ldiskfs/kernel_patches/patches/rhel6.3/ext4-vmalloc.patch

   1 Index: linux-stage/fs/ext4/super.c
   2 ===================================================================
   3 --- linux-stage.orig/fs/ext4/super.c
   4 +++ linux-stage/fs/ext4/super.c
   5 @@ -675,7 +675,12 @@ static void ext4_put_super(struct super_
   6
   7         for (i = 0; i < sbi->s_gdb_count; i++)
   8                 brelse(sbi->s_group_desc[i]);
   9 -       kfree(sbi->s_group_desc);
  10 +
  11 +       if (is_vmalloc_addr(sbi->s_group_desc))
  12 +               vfree(sbi->s_group_desc);
  13 +       else
  14 +               kfree(sbi->s_group_desc);
  15 +
  16         if (is_vmalloc_addr(sbi->s_flex_groups))
  17                 vfree(sbi->s_flex_groups);
  18         else
  19 @@ -2519,12 +2524,13 @@ static int ext4_fill_super(struct super_
  20         unsigned long offset = 0;
  21         unsigned long journal_devnum = 0;
  22         unsigned long def_mount_opts;
  23 -       struct inode *root;
  24 +       struct inode *root = NULL;
  25         char *cp;
  26         const char *descr;
  27         int ret = -EINVAL;
  28         int blocksize;
  29         unsigned int db_count;
  30 +       size_t size;
  31         unsigned int i;
  32         int needs_recovery, has_huge_files;
  33         __u64 blocks_count;
  34 @@ -2850,11 +2856,18 @@ static int ext4_fill_super(struct super_
  35                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
  36         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
  37                    EXT4_DESC_PER_BLOCK(sb);
  38 -       sbi->s_group_desc = kmalloc(db_count * sizeof(struct buffer_head *),
  39 -                                   GFP_KERNEL);
  40 +       size = (size_t)db_count * sizeof(struct buffer_head *);
  41 +       sbi->s_group_desc = kzalloc(size, GFP_KERNEL);
  42         if (sbi->s_group_desc == NULL) {
  43 -               ext4_msg(sb, KERN_ERR, "not enough memory");
  44 -               goto failed_mount;
  45 +               sbi->s_group_desc = vmalloc(size);
  46 +               if (sbi->s_group_desc != NULL) {
  47 +                       memset(sbi->s_group_desc, 0, size);
  48 +               } else {
  49 +                       ext4_msg(sb, KERN_ERR, "no memory for %u groups (%u)\n",
  50 +                                sbi->s_groups_count, (unsigned int)size);
  51 +                       ret = -ENOMEM;
  52 +                       goto failed_mount;
  53 +               }
  54         }
  55
  56  #ifdef __BIG_ENDIAN
  57 @@ -3064,17 +3077,16 @@ no_journal:
  58         if (IS_ERR(root)) {
  59                 ext4_msg(sb, KERN_ERR, "get root inode failed");
  60                 ret = PTR_ERR(root);
  61 +               root = NULL;
  62                 goto failed_mount4;
  63         }
  64         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
  65 -               iput(root);
  66                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
  67                 goto failed_mount4;
  68         }
  69         sb->s_root = d_alloc_root(root);
  70         if (!sb->s_root) {
  71                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
  72 -               iput(root);
  73                 ret = -ENOMEM;
  74                 goto failed_mount4;
  75         }
  76 @@ -3125,6 +3137,7 @@ no_journal:
  77         if (err) {
  78                 ext4_msg(sb, KERN_ERR, "failed to initalize mballoc (%d)",
  79                          err);
  80 +               ret = err;
  81                 goto failed_mount4;
  82         }
  83
  84 @@ -3166,6 +3179,8 @@ cantfind_ext4:
  85         goto failed_mount;
  86
  87  failed_mount4:
  88 +       iput(root);
  89 +       sb->s_root = NULL;
  90         ext4_msg(sb, KERN_ERR, "mount failed");
  91         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
  92  failed_mount_wq:
  93 @@ -3190,7 +3205,11 @@ failed_mount3:
  94  failed_mount2:
  95         for (i = 0; i < db_count; i++)
  96                 brelse(sbi->s_group_desc[i]);
  97 -       kfree(sbi->s_group_desc);
  98 +
  99 +       if (is_vmalloc_addr(sbi->s_group_desc))
 100 +               vfree(sbi->s_group_desc);
 101 +       else
 102 +               kfree(sbi->s_group_desc);
 103  failed_mount:
 104         if (sbi->s_proc) {
 105                 remove_proc_entry(sb->s_id, ext4_proc_root);
 106 Index: linux-stage/fs/ext4/mballoc.c
 107 ===================================================================
 108 --- linux-stage.orig/fs/ext4/mballoc.c
 109 +++ linux-stage/fs/ext4/mballoc.c
 110 @@ -23,6 +23,7 @@
 111
 112  #include "mballoc.h"
 113  #include <linux/debugfs.h>
 114 +#include <linux/vmalloc.h>
 115  #include <trace/events/ext4.h>
 116
 117  /*
 118 @@ -2408,24 +2409,37 @@ static int ext4_mb_init_backend(struct s
 119         while (array_size < sizeof(*sbi->s_group_info) *
 120                num_meta_group_infos_max)
 121                 array_size = array_size << 1;
 122 -       /* An 8TB filesystem with 64-bit pointers requires a 4096 byte
 123 -        * kmalloc. A 128kb malloc should suffice for a 256TB filesystem.
