commit bd3e8ecca937a76318c68ec8e8da72cb5c0239ba
author Primiano Tucci <primiano@google.com> Mon Jun 29 11:28:58 2020 +0100
committer Primiano Tucci <primiano@google.com> Mon Jun 29 11:28:58 2020 +0100
tree 0ab0599b64a0dcdfd157327a60f00b4e21087fc7
parent 8c03b409fdb2f717b3de733b8468457f65c9b027

ftrace: fix pointer stability of events_ vector

This fixes an extremely subtle bug which causes traced_probes to
crash when tracing on some specific devices.
The root cause of the bug is the fact that in rare cirumstances
the events_ vector can be resized after construction, after some
Event* poitners have been handed out, hence invalidating them.
The rare circumstance is:
- The trace config ends up hitting a "generic event" path, i.e.
  the ftrace event is not known at compile time but exists on the
  target device (note this is not that unlikely because the expansion
  of some atrace categories in ftrace_config_muxer.cc can end up
  adding events like sde/sde_cmd_release_bw which are not in the
  compile-time list.
- The ftrace event ID of the generic event above is > than the
  max(ID of events known at compile time)
- The vector resize() ends up hitting the unhappy realloc() path
  which changes the data() pointer and mem-copies everything,
  invalidating all pointers.

Fix: use std::deque, which guarantees pointer stability.

Bug: 159998308
Change-Id: Id21fabf0c940fd0e5302518c0170686d61aaf847