Select Git revision
Search by author
- Any Author
- authors
-
Abhiram Balasubramanian abhiram
-
Anton Burtsev
aburtsev
-
Charlie Jacobsen cjacobsen
-
David Detweiler ddetweiler
-
David Johnson johnsond
-
Michael Quigley MichaelQuigley
-
Mike Hibler hibler
-
Robert Ricci ricci
-
Sarah Spall spall
-
Scotty Bauer ScottyB
-
Vikram Narayanan vikram
- Mar 23, 2011
-
-
Serge E. Hallyn authored
Changelog: Feb 15: Don't set new ipc->user_ns if we didn't create a new ipc_ns. Feb 23: Move extern declaration to ipc_namespace.h, and group fwd declarations at top. Signed-off-by:
Serge E. Hallyn <serge.hallyn@canonical.com> Acked-by:
"Eric W. Biederman" <ebiederm@xmission.com> Acked-by:
Daniel Lezcano <daniel.lezcano@free.fr> Acked-by:
David Howells <dhowells@redhat.com> Cc: James Morris <jmorris@namei.org> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Serge E. Hallyn authored
This allows setuid/setgid in containers. It also fixes some corner cases where kernel logic foregoes capability checks when uids are equivalent. The latter will need to be done throughout the whole kernel. Changelog: Jan 11: Use nsown_capable() as suggested by Bastian Blank. Jan 11: Fix logic errors in uid checks pointed out by Bastian. Feb 15: allow prlimit to current (was regression in previous version) Feb 23: remove debugging printks, uninline set_one_prio_perm and make it bool, and document its return value. Signed-off-by:
-
Serge E. Hallyn authored
So we can let type safety keep things sane, and as a bonus we can remove the declaration of init_user_ns in capability.h. Signed-off-by:
-
Serge E. Hallyn authored
ptrace is allowed to tasks in the same user namespace according to the usual rules (i.e. the same rules as for two tasks in the init user namespace). ptrace is also allowed to a user namespace to which the current task the has CAP_SYS_PTRACE capability. Changelog: Dec 31: Address feedback by Eric: . Correct ptrace uid check . Rename may_ptrace_ns to ptrace_capable . Also fix the cap_ptrace checks. Jan 1: Use const cred struct Jan 11: use task_ns_capable() in place of ptrace_capable(). Feb 23: same_or_ancestore_user_ns() was not an appropriate check to constrain cap_issubset. Rather, cap_issubset() only is meaningful when both capsets are in the same user_ns. Signed-off-by:
-
Serge E. Hallyn authored
Changelog: Dec 8: Fixed bug in my check_kill_permission pointed out by Eric Biederman. Dec 13: Apply Eric's suggestion to pass target task into kill_ok_by_cred() for clarity Dec 31: address comment by Eric Biederman: don't need cred/tcred in check_kill_permission. Jan 1: use const cred struct. Jan 11: Per Bastian Blank's advice, clean up kill_ok_by_cred(). Feb 16: kill_ok_by_cred: fix bad parentheses Feb 23: per akpm, let compiler inline kill_ok_by_cred Signed-off-by:
-
Serge E. Hallyn authored
Changelog: Feb 23: let clone_uts_ns() handle setting uts->user_ns To do so we need to pass in the task_struct who'll get the utsname, so we can get its user_ns. Feb 23: As per Oleg's coment, just pass in tsk, instead of two of its members. Signed-off-by:
-
Serge E. Hallyn authored
- Introduce ns_capable to test for a capability in a non-default user namespace. - Teach cap_capable to handle capabilities in a non-default user namespace. The motivation is to get to the unprivileged creation of new namespaces. It looks like this gets us 90% of the way there, with only potential uid confusion issues left. I still need to handle getting all caps after creation but otherwise I think I have a good starter patch that achieves all of your goals. Changelog: 11/05/2010: [serge] add apparmor 12/14/2010: [serge] fix capabilities to created user namespaces Without this, if user serge creates a user_ns, he won't have capabilities to the user_ns he created. THis is because we were first checking whether his effective caps had the caps he needed and returning -EPERM if not, and THEN checking whether he was the creator. Reverse those checks. 12/16/2010: [serge] security_real_capable needs ns argument in !security case 01/11/2011: [serge] add task_ns_capable helper 01/11/2011: [serge] add nsown_capable() helper per Bastian Blank suggestion 02/16/2011: [serge] fix a logic bug: the root user is always creator of init_user_ns, but should not always have capabilities to it! Fix the check in cap_capable(). 