Kconfig: fix dependency issue when building in-tree without CONFIG_FTRACE

When building in-tree, one could disable CONFIG_FTRACE from kernel
config which will leave CONFIG_TRACEPOINTS selected by LTTNG modules,
but generate a lot of linker errors like below because it leaves out
other stuff, e.g.:

trace.c:(.text+0xd86b): undefined reference to `trace_event_buffer_reserve'
ld: trace.c:(.text+0xd8de): undefined reference to `trace_event_buffer_commit'
ld: trace.c:(.text+0xd926): undefined reference to `event_triggers_call'
ld: trace.c:(.text+0xd942): undefined reference to `trace_event_ignore_this_pid'
ld: net/mac80211/trace.o: in function `trace_event_raw_event_drv_tdls_cancel_channel_switch':

It appears to be caused by the fact that TRACE_EVENT macros in the Linux
kernel depend on the Ftrace ring buffer as soon as CONFIG_TRACEPOINTS is
enabled.

Steps to reproduce:

- Get a clone of an upstream stable kernel and use scripts/built-in.sh on it

- Configure a standard x86-64 build, enable built-in LTTNG but disable
  CONFIG_FTRACE from Kernel Hacking-->Tracers using menuconfig

- Build will fail at linking stage

Signed-off-by: Beniamin Sandu <beniaminsandu@gmail.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: mmap enum flags build failures

Some of the mmap option flags are not available on all architectures and
are defined to zero by include/linux/mman.h. This is probably done as a
way to no-op the use of these flags on configurations that don't support
them.
To fix this, only define these flags in our enumeration if they are
defined and non-zero.

Also, the MAP_HUGE_{2MB,1GB} labels were mistakingly named
MAP_HUGETLB_{2MB,1GB}.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I778a52a0da9da6e04231a52c7f68a22d122dfb83

syscalls: Make mmap()'s fields `prot` and `flags` enums

The `prot` flags is a simple CTF enumeration.

The `flags` field is a CTF struct of 2 CTF enumerations (`type` and
`options`). This is needed to express the two parts of this integer
flag. The 4 least significant bits of the integer are reserved to
express the type of the mapping (MAP_SHARED=0x1, MAP_PRIVATE=0x2, and
MAP_SHARED_VALIDATE=0x3).

The remaining 28 bits are used to specify optional configurations on the
mapping. As opposed to the type part, the options part is bit flag
field where all values are power of 2. This part can be expressed as
ORed bit flag values.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I5ae78754b5863b31d9a3ba1b1173502e1ae284d3

x86: add error code enum to pagefault tracepoints

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ia939eccd1a918958f6a281595e447f33da2d64f7

Fix: TAINT_UNSAFE_SMP renamed to TAINT_CPU_OUT_OF_SPEC in v3.15

See upstream commit:

  commit 8c90487cdc64847b4fdd812ab3047f426fec4d13
  Author: Dave Jones <davej@redhat.com>
  Date:   Wed Feb 26 10:49:49 2014 -0500

    Rename TAINT_UNSAFE_SMP to TAINT_CPU_OUT_OF_SPEC

    Rename TAINT_UNSAFE_SMP to TAINT_CPU_OUT_OF_SPEC, so we can repurpose
    the flag to encompass a wider range of pushing the CPU beyond its
    warrany.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I3e91df01bfbfaa6fab4e3904e59317022a9ec0f8

module_load: change `taints` field to `ctf_enum`

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I67b5aad0bd2bc43e06a5708f0f5e1fea56f31436

Fix: Lock metadata cache on session destroy

commit 92143b2c5656 ("Fix: metadata stream leak, missing list removal and locking")
missed taking a lock protecting the metadata stream list iteration on
session destroy. This opens a race window between iteration and item
removal/free which triggers kernel OOPS.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: metadata stream leak, missing list removal and locking

The metadata stream is part of a list of metadata streams in the
metadata cache. Its addition to the list should be protected by
the metadata cache lock. It needs to be paired with protection
of list iteration with the same lock.

Removal from the list is entirely missing, and should be added
to lttng_metadata_ring_buffer_release (with proper locking).

This missing list removal was probably not causing issues because the
metadata stream structure was leaked: a kfree() is missing from
lttng_metadata_ring_buffer_release as well.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: coherent state not changed atomically with metadata written

commit 122c63cb4310 ("Fix: Implement RING_BUFFER_GET_NEXT_SUBBUF_METADATA_CHECK")
introduces a new ioctl which returns a flag indicating whether the
metadata is in consistent state at the end of the sub-buffer.

That commit is meant to address metadata consistency issues observable
in live sessions.

