Building on the email that Ricky Benitez sent out earlier this month ("Tool
to view extent metadata"), I really wanted to look at the extent descriptors
while working on the O_DIRECT/fallocate/page cache issue, so I created this
patch to debugfs; it adds to the 'stat' command, and only for extent-based
files.
The output pretty much mirrors the BLOCKS output; it shows each extent
simply as a range of logical blocks to the file, plus "[uninit]" if this bit
is set.
Here's some sample output from the 'stat' command for a ~100MB file (that's
been fallocate'd and written):
===========================================================
...
atime: 0x4a679eb7 -- Wed Jul 22 16:20:23 2009
mtime: 0x4a68b71f -- Thu Jul 23 12:16:47 2009
Extents (logical blocks):
(0-25599), (25600-30719 [uninit]), (30720-61439 [uninit]),
(61440-63487 [uninit]), (63488-94207 [uninit]), (94208-96255
[uninit]), (96256-124927 [uninit]), (124928-131071 [uninit])
BLOCKS:
(IND):133120, (0-63487):34816-98303, (63488-96255):100352-133119,
(96256-124927):135168-163839, (124928-131071):165888-172031
TOTAL: 131073
===========================================================
Does this seem reasonable? Generally useful?
Signed-off-by: Curt Wohlgemuth <[email protected]>
---
--- debugfs/debugfs.c.orig 2009-06-30 20:41:09.000000000 -0700
+++ debugfs/debugfs.c 2009-07-23 13:15:32.000000000 -0700
@@ -552,6 +552,45 @@
}
+static void dump_extents(FILE *f, const char *prefix, ext2_ino_t ino)
+{
+ ext2_extent_handle_t handle;
+ struct ext2fs_extent extent;
+ int op = EXT2_EXTENT_ROOT;
+ unsigned int printed = 0;
+ errcode_t errcode;
+
+ errcode = ext2fs_extent_open(current_fs, ino, &handle);
+ if (errcode)
+ return;
+
+ fprintf(f, "%sExtents (logical blocks):\n%s", prefix, prefix);
+
+ while (1) {
+ errcode = ext2fs_extent_get(handle, op, &extent);
+
+ if (errcode)
+ break;
+
+ op = EXT2_EXTENT_NEXT;
+
+ if (!(extent.e_flags & EXT2_EXTENT_FLAGS_LEAF))
+ continue;
+
+ fprintf(f,
+ "%s(%lld-%lld%s)",
+ printed ? ", " : "",
+ extent.e_lblk,
+ extent.e_lblk + (extent.e_len - 1),
+ extent.e_flags & EXT2_EXTENT_FLAGS_UNINIT ?
+ " [uninit]" : "");
+ printed = 1;
+ }
+ if (printed)
+ fprintf(f, "\n");
+}
+
+
void internal_dump_inode(FILE *out, const char *prefix,
ext2_ino_t inode_num, struct ext2_inode *inode,
int do_dump_blocks)
@@ -649,6 +688,11 @@
if (inode->i_dtime)
fprintf(out, "%sdtime: 0x%08x -- %s", prefix, inode->i_dtime,
time_to_string(inode->i_dtime));
+
+ if (inode->i_flags & EXT4_EXTENTS_FL) {
+ dump_extents(out, prefix, inode_num);
+ }
+
if (EXT2_INODE_SIZE(current_fs->super) > EXT2_GOOD_OLD_INODE_SIZE)
internal_dump_inode_extra(out, prefix, inode_num,
(struct ext2_inode_large *) inode);
On Thu, Jul 23, 2009 at 01:36:29PM -0700, Curt Wohlgemuth wrote:
> Here's some sample output from the 'stat' command for a ~100MB file (that's
> been fallocate'd and written):
>
> ===========================================================
> ...
