Whamcloud - gitweb
tools/e2fsprogs.git
21 months agodebugfs: teach logdump the -n <num_trans> option
lihaoxiang (F) [Thu, 14 Jul 2022 01:32:48 +0000 (09:32 +0800)]
debugfs: teach logdump the -n <num_trans> option

The current version's debugfs possessed the function
logdump. Executing with option -O could output the log history. But
when it occurred the block which had no magic number in it's header,
the program would exit.

Sometimes we were locating problems, needed for more transactions that
had replayed instead of the latest batch of transactions and we
weren't hope to display all the history in the meanwhile. So we
introduced the option -n used for controlling the print of history
transactions.  Specially, this parameter was depending on the option
-O otherwise it couldn't work.

So in this modification, we used logdump with -O -n <num_trans>.  The
-n options causes logdump to continue past a block with a missing
magic nuber.  Instead, it will terminate only when the entire log has
been printed or after <num_trans> transactions.

Link: https://lore.kernel.org/r/608df030-593f-8c69-cb65-632a34729d23@huawei.com
Signed-off-by: lihaoxiang <lihaoxiang9@huawei.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agotune2fs: allow disabling casefold feature
Slava Bacherikov [Fri, 8 Jul 2022 12:26:58 +0000 (15:26 +0300)]
tune2fs: allow disabling casefold feature

Casefold can be safely disabled if there are no directories with +F
attribute ( EXT4_CASEFOLD_FL ). This checks all inodes for that flag and in
case there isn't any, it disables casefold FS feature. When FS has
directories with +F attributes, user could convert these directories,
probably by mounting FS and executing some script or by doing it
manually. Afterwards, it would be possible to disable casefold FS flag
via tune2fs.

Link: https://lore.kernel.org/r/20220708122658.17907-1-slava@bacher09.org
Signed-off-by: Slava Bacherikov <slava@bacher09.org>
Reviewed-by: Gabriel Krisman Bertazi <krisman@collabora.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoMerge branch 'maint' into next
Theodore Ts'o [Sat, 13 Aug 2022 02:34:57 +0000 (22:34 -0400)]
Merge branch 'maint' into next

21 months agoe2fsprogs: fix device name parsing to resolve names containing '='
Lukas Czerner [Fri, 12 Aug 2022 13:01:22 +0000 (15:01 +0200)]
e2fsprogs: fix device name parsing to resolve names containing '='

Currently in varisous e2fsprogs tools, most notably tune2fs and e2fsck
we will get the device name by passing the user provided string into
blkid_get_devname(). This library function however is primarily intended
for parsing "NAME=value" tokens. It will return the device matching the
specified token, NULL if nothing is found, or copy of the string if it's
not in "NAME=value" format.

However in case where we're passing in a file name that contains an
equal sign blkid_get_devname() will treat it as a token and will attempt
to find the device with the match. Likely finding nothing.

Fix it by checking existence of the file first and then attempt to call
blkid_get_devname(). In case of a collision, notify the user and
automatically prefer the one returned by blkid_get_devname(). Otherwise
return either the existing file, or NULL.

We do it this way to avoid some existing file in working directory (for
example LABEL=volume-name) masking an actual device containing the
matchin LABEL. User can specify full, or relative path (e.g.
./LABEL=volume-name) to make sure the file is used instead.

Link: https://lore.kernel.org/r/20220812130122.69468-1-lczerner@redhat.com
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reported-by: Daniel Ng <danielng@google.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoe2fsck: validate i_extra_size in ext4_fc_handle_inode
Theodore Ts'o [Fri, 12 Aug 2022 19:48:04 +0000 (15:48 -0400)]
e2fsck: validate i_extra_size in ext4_fc_handle_inode

Addresses-Coverity-Bug: 1500765
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agodebugfs: make blocksize be unsigned in logdump.c
Theodore Ts'o [Fri, 12 Aug 2022 19:15:14 +0000 (15:15 -0400)]
debugfs: make blocksize be unsigned in logdump.c

Blocksize can never be negative, and this makes the use of signed vs
unsigned variables for offsets be consistent.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agolibext2fs: avoid NULL dereference if the root file system isn't in /etc/mtab
Theodore Ts'o [Fri, 12 Aug 2022 04:08:16 +0000 (00:08 -0400)]
libext2fs: avoid NULL dereference if the root file system isn't in /etc/mtab

Addresses-Coverity-Bug: 1497609
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoMerge branch 'maint' into next
Theodore Ts'o [Fri, 12 Aug 2022 03:51:03 +0000 (23:51 -0400)]
Merge branch 'maint' into next

21 months agoAvoid potential NULL dereference when argv[0]
Theodore Ts'o [Fri, 12 Aug 2022 03:45:21 +0000 (23:45 -0400)]
Avoid potential NULL dereference when argv[0]

Addresses-Coverity-Bug: 1500772
Addresses-Coverity-Bug: 1500769
Addresses-Coverity-Bug: 1500767
Addresses-Coverity-Bug: 1500758
Addresses-Coverity-Bug: 1500756
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoe2fsck: mark that we don't care about the return value of e2fsck_lookup()
Theodore Ts'o [Fri, 12 Aug 2022 03:14:33 +0000 (23:14 -0400)]
e2fsck: mark that we don't care about the return value of e2fsck_lookup()

We only print the parent directory to help provide context to the
user, but it's possible that a corrupted directory doesn't have a '..'
link.

Addresses-Coverity-Bug: 1507762
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoe2fsck: streamline problem latch handling
Theodore Ts'o [Fri, 12 Aug 2022 03:01:42 +0000 (23:01 -0400)]
e2fsck: streamline problem latch handling

No functional changes, but streamline the logic, and avoid a coverity
warning.

Addresses-Coverity-Bug: 1507763
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoe2fsck: when mutating file name make sure its length never exceeds 255
Theodore Ts'o [Fri, 12 Aug 2022 02:16:41 +0000 (22:16 -0400)]
e2fsck: when mutating file name make sure its length never exceeds 255

E2fsck will attempt to mutate filenames to ensure uniqueness if
necessary.  If there are two unique filenames that are 254 or 255
characters in length and do not contain the '~' character, the
mutate_name() function would create a filename which is 256 bytes
long, which is not a legal filename in Linux.  Adjust the mutate_name
function to avoid this possibility.

