last_event = last_holder->event;
+	if (event_compare(last_event, event))
+		fsnotify_get_event(last_event);
+	else
+		last_event = NULL;
+
+	spin_unlock(&event->lock);
+
+	return last_event;
+}
+
+static int inotify_handle_event(struct fsnotify_group *group,
+				struct fsnotify_mark *inode_mark,
+				struct fsnotify_mark *vfsmount_mark,
+				struct fsnotify_event *event)
+{
+	struct inotify_inode_mark *i_mark;
 	struct inode *to_tell;
 	struct inotify_event_private_data *event_priv;
 	struct fsnotify_event_private_data *fsn_event_priv;
-	int wd, ret;
+	struct fsnotify_event *added_event;
+	int wd, ret = 0;
+
+	BUG_ON(vfsmount_mark);
+
+	pr_debug("%s: group=%p event=%p to_tell=%p mask=%x\n", __func__, group,
+		 event, event->to_tell, event->mask);
 
 	to_tell = event->to_tell;
 
-	spin_lock(&to_tell->i_lock);
-	entry = fsnotify_find_mark_entry(group, to_tell);
-	spin_unlock(&to_tell->i_lock);
-	/* race with watch removal?  We already passes should_send */
-	if (unlikely(!entry))
-		return 0;
-	ientry = container_of(entry, struct inotify_inode_mark_entry,
-			      fsn_entry);
-	wd = ientry->wd;
+	i_mark = container_of(inode_mark, struct inotify_inode_mark,
+			      fsn_mark);
+	wd = i_mark->wd;
 
 	event_priv = kmem_cache_alloc(event_priv_cachep, GFP_KERNEL);
 	if (unlikely(!event_priv))
@@ -62,48 +121,40 @@ static int inotify_handle_event(struct fsnotify_group *group, struct fsnotify_ev
 	fsn_event_priv->group = group;
 	event_priv->wd = wd;
 
-	ret = fsnotify_add_notify_event(group, event, fsn_event_priv);
-	if (ret) {
+	added_event = fsnotify_add_notify_event(group, event, fsn_event_priv, inotify_merge);
+	if (added_event) {
 		inotify_free_event_priv(fsn_event_priv);
-		/* EEXIST says we tail matched, EOVERFLOW isn't something
-		 * to report up the stack. */
-		if ((ret == -EEXIST) ||
-		    (ret == -EOVERFLOW))
-			ret = 0;
+		if (!IS_ERR(added_event))
+			fsnotify_put_event(added_event);
+		else
+			ret = PTR_ERR(added_event);
 	}
 
-	/*
-	 * If we hold the entry until after the event is on the queue
-	 * IN_IGNORED won't be able to pass this event in the queue
-	 */
-	fsnotify_put_mark(entry);
+	if (inode_mark->mask & IN_ONESHOT)
+		fsnotify_destroy_mark(inode_mark);
 
 	return ret;
 }
 
-static void inotify_freeing_mark(struct fsnotify_mark_entry *entry, struct fsnotify_group *group)
+static void inotify_freeing_mark(struct fsnotify_mark *fsn_mark, struct fsnotify_group *group)
 {
-	inotify_ignored_and_remove_idr(entry, group);
+	inotify_ignored_and_remove_idr(fsn_mark, group);
 }
 
-static bool inotify_should_send_event(struct fsnotify_group *group, struct inode *inode, __u32 mask)
+static bool inotify_should_send_event(struct fsnotify_group *group, struct inode *inode,
+				      struct fsnotify_mark *inode_mark,
+				      struct fsnotify_mark *vfsmount_mark,
+				      __u32 mask, void *data, int data_type)
 {
-	struct fsnotify_mark_entry *entry;
-	bool send;
-
-	spin_lock(&inode->i_lock);
-	entry = fsnotify_find_mark_entry(group, inode);
-	spin_unlock(&inode->i_lock);
-	if (!entry)
-		return false;
+	if ((inode_mark->mask & FS_EXCL_UNLINK) &&
+	    (data_type == FSNOTIFY_EVENT_FILE)) {
+		struct file *file  = data;
 
-	mask = (mask & ~FS_EVENT_ON_CHILD);
-	send = (entry->mask & mask);
-
-	/* find took a reference */
-	fsnotify_put_mark(entry);
+		if (d_unlinked(file->f_path.dentry))
+			return false;
+	}
 
-	return send;
+	return true;
 }
 
 /*
@@ -115,18 +166,18 @@ static bool inotify_should_send_event(struct fsnotify_group *group, struct inode
  */
 static int idr_callback(int id, void *p, void *data)
 {
-	struct fsnotify_mark_entry *entry;
-	struct inotify_inode_mark_entry *ientry;
+	struct fsnotify_mark *fsn_mark;
+	struct inotify_inode_mark *i_mark;
 	static bool warned = false;
 
 	if (warned)
 		return 0;
 
 	warned = true;
-	entry = p;
-	ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
+	fsn_mark = p;
+	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 
-	WARN(1, "inotify closing but id=%d for entry=%p in group=%p still in "
+	WARN(1, "inotify closing but id=%d for fsn_mark=%p in group=%p still in "
 		"idr.  Probably leaking memory\n", id, p, data);
 
 	/*
@@ -135,9 +186,9 @@ static int idr_callback(int id, void *p, void *data)
 	 * out why we got here and the panic is no worse than the original
 	 * BUG() that was here.
 	 */
-	if (entry)
-		printk(KERN_WARNING "entry->group=%p inode=%p wd=%d\n",
-			entry->group, entry->inode, ientry->wd);
+	if (fsn_mark)
+		printk(KERN_WARNING "fsn_mark->group=%p inode=%p wd=%d\n",
+			fsn_mark->group, fsn_mark->i.inode, i_mark->wd);
 	return 0;
 }
 

fs/notify/inotify/inotify_user.c

@@ -46,17 +46,11 @@
 /* these are configurable via /proc/sys/fs/inotify/ */
 static int inotify_max_user_instances __read_mostly;
 static int inotify_max_queued_events __read_mostly;
-int inotify_max_user_watches __read_mostly;
+static int inotify_max_user_watches __read_mostly;
 
 static struct kmem_cache *inotify_inode_mark_cachep __read_mostly;
 struct kmem_cache *event_priv_cachep __read_mostly;
 
-/*
- * When inotify registers a new group it increments this and uses that
- * value as an offset to set the fsnotify group "name" and priority.
- */
-static atomic_t inotify_grp_num;
-
 #ifdef CONFIG_SYSCTL
 
 #include <linux/sysctl.h>
@@ -96,11 +90,14 @@ static inline __u32 inotify_arg_to_mask(u32 arg)
 {
 	__u32 mask;
 
