ldiskfs/kernel_patches/patches/ext3-fiemap-2.6.18-vanilla.patch

   1 Index: linux-2.6.18/fs/ext3/ioctl.c
   2 ===================================================================
   3 --- linux-2.6.18.orig/fs/ext3/ioctl.c
   4 +++ linux-2.6.18/fs/ext3/ioctl.c
   5 @@ -15,7 +15,7 @@
   6  #include <linux/time.h>
   7  #include <asm/uaccess.h>
   8  #include <linux/namei.h>
   9 -
  10 +#include "fiemap.h"
  11
  12  int ext3_ioctl (struct inode * inode, struct file * filp, unsigned int cmd,
  13                 unsigned long arg)
  14 @@ -272,6 +272,9 @@ flags_err:
  15
  16                 return err;
  17         }
  18 +       case EXT3_IOC_FIEMAP: {
  19 +               return ext3_fiemap(inode, filp, cmd, arg);
  20 +       }
  21
  22
  23         default:
  24 Index: linux-2.6.18/include/linux/ext3_fs.h
  25 ===================================================================
  26 --- linux-2.6.18.orig/include/linux/ext3_fs.h
  27 +++ linux-2.6.18/include/linux/ext3_fs.h
  28 @@ -249,7 +249,6 @@ struct ext3_new_group_data {
  29         __u32 free_blocks_count;
  30  };
  31
  32 -
  33  /*
  34   * ioctl commands
  35   */
  36 @@ -257,15 +256,16 @@ struct ext3_new_group_data {
  37  #define        EXT3_IOC_SETFLAGS               _IOW('f', 2, long)
  38  #define        EXT3_IOC_GETVERSION             _IOR('f', 3, long)
  39  #define        EXT3_IOC_SETVERSION             _IOW('f', 4, long)
  40 -#define EXT3_IOC_GROUP_EXTEND          _IOW('f', 7, unsigned long)
  41 +#define        EXT3_IOC_GETRSVSZ               _IOR('f', 5, long)
  42 +#define        EXT3_IOC_SETRSVSZ               _IOW('f', 6, long)
  43 +#define        EXT3_IOC_GROUP_EXTEND           _IOW('f', 7, unsigned long)
  44  #define EXT3_IOC_GROUP_ADD             _IOW('f', 8,struct ext3_new_group_input)
  45 +#define        EXT3_IOC_FIEMAP                 _IOWR('f', 10, struct fiemap)
  46  #define        EXT3_IOC_GETVERSION_OLD         _IOR('v', 1, long)
  47  #define        EXT3_IOC_SETVERSION_OLD         _IOW('v', 2, long)
  48  #ifdef CONFIG_JBD_DEBUG
  49  #define EXT3_IOC_WAIT_FOR_READONLY     _IOR('f', 99, long)
  50  #endif
  51 -#define EXT3_IOC_GETRSVSZ              _IOR('f', 5, long)
  52 -#define EXT3_IOC_SETRSVSZ              _IOW('f', 6, long)
  53
  54  /*
  55   *  Mount options
  56 @@ -1117,6 +1117,8 @@ ext3_get_blocks_wrap(handle_t *handle, s
  57                 bh->b_size = (ret << inode->i_blkbits);
  58         return ret;
  59  }
  60 +extern int ext3_fiemap(struct inode *, struct file *, unsigned int,
  61 +                      unsigned long);
  62
  63
  64  #endif /* __KERNEL__ */
  65 Index: linux-2.6.18/include/linux/ext3_extents.h
  66 ===================================================================
  67 --- linux-2.6.18.orig/include/linux/ext3_extents.h
  68 +++ linux-2.6.18/include/linux/ext3_extents.h
  69 @@ -142,8 +142,9 @@ struct ext3_ext_path {
  70   * callback must return valid extent (passed or newly created)
  71   */
  72  typedef int (*ext_prepare_callback)(struct inode *, struct ext3_ext_path *,
  73 -                                       struct ext3_ext_cache *,
  74 -                                       void *);
  75 +                                   struct ext3_ext_cache *,
  76 +                                   struct ext3_extent *, void *);
  77 +#define HAVE_EXT_PREPARE_CB_EXTENT
  78
  79  #define EXT_CONTINUE   0
  80  #define EXT_BREAK      1
  81 @@ -152,6 +152,26 @@ typedef int (*ext_prepare_callback)(stru
  82
  83  #define EXT_MAX_BLOCK  0xffffffff
  84
  85 +/*
  86 + * EXT_INIT_MAX_LEN is the maximum number of blocks we can have in an
  87 + * initialized extent. This is 2^15 and not (2^16 - 1), since we use the
  88 + * MSB of ee_len field in the extent datastructure to signify if this
  89 + * particular extent is an initialized extent or an uninitialized (i.e.
