fix: mm/page_alloc: fix tracepoint mm_page_alloc_zone_locked() (v5.15.171)

See upstream backported commit:

  commit 28e7a507196fefd119e7ca2286840f1a9aad5e8a
  Author: Wonhyuk Yang <vvghjk1234@gmail.com>
  Date:   Thu May 19 14:08:54 2022 -0700

    mm/page_alloc: fix tracepoint mm_page_alloc_zone_locked()

    [ Upstream commit 10e0f7530205799e7e971aba699a7cb3a47456de ]

    Currently, trace point mm_page_alloc_zone_locked() doesn't show correct
    information.

    First, when alloc_flag has ALLOC_HARDER/ALLOC_CMA, page can be allocated
    from MIGRATE_HIGHATOMIC/MIGRATE_CMA.  Nevertheless, tracepoint use
    requested migration type not MIGRATE_HIGHATOMIC and MIGRATE_CMA.

    Second, after commit 44042b4498728 ("mm/page_alloc: allow high-order pages
    to be stored on the per-cpu lists") percpu-list can store high order
    pages.  But trace point determine whether it is a refiil of percpu-list by
    comparing requested order and 0.

    To handle these problems, make mm_page_alloc_zone_locked() only be called
    by __rmqueue_smallest with correct migration type.  With a new argument
    called percpu_refill, it can show roughly whether it is a refill of
    percpu-list.

Link: https://lkml.kernel.org/r/20220512025307.57924-1-vvghjk1234@gmail.com
Change-Id: Ib76feb79d95e9f93c84c3aa1b946e57ac2e2666a
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: uprobes: make uprobe_register() return struct uprobe * (v6.12)

See upstream commits :

  commit 3c83a9ad0295eb63bdeb81d821b8c3b9417fbcac
  Author: Oleg Nesterov <oleg@redhat.com>
  Date:   Thu Aug 1 15:27:34 2024 +0200

    uprobes: make uprobe_register() return struct uprobe *

    This way uprobe_unregister() and uprobe_apply() can use "struct uprobe *"
    rather than inode + offset. This simplifies the code and allows to avoid
    the unnecessary find_uprobe() + put_uprobe() in these functions.

    TODO: uprobe_unregister() still needs get_uprobe/put_uprobe to ensure that
    this uprobe can't be freed before up_write(&uprobe->register_rwsem).

  commit 04b01625da130c7521b768996cd5e48052198b97
  Author: Peter Zijlstra <peterz@infradead.org>
  Date:   Tue Sep 3 10:46:00 2024 -0700

    perf/uprobe: split uprobe_unregister()

    With uprobe_unregister() having grown a synchronize_srcu(), it becomes
    fairly slow to call. Esp. since both users of this API call it in a
    loop.

    Peel off the sync_srcu() and do it once, after the loop.

    We also need to add uprobe_unregister_sync() into uprobe_register()'s
    error handling path, as we need to be careful about returning to the
    caller before we have a guarantee that partially attached consumer won't
    be called anymore. This is an unlikely slow path and this should be
    totally fine to be slow in the case of a failed attach.

  commit e04332ebc8ac128fa551e83f1161ab1c094d13a9
  Author: Oleg Nesterov <oleg@redhat.com>
  Date:   Thu Aug 1 15:27:28 2024 +0200

    uprobes: kill uprobe_register_refctr()

    It doesn't make any sense to have 2 versions of _register(). Note that
    trace_uprobe_enable(), the only user of uprobe_register(), doesn't need
    to check tu->ref_ctr_offset to decide which one should be used, it could
    safely pass ref_ctr_offset == 0 to uprobe_register_refctr().

    Add this argument to uprobe_register(), update the callers, and kill
    uprobe_register_refctr().

Change-Id: I8d1f9a5db1f19c2bc2029709ae36f82e86f6fe58
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: silence 'non-consumed' message for non-started sessions

Destroying a session with at least one enabled event and which has never
been started will currently result in an error message in the kernel log
about 'non-consumed data' for each of the per-cpu buffer. This happens
because a packet header is created in the buffer but never consumed if
the session is not started.

Add a check in the buffer cleanup code to avoid printing 'non-consumed
data' errors for buffers associated with a session taht was never
started.

Change-Id: I1358e1ae49d03544a961515b97b115a488434e27
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: writeback: Refine the show_inode_state() macro definition (v6.12)

See upstream commit :

  commit 459ca85ae1feff78d1518344df88bb79a092780c
  Author: Julian Sun <sunjunchao2870@gmail.com>
  Date:   Wed Aug 28 16:13:59 2024 +0800

    writeback: Refine the show_inode_state() macro definition

    Currently, the show_inode_state() macro only prints
    part of the state of inode->i_state. Let’s improve it
    to display more of its state.

