summaryrefslogtreecommitdiffstats
path: root/kernel/trace
AgeCommit message (Collapse)AuthorFilesLines
2013-11-16Merge tag 'trace-3.13' of ↵Linus Torvalds14-209/+518
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace Pull tracing update from Steven Rostedt: "This batch of changes is mostly clean ups and small bug fixes. The only real feature that was added this release is from Namhyung Kim, who introduced "set_graph_notrace" filter that lets you run the function graph tracer and not trace particular functions and their call chain. Tom Zanussi added some updates to the ftrace multibuffer tracing that made it more consistent with the top level tracing. One of the fixes for perf function tracing required an API change in RCU; the addition of "rcu_is_watching()". As Paul McKenney is pushing that change in this release too, he gave me a branch that included all the changes to get that working, and I pulled that into my tree in order to complete the perf function tracing fix" * tag 'trace-3.13' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace: tracing: Add rcu annotation for syscall trace descriptors tracing: Do not use signed enums with unsigned long long in fgragh output tracing: Remove unused function ftrace_off_permanent() tracing: Do not assign filp->private_data to freed memory tracing: Add helper function tracing_is_disabled() tracing: Open tracer when ftrace_dump_on_oops is used tracing: Add support for SOFT_DISABLE to syscall events tracing: Make register/unregister_ftrace_command __init tracing: Update event filters for multibuffer recordmcount.pl: Add support for __fentry__ ftrace: Have control op function callback only trace when RCU is watching rcu: Do not trace rcu_is_watching() functions ftrace/x86: skip over the breakpoint for ftrace caller trace/trace_stat: use rbtree postorder iteration helper instead of opencoding ftrace: Add set_graph_notrace filter ftrace: Narrow down the protected area of graph_lock ftrace: Introduce struct ftrace_graph_data ftrace: Get rid of ftrace_graph_filter_enabled tracing: Fix potential out-of-bounds in trace_get_user() tracing: Show more exact help information about snapshot
2013-11-14Merge branch 'for-3.13/core' of git://git.kernel.dk/linux-blockLinus Torvalds1-8/+28
Pull block IO core updates from Jens Axboe: "This is the pull request for the core changes in the block layer for 3.13. It contains: - The new blk-mq request interface. This is a new and more scalable queueing model that marries the best part of the request based interface we currently have (which is fully featured, but scales poorly) and the bio based "interface" which the new drivers for high IOPS devices end up using because it's much faster than the request based one. The bio interface has no block layer support, since it taps into the stack much earlier. This means that drivers end up having to implement a lot of functionality on their own, like tagging, timeout handling, requeue, etc. The blk-mq interface provides all these. Some drivers even provide a switch to select bio or rq and has code to handle both, since things like merging only works in the rq model and hence is faster for some workloads. This is a huge mess. Conversion of these drivers nets us a substantial code reduction. Initial results on converting SCSI to this model even shows an 8x improvement on single queue devices. So while the model was intended to work on the newer multiqueue devices, it has substantial improvements for "classic" hardware as well. This code has gone through extensive testing and development, it's now ready to go. A pull request is coming to convert virtio-blk to this model will be will be coming as well, with more drivers scheduled for 3.14 conversion. - Two blktrace fixes from Jan and Chen Gang. - A plug merge fix from Alireza Haghdoost. - Conversion of __get_cpu_var() from Christoph Lameter. - Fix for sector_div() with 64-bit divider from Geert Uytterhoeven. - A fix for a race between request completion and the timeout handling from Jeff Moyer. This is what caused the merge conflict with blk-mq/core, in case you are looking at that. - A dm stacking fix from Mike Snitzer. - A code consolidation fix and duplicated code removal from Kent Overstreet. - A handful of block bug fixes from Mikulas Patocka, fixing a loop crash and memory corruption on blk cg. - Elevator switch bug fix from Tomoki Sekiyama. A heads-up that I had to rebase this branch. Initially the immutable bio_vecs had been queued up for inclusion, but a week later, it became clear that it wasn't fully cooked yet. So the decision was made to pull this out and postpone it until 3.14. It was a straight forward rebase, just pruning out the immutable series and the later fixes of problems with it. The rest of the patches applied directly and no further changes were made" * 'for-3.13/core' of git://git.kernel.dk/linux-block: (31 commits) block: replace IS_ERR and PTR_ERR with PTR_ERR_OR_ZERO block: replace IS_ERR and PTR_ERR with PTR_ERR_OR_ZERO block: Do not call sector_div() with a 64-bit divisor kernel: trace: blktrace: remove redundent memcpy() in compat_blk_trace_setup() block: Consolidate duplicated bio_trim() implementations block: Use rw_copy_check_uvector() block: Enable sysfs nomerge control for I/O requests in the plug list block: properly stack underlying max_segment_size to DM device elevator: acquire q->sysfs_lock in elevator_change() elevator: Fix a race in elevator switching and md device initialization block: Replace __get_cpu_var uses bdi: test bdi_init failure block: fix a probe argument to blk_register_region loop: fix crash if blk_alloc_queue fails blk-core: Fix memory corruption if blkcg_init_queue fails block: fix race between request completion and timeout handling blktrace: Send BLK_TN_PROCESS events to all running traces blk-mq: don't disallow request merges for req->special being set blk-mq: mq plug list breakage blk-mq: fix for flush deadlock ...
2013-11-12Merge branch 'sched-core-for-linus' of ↵Linus Torvalds3-3/+20
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip Pull scheduler changes from Ingo Molnar: "The main changes in this cycle are: - (much) improved CONFIG_NUMA_BALANCING support from Mel Gorman, Rik van Riel, Peter Zijlstra et al. Yay! - optimize preemption counter handling: merge the NEED_RESCHED flag into the preempt_count variable, by Peter Zijlstra. - wait.h fixes and code reorganization from Peter Zijlstra - cfs_bandwidth fixes from Ben Segall - SMP load-balancer cleanups from Peter Zijstra - idle balancer improvements from Jason Low - other fixes and cleanups" * 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (129 commits) ftrace, sched: Add TRACE_FLAG_PREEMPT_RESCHED stop_machine: Fix race between stop_two_cpus() and stop_cpus() sched: Remove unnecessary iteration over sched domains to update nr_busy_cpus sched: Fix asymmetric scheduling for POWER7 sched: Move completion code from core.c to completion.c sched: Move wait code from core.c to wait.c sched: Move wait.c into kernel/sched/ sched/wait: Fix __wait_event_interruptible_lock_irq_timeout() sched: Avoid throttle_cfs_rq() racing with period_timer stopping sched: Guarantee new group-entities always have weight sched: Fix hrtimer_cancel()/rq->lock deadlock sched: Fix cfs_bandwidth misuse of hrtimer_expires_remaining sched: Fix race on toggling cfs_bandwidth_used sched: Remove extra put_online_cpus() inside sched_setaffinity() sched/rt: Fix task_tick_rt() comment sched/wait: Fix build breakage sched/wait: Introduce prepare_to_wait_event() sched/wait: Add ___wait_cond_timeout() to wait_event*_timeout() too sched: Remove get_online_cpus() usage sched: Fix race in migrate_swap_stop() ...