Addresses-Coverity-Bug: 1500768
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoFix Coverity unintentional integer overflow warnings
Theodore Ts'o [Fri, 12 Aug 2022 02:03:08 +0000 (22:03 -0400)]
Fix Coverity unintentional integer overflow warnings

Neither of these two warnings can actually happen (other limits will
be hit first), but widening the integer to a 64-bit unsigned integer
is an cheap and effective way to silence the Coverity warnings.

Addresses-Coverity-Bug: 1500760
Addresses-Coverity-Bug: 1507886
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoresize2fs: use ext2fs_get_arrayzero() instead of ext2fs_get_array() + memset()
Theodore Ts'o [Thu, 11 Aug 2022 22:37:26 +0000 (18:37 -0400)]
resize2fs: use ext2fs_get_arrayzero() instead of ext2fs_get_array() + memset()

The use of ext2fs_get_arrayzero() to replace using ext2fs_get_array()
+ memset() does not result in any functional change, but it (a) is
slightly more efficient, and (b) makes it easier for Coverity to avoid
signalling a false positive.

Addresses-Coverity-Bug: 1500763
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agomisc: use ext2_ino_t instead of ino_t
Andreas Dilger [Fri, 5 Aug 2022 21:28:54 +0000 (15:28 -0600)]
misc: use ext2_ino_t instead of ino_t

Some of the new fastcommit and casefold changes used the system
"ino_t" instead of "ext2_ino_t" for handling filesystem inodes.
This causes printf warnings if the system "ino_t" is of a different
size.  Use the library "ext2_ino_t" for consistency.

Signed-off-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Gabriel Krisman Bertazi <krisman@collabora.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoMerge branch 'maint' into next
Theodore Ts'o [Thu, 11 Aug 2022 17:00:13 +0000 (13:00 -0400)]
Merge branch 'maint' into next

21 months agotune2fs: do not change j_tail_sequence in journal superblock
zhanchengbin [Thu, 4 Aug 2022 10:33:39 +0000 (18:33 +0800)]
tune2fs: do not change j_tail_sequence in journal superblock

The function recover_ext3_journal() in debugfs/journal.c, if the log
replay is over, the j_tail_sequence in journal superblock is not
changed to the value of the last transaction sequence.  This will
cause subsequent log commitids to count from the commitid in last
time.  After tune2fs -e, the log commitid is counted from the commitid
in last time, if the log ID of the current operation overlaps with
that of the last operation, this will cause logs that were previously
replayed by tune2fs to be replayed here.

Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
Signed-off-by: Zhiqiang Liu <liuzhiqiang26@huawei.com>
Signed-off-by: liangyun <liangyun2@huawei.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agodebugfs: allow <inode> for ncheck
Li Dongyang [Fri, 5 Aug 2022 21:52:21 +0000 (15:52 -0600)]
debugfs: allow <inode> for ncheck

If the ncheck argument is of the form "<ino>", allow it for ncheck
for consistency with other commands that accept an inode number.

Improve the error message, use "Invalid inode number" instead
of "Bad inode", which implies the inode content being bad.

Signed-off-by: Li Dongyang <dongyangli@ddn.com>
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agodebugfs: quiet debugfs 'catastrophic' message
Andreas Dilger [Fri, 5 Aug 2022 22:06:07 +0000 (16:06 -0600)]
debugfs: quiet debugfs 'catastrophic' message

When debugfs runs with "-c", it prints a scary-looking message:

    catastrophic mode - not reading inode or group bitmaps

that is often misunderstood by users to mean that there is something
wrong with the filesystem, when there is no problem at all.

Not reading the bitmaps is totally normal and expected behavior for
the "-c" option, which is used to significantly shorten the debugfs
command execution time by not reading metadata that isn't needed for
commands run against very large filesystems.

Since there is often confusion about what this message means, it
would be better to just avoid printing anything at all, since the
use of "-c" is expressly requesting this behavior, and there are
no messages printed out for other options.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Reviewed-by: Dongyang Li <dongyang@ddn.com>
Change-Id: I59b26a601780544ab995aa4ca7ab0c2123c70118
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoresize2fs: open device read-only when -P is passed
Michael Hudson-Doyle [Thu, 26 May 2022 01:08:28 +0000 (13:08 +1200)]
resize2fs: open device read-only when -P is passed

We ran into this because we noticed that resize2fs -P $device was
triggering udev events.

I added a very simple test that just checks resize2fs -P on a file
lacking the w bit succeeds.

Signed-off-by: Michael Hudson-Doyle <michael.hudson@ubuntu.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agotests: fix ACL-printing tests
Andreas Dilger [Tue, 1 Mar 2022 04:17:06 +0000 (21:17 -0700)]
tests: fix ACL-printing tests

Fix the ACL-printing tests to be more flexible for different systems.
If the MKFS_DIR is on tmpfs, it will not list "system.posix_acl*"
xattrs, so they will not be copied.  Create this on a real filesystem
or skip the test if that doesn't work.

Filter out the security.selinux xattr if it is printed, since this
depends on the selinux configuration of the host system.  However,
this also spills xattrs for "acl_dir/file" into an external xattr
block, and causes it to fail due to different block allocations.
Increase the filesystem inode size so that the allocation is the same
regardless of whether selinux is enabled or not.

Fixes: 67e6ae0a35 ("mke2fs: fix a importing a directory with an ACL")
Signed-off-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Artem Blagodarenko <artem.blagodarenko@hpe.com>
Reviewed-by: Li Dongyang <dongyangli@ddn.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoe2fsck: always probe filesystem blocksize with simple io_manager
Gabriel Krisman Bertazi [Mon, 25 Apr 2022 22:01:00 +0000 (18:01 -0400)]
e2fsck: always probe filesystem blocksize with simple io_manager

Combining superblock (-b) with undo file (-z) fails iff the block size
is not specified (-B) and is different from the first blocksize probed
in try_open_fs (1k).  The reason is as follows:

try_open_fs() will probe different blocksizes if none is provided on
the command line. It is done by opening and closing the filesystem
until it finds a blocksize that makes sense. This is fine for all
io_managers, but undo_io creates the undo file with that blocksize
during ext2fs_open.  Once try_open_fs realizes it had the wrong
blocksize and retries with a different blocksize, undo_io will read
the previously created file and think it's corrupt for this
filesystem.

