5beb95d lua: additionally return name when looking up sections
ff33bb2 lua: support extended section notation
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Including version.mk sets PKG_CONFIG_DEPENDS to config entries used for
VERSION_SED command. We should keep these configs to make sure package
gets refreshed when needed.
Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
All the relevant options used for distfeeds.conf are part of base-files,
so it makes more sense to move the file there as well.
This has the added benefit that the we can share the opkg package again,
reducing the amount of target specific packages.
Acked-by: Jo-Philipp Wich <jo@mein.io>
Signed-off-by: Jonas Gorski <jonas.gorski@gmail.com>
9c13096 ubus: Remove unnecessary memset calls.
6d1ea6c libubus: Fix deletion from context's object AVL tree when removing object
e02813b ubusd: don't free messages in ubus_send_msg() anymore
be146ad ubusd: rename goto label from `error` to `out`
27d712d ubusd_monitor: alloc & free the buffer outside of the loop
5f87f54 ubusd: move global retmsg per client
Signed-off-by: Felix Fietkau <nbd@nbd.name>
7a49632 logd: use uloop instead of ustream_fd for syslog
69d6542 logd: only create pipe in stream mode
df30c8c logread: terminate after EOF
bdcacad logd: implement oneshot mode for stream log read
4a10d4e logread: use oneshot mode without -f, wait for logd to close
ea3d7fa logd: enforce line length limit for ubus based log messages as well
960a29d logread: remove leftover debug code
a081904 logread: fix line buffer size
2c0d9cf logd: move stripping of newlines to log_add()
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Without this change, when a user disables seccomp support in .config,
procd does not get recompiled unless the package is cleaned manually.
It is because when -D option is missing from cmake command line, cmake
uses cached value from the previous run where seccomp was enabled.
Signed-off-by: Michal Sojka <sojkam1@fel.cvut.cz>
ee582d1 instance: properly compare and reload respawn config
260a4cd utrace: Start the tracee only after uloop initialization
520ad3c utrace: Switch all logging to ulog
1c48104 utrace: Support non-contiguous syscall numbers
582cf97 utrace: Forward SIGTERM to the traced process
32534f7 utrace: Report ptrace errors
ccde3fb seccomp: Improve error message
7f9b174 preload-seccomp: Use proper log level for error messages
e3c4302 Start seccomp-enabled services via seccomp-trace
5e4ad02 seccomp: Log seccomp violations with utrace
2661b2f utrace: Use PTHREAD_SEIZE instead of PTHREAD_TRACEME
b5d53c6 utrace: Deliver signals to traced processes
b416ed9 utrace: Support tracing multi-threaded processes and vfork
8b7d47a utrace: Trace processes across forks
c6b6ec6 utrace: Sort syscalls by number of invocations
592c532 Update trace attribute
c8faedc Do not disable seccomp when configuration is not found
017f3a1 utrace: Fix off-by-one errors
5acaf15 utrace: Fix environment initialization
Signed-off-by: John Crispin <john@phrozen.org>
b1bc8d5 kmodloader: log error message in case of out of memory
f346111 kmodloader: lift restriction on module alias info
f1ef2c3 kmodloader: fix possible segfaults
9cb63df kmodloader: fix endianess check
2cff779 kmodloader: Check module endian before loading
d54f38a kmodloader/get_module_info: initialized aliases to make it more clean
a0b6fef kmodloader: insmod: fix a memoryleak in error case
278c4c4 kmodloader/get_module_name: null-terminate the string
16f7e16 syslog: remove unnecessary sizeof struct between messages
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Properly quote the arguments so that you can register a service with TXT
entries that contains spaces.
Example:
procd_add_mdns myservice tcp 9999 "key=descriptive text field 1" \
"another=something equally verbose"
Output before:
$ avahi-browse -r -v _myservice._tcp
_myservice._tcp local
hostname = [blah.local]
address = [192.168.255.74]
port = [9999]
txt = ["verbose" "equally" "another=something" "1" "field" "text" "key=descriptive"]
Output now:
$ avahi-browse -r -v _myservice._tcp
_myservice._tcp local
hostname = [blah.local]
address = [192.168.255.74]
port = [9999]
txt = ["another=something equally verbose" "key=descriptive text field 1"]
Signed-off-by: Karl Palsson <karlp@etactica.com>
Commits since last 2017-07-11:
4bd8601 pkg_parse: fix segfault when parsing descriptions with leading newlines
Fixes FS#933.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Expose "term_timeout" parameter in procd.sh to allow init scripts to
request a longer termination timeout.