2013-11-11tracing: Add rcu annotation for syscall trace descriptorsSteven Rostedt (Red Hat)1-2/+2
sparse complains about the enter/exit_sysycall_files[] variables being dereferenced with rcu_dereference_sched(). The fields need to be annotated with __rcu. Reported-by: kbuild test robot <fengguang.wu@intel.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-11-11ftrace, sched: Add TRACE_FLAG_PREEMPT_RESCHEDPeter Zijlstra3-3/+20
Since the introduction of PREEMPT_NEED_RESCHED in: f27dde8deef3 ("sched: Add NEED_RESCHED to the preempt_count") we need to be able to look at both TIF_NEED_RESCHED and PREEMPT_NEED_RESCHED to understand the full preemption behaviour. Add it to the trace output. Signed-off-by: Peter Zijlstra <peterz@infradead.org> Acked-by: Steven Rostedt <rostedt@goodmis.org> Cc: Fengguang Wu <fengguang.wu@intel.com> Cc: Huang Ying <ying.huang@intel.com> Cc: Yuanhan Liu <yuanhan.liu@linux.intel.com> Link: http://lkml.kernel.org/r/20131004152826.GP3081@twins.programming.kicks-ass.net Signed-off-by: Ingo Molnar <mingo@kernel.org>
2013-11-08kernel: trace: blktrace: remove redundent memcpy() in compat_blk_trace_setup()Chen Gang1-2/+1
do_blk_trace_setup() will fully initialize 'buts.name', so can remove the related memcpy(). And also use BLKTRACE_BDEV_SIZE and ARRAY_SIZE instead of hard code number '32'. Signed-off-by: Chen Gang <gang.chen@asianux.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2013-11-08blktrace: Send BLK_TN_PROCESS events to all running tracesJan Kara1-6/+27
Currently each task sends BLK_TN_PROCESS event to the first traced device it interacts with after a new trace is started. When there are several traced devices and the task accesses more devices, this logic can result in BLK_TN_PROCESS being sent several times to some devices while it is never sent to other devices. Thus blkparse doesn't display command name when parsing some blktrace files. Fix the problem by sending BLK_TN_PROCESS event to all traced devices when a task interacts with any of them. Signed-off-by: Jan Kara <jack@suse.cz> Review-by: Jeff Moyer <jmoyer@redhat.com> Signed-off-by: Jens Axboe <axboe@kernel.dk>
2013-11-06tracing: Do not use signed enums with unsigned long long in fgragh outputSteven Rostedt (Red Hat)2-11/+13
The duration field of print_graph_duration() can also be used to do the space filling by passing an enum in it: DURATION_FILL_FULL DURATION_FILL_START DURATION_FILL_END The problem is that these are enums and defined as negative, but the duration field is unsigned long long. Most archs are fine with this but blackfin fails to compile because of it: kernel/built-in.o: In function `print_graph_duration': kernel/trace/trace_functions_graph.c:782: undefined reference to `__ucmpdi2' Overloading a unsigned long long with an signed enum is just bad in principle. We can accomplish the same thing by using part of the flags field instead. Cc: Mike Frysinger <vapier@gentoo.org> Cc: Frederic Weisbecker <fweisbec@gmail.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-11-06tracing: Remove unused function ftrace_off_permanent()Steven Rostedt (Red Hat)1-15/+0
In the past, ftrace_off_permanent() was called if something strange was detected. But the ftrace_bug() now handles all the anomolies that can happen with ftrace (function tracing), and there are no uses of ftrace_off_permanent(). Get rid of it. Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-11-06tracing: Do not assign filp->private_data to freed memoryGeyslan G. Bem1-1/+8
In system_tr_open(), the filp->private_data can be assigned the 'dir' variable even if it was freed. This is on the error path, and is harmless because the error return code will prevent filp->private_data from being used. But for correctness, we should not assign it to a recently freed variable, as that can cause static tools to give false warnings. Also have both subsystem_open() and system_tr_open() return -ENODEV if tracing has been disabled. Link: http://lkml.kernel.org/r/1383764571-7318-1-git-send-email-geyslan@gmail.com Signed-off-by: Geyslan G. Bem <geyslan@gmail.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-11-06perf/ftrace: Fix paranoid level for enabling function tracerSteven Rostedt1-1/+1
The current default perf paranoid level is "1" which has "perf_paranoid_kernel()" return false, and giving any operations that use it, access to normal users. Unfortunately, this includes function tracing and normal users should not be allowed to enable function tracing by default. The proper level is defined at "-1" (full perf access), which "perf_paranoid_tracepoint_raw()" will only give access to. Use that check instead for enabling function tracing. Reported-by: Dave Jones <davej@redhat.com> Reported-by: Vince Weaver <vincent.weaver@maine.edu> Tested-by: Vince Weaver <vincent.weaver@maine.edu> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Frederic Weisbecker <fweisbec@gmail.com> Cc: stable@vger.kernel.org # 3.4+ CVE: CVE-2013-2930 Fixes: ced39002f5ea ("ftrace, perf: Add support to use function tracepoint in perf") Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-11-06tracing: Add helper function tracing_is_disabled()Geyslan G. Bem2-0/+6
This patch creates the function 'tracing_is_disabled', which can be used outside of trace.c. Link: http://lkml.kernel.org/r/1382141754-12155-1-git-send-email-geyslan@gmail.com Signed-off-by: Geyslan G. Bem <geyslan@gmail.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-11-06tracing: Open tracer when ftrace_dump_on_oops is usedCody P Schafer1-0/+11
With ftrace_dump_on_oops, we previously did not open the tracer in question, sometimes causing the trace output to be useless. For example, the function_graph tracer with tracing_thresh set dumped via ftrace_dump_on_oops would show a series of '}' indented at different levels, but no function names. call trace->open() (and do a few other fixups copied from the normal dump path) to make the output more intelligible. Link: http://lkml.kernel.org/r/1382554197-16961-1-git-send-email-cody@linux.vnet.ibm.com Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-11-05tracing: Add support for SOFT_DISABLE to syscall eventsTom Zanussi2-12/+34
The original SOFT_DISABLE patches didn't add support for soft disable of syscall events; this adds it. Add an array of ftrace_event_file pointers indexed by syscall number to the trace array and remove the existing enabled bitmaps, which as a result are now redundant. The ftrace_event_file structs in turn contain the soft disable flags we need for per-syscall soft disable accounting. Adding ftrace_event_files also means we can remove the USE_CALL_FILTER bit, thus enabling multibuffer filter support for syscall events. Link: http://lkml.kernel.org/r/6e72b566e85d8df8042f133efbc6c30e21fb017e.1382620672.git.tom.zanussi@linux.intel.com Signed-off-by: Tom Zanussi <tom.zanussi@linux.intel.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-11-05tracing: Make register/unregister_ftrace_command __initTom Zanussi2-4/+12
