mempolicy: use struct mempolicy pointer in shmem_sb_info

This patch replaces the mempolicy mode, mode_flags, and nodemask in the shmem_sb_info struct with a struct mempolicy pointer, initialized to NULL. This removes dependency on the details of mempolicy from shmem.c and hugetlbfs inode.c and simplifies the interfaces. mpol_parse_str() in mempolicy.c is changed to return, via a pointer to a pointer arg, a struct mempolicy pointer on success. For MPOL_DEFAULT, the returned pointer is NULL. Further, mpol_parse_str() now takes a 'no_context' argument that causes the input nodemask to be stored in the w.user_nodemask of the created mempolicy for use when the mempolicy is installed in a tmpfs inode shared policy tree. At that time, any cpuset contextualization is applied to the original input nodemask. This preserves the previous behavior where the input nodemask was stored in the superblock. We can think of the returned mempolicy as "context free". Because mpol_parse_str() is now calling mpol_new(), we can remove from mpol_to_str() the semantic checks that mpol_new() already performs. Add 'no_context' parameter to mpol_to_str() to specify that it should format the nodemask in w.user_nodemask for 'bind' and 'interleave' policies. Change mpol_shared_policy_init() to take a pointer to a "context free" struct mempolicy and to create a new, "contextualized" mempolicy using the mode, mode_flags and user_nodemask from the input mempolicy. Note: we know that the mempolicy passed to mpol_to_str() or mpol_shared_policy_init() from a tmpfs superblock is "context free". This is currently the only instance thereof. However, if we found more uses for this concept, and introduced any ambiguity as to whether a mempolicy was context free or not, we could add another internal mode flag to identify context free mempolicies. Then, we could remove the 'no_context' argument from mpol_to_str(). Added shmem_get_sbmpol() to return a reference counted superblock mempolicy, if one exists, to pass to mpol_shared_policy_init(). We must add the reference under the sb stat_lock to prevent races with replacement of the mpol by remount. This reference is removed in mpol_shared_policy_init(). [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: another build fix] [akpm@linux-foundation.org: yet another build fix] Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Christoph Lameter <clameter@sgi.com> Cc: David Rientjes <rientjes@google.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Andi Kleen <ak@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

mempolicy: use struct mempolicy pointer in shmem_sb_info
This patch replaces the mempolicy mode, mode_flags, and nodemask in the shmem_sb_info struct with a struct mempolicy pointer, initialized to NULL. This removes dependency on the details of mempolicy from shmem.c and hugetlbfs inode.c and simplifies the interfaces. mpol_parse_str() in mempolicy.c is changed to return, via a pointer to a pointer arg, a struct mempolicy pointer on success. For MPOL_DEFAULT, the returned pointer is NULL. Further, mpol_parse_str() now takes a 'no_context' argument that causes the input nodemask to be stored in the w.user_nodemask of the created mempolicy for use when the mempolicy is installed in a tmpfs inode shared policy tree. At that time, any cpuset contextualization is applied to the original input nodemask. This preserves the previous behavior where the input nodemask was stored in the superblock. We can think of the returned mempolicy as "context free". Because mpol_parse_str() is now calling mpol_new(), we can remove from mpol_to_str() the semantic checks that mpol_new() already performs. Add 'no_context' parameter to mpol_to_str() to specify that it should format the nodemask in w.user_nodemask for 'bind' and 'interleave' policies. Change mpol_shared_policy_init() to take a pointer to a "context free" struct mempolicy and to create a new, "contextualized" mempolicy using the mode, mode_flags and user_nodemask from the input mempolicy. Note: we know that the mempolicy passed to mpol_to_str() or mpol_shared_policy_init() from a tmpfs superblock is "context free". This is currently the only instance thereof. However, if we found more uses for this concept, and introduced any ambiguity as to whether a mempolicy was context free or not, we could add another internal mode flag to identify context free mempolicies. Then, we could remove the 'no_context' argument from mpol_to_str(). Added shmem_get_sbmpol() to return a reference counted superblock mempolicy, if one exists, to pass to mpol_shared_policy_init(). We must add the reference under the sb stat_lock to prevent races with replacement of the mpol by remount. This reference is removed in mpol_shared_policy_init(). [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: another build fix] [akpm@linux-foundation.org: yet another build fix] Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com> Cc: Christoph Lameter <clameter@sgi.com> Cc: David Rientjes <rientjes@google.com> Cc: Mel Gorman <mel@csn.ul.ie> Cc: Andi Kleen <ak@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Lee Schermerhorn · Linus Torvalds
1 parent 3f226aa1cb
Showing 5 changed files with 134 additions and 95 deletions Side-by-side Diff
fs/hugetlbfs/inode.c
include/linux/mempolicy.h
include/linux/shmem_fs.h
mm/mempolicy.c
mm/shmem.c
@@ -504,7 +504,7 @@
 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 		INIT_LIST_HEAD(&inode->i_mapping->private_list);
 		info = HUGETLBFS_I(inode);
-		mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 0, NULL);
+		mpol_shared_policy_init(&info->policy, NULL);
 		switch (mode & S_IFMT) {
 		default:
 			init_special_inode(inode, mode, dev);
@@ -182,8 +182,7 @@
 	spinlock_t lock;
 };
  