 124 -        * So a two level scheme suffices for now. */
 125 -       sbi->s_group_info = kmalloc(array_size, GFP_KERNEL);
 126 +       /* A 16TB filesystem with 64-bit pointers requires an 8192 byte
 127 +        * kmalloc(). Filesystems larger than 2^32 blocks (16TB normally)
 128 +        * have group descriptors at least twice as large (64 bytes or
 129 +        * more vs. 32 bytes for traditional ext3 filesystems), so a 128TB
 130 +        * filesystem needs a 128kB allocation, which may need vmalloc(). */
 131 +       sbi->s_group_info = kzalloc(array_size, GFP_KERNEL);
 132         if (sbi->s_group_info == NULL) {
 133 -               printk(KERN_ERR "EXT4-fs: can't allocate buddy meta group\n");
 134 -               return -ENOMEM;
 135 +               sbi->s_group_info = vmalloc(array_size);
 136 +               if (sbi->s_group_info != NULL) {
 137 +                       memset(sbi->s_group_info, 0, array_size);
 138 +               } else {
 139 +                       ext4_msg(sb, KERN_ERR, "no memory for groupinfo (%u)\n",
 140 +                                array_size);
 141 +                       return -ENOMEM;
 142 +               }
 143         }
 144         sbi->s_buddy_cache = new_inode(sb);
 145         if (sbi->s_buddy_cache == NULL) {
 146 -               printk(KERN_ERR "EXT4-fs: can't get new inode\n");
 147 +               ext4_msg(sb, KERN_ERR, "can't get new inode\n");
 148                 goto err_freesgi;
 149         }
 150 +       /* To avoid potentially colliding with an valid on-disk inode number,
 151 +        * use EXT4_BAD_INO for the buddy cache inode number.  This inode is
 152 +        * not in the inode hash, so it should never be found by iget(), but
 153 +        * this will avoid confusion if it ever shows up during debugging. */
 154 +       sbi->s_buddy_cache->i_ino = EXT4_BAD_INO;
 155         EXT4_I(sbi->s_buddy_cache)->i_disksize = 0;
 156         for (i = 0; i < ngroups; i++) {
 157                 desc = ext4_get_group_desc(sb, i, NULL);
 158                 if (desc == NULL) {
 159 -                       printk(KERN_ERR
 160 +                       ext4_msg(sb, KERN_ERR,
 161                                 "EXT4-fs: can't read descriptor %u\n", i);
 162                         goto err_freebuddy;
 163                 }
 164 @@ -2461,7 +2474,10 @@ err_freebuddy:
 165                 kfree(sbi->s_group_info[i]);
 166         iput(sbi->s_buddy_cache);
 167  err_freesgi:
 168 -       kfree(sbi->s_group_info);
 169 +       if (is_vmalloc_addr(sbi->s_group_info))
 170 +               vfree(sbi->s_group_info);
 171 +       else
 172 +               kfree(sbi->s_group_info);
 173         return -ENOMEM;
 174  }
 175
 176 @@ -2639,7 +2647,10 @@ int ext4_mb_release(struct super_block *
 177                         EXT4_DESC_PER_BLOCK_BITS(sb);
 178                 for (i = 0; i < num_meta_group_infos; i++)
 179                         kfree(sbi->s_group_info[i]);
 180 -               kfree(sbi->s_group_info);
 181 +               if (is_vmalloc_addr(sbi->s_group_info))
 182 +                       vfree(sbi->s_group_info);
 183 +               else
 184 +                       kfree(sbi->s_group_info);
 185         }
 186         kfree(sbi->s_mb_prealloc_table);
 187         kfree(sbi->s_mb_offsets);