-	/* everything should accept their own ignored and cares about children */
-	mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD);
+	/*
+	 * everything should accept their own ignored, cares about children,
+	 * and should receive events when the inode is unmounted
+	 */
+	mask = (FS_IN_IGNORED | FS_EVENT_ON_CHILD | FS_UNMOUNT);
 
 	/* mask off the flags used to open the fd */
-	mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT));
+	mask |= (arg & (IN_ALL_EVENTS | IN_ONESHOT | IN_EXCL_UNLINK));
 
 	return mask;
 }
@@ -144,6 +141,8 @@ static struct fsnotify_event *get_one_event(struct fsnotify_group *group,
 
 	event = fsnotify_peek_notify_event(group);
 
+	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
+
 	if (event->name_len)
 		event_size += roundup(event->name_len + 1, event_size);
 
@@ -173,6 +172,8 @@ static ssize_t copy_event_to_user(struct fsnotify_group *group,
 	size_t event_size = sizeof(struct inotify_event);
 	size_t name_len = 0;
 
+	pr_debug("%s: group=%p event=%p\n", __func__, group, event);
+
 	/* we get the inotify watch descriptor from the event private data */
 	spin_lock(&event->lock);
 	fsn_priv = fsnotify_remove_priv_from_event(group, event);
@@ -245,6 +246,8 @@ static ssize_t inotify_read(struct file *file, char __user *buf,
 		kevent = get_one_event(group, count);
 		mutex_unlock(&group->notification_mutex);
 
+		pr_debug("%s: group=%p kevent=%p\n", __func__, group, kevent);
+
 		if (kevent) {
 			ret = PTR_ERR(kevent);
 			if (IS_ERR(kevent))
@@ -289,6 +292,8 @@ static int inotify_release(struct inode *ignored, struct file *file)
 	struct fsnotify_group *group = file->private_data;
 	struct user_struct *user = group->inotify_data.user;
 
+	pr_debug("%s: group=%p\n", __func__, group);
+
 	fsnotify_clear_marks_by_group(group);
 
 	/* free this group, matching get was inotify_init->fsnotify_obtain_group */
@@ -312,6 +317,8 @@ static long inotify_ioctl(struct file *file, unsigned int cmd,
 	group = file->private_data;
 	p = (void __user *) arg;
 
+	pr_debug("%s: group=%p cmd=%u\n", __func__, group, cmd);
+
 	switch (cmd) {
 	case FIONREAD:
 		mutex_lock(&group->notification_mutex);
@@ -357,59 +364,159 @@ static int inotify_find_inode(const char __user *dirname, struct path *path, uns
 	return error;
 }
 
+static int inotify_add_to_idr(struct idr *idr, spinlock_t *idr_lock,
+			      int *last_wd,
+			      struct inotify_inode_mark *i_mark)
+{
+	int ret;
+
+	do {
+		if (unlikely(!idr_pre_get(idr, GFP_KERNEL)))
+			return -ENOMEM;
+
+		spin_lock(idr_lock);
+		ret = idr_get_new_above(idr, i_mark, *last_wd + 1,
+					&i_mark->wd);
+		/* we added the mark to the idr, take a reference */
+		if (!ret) {
+			*last_wd = i_mark->wd;
+			fsnotify_get_mark(&i_mark->fsn_mark);
+		}
+		spin_unlock(idr_lock);
+	} while (ret == -EAGAIN);
+
+	return ret;
+}
+
+static struct inotify_inode_mark *inotify_idr_find_locked(struct fsnotify_group *group,
+								int wd)
+{
+	struct idr *idr = &group->inotify_data.idr;
+	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
+	struct inotify_inode_mark *i_mark;
+
+	assert_spin_locked(idr_lock);
+
+	i_mark = idr_find(idr, wd);
+	if (i_mark) {
+		struct fsnotify_mark *fsn_mark = &i_mark->fsn_mark;
+
+		fsnotify_get_mark(fsn_mark);
+		/* One ref for being in the idr, one ref we just took */
+		BUG_ON(atomic_read(&fsn_mark->refcnt) < 2);
+	}
+
+	return i_mark;
+}
+
+static struct inotify_inode_mark *inotify_idr_find(struct fsnotify_group *group,
+							 int wd)
+{
+	struct inotify_inode_mark *i_mark;
+	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
+
+	spin_lock(idr_lock);
+	i_mark = inotify_idr_find_locked(group, wd);
+	spin_unlock(idr_lock);
+
+	return i_mark;
+}
+
+static void do_inotify_remove_from_idr(struct fsnotify_group *group,
+				       struct inotify_inode_mark *i_mark)
+{
+	struct idr *idr = &group->inotify_data.idr;
+	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
+	int wd = i_mark->wd;
+
+	assert_spin_locked(idr_lock);
+
+	idr_remove(idr, wd);
+
+	/* removed from the idr, drop that ref */
+	fsnotify_put_mark(&i_mark->fsn_mark);
+}
+
 /*
  * Remove the mark from the idr (if present) and drop the reference
  * on the mark because it was in the idr.
  */
 static void inotify_remove_from_idr(struct fsnotify_group *group,
-				    struct inotify_inode_mark_entry *ientry)
+				    struct inotify_inode_mark *i_mark)
 {
-	struct idr *idr;
-	struct fsnotify_mark_entry *entry;
-	struct inotify_inode_mark_entry *found_ientry;
+	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
+	struct inotify_inode_mark *found_i_mark = NULL;
 	int wd;
 
-	spin_lock(&group->inotify_data.idr_lock);
-	idr = &group->inotify_data.idr;
-	wd = ientry->wd;
+	spin_lock(idr_lock);
+	wd = i_mark->wd;
 
-	if (wd == -1)
+	/*
+	 * does this i_mark think it is in the idr?  we shouldn't get called
+	 * if it wasn't....
+	 */
+	if (wd == -1) {
+		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
+			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
+			i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
 		goto out;
+	}
 
-	entry = idr_find(&group->inotify_data.idr, wd);
-	if (unlikely(!entry))
+	/* Lets look in the idr to see if we find it */
+	found_i_mark = inotify_idr_find_locked(group, wd);
+	if (unlikely(!found_i_mark)) {
+		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
+			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
+			i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
 		goto out;
+	}
 
