[PATCH v2 0/2] Introduce tracepoint for hugetlbfs

[PATCH v2 0/2] Introduce tracepoint for hugetlbfs

[Hongbo Li]

Here we add some basic tracepoints for debugging hugetlbfs: {alloc, free,
evict}_inode, setattr and fallocate.

v1 can be found at:
https://lore.kernel.org/linux-mm/20240701194906.3a9b6765@gandalf.local.home/T/

Changes since v1:
  - Decrease the parameters for setattr tracer suggested by Steve and Mathieu.
  - Replace current_user_ns() with init_user_ns when translate uid/gid.

Hongbo Li (2):
  hugetlbfs: support tracepoint
  hugetlbfs: use tracepoints in hugetlbfs functions.

 MAINTAINERS                      |   1 +
 fs/hugetlbfs/inode.c             |  17 +++-
 include/trace/events/hugetlbfs.h | 160 +++++++++++++++++++++++++++++++
 3 files changed, 176 insertions(+), 2 deletions(-)
 create mode 100644 include/trace/events/hugetlbfs.h

-- 
2.34.1

[PATCH v2 1/2] hugetlbfs: support tracepoint

[Hongbo Li]

Add basic tracepoints for {alloc, evict, free}_inode, setattr and
fallocate. These can help users to debug hugetlbfs more conveniently.

Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
---
 MAINTAINERS                      |   1 +
 include/trace/events/hugetlbfs.h | 160 +++++++++++++++++++++++++++++++
 2 files changed, 161 insertions(+)
 create mode 100644 include/trace/events/hugetlbfs.h