Change-Id: Idaebd56f5775205f8a5c76e117c5ab65f7f1754b
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: scsi: sd: Atomic write support added in 6.11-rc1

See upstream commit:

    commit bf4ae8f2e6407a779c0368eb0f3e047a8333be17
    Author: John Garry <john.g.garry@oracle.com>
    Date:   Thu Jun 20 12:53:57 2024 +0000

        scsi: sd: Atomic write support

        Support is divided into two main areas:
        - reading VPD pages and setting sdev request_queue limits
        - support WRITE ATOMIC (16) command and tracing

        The relevant block limits VPD page need to be read to allow the block layer
        request_queue atomic write limits to be set. These VPD page limits are
        described in sbc4r22 section 6.6.4 - Block limits VPD page.

        There are five limits of interest:
        - MAXIMUM ATOMIC TRANSFER LENGTH
        - ATOMIC ALIGNMENT
        - ATOMIC TRANSFER LENGTH GRANULARITY
        - MAXIMUM ATOMIC TRANSFER LENGTH WITH BOUNDARY
        - MAXIMUM ATOMIC BOUNDARY SIZE

        MAXIMUM ATOMIC TRANSFER LENGTH is the maximum length for a WRITE ATOMIC
        (16) command. It will not be greater than the device MAXIMUM TRANSFER
        LENGTH.

        ATOMIC ALIGNMENT and ATOMIC TRANSFER LENGTH GRANULARITY are the minimum
        alignment and length values for an atomic write in terms of logical blocks.

        Unlike NVMe, SCSI does not specify an LBA space boundary, but does specify
        a per-IO boundary granularity. The maximum boundary size is specified in
        MAXIMUM ATOMIC BOUNDARY SIZE. When used, this boundary value is set in the
        WRITE ATOMIC (16) ATOMIC BOUNDARY field - layout for the WRITE_ATOMIC_16
        command can be found in sbc4r22 section 5.48. This boundary value is the
        granularity size at which the device may atomically write the data. A value
        of zero in WRITE ATOMIC (16) ATOMIC BOUNDARY field means that all data must
        be atomically written together.

        MAXIMUM ATOMIC TRANSFER LENGTH WITH BOUNDARY is the maximum atomic write
        length if a non-zero boundary value is set.

        For atomic write support, the WRITE ATOMIC (16) boundary is not of much
        interest, as the block layer expects each request submitted to be executed
        be atomically written together.

        MAXIMUM ATOMIC TRANSFER LENGTH WITH BOUNDARY is the maximum atomic write
        length if a non-zero boundary value is set.

        For atomic write support, the WRITE ATOMIC (16) boundary is not of much
        interest, as the block layer expects each request submitted to be executed
        atomically. However, the SCSI spec does leave itself open to a quirky
        scenario where MAXIMUM ATOMIC TRANSFER LENGTH is zero, yet MAXIMUM ATOMIC
        TRANSFER LENGTH WITH BOUNDARY and MAXIMUM ATOMIC BOUNDARY SIZE are both
        non-zero. This case will be supported.

        To set the block layer request_queue atomic write capabilities, sanitize
        the VPD page limits and set limits as follows:
        - atomic_write_unit_min is derived from granularity and alignment values.
          If no granularity value is not set, use physical block size
        - atomic_write_unit_max is derived from MAXIMUM ATOMIC TRANSFER LENGTH. In
          the scenario where MAXIMUM ATOMIC TRANSFER LENGTH is zero and boundary
          limits are non-zero, use MAXIMUM ATOMIC BOUNDARY SIZE for
          atomic_write_unit_max. New flag scsi_disk.use_atomic_write_boundary is
          set for this scenario.
        - atomic_write_boundary_bytes is set to zero always

        SCSI also supports a WRITE ATOMIC (32) command, which is for type 2
        protection enabled. This is not going to be supported now, so check for
        T10_PI_TYPE2_PROTECTION when setting any request_queue limits.

        To handle an atomic write request, add support for WRITE ATOMIC (16)
        command in handler sd_setup_atomic_cmnd(). Flag use_atomic_write_boundary
        is checked here for encoding ATOMIC BOUNDARY field.

        Trace info is also added for WRITE_ATOMIC_16 command.

Change-Id: Ie072002fe2184615c72531ac081a324ef18cfb03
Signed-off-by: Kienan Stewart <kstewart@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: block_start removed from btrfs_get_extent in 6.11-rc1

See upstream commit:

    commit c77a8c61002e91d859e118008fd495efbe1d9373
    Author: Qu Wenruo <wqu@suse.com>
    Date:   Tue Apr 30 07:53:06 2024 +0930

        btrfs: remove extent_map::block_start member

        The member extent_map::block_start can be calculated from
        extent_map::disk_bytenr + extent_map::offset for regular extents.
        And otherwise just extent_map::disk_bytenr.