> atime: 0x4a679eb7 -- Wed Jul 22 16:20:23 2009
> mtime: 0x4a68b71f -- Thu Jul 23 12:16:47 2009
> Extents (logical blocks):
> (0-25599), (25600-30719 [uninit]), (30720-61439 [uninit]),
> (61440-63487 [uninit]), (63488-94207 [uninit]), (94208-96255
> [uninit]), (96256-124927 [uninit]), (124928-131071 [uninit])
> BLOCKS:
> (IND):133120, (0-63487):34816-98303, (63488-96255):100352-133119,
> (96256-124927):135168-163839, (124928-131071):165888-172031
> TOTAL: 131073
> ===========================================================
The Extents and Blocks information display redundant information, so
what I've done is to change the patch so that if the file uses
extent-based block maps, the Extents information is displayed instead
of the BLOCKS information, and it is extended to include more
information, like this:
...
atime: 0x4a6d164e:82c08b38 -- Sun Jul 26 22:51:58 2009
mtime: 0x4a6d164e:861ee098 -- Sun Jul 26 22:51:58 2009
crtime: 0x4a6d164e:82c08b38 -- Sun Jul 26 22:51:58 2009
Size of extra inode fields: 28
EXTENTS:
(0-60 #61): 342272-342332, (61-127 #67 [uninit]): 537822-537888
I've also added a dump_extents command which displays even more
information for extremely complex extent trees, so that someone
examining a filesystem can display the interior nodes of the extent
tree.
- Ted
commit ed166f9b449c53bb8df4ec277fa476d62a160367
Author: Theodore Ts'o <[email protected]>
Date: Sun Jul 26 22:29:47 2009 -0400
debugfs: Add the new command dump_extents and extent the stat command
Extend the stat command to display more detailed extent information if
the file uses extent mapping instead of displaying the block map using
the block_iterate funtion.
Add the command dump_extents which displays even more detailed
information about an inode's extent tree.
This commit is an extension of a patch from Curt Wohlgemuth.
Signed-off-by: Curt Wohlgemuth <[email protected]>
Signed-off-by: "Theodore Ts'o" <[email protected]>
diff --git a/debugfs/debug_cmds.ct b/debugfs/debug_cmds.ct
index 5355764..95dea0d 100644
--- a/debugfs/debug_cmds.ct
+++ b/debugfs/debug_cmds.ct
@@ -43,6 +43,9 @@ request do_list_dir, "List directory",
request do_stat, "Show inode information ",
show_inode_info, stat;
+request do_dump_extents, "Dump extents information ",
+ dump_extents, extents, ex;
+
request do_link, "Create directory link",
link, ln;
diff --git a/debugfs/debugfs.8.in b/debugfs/debugfs.8.in
index 4dcc48b..a0a2dbd 100644
--- a/debugfs/debugfs.8.in
+++ b/debugfs/debugfs.8.in
@@ -192,6 +192,19 @@ option is given set the owner, group and permissions information on
to match
.IR filespec .
.TP
+.I dump_extents [-n] [-l] filespec
+Dump the the extent tree of the inode
+.IR filespec .
+The
+.I -n
+flag will cause
+.I dump_extents
+to only display the interior nodes in the extent tree. The
+.I -l
+flag cause
+.I dump_extents
+to only display the leaf nodes in the extent tree.
+.TP
.I expand_dir filespec
Expand the directory
.IR filespec .