-void mpol_shared_policy_init(struct shared_policy *info, unsigned short mode,
-				unsigned short flags, nodemask_t *nodes);
+void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol);
 int mpol_set_shared_policy(struct shared_policy *info,
 				struct vm_area_struct *vma,
 				struct mempolicy *new);
  
@@ -216,10 +215,10 @@
  
  
 #ifdef CONFIG_TMPFS
-extern int mpol_parse_str(char *str, unsigned short *mode,
-			unsigned short *mode_flags, nodemask_t *policy_nodes);
+extern int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context);
  
-extern int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol);
+extern int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol,
+			int no_context);
 #endif
 #else
  
@@ -262,8 +261,8 @@
 	return -EINVAL;
 }
  
-static inline void mpol_shared_policy_init(struct shared_policy *info,
-		unsigned short mode, unsigned short flags, nodemask_t *nodes)
+static inline void mpol_shared_policy_init(struct shared_policy *sp,
+						struct mempolicy *mpol)
 {
 }
  
  
  
@@ -322,13 +321,14 @@
 }
  
 #ifdef CONFIG_TMPFS
-static inline int mpol_parse_str(char *value, unsigned short *policy,
-				unsigned short flags, nodemask_t *policy_nodes)
+static inline int mpol_parse_str(char *str, struct mempolicy **mpol,
+				int no_context)
 {
-	return 1;
+	return 1;	/* error */
 }
  
-static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
+static inline int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol,
+				int no_context)
 {
 	return 0;
 }
@@ -34,9 +34,7 @@
 	uid_t uid;		    /* Mount uid for root directory */
 	gid_t gid;		    /* Mount gid for root directory */
 	mode_t mode;		    /* Mount mode for root directory */
-	unsigned short policy;	    /* Default NUMA memory alloc policy */
-	unsigned short flags;	    /* Optional mempolicy flags */
-	nodemask_t policy_nodes;    /* nodemask for preferred and bind */
+	struct mempolicy *mpol;     /* default memory policy for mappings */
 };
  
 static inline struct shmem_inode_info *SHMEM_I(struct inode *inode)
@@ -1828,27 +1828,35 @@
 	return 0;
 }
  
-void mpol_shared_policy_init(struct shared_policy *info, unsigned short policy,
-			unsigned short flags, nodemask_t *policy_nodes)
+/**
+ * mpol_shared_policy_init - initialize shared policy for inode
+ * @sp: pointer to inode shared policy
+ * @mpol:  struct mempolicy to install
+ *
+ * Install non-NULL @mpol in inode's shared policy rb-tree.
+ * On entry, the current task has a reference on a non-NULL @mpol.
+ * This must be released on exit.
+ */
+void mpol_shared_policy_init(struct shared_policy *sp, struct mempolicy *mpol)
 {
-	info->root = RB_ROOT;
-	spin_lock_init(&info->lock);
+	sp->root = RB_ROOT;		/* empty tree == default mempolicy */
+	spin_lock_init(&sp->lock);
  
-	if (policy != MPOL_DEFAULT) {
-		struct mempolicy *newpol;
+	if (mpol) {
+		struct vm_area_struct pvma;
+		struct mempolicy *new;
  
-		/* Falls back to NULL policy [MPOL_DEFAULT] on any error */
-		newpol = mpol_new(policy, flags, policy_nodes);
-		if (!IS_ERR(newpol)) {
-			/* Create pseudo-vma that contains just the policy */
-			struct vm_area_struct pvma;
+		/* contextualize the tmpfs mount point mempolicy */
+		new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
+		mpol_put(mpol);	/* drop our ref on sb mpol */
+		if (IS_ERR(new))
+			return;		/* no valid nodemask intersection */
  
-			memset(&pvma, 0, sizeof(struct vm_area_struct));
-			/* Policy covers entire file */
-			pvma.vm_end = TASK_SIZE;
-			mpol_set_shared_policy(info, &pvma, newpol);
-			mpol_put(newpol);
-		}
+		/* Create pseudo-vma that contains just the policy */
+		memset(&pvma, 0, sizeof(struct vm_area_struct));
+		pvma.vm_end = TASK_SIZE;	/* policy covers entire file */
+		mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
+		mpol_put(new);			/* drop initial ref */
 	}
 }
  
  
  
  
@@ -1962,18 +1970,27 @@
 /**
  * mpol_parse_str - parse string to mempolicy
  * @str:  string containing mempolicy to parse
- * @mode:  pointer to returned policy mode
- * @mode_flags:  pointer to returned flags
- * @policy_nodes:  pointer to returned nodemask
+ * @mpol:  pointer to struct mempolicy pointer, returned on success.
+ * @no_context:  flag whether to "contextualize" the mempolicy
  *
  * Format of input:
  *	<mode>[=<flags>][:<nodelist>]
  *
- * Currently only used for tmpfs/shmem mount options
+ * if @no_context is true, save the input nodemask in w.user_nodemask in
+ * the returned mempolicy.  This will be used to "clone" the mempolicy in
+ * a specific context [cpuset] at a later time.  Used to parse tmpfs mpol
+ * mount option.  Note that if 'static' or 'relative' mode flags were
+ * specified, the input nodemask will already have been saved.  Saving
+ * it again is redundant, but safe.
+ *
+ * On success, returns 0, else 1
  */
-int mpol_parse_str(char *str, unsigned short *mode, unsigned short *mode_flags,
-			nodemask_t *policy_nodes)
+int mpol_parse_str(char *str, struct mempolicy **mpol, int no_context)
 {
+	struct mempolicy *new = NULL;
+	unsigned short uninitialized_var(mode);
+	unsigned short uninitialized_var(mode_flags);
+	nodemask_t nodes;
 	char *nodelist = strchr(str, ':');
 	char *flags = strchr(str, '=');
 	int i;
  