  90 + * preallocated).
  91 + * EXT_UNINIT_MAX_LEN is the maximum number of blocks we can have in an
  92 + * uninitialized extent.
  93 + * If ee_len is <= 0x8000, it is an initialized extent. Otherwise, it is an
  94 + * uninitialized one. In other words, if MSB of ee_len is set, it is an
  95 + * uninitialized extent with only one special scenario when ee_len = 0x8000.
  96 + * In this case we can not have an uninitialized extent of zero length and
  97 + * thus we make it as a special case of initialized extent with 0x8000 length.
  98 + * This way we get better extent-to-group alignment for initialized extents.
  99 + * Hence, the maximum number of blocks we can have in an *initialized*
 100 + * extent is 2^15 (32768) and in an *uninitialized* extent is 2^15-1 (32767).
 101 + */
 102 +#define EXT_INIT_MAX_LEN       (1UL << 15)
 103 +#define EXT_UNINIT_MAX_LEN     (EXT_INIT_MAX_LEN - 1)
 104 +
 105  #define EXT_FLAGS_CLR_UNKNOWN  0x7  /* Flags cleared on modification */
 106  #define EXT_HDR_GEN_BITS       24
 107  #define EXT_HDR_GEN_MASK       ((1 << EXT_HDR_GEN_BITS) - 1)
 108 @@ -219,6 +239,13 @@ ext3_ext_invalidate_cache(struct inode *
 109         EXT3_I(inode)->i_cached_extent.ec_type = EXT3_EXT_CACHE_NO;
 110  }
 111
 112 +static inline int ext3_ext_is_uninitialized(struct ext3_extent *ext)
 113 +{
 114 +       /* Extent with ee_len of 0x8000 is treated as an initialized extent */
 115 +       return (le16_to_cpu(ext->ee_len) > EXT_INIT_MAX_LEN);
 116 +}
 117 +
 118 +
 119  extern int ext3_ext_search_left(struct inode *, struct ext3_ext_path *, unsigned long *, unsigned long *);
 120  extern int ext3_ext_search_right(struct inode *, struct ext3_ext_path *, unsigned long *, unsigned long *);
 121  extern int ext3_extent_tree_init(handle_t *, struct inode *);
 122 Index: linux-2.6.18/fs/ext3/extents.c
 123 ===================================================================
 124 --- linux-2.6.18.orig/fs/ext3/extents.c
 125 +++ linux-2.6.18/fs/ext3/extents.c
 126 @@ -42,7 +42,7 @@
 127  #include <linux/slab.h>
 128  #include <linux/ext3_extents.h>
 129  #include <asm/uaccess.h>
 130 -
 131 +#include "fiemap.h"
 132
 133  static handle_t *ext3_ext_journal_restart(handle_t *handle, int needed)
 134  {
 135 @@ -1477,7 +1477,7 @@ int ext3_ext_walk_space(struct inode *in
 136                 }
 137
 138                 BUG_ON(cbex.ec_len == 0);
 139 -               err = func(inode, path, &cbex, cbdata);
 140 +               err = func(inode, path, &cbex, ex, cbdata);
 141                 ext3_ext_drop_refs(path);
 142
 143                 if (err < 0)
 144 @@ -2289,6 +2289,143 @@ int ext3_ext_writepage_trans_blocks(stru
 145         return needed;
 146  }
 147
 148 +struct fiemap_internal {
 149 +       struct fiemap           *fiemap_s;
 150 +       struct fiemap_extent    fm_extent;
 151 +       size_t                  tot_mapping_len;
 152 +       char                    *cur_ext_ptr;
 153 +       int                     current_extent;
 154 +       int                     err;
 155 +};
 156 +
 157 +/*
 158 + * Callback function called for each extent to gather fiemap information.
 159 + */
 160 +int ext3_ext_fiemap_cb(struct inode *inode, struct ext3_ext_path *path,
 161 +                      struct ext3_ext_cache *newex, struct ext3_extent *ex,
 162 +                      void *data)
 163 +{
 164 +       struct fiemap_internal *fiemap_i = data;
 165 +       struct fiemap *fiemap_s = fiemap_i->fiemap_s;
 166 +       struct fiemap_extent *fm_extent = &fiemap_i->fm_extent;
 167 +       int current_extent = fiemap_i->current_extent;
 168 +       unsigned long blksize_bits = inode->i_sb->s_blocksize_bits;
 169 +
 170 +       /*
 171 +        * ext3_ext_walk_space returns a hole for extents that have not been
 172 +        * allocated yet.
