Linus Torvalds [Mon, 18 Jan 2010 23:09:21 +0000 (00:09 +0100)]
libblkid: disable read-ahead when probing device files
Read-ahead doesn't work very well on device probing, and can hurt a lot
when we do essentially random accesses on very slow devices. So disable it
if possible.
Karel Zak [Mon, 18 Jan 2010 14:43:25 +0000 (15:43 +0100)]
libblkid: read() optimization for small devices
- don't read the begin (69kB) of the device by one large read()
- fill in the SB buffer dynamically
- use extra buffer for FATs root dir entries (FAT FS label)
on small devices to avoid large reads
Karel Zak [Mon, 18 Jan 2010 11:50:36 +0000 (12:50 +0100)]
libblkid: restrict RAID/FS proving for small devices (1.4MiB)
- don't ignore @size in blkid_probe_set_device()
- ignore extremely small devices (<= 1024 bytes, e.g. extended partitions)
- don't use safe probing for small devices
- don't probe for RAIDs on small devices
Karel Zak [Wed, 6 Jan 2010 10:12:43 +0000 (11:12 +0100)]
fdisk: sleep-after-sync and fsync usage
It seems that sleep() after sync() is unnecessary legacy. It's very
probably unnecessary since kernel 1.3.20. For example the libparted
does not to use sleep() at all.
It seems that more important is fsync() usage in fdisks. For more
details see
DRBD is the Distributed Replicated Block Device, a replication service for low
level block devices.
The attached patch provides libblkid detection for v08 type drbd devices
(v08 is the current one).
[kzak@redhat.com: - port to libblkid 2.17
- use BLKID_USAGE_RAID flag
- remove BLKID_IDINFO_TOLERANT flag
- note that DRBD is supported since kernel v2.6.33-rc1]
Signed-off-by: Bastian Friedrich <bastian.friedrich@collax.com> Signed-off-by: Karel Zak <kzak@redhat.com>
Karel Zak [Mon, 4 Jan 2010 10:34:13 +0000 (11:34 +0100)]
sfdisk: confused about disk size
The size of disk in the sfdisk command is based on number of cylinders
(this is probably legacy from CHS epoch). That's wrong because
partitions are addressed in sectors (LBA), so cylinders don't provide
necessary resolution (granularity).
On Sat, Jan 02, 2010 at 01:01:16PM +0100, Giulio wrote:
> $ cat /sys/block/sda/size
> 184549376
>
> $ sfdisk -d /dev/sda > part.dump
> $ cat part.dump
> # partition table of /dev/sda
> unit: sectors
>
> /dev/sda1 : start= 2048, size= 2097152, Id=83
> /dev/sda2 : start= 2099200, size= 12582912, Id=83
> /dev/sda3 : start= 14682112, size= 84934656, Id=83
> /dev/sda4 : start= 99616768, size= 84932608, Id=83
>
>
> $ sfdisk -L /dev/sda < part.dump
> Checking that no-one is using this disk right now ...
> OK
>
> Disk /dev/sda: 11487 cylinders, 255 heads, 63 sectors/track
> Old situation:
> Units = cylinders of 8225280 bytes, blocks of 1024 bytes, counting from 0
>
> Device Boot Start End #cyls #blocks Id System
> /dev/sda1 0+ 130- 131- 1048576 83 Linux
> /dev/sda2 130+ 913- 784- 6291456 83 Linux
> /dev/sda3 913+ 6200- 5287- 42467328 83 Linux
> /dev/sda4 6200+ 11487- 5287- 42466304 83 Linux
> Warning: given size (84932608) exceeds max allowable size (84921887)
Mike Frysinger [Sat, 26 Dec 2009 19:56:54 +0000 (14:56 -0500)]
pg: command enters infinite loop
In a multibyte locale such as en_GB.UTF-8, the pg command cannot handle files
containing a form feed character (ASCII 0x0c) at the start of a line. The
program enters an infinite loop.
I've traced the problem to the function endline_for_mb in file pg.c. The code
assumes that the libc function wcwidth will return a nonnegative value, which
is not true for a form feed character. wcwidth returns -1 and the unsigned
variable "pos" goes into underflow.
I'll attach a patch which tests whether the character is printable before
calling wcwidth. If not, it uses instead the width of the constant L'?' which
is later used to replace nonprintable characters. I trust that we can assume
printability of this constant :-)
Steps to Reproduce:
1. Select a multibyte locale (tested with en_GB.UTF-8)
2. Create a file with a form feed character (0x0c) at the start of a line.
3. Try to display this file using the pg command.
Reported-by: Mark Calderbank <m.calderbank@iname.com> Reported-by: Mike Frysinger <vapier@gentoo.org>
Addresses: https://bugs.gentoo.org/297717 Signed-off-by: Karel Zak <kzak@redhat.com>
Karel Zak [Thu, 10 Dec 2009 10:59:46 +0000 (11:59 +0100)]
lib: bug (typo) in function MD5Final()
On Wed, Dec 09, 2009 at 10:08:38PM +0000, Jochen Voss wrote:
> while experimenting with coccinelle, I accidentally found what I
> believe is a bug in util-linux-ng release 2.17-rc2 (downloaded
> today). The problem is the following code in lib/md5.c (around line
> 153):
>
> void MD5Final(unsigned char digest[16], struct MD5Context *ctx)
> {
> [...]
> memset(ctx, 0, sizeof(ctx)); /* In case it's sensitive */
> }
>
> The third argument of memset should probably be the size of 'struct
> MD5Context' instead of the size of the pointer. So my guess is
> that the memset line should be
>
> memset(ctx, 0, sizeof(*ctx)); /* In case it's sensitive */
>
> instead. I don't know whether this actually causes a problem,
> but the comment makes it seem possible that it does.
