| 1 |
---------- Forwarded message ---------- |
| 2 |
From: Alice Ferrazzi <alicef@g.o> |
| 3 |
Date: Sun, Jun 11, 2017 at 7:28 PM |
| 4 |
Subject: [GSoC 2017 - week 02] Report summary |
| 5 |
To: gentoo-soc@l.g.o |
| 6 |
Cc: Mike Pagano <mpagano@g.o>, Gokturk Yuksek <gokturk@g.o> |
| 7 |
|
| 8 |
|
| 9 |
|
| 10 |
Hello everyone, |
| 11 |
Here is my week 02 report summary. |
| 12 |
|
| 13 |
Day 04 |
| 14 |
|
| 15 |
1-What was my plan for the day? |
| 16 |
|
| 17 |
working on kpatch packaging |
| 18 |
|
| 19 |
2- Did I achieve it? if not, what do I intend to do to remedy |
| 20 |
that? |
| 21 |
|
| 22 |
Working on fixing the kpatch ebuild pull request. |
| 23 |
I fixed the license like Gokturk said. |
| 24 |
but also: |
| 25 |
fixed metadata.xml indentation |
| 26 |
sorted lexical |
| 27 |
Added new line after HOMEPAGE |
| 28 |
|
| 29 |
now i'm rebuilding the kernel for testing it again with some |
| 30 |
new changes |
| 31 |
|
| 32 |
3- what will I do next day? |
| 33 |
Going on on fixing the kpatch package |
| 34 |
https://github.com/gentoo/gentoo/pull/4844 |
| 35 |
|
| 36 |
The next day I will work on the review of the kpatch if there |
| 37 |
is, trying to package ksplice maybe and making my |
| 38 |
gitweb.gentoo.org repository. |
| 39 |
|
| 40 |
Day 05 |
| 41 |
|
| 42 |
1-What was my plan for the day? |
| 43 |
|
| 44 |
working on kpatch packaging |
| 45 |
|
| 46 |
2- Did I achieve it? if not, what do I intend to do to remedy that? |
| 47 |
|
| 48 |
Kpatch is now upstream in the gentoo repository |
| 49 |
https://github.com/gentoo/gentoo/pull/4844 |
| 50 |
|
| 51 |
also created the repository for Gentoo Kernel Live patching system |
| 52 |
and created the mirror repository on git.gentoo.org |
| 53 |
|
| 54 |
now called elivepatch because I thought we can probably use it not |
| 55 |
only for security patch but also for more general patching. |
| 56 |
This repository will try to solve the trouble of live patching |
| 57 |
the kernel on different Gentoo systems. |
| 58 |
https://github.com/aliceinwire/elivepatch |
| 59 |
|
| 60 |
3- what will I do next day? |
| 61 |
Try to understand how Kpatch works |
| 62 |
|
| 63 |
Day 06 |
| 64 |
|
| 65 |
1-What was my plan for the day? |
| 66 |
|
| 67 |
use kpatch and write some documentation. |
| 68 |
|
| 69 |
2- Did I achieve it? if not, what do I intend to do to remedy that? |
| 70 |
|
| 71 |
Kpatch examples are bit systemd hardcode dependant but anyway |
| 72 |
they works. |
| 73 |
Here kpatch is using systemctl for start kpatch systemd service |
| 74 |
https://github.com/dynup/kpatch/blob/810f9243826219cca9276d364512bb |
| 75 |
88411aa863/kpatch/kpatch#L289 |
| 76 |
Would be nice to make it work also with openrc. |
| 77 |
I will start making some documentation for Kpatch and add it |
| 78 |
to the gentoo wiki. |
| 79 |
|
| 80 |
|
| 81 |
3- what will I do next day? |
| 82 |
Going on on working with kpatch and writing documentation. |
| 83 |
|
| 84 |
Day 07 |
| 85 |
|
| 86 |
1-What was my plan for the day? |
| 87 |
|
| 88 |
use kpatch and write some documentation. |
| 89 |
|
| 90 |
2- Did I achieve it? if not, what do I intend to do to remedy that? |
| 91 |
|
| 92 |
today I got aware that instead of unset ARCH i could just run |
| 93 |
set_arch_to_kernel Gentoo script. |
| 94 |
And that was one of the change of today. |
| 95 |