This is required to fix FS#859 in a later commit.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Adjust default permissions and ownership of /dev/tty* nodes from
0600/root:root to 0660/root:tty in order to support granting
unprivileged user access when needed.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
453116e system: introduce new attribute board_name
e5b963a preinit: define _GNU_SOURCE
e5ff8ca upgraded: cmake: Find and include uloop.h
f367ec6 hotplug: fix a memory leak in handle_button_complete()
796ba3b service/service_stopped(): fix a use-after-free
79bbe6d system: return legacy board name
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
We can reuse the kill_remaining and run_ramfs facilities of the stage2 run
by upgraded.
Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
We always want to support staged upgrades now, so it's better to include
upgraded into the main package. /lib/upgrade/nand.sh is moved to
base-files.
The procd-nand-firstboot package is removed for now, it may return later
as a separate package.
Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
992b796 rcS: add missing fcntl.h include
63789e5 init: add support for sysupgrades triggered from preinit
5b1fb35 Remove code that has become unnecessary after sysupgrade changes
5918b6d upgraded: add support for passing a "command" argument to stage2
056d8dd upgraded: link dynamically, chroot during exec
7c6cf55 system: always support staged sysupgrade
d42b21e procd/rcS: Use /dev/null as stdin
e0098d4 service/instance: add an auto start option
1247db1 procd: Log initscript output prefixed with script name
8d720b2 procd: Don't use syslog before its initialization
2555474 procd: Add missing \n in debug message
Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
Use install_bin to copy upgraded with all dependencies. The old name
/tmp/upgraded is temporarily retained as a symlink to avoid breaking
things.
Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
88d48d5 libfstools: silence mkfs.{ext4,f2fs}
a19f2b3 build: disable the format-truncation warning error to fix gcc 7 build errors
633a8d0 libfstools: fix multiple volume_identify usages with the same volume
c43ae11 fstools: use -Wno-format-truncation instead of -Wno-error=format-truncation
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Changes since last version
546bc72 pkg: alternatives support
7a96972 libbb: xreadlink: fix memory leak on failure case
3f13edd pkg_run_script: use pkg->dest in half installed case
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
Update to latest Git in order to import the following fixes:
1d0263b check_data_file_clashes_change: remove duplicated offline root in paths
a00a6a9 buildReplaces: do not add duplicated replacees
9fbedd8 opkg: compare-versions doesn't need any state
45b54f6 opkg_cmd: fix segmentation fault in opkg_compare_versions_cmd()
93de62b opkg_cmd: fix return of opkg compare-versions
51275a8 pkg_src_list_push: remove unused function
947d3d4 pkg_src_list_pop: remove unused function
980cfb2 str_list_push: remove unused function
52c31c1 str_list_prev: remove unused function
7d24212 str_list_last: remove unused function
48142a3 conffile_list_pop: remove unused function
46c5de5 pkg_dest_list_push: remove unused functions
368bb62 nv_pair_list_prev: remove unused function
2985c00 nv_pair_list_last: remove unused function
f5082ac conffile_list_pop: remove unused function
89bf8b9 conffile_list_push: remove unused function
23d31fb active_list_sort: remove unused function
7fe45f2 active_list_add_depend: remove unused function
86b584d pkg_dependence_satisfied: remove unused function
816d330 Fix typo in pkg_hash_fetch_best_installation_candidate.
07f3b02 set_flags_from_control: remove function
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
20c16fc cmake: Make blockd link against libjson-c
92b4c2c libfstools: add basic documentation of mount functions
Signed-off-by: John Crispin <john@phrozen.org>
With this patch, the nand.sh script no longer overwrites
CI_KERNELPART and CI_UBIPART if they have been set before.
This allows platforms specify alternative values for these
variables.
Signed-off-by: Christian Lamparter <chunkeey@gmail.com>
continue work started here: http://patchwork.ozlabs.org/patch/520859
Extend /etc/config/system with parameters to set the default respawn treshold and respawn timeout
for procd launched services that have respawn enabled.
This results in cleaner init scripts, while making sure services have respawn parameters set.