However, the "consistent" state is false as soon as a producer is
active (between an outermost metadata_begin/end pair). Unfortunately,
if the last "RING_BUFFER_GET_NEXT_SUBBUF_METADATA_CHECK" operation is
done between the last metadata printf and "end" of the transaction, the
last consistency state will be false, and the consumer daemon will never
send metadata to the relay daemon. This in turn causes a live viewer to
wait for metadata endlessly.

This issue can be reproduced by running lttng-tools:
tests/regression/tools/live/test_kernel

as root in a loop.

We observe two things:
1) the poll operation blocks when there is no more metadata to send,
   which means there is no mean to unblock when the consistency state
   changes back to "true" without producing additional metadata,

2) Even if (1) was fixed, the expectation from an ABI perspective is
   that the "coherent" state is only populated when
   RING_BUFFER_GET_NEXT_SUBBUF_METADATA_CHECK succeeds. Therefore,
   there is no way to let user-space know about conherency transition
   unless additional metadata is generated.

Fixing this requires to hold the metadata cache lock across the entire
production of a coherent metadata transaction. This simpler scheme is
possible because the metadata is generated in a reallocated memory area
and not directly into a ring buffer anymore. This was not the case in
earlier lttng-modules versions, when the metadata was generated directly
into a ring buffer, which explains why this simpler scheme was not
implemented.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: include module.h for EXPORT_SYMBOL_GPL

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ic337e1eb375791ace08560555dd02b37cbefcf25

fix: __lttng_vmalloc_node_range const caller introduced in v3.6

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ib13cf03b5ab11830a8732318a12713720cf1b3e3

fix: version range for overflow_callback

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I1b8f1d59552a1723d3f4ed74780a2b57d13d0e52

fix: global_dirty_limit was introduced in v3.1

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Id97dbb2d0181a45c45cfed36c4be8753cabac283

fix: wrapper_uprobe_unregister is a void function

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ib4438da02aac3defd1245324d1b48f400f806d58

fix: prior to v4.0, __vmalloc_node_range had no vm_flags param

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ib476e32d109298d9ca3e6b6ab7ac8f63c50fb09f

fix: vmalloc on v5.8 without KALLSYMS

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ic945dad92e78a5bc2895a969a10c527e1349decf

Detect missing symbols used with kallsyms_lookup at compile time

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I19a9a31c386196899517899d861fe63611272139

Add time namespace context

Add a context for the new time namespace introduced in v5.6.

Change-Id: Ic3393f65702b80c87670bb21049ee2a19413111d
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Use exported symbol bdevname() instead of disk_name()

bdevname() is a simple wrapper over disk_name() but has the honor to be
exported. Using it removes the need for a kallsym wrapper.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ic2b2233c4db7826175c68edea69751ddcb17a5e6

Add git-review config

Add .gitreview for contributors wishing to use gerrit for patch
reviews.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I663e66a433ddb645f580c4b9f885db9c3a08e02f

fix: mm: remove vmalloc_sync_(un)mappings() (v5.8)

See upstream commit:

  commit 73f693c3a705756032c2863bfb37570276902d7d
  Author: Joerg Roedel <jroedel@suse.de>
  Date:   Mon Jun 1 21:52:36 2020 -0700

    mm: remove vmalloc_sync_(un)mappings()

    These functions are not needed anymore because the vmalloc and ioremap
    mappings are now synchronized when they are created or torn down.

    Remove all callers and function definitions.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ifdefa35b25b4906cde407360e608b77e47cc3808

Update design document

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add lttng-modules design document

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: callstack: initialize nested sequence length field name

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: callstack: NULL pointer dereference: length field also need fdata

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: callstack context memory corruption

commit ceabb767180e "tracepoint: Refactor representation of nested types"
introduces two context fields for callstack contexts. Keeping a pointer
to the first field is not valid when adding the second context field to
the array, because the array is reallocated.

Fix this by introducing new context APIs which operate on indexes rather
than pointers:
- lttng_append_context_index,
- lttng_get_context_field_from_index,
- lttng_remove_context_field_index.

Add a NULL check to lttng_find_context so it can be used before adding
the first context.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: mm/writeback: discard NR_UNSTABLE_NFS, use NR_WRITEBACK (v5.8)

See upstream commit:

  commit 8d92890bd6b8502d6aee4b37430ae6444ade7a8c
  Author: NeilBrown <neilb@suse.de>
  Date:   Mon Jun 1 21:48:21 2020 -0700

    mm/writeback: discard NR_UNSTABLE_NFS, use NR_WRITEBACK instead

    After an NFS page has been written it is considered "unstable" until a
    COMMIT request succeeds.  If the COMMIT fails, the page will be
    re-written.