Ideally, undo_io would know this is a probe and would fix the undo file.
It is not simple, though, because it would require undo_io to know the
file was just created by the probe code, since an undo file survives
through different fsck sessions.  We'd have to pass this information
around somehow.  This seems like a complex change to solve a corner
case.

Instead, this patch changes the blocksize probe to always use the
unix_io_manager. This way, we safely probe for the blocksize without
side effects.  Once the blocksize is known, we can safely reopen the
filesystem under the proper io_manager.

An easily reproducer for this issue (from Ted, adapted by me) is:

 mke2fs -b 4k -q -t ext4 /tmp/foo.img 2G
 e2fsck -b 32768 -z /tmp/undo /tmp/foo.img

Reported-by: Peter Urbanec <linux-ext4.vger.kernel.org@urbanec.net>
Signed-off-by: Gabriel Krisman Bertazi <krisman@collabora.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoQuiet unused variable warnings
Andreas Dilger [Thu, 4 Aug 2022 17:18:32 +0000 (11:18 -0600)]
Quiet unused variable warnings

Quiet various compiler warnings about unreferenced or unset variables.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoresize2fs: trim resize to cluster boundary
Kiselev, Oleg [Sat, 14 May 2022 04:17:09 +0000 (04:17 +0000)]
resize2fs: trim resize to cluster boundary

This patch rounds down the size provided to resize2fs to the nearest
cluster boundary for bigalloc filesystems.  This is similar to the
trimming already done for page boundary alignment.  Aligning the size in
the user space provides the right value feedback from the resize2fs
command, which is a better user experience than trimming the size
in the kernel.

Signed-off-by: Oleg Kiselev <okiselev@amazon.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoMerge branch 'maint' into next
Theodore Ts'o [Thu, 11 Aug 2022 03:46:16 +0000 (23:46 -0400)]
Merge branch 'maint' into next

21 months agoe2fsck: handle invalid percent expansions in the log filename
Theodore Ts'o [Thu, 11 Aug 2022 02:48:13 +0000 (22:48 -0400)]
e2fsck: handle invalid percent expansions in the log filename

Add a missing default: case when expanding percent expansions in the
log file specified in /etc/e2fsck.conf.

Addresses-Coverity-Bug: 1500757
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agolibext2fs: validate block # of the inode table in ext2fs_image_inode_write()
Theodore Ts'o [Thu, 11 Aug 2022 02:43:00 +0000 (22:43 -0400)]
libext2fs: validate block # of the inode table in ext2fs_image_inode_write()

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoBuild the fuzzers from oss-fuzz
Theodore Ts'o [Sat, 6 Aug 2022 22:35:30 +0000 (18:35 -0400)]
Build the fuzzers from oss-fuzz

The fuzzers from oss-fuzz in projects/e2fsprogs/fuzz (as of commit
78ecd3f07fca with some slight modifications for better error
reporting) have been placed in the tests/fuzz directory and the
configure script now supports a new option --enable-fuzzing which will
build these fuzzers using clang's -fsanitize=fuzzer command line
option.

In general, some sanitizer such as --enable-addrsan or --enable-ubsan
(to enable ASAN or UBSAN, respectively) should be enabled alongside
--enable-fuzzing.

A typical configure command to build the fuzzers might be:

  configure CC=clang CXX=clang++ CFLAGS=-g --enable-fuzzing --enable-addrsan

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agolibext2fs: fix potential integer overflow in bitmap accessors
Theodore Ts'o [Tue, 9 Aug 2022 15:16:47 +0000 (11:16 -0400)]
libext2fs: fix potential integer overflow in bitmap accessors

bmap->cluster_bits has a maximum value of 19, but Coverity doesn't
know that.  To make it happy, and just in case there is a bug where
somehow the cluster size does get set to an invalid value and the rest
of the library doesn't check it, use 1ULL instead of 1 to avoid the
integer overflow.

Addresses-Coverity-Bug: 1500759
Addresses-Coverity-Bug: 1500764
Addresses-Coverity-Bug: 1500771
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoe2fsck: fix potential fencepost error in e2fsck_should_rebuild_extents()
Theodore Ts'o [Tue, 9 Aug 2022 14:52:57 +0000 (10:52 -0400)]
e2fsck: fix potential fencepost error in e2fsck_should_rebuild_extents()

The ext2_extent_info.max_depth is zero-based (e.g., it is zero when
the entire extent tree fits in the inode).  Hence, if it is equal to
MAX_EXTENT_DEPTH_COUNT we should always rebuild the extent tree to
shorten it.

Also, for 1k block file systems, it's possible for the worst-case
extent tree in its most compact form to have a maximum depth of 6, not
5.  So set MAX_EXTENT_DEPTH_COUNT to 8 just to be sure we have plenty
of headroom.  (The kernel supports an extent depth up to 2**16, but
e2fsck only keeps statistics up to MAX_EXTENT_DEPTH_COUNT, and if it's
deeper than that, we know that it will be profitable to rebuild the
extent tree in any case.)

Addresses-Coverity-Bug: 1507761
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agolibext2fs: make sure the bitmap locations are valid when writing bitmaps
Theodore Ts'o [Tue, 9 Aug 2022 01:02:51 +0000 (21:02 -0400)]
libext2fs: make sure the bitmap locations are valid when writing bitmaps

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agolibext2fs: reject various bitmap and inode operations for journal_dev file systems
Theodore Ts'o [Tue, 9 Aug 2022 00:52:43 +0000 (20:52 -0400)]
libext2fs: reject various bitmap and inode operations for journal_dev file systems

The ext2fs_open() function will only allow journal_dev file systems to
be open if explicitly requested by programs using the
EXT2_FLAG_JOURNAL_DEV_OK flag.  Those programs will not try to call
functions that make no sense, such as ext2fs_read_inode(),
ext2fs_read_bitmaps(), etc.  Just to make things the library more
robust against buggy programs (or unrealistic fuzzers) add a check for
journal_dev file systems to various ext2fs library functions to return
a new error, EXT2_ET_EXTERNAL_JOURNAL_NOSUPP.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agolibext2fs: in ext2fs_open[2](), return an error if s_desc_size is too large
Theodore Ts'o [Tue, 9 Aug 2022 00:17:40 +0000 (20:17 -0400)]
libext2fs: in ext2fs_open[2](), return an error if s_desc_size is too large

Previously, ext2fs_open() and ext2fs_open2() would return an error if
s_desc_size is too small.  Add a check so it will return an error if
s_desc_size is too large, as well.