register/unregister_ftrace_command() are only ever called from __init functions, so can themselves be made __init. Also make register_snapshot_cmd() __init for the same reason. Link: http://lkml.kernel.org/r/d4042c8cadb7ae6f843ac9a89a24e1c6a3099727.1382620672.git.tom.zanussi@linux.intel.com Signed-off-by: Tom Zanussi <tom.zanussi@linux.intel.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-11-05tracing: Update event filters for multibufferTom Zanussi12-91/+239
The trace event filters are still tied to event calls rather than event files, which means you don't get what you'd expect when using filters in the multibuffer case: Before: # echo 'bytes_alloc > 8192' > /sys/kernel/debug/tracing/events/kmem/kmalloc/filter # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/filter bytes_alloc > 8192 # mkdir /sys/kernel/debug/tracing/instances/test1 # echo 'bytes_alloc > 2048' > /sys/kernel/debug/tracing/instances/test1/events/kmem/kmalloc/filter # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/filter bytes_alloc > 2048 # cat /sys/kernel/debug/tracing/instances/test1/events/kmem/kmalloc/filter bytes_alloc > 2048 Setting the filter in tracing/instances/test1/events shouldn't affect the same event in tracing/events as it does above. After: # echo 'bytes_alloc > 8192' > /sys/kernel/debug/tracing/events/kmem/kmalloc/filter # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/filter bytes_alloc > 8192 # mkdir /sys/kernel/debug/tracing/instances/test1 # echo 'bytes_alloc > 2048' > /sys/kernel/debug/tracing/instances/test1/events/kmem/kmalloc/filter # cat /sys/kernel/debug/tracing/events/kmem/kmalloc/filter bytes_alloc > 8192 # cat /sys/kernel/debug/tracing/instances/test1/events/kmem/kmalloc/filter bytes_alloc > 2048 We'd like to just move the filter directly from ftrace_event_call to ftrace_event_file, but there are a couple cases that don't yet have multibuffer support and therefore have to continue using the current event_call-based filters. For those cases, a new USE_CALL_FILTER bit is added to the event_call flags, whose main purpose is to keep the old behavior for those cases until they can be updated with multibuffer support; at that point, the USE_CALL_FILTER flag (and the new associated call_filter_check_discard() function) can go away. The multibuffer support also made filter_current_check_discard() redundant, so this change removes that function as well and replaces it with filter_check_discard() (or call_filter_check_discard() as appropriate). Link: http://lkml.kernel.org/r/f16e9ce4270c62f46b2e966119225e1c3cca7e60.1382620672.git.tom.zanussi@linux.intel.com Signed-off-by: Tom Zanussi <tom.zanussi@linux.intel.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-11-05ftrace: Have control op function callback only trace when RCU is watchingSteven Rostedt (Red Hat)1-0/+9
Dave Jones reported that trinity would be able to trigger the following back trace: =============================== [ INFO: suspicious RCU usage. ] 3.10.0-rc2+ #38 Not tainted ------------------------------- include/linux/rcupdate.h:771 rcu_read_lock() used illegally while idle! other info that might help us debug this: RCU used illegally from idle CPU! rcu_scheduler_active = 1, debug_locks = 0 RCU used illegally from extended quiescent state! 1 lock held by trinity-child1/18786: #0: (rcu_read_lock){.+.+..}, at: [<ffffffff8113dd48>] __perf_event_overflow+0x108/0x310 stack backtrace: CPU: 3 PID: 18786 Comm: trinity-child1 Not tainted 3.10.0-rc2+ #38 0000000000000000 ffff88020767bac8 ffffffff816e2f6b ffff88020767baf8 ffffffff810b5897 ffff88021de92520 0000000000000000 ffff88020767bbf8 0000000000000000 ffff88020767bb78 ffffffff8113ded4 ffffffff8113dd48 Call Trace: [<ffffffff816e2f6b>] dump_stack+0x19/0x1b [<ffffffff810b5897>] lockdep_rcu_suspicious+0xe7/0x120 [<ffffffff8113ded4>] __perf_event_overflow+0x294/0x310 [<ffffffff8113dd48>] ? __perf_event_overflow+0x108/0x310 [<ffffffff81309289>] ? __const_udelay+0x29/0x30 [<ffffffff81076054>] ? __rcu_read_unlock+0x54/0xa0 [<ffffffff816f4000>] ? ftrace_call+0x5/0x2f [<ffffffff8113dfa1>] perf_swevent_overflow+0x51/0xe0 [<ffffffff8113e08f>] perf_swevent_event+0x5f/0x90 [<ffffffff8113e1c9>] perf_tp_event+0x109/0x4f0 [<ffffffff8113e36f>] ? perf_tp_event+0x2af/0x4f0 [<ffffffff81074630>] ? __rcu_read_lock+0x20/0x20 [<ffffffff8112d79f>] perf_ftrace_function_call+0xbf/0xd0 [<ffffffff8110e1e1>] ? ftrace_ops_control_func+0x181/0x210 [<ffffffff81074630>] ? __rcu_read_lock+0x20/0x20 [<ffffffff81100cae>] ? rcu_eqs_enter_common+0x5e/0x470 [<ffffffff8110e1e1>] ftrace_ops_control_func+0x181/0x210 [<ffffffff816f4000>] ftrace_call+0x5/0x2f [<ffffffff8110e229>] ? ftrace_ops_control_func+0x1c9/0x210 [<ffffffff816f4000>] ? ftrace_call+0x5/0x2f [<ffffffff81074635>] ? debug_lockdep_rcu_enabled+0x5/0x40 [<ffffffff81074635>] ? debug_lockdep_rcu_enabled+0x5/0x40 [<ffffffff81100cae>] ? rcu_eqs_enter_common+0x5e/0x470 [<ffffffff8110112a>] rcu_eqs_enter+0x6a/0xb0 [<ffffffff81103673>] rcu_user_enter+0x13/0x20 [<ffffffff8114541a>] user_enter+0x6a/0xd0 [<ffffffff8100f6d8>] syscall_trace_leave+0x78/0x140 [<ffffffff816f46af>] int_check_syscall_exit_work+0x34/0x3d ------------[ cut here ]------------ Perf uses rcu_read_lock() but as the function tracer can trace functions even when RCU is not currently active, this makes the rcu_read_lock() used by perf ineffective. As perf is currently the only user of the ftrace_ops_control_func() and perf is also the only function callback that actively uses rcu_read_lock(), the quick fix is to prevent the ftrace_ops_control_func() from calling its callbacks if RCU is not active. With Paul's new "rcu_is_watching()" we can tell if RCU is active or not. Reported-by: Dave Jones <davej@redhat.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Jiri Olsa <jolsa@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-11-05trace/trace_stat: use rbtree postorder iteration helper instead of opencodingCody P Schafer1-36/+5
Use rbtree_postorder_for_each_entry_safe() to destroy the rbtree instead of opencoding an alternate postorder iteration that modifies the tree Link: http://lkml.kernel.org/r/1383345566-25087-2-git-send-email-cody@linux.vnet.ibm.com Signed-off-by: Cody P Schafer <cody@linux.vnet.ibm.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-10-18ftrace: Add set_graph_notrace filterNamhyung Kim3-3/+108