    These "unstable" pages are currently accounted as "reclaimable", either
    in WB_RECLAIMABLE, or in NR_UNSTABLE_NFS which is included in a
    'reclaimable' count.  This might have made sense when sending the COMMIT
    required a separate action by the VFS/MM (e.g.  releasepage() used to
    send a COMMIT).  However now that all writes generated by ->writepages()
    will automatically be followed by a COMMIT (since commit 919e3bd9a875
    ("NFS: Ensure we commit after writeback is complete")) it makes more
    sense to treat them as writeback pages.

    So this patch removes NR_UNSTABLE_NFS and accounts unstable pages in
    NR_WRITEBACK and WB_WRITEBACK.

    A particular effect of this change is that when
    wb_check_background_flush() calls wb_over_bg_threshold(), the latter
    will report 'true' a lot less often as the 'unstable' pages are no
    longer considered 'dirty' (as there is nothing that writeback can do
    about them anyway).

    Currently wb_check_background_flush() will trigger writeback to NFS even
    when there are relatively few dirty pages (if there are lots of unstable
    pages), this can result in small writes going to the server (10s of
    Kilobytes rather than a Megabyte) which hurts throughput.  With this
    patch, there are fewer writes which are each larger on average.

    Where the NR_UNSTABLE_NFS count was included in statistics
    virtual-files, the entry is retained, but the value is hard-coded as
    zero.  static trace points and warning printks which mentioned this
    counter no longer report it.

Change-Id: I18080ca62bc6c1cd7d6da4cb27cc1521fbdca5e1
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: block: remove the error argument to the block_bio_complete (v5.8)

See upstream commit:

  commit d24de76af836260a99ca2ba281a937bd5bc55591
  Author: Christoph Hellwig <hch@lst.de>
  Date:   Wed Jun 3 07:14:43 2020 +0200

    block: remove the error argument to the block_bio_complete tracepoint

    The status can be trivially derived from the bio itself.  That also avoid
    callers like NVMe to incorrectly pass a blk_status_t instead of the errno,
    and the overhead of translating the blk_status_t to the errno in the I/O
    completion fast path when no tracing is enabled.

Fixes: 35fe0d12c8a3 ("nvme: trace bio completion")
Change-Id: I8d1463184d79bfab418a1755bfc6a0200170fff3
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: pipe_buf_operations rework (v5.8)

See upstream commits:

  commit c928f642c29a5ffb02e16f2430b42b876dde69de
  Author: Christoph Hellwig <hch@lst.de>
  Date:   Wed May 20 17:58:16 2020 +0200

    fs: rename pipe_buf ->steal to ->try_steal

    And replace the arcane return value convention with a simple bool
    where true means success and false means failure.

    [AV: braino fix folded in]

  commit b8d9e7f2411b0744df2ec33e80d7698180fef21a
  Author: Christoph Hellwig <hch@lst.de>
  Date:   Wed May 20 17:58:15 2020 +0200

    fs: make the pipe_buf_operations ->confirm operation optional

    Just return 0 for success if it is not present.

  commit 76887c256744740d6121af9bc4aa787712a1f694
  Author: Christoph Hellwig <hch@lst.de>
  Date:   Wed May 20 17:58:14 2020 +0200

    fs: make the pipe_buf_operations ->steal operation optional

    Just return 1 for failure if it is not present.

Change-Id: Ic185632202470db1eb5b012e95e793ff2cb26be7
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: syscalls: Ignore fcntl cmd specific to 32-bit in 64-bit only config

When CONFIG_64BIT is defined and CONFIG_COMPAT is not defined, the fcntl system call
"F_GETLK64", "F_SETLK64" and "F_SETLKW64" should be ignored.

Signed-off-by: Ruiqiang Hao <Ruiqiang.Hao@windriver.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: Implement RING_BUFFER_GET_NEXT_SUBBUF_METADATA_CHECK

Get next metadata subbuffer, returning a flag indicating whether the
metadata is guaranteed to be in a consistent state at the end of this
sub-buffer (can be parsed).

This can be used by the consumer to know whether the metadata can be
parsed at the end of this sub-buffer, which is useful to distinguish
between errors and incomplete metadata in live tracing.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: vmalloc_sync_mappings was backported to v5.5.12

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ie554d9c956afc2a8e114fe41e4b3c225d8af40a1

Update: Additional kernel ranges for vmalloc_sync_mappings

Some Ubuntu kernels cannot be directly mapped to an upstream stable
version. Define distro specific ranges for those (4.15, 5.0, 5.3).

