[RHEL8,COMMIT] modules: use kvmalloc when creating sysfs attributes for ELF sections

Submitted by Konstantin Khorenko on Sept. 22, 2020, 12:31 p.m.

Details

Message ID 202009221231.08MCVxmU007968@finist-co8.sw.ru
State New
Series "modules: use kvmalloc when creating sysfs attributes for ELF sections"
Headers show

Commit Message

Konstantin Khorenko Sept. 22, 2020, 12:31 p.m.
The commit is pushed to "branch-rh8-4.18.0-193.6.3.vz8.4.x-ovz" and will appear at https://src.openvz.org/scm/ovz/vzkernel.git
after rh8-4.18.0-193.6.3.vz8.4.6
------>
commit 9175123cf76678492bf559facbce177e90e5c07e
Author: Evgenii Shatokhin <eshatokhin@virtuozzo.com>
Date:   Tue Sep 22 15:31:59 2020 +0300

    modules: use kvmalloc when creating sysfs attributes for ELF sections
    
    A kernel module containing a ReadyKernel patch could have lots of ELF
    sections: one per each new or patched function, one per each new static
    or global variable, etc.
    
    The kernel creates a sysfs file
    /sys/module/<module_name>/sections/<section_name> for each loaded
    section when the patch module is being loaded, see add_sect_attrs() in
    kernel/module.c. A big chunk of memory is allocated for all these with
    kzalloc.
    
    For the ReadyKernel patches we have already released, the amount of
    memory could be as high as 34528 bytes (48 + 80 * 431 loaded sections),
    which is a 4th order allocation. 3rd order allocations are also common
    here, see https://jira.sw.ru/browse/PSBM-95050.
    
    Not only it is a waste (contiguous memory is not needed there), but the
    allocation may also fail when the memory is fragmented. ReadyKernel
    patches are often used in the systems with rather significant uptime, so
    the fragmentation is possible.
    
    It could be better if the patch modules did not use too many ELF
    sections. However, the KPatch maintainers pointed out
    (https://github.com/dynup/kpatch/pull/1131) that the same problem would
    affect regular kernel modules as well after FGKASLR has been merged into
    the mainline kernel. Combining the sections of the kernel modules
    destroys the purpose of FGKASLR, so, it was agreed that we probably
    should just switch to kvmalloc+kvfree in add_sect_attrs/free_sect_attrs.
    
    Details and discussion:
    https://www.spinics.net/lists/live-patching/msg06364.html
    
    https://jira.sw.ru/browse/PSBM-108017
    Signed-off-by: Evgenii Shatokhin <eshatokhin@virtuozzo.com>
---
 kernel/module.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Patch hide | download patch | download mbox

diff --git a/kernel/module.c b/kernel/module.c
index d4702f0d711a..e58ad01de8bf 100644
--- a/kernel/module.c
+++ b/kernel/module.c
@@ -1492,7 +1492,7 @@  static void free_sect_attrs(struct module_sect_attrs *sect_attrs)
 
 	for (section = 0; section < sect_attrs->nsections; section++)
 		kfree(sect_attrs->attrs[section].name);
-	kfree(sect_attrs);
+	kvfree(sect_attrs);
 }
 
 static void add_sect_attrs(struct module *mod, const struct load_info *info)
@@ -1510,7 +1510,7 @@  static void add_sect_attrs(struct module *mod, const struct load_info *info)
 			+ nloaded * sizeof(sect_attrs->attrs[0]),
 			sizeof(sect_attrs->grp.attrs[0]));
 	size[1] = (nloaded + 1) * sizeof(sect_attrs->grp.attrs[0]);
-	sect_attrs = kzalloc(size[0] + size[1], GFP_KERNEL);
+	sect_attrs = kvzalloc(size[0] + size[1], GFP_KERNEL);
 	if (sect_attrs == NULL)
 		return;