- 03 Apr, 2009 40 commits
-
-
Oleg Nesterov authored
If task_detached(p) == T, then either a) p is not the main thread, we will find the group leader on the ->children list. or b) p is the group leader but its ->exit_state = EXIT_DEAD. This can only happen when the last sub-thread has died, but in that case that thread has already called kill_orphaned_pgrp() from exit_notify(). In both cases kill_orphaned_pgrp() looks bogus. Move the task_detached() check up and simplify the code, this is also right from the "common sense" pov: we should do nothing with the detached childs, except move them to the new parent's ->children list. Signed-off-by:Oleg Nesterov <oleg@redhat.com> Cc: Roland McGrath <roland@redhat.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Oleg Nesterov authored
When ptrace_detach() takes tasklist, the tracee can be SIGKILL'ed. If it has already passed exit_notify() we can leak a zombie, because a) ptracing disables the auto-reaping logic, and b) ->real_parent was not notified about the child's death. ptrace_detach() should follow the ptrace_exit's logic, change the code accordingly. Signed-off-by:
Denys Vlasenko <dvlasenk@redhat.com> Signed-off-by:
-
Oleg Nesterov authored
No functional changes, preparation for the next patch. Move the "should we release this child" logic into the separate handler, __ptrace_detach(). Signed-off-by:
-
Oleg Nesterov authored
ignoring_children() takes parent->sighand->siglock and checks k_sigaction[SIGCHLD] atomically. But this buys nothing, we can't get the "really" wrong result even if we race with sigaction(SIGCHLD). If we read the "stale" sa_handler/sa_flags we can pretend it was changed right after the check. Remove spin_lock(->siglock), and kill "int ign" which caches the result of ignoring_children() which becomes rather trivial. Perhaps it makes sense to export this helper, do_notify_parent() can use it too. Signed-off-by:
-
Oleg Nesterov authored
Move the code from __ptrace_detach() to its single caller and kill this helper. Also, fix the ->exit_state check, we shouldn't wake up EXIT_DEAD tasks. Actually, I think task_is_stopped_or_traced() makes more sense, but this needs another patch. Signed-off-by:
-
Sukadev Bhattiprolu authored
When sending a signal to a descendant namespace, set ->si_pid to 0 since the sender does not have a pid in the receiver's namespace. Note: - If rt_sigqueueinfo() sets si_code to SI_USER when sending a signal across a pid namespace boundary, the value in ->si_pid will be cleared to 0. Signed-off-by:
Sukadev Bhattiprolu <sukadev@linux.vnet.ibm.com> Cc: Oleg Nesterov <oleg@tv-sign.ru> Cc: Roland McGrath <roland@redhat.com> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Daniel Lezcano <daniel.lezcano@free.fr> Signed-off-by:
-
Sukadev Bhattiprolu authored
Normally SIG_DFL signals to global and container-init are dropped early. But if a signal is blocked when it is posted, we cannot drop the signal since the receiver may install a handler before unblocking the signal. Once this signal is queued however, the receiver container-init has no way of knowing if the signal was sent from an ancestor or descendant namespace. This patch ensures that contianer-init drops all SIG_DFL signals in get_signal_to_deliver() except SIGKILL/SIGSTOP. If SIGSTOP/SIGKILL originate from a descendant of container-init they are never queued (i.e dropped in sig_ignored() in an earler patch). If SIGSTOP/SIGKILL originate from parent namespace, the signal is queued and container-init processes the signal. IOW, if get_signal_to_deliver() sees a sig_kernel_only() signal for global or container-init, the signal must have been generated internally or must have come from an ancestor ns and we process the signal. Further, the signal_group_exit() check was needed to cover the case of a multi-threaded init sending SIGKILL to other threads when doing an exit() or exec(). But since the new sig_kernel_only() check covers the SIGKILL, the signal_group_exit() check is no longer needed and can be removed. Finally, now that we have all pieces in place, set SIGNAL_UNKILLABLE for container-inits. Signed-off-by:
-
Sukadev Bhattiprolu authored
send_signal() assumes that signals with SEND_SIG_PRIV are generated from within the same namespace. So any nested container-init processes become immune to the SIGKILL generated by kill_proc_info() in zap_pid_ns_processes(). Use force_sig() in zap_pid_ns_processes() instead - force_sig() clears the SIGNAL_UNKILLABLE flag ensuring the signal is processed by container-inits. Signed-off-by:
-
Sukadev Bhattiprolu authored
Drop early any SIG_DFL or SIG_IGN signals to container-init from within the same container. But queue SIGSTOP and SIGKILL to the container-init if they are from an ancestor container. Blocked, fatal signals (i.e when SIG_DFL is to terminate) from within the container can still terminate the container-init. That will be addressed in the next patch. Note: To be bisect-safe, SIGNAL_UNKILLABLE will be set for container-inits in a follow-on patch. Until then, this patch is just a preparatory step. Signed-off-by:
-
Sukadev Bhattiprolu authored
send_signal() (or its helper) needs to determine the pid namespace of the sender. But a signal sent via kill_pid_info_as_uid() comes from within the kernel and send_signal() does not need to determine the pid namespace of the sender. So define a helper for send_signal() which takes an additional parameter, 'from_ancestor_ns' and have kill_pid_info_as_uid() use that helper directly. The 'from_ancestor_ns' parameter will be used in a follow-on patch. Signed-off-by:
-
Oleg Nesterov authored
(This is a modified version of the patch submitted by Oleg Nesterov http://lkml.org/lkml/2008/11/18/249 and tries to address comments that came up in that discussion) init ignores the SIG_DFL signals but we queue them anyway, including SIGKILL. This is mostly OK, the signal will be dropped silently when dequeued, but the pending SIGKILL has 2 bad implications: - it implies fatal_signal_pending(), so we confuse things like wait_for_completion_killable/lock_page_killable. - for the sub-namespace inits, the pending SIGKILL can mask (legacy_queue) the subsequent SIGKILL from the parent namespace which must kill cinit reliably. (preparation, cinits don't have SIGNAL_UNKILLABLE yet) The patch can't help when init is ptraced, but ptracing of init is not "safe" anyway. Signed-off-by:
Roland McGrath <roland@redhat.com> Signed-off-by:
Oleg Nesterov <oleg@tv-sign.ru> Cc: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Daniel Lezcano <daniel.lezcano@free.fr> Signed-off-by:
-
Oleg Nesterov authored
Container-init must behave like global-init to processes within the container and hence it must be immune to unhandled fatal signals from within the container (i.e SIG_DFL signals that terminate the process). But the same container-init must behave like a normal process to processes in ancestor namespaces and so if it receives the same fatal signal from a process in ancestor namespace, the signal must be processed. Implementing these semantics requires that send_signal() determine pid namespace of the sender but since signals can originate from workqueues/ interrupt-handlers, determining pid namespace of sender may not always be possible or safe. This patchset implements the design/simplified semantics suggested by Oleg Nesterov. The simplified semantics for container-init are: - container-init must never be terminated by a signal from a descendant process. - container-init must never be immune to SIGKILL from an ancestor namespace (so a process in parent namespace must always be able to terminate a descendant container). - container-init may be immune to unhandled fatal signals (like SIGUSR1) even if they are from ancestor namespace. SIGKILL/SIGSTOP are the only reliable signals to a container-init from ancestor namespace. This patch: Based on an earlier patch submitted by Oleg Nesterov and comments from Roland McGrath (http://lkml.org/lkml/2008/11/19/258). The handler parameter is currently unused in the tracehook functions. Besides, the tracehook functions are called with siglock held, so the functions can check the handler if they later need to. Removing the parameter simiplifies changes to sig_ignored() in a follow-on patch. Signed-off-by:
-
Oleg Nesterov authored