I also went on starting to use kpatch and checking that everything |
| 96 |
is working correctly. |
| 97 |
Kpatch is actually working correctly, it could load module but I |
| 98 |
had some problem |
| 99 |
unloading module |
| 100 |
|
| 101 |
kpatch - hot patch module management |
| 102 |
|
| 103 |
loading the module: |
| 104 |
gsoc-2017 testmod (gentoo*) # kpatch load testmod.ko |
| 105 |
loading patch module: testmod.ko |
| 106 |
dmesg output: |
| 107 |
[98196.595313] kpatch: loading out-of-tree module taints kernel. |
| 108 |
[98196.595329] kpatch: module verification failed: signature and/or |
| 109 |
required key missing - tainting kernel |
| 110 |
|
| 111 |
unloading the module: |
| 112 |
gsoc-2017 testmod (gentoo*) # kpatch unload testmod.ko |
| 113 |
kpatch: patch module testmod.ko is not loaded |
| 114 |
|
| 115 |
but I could unload correctly using rmmod testmod |
| 116 |
|
| 117 |
|
| 118 |
kpatch list is also working as expected: |
| 119 |
gsoc-2017 testmod (gentoo*) # kpatch list |
| 120 |
Loaded patch modules: |
| 121 |
|
| 122 |
Installed patch modules: |
| 123 |
testmod (4.9.16-gentoo) |
| 124 |
|
| 125 |
and kpatch info |
| 126 |
gsoc-2017 testmod (gentoo*) # kpatch info testmod |
| 127 |
Patch information for testmod: |
| 128 |
filename: /var/lib/kpatch/4.9.16-gentoo/testmod.ko |
| 129 |
license: GPL |
| 130 |
depends: |
| 131 |
vermagic: 4.9.16-gentoo SMP mod_unload modversions |
| 132 |
|
| 133 |
|
| 134 |
kpatch install and uninstall looks also working but is not sure |
| 135 |
|
| 136 |
gsoc-2017 testmod (gentoo*) # kpatch install testmod.ko |
| 137 |
installing testmod.ko (4.9.16-gentoo) |
| 138 |
/usr/local/sbin/kpatch: line 289: systemctl: command not found |
| 139 |
|
| 140 |
kpatch uninstall testmod.ko |
| 141 |
uninstalling testmod.ko (4.9.16-gentoo) |
| 142 |
|
| 143 |
|
| 144 |
The problems became more obvious by using kpatch-build |
| 145 |
kpatch-build is the script for automatically make live patch kernel |
| 146 |
module from |
| 147 |
patch files. |
| 148 |
|
| 149 |
gsoc-2017 kpatch (master) # kpatch-build -t -v |
| 150 |
/boot/vmlinuz-4.4.44-gentoo examples/tcp_cubic-better- |
| 151 |
follow-cubic-curve-converted.patch --skip-gcc-check |
| 152 |
WARNING: Skipping gcc version matching check (not recommended) |
| 153 |
ERROR: Unsupported distribution. |
| 154 |
|
| 155 |
and that is because is hard checking the distribution from |
| 156 |
/etc/os-release |
| 157 |
so I started writing a patch |
| 158 |
https://github.com/aliceinwire/kpatch/commit/ |
| 159 |
9f7f236cfe9d4592267a6b08483751ad5a1da069 |
| 160 |
|
| 161 |
but others problem come out like claming that some binary is not |
| 162 |
there |
| 163 |
I'm thinking that one way to go is by using the -s option Specify |
| 164 |
kernel source directory |
| 165 |
gsoc-2017 kpatch (gentoo) # kpatch-build/kpatch-build -t -v |
| 166 |
/boot/vmlinuz-4.4.44-gentoo examples/tcp_cubic-better- |
| 167 |
follow-cubic-curve-converted.patch --skip-gcc-check |
| 168 |
WARNING: Skipping gcc version matching check (not recommended) |
| 169 |
ERROR: kernel-debuginfo-4.9.16-gentoo not installed. |
| 170 |
|
| 171 |
by using -s option is going on with the work but still stopping |
| 172 |
little after |
| 173 |
gsoc-2017 kpatch (gentoo) # kpatch-build/kpatch-build -t -s |
| 174 |
/usr/src/linux-4.9.16-gentoo/vmlinux -v /boot/vmlinuz-4.9.16-gentoo |
| 175 |
