Once upon a time (and, indeed, to this very day if you're on LP32) the
soinfo struct used a fixed-length buffer for the soname. This caused
some issues, mainly with app developers who accidentally included a full
Windows "C:\My Computer\...\libfoo.so" style path. To avoid all this we
switched to just pointing into the ELF file itself, where the DT_SONAME
is already stored as a NUL-terminated string. And all was well for many
years.
Now though, we've seen a bunch of slow startup traces from dogfood where
`dlopen("libnativebridge.so")` in a cold start takes 125-200ms on a recent
device, despite no IO contention. Even though libnativebridge.so is only
20KiB.
Measurement showed that every library whose soname we check required
pulling in a whole page just for the (usually) very short string. Worse,
there's readahead. In one trace we saw 18 pages of libhwui.so pulled
in just for `"libhwui.so\0"`. In fact, there were 3306 pages (~13MiB)
added to the page cache during `dlopen("libnativebridge.so")`. 13MiB for
a 20KiB shared library!
This is the obvious change to use a std::string to copy the sonames
instead. This will dirty slightly more memory, but massively improve
locality.
Testing with the same pathological setup took `dlopen("libnativebridge.so")`
down from 192ms to 819us.
Bug: http://b/177102905
Test: tested with a pathologically modified kernel
Change-Id: I33837f4706adc25f93c6fa6013e8ba970911dfb9
This replaces a single 2Gb readable memory region with a bunch of tiny
regions, and leaves the bulk of 2Gb mapped but unaccessible. This makes
it harder to defeat ASLR by probing for the CFI shadow region.
Sample CFI shadow mapping with this change:
7165151000-716541f000 ---p 00000000 00:00 0 [anon:cfi shadow]
716541f000-7165420000 r--p 00000000 00:00 0 [anon:cfi shadow]
7165420000-71654db000 ---p 00000000 00:00 0 [anon:cfi shadow]
71654db000-71654dc000 r--p 00000000 00:00 0 [anon:cfi shadow]
71654dc000-71654dd000 r--p 00000000 00:00 0 [anon:cfi shadow]
71654dd000-71654f0000 ---p 00000000 00:00 0 [anon:cfi shadow]
71654f0000-71654f1000 r--p 00000000 00:00 0 [anon:cfi shadow]
71654f1000-71e5151000 ---p 00000000 00:00 0 [anon:cfi shadow]
This change degrades CFI diagnostics for wild jumps and casts (i.e. when
the target of a CFI check is outside of any known library bounds). This
is acceptable, because CFI does not have much to tell about those cases
anyway. Such bugs will show up as SEGV_ACCERR crashes inside
__cfi_slowpath in libdl.so from now on.
Bug: 158113540
Test: bionic-unit-tests/cfi_test.*
Test: adb shell cat /proc/$PID/maps | grep cfi
Change-Id: I57cbd0d3f87eb1610ad99b48d98ffd497ba214b4
Merged-In: I57cbd0d3f87eb1610ad99b48d98ffd497ba214b4
Validate the list of defined versions explicitly, during library
prelinking, rather than implicitly as part of constructing the
VersionTracker in soinfo::link_image.
Doing the validation upfront allows removing the symbol lookup failure
code paths, which only happen on a library with invalid version
information.
Helps on the walleye 64-bit linker relocation benchmark (146.2ms ->
131.6ms)
Bug: none
Test: bionic unit tests
Change-Id: Id17508aba3af2863909f0526897c4277419322b7
We've copied & pasted these to too many places. And if we're going to
have another go at upstreaming these, that's probably yet another reason
to have the *values* in just one place. (Even if upstream wants different
names, we'll likely keep the legacy names around for a while for source
compatibility.)
Bug: http://b/111903542
Test: ran tests
Change-Id: I8ccc557453d69530e5b74f865cbe0b458c84e3ba
Historically we had part of the linker licensed under BSD and
another part under Apache 2 license. This commit makes all the
linker code licensed under BSD license.
Test: m
Change-Id: I11b8163ae75966b5768d3fe992679de376106515
Fallback to unchecked if the shadow offset overflows int16_t.
This may happen when a library's data segment is larger than 256MB.
Also updated some comments.
Bug: 22033465
Test: bionic device tests
Change-Id: I8eef42f75099f24aed566499ff1731a0bbf01ff3