do_wait(WSTOPPED) assumes that p->state must be == TASK_STOPPED, this is not true if the leader is already dead. Check SIGNAL_STOP_STOPPED instead and use signal->group_exit_code. Trivial test-case: void *tfunc(void *arg) { pause(); return NULL; } int main(void) { pthread_t thr; pthread_create(&thr, NULL, tfunc, NULL); pthread_exit(NULL); return 0; } It doesn't react to ^Z (and then to ^C or ^\). The task is stopped, but bash can't see this. The bug is very old, and it was reported multiple times. This patch was sent more than a year ago (http://marc.info/?t=119713920000003) but it was ignored. This change also fixes other oddities (but not all) in this area. For example, before this patch: $ sleep 100 ^Z [1]+ Stopped sleep 100 $ strace -p `pidof sleep` Process 11442 attached - interrupt to quit strace hangs in do_wait(), because ->exit_code was already consumed by bash. After this patch, strace happily proceeds: --- SIGTSTP (Stopped) @ 0 (0) --- restart_syscall(<... resuming interrupted call ...> To me, this looks much more "natural" and correct. Another example. Let's suppose we have the main thread M and sub-thread T, the process is stopped, and its parent did wait(WSTOPPED). Now we can ptrace T but not M. This looks at least strange to me. Imho, do_wait() should not confuse the per-thread ptrace stops with the per-process job control stops. Signed-off-by: -
David Rientjes authored
Kthreads that have the PF_THREAD_BOUND bit set in their flags are bound to a specific cpu. Thus, their set of allowed cpus shall not change. This patch prevents such threads from attaching to non-root cpusets. They do not have mempolicies that restrict them to a subset of system nodes and, since their cpumask may never change, they cannot use any of the features of cpusets. The tasks will forever be a member of the root cpuset and will be returned when listing the tasks attached to that cpuset. Cc: Paul Menage <menage@google.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Dhaval Giani <dhaval@linux.vnet.ibm.com> Signed-off-by:
David Rientjes <rientjes@google.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by:
-
Paul Menage authored
Allow cpusets to be configured/built on non-SMP systems Currently it's impossible to build cpusets under UML on x86-64, since cpusets depends on SMP and x86-64 UML doesn't support SMP. There's code in cpusets that doesn't depend on SMP. This patch surrounds the minimum amount of cpusets code with #ifdef CONFIG_SMP in order to allow cpusets to build/run on UP systems (for testing purposes under UML). Reviewed-by:
Li Zefan <lizf@cn.fujitsu.com> Signed-off-by:
Paul Menage <menage@google.com> Cc: Ingo Molnar <mingo@elte.hu> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by:
-
David Rientjes authored
The cpuset_zone_allowed() variants are actually only a function of the zone's node. Cc: Paul Menage <menage@google.com> Acked-by:
Christoph Lameter <cl@linux-foundation.org> Cc: Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by:
-
Li Zefan authored
Use cgroup_scanner.data, instead of introducing cpuset_hotplug_scanner. Signed-off-by:
-
Li Zefan authored
When writing to cpuset.mems, cpuset has to update its mems_allowed before calling update_tasks_nodemask(), but this function might return -ENOMEM. To avoid this rare case, we allocate the memory before changing mems_allowed, and then pass to update_tasks_nodemask(). Similar to what update_cpumask() does. Signed-off-by:
-
Li Zefan authored
This patch uses cgroup_scan_tasks() to rebind tasks' vmas to new cpuset's mems_allowed. Not only simplify the code largely, but also avoid allocating an array to hold mm pointers of all the tasks in the cpuset. This array can be big (size > PAGESIZE) if we have lots of tasks in that cpuset, thus has a chance to fail the allocation when under memory stress. Signed-off-by:
-
Li Zefan authored
We need to pass some data to test_task() or process_task() in some cases. Will be used later. Signed-off-by:
-
Li Zefan authored
Change to cpuset->cpus_allowed and cpuset->mems_allowed should be protected by callback_mutex, otherwise the reader may read wrong cpus/mems. This is cpuset's lock rule. Signed-off-by:
-
Daisuke Nishimura authored
Current mem_cgroup_cache_charge is a bit complicated especially in the case of shmem's swap-in. This patch cleans it up by using try_charge_swapin and commit_charge_swapin. Signed-off-by:
Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Acked-by:
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