examples/tcp_cubic-better-follow-cubic-curve-converted.patch |
| 176 |
--skip-gcc-check |
| 177 |
WARNING: Skipping gcc version matching check (not recommended) |
| 178 |
ERROR: can't find supporting tools. |
| 179 |
|
| 180 |
and this is because of this code |
| 181 |
The tools mentioned are the TOOLSDIR/SCRIPTDIR |
| 182 |
118 if [[ -e "$SCRIPTDIR/create-diff-object" ]]; then |
| 183 |
119 # git repo |
| 184 |
120 TOOLSDIR="$SCRIPTDIR" |
| 185 |
121 DATADIR="$(readlink -f $SCRIPTDIR/../kmod)" |
| 186 |
122 elif [[ -e "$SCRIPTDIR/../libexec/kpatch/create-diff-object" |
| 187 |
]]; then |
| 188 |
123 # installation path |
| 189 |
124 TOOLSDIR="$(readlink -f $SCRIPTDIR/../libexec/kpatch)" |
| 190 |
125 DATADIR="$(readlink -f $SCRIPTDIR/../share/kpatch)" |
| 191 |
126 else |
| 192 |
127 return 1 |
| 193 |
128 fi |
| 194 |
|
| 195 |
SCRIPTDIR="$(readlink -f $(dirname $(type -p $0)))" |
| 196 |
|
| 197 |
|
| 198 |
+ find_dirs |
| 199 |
+ [[ -e /root/kpatch/kpatch-build/create-diff-object ]] |
| 200 |
+ [[ -e /root/kpatch/kpatch-build/../libexec/kpatch/create-diff-object |
| 201 |
]] |
| 202 |
+ return 1 |
| 203 |
+ die 'can'\''t find supporting tools' |
| 204 |
|
| 205 |
and i'm still investigating this part |
| 206 |
|
| 207 |
Yesterday I tried to use some real patch downloaded from linux git |
| 208 |
and I could load also such patch after converting it with the |
| 209 |
testing tools in the |
| 210 |
testing folder of kpatch |
| 211 |
but other dmesg saying I loaded the patch I couldn't understand |
| 212 |
what it was doing |
| 213 |
|
| 214 |
so would be better to write some fake module by myself with some |
| 215 |
real output |
| 216 |
like just adding some printk |
| 217 |
printk("Module being loaded.\n"); |
| 218 |
would be enough i susptect |
| 219 |
|
| 220 |
after changing the patch file like this |
| 221 |
|
| 222 |
--- testmod_drv.c 2017-06-09 02:37:43.980000000 +0000 |
| 223 |
+++ testmod_drv.c.orig 2017-06-09 02:36:56.100000000 +0000 |
| 224 |
@@ -11,7 +11,7 @@ int value = 2; |
| 225 |
static ssize_t value_show(struct kobject *kobj, |
| 226 |
struct |
| 227 |
kobj_attribute *attr, char *buf) |
| 228 |
{ |
| 229 |
- return sprintf(buf, "%d\n", value+1); |
| 230 |
+ return sprintf(buf, "%d\n", value); |
| 231 |
} |
| 232 |
|
| 233 |
static struct kobj_attribute testmod_value_attr = |
| 234 |
__ATTR_RO(value); |
| 235 |
+ @@ -29,7 +29,6 @@ static int testmod_init(void) |
| 236 |
+ kobject_put(testmod_kobj); |
| 237 |
+ return ret; |
| 238 |
+ } |
| 239 |
+ - printk("Module being loaded.\n"); |
| 240 |
+ |
| 241 |
+ return 0; |
| 242 |
|
| 243 |
i could get correctly: |
| 244 |
[98196.595313] kpatch: loading out-of-tree module taints kernel. |
| 245 |
[98196.595329] kpatch: module verification failed: signature and/or |
| 246 |
required key missing - tainting kernel |
| 247 |
[207928.744194] Module being loaded. |
| 248 |
|
| 249 |
|
| 250 |
generated patch |
| 251 |
gsoc-2017 testmod (master*) # file testmod.ko |
| 252 |
testmod.ko: ELF 64-bit LSB relocatable, x86-64, version 1 (SYSV), |
| 253 |
BuildID[sha1]=9f7ab4eec1a8ee28b6e0bd7850388aca35da65b6, not stripped, with |
| 254 |
debug_info |
| 255 |
|
| 256 |
gcc-version get checked correctly |
| 257 |
|
| 258 |
readelf -p .comment /usr/src/linux-4.9.16-gentoo/vmlinux |
| 259 |
|
| 260 |
String dump of section '.comment': |
| 261 |