  
  
  
  
@@ -1982,26 +1999,30 @@
 	if (nodelist) {
 		/* NUL-terminate mode or flags string */
 		*nodelist++ = '\0';
-		if (nodelist_parse(nodelist, *policy_nodes))
+		if (nodelist_parse(nodelist, nodes))
 			goto out;
-		if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY]))
+		if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY]))
 			goto out;
-	}
+	} else
+		nodes_clear(nodes);
+
 	if (flags)
 		*flags++ = '\0';	/* terminate mode string */
  
 	for (i = 0; i <= MPOL_LOCAL; i++) {
 		if (!strcmp(str, policy_types[i])) {
-			*mode = i;
+			mode = i;
 			break;
 		}
 	}
 	if (i > MPOL_LOCAL)
 		goto out;
  
-	switch (*mode) {
+	switch (mode) {
 	case MPOL_PREFERRED:
-		/* Insist on a nodelist of one node only */
+		/*
+		 * Insist on a nodelist of one node only
+		 */
 		if (nodelist) {
 			char *rest = nodelist;
 			while (isdigit(*rest))
  
  
  
  
  
  
  
  
  
  
  
  
  
@@ -2010,63 +2031,73 @@
 				err = 0;
 		}
 		break;
-	case MPOL_BIND:
-		/* Insist on a nodelist */
-		if (nodelist)
-			err = 0;
-		break;
 	case MPOL_INTERLEAVE:
 		/*
 		 * Default to online nodes with memory if no nodelist
 		 */
 		if (!nodelist)
-			*policy_nodes = node_states[N_HIGH_MEMORY];
+			nodes = node_states[N_HIGH_MEMORY];
 		err = 0;
 		break;
-	default:
+	case MPOL_LOCAL:
 		/*
-		 * MPOL_DEFAULT or MPOL_LOCAL
-		 * Don't allow a nodelist nor flags
+		 * Don't allow a nodelist;  mpol_new() checks flags
 		 */
-		if (!nodelist && !flags)
-			err = 0;
-		if (*mode == MPOL_DEFAULT)
+		if (nodelist)
 			goto out;
-		/* else MPOL_LOCAL */
-		*mode = MPOL_PREFERRED;
-		nodes_clear(*policy_nodes);
+		mode = MPOL_PREFERRED;
 		break;
+
+	/*
+	 * case MPOL_BIND:    mpol_new() enforces non-empty nodemask.
+	 * case MPOL_DEFAULT: mpol_new() enforces empty nodemask, ignores flags.
+	 */
 	}
  
-	*mode_flags = 0;
+	mode_flags = 0;
 	if (flags) {
 		/*
 		 * Currently, we only support two mutually exclusive
 		 * mode flags.
 		 */
 		if (!strcmp(flags, "static"))
-			*mode_flags |= MPOL_F_STATIC_NODES;
+			mode_flags |= MPOL_F_STATIC_NODES;
 		else if (!strcmp(flags, "relative"))
-			*mode_flags |= MPOL_F_RELATIVE_NODES;
+			mode_flags |= MPOL_F_RELATIVE_NODES;
 		else
 			err = 1;
 	}
+
+	new = mpol_new(mode, mode_flags, &nodes);
+	if (IS_ERR(new))
+		err = 1;
+	else if (no_context)
+		new->w.user_nodemask = nodes;	/* save for contextualization */
+
 out:
 	/* Restore string for error message */
 	if (nodelist)
 		*--nodelist = ':';
 	if (flags)
 		*--flags = '=';
+	if (!err)
+		*mpol = new;
 	return err;
 }
 #endif /* CONFIG_TMPFS */
  
-/*
+/**
+ * mpol_to_str - format a mempolicy structure for printing
+ * @buffer:  to contain formatted mempolicy string
+ * @maxlen:  length of @buffer
+ * @pol:  pointer to mempolicy to be formatted
+ * @no_context:  "context free" mempolicy - use nodemask in w.user_nodemask
+ *
  * Convert a mempolicy into a string.
  * Returns the number of characters in buffer (if positive)
  * or an error (negative)
  */
-int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol)
+int mpol_to_str(char *buffer, int maxlen, struct mempolicy *pol, int no_context)
 {
 	char *p = buffer;
 	int l;
@@ -2100,7 +2131,10 @@
 	case MPOL_BIND:
 		/* Fall through */
 	case MPOL_INTERLEAVE:
-		nodes = pol->v.nodes;
+		if (no_context)
+			nodes = pol->w.user_nodemask;
+		else
+			nodes = pol->v.nodes;
 		break;
  
 	default:
@@ -2231,7 +2265,7 @@
 		return 0;
  
 	pol = get_vma_policy(priv->task, vma, vma->vm_start);
-	mpol_to_str(buffer, sizeof(buffer), pol);
+	mpol_to_str(buffer, sizeof(buffer), pol, 0);
 	mpol_cond_put(pol);
  
