Whamcloud - gitweb
c1c8dc7ef9cf6fd0e29382c64e1198ff987ec311
[fs/lustre-release.git] / ldiskfs / kernel_patches / patches / ext3-fiemap-2.6-sles10.patch
1 A large part of this code is from the generic VFS code in fs/ioctl.c in the
2 upstream kernel.
3
4 Index: linux-2.6.16.60-0.23/fs/ext3/ioctl.c
5 ===================================================================
6 --- linux-2.6.16.60-0.23.orig/fs/ext3/ioctl.c
7 +++ linux-2.6.16.60-0.23/fs/ext3/ioctl.c
8 @@ -15,7 +15,159 @@
9  #include <linux/time.h>
10  #include <asm/uaccess.h>
11  #include <linux/namei.h>
12 +#include "fiemap.h"
13  
14 +/* So that the fiemap access checks can't overflow on 32 bit machines. */
15 +#define FIEMAP_MAX_EXTENTS     (UINT_MAX / sizeof(struct fiemap_extent))
16 +
17 +/**
18 + * fiemap_fill_next_extent - Fiemap helper function
19 + * @fieinfo:   Fiemap context passed into ->fiemap
20 + * @logical:   Extent logical start offset, in bytes
21 + * @phys:      Extent physical start offset, in bytes
22 + * @len:       Extent length, in bytes
23 + * @flags:     FIEMAP_EXTENT flags that describe this extent
24 + * @lun:       LUN on which this extent resides
25 + *
26 + * Called from file system ->fiemap callback. Will populate extent
27 + * info as passed in via arguments and copy to user memory. On
28 + * success, extent count on fieinfo is incremented.
29 + *
30 + * Returns 0 on success, -errno on error, 1 if this was the last
31 + * extent that will fit in user array.
32 + */
33 +#define SET_UNKNOWN_FLAGS      (FIEMAP_EXTENT_DELALLOC)
34 +#define SET_NO_DIRECT_FLAGS    (FIEMAP_EXTENT_DATA_COMPRESSED  \
35 +                               |FIEMAP_EXTENT_DATA_ENCRYPTED   \
36 +                               |FIEMAP_EXTENT_NET)
37 +#define SET_NOT_ALIGNED_FLAGS  (FIEMAP_EXTENT_DATA_TAIL|FIEMAP_EXTENT_DATA_INLINE)
38 +int fiemap_fill_next_extent(struct fiemap_extent_info *fieinfo, u64 logical,
39 +                           u64 phys, u64 len, u32 flags, dev_t dev)
40 +{
41 +       struct fiemap_extent extent = { 0 };
42 +       struct fiemap_extent *dest = fieinfo->fi_extents_start;
43 +
44 +       /* only count the extents */
45 +       if (fieinfo->fi_extents_max == 0) {
46 +                fieinfo->fi_extents_mapped++;
47 +                return (flags & FIEMAP_EXTENT_LAST) ? 1 : 0;
48 +       }
49 +
50 +       if (fieinfo->fi_extents_mapped >= fieinfo->fi_extents_max)
51 +                return 1;
52 +
53 +       if (flags & SET_UNKNOWN_FLAGS)
54 +                flags |= FIEMAP_EXTENT_UNKNOWN;
55 +       if (flags & SET_NO_DIRECT_FLAGS)
56 +                flags |= FIEMAP_EXTENT_NO_DIRECT;
57 +       if (flags & SET_NOT_ALIGNED_FLAGS)
58 +                flags |= FIEMAP_EXTENT_NOT_ALIGNED;
59 +
60 +       extent.fe_logical = logical;
61 +       extent.fe_physical = phys;
62 +       extent.fe_length = len;
63 +       extent.fe_flags = flags;
64 +       extent.fe_device = new_encode_dev(dev);
65 +
66 +       dest += fieinfo->fi_extents_mapped;
67 +       if (copy_to_user(dest, &extent, sizeof(extent)))
68 +                return -EFAULT;
69 +
70 +       fieinfo->fi_extents_mapped++;
71 +       if (fieinfo->fi_extents_mapped == fieinfo->fi_extents_max)
72 +                return 1;
73 +
74 +       return (flags & FIEMAP_EXTENT_LAST) ? 1 : 0;
75 +}
76 +
77 +static int fiemap_check_ranges(struct super_block *sb,
78 +                              u64 start, u64 len, u64 *new_len)
79 +{
80 +       *new_len = len;
81 +
82 +       if (len == 0)
83 +               return -EINVAL;
84 +
85 +       if (start > sb->s_maxbytes)
86 +               return -EFBIG;
87 +
88 +       /*
89 +        * Shrink request scope to what the fs can actually handle.