 173 +        */
 174 +       if (((u64)(newex->ec_block + newex->ec_len) << blksize_bits >=
 175 +            inode->i_size) && !ext3_ext_is_uninitialized(ex) &&
 176 +           newex->ec_type == EXT3_EXT_CACHE_GAP)
 177 +               return EXT_BREAK;
 178 +
 179 +       /*
 180 +        * We only need to return number of extents.
 181 +        */
 182 +       if (fiemap_s->fm_flags & FIEMAP_FLAG_NUM_EXTENTS)
 183 +               goto count_extents;
 184 +
 185 +       if (current_extent >= fiemap_s->fm_extent_count)
 186 +               return EXT_BREAK;
 187 +
 188 +       memset(fm_extent, 0, sizeof(*fm_extent));
 189 +       fm_extent->fe_offset = (__u64)newex->ec_start << blksize_bits;
 190 +       fm_extent->fe_length = (__u64)newex->ec_len << blksize_bits;
 191 +       fiemap_i->tot_mapping_len += fm_extent->fe_length;
 192 +
 193 +       if (newex->ec_type == EXT3_EXT_CACHE_GAP)
 194 +               fm_extent->fe_flags |= FIEMAP_EXTENT_HOLE;
 195 +
 196 +       if (ext3_ext_is_uninitialized(ex))
 197 +               fm_extent->fe_flags |= (FIEMAP_EXTENT_DELALLOC |
 198 +                                       FIEMAP_EXTENT_UNMAPPED);
 199 +
 200 +       /*
 201 +        * Mark this fiemap_extent as FIEMAP_EXTENT_EOF if it's past the end
 202 +        * of file.
 203 +        */
 204 +       if ((u64)(newex->ec_block + newex->ec_len) << blksize_bits >=
 205 +                                                               inode->i_size)
 206 +               fm_extent->fe_flags |= FIEMAP_EXTENT_EOF;
 207 +
 208 +       if (!copy_to_user(fiemap_i->cur_ext_ptr, fm_extent,
 209 +                         sizeof(struct fiemap_extent))) {
 210 +               fiemap_i->cur_ext_ptr += sizeof(struct fiemap_extent);
 211 +       } else {
 212 +               fiemap_i->err = -EFAULT;
 213 +               return EXT_BREAK;
 214 +       }
 215 +
 216 +count_extents:
 217 +       fiemap_i->current_extent++;
 218 +
 219 +       /*
 220 +        * Stop if we are beyond requested mapping size but return complete last
 221 +        * extent.
 222 +        */
 223 +       if ((u64)(newex->ec_block + newex->ec_len) << blksize_bits >=
 224 +           fiemap_s->fm_length)
 225 +               return EXT_BREAK;
 226 +
 227 +       return EXT_CONTINUE;
 228 +}
 229 +
 230 +int ext3_fiemap(struct inode *inode, struct file *filp, unsigned int cmd,
 231 +               unsigned long arg)
 232 +{
 233 +       struct fiemap *fiemap_s;
 234 +       struct fiemap_internal fiemap_i;
 235 +       struct fiemap_extent *last_extent;
 236 +       ext3_fsblk_t start_blk;
 237 +       int err = 0;
 238 +
 239 +       if (!(EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL))
 240 +               return -EOPNOTSUPP;
 241 +
 242 +       fiemap_s = kmalloc(sizeof(*fiemap_s), GFP_KERNEL);
 243 +       if (fiemap_s == NULL)
 244 +               return -ENOMEM;
 245 +       if (copy_from_user(fiemap_s, (struct fiemap __user *)arg,
 246 +                          sizeof(*fiemap_s)))
 247 +               return -EFAULT;
 248 +
 249 +       if (fiemap_s->fm_flags & FIEMAP_FLAG_INCOMPAT)
 250 +               return -EOPNOTSUPP;
 251 +
 252 +       if (fiemap_s->fm_flags & FIEMAP_FLAG_SYNC)
 253 +               ext3_sync_file(filp, filp->f_dentry, 1);
 254 +
 255 +       start_blk = (fiemap_s->fm_start + inode->i_sb->s_blocksize - 1) >>
 256 +                                               inode->i_sb->s_blocksize_bits;
 257 +       fiemap_i.fiemap_s = fiemap_s;
 258 +       fiemap_i.tot_mapping_len = 0;
 259 +       fiemap_i.cur_ext_ptr = (char *)(arg + sizeof(*fiemap_s));
 260 +       fiemap_i.current_extent = 0;
 261 +       fiemap_i.err = 0;
 262 +
 263 +       /*
 264 +        * Walk the extent tree gathering extent information