-	found_ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
-	if (unlikely(found_ientry != ientry)) {
-		/* We found an entry in the idr with the right wd, but it's
-		 * not the entry we were told to remove.  eparis seriously
-		 * fucked up somewhere. */
-		WARN_ON(1);
-		ientry->wd = -1;
+	/*
+	 * We found an mark in the idr at the right wd, but it's
+	 * not the mark we were told to remove.  eparis seriously
+	 * fucked up somewhere.
+	 */
+	if (unlikely(found_i_mark != i_mark)) {
+		WARN_ONCE(1, "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p "
+			"mark->inode=%p found_i_mark=%p found_i_mark->wd=%d "
+			"found_i_mark->group=%p found_i_mark->inode=%p\n",
+			__func__, i_mark, i_mark->wd, i_mark->fsn_mark.group,
+			i_mark->fsn_mark.i.inode, found_i_mark, found_i_mark->wd,
+			found_i_mark->fsn_mark.group,
+			found_i_mark->fsn_mark.i.inode);
 		goto out;
 	}
 
-	/* One ref for being in the idr, one ref held by the caller */
-	BUG_ON(atomic_read(&entry->refcnt) < 2);
-
-	idr_remove(idr, wd);
-	ientry->wd = -1;
+	/*
+	 * One ref for being in the idr
+	 * one ref held by the caller trying to kill us
+	 * one ref grabbed by inotify_idr_find
+	 */
+	if (unlikely(atomic_read(&i_mark->fsn_mark.refcnt) < 3)) {
+		printk(KERN_ERR "%s: i_mark=%p i_mark->wd=%d i_mark->group=%p"
+			" i_mark->inode=%p\n", __func__, i_mark, i_mark->wd,
+			i_mark->fsn_mark.group, i_mark->fsn_mark.i.inode);
+		/* we can't really recover with bad ref cnting.. */
+		BUG();
+	}
 
-	/* removed from the idr, drop that ref */
-	fsnotify_put_mark(entry);
+	do_inotify_remove_from_idr(group, i_mark);
 out:
-	spin_unlock(&group->inotify_data.idr_lock);
+	/* match the ref taken by inotify_idr_find_locked() */
+	if (found_i_mark)
+		fsnotify_put_mark(&found_i_mark->fsn_mark);
+	i_mark->wd = -1;
+	spin_unlock(idr_lock);
 }
 
 /*
  * Send IN_IGNORED for this wd, remove this wd from the idr.
  */
-void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
+void inotify_ignored_and_remove_idr(struct fsnotify_mark *fsn_mark,
 				    struct fsnotify_group *group)
 {
-	struct inotify_inode_mark_entry *ientry;
-	struct fsnotify_event *ignored_event;
+	struct inotify_inode_mark *i_mark;
+	struct fsnotify_event *ignored_event, *notify_event;
 	struct inotify_event_private_data *event_priv;
 	struct fsnotify_event_private_data *fsn_event_priv;
 	int ret;
@@ -420,7 +527,7 @@ void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
 	if (!ignored_event)
 		return;
 
-	ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
+	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 
 	event_priv = kmem_cache_alloc(event_priv_cachep, GFP_NOFS);
 	if (unlikely(!event_priv))
@@ -429,37 +536,44 @@ void inotify_ignored_and_remove_idr(struct fsnotify_mark_entry *entry,
 	fsn_event_priv = &event_priv->fsnotify_event_priv_data;
 
 	fsn_event_priv->group = group;
-	event_priv->wd = ientry->wd;
-
-	ret = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv);
-	if (ret)
+	event_priv->wd = i_mark->wd;
+
+	notify_event = fsnotify_add_notify_event(group, ignored_event, fsn_event_priv, NULL);
+	if (notify_event) {
+		if (IS_ERR(notify_event))
+			ret = PTR_ERR(notify_event);
+		else
+			fsnotify_put_event(notify_event);
 		inotify_free_event_priv(fsn_event_priv);
+	}
 
 skip_send_ignore:
 
 	/* matches the reference taken when the event was created */
 	fsnotify_put_event(ignored_event);
 
-	/* remove this entry from the idr */
-	inotify_remove_from_idr(group, ientry);
+	/* remove this mark from the idr */
+	inotify_remove_from_idr(group, i_mark);
 
 	atomic_dec(&group->inotify_data.user->inotify_watches);
 }
 
 /* ding dong the mark is dead */
-static void inotify_free_mark(struct fsnotify_mark_entry *entry)
+static void inotify_free_mark(struct fsnotify_mark *fsn_mark)
 {
-	struct inotify_inode_mark_entry *ientry = (struct inotify_inode_mark_entry *)entry;
+	struct inotify_inode_mark *i_mark;
+
+	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 
-	kmem_cache_free(inotify_inode_mark_cachep, ientry);
+	kmem_cache_free(inotify_inode_mark_cachep, i_mark);
 }
 
 static int inotify_update_existing_watch(struct fsnotify_group *group,
 					 struct inode *inode,
 					 u32 arg)
 {
-	struct fsnotify_mark_entry *entry;
-	struct inotify_inode_mark_entry *ientry;
+	struct fsnotify_mark *fsn_mark;
+	struct inotify_inode_mark *i_mark;
 	__u32 old_mask, new_mask;
 	__u32 mask;
 	int add = (arg & IN_MASK_ADD);
@@ -467,52 +581,43 @@ static int inotify_update_existing_watch(struct fsnotify_group *group,
 
 	/* don't allow invalid bits: we don't want flags set */
 	mask = inotify_arg_to_mask(arg);
-	if (unlikely(!mask))
+	if (unlikely(!(mask & IN_ALL_EVENTS)))
 		return -EINVAL;
 
-	spin_lock(&inode->i_lock);
-	entry = fsnotify_find_mark_entry(group, inode);
-	spin_unlock(&inode->i_lock);
-	if (!entry)
+	fsn_mark = fsnotify_find_inode_mark(group, inode);
+	if (!fsn_mark)
 		return -ENOENT;
 
-	ientry = container_of(entry, struct inotify_inode_mark_entry, fsn_entry);
+	i_mark = container_of(fsn_mark, struct inotify_inode_mark, fsn_mark);
 
-	spin_lock(&entry->lock);
+	spin_lock(&fsn_mark->lock);
 
-	old_mask = entry->mask;
-	if (add) {
-		entry->mask |= mask;
-		new_mask = entry->mask;
-	} else {
-		entry->mask = mask;
-		new_mask = entry->mask;
-	}
+	old_mask = fsn_mark->mask;
+	if (add)
+		fsnotify_set_mark_mask_locked(fsn_mark, (fsn_mark->mask | mask));
+	else
+		fsnotify_set_mark_mask_locked(fsn_mark, mask);
+	new_mask = fsn_mark->mask;
 