90 +        */
91 +       if ((len > sb->s_maxbytes) ||
92 +           (sb->s_maxbytes - len) < start)
93 +               *new_len = sb->s_maxbytes - start;
94 +
95 +       return 0;
96 +}
97 +
98 +/*
99 + * fiemap_check_flags - check validity of requested flags for fiemap
100 + * @fieinfo:   Fiemap context passed into ->fiemap
101 + * @fs_flags:  Set of fiemap flags that the file system understands
102 + *
103 + * Called from file system ->fiemap callback. This will compute the
104 + * intersection of valid fiemap flags and those that the fs supports. That
105 + * value is then compared against the user supplied flags. In case of bad user
106 + * flags, the invalid values will be written into the fieinfo structure, and
107 + * -EBADR is returned, which tells ioctl_fiemap() to return those values to
108 + * userspace. For this reason, a return code of -EBADR should be preserved.
109 + *
110 + * Returns 0 on success, -EBADR on bad flags.
111 +*/
112 +int fiemap_check_flags(struct fiemap_extent_info *fieinfo, u32 fs_flags)
113 +{
114 +       u32 incompat_flags;
115 +
116 +       incompat_flags = fieinfo->fi_flags & ~(FIEMAP_FLAGS_COMPAT & fs_flags);
117 +       if (incompat_flags) {
118 +               fieinfo->fi_flags = incompat_flags;
119 +               return -EBADR;
120 +       }
121 +
122 +       return 0;
123 +}
124 +
125 +int ioctl_fiemap(struct inode *inode, struct file *filp, unsigned long arg)
126 +{
127 +       struct fiemap fiemap;
128 +       u64 len;
129 +       struct fiemap_extent_info fieinfo = {0, };
130 +       struct super_block *sb = inode->i_sb;
131 +       int error = 0;
132 +
133 +       if (copy_from_user(&fiemap, (struct fiemap __user *) arg,
134 +                            sizeof(struct fiemap)))
135 +                return -EFAULT;
136 +
137 +       if (fiemap.fm_extent_count > FIEMAP_MAX_EXTENTS)
138 +                return -EINVAL;
139 +
140 +       error = fiemap_check_ranges(sb, fiemap.fm_start, fiemap.fm_length,
141 +                                   &len);
142 +       if (error)
143 +               return error;
144 +
145 +       fieinfo.fi_flags = fiemap.fm_flags;
146 +       fieinfo.fi_extents_max = fiemap.fm_extent_count;
147 +       fieinfo.fi_extents_start = (struct fiemap_extent *)(arg + sizeof(fiemap));
148 +
149 +       if (fiemap.fm_extent_count != 0 &&
150 +           !access_ok(VERIFY_WRITE, (void *)arg,
151 +                      offsetof(typeof(fiemap), fm_extents[fiemap.fm_extent_count])))
152 +                      return -EFAULT;
153 +
154 +       if (fieinfo.fi_flags & FIEMAP_FLAG_SYNC)
155 +               filemap_write_and_wait(inode->i_mapping);
156 +
157 +       error = ext3_fiemap(inode, &fieinfo, fiemap.fm_start, len);
158 +       fiemap.fm_flags = fieinfo.fi_flags;
159 +       fiemap.fm_mapped_extents = fieinfo.fi_extents_mapped;
160 +       if (copy_to_user((char *)arg, &fiemap, sizeof(fiemap)))
161 +               error = -EFAULT;
162 +
163 +       return error;
164 +}
165  
166  int ext3_ioctl (struct inode * inode, struct file * filp, unsigned int cmd,
167                 unsigned long arg)
168 @@ -262,7 +414,9 @@ flags_err:
169  
170                 return err;
171         }
172 -
173 +       case EXT3_IOC_FIEMAP: {
174 +               return ioctl_fiemap(inode, filp, arg);
175 +       }
176  
177         default:
178                 return -ENOTTY;
