03 Feb, 2011
1 commit
-
Make the tracepoints more robust, making them solid enough to handle compiler
changes by not relying on anything based on compiler-specific behavior with
respect to structure alignment. Implement an approach proposed by David Miller:
use an array of const pointers to refer to the individual structures, and export
this pointer array through the linker script rather than the structures per se.
It will consume 32 extra bytes per tracepoint (24 for structure padding and 8
for the pointers), but are less likely to break due to compiler changes.History:
commit 7e066fb8 tracepoints: add DECLARE_TRACE() and DEFINE_TRACE()
added the aligned(32) type and variable attribute to the tracepoint structures
to deal with gcc happily aligning statically defined structures on 32-byte
multiples.One attempt was to use a 8-byte alignment for tracepoint structures by applying
both the variable and type attribute to tracepoint structures definitions and
declarations. It worked fine with gcc 4.5.1, but broke with gcc 4.4.4 and 4.4.5.The reason is that the "aligned" attribute only specify the _minimum_ alignment
for a structure, leaving both the compiler and the linker free to align on
larger multiples. Because tracepoint.c expects the structures to be placed as an
array within each section, up-alignment cause NULL-pointer exceptions due to the
extra unexpected padding.(this patch applies on top of -tip)
Signed-off-by: Mathieu Desnoyers
Acked-by: David S. Miller
LKML-Reference:
CC: Frederic Weisbecker
CC: Ingo Molnar
CC: Thomas Gleixner
CC: Andrew Morton
CC: Peter Zijlstra
CC: Rusty Russell
Signed-off-by: Steven Rostedt
23 Dec, 2010
2 commits
-
The commit:
84e1c6bb38eb318e456558b610396d9f1afaabf0
x86: Add RO/NX protection for loadable kernel modulesBroke the function tracer with this output:
------------[ cut here ]------------
WARNING: at kernel/trace/ftrace.c:1014 ftrace_bug+0x114/0x171()
Hardware name: Precision WorkStation 470
Modules linked in: i2c_core(+)
Pid: 86, comm: modprobe Not tainted 2.6.37-rc2+ #68
Call Trace:
[] warn_slowpath_common+0x85/0x9d
[] ? __process_new_adapter+0x7/0x34 [i2c_core]
[] ? __process_new_adapter+0x7/0x34 [i2c_core]
[] warn_slowpath_null+0x1a/0x1c
[] ftrace_bug+0x114/0x171
[] ? __process_new_adapter+0x7/0x34 [i2c_core]
[] ftrace_process_locs+0x1ae/0x274
[] ? __process_new_adapter+0x7/0x34 [i2c_core]
[] ftrace_module_notify+0x39/0x44
[] notifier_call_chain+0x37/0x63
[] __blocking_notifier_call_chain+0x46/0x5b
[] blocking_notifier_call_chain+0x14/0x16
[] sys_init_module+0x73/0x1f3
[] system_call_fastpath+0x16/0x1b
---[ end trace 2aff4f4ca53ec746 ]---
ftrace faulted on writing []
__process_new_adapter+0x7/0x34 [i2c_core]The cause was that the module text was set to read only before ftrace
could convert the calls to mcount to nops. Thus, the conversions failed
due to not being able to write to the text locations.The simple fix is to move setting the module to read only after the
module notifiers are called (where ftrace sets the module mcounts to nops).Reported-by: Peter Zijlstra
Acked-by: Rusty Russell
Signed-off-by: Steven Rostedt
18 Nov, 2010
1 commit
-
This patch is a logical extension of the protection provided by
CONFIG_DEBUG_RODATA to LKMs. The protection is provided by
splitting module_core and module_init into three logical parts
each and setting appropriate page access permissions for each
individual section:1. Code: RO+X
2. RO data: RO+NX
3. RW data: RW+NXIn order to achieve proper protection, layout_sections() have
been modified to align each of the three parts mentioned above
onto page boundary. Next, the corresponding page access
permissions are set right before successful exit from
load_module(). Further, free_module() and sys_init_module have
been modified to set module_core and module_init as RW+NX right
before calling module_free().By default, the original section layout and access flags are
preserved. When compiled with CONFIG_DEBUG_SET_MODULE_RONX=y,
the patch will page-align each group of sections to ensure that
each page contains only one type of content and will enforce
RO/NX for each group of pages.-v1: Initial proof-of-concept patch.