Signed-off-by: Claudiu Brasovean <cbrasho@gmail.com>
8f218f5 procd: service gets deleted when its last instance is freed
35209a0 procd: update modprobe path
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Update to latest Git in order to import the following fixes:
19070b6 unarchive: fix possible segmentation fault in deb_extract()
758527e pkg: forward "provided_by_hand" flag in pkg_merge()
1fdc2ca pkg: mark bit fields in pkg_t unsigned
15fc1ee opkg_cmd: avoid null pointer dereference in opkg_what_depends_conflicts_cmd()
439c89e opkg_cmd: fix endless loop in whatprovides/whatreplaces
0fb3e9f pkg_depends: propagate virtual provided detail flag to providers
793155f pkg_depends: fix unsatisfied dependency installation order
b1dc770 libopkg: fix some overly aggressive line wrapping in pkg_depends.c
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
c09e4f0 ubusd: fix incomplete copy of shared buf during queue-ing
453b87f cli: add support for subscribing to objects
6eb3c96 cli: do not use default timeout for listen
dfe3383 libubus: reset ctx->sock.error when doing ubus reconnect
34c6e81 cli: fix listen_timeout compile issue
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Update to latest Git head to get rid of a stray printf() causing unwanted
output in the "opkg list-upgradable" command.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
A missing list pointer initialization caused opkg to ignore conffile entries
from status files, breaking the conffile tracking on the target.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
c553354 cmake: fix typo
8973576 kmodloader: fix not being able to find some modules
fce9382 cmake: Check for getrandom system call
Signed-off-by: Felix Fietkau <nbd@nbd.name>
The switch to cmake caused the -DHAVE_USIGN flag to get lost, disabling
compilation of the correspondinf support code.
Update to latest Git head which enables usign support by default.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Switch to our own fork of opkg to significantly reduce the required amount
of memory when updating lists or installing packages.
Preliminary tests showed a usage drop of about 90% during these operations,
from ~3.7MB with unmodified opkg to ~360KB with our custom fork.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
5f91241 procd: add cancel_timeout on rc scripts when a runtime_timeout is specified
961dc69 procd: stop service using SIGKILL if SIGTERM failed to do so
Signed-off-by: John Crispin <john@phrozen.org>
Default trigger action timeout was added to procd.sh in commit f88e3a4c0
(procd: add default timeout for reload trigger actions)
However, the timeout value was not placed under the correct JSON-script
array nesting level and thus did not apply.
To fix this and make the timeout actually apply to the reload triggers,
we place it in the correct scope, that is the per-trigger array.
Fixes: f88e3a4c0a
Signed-off-by: Denis Osvald <denis.osvald@sartura.hr>
Adds the following fixes:
91acde6 libubus: do not modify uloop_cancelled
763b9b2 libubus: reset ctx->sock.eof to fix reconnect issues
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Update ubox to latest Git HEAD in order to import the following fixes:
ac2d43e kmodloader: support '-q' quiet option
f8d3d16 ubox: Add an option for more accurate timestamps in log
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Fixes build issues with some toolchains that don't add lcrypt in the default
search paths:
CMakeFiles/rpcd.dir/session.c.o: In function `rpc_login_test_password':
build_dir/target-mipsel-linux-gnu/rpcd-2016-12-03-0577cfc1/session.c:823: undefined reference to `crypt'
collect2: error: ld returned 1 exit status
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Update ubox to latest Git head in order to import the following fixes:
14839f0 kmodloader: make insert_module() idempotent
6e3c6dc kmodloader: add module alias awareness
9371411 kmodloader: fix out-of-bound access when parsing .modinfo
a62c946 kmodloader: modprobe: skip possible command line arguments
46a4b5f kmodloader: log to kmsg when loading directories of modules
eacc426 kmodloader: remove redundant glob wildcard char
8488bb5 ubox: Initialize conditionally uninitialized variable
db070f1 ubox: Fix some memory leaks
acc48b5 kmodloader: Fix typo in error message
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
This is how other Linux distributions are doing and kernel
modprobe_path[] of request_module() also has a default value of
/sbin/modprobe
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
A given signal-name is now converted to the corresonding number. In general
it's good style to use names (readability) and it's more portable: signal
numbers can be architecture-dependent, so we are more safe giving names.
A real world example is signal 10, which is BUS on ramips and USR1 on PPC.
All users of 'procd_send_signal' must change their code to reflect this.
Signed-off-by: Bastian Bittorf <bb@npl.de>
Clarify opkg's messages related to downloads:
* more visible error message for package list download failure
* separate error message for signature file download error
* if wget returns 4, signal the network error more clearly
* remove '.' from end of filenames and URLs
* try signature check only if the package list was downloaded ok.
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
- Adds support for passing file descriptors in ubus invoke requests
- Fixes clearing pending timers on ubus_shutdown()
- Fixes checking the amount of written data in ubusd
- Fixes an ubusd crash when trying to subscribe to system objects
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Currently system log is always included as a part of ubox. Add logd as a
seperate package and add it to default packages list.