[ 0] GCC: (Gentoo 5.4.0-r3 p1.3, pie-0.6.5) 5.4.0 |
| 262 |
|
| 263 |
I also wondered if would be nice to try also ksplice but from what |
| 264 |
i could see |
| 265 |
looks like ksplice works only for kernel 3.* |
| 266 |
if this is the case so I don't think is needed for our project |
| 267 |
as Gentoo Kernel is moving to releasing manly Gentoo kernel 4.* |
| 268 |
https://github.com/jirislaby/ksplice/issues/7 |
| 269 |
|
| 270 |
so in conclusion using kpatch by hand is working as far as I know |
| 271 |
|
| 272 |
3- what will I do next day? |
| 273 |
Going on writing script for making kpatch-build work on Gentoo |
| 274 |
|
| 275 |
Day 08 |
| 276 |
|
| 277 |
What i have to do today? |
| 278 |
|
| 279 |
Going on investigating using kpatch in different machine. |
| 280 |
|
| 281 |
What i did today? |
| 282 |
|
| 283 |
gsoc-2017 kpatch (gentoo*) # kpatch-build/kpatch-build -s |
| 284 |
/usr/src/linux-4.9.16-gentoo/ -v /usr/src/linux-4.9.16-gentoo/vmlinux |
| 285 |
examples/test.patch --skip-gcc-check |
| 286 |
WARNING: Skipping gcc version matching check (not recommended) |
| 287 |
Using source directory at /usr/src/linux-4.9.16-gentoo |
| 288 |
Testing patch file |
| 289 |
checking file fs/xfs/libxfs/xfs_bmap.c |
| 290 |
Reading special section data |
| 291 |
Building original kernel |
| 292 |
Building patched kernel |
| 293 |
Extracting new and modified ELF sections |
| 294 |
xfs_bmap.o: changed function: xfs_bmap_add_extent_hole_real |
| 295 |
xfs_bmap.o: changed function: xfs_bmap_split_extent_at |
| 296 |
xfs_bmap.o: changed function: xfs_bmap_add_extent_unwritten_real |
| 297 |
xfs_bmap.o: changed function: xfs_bmap_add_extent_delay_real |
| 298 |
xfs_bmap.o: changed function: xfs_bmse_merge.isra.4.constprop.10 |
| 299 |
xfs_bmap.o: changed function: xfs_bmse_shift_one.constprop.7 |
| 300 |
xfs_bmap.o: changed function: xfs_bmap_del_extent |
| 301 |
xfs_bmap.o: changed function: xfs_bmapi_read |
| 302 |
xfs_bmap.o: changed function: xfs_bmapi_write |
| 303 |
xfs_bmap.o: changed function: __xfs_bunmapi |
| 304 |
xfs_bmap.o: changed function: xfs_bmap_shift_extents |
| 305 |
Patched objects: vmlinux |
| 306 |
Building patch module: kpatch-test.ko |
| 307 |
SUCCESS |
| 308 |
|
| 309 |
kpatch load kpatch-test.ko |
| 310 |
dmesg: |
| 311 |
[Jun 9 22:57] kpatch: tainting kernel with TAINT_LIVEPATCH |
| 312 |
[ +0.000002] kpatch: loaded patch module 'kpatch_test' |
| 313 |
|
| 314 |
looks like is a success for now! |
| 315 |
|
| 316 |
I will start to build the patch for a different system |
| 317 |
|
| 318 |
So I will set a second Gentoo enviroment with 4.9.16 on qemu |
| 319 |
|
| 320 |
* CONFIG_FUNCTION_TRACER must be enabled in the kernel's config file |
| 321 |
* CONFIG_KALLSYMS_ALL must be enabled in the kernel's config file |
| 322 |
|
| 323 |
made the livepatch worker for elivepatch waiting compiling the kernel |
| 324 |
https://github.com/aliceinwire/elivepatch/blob/master/elivepatch_server/ |
| 325 |
resources/livepatch.py |
| 326 |
|
| 327 |
kernel compile is finished and I moved the kpatch.ko to a different |
| 328 |
machine |
| 329 |
and it dosen't work... |
| 330 |
|
| 331 |
this are the difference in the two kernel configuration files |
| 332 |
|
| 333 |
--- /usr/src/linux/.config 2017-06-10 19:51:48.930000000 +0000 |
| 334 |
+++ config 2017-06-10 21:01:18.340000000 +0000 |
| 335 |