It's pointed out that swap_cgroup's message at swapon() is nonsense. Because * It can be calculated very easily if all necessary information is written in Kconfig. * It's not necessary to annoying people at every swapon(). In other view, now, memory usage per swp_entry is reduced to 2bytes from 8bytes(64bit) and I think it's reasonably small. Reported-by:Hugh Dickins <hugh@veritas.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Try to use CSS ID for records in swap_cgroup. By this, on 64bit machine, size of swap_cgroup goes down to 2 bytes from 8bytes. This means, when 2GB of swap is equipped, (assume the page size is 4096bytes) From size of swap_cgroup = 2G/4k * 8 = 4Mbytes. To size of swap_cgroup = 2G/4k * 2 = 1Mbytes. Reduction is large. Of course, there are trade-offs. This CSS ID will add overhead to swap-in/swap-out/swap-free. But in general, - swap is a resource which the user tend to avoid use. - If swap is never used, swap_cgroup area is not used. - Reading traditional manuals, size of swap should be proportional to size of memory. Memory size of machine is increasing now. I think reducing size of swap_cgroup makes sense. Note: - ID->CSS lookup routine has no locks, it's under RCU-Read-Side. - memcg can be obsolete at rmdir() but not freed while refcnt from swap_cgroup is available. Changelog v4->v5: - reworked on to memcg-charge-swapcache-to-proper-memcg.patch Changlog ->v4: - fixed not configured case. - deleted unnecessary comments. - fixed NULL pointer bug. - fixed message in dmesg. [nishimura@mxp.nes.nec.co.jp: css_tryget can be called twice in !PageCgroupUsed case] Signed-off-by: -
Daisuke Nishimura authored
memcg_test.txt says at 4.1: This swap-in is one of the most complicated work. In do_swap_page(), following events occur when pte is unchanged. (1) the page (SwapCache) is looked up. (2) lock_page() (3) try_charge_swapin() (4) reuse_swap_page() (may call delete_swap_cache()) (5) commit_charge_swapin() (6) swap_free(). Considering following situation for example. (A) The page has not been charged before (2) and reuse_swap_page() doesn't call delete_from_swap_cache(). (B) The page has not been charged before (2) and reuse_swap_page() calls delete_from_swap_cache(). (C) The page has been charged before (2) and reuse_swap_page() doesn't call delete_from_swap_cache(). (D) The page has been charged before (2) and reuse_swap_page() calls delete_from_swap_cache(). memory.usage/memsw.usage changes to this page/swp_entry will be Case (A) (B) (C) (D) Event Before (2) 0/ 1 0/ 1 1/ 1 1/ 1 =========================================== (3) +1/+1 +1/+1 +1/+1 +1/+1 (4) - 0/ 0 - -1/ 0 (5) 0/-1 0/ 0 -1/-1 0/ 0 (6) - 0/-1 - 0/-1 =========================================== Result 1/ 1 1/ 1 1/ 1 1/ 1 In any cases, charges to this page should be 1/ 1. In case of (D), mem_cgroup_try_get_from_swapcache() returns NULL (because lookup_swap_cgroup() returns NULL), so "+1/+1" at (3) means charges to the memcg("foo") to which the "current" belongs. OTOH, "-1/0" at (4) and "0/-1" at (6) means uncharges from the memcg("baa") to which the page has been charged. So, if the "foo" and "baa" is different(for example because of task move), this charge will be moved from "baa" to "foo". I think this is an unexpected behavior. This patch fixes this by modifying mem_cgroup_try_get_from_swapcache() to return the memcg to which the swapcache has been charged if PCG_USED bit is set. IIUC, checking PCG_USED bit of swapcache is safe under page lock. Signed-off-by: -
KOSAKI Motohiro authored
commit 4f98a2fe (vmscan: split LRU lists into anon & file sets) removed mem_cgroup_reclaim_imbalance(), but there are some leftovers in memcontrol.h. Signed-off-by:
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Balbir Singh <balbir@linux.vnet.ibm.com> Signed-off-by:
-
KOSAKI Motohiro authored
Currently, mem_cgroup_calc_mapped_ratio() is unused at all. it can be removed and KAMEZAWA-san suggested it. Signed-off-by:
Balbir Singh <balbir@linux.vnet.ibm.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by:
-
Balbir Singh authored
Add RSS and swap to OOM output from memcg Display memcg values like failcnt, usage and limit when an OOM occurs due to memcg. Thanks to Johannes Weiner, Li Zefan, David Rientjes, Kamezawa Hiroyuki, Daisuke Nishimura and KOSAKI Motohiro for review. Sample output ------------- Task in /a/x killed as a result of limit of /a memory: usage 1048576kB, limit 1048576kB, failcnt 4183 memory+swap: usage 1400964kB, limit 9007199254740991kB, failcnt 0 [akpm@linux-foundation.org: compilation fix] [akpm@linux-foundation.org: fix kerneldoc and whitespace] [akpm@linux-foundation.org: add printk facility level] Signed-off-by:
-
KAMEZAWA Hiroyuki authored