        And this is already validated by the validate_extent_map().  Now we can
        remove the member.

        However there is a special case in btrfs_create_dio_extent() where we
        for NOCOW/PREALLOC ordered extents cannot directly use the resulting
        btrfs_file_extent, as btrfs_split_ordered_extent() cannot handle them
        yet.

        So for that call site, we pass file_extent->disk_bytenr +
        file_extent->num_bytes as disk_bytenr for the ordered extent, and 0 for
        offset.

Change-Id: I2e3245bb0d1f5263e902659aa05848d5e231909b
Signed-off-by: Kienan Stewart <kstewart@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: block_len removed frmo btrfs_get_extent in 6.11-rc1

See upstream commit:

    commit e28b851ed9b232c3b84cb8d0fedbdfa8ca881386
    Author: Qu Wenruo <wqu@suse.com>
    Date:   Tue Apr 30 07:53:05 2024 +0930

        btrfs: remove extent_map::block_len member

        The extent_map::block_len is either extent_map::len (non-compressed
        extent) or extent_map::disk_num_bytes (compressed extent).

        Since we already have sanity checks to do the cross-checks between the
        new and old members, we can drop the old extent_map::block_len now.

        For most call sites, they can manually select extent_map::len or
        extent_map::disk_num_bytes, since most if not all of them have checked
        if the extent is compressed.

Change-Id: Ib03fc685b4e876bf4e53afdd28ca9826342a0e4e
Signed-off-by: Kienan Stewart <kstewart@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: orig_start removed from btrfs_get_extent in 6.11-rc1

See upstream commit:

    commit 4aa7b5d1784f510c0f42afc1d74efb41947221d7
    Author: Qu Wenruo <wqu@suse.com>
    Date:   Tue Apr 30 07:53:04 2024 +0930

        btrfs: remove extent_map::orig_start member

        Since we have extent_map::offset, the old extent_map::orig_start is just
        extent_map::start - extent_map::offset for non-hole/inline extents.

        And since the new extent_map::offset is already verified by
        validate_extent_map() while the old orig_start is not, let's just remove
        the old member from all call sites.

Change-Id: I025a30d49b3e3ddc37d7846acc191ebbdf2ff19e
Signed-off-by: Kienan Stewart <kstewart@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: ext4_da_reserve_space changed in 6.11-rc1

See upstream commit:

    commit 0d66b23d79c750276f791411d81a524549a64852
    Author: Zhang Yi <yi.zhang@huawei.com>
    Date:   Fri May 17 20:40:02 2024 +0800

        ext4: make ext4_da_reserve_space() reserve multi-clusters

        Add 'nr_resv' parameter to ext4_da_reserve_space(), which indicates the
        number of clusters wants to reserve, make it reserve multiple clusters
        at a time.

Change-Id: Ib1ce8c3023d53a6d22ec444a435fdb3c871f64c5
Signed-off-by: Kienan Stewart <kstewart@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: kfree_skb changed in 6.11-rc1

See upstream commit:

    commit c53795d48ee8f385c6a9e394651e7ee914baaeba
    Author: Yan Zhai <yan@cloudflare.com>
    Date:   Mon Jun 17 11:09:04 2024 -0700

        net: add rx_sk to trace_kfree_skb

        skb does not include enough information to find out receiving
        sockets/services and netns/containers on packet drops. In theory
        skb->dev tells about netns, but it can get cleared/reused, e.g. by TCP
        stack for OOO packet lookup. Similarly, skb->sk often identifies a local
        sender, and tells nothing about a receiver.

        Allow passing an extra receiving socket to the tracepoint to improve
        the visibility on receiving drops.

Change-Id: I33c8ce1a48006456f198ab1592f733b55be01016
Signed-off-by: Kienan Stewart <kstewart@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Introduce extension points for trace hit counters

Introduce two extension points for trace hit counters:

1) Future "actions" to perform other than "increment",

2) Future dimension indexing schemes (keys) other than tokens.

Change the layout of struct lttng_kernel_abi_counter_key_dimension
by adding a "key_type" field. A new struct lttng_kernel_abi_counter_key_dimension_tokens
inherits from struct lttng_kernel_abi_counter_key_dimension, and contains
the uint32_t nr_key_tokens field. The only currently supported key_type
is LTTNG_KERNEL_ABI_KEY_TYPE_TOKENS = 0.

Change the layout of struct lttng_kernel_abi_counter_event by adding an
"action" field. The only currently supported action is
LTTNG_KERNEL_ABI_COUNTER_ACTION_INCREMENT = 0.

Change the struct lttng_kernel_abi_key_token_string so it inherits from
struct lttng_kernel_abi_key_token. The "len" field of
struct lttng_kernel_abi_key_token now includes the length of the entire
child structure.