 	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
@@ -1079,23 +1079,29 @@
  
 #ifdef CONFIG_NUMA
 #ifdef CONFIG_TMPFS
-static void shmem_show_mpol(struct seq_file *seq, unsigned short mode,
-			unsigned short flags, const nodemask_t policy_nodes)
+static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 {
-	struct mempolicy temp;
 	char buffer[64];
  
-	if (mode == MPOL_DEFAULT)
+	if (!mpol || mpol->mode == MPOL_DEFAULT)
 		return;		/* show nothing */
  
-	temp.mode = mode;
-	temp.flags = flags;
-	temp.v.nodes = policy_nodes;
+	mpol_to_str(buffer, sizeof(buffer), mpol, 1);
  
-	mpol_to_str(buffer, sizeof(buffer), &temp);
-
 	seq_printf(seq, ",mpol=%s", buffer);
 }
+
+static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
+{
+	struct mempolicy *mpol = NULL;
+	if (sbinfo->mpol) {
+		spin_lock(&sbinfo->stat_lock);	/* prevent replace/use races */
+		mpol = sbinfo->mpol;
+		mpol_get(mpol);
+		spin_unlock(&sbinfo->stat_lock);
+	}
+	return mpol;
+}
 #endif /* CONFIG_TMPFS */
  
 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
@@ -1135,8 +1141,7 @@
 }
 #else /* !CONFIG_NUMA */
 #ifdef CONFIG_TMPFS
-static inline void shmem_show_mpol(struct seq_file *seq, unsigned short policy,
-			unsigned short flags, const nodemask_t policy_nodes)
+static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
 {
 }
 #endif /* CONFIG_TMPFS */
@@ -1154,6 +1159,13 @@
 }
 #endif /* CONFIG_NUMA */
  
+#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
+static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
+{
+	return NULL;
+}
+#endif
+
 /*
  * shmem_getpage - either get the page from swap or allocate a new one
  *
@@ -1508,8 +1520,8 @@
 		case S_IFREG:
 			inode->i_op = &shmem_inode_operations;
 			inode->i_fop = &shmem_file_operations;
-			mpol_shared_policy_init(&info->policy, sbinfo->policy,
-					sbinfo->flags, &sbinfo->policy_nodes);
+			mpol_shared_policy_init(&info->policy,
+						 shmem_get_sbmpol(sbinfo));
 			break;
 		case S_IFDIR:
 			inc_nlink(inode);
@@ -1523,8 +1535,7 @@
 			 * Must not load anything in the rbtree,
 			 * mpol_free_shared_policy will not be called.
 			 */
-			mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 0,
-						NULL);
+			mpol_shared_policy_init(&info->policy, NULL);
 			break;
 		}
 	} else
@@ -2139,8 +2150,7 @@
 			if (*rest)
 				goto bad_val;
 		} else if (!strcmp(this_char,"mpol")) {
-			if (mpol_parse_str(value, &sbinfo->policy,
-					 &sbinfo->flags, &sbinfo->policy_nodes))
+			if (mpol_parse_str(value, &sbinfo->mpol, 1))
 				goto bad_val;
 		} else {
 			printk(KERN_ERR "tmpfs: Bad mount option %s\n",
@@ -2191,9 +2201,9 @@
 	sbinfo->free_blocks = config.max_blocks - blocks;
 	sbinfo->max_inodes  = config.max_inodes;
 	sbinfo->free_inodes = config.max_inodes - inodes;
-	sbinfo->policy      = config.policy;
-	sbinfo->flags	    = config.flags;
-	sbinfo->policy_nodes = config.policy_nodes;
+
+	mpol_put(sbinfo->mpol);
+	sbinfo->mpol        = config.mpol;	/* transfers initial ref */
 out:
 	spin_unlock(&sbinfo->stat_lock);
 	return error;
@@ -2214,8 +2224,7 @@
 		seq_printf(seq, ",uid=%u", sbinfo->uid);
 	if (sbinfo->gid != 0)
 		seq_printf(seq, ",gid=%u", sbinfo->gid);
-	shmem_show_mpol(seq, sbinfo->policy, sbinfo->flags,
-			sbinfo->policy_nodes);
+	shmem_show_mpol(seq, sbinfo->mpol);
 	return 0;
 }
 #endif /* CONFIG_TMPFS */
@@ -2245,9 +2254,7 @@
 	sbinfo->mode = S_IRWXUGO | S_ISVTX;
 	sbinfo->uid = current->fsuid;
 	sbinfo->gid = current->fsgid;
-	sbinfo->policy = MPOL_DEFAULT;
-	sbinfo->flags = 0;
-	sbinfo->policy_nodes = node_states[N_HIGH_MEMORY];
+	sbinfo->mpol = NULL;
 	sb->s_fs_info = sbinfo;
  