diff --git a/debugfs/debugfs.c b/debugfs/debugfs.c
index 1710aa2..6d2e853 100644
--- a/debugfs/debugfs.c
+++ b/debugfs/debugfs.c
@@ -551,6 +551,98 @@ static void dump_blocks(FILE *f, const char *prefix, ext2_ino_t inode)
fprintf(f,"\n");
}
+#define DUMP_LEAF_EXTENTS 0x01
+#define DUMP_NODE_EXTENTS 0x02
+#define DUMP_EXTENT_LEVEL 0x04
+#define DUMP_ALL_EXTENTS 0x07
+
+static void dump_extents(FILE *f, const char *prefix, ext2_ino_t ino,
+ int flags)
+{
+ ext2_extent_handle_t handle;
+ struct ext2fs_extent extent;
+ struct ext2_extent_info info;
+ int op = EXT2_EXTENT_ROOT;
+ unsigned int printed = 0;
+ errcode_t errcode;
+
+ errcode = ext2fs_extent_open(current_fs, ino, &handle);
+ if (errcode)
+ return;
+
+ fprintf(f, "%sEXTENTS:\n%s", prefix, prefix);
+
+ while (1) {
+ errcode = ext2fs_extent_get(handle, op, &extent);
+
+ if (errcode)
+ break;
+
+ op = EXT2_EXTENT_NEXT;
+
+ if (extent.e_flags & EXT2_EXTENT_FLAGS_SECOND_VISIT)
+ continue;
+
+ if (extent.e_flags & EXT2_EXTENT_FLAGS_LEAF) {
+ if ((flags & DUMP_LEAF_EXTENTS) == 0)
+ continue;
+ } else {
+ if ((flags & DUMP_NODE_EXTENTS) == 0)
+ continue;
+ }
+
+ errcode = ext2fs_extent_get_info(handle, &info);
+ if (errcode)
+ continue;
+ if (info.curr_entry == 1 && (flags & DUMP_EXTENT_LEVEL)) {
+ fprintf(f, "%s[L%d/%d %d/%d] ",
+ printed ? ", " : "",
+ info.curr_level, info.max_depth,
+ info.num_entries, info.max_entries);
+ printed = 0;
+ }
+
+ if (!(extent.e_flags & EXT2_EXTENT_FLAGS_LEAF)) {
+ if (extent.e_flags & EXT2_EXTENT_FLAGS_SECOND_VISIT)
+ continue;
+
+ fprintf(f, "%s(NODE #%d, %lld-%lld #%u, blk %lld)",
+ printed ? ", " : "",
+ info.curr_entry,
+ extent.e_lblk,
+ extent.e_lblk + (extent.e_len - 1),
+ extent.e_len,
+ extent.e_pblk);
+ printed = 1;
+ continue;
+ }
+
+ if (extent.e_len == 0)
+ continue;
+ else if (extent.e_len == 1)
+ fprintf(f,
+ "%s(%lld%s): %lld",
+ printed ? ", " : "",
+ extent.e_lblk,
+ extent.e_flags & EXT2_EXTENT_FLAGS_UNINIT ?
+ " [uninit]" : "",
+ extent.e_pblk);
+ else
+ fprintf(f,
+ "%s(%lld-%lld #%u%s): %lld-%lld",
+ printed ? ", " : "",
+ extent.e_lblk,
+ extent.e_lblk + (extent.e_len - 1),
+ extent.e_len,
+ extent.e_flags & EXT2_EXTENT_FLAGS_UNINIT ?
+ " [uninit]" : "",
+ extent.e_pblk,
+ extent.e_pblk + (extent.e_len - 1));
+ printed = 1;
+ }
+ if (printed)
+ fprintf(f, "\n");
+}
void internal_dump_inode(FILE *out, const char *prefix,
ext2_ino_t inode_num, struct ext2_inode *inode,
@@ -671,9 +763,12 @@ void internal_dump_inode(FILE *out, const char *prefix,
}
fprintf(out, "%sDevice major/minor number: %02d:%02d (hex %02x:%02x)\n",
devnote, major, minor, major, minor);
+ } else if (do_dump_blocks) {
+ if (inode->i_flags & EXT4_EXTENTS_FL)
+ dump_extents(out, prefix, inode_num, DUMP_LEAF_EXTENTS);
+ else
+ dump_blocks(out, prefix, inode_num);
}
- else if (do_dump_blocks)
- dump_blocks(out, prefix, inode_num);
}
static void dump_inode(ext2_ino_t inode_num, struct ext2_inode *inode)
@@ -716,6 +811,59 @@ void do_stat(int argc, char *argv[])
return;
}
+void do_dump_extents(int argc, char *argv[])
+{
+ struct ext2_inode inode;