@@ -621,7 +621,7 @@ CONFIG_LEGACY_VSYSCALL_EMULATE=y |
| 336 |
# CONFIG_CMDLINE_BOOL is not set |
| 337 |
CONFIG_MODIFY_LDT_SYSCALL=y |
| 338 |
CONFIG_HAVE_LIVEPATCH=y |
| 339 |
-CONFIG_LIVEPATCH=y |
| 340 |
+# CONFIG_LIVEPATCH is not set |
| 341 |
CONFIG_ARCH_ENABLE_MEMORY_HOTPLUG=y |
| 342 |
CONFIG_ARCH_ENABLE_MEMORY_HOTREMOVE=y |
| 343 |
CONFIG_USE_PERCPU_NUMA_NODE_ID=y |
| 344 |
@@ -3758,7 +3758,11 @@ CONFIG_MESSAGE_LOGLEVEL_DEFAULT=4 |
| 345 |
# |
| 346 |
# Compile-time checks and compiler options |
| 347 |
# |
| 348 |
-# CONFIG_DEBUG_INFO is not set |
| 349 |
+CONFIG_DEBUG_INFO=y |
| 350 |
+# CONFIG_DEBUG_INFO_REDUCED is not set |
| 351 |
+# CONFIG_DEBUG_INFO_SPLIT is not set |
| 352 |
+CONFIG_DEBUG_INFO_DWARF4=y |
| 353 |
+# CONFIG_GDB_SCRIPTS is not set |
| 354 |
CONFIG_ENABLE_WARN_DEPRECATED=y |
| 355 |
CONFIG_ENABLE_MUST_CHECK=y |
| 356 |
CONFIG_FRAME_WARN=2048 |
| 357 |
|
| 358 |
|
| 359 |
ok Let's try with the same kernel config and then give the summary. |
| 360 |
looks like is failing as same |
| 361 |
|
| 362 |
kernel_dev ~ # kpatch load testmod.ko |
| 363 |
loading core module: /usr/local/lib/kpatch/4.9.16-gentoo/kpatch.ko |
| 364 |
insmod: ERROR: could not insert module |
| 365 |
/usr/local/lib/kpatch/4.9.16-gentoo/kpatch.ko: |
| 366 |
Invalid module format |
| 367 |
kpatch: failed to load core module |
| 368 |
|
| 369 |
|
| 370 |
Tried to use insmod: |
| 371 |
[ 883.481889] kpatch: disagrees about version of symbol |
| 372 |
module_layout |
| 373 |
[ 940.346754] testmod: loading out-of-tree module taints kernel. |
| 374 |
[ 940.346767] testmod: module verification failed: signature |
| 375 |
and/or required key missing - tainting kernel |
| 376 |
[ 940.346921] Module being loaded. |
| 377 |
|
| 378 |
module being loaded. is the correct output of the live patch. |
| 379 |
|
| 380 |
about the kpatch ebuild i divided 'make all' in src_compile and 'make |
| 381 |
install' in src_install |
| 382 |
on Gokturk suggestion |
| 383 |
https://github.com/aliceinwire/elivepatch/blob/master/sys-kernel/kpatch/ |
| 384 |
kpatch-0.4.0.ebuild |
| 385 |
|
| 386 |
|
| 387 |
Different live patch system: |
| 388 |
kGraft |
| 389 |
This system of live patching consist in keeping active the old |
| 390 |
function and the new function and redirect the call to the old function to |
| 391 |
the new one. |
| 392 |
Not compatible with 4.x |
| 393 |
kpatch |
| 394 |
Is using stop_machine() for halting all the cpu and examine the |
| 395 |
stack for check that nothing is using the function, and then use ftrace |
| 396 |
handler for bypass the oldfunction in the new one. |
| 397 |
Here the problem come with function that are always active and this |
| 398 |
means that such functions cannot be changed (like irq_thread()). |
| 399 |
LIVEPATCH |
| 400 |
Recently introduced in the kernel is similar to kpatch but with |
| 401 |
less feature. |
| 402 |
|
| 403 |
|
| 404 |
kpatch code module. |
| 405 |
For each function patched by the module we must: |
| 406 |
Call stop_machine |
| 407 |
Ensure that no task has the old function in its call stack |
| 408 |
Add the new function address to kpatch_func_hash |
| 409 |
|
| 410 |
-- |
| 411 |
アリス フェッラッツィ |
| 412 |
Alice Ferrazzi |
| 413 |
|
| 414 |
Gentoo Kernel Project Leader |
| 415 |
Mail: Alice Ferrazzi <alicef@g.o> |
| 416 |
PGP: 2E4E 0856 461C 0585 1336 F496 5621 A6B2 8638 781A |
Archives