Add a fast path optimization in e2fsck's pass 5 for the common case
where the block bitmap is correct. The optimization works by
extracting each block group's block allocation bitmap into a memory
buffer, and comparing it with the expected allocation bitmap using
memcmp(). If it matches, then we can just update the free block
counts and be on our way, and skip checking each bit individually.
Addresses-Google-Bug: #7534813
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
Optimize e2fsck pass 1 by marking entire extents as being in use at a
time, instead of block by block. This optimization only works for
non-bigalloc file systems for now (it's tricky to handle bigalloc file
systems since this code is also responsible for dealing with blocks
that are not correctly aligned within a cluster). When the
optimization works, the CPU savings can be significant: ove a full CPU
minute for a mostly full 4T disk.
Addresses-Google-Bug: #7534813
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Lukas Czerner <lczerner@redhat.com>
In e2fsck_pass4(), we were consulting inode_dir_map using
ext2fs_test_inode_bitmap2() for every single inode in the file system.
However, there were many cases where we never needed the result of the
test --- most notably if the inode is not in use.
I was a bit surprised that GCC 4.7 with CFLAGS set to "-g -O2" wasn't
able to optimize this out for us, but here is the pass 4 timing for an
empty 3T file system before this patch:
Pass 4: Memory used: 672k/772k (422k/251k), time: 3.67/ 3.66/ 0.00
and afterwards, we see a 43% improvement:
Pass 4: Memory used: 672k/772k (422k/251k), time: 2.09/ 2.08/ 0.00
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Also fix a bug caused by a stray continuation backslash which caused
the e2fsck/Makefile to fail when profiling is enabled.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
Since clang uses C99 semantics by default, the main changes required
to allow clang to build e2fsprogs was to add support the C99 inline
semantics, while still allowing us to be built when the legacy (but
still default for gcc) GNU C89 inline semantics are in force.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Commit 07307114de didn't correctly handle the lost+found directory
when it added support for metadata checksums. First of all,
e2fsck_get_lost_and_found() assumed that the inode_dir_map bitmap was
initialized, and it wasn't when it was called earlier by a change in
that commit. Secondly, it's important that lost+found dirctory is
processed in case its directory checksums are incorrect, but should
preserve any empty dirctory blocks so there space available for e2fsck
to reconnect any orphan inodes.
Fix these problems, to fix test failures: f_holedir2 and f_rehash_dir
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When passed a negative count (indicating a byte count rather than
a block count) e2fsck_handle_read_error() treats the data as a full
block, causing unix_write_blk64() (which can handle negative counts
just fine) to try to write too much. Given a faulty block device,
this resulted in a SEGV when unix_write_blk64() read past the bottom
of the stack copying the data to cache. (check_backup_super_block ->
unix_read_blk64 -> raw_read_blk -> e2fsck_handle_read_error)
Reported-by: Alex Friedman <alexfr@il.ibm.com>
Signed-off-by: Jim Keniston <jkenisto@us.ibm.com>
Signed-off-by: Dan Streetman <ddstreet@us.ibm.com>
Reviewed-by: Mingming Cao <mcao@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Remove crc32_be in favor of the implementation in libext2fs.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Check the data block checksums when recovering the journal.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When recovering a journal with checksum v2, verify the commit block
checksum.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Verify the descriptor block checksum when recovering a journal.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Verify the revoke block checksum when recovering the journal.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Ensure that the journal superblock passes checksum before recovering the
filesystem.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Define flags and change journal structure definitions to support v2 journal
checksumming.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Check and handle MMP checksum problems by resetting the block.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Use the helper function to determine if group descriptors have a
checksum. Ensure that metadata_csum and uninit_bg flags are not set
simultaneously, as part of pass 0.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If e2fsck finds a superblock with an invalid checksum, try the
backups.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Verify the checksums of separate extended attribute blocks and offer
to clear it if there is a mismatch.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Calculate and verify the checksum for separate (i.e. not in the inode)
extended attribute blocks; the checksum lives in the header.