The set_graph_notrace filter is analogous to set_ftrace_notrace and can be used for eliminating uninteresting part of function graph trace output. It also works with set_graph_function nicely. # cd /sys/kernel/debug/tracing/ # echo do_page_fault > set_graph_function # perf ftrace live true 2) | do_page_fault() { 2) | __do_page_fault() { 2) 0.381 us | down_read_trylock(); 2) 0.055 us | __might_sleep(); 2) 0.696 us | find_vma(); 2) | handle_mm_fault() { 2) | handle_pte_fault() { 2) | __do_fault() { 2) | filemap_fault() { 2) | find_get_page() { 2) 0.033 us | __rcu_read_lock(); 2) 0.035 us | __rcu_read_unlock(); 2) 1.696 us | } 2) 0.031 us | __might_sleep(); 2) 2.831 us | } 2) | _raw_spin_lock() { 2) 0.046 us | add_preempt_count(); 2) 0.841 us | } 2) 0.033 us | page_add_file_rmap(); 2) | _raw_spin_unlock() { 2) 0.057 us | sub_preempt_count(); 2) 0.568 us | } 2) | unlock_page() { 2) 0.084 us | page_waitqueue(); 2) 0.126 us | __wake_up_bit(); 2) 1.117 us | } 2) 7.729 us | } 2) 8.397 us | } 2) 8.956 us | } 2) 0.085 us | up_read(); 2) + 12.745 us | } 2) + 13.401 us | } ... # echo handle_mm_fault > set_graph_notrace # perf ftrace live true 1) | do_page_fault() { 1) | __do_page_fault() { 1) 0.205 us | down_read_trylock(); 1) 0.041 us | __might_sleep(); 1) 0.344 us | find_vma(); 1) 0.069 us | up_read(); 1) 4.692 us | } 1) 5.311 us | } ... Link: http://lkml.kernel.org/r/1381739066-7531-5-git-send-email-namhyung@kernel.org Signed-off-by: Namhyung Kim <namhyung@kernel.org> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-10-18ftrace: Narrow down the protected area of graph_lockNamhyung Kim1-13/+9
The parser set up is just a generic utility that uses local variables allocated by the function. There's no need to hold the graph_lock for this set up. This also makes the code simpler. Link: http://lkml.kernel.org/r/1381739066-7531-4-git-send-email-namhyung@kernel.org Signed-off-by: Namhyung Kim <namhyung@kernel.org> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-10-18ftrace: Introduce struct ftrace_graph_dataNamhyung Kim1-19/+62
The struct ftrace_graph_data is for generalizing the access to set_graph_function file. This is a preparation for adding support to set_graph_notrace. Link: http://lkml.kernel.org/r/1381739066-7531-3-git-send-email-namhyung@kernel.org Signed-off-by: Namhyung Kim <namhyung@kernel.org> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-10-18ftrace: Get rid of ftrace_graph_filter_enabledNamhyung Kim2-7/+2
The ftrace_graph_filter_enabled means that user sets function filter and it always has same meaning of ftrace_graph_count > 0. Link: http://lkml.kernel.org/r/1381739066-7531-2-git-send-email-namhyung@kernel.org Signed-off-by: Namhyung Kim <namhyung@kernel.org> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-10-18tracing: Fix potential out-of-bounds in trace_get_user()Steven Rostedt1-1/+4
Andrey reported the following report: ERROR: AddressSanitizer: heap-buffer-overflow on address ffff8800359c99f3 ffff8800359c99f3 is located 0 bytes to the right of 243-byte region [ffff8800359c9900, ffff8800359c99f3) Accessed by thread T13003: #0 ffffffff810dd2da (asan_report_error+0x32a/0x440) #1 ffffffff810dc6b0 (asan_check_region+0x30/0x40) #2 ffffffff810dd4d3 (__tsan_write1+0x13/0x20) #3 ffffffff811cd19e (ftrace_regex_release+0x1be/0x260) #4 ffffffff812a1065 (__fput+0x155/0x360) #5 ffffffff812a12de (____fput+0x1e/0x30) #6 ffffffff8111708d (task_work_run+0x10d/0x140) #7 ffffffff810ea043 (do_exit+0x433/0x11f0) #8 ffffffff810eaee4 (do_group_exit+0x84/0x130) #9 ffffffff810eafb1 (SyS_exit_group+0x21/0x30) #10 ffffffff81928782 (system_call_fastpath+0x16/0x1b) Allocated by thread T5167: #0 ffffffff810dc778 (asan_slab_alloc+0x48/0xc0) #1 ffffffff8128337c (__kmalloc+0xbc/0x500) #2 ffffffff811d9d54 (trace_parser_get_init+0x34/0x90) #3 ffffffff811cd7b3 (ftrace_regex_open+0x83/0x2e0) #4 ffffffff811cda7d (ftrace_filter_open+0x2d/0x40) #5 ffffffff8129b4ff (do_dentry_open+0x32f/0x430) #6 ffffffff8129b668 (finish_open+0x68/0xa0) #7 ffffffff812b66ac (do_last+0xb8c/0x1710) #8 ffffffff812b7350 (path_openat+0x120/0xb50) #9 ffffffff812b8884 (do_filp_open+0x54/0xb0) #10 ffffffff8129d36c (do_sys_open+0x1ac/0x2c0) #11 ffffffff8129d4b7 (SyS_open+0x37/0x50) #12 ffffffff81928782 (system_call_fastpath+0x16/0x1b) Shadow bytes around the buggy address: ffff8800359c9700: fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd fd ffff8800359c9780: fd fd fd fd fd fd fd fd fa fa fa fa fa fa fa fa ffff8800359c9800: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa ffff8800359c9880: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa ffff8800359c9900: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 =>ffff8800359c9980: 00 00 00 00 00 00 00 00 00 00 00 00 00 00[03]fb ffff8800359c9a00: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa ffff8800359c9a80: fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa fa ffff8800359c9b00: fa fa fa fa fa fa fa fa 00 00 00 00 00 00 00 00 ffff8800359c9b80: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ffff8800359c9c00: 00 00 00 00 00 00 00 00 fa fa fa fa fa fa fa fa Shadow byte legend (one shadow byte represents 8 application bytes): Addressable: 00 Partially addressable: 01 02 03 04 05 06 07 Heap redzone: fa Heap kmalloc redzone: fb Freed heap region: fd Shadow gap: fe The out-of-bounds access happens on 'parser->buffer[parser->idx] = 0;' Although the crash happened in ftrace_regex_open() the real bug occurred in trace_get_user() where there's an incrementation to parser->idx without a check against the size. The way it is triggered is if userspace sends in 128 characters (EVENT_BUF_SIZE + 1), the loop that reads the last character stores it and then breaks out because there is no more characters. Then the last character is read to determine what to do next, and the index is incremented without checking size. Then the caller of trace_get_user() usually nulls out the last character with a zero, but since the index is equal to the size, it writes a nul character after the allocated space, which can corrupt memory. Luckily, only root user has write access to this file. Link: http://lkml.kernel.org/r/20131009222323.04fd1a0d@gandalf.local.home Reported-by: Andrey Konovalov <andreyknvl@google.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-10-09tracing: Show more exact help information about snapshotWang YanQing1-1/+1
The current "help" that comes out of the snapshot file when it is not allocated looks like this: # * Snapshot is freed * # # Snapshot commands: # echo 0 > snapshot : Clears and frees snapshot buffer # echo 1 > snapshot : Allocates snapshot buffer, if not already allocated. # Takes a snapshot of the main buffer. # echo 2 > snapshot : Clears snapshot buffer (but does not allocate) # (Doesn't have to be '2' works with any number that # is not a '0' or '1') Echo 2 says that it does not allocate the buffer, which is correct, but to be more consistent with "echo 0" it should also state that it does not free. Link: http://lkml.kernel.org/r/20130914045916.GA4243@udknight Signed-off-by: Wang YanQing <udknight@gmail.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-09-09Merge tag 'trace-3.12' of ↵Linus Torvalds5-208/+64