179 Index: linux-2.6.16.60-0.23/include/linux/ext3_fs.h
180 ===================================================================
181 --- linux-2.6.16.60-0.23.orig/include/linux/ext3_fs.h
182 +++ linux-2.6.16.60-0.23/include/linux/ext3_fs.h
183 @@ -260,15 +260,19 @@ struct ext3_new_group_data {
184  #define        EXT3_IOC_SETFLAGS               _IOW('f', 2, long)
185  #define        EXT3_IOC_GETVERSION             _IOR('f', 3, long)
186  #define        EXT3_IOC_SETVERSION             _IOW('f', 4, long)
187 +#define EXT3_IOC_GETRSVSZ              _IOR('f', 5, long)
188 +#define EXT3_IOC_SETRSVSZ              _IOW('f', 6, long)
189  #define EXT3_IOC_GROUP_EXTEND          _IOW('f', 7, unsigned long)
190  #define EXT3_IOC_GROUP_ADD             _IOW('f', 8,struct ext3_new_group_input)
191 +#define EXT3_IOC_FIEMAP                        _IOWR('f', 10, struct fiemap)
192  #define        EXT3_IOC_GETVERSION_OLD         _IOR('v', 1, long)
193  #define        EXT3_IOC_SETVERSION_OLD         _IOW('v', 2, long)
194  #ifdef CONFIG_JBD_DEBUG
195  #define EXT3_IOC_WAIT_FOR_READONLY     _IOR('f', 99, long)
196  #endif
197 -#define EXT3_IOC_GETRSVSZ              _IOR('f', 5, long)
198 -#define EXT3_IOC_SETRSVSZ              _IOW('f', 6, long)
199 +
200 +/* FIEMAP flags supported by ext3 */
201 +#define EXT3_FIEMAP_FLAGS_COMPAT (FIEMAP_FLAG_SYNC)
202  
203  /*
204   *  Mount options
205 @@ -996,6 +1000,9 @@ extern int ext3_writepage_trans_blocks(s
206  /* ioctl.c */
207  extern int ext3_ioctl (struct inode *, struct file *, unsigned int,
208                        unsigned long);
209 +struct fiemap_extent_info;
210 +extern int ext3_fiemap(struct inode *, struct fiemap_extent_info *, __u64,
211 +                      __u64);
212  
213  /* namei.c */
214  extern int ext3_orphan_add(handle_t *, struct inode *);
215 Index: linux-2.6.16.60-0.23/include/linux/ext3_extents.h
216 ===================================================================
217 --- linux-2.6.16.60-0.23.orig/include/linux/ext3_extents.h
218 +++ linux-2.6.16.60-0.23/include/linux/ext3_extents.h
219 @@ -170,7 +170,10 @@ struct ext3_extents_helpers {
220   */
221  typedef int (*ext_prepare_callback)(struct ext3_extents_tree *,
222                                     struct ext3_ext_path *,
223 -                                   struct ext3_ext_cache *);
224 +                                   struct ext3_ext_cache *,
225 +                                   struct ext3_extent *);
226 +
227 +#define HAVE_EXT_PREPARE_CB_EXTENT
228  
229  #define EXT_CONTINUE   0
230  #define EXT_BREAK      1
231 @@ -179,6 +182,25 @@ typedef int (*ext_prepare_callback)(stru
232  
233  #define EXT_MAX_BLOCK  0xffffffff
234  
235 +/*
236 + * EXT_INIT_MAX_LEN is the maximum number of blocks we can have in an
237 + * initialized extent. This is 2^15 and not (2^16 - 1), since we use the
238 + * MSB of ee_len field in the extent datastructure to signify if this
239 + * particular extent is an initialized extent or an uninitialized (i.e.
240 + * preallocated).
241 + * EXT_UNINIT_MAX_LEN is the maximum number of blocks we can have in an
242 + * uninitialized extent.
243 + * If ee_len is <= 0x8000, it is an initialized extent. Otherwise, it is an
244 + * uninitialized one. In other words, if MSB of ee_len is set, it is an
245 + * uninitialized extent with only one special scenario when ee_len = 0x8000.
