It turns out that it's very hard to calculate overheads in the face of
clustered allocation (bigalloc). This is because multiple metadata
blocks from different block groups can end up in the same allocation
cluster. Calculating the exact overhead requires O(all block bitmaps)
in memory, or O(number of block groups**2) in time. So we will
calculate this at mkfs time and stash it in the superblock.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This adds the superblock fields needed so that dumpe2fs works and the
code points and renames the superblock fields from describing
fragments to clusters.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This patch adds support for detecting the new 'quota' feature in ext4.
The patch reserves code points for usr and group quota inodes and also
for the feature flag EXT4_FEATURE_RO_COMPAT_QUOTA.
Signed-off-by: Aditya Kali <adityakali@google.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
User namespace xattrs are generally useful, and I think extN
is the only filesystem requiring a special mount option to
enable them, when xattrs are otherwise available. So this
change sets that mount option into the defaults, via a
mke2fs.conf option.
Note that if xattrs are config'd off, this will lead to a
mostly-harmless:
EXT4-fs (sdc1): (no)user_xattr options not supported
message at mount time...
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The forced fsck often comes at unexpected and inopportune moments,
and even enterprise customers are often caught by surprise when
this happens. Because a filesystem with an error condition will
be marked as requiring fsck anyway, I submit that the time-based
and mount-based checks are not particularly useful, and that
administrators can schedule fscks on their own time, or tune2fs
the enforced intervals if they so choose. This patch disables the
intervals by default, and I've added a new mkfs.conf option to
turn on the old behavior of random, unexpected, time-consuming
fscks at boot time. ;)
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
After debian bug #192277, debian/rules started making a symlink
to com_err.h in /usr/include. Now I have Fedora bug #550889
for the same issue, and perhaps it's time to make this link
by default, rather than fixing it up in packaging steps?
[ Changed by tytso to remove the explicit -s option; this will default
to creating a hard link by default, which slightly faster. If
people want to use symlinks for all links during the install
process, they can use configure option --enable-symlink-install.
The reason for this change is that some file systems, like AFS,
don't support symlinks, and AFS users complain when they can't build
or install into AFS. So I don't want to use symlinks
unconditionally without a way of switching things back and forth,
and it's easier if we just make all links made during the install
process to be hard links or sym links. ]
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
If the file system has a blocksize less than 64k, then don't use the
extended rec_len encoding, to be consistent with what the kernel will
do.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Flags used during RHEL/Fedora builds lead to a couple type-punning
warnings:
recovery.c: In function 'do_one_pass':
recovery.c:539: warning: dereferencing type-punned pointer will break strict-aliasing rules
./csum.c: In function 'print_csum':
./csum.c:170: warning: dereferencing type-punned pointer will break strict-aliasing rules
The two changes below fix this up.
Note that the csum test binary output changes slightly, but this does
not break any tests.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The development branch of e2fsprogs already has a code point assigned
in conflict with EXT2_FLAG_DIRECT_IO. Fix this.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Allocate various memory structures to be properly aligned to avoid
needing to use a bounce buffer when doing direct I/O read/writes.
This should also help on FreeBSD systems which require aligned buffers
unconditionally.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This adds the basic support for Direct I/O to unix_io.c, and adds a
new flag EXT_FLAG_DIRECT_IO which can be passed to ext2fs_open() or
ext2fs_open2() to request Direct I/O support.
Note that device mapper devices in Linux don't support Direct I/O, and
in some circumstances using Direct I/O can actually make performance
*worse*!
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The test now checks to make sure the superblock fields are correctly
aligned and prints them out so they can be manually checked to make
sure they are where we expect them to be.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Add superblock fields which track where and when the first and most
recent file system errors occured. These fields are displayed by
dumpe2fs and cleared by e2fsck.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
We also support for byte-swapping the Next3 fields, although the
current Next3 implementation doesn't support big-endian systems.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reserve the EXT4_FEATURE_INCOMPAT_DIRDATA feature flag for adding
extra file data in ext2_dir_entry_2 entries.
This changes the on-disk layout in the following way.
Firstly, the ext2_dir_entry_2 file_type field now has a mask: that
limits the "filetype" information to the low 4 bits of this field.
Since these values are sequentially assigned, this allows for up to 7
more filetypes to be assigned. When reading the "filetype" field, the
high 4 bits should be masked off when converting to DT_* filetypes for
userspace.