-v2: The patch have been re-written to reduce the number of #ifdefs
and to make it architecture-agnostic. Code formatting has also
been corrected.
-v3: Opportunistic RO/NX protection is now unconditional. Section
page-alignment is enabled when CONFIG_DEBUG_RODATA=y.
-v4: Removed most macros and improved coding style.
-v5: Changed page-alignment and RO/NX section size calculation
-v6: Fixed comments. Restricted RO/NX enforcement to x86 only
-v7: Introduced CONFIG_DEBUG_SET_MODULE_RONX, added
calls to set_all_modules_text_rw() and set_all_modules_text_ro()
in ftrace
-v8: updated for compatibility with linux 2.6.33-rc5
-v9: coding style fixes
-v10: more coding style fixes
-v11: minor adjustments for -tip
-v12: minor adjustments for v2.6.35-rc2-tip
-v13: minor adjustments for v2.6.37-rc1-tipSigned-off-by: Siarhei Liakh
Signed-off-by: Xuxian Jiang
Acked-by: Arjan van de Ven
Reviewed-by: James Morris
Signed-off-by: H. Peter Anvin
Cc: Andi Kleen
Cc: Rusty Russell
Cc: Stephen Rothwell
Cc: Dave Jones
Cc: Kees Cook
Cc: Linus Torvalds
LKML-Reference:
[ minor cleanliness edits, -v14: build failure fix ]
Signed-off-by: Ingo Molnar
11 Nov, 2010
1 commit
-
On use of trace_printk() there's a macro that determines if the format
is static or a variable. If it is static, it defaults to __trace_bprintk()
otherwise it uses __trace_printk().A while ago, Lai Jiangshan added __trace_bprintk(). In that patch, we
discussed a way to allow modules to use it. The difference between
__trace_bprintk() and __trace_printk() is that for faster processing,
just the format and args are stored in the trace instead of running
it through a sprintf function. In order to do this, the format used
by the __trace_bprintk() had to be persistent.See commit 1ba28e02a18cbdbea123836f6c98efb09cbf59ec
The problem comes with trace_bprintk() where the module is unloaded.
The pointer left in the buffer is still pointing to the format.To solve this issue, the formats in the module were copied into kernel
core. If the same format was used, they would use the same copy (to prevent
memory leak). This all worked well until we tried to merge everything.At the time this was written, Lai Jiangshan, Frederic Weisbecker,
Ingo Molnar and myself were all touching the same code. When this was
merged, we lost the part of it that was in module.c. This kept out the
copying of the formats and unloading the module could cause bad pointers
left in the ring buffer.This patch adds back (with updates required for current kernel) the
module code that sets up the necessary pointers.Cc: Lai Jiangshan
Cc: Rusty Russell
Signed-off-by: Steven Rostedt
27 Oct, 2010
1 commit
-
Building with CONFIG_KALLSYMS=n gives following warning:
/mnt/src/linux-git/kernel/module.c: In function ‘post_relocation’:
/mnt/src/linux-git/kernel/module.c:2534:2: warning: passing argument 2 of ‘add_kallsyms’ discards qualifiers from pointer target type
/mnt/src/linux-git/kernel/module.c:2038:13: note: expected ‘struct load_info *’ but argument is of type ‘const struct load_info *’Signed-off-by: Michał Mirosław
Signed-off-by: Rusty Russell
08 Oct, 2010
1 commit
-
Conflicts:
arch/x86/kernel/module.cMerge reason: Resolve the conflict, pick up fixes.