-	spin_unlock(&entry->lock);
+	spin_unlock(&fsn_mark->lock);
 
 	if (old_mask != new_mask) {
 		/* more bits in old than in new? */
 		int dropped = (old_mask & ~new_mask);
-		/* more bits in this entry than the inode's mask? */
+		/* more bits in this fsn_mark than the inode's mask? */
 		int do_inode = (new_mask & ~inode->i_fsnotify_mask);
-		/* more bits in this entry than the group? */
-		int do_group = (new_mask & ~group->mask);
 
-		/* update the inode with this new entry */
+		/* update the inode with this new fsn_mark */
 		if (dropped || do_inode)
 			fsnotify_recalc_inode_mask(inode);
 
-		/* update the group mask with the new mask */
-		if (dropped || do_group)
-			fsnotify_recalc_group_mask(group);
 	}
 
 	/* return the wd */
-	ret = ientry->wd;
+	ret = i_mark->wd;
 
-	/* match the get from fsnotify_find_mark_entry() */
-	fsnotify_put_mark(entry);
+	/* match the get from fsnotify_find_mark() */
+	fsnotify_put_mark(fsn_mark);
 
 	return ret;
 }
@@ -521,73 +626,51 @@ static int inotify_new_watch(struct fsnotify_group *group,
 			     struct inode *inode,
 			     u32 arg)
 {
-	struct inotify_inode_mark_entry *tmp_ientry;
+	struct inotify_inode_mark *tmp_i_mark;
 	__u32 mask;
 	int ret;
+	struct idr *idr = &group->inotify_data.idr;
+	spinlock_t *idr_lock = &group->inotify_data.idr_lock;
 
 	/* don't allow invalid bits: we don't want flags set */
 	mask = inotify_arg_to_mask(arg);
-	if (unlikely(!mask))
+	if (unlikely(!(mask & IN_ALL_EVENTS)))
 		return -EINVAL;
 
-	tmp_ientry = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
-	if (unlikely(!tmp_ientry))
+	tmp_i_mark = kmem_cache_alloc(inotify_inode_mark_cachep, GFP_KERNEL);
+	if (unlikely(!tmp_i_mark))
 		return -ENOMEM;
 
-	fsnotify_init_mark(&tmp_ientry->fsn_entry, inotify_free_mark);
-	tmp_ientry->fsn_entry.mask = mask;
-	tmp_ientry->wd = -1;
+	fsnotify_init_mark(&tmp_i_mark->fsn_mark, inotify_free_mark);
+	tmp_i_mark->fsn_mark.mask = mask;
+	tmp_i_mark->wd = -1;
 
 	ret = -ENOSPC;
 	if (atomic_read(&group->inotify_data.user->inotify_watches) >= inotify_max_user_watches)
 		goto out_err;
-retry:
-	ret = -ENOMEM;
-	if (unlikely(!idr_pre_get(&group->inotify_data.idr, GFP_KERNEL)))
-		goto out_err;
 
-	/* we are putting the mark on the idr, take a reference */
-	fsnotify_get_mark(&tmp_ientry->fsn_entry);
-
-	spin_lock(&group->inotify_data.idr_lock);
-	ret = idr_get_new_above(&group->inotify_data.idr, &tmp_ientry->fsn_entry,
-				group->inotify_data.last_wd+1,
-				&tmp_ientry->wd);
-	spin_unlock(&group->inotify_data.idr_lock);
-	if (ret) {
-		/* we didn't get on the idr, drop the idr reference */
-		fsnotify_put_mark(&tmp_ientry->fsn_entry);
-
-		/* idr was out of memory allocate and try again */
-		if (ret == -EAGAIN)
-			goto retry;
+	ret = inotify_add_to_idr(idr, idr_lock, &group->inotify_data.last_wd,
+				 tmp_i_mark);
+	if (ret)
 		goto out_err;
-	}
 
 	/* we are on the idr, now get on the inode */
-	ret = fsnotify_add_mark(&tmp_ientry->fsn_entry, group, inode);
+	ret = fsnotify_add_mark(&tmp_i_mark->fsn_mark, group, inode, NULL, 0);
 	if (ret) {
 		/* we failed to get on the inode, get off the idr */
-		inotify_remove_from_idr(group, tmp_ientry);
+		inotify_remove_from_idr(group, tmp_i_mark);
 		goto out_err;
 	}
 
-	/* update the idr hint, who cares about races, it's just a hint */
-	group->inotify_data.last_wd = tmp_ientry->wd;
-
 	/* increment the number of watches the user has */
 	atomic_inc(&group->inotify_data.user->inotify_watches);
 
-	/* return the watch descriptor for this new entry */
-	ret = tmp_ientry->wd;
-
-	/* if this mark added a new event update the group mask */
-	if (mask & ~group->mask)
-		fsnotify_recalc_group_mask(group);
+	/* return the watch descriptor for this new mark */
+	ret = tmp_i_mark->wd;
 
 out_err:
-	/* match the ref from fsnotify_init_markentry() */
-	fsnotify_put_mark(&tmp_ientry->fsn_entry);
+	/* match the ref from fsnotify_init_mark() */
+	fsnotify_put_mark(&tmp_i_mark->fsn_mark);
 
 	return ret;
 }
@@ -616,11 +699,8 @@ static int inotify_update_watch(struct fsnotify_group *group, struct inode *inod
 static struct fsnotify_group *inotify_new_group(struct user_struct *user, unsigned int max_events)
 {
 	struct fsnotify_group *group;
-	unsigned int grp_num;
 
-	/* fsnotify_obtain_group took a reference to group, we put this when we kill the file in the end */
-	grp_num = (INOTIFY_GROUP_NUM - atomic_inc_return(&inotify_grp_num));
-	group = fsnotify_obtain_group(grp_num, 0, &inotify_fsnotify_ops);
+	group = fsnotify_alloc_group(&inotify_fsnotify_ops);
 	if (IS_ERR(group))
 		return group;
 
@@ -726,7 +806,7 @@ SYSCALL_DEFINE3(inotify_add_watch, int, fd, const char __user *, pathname,
 SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
 {
 	struct fsnotify_group *group;
-	struct fsnotify_mark_entry *entry;
+	struct inotify_inode_mark *i_mark;
 	struct file *filp;
 	int ret = 0, fput_needed;
 
@@ -735,25 +815,23 @@ SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
 		return -EBADF;
 
 	/* verify that this is indeed an inotify instance */
-	if (unlikely(filp->f_op != &inotify_fops)) {
-		ret = -EINVAL;
+	ret = -EINVAL;
+	if (unlikely(filp->f_op != &inotify_fops))
 		goto out;
-	}
 
 	group = filp->private_data;
 