Signed-off-by: Stefan Bader <stefan.bader@canonical.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Update: Use vmalloc_sync_mappings for stable kernels

Starting from v5.4.28/v5.2.37/v4.19.113/v4.14.175/v4.9.218/v4.4.218, stable
kernel branches backported v5.6 upstream commit [1], causing the following
warnings:
...
[  483.242037] LTTng: vmalloc_sync_all symbol lookup failed.
[  483.257056] Page fault handler and NMI tracing might trigger faults.
...

Extend check for vmalloc_sync_mappings for stable kernels as well.

[1] https://github.com/torvalds/linux/commit/763802b53a427ed3cbd419dbba255c414fdd9e7c

[ Edit: minor coding style fix by Mathieu Desnoyers. ]

Signed-off-by: Ovidiu Panait <ovidiu.panait@windriver.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: Use vmalloc_sync_mappings on kernel 5.6 as well

Upstream commit [1], that got rid of vmalloc_sync_all and introduced
vmalloc_sync_mappings, is a v5.6 commit:
$ git tag --contains 763802b53a427ed3cbd419dbba255c414fdd9e7c
v5.6
v5.6-rc7
v5.7-rc1
v5.7-rc2
v5.7-rc3

Extend the LINUX_VERSION_CODE check to v5.6 to fix the following warnings:
...
[  483.242037] LTTng: vmalloc_sync_all symbol lookup failed.
[  483.257056] Page fault handler and NMI tracing might trigger faults.
...

[1] https://github.com/torvalds/linux/commit/763802b53a427ed3cbd419dbba255c414fdd9e7c

Signed-off-by: Ovidiu Panait <ovidiu.panait@windriver.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: remove unsupported `ctf_float()` macros

Tracing floats is not supported for the kernel tracer. Disallow building
kernel probes with those fields, rather than silently ignoring them.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I1cf9751df96d2af3b54f725797bd20d7b05f2b38

Cleanup: have interpreter functions return _DISCARD instead of 0

It's easier to understand the meaning of the zero return value of these
function using the enum. It makes it obvious.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I1df8f704fa9f6768f413c12c3c1de61a94b3aff8

Cleanup: bytecode: typo: "s16" -> "u16"

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I68901ca2d89d08f2cb69853816e0214c588aa7f8

Cleanup: Rename patches.i to patches.h

This generated header file contains a list of patches applied on the
lttng-modules tree. Based on the C99 specification, ".i" files are not
supposed to be preprocessed, although this header header file is
expected to be preprocessed.

Rename it from ".i" to ".h" to convey that it is a C header meant to be
preprocessed.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: Move all source files to src/

This includes *.c, lib/*/*.c, probes/*.c, wrapper/*.c.

Adapt Makefile and Kbuild accordingly. Introduce src/Kbuild.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: Move patches.i to include/generated/

Move patches.i from /extra_version to include/generated/ so we
can include them without using relative path includes.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I87927a372ffeb244f3c097c9bb80eeca7d9f99eb

Cleanup: Move lttng-modules instrumentation headers

The directory hierarchy "instrumentation/events/lttng-module/" only
exists for historical reasons and is not needed anymore. Move all
its contents into "instrumentation/events/".

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: Remove toplevel directory from include search path

Now that all include files are moved to include/ (except for those
meant to be included with "#include "...h"), we can remove the toplevel
directory from the include search path.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: Move blacklist/ headers to include/blacklist/

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: Move wrapper/ headers to include/wrapper/

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: Move instrumentation/ headers to include/instrumentation/

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: Remove deprecated TODO file

All relevant items that were left were moved to
https://bugs.lttng.org/projects/lttng-modules "Feature".

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: add missing guid_t type to wrapper

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I0de39c24a7925b580fabbdaa12dbe05c43cfcd98

Fix: missing wrapper rename to wrapper_vmalloc_sync_mappings

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Idf7082a980c5a604bfef5c69906678b5083a9bbf

Cleanup: Move headers from toplevel to include/lttng/

- Remove extra "lttng-" from filename (now implied by the path).
- Adapt includes accordingly.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: Move headers from probes/ to include/lttng/

- Remove extra "lttng-" from filename (now implied by the path).
- Adapt includes accordingly.
- Adapt lttng-syscalls-generate-headers.sh header generation script
  accordingly.
- Remove probes/lttng.h, include its PARAMS() define in the two
  user headers.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: Move headers from lib/ to include/lttng/

Adapt includes accordingly.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: Move lib/ringbuffer/ headers to include/ringbuffer/

Remove the <wrapper/ringbuffer/...> proxy include files, and add the
include/ directory to the preprocessor include search patch.