diff --git a/MAINTAINERS b/MAINTAINERS
index cd2ca0c3158e..865c48e92d40 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -10188,6 +10188,7 @@ F:	Documentation/mm/hugetlbfs_reserv.rst
 F:	Documentation/mm/vmemmap_dedup.rst
 F:	fs/hugetlbfs/
 F:	include/linux/hugetlb.h
+F:	include/trace/events/hugetlbfs.h
 F:	mm/hugetlb.c
 F:	mm/hugetlb_vmemmap.c
 F:	mm/hugetlb_vmemmap.h
diff --git a/include/trace/events/hugetlbfs.h b/include/trace/events/hugetlbfs.h
new file mode 100644
index 000000000000..975f584f6f51
--- /dev/null
+++ b/include/trace/events/hugetlbfs.h
@@ -0,0 +1,160 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM hugetlbfs
+
+#if !defined(_TRACE_HUGETLBFS_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_HUGETLBFS_H
+
+#include <linux/tracepoint.h>
+
+TRACE_EVENT(hugetlbfs_alloc_inode,
+
+	TP_PROTO(struct inode *inode, struct inode *dir, int mode),
+
+	TP_ARGS(inode, dir, mode),
+
+	TP_STRUCT__entry(
+		__field(dev_t,		dev)
+		__field(ino_t,		ino)
+		__field(ino_t,		dir)
+		__field(__u16,		mode)
+	),
+
+	TP_fast_assign(
+		__entry->dev		= inode->i_sb->s_dev;
+		__entry->ino		= inode->i_ino;
+		__entry->dir		= dir->i_ino;
+		__entry->mode		= mode;
+	),
+
+	TP_printk("dev = (%d,%d), ino = %lu, dir = %lu, mode = 0%o",
+		MAJOR(__entry->dev), MINOR(__entry->dev),
+		(unsigned long) __entry->ino,
+		(unsigned long) __entry->dir, __entry->mode)
+);
+
+DECLARE_EVENT_CLASS(hugetlbfs__inode,
+
+	TP_PROTO(struct inode *inode),
+
+	TP_ARGS(inode),
+
+	TP_STRUCT__entry(
+		__field(dev_t,		dev)
+		__field(ino_t,		ino)
+		__field(__u16,		mode)
+		__field(loff_t,		size)
+		__field(unsigned int,	nlink)
+		__field(unsigned int,	seals)
+		__field(blkcnt_t,	blocks)
+	),
+
+	TP_fast_assign(
+		__entry->dev		= inode->i_sb->s_dev;
+		__entry->ino		= inode->i_ino;
+		__entry->mode		= inode->i_mode;
+		__entry->size		= inode->i_size;
+		__entry->nlink		= inode->i_nlink;
+		__entry->seals		= HUGETLBFS_I(inode)->seals;
+		__entry->blocks		= inode->i_blocks;
+	),
+
+	TP_printk("dev = (%d,%d), ino = %lu, i_mode = 0%o, i_size = %lld, i_nlink = %u, seals = %u, i_blocks = %llu",
+		MAJOR(__entry->dev), MINOR(__entry->dev), (unsigned long) __entry->ino,
+		__entry->mode, __entry->size, __entry->nlink, __entry->seals,
+		(unsigned long long)__entry->blocks)
+);
+
+DEFINE_EVENT(hugetlbfs__inode, hugetlbfs_evict_inode,
+
+	TP_PROTO(struct inode *inode),
+
+	TP_ARGS(inode)
+);
+
+DEFINE_EVENT(hugetlbfs__inode, hugetlbfs_free_inode,
+
+	TP_PROTO(struct inode *inode),
+
+	TP_ARGS(inode)
+);
+
+TRACE_EVENT(hugetlbfs_setattr,
+
+	TP_PROTO(struct inode *inode, struct dentry *dentry,
+		struct iattr *attr),
+
+	TP_ARGS(inode, dentry, attr),
+
+	TP_STRUCT__entry(
+		__field(dev_t,		dev)
+		__field(ino_t,		ino)
+		__field(unsigned int,	d_len)
+		__string(d_name,	dentry->d_name.name)
+		__field(unsigned int,	ia_valid)
+		__field(unsigned int,	ia_mode)
+		__field(unsigned int,	ia_uid)
+		__field(unsigned int,	ia_gid)
+		__field(loff_t,		old_size)
+		__field(loff_t,		ia_size)
+	),
+
+	TP_fast_assign(
+		__entry->dev		= inode->i_sb->s_dev;
+		__entry->ino		= inode->i_ino;
+		__entry->d_len		= dentry->d_name.len;
+		__assign_str(d_name);
+		__entry->ia_valid	= attr->ia_valid;
+		__entry->ia_mode	= attr->ia_mode;
+		__entry->ia_uid		= from_kuid_munged(current_user_ns(), attr->ia_uid);
+		__entry->ia_gid		= from_kgid_munged(current_user_ns(), attr->ia_gid);
+		__entry->old_size	= inode->i_size;
+		__entry->ia_size	= attr->ia_size;
+	),
+
+	TP_printk("dev = (%d,%d), ino = %lu, name = %.*s, ia_valid = %u, ia_mode = 0%o, ia_uid = %u, ia_gid = %u, old_size = %lld, ia_size = %lld",
+		MAJOR(__entry->dev), MINOR(__entry->dev), (unsigned long)__entry->ino,
+		__entry->d_len, __get_str(d_name), __entry->ia_valid, __entry->ia_mode,
+		__entry->ia_uid, __entry->ia_gid, __entry->old_size, __entry->ia_size)
+);
+
+TRACE_EVENT(hugetlbfs_fallocate,
+
+	TP_PROTO(struct inode *inode, int mode,
+		loff_t offset, loff_t len, int ret),
+
+	TP_ARGS(inode, mode, offset, len, ret),
+
+	TP_STRUCT__entry(
+		__field(dev_t,		dev)
+		__field(ino_t,		ino)
+		__field(int,		mode)
+		__field(loff_t,		offset)
+		__field(loff_t,		len)
+		__field(loff_t,		size)
+		__field(int,		ret)
+	),
+
+	TP_fast_assign(
+		__entry->dev		= inode->i_sb->s_dev;
+		__entry->ino		= inode->i_ino;
+		__entry->mode		= mode;
+		__entry->offset		= offset;
+		__entry->len		= len;
+		__entry->size		= inode->i_size;
+		__entry->ret		= ret;
+	),
+
+	TP_printk("dev = (%d,%d), ino = %lu, mode = %x, offset = %lld, len = %lld,  i_size = %lld, ret = %d",
+		MAJOR(__entry->dev), MINOR(__entry->dev),
+		(unsigned long)__entry->ino, __entry->mode,
+		(unsigned long long)__entry->offset,
+		(unsigned long long)__entry->len,
+		(unsigned long long)__entry->size,
+		__entry->ret)
+);
+
+#endif /* _TRACE_HUGETLBFS_H */
+
+ /* This part must be outside protection */
+#include <trace/define_trace.h>
-- 
2.34.1

Re: [PATCH v2 1/2] hugetlbfs: support tracepoint

[Matthew Wilcox]

On Thu, Jul 04, 2024 at 11:07:03AM +0800, Hongbo Li wrote:
> +	TP_printk("dev = (%d,%d), ino = %lu, dir = %lu, mode = 0%o",
> +		MAJOR(__entry->dev), MINOR(__entry->dev),
> +		(unsigned long) __entry->ino,
> +		(unsigned long) __entry->dir, __entry->mode)

erofs and f2fs are the only two places that print devices like this.