Signed-off-by: Andrej Vlasic <andrej.vlasic@sartura.hr>
Signed-off-by: Luka Perkov <luka.perkov@sartura.hr>
opkg doesn't have BUILD_VARIANTs anymore, so the previously defined
PKG_BUILD_DIR would lead to a weird 'opkg-' path component.
Signed-off-by: Matthias Schiffer <mschiffer@universe-factory.net>
This reverts the following commits:
fbe522d120278ad007ee863888e44f96daf6352fcfd83555fc
This seems to trigger some mconf bugs when built with all feeds
packages, so I will try to find a less intrusive solution before the
release.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
It has never been used by default (due to being too bloated), and it is
properly replaced by usign (which has been the default for a long time
now).
Remove this feature to simplify the build system
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Brings in the following changes:
84b530a732b1 libfstools: Check return values for fread and system
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Opkg's builtin decompression code is unsuitable to process nested archives as
it uses a single shared state and relies on undefined seek behaviour for pipes.
Rework the extraction logic to use the external gzip command as I/O filter for
decompressing data and remove the builtin inflate code entirely.
This shrinks the final opkg binary by about 4KB and results in less runtime
memory consumption due to efficient use of vfork() and less copy-on-write
operations in the forked child.
Rework by Felix: create a thread that relays data to the gzip process
instead of using a fragile poll loop
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Opkg's builtin decompression code is unsuitable to process nested archives as
it uses a single shared state and relies on undefined seek behaviour for pipes.
Rework the extraction logic to use the external gzip command as I/O filter for
decompressing data and remove the builtin inflate code entirely.
This shrinks the final opkg binary by about 4KB and results in less runtime
memory consumption due to efficient use of vfork() and less copy-on-write
operations in the forked child.
Rework by Felix: create a thread that relays data to the gzip process
instead of using a fragile poll loop
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Add procd_send_signal which allows to send an optional specified
kill signal to one specified or all instances of a given service.
By default SIGHUP is sent if no signal is specified
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Update procd to latest HEAD in order to introduce support for services signals:
- Adds a new service.signal ubus call to send a kill() signal to one or all
running instances of a given service
- Adds a new "reload_signal" property which allows service init scripts to
request procd to send a specific kill() signal on reload, instead of
stopping and restarting running processes
Also fixes some potential memory leaks reported by cppcheck and an environment
variable corruption in the trace command.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
To make sure we properly restart services on upgrade we need to
call the prerm script of the old package, in case the init script
changes (or vanishes).
Signed-off-by: Jonas Gorski <jonas.gorski@gmail.com>
Acked-by: Jo-Philipp Wich <jo@mein.io>
Allow scripts from the package to be upgraded to be aware of being
upgraded.
Signed-off-by: Jonas Gorski <jonas.gorski@gmail.com>
Acked-by: Jo-Philipp Wich <jo@mein.io>
This will be used to append extra information to images which allows the
system to verify if an image is compatible with the system.
The extra data is appended to the end of the image, where it will be
ignored when upgrading from systems that do not process this data yet:
If the image is a squashfs or jffs2 image, the extra data will land
after the end-of-filesystem marker, where it will be overwritten once
the system boots for the first timee.
If the image is a sysupgrade tar file, tar will simply ignore the extra
data when unpacking.
The layout of the metadata/signature chunks is constructed in a way
that the last part contains just a magic and size information, so that
the tool can quickly check if any valid data is present without having
to do a pattern search throughout the full image.
Chunks also contain CRC32 information to detect file corruption, even
when the image is not signed.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
The libblkid probe support in fstools git head requires blkid/blkid.h for
compilation, so add a build dependency on util-linux which provides libblkid.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
Commit f5c741b5e0 updated procd to a more recent version, but did not
change the hash of the tar. Update it to the one matching the file on
the download servers.
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
* Change git packages to xz
* Update mirror checksums in packages where they are used
* Change a few source tarballs to xz if available upstream
* Remove unused lines in packages we're touching, requested by jow- and blogic
* We're relying more on xz-utils so add official mirror as primary source, master site as secondary.
* Add SHA256 checksums to multiple git tarball packages
Signed-off-by: Daniel Engberg <daniel.engberg.lists@pyret.net>
Enable CONFIG_PROCD_ZRAM_TMPFS compatibility via two changes to list_cpu_idx():
* detect if /tmp is being used by /dev/zram0; if yes, offset initial value by 1 to skip first zram device.
* hot-add /dev/zram1, if not already present.