Note, this typo does not have any impact on the utils in the
util-linux-ng project, because we don't use MD5 for any security
sensitive data or cryptographic stuff. The typo also does not have any
impact to the final MD5 hashes.
Reported-by: Jochen Voss <voss@seehuhn.de> Signed-off-by: Karel Zak <kzak@redhat.com>
Mike Frysinger [Mon, 7 Dec 2009 14:18:17 +0000 (15:18 +0100)]
flock: fix hang when parent ignores SIGCHLD
If flock is executed from a process which has set SIGCHLD to SIG_IGN, then
flock will eat cpu and hang indefinitely if given a command to execute.
So before we fork(), make sure to set SIGCHLD handling back to the default
so that the later waitpid() doesn't freak out on us.
[kzak@redhat.com: - add a check for waitpid() return value]
Signed-off-by: Mike Frysinger <vapier@gentoo.org> Signed-off-by: Karel Zak <kzak@redhat.com>
Karel Zak [Fri, 4 Dec 2009 14:45:19 +0000 (15:45 +0100)]
build-sys: rewrite TLS detection
* use more robust tls.m4 from gcc project
The old version (from util-linux-ng) used AC_TRY_COMPILE. That's
wrong. We need to use AC_RUN_IFELSE to check that the result is
link-able and executable.
The new version also test it TLS really works in multi-thread
applications.
* we need to detect TLS usability for cross-compiling
* this new version supports __thread keyword only, it seems that we
needn't to care about anything other
Ludwig Nussel [Fri, 27 Nov 2009 09:15:53 +0000 (10:15 +0100)]
fsck: honor nofail option in fsck
analog to mount gracefully ignoring non existing devices if the "nofail"
option is specified in fstab, also have fsck -A skip them. This way it's
possible to have devices optionally not available during boot but still
have them fsck'd if they are there.
Signed-off-by: Ludwig Nussel <ludwig.nussel@suse.de>
On Wed, Nov 18, 2009 at 03:33:12PM +0000, Daniel Drake wrote:
> Booting into a system this way just leads to problems because
> you cannot remount the root read-only at shutdown (leading to unclean
> shutdowns).
> Miklos Szeredi pointed out a trick to turn any directory into a
> mount point which avoids this problem. Therefore we can simplify
> switch_root again and simply document that its users should set
Karel Zak [Thu, 19 Nov 2009 14:56:12 +0000 (15:56 +0100)]
mount: check for unsuccessful read-only bind mounts
Linux kernel allows to use MS_RDONLY together with MS_BIND,
unfortunately the MS_RDONLY is silently ignored and the target
mountpoint is still read-write. Then we have 'ro' in mtab and 'rw' in
/proc/mounts.
This patch checks for this situation by access(2) or futimens(2)
(change atime) and mtab is properly updated and user informed.
Reported-by: Terry Burton <tez@terryburton.co.uk> Signed-off-by: Karel Zak <kzak@redhat.com>
Karel Zak [Wed, 4 Nov 2009 14:14:04 +0000 (15:14 +0100)]
fdisk: offer aligned first sector
Typical "new partition" dialog looks like:
Partition number (1-4): 1
First sector (4-818687, default 4):
^^^^^^^^^
The range (e.g. 4-818687) depends on fdisk mode (DOS/non-DOS), but the
default value should be always aligned.
For example RAID5 device in the DOS mode:
Disk /dev/md0: 419 MB, 419168256 bytes
2 heads, 4 sectors/track, 102336 cylinders, total 818688 sectors
Units = sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 65536 bytes
Disk identifier: 0x081479c3
....
Command (m for help): n
Command action
e extended
p primary partition (1-4)
p
Partition number (1-4): 1
First sector (4-818687, default 128): <---- !!!
Using default value 128
Last sector, +sectors or +size{K,M,G} (128-818687, default 818687): +10M
Command (m for help): p
....
Device Boot Start End Blocks Id System
/dev/md0p1 128 20607 10240 83 Linux
Karel Zak [Tue, 3 Nov 2009 11:00:23 +0000 (12:00 +0100)]
fdisk: print info and recommendations about alignment
* inform user that phy.sector > log.sector
* warn user when alignment_offset is not provided does, DOS-compatible
mode is enabled and the default geo.sectors are not aligned
* suggest to change display units to sectors (oh yes, fdisk
default are cylinders...)
Karel Zak [Thu, 29 Oct 2009 10:25:59 +0000 (11:25 +0100)]
fdisk: read topology info from libblkid
and print ('p' command) info about logical and physical sectors and
alignment_offset.
minimum_io_size
we don't use physical sector size directly, because on RAIDs is
better to use minimum_io_size (aka stripe chunk size). For disk drives
is minimum_io_size the same value as physical sector size.
alignment_offset
For compatibility with legacy operating systems some vendors provide
disks where logical and physical sectors are aligned at sector 63
(= geometry.sectors). In other words the physical 4KB sectors
start at LBA -1. Then the physical boundary is at:
Karel Zak [Tue, 10 Nov 2009 11:48:34 +0000 (12:48 +0100)]
Revert "libblkid: fix buffer usage in FAT prober"
This patch is unnecessary, the library uses two buffers and the VFAT
superblock is always stored in the probe->sbbuf buffer which is never
overwritten. It seems that FAT fsinfo is also in the superblock
buffer. I was too paranoid :-) Sorry.
projects / util-linux / log