[ Merged in change from Tao so that we always use the fs checksum seed
for the xattr blocks. ]
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Tao Ma <boyu.mt@taobao.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Checks that directory leaf blocks have the necessary fake dir_entry at
the end of the block to hold a checksum and that the checksum is
valid. It will resize the block and/or rebuild the directory if
necessary.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Introduce small structures for recording directory tree checksums, and
some API changes to support writing out directory blocks with
checksums.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Check htree internal node checksums. If broken, ask user to clear
the htree index and recreate it later.
[ Move the check for not rehashing the lost+found directory to pass1
so that we don't end up truncating lost+found when the metadata
checksum feature is enabled. -- TYT ]
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When we encounter an extent tree block that passes the header check
but fails the checksum, offer to clear just that extent block instead
of failing the whole tree, which results in the entire inode being
wiped out.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Since the correct inode and block bitmaps are calculated in pass 5,
don't fail the bitmap read operation in prior passes since (a)
incorrect results won't kill us and (b) if we fail early, we'll never
_get_ to pass 5.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Check block bitmap checksum and write a new checksum if the
verification fails. This is ok because e2fsck has already computed
the correct block bitmap.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Rewrite the block bitmap when the checksum doesn't match. This is
ok since e2fsck will have already computed the correct inode bitmap.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Detect mismatches of the inode and checksum, and prompt the user to
fix the situation.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Precompute the FS UUID checksum seed that is used for all metadata
checksumming operations and store it in ext2_filsys.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Change libext2fs to read and write full-size inodes in preparation for
the metadata checksumming patchset, which will require this. Due to
ABI compatibility requirements, this change must be hidden from client
programs.
Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In addition to the free blocks and free inodes, also print the number
of blocks and inodes in the verbose statistics.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Disks have gotten bigger, so 8 digits might not be enough. Allow for
12 digits worth of blocks, which is more than enough for 3 petabytes.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Add report_time, report_verbose, and report_features options to
e2fsck.conf which enable additional, more verbose reporting by e2fsck.
This is useful for large cloud installations where there are a large
number file systems being managed, and where it may not be obvious
from the e2fsck log files exactly how a particular file system is
configured.
The report_time and report_verbose options, which are the same as the
-tt and -v command line options, respectively, are useful because they
are options specific to e2fsck, and the fsck program does not have a
way of passing certain options only to a specific /sbin/fsck.<fstype>
program.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
When checking to see whether or not a new name is unique, the code was
using the wrong length parameter, which could cause the anti-collision
loop for a long time trying to find what it thinks is a unique name.
Addresses-Sourceforge-Bug: #3540545
Reported-by: Vitaly Oratovsky <vmo@users.sourceforge.net>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Previously e2fsck would only allow a mounted file system to be checked
if it was the root file system and it was mounted read-only. Now
allow any file system mounted read-only if the -f option is specified.
This makes it easier to test how e2fsck handles checking file systems
which are mounted without having to test on the root file system.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
This reverts commit 47c1b8e166.
The original reason for this commit was to speed up boots for hard
drives. However, I've measured the time difference on a 1TB laptop
drive, and it's not significant: 70ms vs 10ms when running e2fsck on a
clean file system.
The problem with this optimization is that we don't notice if the
journal superblock has a non-zero s_errno field. If we don't transfer
the error indicator from the journal superblock to the file system
superblock, then the kernel will transfer it when the file system is
remounted read-write, causing scary messages to appear in the syslog.
(And since there was a bug in the kernel code which didn't clear the
error indicator in the journal superblock, it would never get
cleared.)
Reported-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If user chooses to not fix quota info, then the FS should be
marked as having errors. PR_NO_OK prevented this from happening.
Signed-off-by: Aditya Kali <adityakali@google.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Since "bool" is a valid C type, declarations of the form "int bool"
will cause compiler errors if <stdbool.h> is included. Rename these
variables to avoid this name clash.
Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The Build Log Hardening Check is a debian tool which scans the output
of a package build making sure that the security hardening flags are
used when compiling and linking all of binaries in a package.
For the most part we were passing CFLAGS, CPPFLAGS, and LDFLAGS down
to the compiler and link commands, but there there were one or two
exceptions. In addition, there where a few places in "make install"
where the V=1 option was not being honored, which triggered blhc
warnings since it couldn't analyze those commands.