These checks will be skipped for e2fsck when it uses the flag
EXT2_FLAG_IGNORE_SB_ERRORS.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoMerge branch 'maint' into next
Theodore Ts'o [Mon, 8 Aug 2022 01:31:18 +0000 (21:31 -0400)]
Merge branch 'maint' into next

21 months agoUpdate configure/configure.ac/aclocal.m4 to use autoconf 2.71
Theodore Ts'o [Sat, 6 Aug 2022 18:58:40 +0000 (14:58 -0400)]
Update configure/configure.ac/aclocal.m4 to use autoconf 2.71

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
21 months agoFix UBSAN if s_log_groups_per_flex is 31
Theodore Ts'o [Sun, 7 Aug 2022 23:47:25 +0000 (19:47 -0400)]
Fix UBSAN if s_log_groups_per_flex is 31

It is logal (albeit rare) for the number of block groups per flex_bg
to 2**31 (which effectively means to put all of the block groups into
a single flex_bg).  However, in that case "1 << 31" is undefined on
architectures with a 32-bit integer.  Fix this UBSAN complaint by
using "1U << 31" instead.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
22 months agoMerge branch 'maint' into next
Theodore Ts'o [Sat, 6 Aug 2022 06:37:24 +0000 (02:37 -0400)]
Merge branch 'maint' into next

22 months agolibext2fs: teach ext2fs_open() to reject file systems with an invalid flex_bg size
Theodore Ts'o [Sat, 6 Aug 2022 06:21:49 +0000 (02:21 -0400)]
libext2fs: teach ext2fs_open() to reject file systems with an invalid flex_bg size

If s_log_groups_per_flex is greater than 31, it will result in an
UBSAN error, since it will result in an invalid shift exponent when
calculating the flex_bg size.  So reject such file systems when they
are opened.  (The mke2fs program will not allow the creation of such
file systems, so they can only occur due to corruption.)

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
22 months agolibext2fs: teach ext2fs_open() to reject file systems with an invalid cluster size
Theodore Ts'o [Sat, 6 Aug 2022 05:37:20 +0000 (01:37 -0400)]
libext2fs: teach ext2fs_open() to reject file systems with an invalid cluster size

If the cluster size is smaller than the block size, this can result in
a negative shift, which is undefined.  When such a file system is
opened, immediately return an error indicating that the file system is
corrupted.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
22 months agoresize2fs: fix to respect the environment variable E2FSPROGS_FAKE_TIME
Theodore Ts'o [Thu, 4 Aug 2022 19:18:15 +0000 (15:18 -0400)]
resize2fs: fix to respect the environment variable E2FSPROGS_FAKE_TIME

When performing an off-line resize, if an inode's block map needs to
be updated, resize2fs will update the inode's ctime.  In addition, if
inode numbers need to be renumbered due to the file system shrinking
forcing the inode table to be shrunk, any directories which need to be
modified will have their ctime and mtime updated.

If the E2FSPROGS_FAkE_TIME environment variable is set, when the file
system is opened, fs->now will be set to this value, and resize2fs
needs to use it instead of calling time(0) to get their current time.

Addresses-Google-Bug: 230874381
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
23 months agolibext2fs: check for invalid blocks in ext2fs_punch_blocks()
Theodore Ts'o [Tue, 7 Jun 2022 03:48:37 +0000 (23:48 -0400)]
libext2fs: check for invalid blocks in ext2fs_punch_blocks()

If the extent tree has out-of-range physical block numbers, don't try
to release them.

Also add a similar check in ext2fs_block_alloc_stats2() to avoid a
NULL pointer dereference.

Reported-by: Nils Bars <nils.bars@rub.de>
Reported-by: Moritz Schlögel <moritz.schloegel@rub.de>
Reported-by: Nico Schiller <nico.schiller@rub.de>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
23 months agolibext2fs: check for cyclic loops in the extent tree
Theodore Ts'o [Tue, 7 Jun 2022 02:49:47 +0000 (22:49 -0400)]
libext2fs: check for cyclic loops in the extent tree

In the extent tree handling code in libext2fs, when we go move down
the extent tree, if a cyclic loop is detected, return an error.

Reported-by: Nils Bars <nils.bars@rub.de>
Reported-by: Moritz Schlögel <moritz.schloegel@rub.de>
Reported-by: Nico Schiller <nico.schiller@rub.de>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
23 months agoe2fsck: avoid out-of-bounds write for very deep extent trees
Theodore Ts'o [Tue, 7 Jun 2022 02:44:35 +0000 (22:44 -0400)]
e2fsck: avoid out-of-bounds write for very deep extent trees

The kernel doesn't support extent trees deeper than 5
(EXT4_MAX_EXTENT_DEPTH).  For this reason we only maintain the extent
tree statistics for 5 levels.  Avoid out-of-bounds writes and reads if
the extent tree is deeper than this.

We keep these statistics to determine whether we should rebuild the
extent tree.  If the extent tree is too deep, we don't need the
statistics because we should always rebuild the it.

Reported-by: Nils Bars <nils.bars@rub.de>
Reported-by: Moritz Schlögel <moritz.schloegel@rub.de>
Reported-by: Nico Schiller <nico.schiller@rub.de>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoe2fsck: check for xattr value size integer wraparound
Theodore Ts'o [Mon, 6 Jun 2022 17:34:08 +0000 (13:34 -0400)]
e2fsck: check for xattr value size integer wraparound

When checking an extended attrbiute block for correctness, we check if
the starting offset plus the value size exceeds the end of the block.
However, we weren't checking if the size was too large, and if it is
so large that it triggers a wraparound when we added the starting
offset, we won't notice the problem.  Add the missing check.

Reported-by: Nils Bars <nils.bars@rub.de>
Reported-by: Moritz Schlögel <moritz.schloegel@rub.de>
Reported-by: Nico Schiller <nico.schiller@rub.de>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agolibext2fs: add check for too-short directory blocks
Theodore Ts'o [Mon, 6 Jun 2022 16:03:36 +0000 (12:03 -0400)]
libext2fs: add check for too-short directory blocks

If there is an inline data directory which is smaller than 8 bytes
(which should never happen but for corrupted or fuzzed file systems),
ext2fs_process_dir_block() will now abort EXT2_ET_DIR_CORRUPTED to
avoid an out-of-bounds read.