Adapt all includes accordingly.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: wrapper random documentation

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Update for kernel 5.7: use vmalloc_sync_mappings on kernels >= 5.7

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Unbreak LTTng for kernel 5.7

Linux commit 0bd476e6c67190b5eb7b6e105c8db8ff61103281 ("kallsyms:
unexport kallsyms_lookup_name() and kallsyms_on_each_symbol()") breaks
LTTng-modules by removing symbols used by the LTTng-modules out-of-tree
tracer.

I pointed this out when the change was originally considered before the
5.7 merge window. This generated some discussion but it did not lead to
any concrete proposal to fix the issue. [1]

The commit has been merged in the 5.7 merge window. At that point, as
maintainer of LTTng, I immediately raised a flag about this issue,
proposing an alternative approach to solve this: expose the few symbols
needed by LTTng to GPL modules. This was NACKed on the ground that the
Linux kernel cannot export GPL symbols when there are no in-tree
users. [2]

Steven Rostedt has shown interest in merging LTTng-modules upstream.
LTTng-modules being LGPL, this is very much doable. I have prepared a
tree of LTTng-modules "for upstreaming" and sent it to him privately so
he can review it. Even if in an ideal scenario LTTng-modules is merged
for the following merge window, it leaves LTTng-modules broken on the
5.7 kernel.

In order to ensure that the LTTng-modules kernel tracer continues working
for my end users on kernels 5.7 onwards, as a very last resort, this is
with great reluctance that I created this fix for LTTng modules. It
basically uses kprobes to lookup the kallsyms_lookup_name symbol, and
continues using kallsyms_lookup_name as before.

Link: https://lore.kernel.org/r/20200302192811.n6o5645rsib44vco@localhost
Link: https://lore.kernel.org/r/20200409193543.18115-1-mathieu.desnoyers@efficios.com
Link: https://lwn.net/Articles/817988/
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
CC: Thomas Gleixner <tglx@linutronix.de>
CC: Will Deacon <will@kernel.org>
CC: akpm@linux-foundation.org
CC: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
CC: Masami Hiramatsu <mhiramat@kernel.org>
CC: rostedt@goodmis.org
CC: Alexei Starovoitov <ast@kernel.org>

Move lttng wrappers into own module

Currently, we only pull the wrapper symbols into a single sub-module,
either:

lttng-tracer.o:
  - wrapper/random.o
  - wrapper/trace-clock.o
  - wrapper/page_alloc.o

or

lttng-statedump.o:
  - wrapper/irqdesc.o
  - wrapper/fdtable.o

Because lttng-tracer depends on lttng-statedump, we cannot just put all
wrappers into lttng-tracer.o, because it would create a circular
dependency. This will be an issue if we introduce common wrappers which
are used in both lttng-tracer.o and in lttng-statedump.o.

Introduce a new lttng-wrapper.o to contain all wrapper symbols for all
lttng modules.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Introduce lttng_guid_gen wrapper for kernels >= 5.7.0

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

instrumentation: update x86 kvm instrumentation for kernel >= 5.7.0

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

instrumentation: update mm_vmscan for kernel >= 5.7.0

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

filter: bytecode already in the list should go before

Background
==========
This `seqnum` (sequence number) feature is currently unused. It was
designed so that the session daemon could tell the tracer the order in
which the bytecode should be run.

Issue
=====
The current implementation of the session daemon doesn't use this
feature so there is only ever a single bytecode to execute per callsite.

During work on an upcoming feature uses this `seqnum` became useful and
it was realized that the current bytecode linking code would reverse the
order in which the bytecode were executed when all bytecodes have the
same `seqnum` value.

This is due to the fact that the `cds_list_for_each_entry_reverse` loops
until it finds a `seqnum` smaller than the new one.

So if all bytecodes have the same `seqnum`, the new bytecode will be
added at the beginning of the list.

This is not technically a problem since it's the session daemon's job to
set the sequence number if it wants a particular ordering. Even
considering that, we found it counterintuitive that new bytecodes are
added at the beginning of the list in those cases.

Solution
========
This commit makes it so that on equality, the insertion is done after
the existing bytecodes.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I784887e3e6085f9344a2bb429d4f0d30586ebc57

tracepoint: Refactor representation of nested types

Refactor enumeration, sequence, array, structure, and variant types.

Implement internal data structures to support nested types.

All probe providers using ctf_enum(), ctf_array*() and ctf_sequence*()
are switched to this new internal type representation.