02/21/2011: Add the required user_ns parameter to security_capable, fixing a compile failure. 02/23/2011: Convert some macros to functions as per akpm comments. Some couldn't be converted because we can't easily forward-declare them (they are inline if !SECURITY, extern if SECURITY). Add a current_user_ns function so we can use it in capability.h without #including cred.h. Move all forward declarations together to the top of the #ifdef __KERNEL__ section, and use kernel-doc format. 02/23/2011: Per dhowells, clean up comment in cap_capable(). 02/23/2011: Per akpm, remove unreachable 'return -EPERM' in cap_capable. (Original written and signed off by Eric; latest, modified version acked by him) [akpm@linux-foundation.org: fix build] [akpm@linux-foundation.org: export current_user_ns() for ecryptfs] [serge.hallyn@canonical.com: remove unneeded extra argument in selinux's task_has_capability] Signed-off-by:
-
Serge E. Hallyn authored
The expected course of development for user namespaces targeted capabilities is laid out at https://wiki.ubuntu.com/UserNamespace . Goals: - Make it safe for an unprivileged user to unshare namespaces. They will be privileged with respect to the new namespace, but this should only include resources which the unprivileged user already owns. - Provide separate limits and accounting for userids in different namespaces. Status: Currently (as of 2.6.38) you can clone with the CLONE_NEWUSER flag to get a new user namespace if you have the CAP_SYS_ADMIN, CAP_SETUID, and CAP_SETGID capabilities. What this gets you is a whole new set of userids, meaning that user 500 will have a different 'struct user' in your namespace than in other namespaces. So any accounting information stored in struct user will be unique to your namespace. However, throughout the kernel there are checks which - simply check for a capability. Since root in a child namespace has all capabilities, this means that a child namespace is not constrained. - simply compare uid1 == uid2. Since these are the integer uids, uid 500 in namespace 1 will be said to be equal to uid 500 in namespace 2. As a result, the lxc implementation at lxc.sf.net does not use user namespaces. This is actually helpful because it leaves us free to develop user namespaces in such a way that, for some time, user namespaces may be unuseful. Bugs aside, this patchset is supposed to not at all affect systems which are not actively using user namespaces, and only restrict what tasks in child user namespace can do. They begin to limit privilege to a user namespace, so that root in a container cannot kill or ptrace tasks in the parent user namespace, and can only get world access rights to files. Since all files currently belong to the initila user namespace, that means that child user namespaces can only get world access rights to *all* files. While this temporarily makes user namespaces bad for system containers, it starts to get useful for some sandboxing. I've run the 'runltplite.sh' with and without this patchset and found no difference. This patch: copy_process() handles CLONE_NEWUSER before the rest of the namespaces. So in the case of clone(CLONE_NEWUSER|CLONE_NEWUTS) the new uts namespace will have the new user namespace as its owner. That is what we want, since we want root in that new userns to be able to have privilege over it. Changelog: Feb 15: don't set uts_ns->user_ns if we didn't create a new uts_ns. Feb 23: Move extern init_user_ns declaration from init/version.c to utsname.h. Signed-off-by:
-
Eric W. Biederman authored
Reorganize proc_get_sb() so it can be called before the struct pid of the first process is allocated. Signed-off-by:
Serge E. Hallyn <serge@hallyn.com> Signed-off-by:
-
Eric W. Biederman authored
This patchset is a cleanup and a preparation to unshare the pid namespace. These prerequisites prepare for Eric's patchset to give a file descriptor to a namespace and join an existing namespace. This patch: It turns out that the existing assignment in copy_process of the child_reaper can handle the initial assignment of child_reaper we just need to generalize the test in kernel/fork.c Signed-off-by:
-
Richard Weinberger authored