Reported-by: Nils Bars <nils.bars@rub.de>
Reported-by: Moritz Schlögel <moritz.schloegel@rub.de>
Reported-by: Nico Schiller <nico.schiller@rub.de>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoe2fsck: fix potential out-of-bounds read in inc_ea_inode_refs()
Theodore Ts'o [Mon, 6 Jun 2022 15:39:23 +0000 (11:39 -0400)]
e2fsck: fix potential out-of-bounds read in inc_ea_inode_refs()

If there isn't enough space for a full extended attribute entry,
inc_ea_inode_refs() might end up reading beyond the allocated memory
buffer.

Reported-by: Nils Bars <nils.bars@rub.de>
Reported-by: Moritz Schlögel <moritz.schloegel@rub.de>
Reported-by: Nico Schiller <nico.schiller@rub.de>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoe2fsck: sanity check the journal inode number
Theodore Ts'o [Mon, 30 May 2022 23:17:30 +0000 (19:17 -0400)]
e2fsck: sanity check the journal inode number

E2fsck replays the journal before sanity checking the full superblock.
So it's possible that the journal inode number is not valid relative
to the number of block groups.  So to avoid potentially an array
bounds overrun, sanity check this before trying to find the journal
inode.

Reported-by: Nils Bars <nils.bars@rub.de>
Reported-by: Moritz Schlögel <moritz.schloegel@rub.de>
Reported-by: Nico Schiller <nico.schiller@rub.de>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agolibsupport: avoid possible null dereference in quota_set_sb_inum()
zhanchengbin [Fri, 31 Dec 2021 07:42:17 +0000 (15:42 +0800)]
libsupport: avoid possible null dereference in quota_set_sb_inum()

If the quota type is invalid, quota_sb_inump will return NULL; this
should not cause the program to crash.

Link: https://lore.kernel.org/r/ee0b034c-71f3-63b7-a8de-d8e7760b9545@huawei.com
Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agolibext2fs: fix memory leak in error path while opening test_io manager
zhanchengbin [Fri, 31 Dec 2021 07:43:36 +0000 (15:43 +0800)]
libext2fs: fix memory leak in error path while opening test_io manager

Link: https://lore.kernel.org/r/d0632bbc-9713-38a9-c914-137b702f6ae1@huawei.com
Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoe2fsck: avoid theoretical null dereference in end_problem_latch()
zhanchengbin [Fri, 31 Dec 2021 07:43:10 +0000 (15:43 +0800)]
e2fsck: avoid theoretical null dereference in end_problem_latch()

This should only happen if there is a programming bug, but better safe
than sorry.

Link: https://lore.kernel.org/r/9a9c6658-a8b3-794a-85df-c3bdf0470111@huawei.com
Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agodumpe2fs, resize2fs: avoid memory leak on error path
zhanchengbin [Fri, 31 Dec 2021 07:42:40 +0000 (15:42 +0800)]
dumpe2fs, resize2fs: avoid memory leak on error path

Link: https://lore.kernel.org/r/cbfd9852-bc89-1e83-f101-36fd29a0e70e@huawei.com
Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoe2fsck: handle malloc() failure when computing the log file name
zhanchengbin [Fri, 31 Dec 2021 07:41:41 +0000 (15:41 +0800)]
e2fsck: handle malloc() failure when computing the log file name

Link: https://lore.kernel.org/r/6d2844c7-0fd2-e432-3c7e-bb8de8c8a186@huawei.com
Signed-off-by: zhanchengbin <zhanchengbin1@huawei.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoe2fsck: no parent lookup in disconnected dir
Andreas Dilger [Mon, 13 Dec 2021 06:35:30 +0000 (23:35 -0700)]
e2fsck: no parent lookup in disconnected dir

Don't call into ext2fs_get_pathname() to do a name lookup for a
disconnected directory, since the directory block traversal in
pass1 has already scanned all of the leaf blocks and never finds
the entry, always printing "???".  If the name entry had been
found earlier, the directory would not be disconnected in pass3.

Instead, lookup ".." and print the parent name in the prompt, and
then do not search for the current directory name at all.  This
avoids a useless full directory scan for each disconnected entry,
which can potentially be slow if the parent directory is large.

Separate the recursively looped directory case to a new error code,
since it is a different problem that should use its own descriptive
text, and a proper pathname can be shown in this case.

Lustre-bug-Id: https://jira.whamcloud.com/browse/LU-15330
Change-Id: If17a92689f24f365ca1fbe5c837e7d5f383ebbe5
Signed-off-by: Andreas Dilger <adilger@whamcloud.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoe2fsck: map PROMPT_* values to prompt messages
Andreas Dilger [Wed, 8 Dec 2021 07:51:12 +0000 (00:51 -0700)]
e2fsck: map PROMPT_* values to prompt messages

It isn't totally clear when searching the code for PROMPT_*
constants from problem codes where these messages come from.
Similarly, there isn't a direct mapping from the prompt string
to the constant.

Add comments that make this mapping more clear.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agomisc: fix chattr usage message for project ID
Andreas Dilger [Wed, 1 Dec 2021 22:56:51 +0000 (15:56 -0700)]
misc: fix chattr usage message for project ID

Fix the "chattr -h" usage message to properly document that the
"-p" option takes a project argument, like "-v" takes a version.

Update the man page formatting to emphasize literal strings.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoUse mallinfo2 instead of mallinfo if available
Lukas Czerner [Thu, 17 Feb 2022 09:25:00 +0000 (10:25 +0100)]
Use mallinfo2 instead of mallinfo if available

mallinfo has been deprecated with GNU C library version 2.33 in favor of
mallinfo2 which works exactly the same as mallinfo but with larger field
widths. Use mallinfo2 if available.

Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agolibss: fix possible NULL pointer dereferece on allocation failure
Lukas Czerner [Thu, 17 Feb 2022 09:24:59 +0000 (10:24 +0100)]
libss: fix possible NULL pointer dereferece on allocation failure

Currently in ss_execute_command() we're missng a check to see if the
memory allocation was succesful. Fix it by checking the return from
malloc and returning ENOMEM if it had failed.