Each of sequence, array, struct and variant gain a "alignment" property,
which is a feature which was needed in lttng-modules to express
alignment for an array or sequence of bits.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

wrapper/compiler.h: Implement __LTTNG_COMPOUND_LITERAL

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Update to SPDX v3.0 identifiers

The short form of GPL-2.0 and LGPL-2.1 were deprecated in favour of the
clearer GPL-2.0-only and GPL-2.0-or-later in the SPDX license list v3.0.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: If2337f5c67a2548d7f25043e67006211213cbe3e

fix: uaccess wrapper for CentOS >= 4.18.0-147

Fixes: #1253
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I2a79c1c0e897a6148e60e5599949cd2778d09d50

fix: ext4 instrumentation for CentOS >= 4.18.0-147

Fixes: #1253
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I1fd54af16fbb02cd4b3ab7fc7d9232708088f1fd

fix: signal instrumentation for CentOS >= 4.18.0-147

Fixes: #1253
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I280013402df6f14222fbb912cdf64d80af3ab265

fix: kvm instrumentation for CentOS >= 4.18.0-147

Fixes: #1253
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ide20ebf51bec503866ffc96dda3e0b09ebeb14d6

fix: rcu instrumentation for CentOS >= 4.18.0-80

Fixes: #1253
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I1735d2caa7215ce94272aaaa98cbbc8f3a10743d

Fix: update kvm instrumentation for Ubuntu 5.3.0-45

This commit introduced in 5.3.0-43 was dropped in 5.3.0-45 and reintroduced
in 5.3.0-46:

  commit 795f8a34f279e17c279bba46da10f15c5dd00264
  Author: Sean Christopherson <sean.j.christopherson@intel.com>
  Date:   Fri Dec 6 15:57:14 2019 -0800

    KVM: x86: Use gpa_t for cr2/gpa to fix TDP support on 32-bit KVM

BugLink: https://bugs.launchpad.net/bugs/1867051
    [ Upstream commit 736c291c9f36b07f8889c61764c28edce20e715d ]

Fun times!

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ia5f1a4ba355f592f09e964038b6334ddb3ad5153

Fix: update kvm instrumentation for Ubuntu 5.3.0-43

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I1ce5f9ebba997fcc4cfbae6901eed479e2e1a79e

Fix: fcntl enum: only emit F_GETOWNER_UIDS for kernels >= 3.6

It appears in upstream Linux kernel
commit 1d151c337d79 "c/r: fcntl: add F_GETOWNER_UIDS option"
which was first released in v3.6.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

syscalls: Make the cmd field of fcntl an enum

The fcntl system call has a cmd parameter which is one of multiple
values defined in the linux/fcntl.h file. This field is now an
enumeration of the various possible values.

Here's an example babeltrace output of the fcntl system call:

[...] syscall_entry_fcntl: { cpu_id = 3 }, { fd = 35,
cmd = ( "F_GETFL" : container = 3 ), arg = 140388746332169 }

Change-Id: I06f511d5dba986aaf01b50057cbe6eac789bbea7
Signed-off-by: Geneviève Bastien <gbastien+lttng@versatic.net>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

syscalls: Make the flags and mode fields of open[at] enumerations

The open and openat system call have a flags and mode fields, whose values
are defined in the linux/fcntl.h file. These fields are now
enumerations that can be read as a bit field enum, to make the values more
readable / meaningful.

Here's an example babeltrace output of the open system call:

[...] syscall_entry_openat: { cpu_id = 0 }, { dfd = -100,
filename = "/tmp/edg0_383407",
flags = ( "O_RDWR" | "O_CREAT" | "O_TRUNC" : container = 578 ),
mode = ( "S_IWOTH" | "S_IROTH" | "S_IWGRP" | "S_IRGRP" |
   "S_IWUSR" | "S_IRUSR" : container = 438 ) }

Change-Id: Id7a516670b03e52fc75f9ff3c6ba8114c61a3865
Signed-off-by: Geneviève Bastien <gbastien@versatic.net>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: update kvm instrumentation for Ubuntu 4.15.0-92

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ib367b9a0ce3846f45313906e710a9a6d644e3955

Record event as soon as one filter evaluates to TRUE

This is a tiny optimization but it can reduce tracing cost in cases
where multiple filters are attached to a probe.

Since we want to trace an event if any of its filter expressions
evaluates to TRUE, there is no need to keep on evaluating expressions
once one is found to be TRUE.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I6b5b7697869d3e25dc24d38dc39c34ae1d49dcc9

Add UDP and ICMP packet header information to the tracepoint

* UDP transport header
* ICMP transport header

Signed-off-by: Florian Walbroel <walbroel@silexica.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

statedump: introduce file_table_address

Currently the LTTng-modules statedump simply iterates over all processes
in the system and assumes all threads share the same file descriptor
table, which is only true if threads were created with clone
CLONE_FILES.

Directly invoking clone without the CLONE_FILES creates threads which
belong to the same process, but have their own file descriptor table.