git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace Pull tracing updates from Steven Rostedt: "Not much changes for the 3.12 merge window. The major tracing changes are still in flux, and will have to wait for 3.13. The changes for 3.12 are mostly clean ups and minor fixes. H Peter Anvin added a check to x86_32 static function tracing that helps a small segment of the kernel community. Oleg Nesterov had a few changes from 3.11, but were mostly clean ups and not worth pushing in the -rc time frame. Li Zefan had small clean up with annotating a raw_init with __init. I fixed a slight race in updating function callbacks, but the race is so small and the bug that happens when it occurs is so minor it's not even worth pushing to stable. The only real enhancement is from Alexander Z Lam that made the tracing_cpumask work for trace buffer instances, instead of them all sharing a global cpumask" * tag 'trace-3.12' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace: ftrace/rcu: Do not trace debug_lockdep_rcu_enabled() x86-32, ftrace: Fix static ftrace when early microcode is enabled ftrace: Fix a slight race in modifying what function callback gets traced tracing: Make tracing_cpumask available for all instances tracing: Kill the !CONFIG_MODULES code in trace_events.c tracing: Don't pass file_operations array to event_create_dir() tracing: Kill trace_create_file_ops() and friends tracing/syscalls: Annotate raw_init function with __init
2013-09-03ftrace: Fix a slight race in modifying what function callback gets tracedSteven Rostedt (Red Hat)1-1/+16
There's a slight race when going from a list function to a non list function. That is, when only one callback is registered to the function tracer, it gets called directly by the mcount trampoline. But if this function has filters, it may be called by the wrong functions. As the list ops callback that handles multiple callbacks that are registered to ftrace, it also handles what functions they call. While the transaction is taking place, use the list function always, and after all the updates are finished (only the functions that should be traced are being traced), then we can update the trampoline to call the function directly. Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-09-03Merge branch 'rcu/next' of ↵Ingo Molnar2-0/+22
git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu into core/rcu Pull RCU updates from Paul E. McKenney: " * Update RCU documentation. These were posted to LKML at https://lkml.org/lkml/2013/8/19/611. * Miscellaneous fixes. These were posted to LKML at https://lkml.org/lkml/2013/8/19/619. * Full-system idle detection. This is for use by Frederic Weisbecker's adaptive-ticks mechanism. Its purpose is to allow the timekeeping CPU to shut off its tick when all other CPUs are idle. These were posted to LKML at https://lkml.org/lkml/2013/8/19/648. * Improve rcutorture test coverage. These were posted to LKML at https://lkml.org/lkml/2013/8/19/675. " Signed-off-by: Ingo Molnar <mingo@kernel.org>
2013-08-22tracing: Make tracing_cpumask available for all instancesAlexander Z Lam2-17/+21
Allow tracer instances to disable tracing by cpu by moving the static global tracing_cpumask into trace_array. Link: http://lkml.kernel.org/r/921622317f239bfc2283cac2242647801ef584f2.1375980149.git.azl@google.com Cc: Vaibhav Nagarnaik <vnagarnaik@google.com> Cc: David Sharp <dhsharp@google.com> Cc: Alexander Z Lam <lambchop468@gmail.com> Signed-off-by: Alexander Z Lam <azl@google.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-08-21tracing: Kill the !CONFIG_MODULES code in trace_events.cOleg Nesterov1-12/+6
Move trace_module_nb under CONFIG_MODULES and kill the dummy trace_module_notify(). Imho it doesn't make sense to define "struct notifier_block" and its .notifier_call just to avoid "ifdef" in event_trace_init(), and all other !CONFIG_MODULES code has already gone away. Link: http://lkml.kernel.org/r/20130731173137.GA31043@redhat.com Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-08-21tracing: Don't pass file_operations array to event_create_dir()Oleg Nesterov1-34/+12
Now that event_create_dir() and __trace_add_new_event() always use the same file_operations we can kill these arguments and simplify the code. Link: http://lkml.kernel.org/r/20130731173135.GA31040@redhat.com Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-08-21tracing: Kill trace_create_file_ops() and friendsOleg Nesterov1-144/+9
trace_create_file_ops() allocates the copy of id/filter/format/enable file_operations to set "f_op->owner = mod" for fops_get(). However after the recent changes there is no reason to prevent rmmod even if one of these files is opened. A file operation can do nothing but fail after remove_event_file_dir() clears ->i_private for every file removed by trace_module_remove_events(). Kill "struct ftrace_module_file_ops" and fix the compilation errors. Link: http://lkml.kernel.org/r/20130731173132.GA31033@redhat.com Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-08-21tracing/syscalls: Annotate raw_init function with __initLi Zefan1-5/+5
init_syscall_trace() can only be called during kernel bootup only, so we can mark it and the functions it calls as __init. Link: http://lkml.kernel.org/r/51528E89.6080508@huawei.com Signed-off-by: Li Zefan <lizefan@huawei.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-08-02tracing: Fix reset of time stamps during trace_clock changesAlexander Z Lam1-12/+12
Fixed two issues with changing the timestamp clock with trace_clock: - The global buffer was reset on instance clock changes. Change this to pass the correct per-instance buffer - ftrace_now() is used to set buf->time_start in tracing_reset_online_cpus(). This was incorrect because ftrace_now() used the global buffer's clock to return the current time. Change this to use buffer_ftrace_now() which returns the current time for the correct per-instance buffer. Also removed tracing_reset_current() because it is not used anywhere Link: http://lkml.kernel.org/r/1375493777-17261-2-git-send-email-azl@google.com Cc: Vaibhav Nagarnaik <vnagarnaik@google.com> Cc: David Sharp <dhsharp@google.com> Cc: Alexander Z Lam <lambchop468@gmail.com> Cc: stable@vger.kernel.org # 3.10 Signed-off-by: Alexander Z Lam <azl@google.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-08-02tracing: Make TRACE_ITER_STOP_ON_FREE stop the correct bufferAlexander Z Lam1-1/+1
Releasing the free_buffer file in an instance causes the global buffer to be stopped when TRACE_ITER_STOP_ON_FREE is enabled. Operate on the correct buffer. Link: http://lkml.kernel.org/r/1375493777-17261-1-git-send-email-azl@google.com Cc: Vaibhav Nagarnaik <vnagarnaik@google.com> Cc: David Sharp <dhsharp@google.com> Cc: Alexander Z Lam <lambchop468@gmail.com> Cc: stable@vger.kernel.org # 3.10 Signed-off-by: Alexander Z Lam <azl@google.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-08-02tracing: Fix fields of struct trace_iterator that are zeroed by mistakeAndrew Vagin1-0/+1