[ Removed addition of the SS_ET_ENOMEM entry to the the libss error
  table.  -TYT ]

Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoresize2fs: remove unused variable 'c'
Lukas Czerner [Thu, 17 Feb 2022 09:24:58 +0000 (10:24 +0100)]
resize2fs: remove unused variable 'c'

Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agolibext2fs: add sanity check to extent manipulation
Lukas Czerner [Thu, 21 Apr 2022 17:31:48 +0000 (19:31 +0200)]
libext2fs: add sanity check to extent manipulation

It is possible to have a corrupted extent tree in such a way that a leaf
node contains zero extents in it. Currently if that happens and we try
to traverse the tree we can end up accessing wrong data, or possibly
even uninitialized memory. Make sure we don't do that.

Additionally make sure that we have a sane number of bytes passed to
memmove() in ext2fs_extent_delete().

Note that e2fsck is currently unable to spot and fix such corruption in
pass1.

Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reported-by: Nils Bars <nils_bars@t-online.de>
Addresses: https://bugzilla.redhat.com/show_bug.cgi?id=2068113
Addresses: CVE-2022-1304
Addresses-Debian-Bug: #1010263
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoMerge branch 'maint' into next
Theodore Ts'o [Thu, 13 Jan 2022 13:44:48 +0000 (08:44 -0500)]
Merge branch 'maint' into next

2 years agosetup-schroot: install the udev and systemd packages separately
Theodore Ts'o [Tue, 4 Jan 2022 05:02:22 +0000 (00:02 -0500)]
setup-schroot: install the udev and systemd packages separately

On non-Linux Debian ports (e.g., GNU/Hurd and GNU/kFreeBSD) the udev
and systemd packages don't exist.  So try to install them separately,
so they can fail on their own on those platforms.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agotests: support older versions of timeout in r_corrupt_fs
Theodore Ts'o [Tue, 4 Jan 2022 03:45:37 +0000 (22:45 -0500)]
tests: support older versions of timeout in r_corrupt_fs

Older versions of the timeout program in coreutils don't support the
-v option.  (This is apparently still in use in the GNU/FreeBSD Debain
port since coreutils hasn't built successfully since Coreutils version
8.28.)

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agodebian: suppress the no-debian-changes lintian warning
Theodore Ts'o [Thu, 30 Dec 2021 15:43:49 +0000 (10:43 -0500)]
debian: suppress the no-debian-changes lintian warning

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agotune2fs: implement support for set/get label iocts
Lukas Czerner [Wed, 24 Nov 2021 13:45:42 +0000 (14:45 +0100)]
tune2fs: implement support for set/get label iocts

Implement support for FS_IOC_SETFSLABEL and FS_IOC_GETFSLABEL ioctls.
Try to use the ioctls if possible even before we open the file system
since we don't need it. Only fall back to the old method in the case the
file system is not mounted, is mounted read only in the set label case,
or the ioctls are not suppported by the kernel.

The new ioctls can also be supported by file system drivers other than
ext4. As a result tune2fs and e2label will work for those file systems
as well as long as the file system is mounted. Note that we still truncate
the label exceeds the supported lenghth on extN file system family, while
we keep the label intact for others.

Update tune2fs and e2label as well.

Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoMerge branch 'maint' into next
Theodore Ts'o [Thu, 30 Dec 2021 15:59:52 +0000 (10:59 -0500)]
Merge branch 'maint' into next

2 years agoUpdate release notes, etc., for the 1.46.5 release v1.46.5
Theodore Ts'o [Thu, 30 Dec 2021 05:52:29 +0000 (00:52 -0500)]
Update release notes, etc., for the 1.46.5 release

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agomke2fs.conf.5: fix spelling errors in man page
Theodore Ts'o [Thu, 30 Dec 2021 02:33:27 +0000 (21:33 -0500)]
mke2fs.conf.5: fix spelling errors in man page

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agotst_getsize: use ext2fs_get_device_size2() to support testing large devices
Theodore Ts'o [Tue, 28 Dec 2021 20:17:31 +0000 (15:17 -0500)]
tst_getsize: use ext2fs_get_device_size2() to support testing large devices

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agopo: update zh_CN.po (from translationproject.org)
Wenbin Lv [Tue, 28 Dec 2021 20:16:43 +0000 (15:16 -0500)]
po: update zh_CN.po (from translationproject.org)

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agopo: update uk.po (from translationproject.org)
Yuri Chornoivan [Tue, 28 Dec 2021 20:16:43 +0000 (15:16 -0500)]
po: update uk.po (from translationproject.org)

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agopo: update sv.po (from translationproject.org)
Göran Uddeborg [Tue, 28 Dec 2021 20:16:43 +0000 (15:16 -0500)]
po: update sv.po (from translationproject.org)

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agopo: update sr.po (from translationproject.org)
Мирослав Николић [Tue, 28 Dec 2021 20:16:42 +0000 (15:16 -0500)]
po: update sr.po (from translationproject.org)

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agopo: update ms.po (from translationproject.org)
Sharuzzaman Ahmat Raslan [Tue, 28 Dec 2021 20:16:42 +0000 (15:16 -0500)]
po: update ms.po (from translationproject.org)

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agopo: update es.po (from translationproject.org)
Antonio Ceballos [Tue, 28 Dec 2021 20:16:42 +0000 (15:16 -0500)]
po: update es.po (from translationproject.org)

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoreisze2fs: sanity check free block group counts when calculating minimum size
Theodore Ts'o [Tue, 28 Dec 2021 17:33:15 +0000 (12:33 -0500)]
reisze2fs: sanity check free block group counts when calculating minimum size

If one or more block group descriptor's free blocks count is insane,
it's possible this can lead to a infinite loop in the function
calculate_minimum_resize_size(), which is called by resize2fs -P or
resize2fs -M.