Signed-off-by: Ingo Molnar
06 Oct, 2010
1 commit
-
With all the recent module loading cleanups, we've minimized the code
that sits under module_mutex, fixing various deadlocks and making it
possible to do most of the module loading in parallel.However, that whole conversion totally missed the rather obscure code
that adds a new module to the list for BUG() handling. That code was
doubly obscure because (a) the code itself lives in lib/bugs.c (for
dubious reasons) and (b) it gets called from the architecture-specific
"module_finalize()" rather than from generic code.Calling it from arch-specific code makes no sense what-so-ever to begin
with, and is now actively wrong since that code isn't protected by the
module loading lock any more.So this commit moves the "module_bug_{finalize,cleanup}()" calls away
from the arch-specific code, and into the generic code - and in the
process protects it with the module_mutex so that the list operations
are now safe.Future fixups:
- move the module list handling code into kernel/module.c where it
belongs.
- get rid of 'module_bug_list' and just use the regular list of modules
(called 'modules' - imagine that) that we already create and maintain
for other reasons.Reported-and-tested-by: Thomas Gleixner
Cc: Rusty Russell
Cc: Adrian Bunk
Cc: Andrew Morton
Cc: stable@kernel.org
Signed-off-by: Linus Torvalds
23 Sep, 2010
1 commit
-
base patch to implement 'jump labeling'. Based on a new 'asm goto' inline
assembly gcc mechanism, we can now branch to labels from an 'asm goto'
statment. This allows us to create a 'no-op' fastpath, which can subsequently
be patched with a jump to the slowpath code. This is useful for code which
might be rarely used, but which we'd like to be able to call, if needed.
Tracepoints are the current usecase that these are being implemented for.Acked-by: David S. Miller
Signed-off-by: Jason Baron
LKML-Reference:[ cleaned up some formating ]
Signed-off-by: Steven Rostedt
05 Aug, 2010
19 commits
-
On my (32-bit x86) machine, sys_init_module() uses 124 bytes of stack
once load_module() is inlined.This effectively reverts ffb4ba76 which inlined it due to stack
pressure.Signed-off-by: Rusty Russell
-
This simply hoists more code out of load_module; we also put the
identification of the extable and dynamic debug table in with the
others in find_module_sections().We move the taint check to the actual add/remove of the dynamic debug
info: this is certain (find_module_sections is too early).Signed-off-by: Rusty Russell
Cc: Yehuda Sadeh -
Instead of copying and allocating the args and storing it in
load_info, we can just allocate them right before we need them.Signed-off-by: Rusty Russell
-
Pass the struct load_info into all the other functions in module
loading. This neatens things and makes them more consistent.Signed-off-by: Rusty Russell
-
Restore the stub module_remove_modinfo_attrs, remove the now-unused
!CONFIG_SYSFS module_sysfs_init.Also, rename mod_kobject_remove() to mod_sysfs_teardown() as
it is the logical counterpart to mod_sysfs_setup now.Reported-by: Randy Dunlap
Signed-off-by: Rusty Russell -
We change the sysfs functions to take struct load_info, and call
them all in mod_sysfs_setup().We also clean up the #ifdefs a little.
Signed-off-by: Rusty Russell
-
layout_and_allocate() does everything up to and including the final
struct module placement inside the allocated module memory. We have
to store the symbol layout information in our struct load_info though.This avoids the nasty code we had before where 'mod' pointed first
to the version inside the temporary allocation containing the entire
file, then later was moved to point to the real struct module: now
the main code only ever sees the final module address.(Includes fix for the Tony Luck-found Linus-diagnosed failure path
error).Signed-off-by: Rusty Russell
-
Andrew had the sole pleasure of tickling this bug in linux-next; when we set
up "info->strtab" it's pointing into the temporary copy of the module. For
most uses that is fine, but kallsyms keeps a pointer around during module
load (inside mod->strtab).If we oops for some reason inside a module's init function, kallsyms will use
the mod->strtab pointer into the now-freed temporary module copy.(Later oopses work fine: after init we overwrite mod->strtab to point to a
compacted core-only strtab).Reported-by: Andrew "Grumpy" Morton
Signed-off-by: Rusty "Buggy" Russell
Tested-by: Andrew "Happy" Morton
Signed-off-by: Rusty Russell -
Simple refactor causes us to lift struct definition to top of file.