246 + * In this case we can not have an uninitialized extent of zero length and
247 + * thus we make it as a special case of initialized extent with 0x8000 length.
248 + * This way we get better extent-to-group alignment for initialized extents.
249 + * Hence, the maximum number of blocks we can have in an *initialized*
250 + * extent is 2^15 (32768) and in an *uninitialized* extent is 2^15-1 (32767).
251 + */
252 +#define        EXT_INIT_MAX_LEN        (1UL << 15)
253 +#define        EXT_UNINIT_MAX_LEN      (EXT_INIT_MAX_LEN - 1)
254  
255  #define EXT_FIRST_EXTENT(__hdr__) \
256         ((struct ext3_extent *) (((char *) (__hdr__)) +         \
257 @@ -242,6 +264,12 @@ struct ext3_extent_tree_stats {
258         int leaf_num;
259  };
260  
261 +static inline int ext3_ext_is_uninitialized(struct ext3_extent *ext)
262 +{
263 +       /* Extent with ee_len of 0x8000 is treated as an initialized extent */
264 +       return (le16_to_cpu(ext->ee_len) > EXT_INIT_MAX_LEN);
265 +}
266 +
267  extern int ext3_ext_search_left(struct ext3_extents_tree *, struct ext3_ext_path *, unsigned long *, unsigned long *);
268  extern int ext3_ext_search_right(struct ext3_extents_tree *, struct ext3_ext_path *, unsigned long *, unsigned long *);
269  extern void ext3_init_tree_desc(struct ext3_extents_tree *, struct inode *);
270 Index: linux-2.6.16.60-0.23/fs/ext3/extents.c
271 ===================================================================
272 --- linux-2.6.16.60-0.23.orig/fs/ext3/extents.c
273 +++ linux-2.6.16.60-0.23/fs/ext3/extents.c
274 @@ -41,6 +41,7 @@
275  #include <linux/string.h>
276  #include <linux/slab.h>
277  #include <linux/ext3_extents.h>
278 +#include "fiemap.h"
279  #include <asm/uaccess.h>
280  
281  
282 @@ -1488,7 +1489,7 @@ int ext3_ext_walk_space(struct ext3_exte
283  
284                 EXT_ASSERT(cbex.ec_len > 0);
285                 EXT_ASSERT(path[depth].p_hdr);
286 -               err = func(tree, path, &cbex);
287 +               err = func(tree, path, &cbex, ex);
288                 ext3_ext_drop_refs(path);
289  
290                 if (err < 0)
291 @@ -2502,7 +2503,109 @@ int ext3_ext_calc_blockmap_metadata(stru
292         ext3_init_tree_desc(&tree, inode);
293         return ext3_ext_calc_metadata_amount(&tree, blocks);
294  }
295 -       
296 +
297 +/*
298 + * Callback function called for each extent to gather FIEMAP information.
299 + */
300 +int ext3_ext_fiemap_cb(struct ext3_extents_tree *tree,
301 +                      struct ext3_ext_path *path, struct ext3_ext_cache *newex,
302 +                      struct ext3_extent *ex)
303 +{
304 +       struct fiemap_extent_info *fieinfo = (struct fiemap_extent_info *)tree->private;
305 +       struct inode *inode = tree->inode;
306 +       unsigned long blksize_bits = inode->i_sb->s_blocksize_bits;
307 +       __u64   logical;
308 +       __u64   physical;
309 +       __u64   length;
310 +       __u32   flags = 0;
311 +       int     error;
312 +
313 +       logical =  (__u64)newex->ec_block << blksize_bits;
314 +
315 +       if (newex->ec_type == EXT3_EXT_CACHE_GAP) {
316 +               pgoff_t offset;
317 +               struct page *page;
318 +               struct buffer_head *bh = NULL;
319 +
320 +               offset = logical >> PAGE_SHIFT;
321 +               page = find_get_page(inode->i_mapping, offset);
322 +               if (!page || !page_has_buffers(page))
323 +                       return EXT_CONTINUE;
324 +
325 +               bh = page_buffers(page);
326 +
327 +               if (!bh)
328 +                       return EXT_CONTINUE;
329 +
330 +               if (buffer_delay(bh)) {
331 +                       flags |= FIEMAP_EXTENT_DELALLOC;
332 +                       page_cache_release(page);
333 +               } else {
334 +                       page_cache_release(page);
335 +                       return EXT_CONTINUE;
336 +               }
337 +       }
338 +
339 +       physical = (__u64)newex->ec_start << blksize_bits;
340 +       length =   (__u64)newex->ec_len << blksize_bits;
341 +
342 +       if (ex && ext3_ext_is_uninitialized(ex))
343 +               flags |= FIEMAP_EXTENT_UNWRITTEN;
344 +
345 +       /*
346 +        * If this extent reaches EXT_MAX_BLOCK, it must be last.