Add some sanity checks to avoid this.  In the case where the file
system is corrupt, this will result in resize2fs -P reporting an
incorrect value, but that's OK, since when we try to do an actual
resize operation, resize2fs requires that the file system be freshly
checked using e2fsck.

https://github.com/tytso/e2fsprogs/issues/94

Fixes: ac94445fc01f ("resize2fs: make minimum size estimates more reliable for mounted fs")
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agosetup-schroot: add some additional packages needed to build debian packages
Theodore Ts'o [Tue, 21 Dec 2021 19:55:32 +0000 (14:55 -0500)]
setup-schroot: add some additional packages needed to build debian packages

On older Debian systems, "apt-get build-dep e2fsprogs" might not bring
in all of the packages needed to build in the most recent versions of
e2fsprogs.  So explicitly try to install some additional packages
including dh-exec, udev, systemd, and cron.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agolibuuid: try to use getrandom() or getentropy() if available
Theodore Ts'o [Tue, 21 Dec 2021 19:28:51 +0000 (14:28 -0500)]
libuuid: try to use getrandom() or getentropy() if available

If getrandom() or getentropy() is available, use these interfaces in
favor of opening /dev/[u]random.  This avoids a potential TSAN problem
that could potentially cause a fd leak when trying to open
/dev/urandom.  (Which is not a disaster, but these interfaces are more
foolproof and avoids needing to open a file descriptor in a library,
which is a good thing.)

Addresses-Google-Bug: #198050608
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoe2fsck: update the bg_checksum after fixing problems in the bg descriptor
Theodore Ts'o [Sat, 11 Dec 2021 03:40:40 +0000 (22:40 -0500)]
e2fsck: update the bg_checksum after fixing problems in the bg descriptor

Otherwise, we break the block group descriptor's checksum, and while
this gets fixed by e2fsck, it results unnecessary messages printed or
questions asked of the system administrator.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agolibext2fs: don't old the CACHE_MTX while doing I/O
Theodore Ts'o [Thu, 9 Dec 2021 15:55:54 +0000 (10:55 -0500)]
libext2fs: don't old the CACHE_MTX while doing I/O

A report a deadlock problem caused by I/O errors (caused by e2fsck's
error handler trying to write to a bad block to perform a forced
rewrite) uncovered that we were holding the CACHE_MTX while doing read
operations.  This serialized read operations which destroyed the
performance benefits from doing parallel bitmap loading (or the
parallel e2fsck processing under development).

So restructure the code in unix_read_blk64() so that the read is
always done into the user-provided buffer, and then copied into the
cache afterwards.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoe2fsck: skip sorting extents if there are no valid extents
Harshad Shirwadkar [Wed, 17 Nov 2021 16:50:15 +0000 (08:50 -0800)]
e2fsck: skip sorting extents if there are no valid extents

At the end of a fast commit replay, e2fsck tries merging extents in a
inode. This patch fixes a bug in this logic where we were continuing
this action even if there were no extents to merge resulting in
accessing illegal memory.

Signed-off-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
2 years agolibext2fs: fix missing unlocks in the error path in unix_set_blksize()
ryancaicse [Mon, 1 Nov 2021 05:42:56 +0000 (13:42 +0800)]
libext2fs: fix missing unlocks in the error path in unix_set_blksize()

https://github.com/tytso/e2fsprogs/pull/83

Signed-off-by: Ryan Cai <ryancaicse@gmail.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoext2fs: avoid re-reading inode multiple times
Andreas Dilger [Wed, 7 Apr 2021 07:50:23 +0000 (01:50 -0600)]
ext2fs: avoid re-reading inode multiple times

Reduce the number of times that the inode is read from storage.
Factor ext2fs_xattrs_read() into a new ext2fs_xattrs_read_inode()
function that can accept an in-memory inode, and call that from
within ext2fs_xattrs_read() and in e2fsck_pass1() when the inode
is already available.

Similarly, in e2fsck_pass4() avoid re-reading the inode multiple
times in disconnect_inode(), check_ea_inode(), and in the main
function body if possible.

Signed-off-by: Andreas Dilger <adilger@dilger.ca>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agomke2fs: add extended option for prezeroed storage devices
Sarthak Kukreti [Mon, 27 Sep 2021 10:39:10 +0000 (03:39 -0700)]
mke2fs: add extended option for prezeroed storage devices

This patch adds an extended option "assume_storage_prezeroed" to
mke2fs. When enabled, this option acts as a hint to mke2fs that the
underlying block device was zeroed before mke2fs was called.  This
allows mke2fs to optimize out the zeroing of the inode table and the
journal, which speeds up the filesystem creation time.

Additionally, on thinly provisioned storage devices (like Ceph,
dm-thin, newly created sparse loopback files), reads on unmapped
extents return zero. This property allows mke2fs (with
assume_storage_prezeroed) to avoid pre-allocating metadata space for
inode tables for the entire filesystem and saves space that would
normally be preallocated for zero inode tables.

Tests
-----
1) Running 'mke2fs -t ext4' on 10G sparse files on an ext4
filesystem drops the time taken by mke2fs from 0.09s to 0.04s
and reduces the initial metadata space allocation (stat on
sparse file) from 139736 blocks (545M) to 8672 blocks (34M).

2) On ChromeOS (running linux kernel 4.19) with dm-thin
and 200GB thin logical volumes using 'mke2fs -t ext4 <dev>':

- Time taken by mke2fs drops from 1.07s to 0.08s.
- Avoiding zeroing out the inode table and journal reduces the
  initial metadata space allocation from 0.48% to 0.01%.
- Lazy inode table zeroing results in a further 1.45% of logical
  volume space getting allocated for inode tables, even if no file
  data is added to the filesystem. With assume_storage_prezeroed,
  the metadata allocation remains at 0.01%.

[ Fixed regression test to work on newer versions of e2fsprogs -- TYT ]

Signed-off-by: Sarthak Kukreti <sarthakkukreti@chromium.org>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agodumpe2fs, debugfs, e2image: Add support for orphan file
Jan Kara [Wed, 25 Aug 2021 22:11:34 +0000 (00:11 +0200)]
dumpe2fs, debugfs, e2image: Add support for orphan file

Print inode number of orphan file in outputs, dump e2image file to
filesystem image.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agotune2fs: Add support for orphan_file feature
Jan Kara [Wed, 25 Aug 2021 22:11:33 +0000 (00:11 +0200)]
tune2fs: Add support for orphan_file feature

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoe2fsck: Add support for handling orphan file
Jan Kara [Wed, 25 Aug 2021 22:11:32 +0000 (00:11 +0200)]
e2fsck: Add support for handling orphan file

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agomke2fs: Add support for orphan_file feature
Jan Kara [Wed, 25 Aug 2021 22:11:31 +0000 (00:11 +0200)]
mke2fs: Add support for orphan_file feature

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agolibext2fs: Support for orphan file feature
Jan Kara [Wed, 25 Aug 2021 22:11:30 +0000 (00:11 +0200)]
libext2fs: Support for orphan file feature

Add support for creating and deleting orphan file and a couple of
utility functions that will be used in other tools.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agotests: Add option to print diff output of failed tests
Lukas Czerner [Thu, 2 Sep 2021 10:58:52 +0000 (12:58 +0200)]
tests: Add option to print diff output of failed tests

Add variable $PRINT_FAILED which when set will print the diff output of
a failed test.

Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoresize2fs: optimize resize2fs_calculate_summary_stats()
Theodore Ts'o [Tue, 14 Sep 2021 19:05:45 +0000 (15:05 -0400)]
resize2fs: optimize resize2fs_calculate_summary_stats()

Speed up an off-line resize of a 10GB file system to 64TB located on
tmpfs from 90 seconds to 16 seconds by extracting block group bitmaps
using a population count function to count the blocks in use instead
checking each bit in the block bitmap.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoresize2fs: adjust new size of the file system to allow a successful resize
Theodore Ts'o [Tue, 14 Sep 2021 12:56:46 +0000 (08:56 -0400)]
resize2fs: adjust new size of the file system to allow a successful resize

The previous commit in this series (commit 50088b1996cc: "resize2fs:
attempt to keep the # of inodes valid by removing the last bg") allows
a successful off-line resize of a file system with the default 16k
inode ratio to be grown to support a 64TB storage device by dropping
the last block group so the number of inodes is just below the maximum
2**32-1 number of inodes.

However, this is not a complete solution, for two reasons.  First,
this adjustment happens after resize2fs has started potentially making
changes to the file system in the off-line (unmounted) case, which
means resize2fs will do a lot of unnecessary work.  Secondly, in the
on-line resize case, passing the original requested size to the kernel
causes the kernel fail the online resize request.

So teach resize2fs to adjust the new size of the file system much
earlier, which avoids both problems.

Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Leah Rumancik <leah.rumancik@gmail.com>
2 years agoresize2fs: attempt to keep the # of inodes valid by removing the last bg
Theodore Ts'o [Tue, 14 Sep 2021 12:41:33 +0000 (08:41 -0400)]
resize2fs: attempt to keep the # of inodes valid by removing the last bg

If a the 10GB file system (with the default inode ratio size of 16k)
is resized to 64TB, the number of inodes will become 2**32 --- one
above the maximum allowed number of inodes of 2**32-1.  In
adjust_fs_info(), we already try drop the last block group if there
isn't sufficient space in the last block group to support the metadata
for that block group.  So if dropping the last block group allows the
number of inodes to valid, we should try that as well.  In some cases
this will mean resizing a file system to 64TB will result in it be
resized to a size of 64TB - 128MB, which is close enough for
government work.

Addresses-Google-Bug: 199105099
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Reviewed-by: Leah Rumancik <leah.rumancik@gmail.com>
2 years agotests: update expect files for f_large_dir and f_large_dir_csum
Lukas Czerner [Tue, 24 Aug 2021 12:10:20 +0000 (14:10 +0200)]
tests: update expect files for f_large_dir and f_large_dir_csum

Update expect files for f_large_dir and f_large_dir_csum tests to
include the warning about missing y2038 support with 128-byte inodes.

Fixes: a23b50cd ("mke2fs: warn about missing y2038 support when formatting fresh ext4 fs")
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agotests: update expect files for f_mmp_garbage
Lukas Czerner [Tue, 24 Aug 2021 12:10:19 +0000 (14:10 +0200)]
tests: update expect files for f_mmp_garbage

Update expect file for f_mmp_garbage test to work correctly with the
new default 256 inode size.

Fixes: d730be5ceeba ("tests: update mke2fs.conf to create 256 byte inodes by default")
Signed-off-by: Lukas Czerner <lczerner@redhat.com>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoquota: Drop dead code
Jan Kara [Mon, 23 Aug 2021 15:41:28 +0000 (17:41 +0200)]
quota: Drop dead code

Drop unused function from quota support code.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agodebugfs: Fix headers for quota commands
Jan Kara [Mon, 23 Aug 2021 15:41:27 +0000 (17:41 +0200)]
debugfs: Fix headers for quota commands

list_quota and get_quota commands have 'blocks' header while what they
actually show is a used space in bytes. Fix the header to state 'space'
instead.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agotests: Expand test checking quota and orphan processing interaction
Jan Kara [Mon, 23 Aug 2021 15:41:26 +0000 (17:41 +0200)]
tests: Expand test checking quota and orphan processing interaction

Expand f_orphquot test to also check handling of quotas for non-root
user and verify that quota limits are properly preserved over orphan
replay.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoe2fsck: Do not trash user limits when processing orphan list
Jan Kara [Mon, 23 Aug 2021 15:41:25 +0000 (17:41 +0200)]
e2fsck: Do not trash user limits when processing orphan list

When e2fsck was loading quotas to process orphan list, it was loading
only quota usage. However subsequent quota writeout has effectively
overwritten quota limits, loosing them forever. Make sure quota limits
are preserved over orphan replay.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agotune2fs: Fix conversion of quota files
Jan Kara [Mon, 23 Aug 2021 15:41:24 +0000 (17:41 +0200)]
tune2fs: Fix conversion of quota files

When tune2fs is enabling quota feature, it looks for old-style quota
files and tries to transfer limits stored in these files into newly
created hidded quota files. However the code doing the transfer setups
the quota scan wrongly and instead of transferring limits we transfer
usage. So not only quota limits are lost (at least they can still be
recovered from the old quota files) but also usage information may be
wrong if the accounting in e2fsprogs does not exactly match the
accounting in quota-tools (which is actually the case). Fix the setup of
the quota scan.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
2 years agoquota: Rename quota_update_limits() to quota_read_all_dquots()
Jan Kara [Mon, 23 Aug 2021 15:41:23 +0000 (17:41 +0200)]
quota: Rename quota_update_limits() to quota_read_all_dquots()

quota_update_limits() is a misnomer because what it actually does is
that it updates 'usage' counters and leaves 'limit' counters intact.
Rename quota_update_limits() to quota_read_all_dquots() and while
changing prototype also add a flags argument so that callers can control
which quota information is actually updated from the disk.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Theodore Ts'o <tytso@mit.edu>