Remove struct lttng_kernel_abi_counter_key: it was previously expecting
all key dimensions to have the same size. But because each dimension can
be of a different type, each may have its own distinct size.

Change the newly introduced API between probe providers to change the
"event_counter_add" callback into a "counter_hit" callback, which takes
one less argument (no integer value), but takes additional stack_data,
probe_ctx, and event_counter_ctx arguments for future use.

Change-Id: I4de86a9fac75ce759f58b545dffa2f99c59f5688
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: event notifier: set eval_capture to false for kprobe, kretprobe and uprobe

Trying to capture fields for kprobe, kretprobe, uprobe, event
notifications will end up dereferencing NULL pointers. Prevent execution
of capture code in those cases.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: If86b0602351fb42047b184608b28108e9de0d183

Implement REUSE 3.0 with SPDX identifiers

Implement the full REUSE spec [1] to help with copyright and licensing
audits and compliance. This will reduce a lot of manual work for the
licensing audit required in Debian on each update and also allow using
automated tools.

For files that lacked copyright and licensing information, I used the
following guidelines. If a clear author could be determined from the git
history use it, otherwise use 'EfficiOS Inc.'. For code use
'GPL-2.0-only OR LGPL-2.1-only' unless otherwise stated, for
documentation 'CC-BY-SA-4.0' and for data files 'CC0-1.0'.

Freeform text files were converted to Markdown to allow licensing
comments.

Running the reuse tool on the repo is now succesful:

  $ reuse lint

  # SUMMARY

  * Bad licenses: 0
  * Deprecated licenses: 0
  * Licenses without file extension: 0
  * Missing licenses: 0
  * Unused licenses: 0
  * Used licenses: CC0-1.0, GPL-2.0-only, CC-BY-SA-4.0, MIT, LGPL-2.1-only
  * Read errors: 0
  * files with copyright information: 358 / 358
  * files with license information: 358 / 358

  Congratulations! Your project is compliant with version 3.0 of the REUSE Specification :-)

[1] https://reuse.software/tutorial/

Change-Id: I1755cab24a6fcec7a6c9a2136891418203ec34b8
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: add 'static inline' to lttng_kretprobes_init_event()

Add missing 'static inline' to lttng_kretprobes_init_event() placeholder
function when CONFIG_KRETPROBES is not set. Also some minor reformating
to improve readability.

Change-Id: I23cf83dff99f4168ae0f339c2b4911796e0b0273
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: copy_struct_from_user() for non-LTS branches < v4.19

The 'linux/bits.h' was backported to LTS branches but is not available
on non-LTS before v4.19. Use 'asm/byteorder.h' instead to get the
__LITTLE_ENDIAN define which is available on all kernel versions we
support.

Change-Id: Icfe733ab944616b3bd6d0023ad0869eefb830b34
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: copy_struct_from_user() wrapper

* Add 'static inline' to lttng_check_zeroed_user()
* Add missing parenthesis in lttng_check_zeroed_user()
* Add wrapper for aligned_byte_mask()

Change-Id: I92f3bce14ac91e896148d455de2ae0634b4d7972
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix: lttng-abi: zero-init counter_conf

Both lttng_abi_copy_user_old_counter_conf and
lttng_abi_copy_user_counter_conf should zero-init the counter_conf
destination argument, else the "dimension->flags" field is uninitialized
before being OR'd with flags.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ib2e0ca5871ec6fc9f485ec1b60f362de9568b9d1

counter ABI: Fix too large stack size warning

Fix the following warning by splitting lttng_counter_ioctl (which has a
lot of local variables in the switch/case legs) into many sub-functions.

/home/efficios/git/lttng-modules/src/lttng-abi.c: In function ‘lttng_counter_ioctl’:
/home/efficios/git/lttng-modules/src/lttng-abi.c:1227:1: warning: the frame size of 1032 bytes is larger than 1024 bytes [-Wframe-larger-than=]
 1227 | }
      | ^

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I250c78827a743c024da2e28fe807fdd5c290e686

__lttng_counter_add: skip effect-less code when global_sum_step=0

Skip effect-less code in the common case where global_sum_step=0.
This saves about 1.5ns on x86-64.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I62e8d163143780e64d7130051b1013470fef48fe

ABI: add key_string_len output field in lttng_kernel_abi_counter_map_descriptor

Tell userspace the size of the key string. Useful to adjust memory
allocation for a retry.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ifde8f2acc841579b717b78128f8bfddbc0dfee26

Implement extensible LTTNG_KERNEL_ABI_COUNTER_EVENT ABI

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I427287757a8842ab163791a6b34918ce80adcf51

Implement extensible LTTNG_KERNEL_ABI_COUNTER_MAP_DESCRIPTOR

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ibd16679a5e3354553a9b606a6f3a5f9f2692836d