 265 +        */
 266 +       mutex_lock(&EXT3_I(inode)->truncate_mutex);
 267 +       err = ext3_ext_walk_space(inode, start_blk , EXT_MAX_BLOCK - start_blk,
 268 +                                 ext3_ext_fiemap_cb, &fiemap_i);
 269 +       mutex_unlock(&EXT3_I(inode)->truncate_mutex);
 270 +       if (err)
 271 +               return err;
 272 +
 273 +       fiemap_s->fm_extent_count = fiemap_i.current_extent;
 274 +       fiemap_s->fm_length = fiemap_i.tot_mapping_len;
 275 +       if (fiemap_i.current_extent != 0 &&
 276 +           !(fiemap_s->fm_flags & FIEMAP_FLAG_NUM_EXTENTS)) {
 277 +               last_extent = &fiemap_i.fm_extent;
 278 +               last_extent->fe_flags |= FIEMAP_EXTENT_LAST;
 279 +       }
 280 +       err = copy_to_user((void *)arg, fiemap_s, sizeof(*fiemap_s));
 281 +
 282 +       return err;
 283 +}
 284 +
 285  EXPORT_SYMBOL(ext3_mark_inode_dirty);
 286  EXPORT_SYMBOL(ext3_ext_invalidate_cache);
 287  EXPORT_SYMBOL(ext3_ext_insert_extent);
 288 Index: linux-2.6.18/fs/ext3/fiemap.h
 289 ===================================================================
 290 --- /dev/null
 291 +++ linux-2.6.18/fs/ext3/fiemap.h
 292 @@ -0,0 +1,49 @@
 293 +/*
 294 + * linux/fs/ext3/fiemap.h
 295 + *
 296 + * Copyright 2008 Sun Microsystems, Inc.
 297 + *
 298 + * Author: Kalpak Shah <kalpak@clusterfs.com>
 299 + */
 300 +
 301 +#ifndef _LINUX_EXT3_FIEMAP_H
 302 +#define _LINUX_EXT3_FIEMAP_H
 303 +
 304 +struct fiemap_extent {
 305 +       __u64   fe_offset; /* offset in bytes for the start of the extent */
 306 +       __u64   fe_length; /* length in bytes for the extent */
 307 +       __u32   fe_flags;  /* returned FIEMAP_EXTENT_* flags for the extent */
 308 +       __u32   fe_lun;    /* logical device number for extent (starting at 0)*/
 309 +};
 310 +
 311 +/*
 312 + * fiemap is not ext3-specific and should be moved into fs.h eventually.
 313 + */
 314 +
 315 +struct fiemap {
 316 +       __u64   fm_start;        /* logical starting byte offset (in/out) */
 317 +       __u64   fm_length;       /* logical length of map (in/out) */
 318 +       __u32   fm_flags;        /* FIEMAP_FLAG_* flags for request (in/out) */
 319 +       __u32   fm_extent_count; /* number of extents in fm_extents (in/out) */
 320 +       __u64   fm_unused;
 321 +       struct fiemap_extent    fm_extents[0];
 322 +};
 323 +
 324 +#define        FIEMAP_FLAG_SYNC        0x00000001 /* sync file data before map */
 325 +#define        FIEMAP_FLAG_HSM_READ    0x00000002 /* get data from HSM before map */
 326 +#define        FIEMAP_FLAG_NUM_EXTENTS 0x00000004 /* return only number of extents */
 327 +#define        FIEMAP_FLAG_INCOMPAT    0xff000000 /* error for unknown flags in here */
 328 +
 329 +#define        FIEMAP_EXTENT_HOLE      0x00000001 /* has no data or space allocation */
 330 +#define        FIEMAP_EXTENT_UNWRITTEN 0x00000002 /* space allocated, but no data */
 331 +#define        FIEMAP_EXTENT_UNMAPPED  0x00000004 /* has data but no space allocation*/
 332 +#define        FIEMAP_EXTENT_ERROR     0x00000008 /* mapping error, errno in fe_start*/
 333 +#define        FIEMAP_EXTENT_NO_DIRECT 0x00000010 /* cannot access data directly */
 334 +#define        FIEMAP_EXTENT_LAST      0x00000020 /* last extent in the file */
 335 +#define        FIEMAP_EXTENT_DELALLOC  0x00000040 /* has data but not yet written,
 336 +                                           * must have EXTENT_UNKNOWN set */
 337 +#define        FIEMAP_EXTENT_SECONDARY 0x00000080 /* data (also) in secondary storage,
 338 +                                           * not in primary if EXTENT_UNKNOWN*/
 339 +#define        FIEMAP_EXTENT_EOF       0x00000100 /* if fm_start+fm_len is beyond EOF*/
 340 +
 341 +#endif /* _LINUX_EXT3_FIEMAP_H */