+ ext2_ino_t ino;
+ FILE *out;
+ int c, flags = 0;
+
+ reset_getopt();
+ while ((c = getopt(argc, argv, "nl")) != EOF) {
+ switch (c) {
+ case 'n':
+ flags |= DUMP_NODE_EXTENTS;
+ break;
+ case 'l':
+ flags |= DUMP_LEAF_EXTENTS;
+ break;
+ }
+ }
+
+ if (argc != optind+1) {
+ print_usage:
+ com_err(0, 0, "Usage: dump_extents [-n] [-l] file");
+ return;
+ }
+
+ if (flags == 0)
+ flags = DUMP_ALL_EXTENTS;
+ else if (flags == DUMP_EXTENT_LEVEL)
+ flags |= DUMP_NODE_EXTENTS;
+ flags |= DUMP_EXTENT_LEVEL;
+
+ if (check_fs_open(argv[0]))
+ return;
+
+ ino = string_to_inode(argv[optind]);
+ if (ino == 0)
+ return;
+
+ if (debugfs_read_inode(ino, &inode, argv[0]))
+ return;
+
+ if ((inode.i_flags & EXT4_EXTENTS_FL) == 0) {
+ fprintf(stderr, "%s: does not uses extent block maps\n",
+ argv[optind]);
+ return;
+ }
+
+ out = open_pager();
+ dump_extents(out, "", ino, flags);
+ close_pager(out);
+ return;
+}
+
void do_chroot(int argc, char *argv[])
{
ext2_ino_t inode;
Theodore Tso wrote:
...
> The Extents and Blocks information display redundant information, so
> what I've done is to change the patch so that if the file uses
> extent-based block maps, the Extents information is displayed instead
> of the BLOCKS information, and it is extended to include more
> information, like this:
>
> ...
> atime: 0x4a6d164e:82c08b38 -- Sun Jul 26 22:51:58 2009
> mtime: 0x4a6d164e:861ee098 -- Sun Jul 26 22:51:58 2009
> crtime: 0x4a6d164e:82c08b38 -- Sun Jul 26 22:51:58 2009
> Size of extra inode fields: 28
> EXTENTS:
> (0-60 #61): 342272-342332, (61-127 #67 [uninit]): 537822-537888
Perhaps a bit late, but I find the "#<extent length>" reporting very
unintuitive. Doesn't "#X" usually imply ordering? How about:
(0-60 [61b]): 342272-342332, (61-127 [67b,uninit]): 537822-537888
... or maybe some other ideas. IMHO it's all a bit hard to read anyway
unless it's printed in a table format ala the new filefrag output.
-Eric
On Sun, Jul 26, 2009 at 10:18:02PM -0500, Eric Sandeen wrote:
>
> Perhaps a bit late, but I find the "#<extent length>" reporting very
> unintuitive. Doesn't "#X" usually imply ordering? How about:
>
> (0-60 [61b]): 342272-342332, (61-127 [67b,uninit]): 537822-537888
>
> ... or maybe some other ideas. IMHO it's all a bit hard to read anyway
> unless it's printed in a table format ala the new filefrag output.
That's a good idea. I've switched it to use a table format:
Level Entries Logical Physical Length Flags
0/ 0 1/ 2 0 - 60 342272 - 342332 61
0/ 0 2/ 2 61 - 127 537822 - 537888 67 Uninit
and
Level Entries Logical Physical Length Flags
0/ 2 1/ 4 1 - 274323 24844 274323
1/ 2 1/ 84 1 - 387 8458 387
2/ 2 1/ 84 1 - 1 10242 - 10242 1
2/ 2 2/ 84 4 - 4 10245 - 10245 1
2/ 2 3/ 84 8 - 8 10249 - 10249 1
I've also simplified the debugfs stat output so it looks more like the
"BLOCKS" output:
mtime: 0x4a6d164e:861ee098 -- Sun Jul 26 22:51:58 2009
crtime: 0x4a6d164e:82c08b38 -- Sun Jul 26 22:51:58 2009
Size of extra inode fields: 28
EXTENTS:
(0-60): 342272-342332, (61-127 [uninit]): 537822-537888
Hopefully that's easier for everyone to understand.
- Ted