This patch tries to fix OOM Killer problems caused by hierarchy. Now, memcg itself has OOM KILL function (in oom_kill.c) and tries to kill a task in memcg. But, when hierarchy is used, it's broken and correct task cannot be killed. For example, in following cgroup /groupA/ hierarchy=1, limit=1G, 01 nolimit 02 nolimit All tasks' memory usage under /groupA, /groupA/01, groupA/02 is limited to groupA's 1Gbytes but OOM Killer just kills tasks in groupA. This patch provides makes the bad process be selected from all tasks under hierarchy. BTW, currently, oom_jiffies is updated against groupA in above case. oom_jiffies of tree should be updated. To see how oom_jiffies is used, please check mem_cgroup_oom_called() callers. [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: const fix] Signed-off-by:
-
KAMEZAWA Hiroyuki authored
As pointed out, shrinking memcg's limit should return -EBUSY after reasonable retries. This patch tries to fix the current behavior of shrink_usage. Before looking into "shrink should return -EBUSY" problem, we should fix hierarchical reclaim code. It compares current usage and current limit, but it only makes sense when the kernel reclaims memory because hit limits. This is also a problem. What this patch does are. 1. add new argument "shrink" to hierarchical reclaim. If "shrink==true", hierarchical reclaim returns immediately and the caller checks the kernel should shrink more or not. (At shrinking memory, usage is always smaller than limit. So check for usage < limit is useless.) 2. For adjusting to above change, 2 changes in "shrink"'s retry path. 2-a. retry_count depends on # of children because the kernel visits the children under hierarchy one by one. 2-b. rather than checking return value of hierarchical_reclaim's progress, compares usage-before-shrink and usage-after-shrink. If usage-before-shrink <= usage-after-shrink, retry_count is decremented. Reported-by: -
KAMEZAWA Hiroyuki authored
Clean up memory.stat file routine and show "total" hierarchical stat. This patch does - renamed get_all_zonestat to be get_local_zonestat. - remove old mem_cgroup_stat_desc, which is only for per-cpu stat. - add mcs_stat to cover both of per-cpu/per-lru stat. - add "total" stat of hierarchy (*) - add a callback system to scan all memcg under a root. == "total" is added. [kamezawa@localhost ~]$ cat /opt/cgroup/xxx/memory.stat cache 0 rss 0 pgpgin 0 pgpgout 0 inactive_anon 0 active_anon 0 inactive_file 0 active_file 0 unevictable 0 hierarchical_memory_limit 50331648 hierarchical_memsw_limit 9223372036854775807 total_cache 65536 total_rss 192512 total_pgpgin 218 total_pgpgout 155 total_inactive_anon 0 total_active_anon 135168 total_inactive_file 61440 total_active_file 4096 total_unevictable 0 == (*) maybe the user can do calc hierarchical stat by his own program in userland but if it can be written in clean way, it's worth to be shown, I think. Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Assigning CSS ID for each memcg and use css_get_next() for scanning hierarchy. Assume folloing tree. group_A (ID=3) /01 (ID=4) /0A (ID=7) /02 (ID=10) group_B (ID=5) and task in group_A/01/0A hits limit at group_A. reclaim will be done in following order (round-robin). group_A(3) -> group_A/01 (4) -> group_A/01/0A (7) -> group_A/02(10) -> group_A -> ..... Round robin by ID. The last visited cgroup is recorded and restart from it when it start reclaim again. (More smart algorithm can be implemented..) No cgroup_mutex or hierarchy_mutex is required. Signed-off-by:
-
Li Zefan authored
There is nothing special that has to be protected by cgroup_lock, so introduce devcgroup_mtuex for it's own use. Signed-off-by:
Serge Hallyn <serue@us.ibm.com> Cc: Balbir Singh <balbir@in.ibm.com> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Signed-off-by:
-
Li Zefan authored
Since we are in cgroup write handler, so the cgrp is valid, so we don't have to hold cgroup_mutex when calling cgroup_task_count(). One similar example is in cgroup_tasks_open(). Signed-off-by:
-
Li Zefan authored