-	spin_lock(&group->inotify_data.idr_lock);
-	entry = idr_find(&group->inotify_data.idr, wd);
-	if (unlikely(!entry)) {
-		spin_unlock(&group->inotify_data.idr_lock);
-		ret = -EINVAL;
+	ret = -EINVAL;
+	i_mark = inotify_idr_find(group, wd);
+	if (unlikely(!i_mark))
 		goto out;
-	}
-	fsnotify_get_mark(entry);
-	spin_unlock(&group->inotify_data.idr_lock);
 
-	fsnotify_destroy_mark_by_entry(entry);
-	fsnotify_put_mark(entry);
+	ret = 0;
+
+	fsnotify_destroy_mark(&i_mark->fsn_mark);
+
+	/* match ref taken by inotify_idr_find */
+	fsnotify_put_mark(&i_mark->fsn_mark);
 
 out:
 	fput_light(filp, fput_needed);
@@ -767,7 +845,28 @@ SYSCALL_DEFINE2(inotify_rm_watch, int, fd, __s32, wd)
  */
 static int __init inotify_user_setup(void)
 {
-	inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark_entry, SLAB_PANIC);
+	BUILD_BUG_ON(IN_ACCESS != FS_ACCESS);
+	BUILD_BUG_ON(IN_MODIFY != FS_MODIFY);
+	BUILD_BUG_ON(IN_ATTRIB != FS_ATTRIB);
+	BUILD_BUG_ON(IN_CLOSE_WRITE != FS_CLOSE_WRITE);
+	BUILD_BUG_ON(IN_CLOSE_NOWRITE != FS_CLOSE_NOWRITE);
+	BUILD_BUG_ON(IN_OPEN != FS_OPEN);
+	BUILD_BUG_ON(IN_MOVED_FROM != FS_MOVED_FROM);
+	BUILD_BUG_ON(IN_MOVED_TO != FS_MOVED_TO);
+	BUILD_BUG_ON(IN_CREATE != FS_CREATE);
+	BUILD_BUG_ON(IN_DELETE != FS_DELETE);
+	BUILD_BUG_ON(IN_DELETE_SELF != FS_DELETE_SELF);
+	BUILD_BUG_ON(IN_MOVE_SELF != FS_MOVE_SELF);
+	BUILD_BUG_ON(IN_UNMOUNT != FS_UNMOUNT);
+	BUILD_BUG_ON(IN_Q_OVERFLOW != FS_Q_OVERFLOW);
+	BUILD_BUG_ON(IN_IGNORED != FS_IN_IGNORED);
+	BUILD_BUG_ON(IN_EXCL_UNLINK != FS_EXCL_UNLINK);
+	BUILD_BUG_ON(IN_ISDIR != FS_IN_ISDIR);
+	BUILD_BUG_ON(IN_ONESHOT != FS_IN_ONESHOT);
+
+	BUG_ON(hweight32(ALL_INOTIFY_BITS) != 21);
+
+	inotify_inode_mark_cachep = KMEM_CACHE(inotify_inode_mark, SLAB_PANIC);
 	event_priv_cachep = KMEM_CACHE(inotify_event_private_data, SLAB_PANIC);
 
 	inotify_max_queued_events = 16384;