tracing_read_pipe zeros all fields bellow "seq". The declaration contains a comment about that, but it doesn't help. The first field is "snapshot", it's true when current open file is snapshot. Looks obvious, that it should not be zeroed. The second field is "started". It was converted from cpumask_t to cpumask_var_t (v2.6.28-4983-g4462344), in other words it was converted from cpumask to pointer on cpumask. Currently the reference on "started" memory is lost after the first read from tracing_read_pipe and a proper object will never be freed. The "started" is never dereferenced for trace_pipe, because trace_pipe can't have the TRACE_FILE_ANNOTATE options. Link: http://lkml.kernel.org/r/1375463803-3085183-1-git-send-email-avagin@openvz.org Cc: stable@vger.kernel.org # 2.6.30 Signed-off-by: Andrew Vagin <avagin@openvz.org> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-08-01tracing/uprobes: Fail to unregister if probe event files are in useSteven Rostedt (Red Hat)1-13/+38
Uprobes suffer the same problem that kprobes have. There's a race between writing to the "enable" file and removing the probe. The probe checks for it being in use and if it is not, goes about deleting the probe and the event that represents it. But the problem with that is, after it checks if it is in use it can be enabled, and the deletion of the event (access to the probe) will fail, as it is in use. But the uprobe will still be deleted. This is a problem as the event can reference the uprobe that was deleted. The fix is to remove the event first, and check to make sure the event removal succeeds. Then it is safe to remove the probe. When the event exists, either ftrace or perf can enable the probe and prevent the event from being removed. Link: http://lkml.kernel.org/r/20130704034038.991525256@goodmis.org Acked-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-31tracing/kprobes: Fail to unregister if probe event files are in useSteven Rostedt (Red Hat)1-6/+15
When a probe is being removed, it cleans up the event files that correspond to the probe. But there is a race between writing to one of these files and deleting the probe. This is especially true for the "enable" file. CPU 0 CPU 1 ----- ----- fd = open("enable",O_WRONLY); probes_open() release_all_trace_probes() unregister_trace_probe() if (trace_probe_is_enabled(tp)) return -EBUSY write(fd, "1", 1) __ftrace_set_clr_event() call->class->reg() (kprobe_register) enable_trace_probe(tp) __unregister_trace_probe(tp); list_del(&tp->list) unregister_probe_event(tp) <-- fails! free_trace_probe(tp) write(fd, "0", 1) __ftrace_set_clr_event() call->class->unreg (kprobe_register) disable_trace_probe(tp) <-- BOOM! A test program was written that used two threads to simulate the above scenario adding a nanosleep() interval to change the timings and after several thousand runs, it was able to trigger this bug and crash: BUG: unable to handle kernel paging request at 00000005000000f9 IP: [<ffffffff810dee70>] probes_open+0x3b/0xa7 PGD 7808a067 PUD 0 Oops: 0000 [#1] PREEMPT SMP Dumping ftrace buffer: --------------------------------- Modules linked in: ipt_MASQUERADE sunrpc ip6t_REJECT nf_conntrack_ipv6 CPU: 1 PID: 2070 Comm: test-kprobe-rem Not tainted 3.11.0-rc3-test+ #47 Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./To be filled by O.E.M., BIOS SDBLI944.86P 05/08/2007 task: ffff880077756440 ti: ffff880076e52000 task.ti: ffff880076e52000 RIP: 0010:[<ffffffff810dee70>] [<ffffffff810dee70>] probes_open+0x3b/0xa7 RSP: 0018:ffff880076e53c38 EFLAGS: 00010203 RAX: 0000000500000001 RBX: ffff88007844f440 RCX: 0000000000000003 RDX: 0000000000000003 RSI: 0000000000000003 RDI: ffff880076e52000 RBP: ffff880076e53c58 R08: ffff880076e53bd8 R09: 0000000000000000 R10: ffff880077756440 R11: 0000000000000006 R12: ffffffff810dee35 R13: ffff880079250418 R14: 0000000000000000 R15: ffff88007844f450 FS: 00007f87a276f700(0000) GS:ffff88007d480000(0000) knlGS:0000000000000000 CS: 0010 DS: 0000 ES: 0000 CR0: 000000008005003b CR2: 00000005000000f9 CR3: 0000000077262000 CR4: 00000000000007e0 Stack: ffff880076e53c58 ffffffff81219ea0 ffff88007844f440 ffffffff810dee35 ffff880076e53ca8 ffffffff81130f78 ffff8800772986c0 ffff8800796f93a0 ffffffff81d1b5d8 ffff880076e53e04 0000000000000000 ffff88007844f440 Call Trace: [<ffffffff81219ea0>] ? security_file_open+0x2c/0x30 [<ffffffff810dee35>] ? unregister_trace_probe+0x4b/0x4b [<ffffffff81130f78>] do_dentry_open+0x162/0x226 [<ffffffff81131186>] finish_open+0x46/0x54 [<ffffffff8113f30b>] do_last+0x7f6/0x996 [<ffffffff8113cc6f>] ? inode_permission+0x42/0x44 [<ffffffff8113f6dd>] path_openat+0x232/0x496 [<ffffffff8113fc30>] do_filp_open+0x3a/0x8a [<ffffffff8114ab32>] ? __alloc_fd+0x168/0x17a [<ffffffff81131f4e>] do_sys_open+0x70/0x102 [<ffffffff8108f06e>] ? trace_hardirqs_on_caller+0x160/0x197 [<ffffffff81131ffe>] SyS_open+0x1e/0x20 [<ffffffff81522742>] system_call_fastpath+0x16/0x1b Code: e5 41 54 53 48 89 f3 48 83 ec 10 48 23 56 78 48 39 c2 75 6c 31 f6 48 c7 RIP [<ffffffff810dee70>] probes_open+0x3b/0xa7 RSP <ffff880076e53c38> CR2: 00000005000000f9 ---[ end trace 35f17d68fc569897 ]--- The unregister_trace_probe() must be done first, and if it fails it must fail the removal of the kprobe. Several changes have already been made by Oleg Nesterov and Masami Hiramatsu to allow moving the unregister_probe_event() before the removal of the probe and exit the function if it fails. This prevents the tp structure from being used after it is freed. Link: http://lkml.kernel.org/r/20130704034038.819592356@goodmis.org Acked-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-31tracing: Add comment to describe special break case in probe_remove_event_call()Steven Rostedt (Red Hat)1-0/+6
The "break" used in the do_for_each_event_file() is used as an optimization as the loop is really a double loop. The loop searches all event files for each trace_array. There's only one matching event file per trace_array and after we find the event file for the trace_array, the break is used to jump to the next trace_array and start the search there. As this is not a standard way of using "break" in C code, it requires a comment right before the break to let people know what is going on. Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-31tracing: trace_remove_event_call() should fail if call/file is in useOleg Nesterov1-2/+33
Change trace_remove_event_call(call) to return the error if this call is active. This is what the callers assume but can't verify outside of the tracing locks. Both trace_kprobe.c/trace_uprobe.c need the additional changes, unregister_trace_probe() should abort if trace_remove_event_call() fails. The caller is going to free this call/file so we must ensure that nobody can use them after trace_remove_event_call() succeeds. debugfs should be fine after the previous changes and event_remove() does TRACE_REG_UNREGISTER, but still there are 2 reasons why we need the additional checks: - There could be a perf_event(s) attached to this tp_event, so the patch checks ->perf_refcount. - TRACE_REG_UNREGISTER can be suppressed by FTRACE_EVENT_FL_SOFT_MODE, so we simply check FTRACE_EVENT_FL_ENABLED protected by event_mutex. Link: http://lkml.kernel.org/r/20130729175033.GB26284@redhat.com Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-30ftrace: Check module functions being traced on reloadSteven Rostedt (Red Hat)1-9/+62