Signed-off-by: Rusty Russell
-
We can't do the find_sec after removing the SHF_ALLOC flags; it won't
find the sections.Signed-off-by: Rusty Russell
-
Put all the "rewrite and check section headers" in one place. This
adds another iteration over the sections, but it's far clearer. We
iterate once for every find_section() so we already iterate over many
times.Signed-off-by: Rusty Russell
-
Btw, here's a patch that _looks_ large, but it really pretty trivial, and
sets things up so that it would be way easier to split off pieces of the
module loading.The reason it looks large is that it creates a "module_info" structure
that contains all the module state that we're building up while loading,
instead of having individual variables for all the indices etc.So the patch ends up being large, because every "symindex" access instead
becomes "info.index.sym" etc. That may be a few characters longer, but it
then means that we can just pass a pointer to that "info" structure
around. and let all the pieces fill it in very naturally.As an example of that, the patch also moves the initialization of all
those convenience variables into a "setup_module_info()" function. And at
this point it really does become very natural to start to peel off some of
the error labels and move them into the helper functions - now the
"truncated" case is gone, and is handled inside that setup function
instead.So maybe you don't like this approach, and it does make the variable
accesses a bit longer, but I don't think unreadably so. And the patch
really does look big and scary, but there really should be absolutely no
semantic changes - most of it was a trivial and mindless rename.In fact, it was so mindless that I on purpose kept the existing helper
functions looking like this:- err = check_modinfo(mod, sechdrs, infoindex, versindex);
+ err = check_modinfo(mod, info.sechdrs, info.index.info, info.index.vers);rather than changing them to just take the "info" pointer. IOW, a second
phase (if you think the approach is ok) would change that calling
convention to just doerr = check_modinfo(mod, &info);
(and same for "layout_sections()", "layout_symtabs()" etc.) Similarly,
while right now it makes things _look_ bigger, with things like this:versindex = find_sec(hdr, sechdrs, secstrings, "__versions");
becoming
info->index.vers = find_sec(info->hdr, info->sechdrs, info->secstrings, "__versions");
in the new "setup_module_info()" function, that's again just a result of
it being a search-and-replace patch. By using the 'info' pointer, we could
just change the 'find_sec()' interface so that it ends up beinginfo->index.vers = find_sec(info, "__versions");
instead, and then we'd actually have a shorter and more readable line. So
for a lot of those mindless variable name expansions there's would be room
for separate cleanups.I didn't move quite everything in there - if we do this to layout_symtabs,
for example, we'd want to move the percpu, symoffs, stroffs, *strmap
variables to be fields in that module_info structure too. But that's a
much smaller patch, I moved just the really core stuff that is currently
being set up and used in various parts.But even in this rough form, it removes close to 70 lines from that
function (but adds 22 lines overall, of course - the structure definition,
the helper function declarations and call-sites etc etc).Signed-off-by: Linus Torvalds
Signed-off-by: Rusty Russell -
And now that I'm looking at that call-chain (to see if it would make sense
to use some other more specific lock - doesn't look like it: all the
readers are using RCU and this is the only writer), I also give you this
trivial one-liner. It changes each_symbol() to not put that constant array
on the stack, resulting in changingmovq $C.388.31095, %rsi #, tmp85
subq $376, %rsp #,
movq %rdi, %rbx # fn, fn
leaq -208(%rbp), %rdi #, tmp84
movq %rbx, %rdx # fn,
rep movsl
xorl %esi, %esi #
leaq -208(%rbp), %rdi #, tmp87
movq %r12, %rcx # data,
call each_symbol_in_section.clone.0 #into
xorl %esi, %esi #
subq $216, %rsp #,
movq %rdi, %rbx # fn, fn
movq $arr.31078, %rdi #,
call each_symbol_in_section.clone.0 #which is not so much about being obviously shorter and simpler because we
don't unnecessarily copy that constant array around onto the stack, but