The high 4 bits of "filetype" are used as a bitmask to register up to
4 different "extended" directory entry fields. Extended data fields
are packed without alignment into the directory entry after the "name"
field in order of increasing bitmask value, for each field where bit
is set. In order to avoid the need to "understand" each of the
extended fields, the first byte of each extended data field holds the
size of that data field (including the size itself), so they can be
skipped if not understood. For fields that change the semantics of
the filesystem it is expected that a separate ROCOMPAT or INCOMPAT
field is registered.
There is a single dirent data type defined currently, for Lustre:
which holds a 128-bit file identifier. It is expected that if there
are 64-bit inode values that this will be assigned the 0x20 value.
Should a need ever arise to use all 4 of the extended dirent data
fields, it would be possible to keep the last bit (0x80) for use as a
multiplexor that stores a 1-byte aggregate data size, then a series of
"<u8_size><u8_type><data>" records in the last extended data record.
It is not expected that this will actually be needed in the lifetime
of ext4.
Signed-off-by: Andreas Dilger <adilger@sun.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reserve the EXT4_INCOMPAT_EA_INODE feature flag for use with
large extended attributes that are stored in a separate inode.
This changes the on-disk format in several ways:
First, replace the e_value_block field with e_value_inum, so that
an xattr entry can reference an external inode. This field is
currently unused, as all of the entries live in the same block.
struct ext2_ext_attr_entry {
__u8 e_name_len; /* length of name */
__u8 e_name_index; /* attribute name index */
__le16 e_value_offs; /* offset in disk block of value */
> __le32 e_value_inum; /* inode in which the value is stored */
__le32 e_value_size; /* size of attribute value */
__le32 e_hash; /* hash value of name and value */
char e_name[0]; /* attribute name */
}
Second, add a flag to the inode that indicates it is using a large
(external) extended attribute. This is needed so that when unlinking
an inode the xattrs will be scanned to unlink the xattr inodes
referenced by the main inode.
Third, for inodes that have a number of xattrs that are larger than
a single block, but not large enough to justify an external inode
(less than 64kB total xattr size, due to e_value_offs limitation)
the ext2_ext_attr_header->h_blocks field can grow beyond a single
block to represent a contiguous allocation of blocks for the xattr.
Signed-off-by: Andreas Dilger <adilger@sun.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The system header file can end up causing type conflicts, and
including kernel header files is always dodgy/dangerous (and this case
not needed).
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Some devices, notably 4k sector drives, may have a 512 logical
sector size, mapped onto a 4k physical sector size.
When mke2fs is ratcheting down the blocksize for small filesystems,
or when a blocksize is specified on the commandline, we should not
willingly go below the physical sector size of the device.
When a blocksize is specified, we -must- not go below
the logical sector size of the device.
Add a new library function, ext2fs_get_device_phys_sectsize()
to get the physical sector size if possible, and adjust the
logic in mke2fs to enforce the above rules.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The top-level COPYING file states that the e2p and ext2fs libraries
are available under the LGPLv2. The files were incorrectly labelled.
Alex Thomas/Luster has been consulted wrt to the ext3_extents.h file;
the rest of the files were primarily authored by Theodore Ts'o.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
namei.o is also needed by e2initrd_helper.
Long term, if we care about reduced e2fsprogs builds, we need a more
general solution for deciding what .o files are needed for a
particular build. Given that install floppies are going (gone?) the
way the dodo bird, we probably don't care, though.
Addresses-Sourceforge-Bug: #2911433
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
These options allow e2fsprogs to be built using symlinks instead of
hard links, and to be installed using symlinks instead of hard links,
respectively.
Addresses-Sourceforge-Bug: #1436294
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This fixes a long-standing botch in the com_err library, and solves a
regression test problem for libss that gets tickled by source code
management systems (like Perforce) that don't preserve CRLF line
endings with fidelity.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When ext2fs_block_iterate2() is called on an extent-mapped file with a
depth > 1, it will erroneously calling the callback function starting
all over again with an offset of logical block 0. It shouldn't do
this, and it cases mke2fs to become very slow when creating files with
very large journals.
Fix this.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This is the userspace side of Jiaying's EOFBLOCKS patch. With
Aneesh's patches for .33, Jiaying's patch, and this one, xfstests
013/fsstress (even with direct IO enabled) has held up through many
runs.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Rawhide now has libreadline.so.6 ... add it to the ever-expanding
list of libs to look for.