347 +        *
348 +        * Or if ext3_ext_next_allocated_block is EXT_MAX_BLOCK,
349 +        * this indicates no more allocated blocks.
350 +        *
351 +        * XXX this might miss a single-block extent at EXT_MAX_BLOCK
352 +        */
353 +       if (logical + length - 1 == EXT_MAX_BLOCK ||
354 +           ext3_ext_next_allocated_block(path) == EXT_MAX_BLOCK)
355 +               flags |= FIEMAP_EXTENT_LAST;
356 +
357 +       error = fiemap_fill_next_extent(fieinfo, logical, physical,
358 +                                       length, flags, inode->i_sb->s_dev);
359 +       if (error < 0)
360 +               return error;
361 +       if (error == 1)
362 +               return EXT_BREAK;
363 +
364 +       return EXT_CONTINUE;
365 +}
366 +
367 +int ext3_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
368 +               __u64 start, __u64 len)
369 +{
370 +       struct ext3_extents_tree tree;
371 +       unsigned long start_blk;
372 +       unsigned long len_blks;
373 +       int error = 0;
374 +
375 +       if (!(EXT3_I(inode)->i_flags & EXT3_EXTENTS_FL))
376 +               return -EOPNOTSUPP;
377 +
378 +       if (fiemap_check_flags(fieinfo, EXT3_FIEMAP_FLAGS_COMPAT))
379 +               return -EBADR;
380 +
381 +       start_blk = start >> inode->i_sb->s_blocksize_bits;
382 +       len_blks = (len + inode->i_sb->s_blocksize - 1) >> inode->i_sb->s_blocksize_bits;
383 +
384 +       ext3_init_tree_desc(&tree, inode);
385 +       tree.private = fieinfo;
386 +
387 +       /*
388 +        * Walk the extent tree gathering extent information.
389 +        * ext3_ext_fiemap_cb will push extents back to user.
390 +        */
391 +       down(&EXT3_I(inode)->truncate_sem);
392 +       error = ext3_ext_walk_space(&tree, start_blk, len_blks,
393 +                                   ext3_ext_fiemap_cb);
394 +       up(&EXT3_I(inode)->truncate_sem);
395 +
396 +       return error;
397 +}
398 +
399  EXPORT_SYMBOL(ext3_init_tree_desc);
400  EXPORT_SYMBOL(ext3_mark_inode_dirty);
401  EXPORT_SYMBOL(ext3_ext_invalidate_cache);
402 Index: linux-2.6.16.60-0.23/fs/ext3/fiemap.h
403 ===================================================================
404 --- /dev/null
405 +++ linux-2.6.16.60-0.23/fs/ext3/fiemap.h
406 @@ -0,0 +1,84 @@
407 +/*
408 + * FIEMAP ioctl infrastructure.
409 + *
410 + * Copyright 2008 Sun Microsystems, Inc.