Signed-off-by: Conn O'Griofa >connogriofa@gmail.com>
This reverts commit 763f5d7873.
Currently the vfork() code path in opkg is broken and relies on unsupported
ftello() / fseeko() operations on pipes - we need to restructure the code
before we can reconsider this approach.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
This reverts commit 02e3c718e9.
Currently the vfork() code path in opkg is broken and relies on unsupported
ftello() / fseeko() operations on pipes - we need to restructure the code
before we can reconsider this approach.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
src/linksys_bootcount.c misses to include stdint.h.
Apparently musl doesn't mind and includes this header by default,
but glibc does not and causes the build to fail.
Signed-off-by: Josua Mayer <josua.mayer97@gmail.com>
sysupgrade immediately reboots after flashing an image and doesn't
allow to unmount filesystems. At least in case the image used for
sysupgrade is stored on a FAT formatted usb flash drive, the following
warning is printed during the next mount of the flash drive:
FAT-fs (sda1): Volume was not properly unmounted. Some data may be
corrupt. Please run fsck.
Although a data corruption during read operations is unlikely, there is
no need to scare the users.
Signed-off-by: Mathias Kresin <dev@kresin.me>
Open/close triggers array around service_triggers call to make using
multiple triggers easier to deal with.
The API was quite confusing, because some functions contained implicit
trigger open/close calls and some didn't.
Signed-off-by: Felix Fietkau <nbd@nbd.name>
Fixes duplicate ubiblock entries being listed and improves
find_mount_point to also match against a block device's
major:minor numbers (needed e.g. for /dev/root).
Signed-off-by: Daniel Golle <daniel@makrotopia.org>
Small cleanup. I initially though /dev/kmsg was used for dmsg(and journald
on desktops) but this seems not to be the case. dmsg is still accessible
as non-root(gives output) which begs the question what does this do? Some
googling reveals that permissions are set to 600 for some embedded systems
while 644 for others. I can't find any justification for the latter. Might
as well err on the side of caution.
Signed-off by: Rosen Penev <rosenp@gmail.com>
This changes the default permissions for /dev entries to be more similar to
a desktop distro. Taken from the defaults of Arch Linux and Ubuntu. Also
changed some that were nonsensical. For example, all 660 permissions on
desktop distros were of the form root:x where x is something other than
root. As such, 660 is useless for LEDE where the specific group is missing.
audio seems to be the only group that isn't.
Signed-off by: Rosen Penev <rosenp@gmail.com>
/etc/init.d/fstab had some warnings in it to let users know that they
should call "block mount" directly. these ended up in the bootlog.
fix this by splitting the code into boot() and start(). this way the system
will boot without warning but manually starting the script will lead to the
message.
fixes FS#3
Reported-by: Hannu Nyman <hannu.nyman@iki.fi>
Signed-off-by: John Crispin <john@phrozen.org>
Increase the default system log buffer size
from 16 kB (default both in logd source and in the startup script)
to 64 kB by adjusting the default value in startup script.
Signed-off-by: Hannu Nyman <hannu.nyman@iki.fi>
Some SSL applications requires a certificates bundle rather
than a directory containing certificates. For thos applications
we build the ca-bundle package
Signed-off-by: Daniel Dickinson <lede@daniel.thecshore.com>
the recent fixes to make mount_root work during failsafe caused lots of
unwanted side effects. use the new preinit sentinel file to detect if
we are in preinit. this will also work if logged in via ssh.
Signed-off-by: John Crispin <john@phrozen.org>
Currently system log is always included as a part of ubox.
Add logd as a seperate package and add it to default packages list.
Signed-off-by: Andrej Vlasic <andrej.vlasic@sartura.hr>
SVN-Revision: 49285
So far fixtrx was calculating checksum over amount of data matching
partition erase size. It was mostly a workaround of checksum problem
after changing anything in initial TRX content (e.g. formatting JFFS2).
Its main purpose was to make bootloader accept modified TRX. This didn't
provide much protection of flash data against corruption.
This new option lets caller request calculating checksum over a bigger
amount of data. It may be used e.g. to include whole kernel data for
checksum and hopefully make bootloader go info failsafe mode if
something goes wrong.
Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
We plan to adjust usage of the main buffer to allow reading custom
amount of data for CRC32. This means we need another buffer that will be
always block aligned.
Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
1) Put sanity checks in one place
2) Respect provided offset
3) Read only as much data as needed for MD5 calculation
Thanks to the last change this is a great speedup and memory saver. On
devices with NAND flash we were allocating & reading about 128 MiB while
something about 8 MiB is enough.
Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