 #ifdef CONFIG_TMPFS
...	...	@@ -504,7 +504,7 @@
504	504	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
505	505	INIT_LIST_HEAD(&inode->i_mapping->private_list);
506	506	info = HUGETLBFS_I(inode);
507		- mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 0, NULL);
	507	+ mpol_shared_policy_init(&info->policy, NULL);
508	508	switch (mode & S_IFMT) {
509	509	default:
510	510	init_special_inode(inode, mode, dev);
...	...	@@ -182,8 +182,7 @@
182	182	spinlock_t lock;
183	183	};
184	184
185		-void mpol_shared_policy_init(struct shared_policy *info, unsigned short mode,
186		- unsigned short flags, nodemask_t *nodes);
	185	+void mpol_shared_policy_init(struct shared_policy sp, struct mempolicy mpol);
187	186	int mpol_set_shared_policy(struct shared_policy *info,
188	187	struct vm_area_struct *vma,
189	188	struct mempolicy *new);
190	189
...	...	@@ -216,10 +215,10 @@
216	215
217	216
218	217	#ifdef CONFIG_TMPFS
219		-extern int mpol_parse_str(char str, unsigned short mode,
220		- unsigned short mode_flags, nodemask_t policy_nodes);
	218	+extern int mpol_parse_str(char str, struct mempolicy *mpol, int no_context);
221	219
222		-extern int mpol_to_str(char buffer, int maxlen, struct mempolicy pol);
	220	+extern int mpol_to_str(char buffer, int maxlen, struct mempolicy pol,
	221	+ int no_context);
223	222	#endif
224	223	#else
225	224
...	...	@@ -262,8 +261,8 @@
262	261	return -EINVAL;
263	262	}
264	263
265		-static inline void mpol_shared_policy_init(struct shared_policy *info,
266		- unsigned short mode, unsigned short flags, nodemask_t *nodes)
	264	+static inline void mpol_shared_policy_init(struct shared_policy *sp,
	265	+ struct mempolicy *mpol)
267	266	{
268	267	}
269	268
270	269
271	270
...	...	@@ -322,13 +321,14 @@
322	321	}
323	322
324	323	#ifdef CONFIG_TMPFS
325		-static inline int mpol_parse_str(char value, unsigned short policy,
326		- unsigned short flags, nodemask_t *policy_nodes)
	324	+static inline int mpol_parse_str(char str, struct mempolicy *mpol,
	325	+ int no_context)
327	326	{
328		- return 1;
	327	+ return 1; /* error */
329	328	}
330	329
331		-static inline int mpol_to_str(char buffer, int maxlen, struct mempolicy pol)
	330	+static inline int mpol_to_str(char buffer, int maxlen, struct mempolicy pol,
	331	+ int no_context)
332	332	{
333	333	return 0;
334	334	}
...	...	@@ -34,9 +34,7 @@
34	34	uid_t uid; /* Mount uid for root directory */
35	35	gid_t gid; /* Mount gid for root directory */
36	36	mode_t mode; /* Mount mode for root directory */
37		- unsigned short policy; /* Default NUMA memory alloc policy */
38		- unsigned short flags; /* Optional mempolicy flags */
39		- nodemask_t policy_nodes; /* nodemask for preferred and bind */
	37	+ struct mempolicy mpol; / default memory policy for mappings */
40	38	};
41	39
42	40	static inline struct shmem_inode_info SHMEM_I(struct inode inode)
...	...	@@ -1828,27 +1828,35 @@
1828	1828	return 0;
1829	1829	}
1830	1830
1831		-void mpol_shared_policy_init(struct shared_policy *info, unsigned short policy,
1832		- unsigned short flags, nodemask_t *policy_nodes)
	1831	+/**
	1832	+ * mpol_shared_policy_init - initialize shared policy for inode
	1833	+ * @sp: pointer to inode shared policy
	1834	+ * @mpol: struct mempolicy to install
	1835	+ *
	1836	+ * Install non-NULL @mpol in inode's shared policy rb-tree.
	1837	+ * On entry, the current task has a reference on a non-NULL @mpol.
	1838	+ * This must be released on exit.
	1839	+ */
	1840	+void mpol_shared_policy_init(struct shared_policy sp, struct mempolicy mpol)
1833	1841	{
1834		- info->root = RB_ROOT;
1835		- spin_lock_init(&info->lock);
	1842	+ sp->root = RB_ROOT; /* empty tree == default mempolicy */
	1843	+ spin_lock_init(&sp->lock);
1836	1844
1837		- if (policy != MPOL_DEFAULT) {
1838		- struct mempolicy *newpol;
	1845	+ if (mpol) {
	1846	+ struct vm_area_struct pvma;
	1847	+ struct mempolicy *new;
1839	1848
1840		- /* Falls back to NULL policy [MPOL_DEFAULT] on any error */