also about having a much smaller stack footprint (376 vs 216 bytes - see
the update of 'rsp').Signed-off-by: Linus Torvalds
Signed-off-by: Rusty Russell -
1) Extract out the relocation loop into apply_relocations
2) Extract license and version checks into check_module_license_and_versions
3) Extract icache flushing into flush_module_icache
4) Move __obsparm warning into find_module_sections
5) Move license setting into check_modinfo.Signed-off-by: Rusty Russell
-
Allocate references inside module_unload_init(), clean up inside
module_unload_free().This version fixed to do allocation before __this_cpu_write, thanks to
bug reports from linux-next from Dave Young
and Stephen Rothwell .Signed-off-by: Rusty Russell
-
Extract out the allocation and copying in from userspace, and the
first set of modinfo checks.Signed-off-by: Rusty Russell
-
Here's a second one. It's slightly less trivial - since we now have error
cases - and equally untested so it may well be totally broken. But it also
cleans up a bit more, and avoids one of the goto targets, because the
"move_module()" helper now does both allocations or none.Signed-off-by: Linus Torvalds
Signed-off-by: Rusty Russell -
I'd start from the trivial stuff. There's a fair amount of straight-line
code that just makes the function hard to read just because you have to
page up and down so far. Some of it is trivial to just create a helper
function for.Signed-off-by: Linus Torvalds
Signed-off-by: Rusty Russell -
No need to clear mod->refptr in module_unload_init(), since
alloc_percpu() already clears allocated chunks.Signed-off-by: Eric Dumazet
Signed-off-by: Rusty Russell (removed unused var)
28 Jul, 2010
1 commit
-
The command
echo "file ec.c +p" >/sys/kernel/debug/dynamic_debug/control
causes an oops.
Move the call to ddebug_remove_module() down into free_module(). In this
way it should be called from all error paths. Currently, we are missing
the remove if the module init routine fails.Signed-off-by: Jason Baron
Reported-by: Thomas Renninger
Tested-by: Thomas Renninger
Cc: [2.6.32+]
Signed-off-by: Andrew Morton
Signed-off-by: Linus Torvalds
05 Jul, 2010
1 commit
-
We should initialize the module dynamic debug datastructures
only after determining that the module is not loaded yet. This
fixes a bug that introduced in 2.6.35-rc2, where when a trying
to load a module twice, we also load it's dynamic printing data
twice which causes all sorts of nasty issues. Also handle
the dynamic debug cleanup later on failure.Signed-off-by: Yehuda Sadeh
Signed-off-by: Rusty Russell (removed a #ifdef)
Signed-off-by: Linus Torvalds
05 Jun, 2010
8 commits
-
Problem: it's hard to avoid an init routine stumbling over a
request_module these days. And it's not clear it's always a bad idea:
for example, a module like kvm with dynamic dependencies on kvm-intel
or kvm-amd would be neater if it could simply request_module the right
one.In this particular case, it's libcrc32c:
libcrc32c_mod_init
crypto_alloc_shash
crypto_alloc_tfm
crypto_find_alg
crypto_alg_mod_lookup
crypto_larval_lookup
request_moduleIf another module is waiting inside resolve_symbol() for libcrc32c to
finish initializing (ie. bne2 depends on libcrc32c) then it does so
holding the module lock, and our request_module() can't make progress
until that is released.Waiting inside resolve_symbol() without the lock isn't all that hard:
we just need to pass the -EBUSY up the call chain so we can sleep
where we don't hold the lock. Error reporting is a bit trickier: we
need to copy the name of the unfinished module before releasing the
lock.Other notes:
1) This also fixes a theoretical issue where a weak dependency would allow
symbol version mismatches to be ignored.
2) We rename use_module to ref_module to make life easier for the only
external user (the out-of-tree ksplice patches).Signed-off-by: Rusty Russell
Cc: Linus Torvalds
Cc: Tim Abbot
Tested-by: Brandon Philips -
It disabled preempt so it was "safe", but nothing stops another module
slipping in before this module is added to the global list now we don't
hold the lock the whole time.So we check this just after we check for duplicate modules, and just
before we put the module in the global list.(find_symbol finds symbols in coming and going modules, too).