Implement extensible counter read, aggregate, clear ioctls

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I5984e5442dbfe47e00fe4166d0d6cc686e51d655

Implement counter configuration/dimensions extensible ABI

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I0660738c60a128b57a6975e65db5bf2a1562a323

Implement copy_user_event_param_ext with lttng_copy_struct_from_user

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I55b0c811acc0c09374091dfa2cdc7535d22c9f34

Introduce wrapper lttng_copy_struct_from_user

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I67e3d0249bbfdb5bc6e9d469119220a1fe3f03fd

Introduce event counter extensible ABI structure layout

Introduce event counter ioctl argument structure with extensible layout
for new ioctls.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I120e057d136951973b9af801fd4940fb9866ce94

ABI refactoring: Rename event counter ABI to "old"

The counter ioctls exposed in 2.13 for notifications use a fixed-size
array for the number of dimensions.

We aim to replace this by variable length arrays in 2.14 for event
counters, and introduce new ABIs as extensible structures.

Rename the ioctls as follows:

- LTTNG_KERNEL_ABI_COUNTER -> LTTNG_KERNEL_ABI_OLD_COUNTER
- LTTNG_KERNEL_ABI_COUNTER_READ -> LTTNG_KERNEL_ABI_OLD_COUNTER_READ
- LTTNG_KERNEL_ABI_COUNTER_AGGREGATE -> LTTNG_KERNEL_ABI_OLD_COUNTER_AGGREGATE
- LTTNG_KERNEL_ABI_COUNTER_CLEAR -> LTTNG_KERNEL_ABI_OLD_COUNTER_CLEAR

And rename the following structures:

- lttng_kernel_abi_counter_dimension -> lttng_kernel_abi_old_counter_dimension
- lttng_kernel_abi_counter_conf -> lttng_kernel_abi_old_counter_conf
- lttng_kernel_abi_counter_index -> lttng_kernel_abi_old_counter_index
- lttng_kernel_abi_counter_value -> lttng_kernel_abi_old_counter_value
- lttng_kernel_abi_counter_read -> lttng_kernel_abi_old_counter_read
- lttng_kernel_abi_counter_aggregate -> lttng_kernel_abi_old_counter_aggregate
- lttng_kernel_abi_counter_clear -> lttng_kernel_abi_old_counter_clear

Introduce "TODO" comments where the new ABIs should be implemented.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I7a4f04e89a56ca1415d326976c4ea3afb9715295

kretprobe: Implement kretprobes with event enablers

This enables support for enabling only entry or exit for kretprobes.
This enables event notifier support for kretprobes.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ieef4f2744d5e1ebfc2f3320862e143bfe6401620

kprobe: Implement kprobes with event enablers

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Idd1285baa8da6476620ddfc3c21d93a8d4ab08f7

ABI: introduce lttng_kernel_abi_match_check

Requires introduction of extended event parameters because there are
only 8 bytes of padding left in struct lttng_kernel_abi_event.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I3dd1346093eeee42c1a7a047ec32865fda9c6ff5

kretprobe: implement lttng_kretprobes_match_check

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I7a4584c3fa524664097a262d7b5f74eb93d2a335

kprobe: Introduce lttng_kprobes_match_check

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Iffc9dba158f9802db9f877fabeec288d5072a84c

Implement event notifier kretprobe support

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I32b42ddad79889a19ee8833ed9b52e4a37505f7b

Remove lttng_kernel_event_create warnings

Errors returned by lttng_kernel_event_create are handled by the caller,
and may happen e.g. when a kprobe or kretprobe symbol does not exist.
It should not generate a warning in the kernel console.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I0bd59ec35e5648fa6aa1ece3cf36714c5cc1a83e

kretprobes: implement event counter support

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Id5142ae8d31bba2c9a806f75cbbfefdca303d523

lttng_abi_create_event_notifier: add missing fallthrough

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I8e7011f63ccce4db21b36fcd18eb78e95ca2f406

Implement events-by-key hash table

Allow re-use of counter index for events with matching key.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Id1dff39ff3c58d5664d04ae5d77d1ae66e728cd6

Rename lttng_event_notifier_group events_ht to events_name_ht

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I64313d4b8255a717ef06ef7bf5944d2c1c4e24da

Rename lttng_get_event_ht_from_enabler to lttng_get_events_name_ht_from_enabler

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I53bb9917ae7644ba79f36dc16f9feb095a3c4194

Implement counter maps (for listing)

Also wire up LTTNG_KERNEL_ABI_SYSCALL_MASK in lttng_counter_ioctl.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: If9215bafb010162250636c9407d85fefcd61ebc2

Implement channel counter creation/destroy

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I2e242fe9b50f3b4ae5f97969b587f4306051109f

Cleanup: remove kprobes/kretprobes/uprobes unused exports

Can be removed now that those are embedded in the tracer core.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I14f4b011c562d7a2e669749159fda23ee964554f

Fix: circular dependency on symbol lttng_id_tracker_lookup

Adding lttng_id_tracker_lookup feature into kprobes, uprobes and
kretprobes introduces a circular dependency between lttng-tracer.ko and
the respective probe modules.