1841		- newpol = mpol_new(policy, flags, policy_nodes);
1842		- if (!IS_ERR(newpol)) {
1843		- /* Create pseudo-vma that contains just the policy */
1844		- struct vm_area_struct pvma;
	1849	+ /* contextualize the tmpfs mount point mempolicy */
	1850	+ new = mpol_new(mpol->mode, mpol->flags, &mpol->w.user_nodemask);
	1851	+ mpol_put(mpol); /* drop our ref on sb mpol */
	1852	+ if (IS_ERR(new))
	1853	+ return; /* no valid nodemask intersection */
1845	1854
1846		- memset(&pvma, 0, sizeof(struct vm_area_struct));
1847		- /* Policy covers entire file */
1848		- pvma.vm_end = TASK_SIZE;
1849		- mpol_set_shared_policy(info, &pvma, newpol);
1850		- mpol_put(newpol);
1851		- }
	1855	+ /* Create pseudo-vma that contains just the policy */
	1856	+ memset(&pvma, 0, sizeof(struct vm_area_struct));
	1857	+ pvma.vm_end = TASK_SIZE; /* policy covers entire file */
	1858	+ mpol_set_shared_policy(sp, &pvma, new); /* adds ref */
	1859	+ mpol_put(new); /* drop initial ref */
1852	1860	}
1853	1861	}
1854	1862
1855	1863
1856	1864
1857	1865
...	...	@@ -1962,18 +1970,27 @@
1962	1970	/**
1963	1971	* mpol_parse_str - parse string to mempolicy
1964	1972	* @str: string containing mempolicy to parse
1965		- * @mode: pointer to returned policy mode
1966		- * @mode_flags: pointer to returned flags
1967		- * @policy_nodes: pointer to returned nodemask
	1973	+ * @mpol: pointer to struct mempolicy pointer, returned on success.
	1974	+ * @no_context: flag whether to "contextualize" the mempolicy
1968	1975	*
1969	1976	* Format of input:
1970	1977	* <mode>[=<flags>][:<nodelist>]
1971	1978	*
1972		- * Currently only used for tmpfs/shmem mount options
	1979	+ * if @no_context is true, save the input nodemask in w.user_nodemask in
	1980	+ * the returned mempolicy. This will be used to "clone" the mempolicy in
	1981	+ * a specific context [cpuset] at a later time. Used to parse tmpfs mpol
	1982	+ * mount option. Note that if 'static' or 'relative' mode flags were
	1983	+ * specified, the input nodemask will already have been saved. Saving
	1984	+ * it again is redundant, but safe.
	1985	+ *
	1986	+ * On success, returns 0, else 1
1973	1987	*/
1974		-int mpol_parse_str(char str, unsigned short mode, unsigned short *mode_flags,
1975		- nodemask_t *policy_nodes)
	1988	+int mpol_parse_str(char str, struct mempolicy *mpol, int no_context)
1976	1989	{
	1990	+ struct mempolicy *new = NULL;
	1991	+ unsigned short uninitialized_var(mode);
	1992	+ unsigned short uninitialized_var(mode_flags);
	1993	+ nodemask_t nodes;
1977	1994	char *nodelist = strchr(str, ':');
1978	1995	char *flags = strchr(str, '=');
1979	1996	int i;
1980	1997
1981	1998
1982	1999
1983	2000
1984	2001
...	...	@@ -1982,26 +1999,30 @@
1982	1999	if (nodelist) {
1983	2000	/* NUL-terminate mode or flags string */
1984	2001	*nodelist++ = '\0';
1985		- if (nodelist_parse(nodelist, *policy_nodes))
	2002	+ if (nodelist_parse(nodelist, nodes))
1986	2003	goto out;
1987		- if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY]))
	2004	+ if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY]))
1988	2005	goto out;
1989		- }
	2006	+ } else
	2007	+ nodes_clear(nodes);
	2008	+
1990	2009	if (flags)
1991	2010	flags++ = '\0'; / terminate mode string */
1992	2011
1993	2012	for (i = 0; i <= MPOL_LOCAL; i++) {
1994	2013	if (!strcmp(str, policy_types[i])) {
1995		- *mode = i;
	2014	+ mode = i;
1996	2015	break;
1997	2016	}
1998	2017	}
1999	2018	if (i > MPOL_LOCAL)
2000	2019	goto out;
2001	2020
2002		- switch (*mode) {
	2021	+ switch (mode) {
2003	2022	case MPOL_PREFERRED:
2004		- /* Insist on a nodelist of one node only */
	2023	+ /*
	2024	+ * Insist on a nodelist of one node only
	2025	+ */
2005	2026	if (nodelist) {
2006	2027	char *rest = nodelist;
2007	2028	while (isdigit(*rest))
2008	2029
2009	2030
2010	2031
2011	2032
2012	2033
2013	2034
2014	2035
2015	2036
2016	2037
2017	2038
2018	2039
2019	2040
2020	2041
...	...	@@ -2010,63 +2031,73 @@
2010	2031	err = 0;