When dmesg_restrict is set to 1 CAP_SYS_ADMIN is needed to read the kernel ring buffer. But a root user without CAP_SYS_ADMIN is able to reset dmesg_restrict to 0. This is an issue when e.g. LXC (Linux Containers) are used and complete user space is running without CAP_SYS_ADMIN. A unprivileged and jailed root user can bypass the dmesg_restrict protection. With this patch writing to dmesg_restrict is only allowed when root has CAP_SYS_ADMIN. Signed-off-by:
Richard Weinberger <richard@nod.at> Acked-by:
Dan Rosenberg <drosenberg@vsecurity.com> Acked-by:
-
Petr Holasek authored
Add boundaries of allowed input ranges for: dirty_expire_centisecs, drop_caches, overcommit_memory, page-cluster and panic_on_oom. Signed-off-by:
Petr Holasek <pholasek@redhat.com> Acked-by:
Dave Young <hidave.darkstar@gmail.com> Cc: David Rientjes <rientjes@google.com> Cc: Wu Fengguang <fengguang.wu@intel.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Signed-off-by:
-
Denis Kirjanov authored
Drop dead code. Signed-off-by:
Denis Kirjanov <dkirjanov@kernel.org> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by:
-
Denis Kirjanov authored
Since the for loop checks for the table->procname drop useless table->procname checks inside the loop body Signed-off-by:
-
Li Zefan authored
Chaning cpuset->mems/cpuset->cpus should be protected under callback_mutex. cpuset_clone() doesn't follow this rule. It's ok because it's called when creating and initializing a cgroup, but we'd better hold the lock to avoid subtil break in the future. Signed-off-by:
Li Zefan <lizf@cn.fujitsu.com> Acked-by:
Paul Menage <menage@google.com> Acked-by:
David Rientjes <rientjes@google.com> Cc: Miao Xie <miaox@cn.fujitsu.com> Signed-off-by:
-
Li Zefan authored
Those functions that use NODEMASK_ALLOC() can't propagate errno to users, but will fail silently. Fix it by using a static nodemask_t variable for each function, and those variables are protected by cgroup_mutex; [akpm@linux-foundation.org: fix comment spelling, strengthen cgroup_lock comment] Signed-off-by:
-
Li Zefan authored
oldcs->mems_allowed is not modified during cpuset_attach(), so we don't have to copy it to a buffer allocated by NODEMASK_ALLOC(). Just pass it to cpuset_migrate_mm(). Signed-off-by:
-
Li Zefan authored
It's not necessary to copy cpuset->mems_allowed to a buffer allocated by NODEMASK_ALLOC(). Just pass it to nodelist_scnprintf(). As spotted by Paul, a side effect is we fix a bug that the function can return -ENOMEM but the caller doesn't expect negative return value. Therefore change the return value of cpuset_sprintf_cpulist() and cpuset_sprintf_memlist() from int to size_t. Signed-off-by:
-
Johannes Weiner authored
In struct page_cgroup, we have a full word for flags but only a few are reserved. Use the remaining upper bits to encode, depending on configuration, the node or the section, to enable page_cgroup-to-page lookups without a direct pointer. This saves a full word for every page in a system with memory cgroups enabled. Signed-off-by:
Johannes Weiner <hannes@cmpxchg.org> Acked-by:
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Cc: Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Minchan Kim <minchan.kim@gmail.com> Cc: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
res_counter_read_u64 reads u64 value without lock. It's dangerous in a 32bit environment. Add locking. Signed-off-by:
-
- Mar 22, 2011
-
-
Mandeep Singh Baines authored
We've been burned by regressions/bugs which we later realized could have been triaged quicker if only we'd paid closer attention to dmesg. To make it easier to audit dmesg, we'd like to make DEFAULT_MESSAGE_LEVEL Kconfig-settable. That way we can set it to KERN_NOTICE and audit any messages <= KERN_WARNING. Signed-off-by:
Mandeep Singh Baines <msb@chromium.org> Cc: Ingo Molnar <mingo@elte.hu> Cc: Joe Perches <joe@perches.com> Cc: Olof Johansson <olofj@chromium.org> Cc: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by:
-
Kees Cook authored
In an effort to reduce kernel address leaks that might be used to help target kernel privilege escalation exploits, this patch uses %pK when displaying addresses in /proc/kallsyms, /proc/modules, and /sys/module/*/sections/*. Note that this changes %x to %p, so some legitimately 0 values in /proc/kallsyms would have changed from 00000000 to "(null)". To avoid this, "(null)" is not used when using the "K" format. Anything that was already successfully parsing "(null)" in addition to full hex digits should have no problem with this change. (Thanks to Joe Perches for the suggestion.) Due to the %x to %p, "void *" casts are needed since these addresses are already "unsigned long" everywhere internally, due to their starting life as ELF section offsets. Signed-off-by:
Kees Cook <kees.cook@canonical.com> Cc: Eugene Teo <eugene@redhat.com> Cc: Dan Rosenberg <drosenberg@vsecurity.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Signed-off-by:
-
Feng Tang authored
For a platform with many consoles like: "console=tty1 console=ttyMFD2 console=ttyS0 earlyprintk=mrst" Each time when the non "selected_console" (tty1 and ttyMFD2 here) get registered, the existing kernel message will be printed out on registered consoles again, the "mrst" early console will get some same message for 3 times, and "tty1" will get some for twice. As suggested by Andrew Morton, every time a new console is registered, it will be set as the "exclusive" console which will dump the already existing kernel messages. Signed-off-by:
Feng Tang <feng.tang@intel.com> Cc: Greg KH <gregkh@suse.de> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Signed-off-by:
-
Fabio M. Di Nitto authored
On some architectures, the boot process involves de-registering the boot console (early boot), initialize drivers and then re-register the console. This mechanism introduces a window in which no printk can happen on the console and messages are buffered and then printed once the new console is available. If a kernel crashes during this window, all it's left on the boot console is "console [foo] enabled, bootconsole disabled" making debug of the crash rather 'interesting'. By adding "keep_bootcon" option, do not unregister the boot console, that will allow to printk everything that is happening up to the crash. The option is clearly meant only for debugging purposes as it introduces lots of duplicated info printed on console, but will make bug report from users easier as it doesn't require a kernel build just to figure out where we crash. Signed-off-by:
Fabio M. Di Nitto <fabbione@fabbione.net> Acked-by:
David S. Miller <davem@davemloft.net> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: Greg KH <gregkh@suse.de> Signed-off-by:
-
Don Zickus authored
This patch addresses a couple of problems. One was the case when the hardlockup failed to start, it also failed to start the softlockup. There were valid cases when the hardlockup shouldn't start and that shouldn't block the softlockup (no lapic, bios controls perf counters). The second problem was when the hardlockup failed to start on boxes (from a no lapic or bios controlled perf counter case), it reported failure to the cpu notifier chain. This blocked the notifier from continuing to start other more critical pieces of cpu bring-up (in our case based on a 2.6.32 fork, it was the mce). As a result, during soft cpu online/offline testing, the system would panic when a cpu was offlined because the cpu notifier would succeed in processing a watchdog disable cpu event and would panic in the mce case as a result of un-initialized variables from a never executed cpu up event. I realized the hardlockup/softlockup cases are really just debugging aids and should never impede the progress of a cpu up/down event. Therefore I modified the code to always return NOTIFY_OK and instead rely on printks to inform the user of problems. Signed-off-by:
Don Zickus <dzickus@redhat.com> Acked-by:
Peter Zijlstra <a.p.zijlstra@chello.nl> Reviewed-by:
WANG Cong <xiyou.wangcong@gmail.com> Signed-off-by:
-
Don Zickus authored
When a cpu is considered stuck, instead of limping along and just printing a warning, it is sometimes preferred to just panic, let kdump capture the vmcore and reboot. This gets the machine back into a stable state quickly while saving the info that got it into a stuck state to begin with. Add a Kconfig option to allow users to set the hardlockup to panic by default. Also add in a 'nmi_watchdog=nopanic' to override this. [akpm@linux-foundation.org: fix strncmp length] Signed-off-by:
-
Oleg Nesterov authored