There's been a nasty bug that would show up and not give much info. The bug displayed the following warning: WARNING: at kernel/trace/ftrace.c:1529 __ftrace_hash_rec_update+0x1e3/0x230() Pid: 20903, comm: bash Tainted: G O 3.6.11+ #38405.trunk Call Trace: [<ffffffff8103e5ff>] warn_slowpath_common+0x7f/0xc0 [<ffffffff8103e65a>] warn_slowpath_null+0x1a/0x20 [<ffffffff810c2ee3>] __ftrace_hash_rec_update+0x1e3/0x230 [<ffffffff810c4f28>] ftrace_hash_move+0x28/0x1d0 [<ffffffff811401cc>] ? kfree+0x2c/0x110 [<ffffffff810c68ee>] ftrace_regex_release+0x8e/0x150 [<ffffffff81149f1e>] __fput+0xae/0x220 [<ffffffff8114a09e>] ____fput+0xe/0x10 [<ffffffff8105fa22>] task_work_run+0x72/0x90 [<ffffffff810028ec>] do_notify_resume+0x6c/0xc0 [<ffffffff8126596e>] ? trace_hardirqs_on_thunk+0x3a/0x3c [<ffffffff815c0f88>] int_signal+0x12/0x17 ---[ end trace 793179526ee09b2c ]--- It was finally narrowed down to unloading a module that was being traced. It was actually more than that. When functions are being traced, there's a table of all functions that have a ref count of the number of active tracers attached to that function. When a function trace callback is registered to a function, the function's record ref count is incremented. When it is unregistered, the function's record ref count is decremented. If an inconsistency is detected (ref count goes below zero) the above warning is shown and the function tracing is permanently disabled until reboot. The ftrace callback ops holds a hash of functions that it filters on (and/or filters off). If the hash is empty, the default means to filter all functions (for the filter_hash) or to disable no functions (for the notrace_hash). When a module is unloaded, it frees the function records that represent the module functions. These records exist on their own pages, that is function records for one module will not exist on the same page as function records for other modules or even the core kernel. Now when a module unloads, the records that represents its functions are freed. When the module is loaded again, the records are recreated with a default ref count of zero (unless there's a callback that traces all functions, then they will also be traced, and the ref count will be incremented). The problem is that if an ftrace callback hash includes functions of the module being unloaded, those hash entries will not be removed. If the module is reloaded in the same location, the hash entries still point to the functions of the module but the module's ref counts do not reflect that. With the help of Steve and Joern, we found a reproducer: Using uinput module and uinput_release function. cd /sys/kernel/debug/tracing modprobe uinput echo uinput_release > set_ftrace_filter echo function > current_tracer rmmod uinput modprobe uinput # check /proc/modules to see if loaded in same addr, otherwise try again echo nop > current_tracer [BOOM] The above loads the uinput module, which creates a table of functions that can be traced within the module. We add uinput_release to the filter_hash to trace just that function. Enable function tracincg, which increments the ref count of the record associated to uinput_release. Remove uinput, which frees the records including the one that represents uinput_release. Load the uinput module again (and make sure it's at the same address). This recreates the function records all with a ref count of zero, including uinput_release. Disable function tracing, which will decrement the ref count for uinput_release which is now zero because of the module removal and reload, and we have a mismatch (below zero ref count). The solution is to check all currently tracing ftrace callbacks to see if any are tracing any of the module's functions when a module is loaded (it already does that with callbacks that trace all functions). If a callback happens to have a module function being traced, it increments that records ref count and starts tracing that function. There may be a strange side effect with this, where tracing module functions on unload and then reloading a new module may have that new module's functions being traced. This may be something that confuses the user, but it's not a big deal. Another approach is to disable all callback hashes on module unload, but this leaves some ftrace callbacks that may not be registered, but can still have hashes tracing the module's function where ftrace doesn't know about it. That situation can cause the same bug. This solution solves that case too. Another benefit of this solution, is it is possible to trace a module's function on unload and load. Link: http://lkml.kernel.org/r/20130705142629.GA325@redhat.com Reported-by: Jörn Engel <joern@logfs.org> Reported-by: Dave Jones <davej@redhat.com> Reported-by: Steve Hodgson <steve@purestorage.com> Tested-by: Steve Hodgson <steve@purestorage.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-29ftrace: Consolidate some duplicate code for updating ftrace opsSteven Rostedt (Red Hat)1-6/+10
When ftrace ops modifies the functions that it will trace, the update to the function mcount callers may need to be modified. Consolidate the two places that do the checks to see if an update is required with a wrapper function for those checks. Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-29tracing: Change remove_event_file_dir() to clear "d_subdirs"->i_privateOleg Nesterov1-32/+15
Change remove_event_file_dir() to clear ->i_private for every file we are going to remove. We need to check file->dir != NULL because event_create_dir() can fail. debugfs_remove_recursive(NULL) is fine but the patch moves it under the same check anyway for readability. spin_lock(d_lock) and "d_inode != NULL" check are not needed afaics, but I do not understand this code enough. tracing_open_generic_file() and tracing_release_generic_file() can go away, ftrace_enable_fops and ftrace_event_filter_fops() use tracing_open_generic() but only to check tracing_disabled. This fixes all races with event_remove() or instance_delete(). f_op->read/write/whatever can never use the freed file/call, all event/* files were changed to check and use ->i_private under event_mutex. Note: this doesn't not fix other problems, event_remove() can destroy the active ftrace_event_call, we need more changes but those changes are completely orthogonal. Link: http://lkml.kernel.org/r/20130728183527.GB16723@redhat.com Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-29tracing: Introduce remove_event_file_dir()Oleg Nesterov1-24/+23