fs/notify/mark.c

+/*
+ *  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:
+ *
+ * mark->lock
+ * group->mark_lock
+ * inode->i_lock
+ *
+ * mark->lock protects 2 things, mark->group and mark->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 marks_list anchored inside a given group
+ * and each mark 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_marks list anchored inside a
+ * given inode and each mark 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)
+ * - 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_marks 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/kthread.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/srcu.h>
+#include <linux/writeback.h> /* for inode_lock */
+
+#include <asm/atomic.h>
+
+#include <linux/fsnotify_backend.h>
+#include "fsnotify.h"
+
+struct srcu_struct fsnotify_mark_srcu;
+static DEFINE_SPINLOCK(destroy_lock);
+static LIST_HEAD(destroy_list);
+static DECLARE_WAIT_QUEUE_HEAD(destroy_waitq);
+
+void fsnotify_get_mark(struct fsnotify_mark *mark)
+{
+	atomic_inc(&mark->refcnt);
+}
+
+void fsnotify_put_mark(struct fsnotify_mark *mark)
+{
+	if (atomic_dec_and_test(&mark->refcnt))
+		mark->free_mark(mark);
+}
+
+/*
+ * 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 mark->lock
+ */
+void fsnotify_destroy_mark(struct fsnotify_mark *mark)
+{
+	struct fsnotify_group *group;
+	struct inode *inode = NULL;
+
+	spin_lock(&mark->lock);
+
+	group = mark->group;
+
+	/* something else already called this function on this mark */
+	if (!(mark->flags & FSNOTIFY_MARK_FLAG_ALIVE)) {
+		spin_unlock(&mark->lock);
+		return;
+	}
+
+	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
+
+	/* 1 from caller and 1 for being on i_list/g_list */
+	BUG_ON(atomic_read(&mark->refcnt) < 2);
+
+	spin_lock(&group->mark_lock);
+
+	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE) {
+		inode = mark->i.inode;
+		fsnotify_destroy_inode_mark(mark);
+	} else if (mark->flags & FSNOTIFY_MARK_FLAG_VFSMOUNT)
+		fsnotify_destroy_vfsmount_mark(mark);
+	else
+		BUG();
+
+	list_del_init(&mark->g_list);
+
+	spin_unlock(&group->mark_lock);
+	spin_unlock(&mark->lock);
+
+	spin_lock(&destroy_lock);
+	list_add(&mark->destroy_list, &destroy_list);
+	spin_unlock(&destroy_lock);
+	wake_up(&destroy_waitq);
+
+	/*
+	 * Some groups like to know that marks are being freed.  This is a
+	 * callback to the group function to let it know that this mark
+	 * is being freed.
+	 */
+	if (group->ops->freeing_mark)
+		group->ops->freeing_mark(mark, group);
+
+	/*
+	 * __fsnotify_update_child_dentry_flags(inode);
+	 *
+	 * I really want to call that, but we can't, we have no idea if the inode
+	 * still exists the second we drop the mark->lock.
+	 *
+	 * The next time an event arrive to this inode from one of it's children
+	 * __fsnotify_parent will see that the inode doesn't care about it's
+	 * children and will update all of these flags then.  So really this
+	 * is just a lazy update (and could be a perf win...)
+	 */
+
+	if (inode && (mark->flags & FSNOTIFY_MARK_FLAG_OBJECT_PINNED))
+		iput(inode);
+
+	/*
+	 * 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);
+}
+
+void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
+{
+	assert_spin_locked(&mark->lock);
+
+	mark->mask = mask;
+
+	if (mark->flags & FSNOTIFY_MARK_FLAG_INODE)
+		fsnotify_set_inode_mark_mask_locked(mark, mask);
+}
+
+void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask)
+{
+	assert_spin_locked(&mark->lock);
+
+	mark->ignored_mask = mask;
+}
+
+/*
+ * Attach an initialized mark to a given group and fs object.
+ * These marks may be used for the fsnotify backend to determine which
+ * event types should be delivered to which group.
+ */
+int fsnotify_add_mark(struct fsnotify_mark *mark,
+		      struct fsnotify_group *group, struct inode *inode,
+		      struct vfsmount *mnt, int allow_dups)
+{
+	int ret = 0;
+
+	BUG_ON(inode && mnt);
+	BUG_ON(!inode && !mnt);
+
+	/*
+	 * LOCKING ORDER!!!!
+	 * mark->lock
+	 * group->mark_lock
+	 * inode->i_lock
+	 */
+	spin_lock(&mark->lock);
+	spin_lock(&group->mark_lock);
+
+	mark->flags |= FSNOTIFY_MARK_FLAG_ALIVE;
+
+	mark->group = group;
+	list_add(&mark->g_list, &group->marks_list);
+	atomic_inc(&group->num_marks);
+	fsnotify_get_mark(mark); /* for i_list and g_list */
+
+	if (inode) {
+		ret = fsnotify_add_inode_mark(mark, group, inode, allow_dups);
+		if (ret)
+			goto err;
+	} else if (mnt) {
+		ret = fsnotify_add_vfsmount_mark(mark, group, mnt, allow_dups);
+		if (ret)
+			goto err;
+	} else {
+		BUG();
+	}
+
+	spin_unlock(&group->mark_lock);
+
+	/* this will pin the object if appropriate */
+	fsnotify_set_mark_mask_locked(mark, mark->mask);
+
+	spin_unlock(&mark->lock);
+
+	if (inode)
+		__fsnotify_update_child_dentry_flags(inode);
+
+	return ret;
+err:
+	mark->flags &= ~FSNOTIFY_MARK_FLAG_ALIVE;
+	list_del_init(&mark->g_list);
+	mark->group = NULL;
+	atomic_dec(&group->num_marks);
+
+	spin_unlock(&group->mark_lock);
+	spin_unlock(&mark->lock);
+
+	spin_lock(&destroy_lock);
+	list_add(&mark->destroy_list, &destroy_list);
+	spin_unlock(&destroy_lock);
+	wake_up(&destroy_waitq);
+
+	return ret;
+}
+
+/*
+ * clear any marks in a group in which mark->flags & flags is true
+ */
+void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group,
+					 unsigned int flags)
+{
+	struct fsnotify_mark *lmark, *mark;
+	LIST_HEAD(free_list);
+
+	spin_lock(&group->mark_lock);
+	list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
+		if (mark->flags & flags) {
+			list_add(&mark->free_g_list, &free_list);
+			list_del_init(&mark->g_list);
+			fsnotify_get_mark(mark);
+		}
+	}
+	spin_unlock(&group->mark_lock);
+
+	list_for_each_entry_safe(mark, lmark, &free_list, free_g_list) {
+		fsnotify_destroy_mark(mark);
+		fsnotify_put_mark(mark);
+	}
+}
+
+/*
+ * Given a group, destroy all of the marks associated with that group.
+ */
+void fsnotify_clear_marks_by_group(struct fsnotify_group *group)
+{
+	fsnotify_clear_marks_by_group_flags(group, (unsigned int)-1);
+}
+
+void fsnotify_duplicate_mark(struct fsnotify_mark *new, struct fsnotify_mark *old)
+{
+	assert_spin_locked(&old->lock);
+	new->i.inode = old->i.inode;
+	new->m.mnt = old->m.mnt;
+	new->group = old->group;
+	new->mask = old->mask;
+	new->free_mark = old->free_mark;
+}
+
+/*
+ * Nothing fancy, just initialize lists and locks and counters.
+ */
+void fsnotify_init_mark(struct fsnotify_mark *mark,
+			void (*free_mark)(struct fsnotify_mark *mark))
+{
+	memset(mark, 0, sizeof(*mark));
+	spin_lock_init(&mark->lock);
+	atomic_set(&mark->refcnt, 1);
+	mark->free_mark = free_mark;
+}
+
+static int fsnotify_mark_destroy(void *ignored)
+{
+	struct fsnotify_mark *mark, *next;
+	LIST_HEAD(private_destroy_list);
+
+	for (;;) {
+		spin_lock(&destroy_lock);
+		/* exchange the list head */
+		list_replace_init(&destroy_list, &private_destroy_list);
+		spin_unlock(&destroy_lock);
+
+		synchronize_srcu(&fsnotify_mark_srcu);
+
+		list_for_each_entry_safe(mark, next, &private_destroy_list, destroy_list) {
+			list_del_init(&mark->destroy_list);
+			fsnotify_put_mark(mark);
+		}
+
+		wait_event_interruptible(destroy_waitq, !list_empty(&destroy_list));
+	}
+
+	return 0;
+}
+
+static int __init fsnotify_mark_init(void)
+{
+	struct task_struct *thread;
+
+	thread = kthread_run(fsnotify_mark_destroy, NULL,
+			     "fsnotify_mark");
+	if (IS_ERR(thread))
+		panic("unable to start fsnotify mark destruction thread.");
+
+	return 0;
+}
+device_initcall(fsnotify_mark_init);

fs/notify/notification.c

@@ -31,6 +31,7 @@
  * allocated and used.
  */
 
+#include <linux/file.h>
 #include <linux/fs.h>
 #include <linux/init.h>
 #include <linux/kernel.h>
@@ -56,7 +57,7 @@ static struct kmem_cache *fsnotify_event_holder_cachep;
  * it is needed.  It's refcnt is set 1 at kernel init time and will never
  * get set to 0 so it will never get 'freed'
  */
-static struct fsnotify_event q_overflow_event;
+static struct fsnotify_event *q_overflow_event;
 static atomic_t fsnotify_sync_cookie = ATOMIC_INIT(0);
 
 /**
@@ -87,12 +88,15 @@ void fsnotify_put_event(struct fsnotify_event *event)
 		return;
 
 	if (atomic_dec_and_test(&event->refcnt)) {
-		if (event->data_type == FSNOTIFY_EVENT_PATH)
-			path_put(&event->path);
+		pr_debug("%s: event=%p\n", __func__, event);
+
+		if (event->data_type == FSNOTIFY_EVENT_FILE)
+			fput(event->file);
 