Remount can fail in either case: - wrong mount options is specified, or option 'noprefix' is changed. - a to-be-added subsys is already mounted/active. When using remount to change 'release_agent', for the above former failure case, remount will return errno with release_agent unchanged, but for the latter case, remount will return EBUSY with relase_agent changed, which is unexpected I think: # mount -t cgroup -o cpu xxx /cgrp1 # mount -t cgroup -o cpuset,release_agent=agent1 yyy /cgrp2 # cat /cgrp2/release_agent agent1 # mount -t cgroup -o remount,cpuset,noprefix,release_agent=agent2 yyy /cgrp2 mount: /cgrp2 not mounted already, or bad option # cat /cgrp2/release_agent agent1 <-- ok # mount -t cgroup -o remount,cpu,cpuset,release_agent=agent2 yyy /cgrp2 mount: /cgrp2 is busy # cat /cgrp2/release_agent agent2 <-- unexpected! Signed-off-by:
-
Li Zefan authored
We have some read-only files and write-only files, but currently they are all set to 0644, which is counter-intuitive and cause trouble for some cgroup tools like libcgroup. This patch adds 'mode' to struct cftype to allow cgroup subsys to set it's own files' file mode, and for the most cases cft->mode can be default to 0 and cgroup will figure out proper mode. Acked-by:
-
Li Zefan authored
This won't remove cpuacct from the mounted hierachy: # mount -t cgroup -o cpu,cpuacct xxx /mnt # mount -o remount,cpu /mnt Because for this usage mount(8) will append the new options to the original options. And this will get you right: # mount [-t cgroup] -o remount,cpu xxx /mnt Also document how to specify or change release_agent. Signed-off-by:
-
Jesper Juhl authored
Reduces object file size a bit: Before: $ size kernel/cgroup.o text data bss dec hex filename 21593 7804 4924 34321 8611 kernel/cgroup.o After: $ size kernel/cgroup.o text data bss dec hex filename 21537 7744 4924 34205 859d kernel/cgroup.o Signed-off-by:
Jesper Juhl <jj@chaosbits.net> Cc: Paul Menage <menage@google.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
In following situation, with memory subsystem, /groupA use_hierarchy==1 /01 some tasks /02 some tasks /03 some tasks /04 empty When tasks under 01/02/03 hit limit on /groupA, hierarchical reclaim is triggered and the kernel walks tree under groupA. In this case, rmdir /groupA/04 fails with -EBUSY frequently because of temporal refcnt from the kernel. In general. cgroup can be rmdir'd if there are no children groups and no tasks. Frequent fails of rmdir() is not useful to users. (And the reason for -EBUSY is unknown to users.....in most cases) This patch tries to modify above behavior, by - retries if css_refcnt is got by someone. - add "return value" to pre_destroy() and allows subsystem to say "we're really busy!" Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Patch for Per-CSS(Cgroup Subsys State) ID and private hierarchy code. This patch attaches unique ID to each css and provides following. - css_lookup(subsys, id) returns pointer to struct cgroup_subysys_state of id. - css_get_next(subsys, id, rootid, depth, foundid) returns the next css under "root" by scanning When cgroup_subsys->use_id is set, an id for css is maintained. The cgroup framework only parepares - css_id of root css for subsys - id is automatically attached at creation of css. - id is *not* freed automatically. Because the cgroup framework don't know lifetime of cgroup_subsys_state. free_css_id() function is provided. This must be called by subsys. There are several reasons to develop this. - Saving space .... For example, memcg's swap_cgroup is array of pointers to cgroup. But it is not necessary to be very fast. By replacing pointers(8bytes per ent) to ID (2byes per ent), we can reduce much amount of memory usage. - Scanning without lock. CSS_ID provides "scan id under this ROOT" function. By this, scanning css under root can be written without locks. ex) do { rcu_read_lock(); next = cgroup_get_next(subsys, id, root, &found); /* check sanity of next here */ css_tryget(); rcu_read_unlock(); id = found + 1 } while(...) Characteristics: - Each css has unique ID under subsys. - Lifetime of ID is controlled by subsys. - css ID contains "ID" and "Depth in hierarchy" and stack of hierarchy - Allowed ID is 1-65535, ID 0 is UNUSED ID. Design Choices: - scan-by-ID v.s. scan-by-tree-walk. As /proc's pid scan does, scan-by-ID is robust when scanning is done by following kind of routine. scan -> rest a while(release a lock) -> conitunue from interrupted memcg's hierarchical reclaim does this. - When subsys->use_id is set, # of css in the system is limited to 65535. [bharata@linux.vnet.ibm.com: remove rcu_read_lock() from css_get_next()] Signed-off-by:Bharata B Rao <bharata@linux.vnet.ibm.com> Signed-off-by:
-