Preparation for the next patch. Extract the common code from remove_event_from_tracers() and __trace_remove_event_dirs() into the new helper, remove_event_file_dir(). The patch looks more complicated than it actually is, it also moves remove_subsystem() up to avoid the forward declaration. Link: http://lkml.kernel.org/r/20130726172547.GA3629@redhat.com Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-29tracing: Change f_start() to take event_mutex and verify i_private != NULLOleg Nesterov1-4/+9
trace_format_open() and trace_format_seq_ops are racy, nothing protects ftrace_event_call from trace_remove_event_call(). Change f_start() to take event_mutex and verify i_private != NULL, change f_stop() to drop this lock. This fixes nothing, but now we can change debugfs_remove("format") callers to nullify ->i_private and fix the the problem. Note: the usage of event_mutex is sub-optimal but simple, we can change this later. Link: http://lkml.kernel.org/r/20130726172543.GA3622@redhat.com Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-29tracing: Change event_filter_read/write to verify i_private != NULLOleg Nesterov2-18/+25
event_filter_read/write() are racy, ftrace_event_call can be already freed by trace_remove_event_call() callers. 1. Shift mutex_lock(event_mutex) from print/apply_event_filter to the callers. 2. Change the callers, event_filter_read() and event_filter_write() to read i_private under this mutex and abort if it is NULL. This fixes nothing, but now we can change debugfs_remove("filter") callers to nullify ->i_private and fix the the problem. Link: http://lkml.kernel.org/r/20130726172540.GA3619@redhat.com Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-29tracing: Change event_enable/disable_read() to verify i_private != NULLOleg Nesterov1-10/+20
tracing_open_generic_file() is racy, ftrace_event_file can be already freed by rmdir or trace_remove_event_call(). Change event_enable_read() and event_disable_read() to read and verify "file = i_private" under event_mutex. This fixes nothing, but now we can change debugfs_remove("enable") callers to nullify ->i_private and fix the the problem. Link: http://lkml.kernel.org/r/20130726172536.GA3612@redhat.com Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-29tracing: Turn event/id->i_private into call->event.typeOleg Nesterov1-5/+13
event_id_read() is racy, ftrace_event_call can be already freed by trace_remove_event_call() callers. Change event_create_dir() to pass "data = call->event.type", this is all event_id_read() needs. ftrace_event_id_fops no longer needs tracing_open_generic(). We add the new helper, event_file_data(), to read ->i_private, it will have more users. Note: currently ACCESS_ONCE() and "id != 0" check are not needed, but we are going to change event_remove/rmdir to clear ->i_private. Link: http://lkml.kernel.org/r/20130726172532.GA3605@redhat.com Reviewed-by: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com> Signed-off-by: Oleg Nesterov <oleg@redhat.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-26tracing: Add __tracepoint_string() to export string pointersSteven Rostedt (Red Hat)2-0/+22
There are several tracepoints (mostly in RCU), that reference a string pointer and uses the print format of "%s" to display the string that exists in the kernel, instead of copying the actual string to the ring buffer (saves time and ring buffer space). But this has an issue with userspace tools that read the binary buffers that has the address of the string but has no access to what the string itself is. The end result is just output that looks like: rcu_dyntick: ffffffff818adeaa 1 0 rcu_dyntick: ffffffff818adeb5 0 140000000000000 rcu_dyntick: ffffffff818adeb5 0 140000000000000 rcu_utilization: ffffffff8184333b rcu_utilization: ffffffff8184333b The above is pretty useless when read by the userspace tools. Ideally we would want something that looks like this: rcu_dyntick: Start 1 0 rcu_dyntick: End 0 140000000000000 rcu_dyntick: Start 140000000000000 0 rcu_callback: rcu_preempt rhp=0xffff880037aff710 func=put_cred_rcu 0/4 rcu_callback: rcu_preempt rhp=0xffff880078961980 func=file_free_rcu 0/5 rcu_dyntick: End 0 1 The trace_printk() which also only stores the address of the string format instead of recording the string into the buffer itself, exports the mapping of kernel addresses to format strings via the printk_format file in the debugfs tracing directory. The tracepoint strings can use this same method and output the format to the same file and the userspace tools will be able to decipher the address without any modification. The tracepoint strings need its own section to save the strings because the trace_printk section will cause the trace_printk() buffers to be allocated if anything exists within the section. trace_printk() is only used for debugging and should never exist in the kernel, we can not use the trace_printk sections. Add a new tracepoint_str section that will also be examined by the output of the printk_format file. Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-26tracing: Remove locking trace_types_lock from tracing_reset_all_online_cpus()Steven Rostedt (Red Hat)1-2/+1
Commit a82274151af "tracing: Protect ftrace_trace_arrays list in trace_events.c" added taking the trace_types_lock mutex in trace_events.c as there were several locations that needed it for protection. Unfortunately, it also encapsulated a call to tracing_reset_all_online_cpus() which also takes the trace_types_lock, causing a deadlock. This happens when a module has tracepoints and has been traced. When the module is removed, the trace events module notifier will grab the trace_types_lock, do a bunch of clean ups, and also clears the buffer by calling tracing_reset_all_online_cpus. This doesn't happen often which explains why it wasn't caught right away. Commit a82274151af was marked for stable, which means this must be sent to stable too. Link: http://lkml.kernel.org/r/51EEC646.7070306@broadcom.com Reported-by: Arend van Spril <arend@broadcom.com> Tested-by: Arend van Spriel <arend@broadcom.com> Cc: Alexander Z Lam <azl@google.com> Cc: Vaibhav Nagarnaik <vnagarnaik@google.com> Cc: David Sharp <dhsharp@google.com> Cc: stable@vger.kernel.org # 3.10 Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
2013-07-24ftrace: Add check for NULL regs if ops has SAVE_REGS setSteven Rostedt (Red Hat)1-4/+14
If a ftrace ops is registered with the SAVE_REGS flag set, and there's already a ops registered to one of its functions but without the SAVE_REGS flag, there's a small race window where the SAVE_REGS ops gets added to the list of callbacks to call for that function before the callback trampoline gets set to save the regs. The problem is, the function is not currently saving regs, which opens a small race window where the ops that is expecting regs to be passed to it, wont. This can cause a crash if the callback were to reference the regs, as the SAVE_REGS guarantees that regs will be set. To fix this, we add a check in the loop case where it checks if the ops has the SAVE_REGS flag set, and if so, it will ignore it if regs is not set. Signed-off-by: Steven Rostedt <rostedt@goodmis.org>