 		BUG_ON(!list_empty(&event->private_data_list));
 
 		kfree(event->file_name);
+		put_pid(event->tgid);
 		kmem_cache_free(fsnotify_event_cachep, event);
 	}
 }
@@ -104,7 +108,8 @@ struct fsnotify_event_holder *fsnotify_alloc_event_holder(void)
 
 void fsnotify_destroy_event_holder(struct fsnotify_event_holder *holder)
 {
-	kmem_cache_free(fsnotify_event_holder_cachep, holder);
+	if (holder)
+		kmem_cache_free(fsnotify_event_holder_cachep, holder);
 }
 
 /*
@@ -128,54 +133,21 @@ struct fsnotify_event_private_data *fsnotify_remove_priv_from_event(struct fsnot
 	return priv;
 }
 
-/*
- * Check if 2 events contain the same information.  We do not compare private data
- * but at this moment that isn't a problem for any know fsnotify listeners.
- */
-static bool event_compare(struct fsnotify_event *old, struct fsnotify_event *new)
-{
-	if ((old->mask == new->mask) &&
-	    (old->to_tell == new->to_tell) &&
-	    (old->data_type == new->data_type) &&
-	    (old->name_len == new->name_len)) {
-		switch (old->data_type) {
-		case (FSNOTIFY_EVENT_INODE):
-			/* remember, after old was put on the wait_q we aren't
-			 * allowed to look at the inode any more, only thing
-			 * left to check was if the file_name is the same */
-			if (!old->name_len ||
-			    !strcmp(old->file_name, new->file_name))
-				return true;
-			break;
-		case (FSNOTIFY_EVENT_PATH):
-			if ((old->path.mnt == new->path.mnt) &&
-			    (old->path.dentry == new->path.dentry))
-				return true;
-			break;
-		case (FSNOTIFY_EVENT_NONE):
-			if (old->mask & FS_Q_OVERFLOW)
-				return true;
-			else if (old->mask & FS_IN_IGNORED)
-				return false;
-			return false;
-		};
-	}
-	return false;
-}
-
 /*
  * Add an event to the group notification queue.  The group can later pull this
  * event off the queue to deal with.  If the event is successfully added to the
  * group's notification queue, a reference is taken on event.
  */
-int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_event *event,
-			      struct fsnotify_event_private_data *priv)
+struct fsnotify_event *fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_event *event,
+						 struct fsnotify_event_private_data *priv,
+						 struct fsnotify_event *(*merge)(struct list_head *,
+										 struct fsnotify_event *))
 {
+	struct fsnotify_event *return_event = NULL;
 	struct fsnotify_event_holder *holder = NULL;
 	struct list_head *list = &group->notification_list;
-	struct fsnotify_event_holder *last_holder;
-	struct fsnotify_event *last_event;
-	int ret = 0;
+
+	pr_debug("%s: group=%p event=%p priv=%p\n", __func__, group, event, priv);
 
 	/*
 	 * There is one fsnotify_event_holder embedded inside each fsnotify_event.
@@ -189,18 +161,40 @@ int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_even
 alloc_holder:
 		holder = fsnotify_alloc_event_holder();
 		if (!holder)
-			return -ENOMEM;
+			return ERR_PTR(-ENOMEM);
 	}
 
 	mutex_lock(&group->notification_mutex);
 
 	if (group->q_len >= group->max_events) {
-		event = &q_overflow_event;
-		ret = -EOVERFLOW;
+		event = q_overflow_event;
+
+		/*
+		 * we need to return the overflow event
+		 * which means we need a ref
+		 */
+		fsnotify_get_event(event);
+		return_event = event;
+
 		/* sorry, no private data on the overflow event */
 		priv = NULL;
 	}
 
+	if (!list_empty(list) && merge) {
+		struct fsnotify_event *tmp;
+
+		tmp = merge(list, event);
+		if (tmp) {
+			mutex_unlock(&group->notification_mutex);
+
+			if (return_event)
+				fsnotify_put_event(return_event);
+			if (holder != &event->holder)
+				fsnotify_destroy_event_holder(holder);
+			return tmp;
+		}
+	}
+
 	spin_lock(&event->lock);
 
 	if (list_empty(&event->holder.event_list)) {
@@ -212,19 +206,13 @@ int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_even
 		 * event holder was used, go back and get a new one */
 		spin_unlock(&event->lock);
 		mutex_unlock(&group->notification_mutex);
-		goto alloc_holder;
-	}
 
-	if (!list_empty(list)) {
-		last_holder = list_entry(list->prev, struct fsnotify_event_holder, event_list);
-		last_event = last_holder->event;
-		if (event_compare(last_event, event)) {
-			spin_unlock(&event->lock);
-			mutex_unlock(&group->notification_mutex);
-			if (holder != &event->holder)
-				fsnotify_destroy_event_holder(holder);
-			return -EEXIST;
+		if (return_event) {
+			fsnotify_put_event(return_event);
+			return_event = NULL;
 		}
+
+		goto alloc_holder;
 	}
 
 	group->q_len++;
@@ -238,7 +226,7 @@ int fsnotify_add_notify_event(struct fsnotify_group *group, struct fsnotify_even
 	mutex_unlock(&group->notification_mutex);
 
 	wake_up(&group->notification_waitq);
-	return ret;
+	return return_event;
 }
 
 /*
@@ -253,6 +241,8 @@ struct fsnotify_event *fsnotify_remove_notify_event(struct fsnotify_group *group
 
 	BUG_ON(!mutex_is_locked(&group->notification_mutex));
 
+	pr_debug("%s: group=%p\n", __func__, group);
+
 	holder = list_first_entry(&group->notification_list, struct fsnotify_event_holder, event_list);
 
 	event = holder->event;
@@ -314,25 +304,82 @@ void fsnotify_flush_notify(struct fsnotify_group *group)
 
 static void initialize_event(struct fsnotify_event *event)
 {
-	event->holder.event = NULL;
 	INIT_LIST_HEAD(&event->holder.event_list);
 	atomic_set(&event->refcnt, 1);
 
 	spin_lock_init(&event->lock);
 
-	event->path.dentry = NULL;
-	event->path.mnt = NULL;
-	event->inode = NULL;
-	event->data_type = FSNOTIFY_EVENT_NONE;
-
 	INIT_LIST_HEAD(&event->private_data_list);
+}
 
-	event->to_tell = NULL;
+/*
+ * Caller damn well better be holding whatever mutex is protecting the
+ * old_holder->event_list and the new_event must be a clean event which
+ * cannot be found anywhere else in the kernel.
+ */
+int fsnotify_replace_event(struct fsnotify_event_holder *old_holder,
+			   struct fsnotify_event *new_event)
+{
+	struct fsnotify_event *old_event = old_holder->event;
+	struct fsnotify_event_holder *new_holder = &new_event->holder;
 