The e2fsck.static was the only binary that was not getting built and
packaged with the hardening flags, but I've fixed all of the blhc
warnings so in the future it will be obvious if we regress.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
By the time we start processing the orphan inode list, we have already
calculated the total expected number of free blocks and inodes in
ctx->free_{blocks,inodes}. This is used to set the free blocks/inodes
count in the superblock in the case where we don't need to do a full
e2fsck.
We need to update these expected free block counts as we process the
orphan inode list so that superblock values are set correctly.
Otherwise we could have the following happen:
% e2fsck /tmp/test.img
e2fsck 1.42.3 (14-May-2012)
Truncating orphaned inode 12 (uid=0, gid=0, mode=0100644, size=0)
Setting free blocks count to 46 (was 79)
/tmp/test.img: clean, 12/16 files, 54/100 blocks
% e2fsck /tmp/test.img
e2fsck 1.42.3 (14-May-2012)
Setting free blocks count to 79 (was 46)
/tmp/test.img: clean, 12/16 files, 21/100 blocks
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
The on-disk format for interior nodes in the extent tree does not
encode the length of each entry in the interior node; instead, it is
synthesized/simulated by the extent library code in libext2fs.
Unfortunately, this simulation is not perfect; in particular it does
not work for the last extent in the extent tree if there are
uninitialized blocks allocated using fallocate with
FALLOC_FL_KEEP_SIZE, and it leads to e2fsck incorrectly complaining
about an invalid zero-length extent.
We only need to worry about the extent length for the leaves of the
tree, since it is there were we are checking an on-disk value, as
opposed to a software-generated simulation. So restrict the check of
extent length to leaf nodes in the extent tree.
Reported-by: Tao Ma <tm@tao.ma>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If the file system is mounted read-only after a file system error has
been detected, the fact that an error occurred is written to the
journal. This is important because while the journal is getting
replayed, the error indication in the superblock may very well get
overwritten.
Unfortunately, the code to propagate the error indication from the
journal to superblock was broken because this was being done before
the old file system handle is thrown away and the file system is
re-opened to ensure that no stale data is in the file system handle.
As a result, the error indication in the superblock was never written
out.
To fix this, we need to move the check if the journal's error
indicator has been set after the file system has been freed and
re-open.
Reported-by: Ken Sumrall <ksumrall@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Quite some definitions in quota library are not necessary. Remove them.
Also fold quota.h file into quotaio.h since it didn't contain that many
definitions.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
If a file system was remounted read-only after a file system
corruption is detected, and then that file system is mounted and
unmounted by the kernel, the journal would have been recovered, but
the kernel currently leaves the s_errno field still set. This is
arguably a bug, since it has already propgated the non-zero s_errno
field to the file system superblock, where it will be retained until
e2fsck has been run.
However, e2fsck should handle this case for existing kernel by
checking the journal superblock's s_errno field even if journal
recovery is not required.
Without this commit, e2fsck would not notice anything wrong with the
file system, but a subsequent mount of the file system by the kernel
would mark the file system's superblock as needing checking (since the
journal's s_errno field would still be set), resulting an full e2fsck
run at the next reboot, which would find nothing wrong --- and then
when the file system was mounted, the whole cycle would repeat again.
I had seen reports of this in the past, but it wasn't until recently
that I realized exactly how this had come about, since normally e2fsck
would be run automatically before the file system is mounted again,
thus avoiding this problem. However, a user using a rescue CD who
didn't run e2fsck before mounting the a file system in this condition
could trigger this situation, and unfortunately, with previous
versions of e2fsprogs and the kernel, there would be no way out no
matter what the user tried to do.
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Commit 732e26b98e added checks to
prevent e2fsck from being run in filesystem-modifying mode against
a mounted or otherwise busy device, due to several bug reports of
users doing this even with the verbose warnings in check_mount().
However, it also prevented e2fsck from checking a mounted root
filesystem, which will prevent the node from booting. Once again
allow e2fsck to run against the mounted root filesystem if it is
also mounted read-only at the time.
Signed-off-by: Andreas Dilger <adilger@whamcloud.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>