Unfortunately without commit 06ef971be5
this fails in a rather cryptic way.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When trying to find the best place for the inode table in the last
flex block group, use the true size for the flex_bg's portion of the
inode table instead of the worst case required size of the inode table
fragment if the file system is resized. This fixes a corner case
where if the size of the filesystem is just big enough that there is
only room for a single block group in the last flex_bg, and that
partial block group is too small for the full portion of the inode
table, the inode table is placed in the very first block group:
Group 64: (Blocks 2097152-2099199) [INODE_UNINIT, ITABLE_ZEROED]
Checksum 0xd305, unused inodes 8080
Block bitmap at 2097152 (+0), Inode bitmap at 2097168 (+16)
Inode table at 8626-9130 (+4292878770)
^^^^^^^^^
Thanks to Vyacheslav Dubeyko for pointing this out.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Unfortunately, texi2html gratuitously changed its behavior of where
its output html files are placed when the -split_chapter is in effect.
(First it was in a subdirectory; then it was in the current directory;
now it's back to putting the output html files in a subdirectory
again.)
Support either way of doing things since the texi2html team seems to
be indecisive...
Addresses-Debian-Bug: #552934
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Don't byte-swap the block number when setting i_block[x], since the
write_inode function will take of byte swapping the inode.
The phys_blk parameter contains an input parameter in the SET_BMAP
case, so it must be passed to ext2fs_bmap2() from the legacy function
ext2fs_bmap().
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Linux <= 2.6.19 contained a bug in the /proc/swaps code where the
header would not be displayed (the first line).
This issue has been reported by Mike Frysinger for swapon(8).
Signed-off-by: Karel Zak <kzak@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The f_illitable_flexbg test was failing on ppc, because
e2fsck_move_ext3_journal is doing a direct memcmp of i_block with
s_jnl_blocks, and failing.
This is because we don't swap extent data on read from disk; rather
we do it when we access the extents. However, ext2fs_swap_super
was swapping s_jnl_blocks unconditionally, so these didn't match.
Looks like we need to treat s_jnl_blocks the same as i_block, and
swap it on access, not on read. Except for the last i_size bit...
Reviewed-by: Andreas Dilger <adilger@sun.com>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Clear the function pointer for readline_shutdown() so that if libss is
linking against a readline library which doesn't supply a
readline_shutdown() function, ss_delete_invocation() won't seg fault.
Thanks to Solar Designer <solar@openwall.com> for reporting this
problem to me.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If the RO compat HUGE_FILE feature flag is set, but the inode's
HUGE_FILE_FL flag is not set, we should still pay attention to the
high 32 bits of the i_blocks filed.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Commit 1d9b818: dumpe2fs: Print more information about the inline journal
caused dumpe2fs to use ext2fs_file_open2(). Previously the file_io
functions were only used by debugfs, so if debugfs was disabled,
file_io was not built. Now that dumpe2fs is also using file_io, we
need to build it unconditionally.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This commit forces the use of the system-provided blkid or uuid header
files if we are using the system-provided blkid or uuid libraries.
This avoids using the in-tree header files with the system libraries.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The x86 BT assembly instructure can overshoot the end of a bit array
when testing a bit at the end of the bit array, even if it never needs
to look at those memory locations. This can cause a spurious
segmentation fault. If we allocate a little extra memory, it avoids
this problem. See:
http://faydoc.tripod.com/cpu/bt.htm
This doesn't happen on Linux, probably because of the glibc's malloc()
function works, but apparently it's a major problem on the *BSD
operating systems.
Addresses-Sourceforge-Bug: #2328708
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The Linux kernel modules could be compressed, it means modules.dep
parser in libblid has to support .ko.gz extension too.
(Note, I've talked about this problem with Jon Masters and his
suggestion is to exec(/sbin/modinfo) rather than directly parse
modules.dep. BTW, the modules.dep file is deprecated.)
Address-Red-Hat-Bug: #518572
Signed-off-by: Karel Zak <kzak@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
For filesystms that have the extent feature enabled, we need to grab
the use EXT2_IOC_GETFLAGS so that we don't accidentally end up trying
to request clearing the EXT2_EXTENT_FL, which is not supported and
causes the tune2fs -j error out.
Also fix the error returning in ext2fs_add_journal_inode() so it
returns a proper error code if the fstat() or ioctl() calls fail.
Addresses-Launchpad-bug: #416648
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
We are using a signed int to store a block number in
ext2fs_allocate_group_table. We don't actually do any computation or
comparisons using it, so it shouldn't cause any bugs, but it's
technically incorrect, and it's possible an overly clever compiler
might do something wrong with it.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Avoiding inserting a new extent if it is possible to merge the new
block to the beginning or the end of the previous or next extent.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>