2011	2032	}
2012	2033	break;
2013		- case MPOL_BIND:
2014		- /* Insist on a nodelist */
2015		- if (nodelist)
2016		- err = 0;
2017		- break;
2018	2034	case MPOL_INTERLEAVE:
2019	2035	/*
2020	2036	* Default to online nodes with memory if no nodelist
2021	2037	*/
2022	2038	if (!nodelist)
2023		- *policy_nodes = node_states[N_HIGH_MEMORY];
	2039	+ nodes = node_states[N_HIGH_MEMORY];
2024	2040	err = 0;
2025	2041	break;
2026		- default:
	2042	+ case MPOL_LOCAL:
2027	2043	/*
2028		- * MPOL_DEFAULT or MPOL_LOCAL
2029		- * Don't allow a nodelist nor flags
	2044	+ * Don't allow a nodelist; mpol_new() checks flags
2030	2045	*/
2031		- if (!nodelist && !flags)
2032		- err = 0;
2033		- if (*mode == MPOL_DEFAULT)
	2046	+ if (nodelist)
2034	2047	goto out;
2035		- /* else MPOL_LOCAL */
2036		- *mode = MPOL_PREFERRED;
2037		- nodes_clear(*policy_nodes);
	2048	+ mode = MPOL_PREFERRED;
2038	2049	break;
	2050	+
	2051	+ /*
	2052	+ * case MPOL_BIND: mpol_new() enforces non-empty nodemask.
	2053	+ * case MPOL_DEFAULT: mpol_new() enforces empty nodemask, ignores flags.
	2054	+ */
2039	2055	}
2040	2056
2041		- *mode_flags = 0;
	2057	+ mode_flags = 0;
2042	2058	if (flags) {
2043	2059	/*
2044	2060	* Currently, we only support two mutually exclusive
2045	2061	* mode flags.
2046	2062	*/
2047	2063	if (!strcmp(flags, "static"))
2048		- *mode_flags \|= MPOL_F_STATIC_NODES;
	2064	+ mode_flags \|= MPOL_F_STATIC_NODES;
2049	2065	else if (!strcmp(flags, "relative"))
2050		- *mode_flags \|= MPOL_F_RELATIVE_NODES;
	2066	+ mode_flags \|= MPOL_F_RELATIVE_NODES;
2051	2067	else
2052	2068	err = 1;
2053	2069	}
	2070	+
	2071	+ new = mpol_new(mode, mode_flags, &nodes);
	2072	+ if (IS_ERR(new))
	2073	+ err = 1;
	2074	+ else if (no_context)
	2075	+ new->w.user_nodemask = nodes; /* save for contextualization */
	2076	+
2054	2077	out:
2055	2078	/* Restore string for error message */
2056	2079	if (nodelist)
2057	2080	*--nodelist = ':';
2058	2081	if (flags)
2059	2082	*--flags = '=';
	2083	+ if (!err)
	2084	+ *mpol = new;
2060	2085	return err;
2061	2086	}
2062	2087	#endif /* CONFIG_TMPFS */
2063	2088
2064		-/*
	2089	+/**
	2090	+ * mpol_to_str - format a mempolicy structure for printing
	2091	+ * @buffer: to contain formatted mempolicy string
	2092	+ * @maxlen: length of @buffer
	2093	+ * @pol: pointer to mempolicy to be formatted
	2094	+ * @no_context: "context free" mempolicy - use nodemask in w.user_nodemask
	2095	+ *
2065	2096	* Convert a mempolicy into a string.
2066	2097	* Returns the number of characters in buffer (if positive)
2067	2098	* or an error (negative)
2068	2099	*/
2069		-int mpol_to_str(char buffer, int maxlen, struct mempolicy pol)
	2100	+int mpol_to_str(char buffer, int maxlen, struct mempolicy pol, int no_context)
2070	2101	{
2071	2102	char *p = buffer;
2072	2103	int l;
...	...	@@ -2100,7 +2131,10 @@
2100	2131	case MPOL_BIND:
2101	2132	/* Fall through */
2102	2133	case MPOL_INTERLEAVE:
2103		- nodes = pol->v.nodes;
	2134	+ if (no_context)
	2135	+ nodes = pol->w.user_nodemask;
	2136	+ else
	2137	+ nodes = pol->v.nodes;
2104	2138	break;
2105	2139
2106	2140	default:
...	...	@@ -2231,7 +2265,7 @@
2231	2265	return 0;
2232	2266
2233	2267	pol = get_vma_policy(priv->task, vma, vma->vm_start);
2234		- mpol_to_str(buffer, sizeof(buffer), pol);
	2268	+ mpol_to_str(buffer, sizeof(buffer), pol, 0);
2235	2269	mpol_cond_put(pol);
2236	2270
2237	2271	seq_printf(m, "%08lx %s", vma->vm_start, buffer);
...	...	@@ -1079,23 +1079,29 @@
1079	1079
1080	1080	#ifdef CONFIG_NUMA
1081	1081	#ifdef CONFIG_TMPFS
1082		-static void shmem_show_mpol(struct seq_file *seq, unsigned short mode,
1083		- unsigned short flags, const nodemask_t policy_nodes)
	1082	+static void shmem_show_mpol(struct seq_file seq, struct mempolicy mpol)
1084	1083	{
1085		- struct mempolicy temp;
1086	1084	char buffer[64];
1087	1085
1088		- if (mode == MPOL_DEFAULT)
	1086	+ if (!mpol \|\| mpol->mode == MPOL_DEFAULT)
1089	1087	return; /* show nothing */
1090	1088