There is no real reason for having the kprobes/uprobes/kretprobes
modules separate from the tracer core, so combine those.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Iea80c8054054b341e1836c8675c04e00424608fc

Fix: circular dependency between lttng-events and lib counter

lib counter depends on lttng-events (tracer), not the opposite. Use the
lttng-counter clients callbacks to remove the circular dependency.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I04cdbfb39deb76884092459cab62227b9f39d85d

Implement event counter creation

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I5b407513c5c64ae418fc1846ae10f3fa99609796

Rename lttng_abi_create_event to lttng_abi_create_event_recorder_enabler

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I39b2e288b952e0316b6511f4f79014700cbc42e5

uprobe: implement counter support

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I46909ecc57a15d5f0c6b58e1f2f254219e55e01d

kretprobes: implement counter support

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I8e64712455b2812305250a45311b5b023281a1fe

kprobe: implement counter support

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I8741ab89b03089ec34f17965749f84ddc23a2601

Add missing counter handling

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I5ba6bc03b7c090e6c9b6431d81da4a739c764610

abi: introduce session common

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Id6512b87661d7e29666f41002370d065d540a27a

Implement event counter probe

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I6cdeef84a3b2f5675029a7509ce353a17bbeb0b4

Remove unused lttng_event_enabler_event_name_match_event

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Icd02b897aa6aaa870e503db873f86f80a04ff02c

Match event keys for syscall events

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ia0b3df16abeda9e0a404eeed156e33c17f4f47e3

Skip sync of disabled enablers

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Icf20e16baee3c63b84f1be9f730100229108bcc6

Remove duplicate lookup in lttng_event_enabler_create_tracepoint_events_if_missing

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ic597f34dfc3823ba83728bde3fcd41bb3eed9c91

Implement _lttng_kernel_event_create counter event support

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: If1a2e1d64d8663f592e81611914402d30f111e17

lttng_kernel_event_alloc: handler counters

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ic50018274bd5048181f9da278027e1c175b6bd80

Refactor notification error counters

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I106b6fe85f899297582e131252c591ac5ba7d538

Add counter case in lttng_kernel_event_id_available

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I1967ddd37d677806558a41138f32566ce19bae21

Implement event counters

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I9eadee226c78c6b48158a0670b9b50a8b6aab06e

Add channel counter structures to internal header

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I00b2a4df48d66977deafedeb3e9f5021754bf42a

Wire up LTTNG_KERNEL_EVENT_TYPE_COUNTER

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Iddfaa8de487a376d35d404415bfec0676d7cb304

Refactor _lttng_event_destroy duplicated code

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I42a607101c084ade8ebed673b3a666987cfad98f

struct lttng_kernel_event_session_common_private: remove unused ctx field

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I99a8eb21db55ec8d9df2300853b3d08abec095e5

Rename struct lttng_kernel_session_private events_ht field to events_name_ht

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I2452d6154064fd91e801d0b8487168c7cdfb0a3e

Rename struct lttng_metadata_stream list field to node

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Iaefccfd27d18cbb90726f6b77400409eeea493d0

Rename struct lttng_kernel_session_private list field to node

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I971bb068e51567ef79a30608984137827f1d45c8

Reorder struct lttng_kernel_session_private fields

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I9a7c8f415c1b0b128662081f580df14bd5368893

Rename session events field to events_head

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I8181d86dc2dede4ba15178722e1eb884feb556eb

Rename session chan field to chan_head

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I4447978d05d2979e8c7e8c6d66126da66142e32c

Introduce struct lttng_event_counter_enabler

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I9683c18b4d661d65b128a1edda5adb1154d8d543

Introduce struct lttng_kernel_event_counter_private

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I2cc9d82a18f8222682e413e41a27acb8f8a74754

Move counter key structures to beginning of events-internal.h

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I4b76cac6837a5098720da9ae321169b46d78866b

Introduce struct lttng_kernel_channel_counter and struct lttng_kernel_channel_counter_ops

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Icdc8c1fb6444eca2051b60fa4a7dd564e943da68

Introduce struct lttng_kernel_event_counter

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ie0e9ab234bce48c3ce23cf0d4c664ce4bb3bdaee

Introduce struct lttng_kernel_event_session_common_private

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ibaab6a0a8ce60bbc5ac212106c5c75a8a16d7e2c

