ext4: grow the s_group_info array as needed

Previously we allocated the s_group_info array with enough space for
any future possible growth of the file system via online resize.  This
is unfortunate because it wastes memory, and it doesn't work for the
meta_bg scheme, since there is no limit based on the number of
reserved gdt blocks.  So add the code to grow the s_group_info array
as needed.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>

3 files changed, 50 insertions, 40 deletions

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 464cff711ed6..8b6902c4d7be 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -1233,6 +1233,7 @@ struct ext4_sb_info {
 	spinlock_t s_md_lock;
 	unsigned short *s_mb_offsets;
 	unsigned int *s_mb_maxs;
+	unsigned int s_group_info_size;
 
 	/* tunables */
 	unsigned long s_stripe;
@@ -1971,6 +1972,8 @@ extern void ext4_exit_mballoc(void);
 extern void ext4_free_blocks(handle_t *handle, struct inode *inode,
 			     struct buffer_head *bh, ext4_fsblk_t block,
 			     unsigned long count, int flags);
+extern int ext4_mb_alloc_groupinfo(struct super_block *sb,
+				   ext4_group_t ngroups);
 extern int ext4_mb_add_groupinfo(struct super_block *sb,
 		ext4_group_t i, struct ext4_group_desc *desc);
 extern int ext4_group_add_blocks(handle_t *handle, struct super_block *sb,
diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
index 6873571c9f44..2102c20f7e98 100644
--- a/fs/ext4/mballoc.c
+++ b/fs/ext4/mballoc.c
@@ -24,6 +24,7 @@
 #include "ext4_jbd2.h"
 #include "mballoc.h"
 #include <linux/debugfs.h>
+#include <linux/log2.h>
 #include <linux/slab.h>
 #include <trace/events/ext4.h>
 
@@ -2163,6 +2164,39 @@ static struct kmem_cache *get_groupinfo_cache(int blocksize_bits)
 	return cachep;
 }
 
+/*
+ * Allocate the top-level s_group_info array for the specified number
+ * of groups
+ */
+int ext4_mb_alloc_groupinfo(struct super_block *sb, ext4_group_t ngroups)
+{
+	struct ext4_sb_info *sbi = EXT4_SB(sb);
+	unsigned size;
+	struct ext4_group_info ***new_groupinfo;
+
+	size = (ngroups + EXT4_DESC_PER_BLOCK(sb) - 1) >>
+		EXT4_DESC_PER_BLOCK_BITS(sb);
+	if (size <= sbi->s_group_info_size)
+		return 0;
+
+	size = roundup_pow_of_two(sizeof(*sbi->s_group_info) * size);
+	new_groupinfo = ext4_kvzalloc(size, GFP_KERNEL);
+	if (!new_groupinfo) {
+		ext4_msg(sb, KERN_ERR, "can't allocate buddy meta group");
+		return -ENOMEM;
+	}
+	if (sbi->s_group_info) {
+		memcpy(new_groupinfo, sbi->s_group_info,
+		       sbi->s_group_info_size * sizeof(*sbi->s_group_info));
+		ext4_kvfree(sbi->s_group_info);
+	}
+	sbi->s_group_info = new_groupinfo;
+	sbi->s_group_info_size = size / sizeof(*sbi->s_group_info);
+	ext4_debug("allocated s_groupinfo array for %d meta_bg's\n", 
+		   sbi->s_group_info_size);
+	return 0;
+}
+
 /* Create and initialize ext4_group_info data for the given group. */
 int ext4_mb_add_groupinfo(struct super_block *sb, ext4_group_t group,
 			  struct ext4_group_desc *desc)
@@ -2252,49 +2286,14 @@ static int ext4_mb_init_backend(struct super_block *sb)
 	ext4_group_t ngroups = ext4_get_groups_count(sb);
 	ext4_group_t i;
 	struct ext4_sb_info *sbi = EXT4_SB(sb);
-	struct ext4_super_block *es = sbi->s_es;
-	int num_meta_group_infos;
-	int num_meta_group_infos_max;
-	int array_size;
+	int err;
 	struct ext4_group_desc *desc;
 	struct kmem_cache *cachep;
 
-	/* This is the number of blocks used by GDT */
-	num_meta_group_infos = (ngroups + EXT4_DESC_PER_BLOCK(sb) -
-				1) >> EXT4_DESC_PER_BLOCK_BITS(sb);
-
-	/*
-	 * This is the total number of blocks used by GDT including
-	 * the number of reserved blocks for GDT.
-	 * The s_group_info array is allocated with this value
-	 * to allow a clean online resize without a complex
-	 * manipulation of pointer.
-	 * The drawback is the unused memory when no resize
-	 * occurs but it's very low in terms of pages
-	 * (see comments below)
-	 * Need to handle this properly when META_BG resizing is allowed
-	 */
-	num_meta_group_infos_max = num_meta_group_infos +
-				le16_to_cpu(es->s_reserved_gdt_blocks);
+	err = ext4_mb_alloc_groupinfo(sb, ngroups);
+	if (err)
+		return err;
 
-	/*
-	 * array_size is the size of s_group_info array. We round it
-	 * to the next power of two because this approximation is done
-	 * internally by kmalloc so we can have some more memory
-	 * for free here (e.g. may be used for META_BG resize).
-	 */
-	array_size = 1;
-	while (array_size < sizeof(*sbi->s_group_info) *
-	       num_meta_group_infos_max)
-		array_size = array_size << 1;
-	/* An 8TB filesystem with 64-bit pointers requires a 4096 byte
-	 * kmalloc. A 128kb malloc should suffice for a 256TB filesystem.
-	 * So a two level scheme suffices for now. */
-	sbi->s_group_info = ext4_kvzalloc(array_size, GFP_KERNEL);
-	if (sbi->s_group_info == NULL) {
-		ext4_msg(sb, KERN_ERR, "can't allocate buddy meta group");
-		return -ENOMEM;
-	}
 	sbi->s_buddy_cache = new_inode(sb);
 	if (sbi->s_buddy_cache == NULL) {
 		ext4_msg(sb, KERN_ERR, "can't get new inode");
@@ -2322,7 +2321,7 @@ err_freebuddy:
 	cachep = get_groupinfo_cache(sb->s_blocksize_bits);
 	while (i-- > 0)
 		kmem_cache_free(cachep, ext4_get_group_info(sb, i));
-	i = num_meta_group_infos;
+	i = sbi->s_group_info_size;
 	while (i-- > 0)
 		kfree(sbi->s_group_info[i]);
 	iput(sbi->s_buddy_cache);
diff --git a/fs/ext4/resize.c b/fs/ext4/resize.c
index 3f5c67bf13a2..f288933bf4c0 100644
--- a/fs/ext4/resize.c
+++ b/fs/ext4/resize.c
@@ -1507,6 +1507,10 @@ int ext4_group_add(struct super_block *sb, struct ext4_new_group_data *input)
 	if (err)
 		return err;
 
+	err = ext4_mb_alloc_groupinfo(sb, input->group + 1);
+	if (err)
+		goto out;
+
 	flex_gd.count = 1;
 	flex_gd.groups = input;
 	flex_gd.bg_flags = &bg_flags;
@@ -1732,6 +1736,10 @@ int ext4_resize_fs(struct super_block *sb, ext4_fsblk_t n_blocks_count)
 	if (err)
 		return err;
 
+	err = ext4_mb_alloc_groupinfo(sb, n_group + 1);
+	if (err)
+		goto out;
+
 	flex_gd = alloc_flex_gd(flexbg_size);
 	if (flex_gd == NULL) {
 		err = -ENOMEM;