1091		- temp.mode = mode;
1092		- temp.flags = flags;
1093		- temp.v.nodes = policy_nodes;
	1089	+ mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1094	1090
1095		- mpol_to_str(buffer, sizeof(buffer), &temp);
1096		-
1097	1091	seq_printf(seq, ",mpol=%s", buffer);
1098	1092	}
	1093	+
	1094	+static struct mempolicy shmem_get_sbmpol(struct shmem_sb_info sbinfo)
	1095	+{
	1096	+ struct mempolicy *mpol = NULL;
	1097	+ if (sbinfo->mpol) {
	1098	+ spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
	1099	+ mpol = sbinfo->mpol;
	1100	+ mpol_get(mpol);
	1101	+ spin_unlock(&sbinfo->stat_lock);
	1102	+ }
	1103	+ return mpol;
	1104	+}
1099	1105	#endif /* CONFIG_TMPFS */
1100	1106
1101	1107	static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
...	...	@@ -1135,8 +1141,7 @@
1135	1141	}
1136	1142	#else /* !CONFIG_NUMA */
1137	1143	#ifdef CONFIG_TMPFS
1138		-static inline void shmem_show_mpol(struct seq_file *seq, unsigned short policy,
1139		- unsigned short flags, const nodemask_t policy_nodes)
	1144	+static inline void shmem_show_mpol(struct seq_file seq, struct mempolicy p)
1140	1145	{
1141	1146	}
1142	1147	#endif /* CONFIG_TMPFS */
...	...	@@ -1154,6 +1159,13 @@
1154	1159	}
1155	1160	#endif /* CONFIG_NUMA */
1156	1161
	1162	+#if !defined(CONFIG_NUMA) \|\| !defined(CONFIG_TMPFS)
	1163	+static inline struct mempolicy shmem_get_sbmpol(struct shmem_sb_info sbinfo)
	1164	+{
	1165	+ return NULL;
	1166	+}
	1167	+#endif
	1168	+
1157	1169	/*
1158	1170	* shmem_getpage - either get the page from swap or allocate a new one
1159	1171	*
...	...	@@ -1508,8 +1520,8 @@
1508	1520	case S_IFREG:
1509	1521	inode->i_op = &shmem_inode_operations;
1510	1522	inode->i_fop = &shmem_file_operations;
1511		- mpol_shared_policy_init(&info->policy, sbinfo->policy,
1512		- sbinfo->flags, &sbinfo->policy_nodes);
	1523	+ mpol_shared_policy_init(&info->policy,
	1524	+ shmem_get_sbmpol(sbinfo));
1513	1525	break;
1514	1526	case S_IFDIR:
1515	1527	inc_nlink(inode);
...	...	@@ -1523,8 +1535,7 @@
1523	1535	* Must not load anything in the rbtree,
1524	1536	* mpol_free_shared_policy will not be called.
1525	1537	*/
1526		- mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 0,
1527		- NULL);
	1538	+ mpol_shared_policy_init(&info->policy, NULL);
1528	1539	break;
1529	1540	}
1530	1541	} else
...	...	@@ -2139,8 +2150,7 @@
2139	2150	if (*rest)
2140	2151	goto bad_val;
2141	2152	} else if (!strcmp(this_char,"mpol")) {
2142		- if (mpol_parse_str(value, &sbinfo->policy,
2143		- &sbinfo->flags, &sbinfo->policy_nodes))
	2153	+ if (mpol_parse_str(value, &sbinfo->mpol, 1))
2144	2154	goto bad_val;
2145	2155	} else {
2146	2156	printk(KERN_ERR "tmpfs: Bad mount option %s\n",
...	...	@@ -2191,9 +2201,9 @@
2191	2201	sbinfo->free_blocks = config.max_blocks - blocks;
2192	2202	sbinfo->max_inodes = config.max_inodes;
2193	2203	sbinfo->free_inodes = config.max_inodes - inodes;
2194		- sbinfo->policy = config.policy;
2195		- sbinfo->flags = config.flags;
2196		- sbinfo->policy_nodes = config.policy_nodes;
	2204	+
	2205	+ mpol_put(sbinfo->mpol);
	2206	+ sbinfo->mpol = config.mpol; /* transfers initial ref */
2197	2207	out:
2198	2208	spin_unlock(&sbinfo->stat_lock);
2199	2209	return error;
...	...	@@ -2214,8 +2224,7 @@
2214	2224	seq_printf(seq, ",uid=%u", sbinfo->uid);
2215	2225	if (sbinfo->gid != 0)
2216	2226	seq_printf(seq, ",gid=%u", sbinfo->gid);
2217		- shmem_show_mpol(seq, sbinfo->policy, sbinfo->flags,
2218		- sbinfo->policy_nodes);
	2227	+ shmem_show_mpol(seq, sbinfo->mpol);
2219	2228	return 0;
2220	2229	}
2221	2230	#endif /* CONFIG_TMPFS */
...	...	@@ -2245,9 +2254,7 @@
2245	2254	sbinfo->mode = S_IRWXUGO \| S_ISVTX;
2246	2255	sbinfo->uid = current->fsuid;
2247	2256	sbinfo->gid = current->fsgid;
2248		- sbinfo->policy = MPOL_DEFAULT;
2249		- sbinfo->flags = 0;
2250		- sbinfo->policy_nodes = node_states[N_HIGH_MEMORY];
	2257	+ sbinfo->mpol = NULL;
2251	2258	sb->s_fs_info = sbinfo;
2252	2259
2253	2260	#ifdef CONFIG_TMPFS