Therefore, model-wise, we cannot assume that all threads in a process
have the same fd table content.

Add a new "file_table_address" field to the lttng_statedump_process_state
event, which dumps the address of the thread's struct files_struct
pointer. This pointer is guaranteed to never be re-used while we hold
the RCU read-side lock (so for the entire iteration over
processes/threads).

For the lttng_statedump_file_descriptor event, remove the "pid" field
(which is semantically inaccurate) and add a "file_table_address" field,
which contains the struct files_struct address of the file table
containing the file descriptor.

An optimization is performed to eliminate most duplcated file table
content by skipping file table dump if the same file table address is
encountered consecutively while iterating over a process' threads.

This introduces a semantic change to the statedump fields, and will
therefore be introduced in lttng-modules 2.12 onwards, not backported as
a fix.

Fixes: #1245
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Remove kernel version from syscall headers name

This will reduce the delta when updating the syscall headers and make it
easier to track the changes.

Add a comment in the headers to track the version they were generated
from.

Change-Id: I939e1ebd8ab18d880ff1394e771982b5491d3b19
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Add script to automate the syscall extraction process

Change-Id: Ic8e1e62058616f221569e1f5ed2ffb9d54275af5
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Update lttng-syscalls-extractor for v5.6.0

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I48e4d82484d6edcc7601a40d2f94e5c46ad5184b

Add a Makefile for the lttng-syscalls-extractor module

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I0c1cc650d8594ef01b4039bb3d2af94bfddea54d

Cleanup: Syscall headers scripts shellcheck warnings

Change-Id: Ia616724d942f43f5fde2186bff28446120b17c43
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Remove lttng-ftrace integration

The lttng-ftrace integration (LTTNG_KERNEL_FUNCTION instrumentation
type) was unused for a while now. The "function" probing is actually
done with kprobes and kretprobes (LTTNG_KERNEL_KPROBE and
LTTNG_KERNEL_KRETPROBE).

Remove it so a use of kallsyms_lookup_name() can be removed as well.
Note that in the future we could add back this support by using
register_ftrace_function() which is exported to kernel modules, but
considering that we have not been using this code for a while,
just remove the implementation for now.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Remove dependency on kallsyms for splice_to_pipe (kernel 4.2+)

Upstream commit 2b514574f7e88 "net: af_unix: implement splice for stream
af_unix sockets" exported the "splice_to_pipe" symbol, so use it to
remove a dependency on kallsyms_lookup_name().

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Remove dependency on kallsyms for irq_to_desc (kernel 3.4+)

Upstream commit 3911ff30f5d "genirq: export handle_edge_irq() and
irq_to_desc()" exported the irq_to_desc symbol, so use it to remove a
dependency on kallsyms_lookup_name().

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Remove work-around for signed tracepoint module tainting (kernel 3.15+)

Upstream commit 66cc69e34e86a "Fix: module signature vs tracepoints: add
new TAINT_UNSIGNED_MODULE" fixed an issue where the kernel was
considering unsigned modules as tainting the kernel in the same way as a
force-loaded modules, which was causing the tracepoints within those
modules to be hidden.

This fix was merged in kernel 3.15, so there is no use in applying this
work-around starting from that kernel.

This removes a dependency on kallsyms_lookup_name() for the symbol
"tracepoint_module_notify".

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Change integer base to hex for fields representing addresses

It seems more expected to show addresses in hexadecimal.

I applied this to all fields named "ip" or "addr".

I haven't change the `addr` fields in the i2c.h file as I am not aware
of what is the convention when displaying I2C addresses.

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I6ed00b31e319e1d966ab138471248dcf3d2c7315

Fix: rcu: Fix data-race due to atomic_t copy-by-value (5.5.6, 5.4.22)

The following upstream commit has been backported to stable kernels
5.5.6 and 5.4.22:

  commit 6cf539a87a61a4fbc43f625267dbcbcf283872ed
  Author: Marco Elver <elver@google.com>
  Date:   Wed Oct 9 17:57:43 2019 +0200

    rcu: Fix data-race due to atomic_t copy-by-value

    This fixes a data-race where `atomic_t dynticks` is copied by value. The
    copy is performed non-atomically, resulting in a data-race if `dynticks`
    is updated concurrently.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: y2038: hide timeval/timespec/itimerval/itimerspec types (v5.6)

See upstream commit:

  commit c766d1472c70d25ad475cf56042af1652e792b23
  Author: Arnd Bergmann <arnd@arndb.de>
  Date:   Thu Feb 20 20:03:57 2020 -0800

    y2038: hide timeval/timespec/itimerval/itimerspec types

    There are no in-kernel users remaining, but there may still be users that
    include linux/time.h instead of sys/time.h from user space, so leave the
    types available to user space while hiding them from kernel space.