Introduce struct lttng_event_enabler_session_common

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ia160e559525f9276512506271742703804be6c9b

Trace hit counters: introduce counter private data structures

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: Ie6adadcbab8c41237b2bfb03b1c341b4183715cc

Trace hit counters: ABI

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I886975f27af90e42cbab274f70820a06090c48bc

Fix: libcounter: __lttng_counter_add global sum step alloc vs sync mixup

The global sum step needs to be handled on per-cpu allocated counters,
not per-cpu sync.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I9b6ef221fb099c9fc602dfdb9a93d70ae2192500

Fix: lttng_abi_validate_event_param: use kretprobe enum

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I8c444e16781fb8d8df9f05f0c70775f324ae030f

kvm instrumentation: Fix kvm_mmio event NULL pointer dereference

Upstream Linux commit e39d200fa5bf ("KVM: Fix stack-out-of-bounds read
in write_mmio") introduce a NULL pointer check within TP_fast_assign().

lttng-modules commit 33630522da97 ("Update kvm instrumentation for 4.15")
introduce use of:

  ctf_sequence_hex(unsigned char, val, val, u32, len)

without the required NULL pointer check, which can trigger NULL pointer
dereference in case of unsatisfied MMIO read.

Add the missing NULL pointer check. Record a sequence of length 0 in the
trace when the val pointer is NULL.

Reported-by: Fahad Arslan <fahad.arslan@siemens.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I51a171a56af96e2cf68dba73f7eb473dd6c0ba0e

kvm instrumentation: Cleanup: Eliminate code duplication

Commit 0badc02f82b38 ("Fix: adjust SLE version ranges to build with SP2
and SP3") introduced code duplication. Modify the version match logic to
remove duplicated code.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I918c87d2e6d68e6c71df18fbf66d88445ce68d23

Fix: Build on CentOS 9 Stream 2024-06

Change-Id: I445d2df9d49930e28e57b6a7d075a9f68e914bad
Signed-off-by: Kienan Stewart <kstewart@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: file: Rename fcheck lookup_fd_rcu (v5.10.220)

See upstream backported commit:

  commit c4716bb296504cbc64aeefb370df44e821214c44
  Author: Eric W. Biederman <ebiederm@xmission.com>
  Date:   Fri Nov 20 17:14:27 2020 -0600

    file: Rename fcheck lookup_fd_rcu

    [ Upstream commit 460b4f812a9d473d4b39d87d37844f9fc30a9eb3 ]

    Also remove the confusing comment about checking if a fd exists.  I
    could not find one instance in the entire kernel that still matches
    the description or the reason for the name fcheck.

    The need for better names became apparent in the last round of
    discussion of this set of changes[1].

    [1] https://lkml.kernel.org/r/CAHk-=wj8BQbgJFLa+J0e=iT-1qpmCRTbPAJ8gd6MJQ=kbRPqyQ@mail.gmail.com
Link: https://lkml.kernel.org/r/20201120231441.29911-10-ebiederm@xmission.com
Change-Id: Ib880bd8feef1c5d75d2a018cd93a1d464485ab7b
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: update stale file paths in LICENSE

Change-Id: I4849b19daa235b93a6435e57bd764128e43d691e
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Cleanup: use 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.

Change-Id: I8b59b3689aa38fb5f5a114f9d02f22274a5bff57
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Warn and return on fd overflow fdt

The fdt should only grow and iterate_fd() holds file_lock, which should
ensure the fdt does not change while the lock is taken but be cautious
and check anyway.

Change-Id: Icd6a3263026734cbe3f296f6087f79add4148a8f
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

cleanup: add correct error messages to lttng-get-syscall-inout.sh

The script argument parsing was modified in
commit 6ae73da43362 ("Fix syscall generator scripts") without adjusting
the error messages and comments.

Report the proper errors for the current arguments.

Change-Id: I86f4d16ffede1a6a48bc6ab5fe1e678ef340585b
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

cleanup: document 'pipe' syscall override

Change-Id: Iec84c10b774c1001ff2035024a1d05fdd31c8d9d
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: net: udp: add IP/port data to the tracepoint udp/udp_fail_queue_rcv_skb (v6.10)

See upstream commit:

  commit e9669a00bba79442dd4862c57761333d6a020c24
  Author: Balazs Scheidler <bazsi77@gmail.com>
  Date:   Tue Mar 26 19:05:47 2024 +0100

    net: udp: add IP/port data to the tracepoint udp/udp_fail_queue_rcv_skb

    The udp_fail_queue_rcv_skb() tracepoint lacks any details on the source
    and destination IP/port whereas this information can be critical in case
    of UDP/syslog.