Cleanup: kill the dead code which does nothing but complicates the code and confuses the reader. sys_unshare(CLONE_THREAD/SIGHAND/VM) is not really implemented, and I doubt very much it will ever work. At least, nobody even tried since the original 99d1419d ("unshare system call -v5: system call handler function") was applied more than 4 years ago. And the code is not consistent. unshare_thread() always fails unconditionally, while unshare_sighand() and unshare_vm() pretend to work if there is nothing to unshare. Remove unshare_thread(), unshare_sighand(), unshare_vm() helpers and related variables and add a simple CLONE_THREAD | CLONE_SIGHAND| CLONE_VM check into check_unshare_flags(). Also, move the "CLONE_NEWNS needs CLONE_FS" check from check_unshare_flags() to sys_unshare(). This looks more consistent and matches the similar do_sysvsem check in sys_unshare(). Note: with or without this patch "atomic_read(mm->mm_users) > 1" can give a false positive due to get_task_mm(). Signed-off-by:
Oleg Nesterov <oleg@redhat.com> Acked-by:
Roland McGrath <roland@redhat.com> Cc: Janak Desai <janak@us.ibm.com> Cc: Daniel Lezcano <daniel.lezcano@free.fr> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Acked-by:
-
Michael Rodriguez authored
Change the printk() calls to have the KERN_INFO/KERN_ERROR stuff, and fixes other coding style errors. Not _all_ of them are gone, though. [akpm@linux-foundation.org: revert the bits I disagree with] Signed-off-by:
Michael Rodriguez <dkingston02@gmail.com> Signed-off-by:
-
Amerigo Wang authored
Move setup_nr_cpu_ids(), smp_init() and some other SMP boot parameter setup functions from init/main.c to kenrel/smp.c, saves some #ifdef CONFIG_SMP. Signed-off-by:
WANG Cong <amwang@redhat.com> Cc: Rakib Mullick <rakib.mullick@gmail.com> Cc: David Howells <dhowells@redhat.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Tejun Heo <tj@kernel.org> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Akinobu Mita <akinobu.mita@gmail.com> Signed-off-by:
-
Olaf Hering authored
The oops=panic cmdline option is not x86 specific, move it to generic code. Update documentation. Signed-off-by:
Olaf Hering <olaf@aepfle.de> Cc: Ingo Molnar <mingo@elte.hu> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: "H. Peter Anvin" <hpa@zytor.com> Signed-off-by:
-
Eric Dumazet authored
ksoftirqd, kworker, migration, and pktgend kthreads can be created with kthread_create_on_node(), to get proper NUMA affinities for their stack and task_struct. Signed-off-by:
Eric Dumazet <eric.dumazet@gmail.com> Acked-by:
Andi Kleen <ak@linux.intel.com> Acked-by:
Rusty Russell <rusty@rustcorp.com.au> Acked-by:
Tejun Heo <tj@kernel.org> Cc: Tony Luck <tony.luck@intel.com> Cc: Fenghua Yu <fenghua.yu@intel.com> Cc: David Howells <dhowells@redhat.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by:
-
Eric Dumazet authored
All kthreads being created from a single helper task, they all use memory from a single node for their kernel stack and task struct. This patch suite creates kthread_create_on_node(), adding a 'cpu' parameter to parameters already used by kthread_create(). This parameter serves in allocating memory for the new kthread on its memory node if possible. Signed-off-by:
-
Eric Dumazet authored
Add a node parameter to alloc_thread_info(), and change its name to alloc_thread_info_node() This change is needed to allow NUMA aware kthread_create_on_cpu() Signed-off-by:
-
Eric Dumazet authored
All kthreads being created from a single helper task, they all use memory from a single node for their kernel stack and task struct. This patch suite creates kthread_create_on_cpu(), adding a 'cpu' parameter to parameters already used by kthread_create(). This parameter serves in allocating memory for the new kthread on its memory node if available. Users of this new function are : ksoftirqd, kworker, migration, pktgend... This patch: Add a node parameter to alloc_task_struct(), and change its name to alloc_task_struct_node() This change is needed to allow NUMA aware kthread_create_on_cpu() Signed-off-by:
-
Phil Carmody authored
list_del() leaves poison in the prev and next pointers. The next list_empty() will compare those poisons, and say the list isn't empty. Any list operations that assume the node is on a list because of such a check will be fooled into dereferencing poison. One needs to INIT the node after the del, and fortunately there's already a wrapper for that - list_del_init(). Some of the dels are followed by deallocations, so can be ignored, and one can be merged with an add to make a move. Apart from that, I erred on the side of caution in making nodes list_empty()-queriable. Signed-off-by:
Phil Carmody <ext-phil.2.carmody@nokia.com> Reviewed-by:
Kirill A. Shutemov <kirill@shutemov.name> Cc: <stable@kernel.org> Signed-off-by:
-
- Mar 21, 2011
-
-
Julien Tinnes authored
Userland should be able to trust the pid and uid of the sender of a signal if the si_code is SI_TKILL. Unfortunately, the kernel has historically allowed sigqueueinfo() to send any si_code at all (as long as it was negative - to distinguish it from kernel-generated signals like SIGILL etc), so it could spoof a SI_TKILL with incorrect siginfo values. Happily, it looks like glibc has always set si_code to the appropriate SI_QUEUE, so there are probably no actual user code that ever uses anything but the appropriate SI_QUEUE flag. So just tighten the check for si_code (we used to allow any negative value), and add a (one-time) warning in case there are binaries out there that might depend on using other si_code values. Signed-off-by:
Julien Tinnes <jln@google.com> Acked-by:
-
- Mar 18, 2011
-
-
Ingo Molnar authored
Make sure the 'match' variable always has a value. Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> LKML-Reference: <new-submission> Signed-off-by:Ingo Molnar <mingo@elte.hu>
-
- Mar 17, 2011
-
-
Milton Miller authored
Use the newly added smp_call_func_t in smp_call_function_interrupt for the func variable, and make the comment above the WARN more assertive and explicit. Also, func is a function pointer and does not need an offset, so use %pf not %pS. Signed-off-by:
Milton Miller <miltonm@bga.com> Signed-off-by:
-
Milton Miller authored
Mike Galbraith reported finding a lockup ("perma-spin bug") where the cpumask passed to smp_call_function_many was cleared by other cpu(s) while a cpu was preparing its call_data block, resulting in no cpu to clear the last ref and unlock the block. Having cpus clear their bit asynchronously could be useful on a mask of cpus that might have a translation context, or cpus that need a push to complete an rcu window. Instead of adding a BUG_ON and requiring yet another cpumask copy, just detect the race and handle it. Note: arch_send_call_function_ipi_mask must still handle an empty cpumask because the data block is globally visible before the that arch callback is made. And (obviously) there are no guarantees to which cpus are notified if the mask is changed during the call; only cpus that were online and had their mask bit set during the whole call are guaranteed to be called. Reported-by:Mike Galbraith <efault@gmx.de> Reported-by:
Jan Beulich <JBeulich@novell.com> Acked-by:
-
Milton Miller authored
Paul McKenney's review pointed out two problems with the barriers in the 2.6.38 update to the smp call function many code. First, a barrier that would force the func and info members of data to be visible before their consumption in the interrupt handler was missing. This can be solved by adding a smp_wmb between setting the func and info members and setting setting the cpumask; this will pair with the existing and required smp_rmb ordering the cpumask read before the read of refs. This placement avoids the need a second smp_rmb in the interrupt handler which would be executed on each of the N cpus executing the call request. (I was thinking this barrier was present but was not). Second, the previous write to refs (establishing the zero that we the interrupt handler was testing from all cpus) was performed by a third party cpu. This would invoke transitivity which, as a recient or concurrent addition to memory-barriers.txt now explicitly states, would require a full smp_mb(). However, we know the cpumask will only be set by one cpu (the data owner) and any preivous iteration of the mask would have cleared by the reading cpu. By redundantly writing refs to 0 on the owning cpu before the smp_wmb, the write to refs will follow the same path as the writes that set the cpumask, which in turn allows us to keep the barrier in the interrupt handler a smp_rmb instead of promoting it to a smp_mb (which will be be executed by N cpus for each of the possible M elements on the list). I moved and expanded the comment about our (ab)use of the rcu list primitives for the concurrent walk earlier into this function. I considered moving the first two paragraphs to the queue list head and lock, but felt it would have been too disconected from the code. Cc: Paul McKinney <paulmck@linux.vnet.ibm.com> Cc: stable@kernel.org (2.6.32 and later) Signed-off-by:
-