    Only the __kernel_old_* versions of these types remain now.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I986a813ad8b1c753ab1fa07f726b0cc481f049cb

fix: use timespec64 on kernels that have it

This fixes v5.6 which has hidden 'struct timespec' from kernel code and
makes 32bit archs y2038 compliant on v3.17 and newer.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I1daedc4a93cbbebb8f5a1d99c4619cb27f6a9e07

fix: move lttng_close_on_exec to proper wrapper

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I330c6d63fc9c1af00c9a9471563d322d036b9a9c

fix: 'struct timex' removed upstream (v5.6)

The 'timex' struct was remove in v5.6 and replaced by 2 variants, one
that is y2038 compliant and a compat version for 32bit archs.

Add this temporary fix while we update our syscalls tracepoint headers,
the type of this struct has limited importance since it's only used to
record the adress in the trace.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Change-Id: I085b22f282db57985f1c3d341e7c0866cb20e3c9

Fix: statedump: consistently check task_cred_xxx() return value for NULL

trace_lttng_statedump_process_user_ns() internally checks whether
user_ns is NULL. While this does not appear to be a possible return
value for task_cred_xxx(), err on the safe side and check for NULL here
as well to be consistent with the paranoid behavior of
trace_lttng_statedump_process_user_ns().

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: statedump: check task_active_pid_ns return value for NULL

The lttng-statedump checks the return value of task_active_pid_ns()
before each use within lttng_statedump_process_pid_ns(), but misses
the NULL check before dereferencing pid_ns->parent.

This race happens if a task exists in "dead" state while the statedump
iterates on that task.

Reported-by: Li Zhou <li.zhou@windriver.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: lttng-events.c: variable may be used uninitialized

Fixes the following warning:
/home/frdeso/projets/lttng/modules/lttng-events.c: In function ‘print_metadata_escaped_field’:
/home/frdeso/projets/lttng/modules/lttng-events.c:2563:5: error: ‘ret’ may be used uninitialized in this function [-Werror=maybe-uninitialized]
     if (ret)
           ^

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: If2db5e1ad9782fb49d6d07026976d3d22f89f2ab

Cleanup: remove trailing white spaces across project

Signed-off-by: Francis Deslauriers <francis.deslauriers@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I2de34e4913f9cefbd72a138f2d3dfb9d67ece311

sched: Make the sched_switch task state an enum

This gives meaning to the task state value. Only the bit masks are
enumerated, as defined compositions are not exhaustive listing of all
possible values and there would be a lot of unknown. Interpretation of
combination of bit flags is left to the consumer of the event.

Change-Id: I83c5fbee9cba2701c7238c0ac6abd4c8a351b193
Signed-off-by: Geneviève Bastien <gbastien@versatic.net>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

block: Make the rwbs field as a bit field enum

The rwbs value of block request events is in fact a series of bit fields
set to 0 or 1. The enumeration values are all powers of 2, trace readers
could interpret this as a bit field enum and show the result as a
concatenation of the corresponding flags. For example, with a matching
patch for babeltrace2's pretty sink, the output for a rwbs value of 12
would be:

[13:15:49.024354958] (+0.000003868) wilbrod block_rq_complete: { cpu_id = 4 },
    { dev = 8388624, sector = 375490176, nr_sector = 360, error = 0,
      rwbs = ( "RWBS_FLAG_READ" | "RWBS_FLAG_RAHEAD" : container = 12 ) }

Consumers who have no notion of the bit field enum can still use the
integer value of the field as before.

Change-Id: I2e873f16e5a0507e52cfc85c70b471a9f4d5d5d1
Signed-off-by: Geneviève Bastien <gbastien@versatic.net>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: workqueue: add worker function to workqueue_execute_end tracepoint (v5.6)

See upstream commit :

  commit 1c5da0ec7f20dfb56030fb93f7f52f48e12deb52
  Author: Daniel Jordan <daniel.m.jordan@oracle.com>
  Date:   Mon Jan 13 17:52:39 2020 -0500

    workqueue: add worker function to workqueue_execute_end tracepoint

    It's surprising that workqueue_execute_end includes only the work when
    its counterpart workqueue_execute_start has both the work and the worker
    function.

    You can't set a tracing filter or trigger based on the function, and
    postprocessing scripts interested in specific functions are harder to
    write since they have to remember the work from _start and match it up
    with the same field in _end.

    Add the function name, taking care to use the copy stashed in the
    worker since the work is no longer safe to touch.

Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>