Signed-off-by: Rusty Russell
-
I think Rusty may have made the lock a bit _too_ finegrained there, and
didn't add it to some places that needed it. It looks, for example, like
PATCH 1/2 actually drops the lock in places where it's needed
("find_module()" is documented to need it, but now load_module() didn't
hold it at all when it did the find_module()).Rather than adding a new "module_loading" list, I think we should be able
to just use the existing "modules" list, and just fix up the locking a
bit.In fact, maybe we could just move the "look up existing module" a bit
later - optimistically assuming that the module doesn't exist, and then
just undoing the work if it turns out that we were wrong, just before
adding ourselves to the list.Signed-off-by: Rusty Russell
-
Kay Sievers reports that we still have some
contention over module loading which is slowing boot.Linus also disliked a previous "drop lock and regrab" patch to fix the
bne2 "gave up waiting for init of module libcrc32c" message.This is more ambitious: we only grab the lock where we need it.
Signed-off-by: Rusty Russell
Cc: Brandon Philips
Cc: Kay Sievers
Cc: Linus Torvalds -
These were placed in the header in ef665c1a06 to get the various
SYSFS/MODULE config combintations to compile.That may have been necessary then, but it's not now. These functions
are all local to module.c.Signed-off-by: Rusty Russell
Cc: Randy Dunlap -
This means a little extra work, but is more logical: we don't put
anything in sysfs until we're about to put the module into the
global list an parse its parameters.This also gives us a logical place to put duplicate module detection
in the next patch.Signed-off-by: Rusty Russell
-
Linus changed the structure, and luckily this didn't compile any more.
Reported-by: Stephen Rothwell
Signed-off-by: Rusty Russell
Cc: Jason Wessel
Cc: Martin Hicks -
When adding a module that depends on another one, we used to create a
one-way list of "modules_which_use_me", so that module unloading could
see who needs a module.It's actually quite simple to make that list go both ways: so that we
not only can see "who uses me", but also see a list of modules that are
"used by me".In fact, we always wanted that list in "module_unload_free()": when we
unload a module, we want to also release all the other modules that are
used by that module. But because we didn't have that list, we used to
first iterate over all modules, and then iterate over each "used by me"
list of that module.By making the list two-way, we simplify module_unload_free(), and it
allows for some trivial fixes later too.Signed-off-by: Linus Torvalds
Signed-off-by: Rusty Russell (cleaned & rebased)
01 Jun, 2010
2 commits
-
* 'for-35' of git://repo.or.cz/linux-kbuild: (81 commits)
kbuild: Revert part of e8d400a to resolve a conflict
kbuild: Fix checking of scm-identifier variable
gconfig: add support to show hidden options that have prompts
menuconfig: add support to show hidden options which have prompts
gconfig: remove show_debug option
gconfig: remove dbg_print_ptype() and dbg_print_stype()
kconfig: fix zconfdump()
kconfig: some small fixes
add random binaries to .gitignore
kbuild: Include gen_initramfs_list.sh and the file list in the .d file
kconfig: recalc symbol value before showing search results
.gitignore: ignore *.lzo files
headerdep: perlcritic warning
scripts/Makefile.lib: Align the output of LZO
kbuild: Generate modules.builtin in make modules_install
Revert "kbuild: specify absolute paths for cscope"
kbuild: Do not unnecessarily regenerate modules.builtin
headers_install: use local file handles
headers_check: fix perl warnings
export_report: fix perl warnings
... -
Rafael sees a sometimes crash at precpu_modfree from kernel/module.c; it
only occurred with another (since-reverted) patch, but that patch simply
changed timing to uncover this bug, it was otherwise unrelated.The comment about the mod being freed is self-explanatory, but neither
Tejun nor I read it. This bug was introduced in 259354deaa, after it
had previously been fixed in 6e2b75740b. How embarrassing.Reported-by: "Rafael J. Wysocki"
Signed-off-by: Rusty Russell
Embarrassingly-Acked-by: Tejun Heo
Cc: Masami Hiramatsu
Tested-by: "Rafael J. Wysocki"
Signed-off-by: Linus Torvalds