411 + *
412 + * Author: Kalpak Shah <kalpak.shah@sun.com>
413 + *         Andreas Dilger <adilger@sun.com>
414 + */
415 +
416 +#ifndef _LINUX_EXT3_FIEMAP_H
417 +#define _LINUX_EXT3_FIEMAP_H
418 +
419 +struct fiemap_extent {
420 +       __u64 fe_logical;  /* logical offset in bytes for the start of
421 +                           * the extent from the beginning of the file */
422 +       __u64 fe_physical; /* physical offset in bytes for the start
423 +                           * of the extent from the beginning of the disk */
424 +       __u64 fe_length;   /* length in bytes for this extent */
425 +       __u32 fe_flags;    /* FIEMAP_EXTENT_* flags for this extent */
426 +       __u32 fe_device;   /* device number for this extent */
427 +};
428 +
429 +struct fiemap {
430 +       __u64 fm_start;  /* logical offset (inclusive) at
431 +                                * which to start mapping (in) */
432 +       __u64 fm_length;        /* logical length of mapping which
433 +                                * userspace wants (in) */
434 +       __u32 fm_flags;  /* FIEMAP_FLAG_* flags for request (in/out) */
435 +       __u32 fm_mapped_extents;/* number of extents that were mapped (out) */
436 +       __u32 fm_extent_count;  /* size of fm_extents array (in) */
437 +       __u32 fm_reserved;
438 +       struct fiemap_extent fm_extents[0]; /* array of mapped extents (out) */
439 +};
440 +
441 +/*
442 + * FIEMAP helper definition.
443 + */
444 +struct fiemap_extent_info {
445 +       unsigned int    fi_flags;               /* Flags as passed from user */
446 +       unsigned int    fi_extents_mapped;      /* Number of mapped extents */
447 +       unsigned int    fi_extents_max;         /* Size of fiemap_extent array*/
448 +       struct fiemap_extent *fi_extents_start; /* Start of fiemap_extent array */
449 +};
450 +
451 +int fiemap_check_flags(struct fiemap_extent_info *fieinfo, u32 fs_flags);
452 +int fiemap_fill_next_extent(struct fiemap_extent_info *info, u64 logical,
453 +                           u64 phys, u64 len, u32 flags, u32 lun);
454 +
455 +#define        FIEMAP_MAX_OFFSET       (~0ULL)
456 +
457 +#define        FIEMAP_FLAG_SYNC        0x00000001 /* sync file data before map */
458 +#define        FIEMAP_FLAG_XATTR       0x00000002 /* map extended attribute tree */
459 +
460 +/* ldiskfs only supports FLAG_SYNC flag currently */
461 +#define FIEMAP_FLAGS_COMPAT (FIEMAP_FLAG_SYNC | FIEMAP_FLAG_XATTR)
462 +
463 +
464 +#define FIEMAP_EXTENT_LAST             0x00000001 /* Last extent in file. */
465 +#define FIEMAP_EXTENT_UNKNOWN          0x00000002 /* Data location unknown. */
466 +#define FIEMAP_EXTENT_DELALLOC         0x00000004 /* Location still pending.
467 +                                                  * Sets EXTENT_UNKNOWN. */
468 +#define FIEMAP_EXTENT_NO_DIRECT                0x00000008 /* Data mapping undefined */
469 +#define FIEMAP_EXTENT_SECONDARY                0x00000010 /* Data copied offline. May
470 +                                                  * set EXTENT_NO_DIRECT. */
471 +#define FIEMAP_EXTENT_NET              0x00000020 /* Data stored remotely.
472 +                                                  * Sets EXTENT_NO_DIRECT. */
473 +#define FIEMAP_EXTENT_DATA_COMPRESSED  0x00000040 /* Data is compressed by fs.
474 +                                                  * Sets EXTENT_NO_DIRECT. */
475 +#define FIEMAP_EXTENT_DATA_ENCRYPTED   0x00000080 /* Data is encrypted by fs.
476 +                                                  * Sets EXTENT_NO_DIRECT. */
477 +#define FIEMAP_EXTENT_NOT_ALIGNED      0x00000100 /* Extent offsets may not be
478 +                                                  * block aligned. */
479 +#define FIEMAP_EXTENT_DATA_INLINE      0x00000200 /* Data mixed with metadata.
480 +                                                  * Sets EXTENT_NOT_ALIGNED.*/
481 +#define FIEMAP_EXTENT_DATA_TAIL                0x00000400 /* Multiple files in block.
482 +                                                  * Sets EXTENT_NOT_ALIGNED.*/
483 +#define FIEMAP_EXTENT_UNWRITTEN                0x00000800 /* Space allocated, but
484 +                                                  * no data (i.e. zero). */
485 +#define FIEMAP_EXTENT_MERGED           0x00001000 /* File does not natively
486 +                                                  * support extents. Result
487 +                                                  * merged for efficiency. */
488 +
489 +#endif /* _LINUX_EXT3_FIEMAP_H */
490 +