-	event->file_name = NULL;
-	event->name_len = 0;
+	enum event_spinlock_class {
+		SPINLOCK_OLD,
+		SPINLOCK_NEW,
+	};
 
-	event->sync_cookie = 0;
+	pr_debug("%s: old_event=%p new_event=%p\n", __func__, old_event, new_event);
+
+	/*
+	 * if the new_event's embedded holder is in use someone
+	 * screwed up and didn't give us a clean new event.
+	 */
+	BUG_ON(!list_empty(&new_holder->event_list));
+
+	spin_lock_nested(&old_event->lock, SPINLOCK_OLD);
+	spin_lock_nested(&new_event->lock, SPINLOCK_NEW);
+
+	new_holder->event = new_event;
+	list_replace_init(&old_holder->event_list, &new_holder->event_list);
+
+	spin_unlock(&new_event->lock);
+	spin_unlock(&old_event->lock);
+
+	/* event == holder means we are referenced through the in event holder */
+	if (old_holder != &old_event->holder)
+		fsnotify_destroy_event_holder(old_holder);
+
+	fsnotify_get_event(new_event); /* on the list take reference */
+	fsnotify_put_event(old_event); /* off the list, drop reference */
+
+	return 0;
+}
+
+struct fsnotify_event *fsnotify_clone_event(struct fsnotify_event *old_event)
+{
+	struct fsnotify_event *event;
+
+	event = kmem_cache_alloc(fsnotify_event_cachep, GFP_KERNEL);
+	if (!event)
+		return NULL;
+
+	pr_debug("%s: old_event=%p new_event=%p\n", __func__, old_event, event);
+
+	memcpy(event, old_event, sizeof(*event));
+	initialize_event(event);
+
+	if (event->name_len) {
+		event->file_name = kstrdup(old_event->file_name, GFP_KERNEL);
+		if (!event->file_name) {
+			kmem_cache_free(fsnotify_event_cachep, event);
+			return NULL;
+		}
+	}
+	event->tgid = get_pid(old_event->tgid);
+	if (event->data_type == FSNOTIFY_EVENT_FILE)
+		get_file(event->file);
+
+	return event;
 }
 
 /*
@@ -348,15 +395,18 @@ static void initialize_event(struct fsnotify_event *event)
  * @name the filename, if available
  */
 struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask, void *data,
-					     int data_type, const char *name, u32 cookie,
-					     gfp_t gfp)
+					     int data_type, const unsigned char *name,
+					     u32 cookie, gfp_t gfp)
 {
 	struct fsnotify_event *event;
 
-	event = kmem_cache_alloc(fsnotify_event_cachep, gfp);
+	event = kmem_cache_zalloc(fsnotify_event_cachep, gfp);
 	if (!event)
 		return NULL;
 
+	pr_debug("%s: event=%p to_tell=%p mask=%x data=%p data_type=%d\n",
+		 __func__, event, to_tell, mask, data, data_type);
+
 	initialize_event(event);
 
 	if (name) {
@@ -368,35 +418,36 @@ struct fsnotify_event *fsnotify_create_event(struct inode *to_tell, __u32 mask,
 		event->name_len = strlen(event->file_name);
 	}
 
+	event->tgid = get_pid(task_tgid(current));
 	event->sync_cookie = cookie;
 	event->to_tell = to_tell;
+	event->data_type = data_type;
 
 	switch (data_type) {
 	case FSNOTIFY_EVENT_FILE: {
-		struct file *file = data;
-		struct path *path = &file->f_path;
-		event->path.dentry = path->dentry;
-		event->path.mnt = path->mnt;
-		path_get(&event->path);
-		event->data_type = FSNOTIFY_EVENT_PATH;
-		break;
-	}
-	case FSNOTIFY_EVENT_PATH: {
-		struct path *path = data;
-		event->path.dentry = path->dentry;
-		event->path.mnt = path->mnt;
-		path_get(&event->path);
-		event->data_type = FSNOTIFY_EVENT_PATH;
+		event->file = data;
+		/*
+		 * if this file is about to disappear hold an extra reference
+		 * until we return to __fput so we don't have to worry about
+		 * future get/put destroying the file under us or generating
+		 * additional events.  Notice that we change f_mode without
+		 * holding f_lock.  This is safe since this is the only possible
+		 * reference to this object in the kernel (it was about to be
+		 * freed, remember?)
+		 */
+		if (!atomic_long_read(&event->file->f_count)) {
+			event->file->f_mode |= FMODE_NONOTIFY;
+			get_file(event->file);
+		}
+		get_file(event->file);
 		break;
 	}
 	case FSNOTIFY_EVENT_INODE:
 		event->inode = data;
-		event->data_type = FSNOTIFY_EVENT_INODE;
 		break;
 	case FSNOTIFY_EVENT_NONE:
 		event->inode = NULL;
-		event->path.dentry = NULL;
-		event->path.mnt = NULL;
+		event->file = NULL;
 		break;
 	default:
 		BUG();
@@ -412,8 +463,11 @@ __init int fsnotify_notification_init(void)
 	fsnotify_event_cachep = KMEM_CACHE(fsnotify_event, SLAB_PANIC);
 	fsnotify_event_holder_cachep = KMEM_CACHE(fsnotify_event_holder, SLAB_PANIC);
 
-	initialize_event(&q_overflow_event);
-	q_overflow_event.mask = FS_Q_OVERFLOW;
+	q_overflow_event = fsnotify_create_event(NULL, FS_Q_OVERFLOW, NULL,
+						 FSNOTIFY_EVENT_NONE, NULL, 0,
+						 GFP_KERNEL);
+	if (!q_overflow_event)
+		panic("unable to allocate fsnotify q_overflow_event\n");
 
 	return 0;
 }

fs/open.c

@@ -29,6 +29,7 @@
 #include <linux/falloc.h>
 #include <linux/fs_struct.h>
 #include <linux/ima.h>
+#include <linux/dnotify.h>
 
 #include "internal.h"
 
@@ -887,7 +888,7 @@ long do_sys_open(int dfd, const char __user *filename, int flags, int mode)
 				put_unused_fd(fd);
 				fd = PTR_ERR(f);
 			} else {
-				fsnotify_open(f->f_path.dentry);
+				fsnotify_open(f);
 				fd_install(fd, f);
 			}
 		}

include/linux/Kbuild

@@ -210,6 +210,7 @@ unifdef-y += ethtool.h
 unifdef-y += eventpoll.h
 unifdef-y += signalfd.h
 unifdef-y += ext2_fs.h
+unifdef-y += fanotify.h
 unifdef-y += fb.h
 unifdef-y += fcntl.h
 unifdef-y += filter.h