Change-Id: I0c337c5817b0a120298cbf5088d60671d9625b0d
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: close_on_exec(): pass files_struct instead of fdtable (v6.10)

See upstream commit:

  commit f60d374d2cc88034385265d193a38e3f4a4b430c
  Author: Al Viro <viro@zeniv.linux.org.uk>
  Date:   Thu Jan 4 21:35:38 2024 -0500

    close_on_exec(): pass files_struct instead of fdtable

    both callers are happier that way...

Change-Id: I8cdabb073c2090842b27b74954d86cb486c43b3e
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: btrfs: move ->parent and ->ref_root into btrfs_delayed_ref_node (v6.10)

See upstream commit:

  commit cf4f04325b2b27efa5697ba0ea4c1abdee0035b4
  Author: Josef Bacik <josef@toxicpanda.com>
  Date:   Fri Apr 12 22:57:13 2024 -0400

    btrfs: move ->parent and ->ref_root into btrfs_delayed_ref_node

    These two members are shared by both the tree refs and data refs, so
    move them into btrfs_delayed_ref_node proper.  This allows us to greatly
    simplify the comparison code, as the shared refs always only sort on
    parent, and the non shared refs always sort first on ref_root, and then
    only data refs sort on their specific fields.

Change-Id: Ib7c92cc4bb8d674ac66ccfa25c03476f7adaaf90
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: btrfs: simplify delayed ref tracepoints (v6.10)

See upstream commit:

  commit 1bff6d4f873790cfc675afce9860208576508c5a
  Author: Josef Bacik <josef@toxicpanda.com>
  Date:   Fri Apr 12 20:27:00 2024 -0400

    btrfs: simplify delayed ref tracepoints

    Now that all of the delayed ref information is in the delayed ref node,
    drastically simplify the delayed ref tracepoints by simply passing in
    the btrfs_delayed_ref_node and populating the tracepoints with the
    values from the structure itself.

Change-Id: Ic90bc23d6aa558baec33adc33b4d21e052e83375
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Fix mm_vmscan_lru_isolate tracepoint for RHEL 9.4 kernel

Redhat has moved to using the format first found in the 6.7 kernel
for the mm_vmscan_lru_isolate tracepoint.

Change-Id: I2aa84769c0070458d902e9a0305488d6d8a380e1
Signed-off-by: Martin Hicks <martin@sr-research.com>
Signed-off-by: Kienan Stewart <kstewart@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

fix: Add missing 'pselect6_time32' and 'ppoll_time32' syscall overrides

The instrumentation currently has overrides to the generated syscall
tracepoints of 'ppoll' and 'pselect6' to extract additional information
from the parameters.

On arm-32 and x86-32 these 2 syscalls were renamed to 'ppoll_time32' and
'pselect6_time32' and new syscalls using 64-bit time_t were introduced
with the old names. This results in missing overrides on these
architectures for the 32-bit variants that were renamed.

Add the '_time32' overrides to restore the previous behavior.

Change-Id: I81e3a3ddc3f3cea58d86edcdf4a1fc9b600637c2
Signed-off-by: Michael Jeanson <mjeanson@efficios.com>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

page alloc wrapper: Fix get_pfnblock_flags_mask prototype

The canary __canary__get_pfnblock_flags_mask has never done its job of
detecting changes to the prototype of get_pfnblock_flags_mask because
it was actually calling the wrapper, because the wrapper/page_alloc.h
header maps get_pfnblock_flags_mask to wrapper_get_pfnblock_flags_mask.

Unfortunately, this wrapper is included by page_alloc.c only _after_ the
linux/pageblock-flags.h header is included, which means the
get_pfnblock_flags_mask prototype does _not_ have the wrapper prefix,
which prevents it from being useful for any kind of type validation.

This has been detected by a compiler warning stating that
wrapper_get_pfnblock_flags_mask() does not have a prior declaration.

Move the wrapper/page_alloc.h include _before_ including
pageblock-flags.h. This ensures the declaration has the wrapper_ prefix,
and therefore the compiler compares the declaration with the definition
of wrapper_get_pfnblock_flags_mask within page_alloc.c. The canary
function can be removed because it is redundant with this type check.

With this proper type check in place, we notice the following two
changes upstream:

commit 535b81e209219 ("mm/page_alloc.c: remove unnecessary end_bitidx for [set|get]_pfnblock_flags_mask()")
introduced in v5.9 removes the end_bitidx argument.

commit ca891f41c4c79 ("mm: constify get_pfnblock_flags_mask and get_pfnblock_migratetype")
introduced in v5.14 adds a const qualifier to the struct page pointer.

Adapt the code to match the evolution of this prototype.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I51b7871edfbff0f74ba1cf4d0ad988eb8d642b4e

lttng probe: include events-internal.h

Include events-internal.h for the declarations of lttng_logger_init and
lttng_logger_exit.

Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Change-Id: I14968060c309083f90a282f186f3c635f1ebfd8d