	"dev=%d:%d inode=%lx"

Why do we need dir and mode?

Actually, why do we need a tracepoint on alloc_inode at all?  What
does it help us debug, and why does no other filesystem need an
alloc_inode tracepoint?

Re: [PATCH v2 1/2] hugetlbfs: support tracepoint

[Hongbo Li]

On 2024/7/4 11:37, Matthew Wilcox wrote:
> On Thu, Jul 04, 2024 at 11:07:03AM +0800, Hongbo Li wrote:
>> +	TP_printk("dev = (%d,%d), ino = %lu, dir = %lu, mode = 0%o",
>> +		MAJOR(__entry->dev), MINOR(__entry->dev),
>> +		(unsigned long) __entry->ino,
>> +		(unsigned long) __entry->dir, __entry->mode)
> 
> erofs and f2fs are the only two places that print devices like this.
> 
> 	"dev=%d:%d inode=%lx"
> 
> Why do we need dir and mode?
Thanks for reviewing!

Here dir and mode are used to track the creation of the directory tree.
> 
> Actually, why do we need a tracepoint on alloc_inode at all?  What
> does it help us debug, and why does no other filesystem need an
> alloc_inode tracepoint?
> 
In fact, f2fs and ext4 have added this tracepoint such as 
trace_f2fs_new_inode(in f2fs) and trace_ext4_allocate_inode(in ext4). 
This can trace the lifecycle of an inode comprehensively. These 
tracepoints are used to debug some closed application scenarios, and 
also helping developers to debug the filesystem logic in hugetlbfs.

Thanks,
Hongbo

Re: [PATCH v2 1/2] hugetlbfs: support tracepoint

[Dave Chinner]

On Thu, Jul 04, 2024 at 04:37:30AM +0100, Matthew Wilcox wrote:
> On Thu, Jul 04, 2024 at 11:07:03AM +0800, Hongbo Li wrote:
> > +	TP_printk("dev = (%d,%d), ino = %lu, dir = %lu, mode = 0%o",
> > +		MAJOR(__entry->dev), MINOR(__entry->dev),
> > +		(unsigned long) __entry->ino,
> > +		(unsigned long) __entry->dir, __entry->mode)
> 
> erofs and f2fs are the only two places that print devices like this.
> 
> 	"dev=%d:%d inode=%lx"

"dev %d:%d inode %lx"

i.e. every token in the output should be space separated, and no
commas between values.

Any other format makes it difficult for post processing tracepoint
output with sed, grep, awk, python, etc. Every token then has to be
split into name and value parts, and then the value has to have the
comma stripped from it.

Having to do this is additional work when writing use-once scripts
that get thrown away when the tracepoint output analysis is done
is painful, and it's completely unnecessary if the tracepoint output
is completely space separated from the start.

-Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH v2 1/2] hugetlbfs: support tracepoint

[Steven Rostedt]

On Thu, 4 Jul 2024 22:56:29 +1000
Dave Chinner <david@fromorbit.com> wrote:

> Having to do this is additional work when writing use-once scripts
> that get thrown away when the tracepoint output analysis is done
> is painful, and it's completely unnecessary if the tracepoint output
> is completely space separated from the start.

If you are using scripts to parse the output, then you could just
enable the "fields" options, which will just ignore the TP_printk() and
print the fields in both their hex and decimal values:

 # trace-cmd start -e filemap -O fields

// the above fields change can also be done with:
//  echo 1 > /sys/kernel/tracing/options/fields

 # trace-cmd show
# tracer: nop
#
# entries-in-buffer/entries-written: 8/8   #P:8
#
#                                _-----=> irqs-off/BH-disabled
#                               / _----=> need-resched
#                              | / _---=> hardirq/softirq
#                              || / _--=> preempt-depth
#                              ||| / _-=> migrate-disable
#                              |||| /     delay
#           TASK-PID     CPU#  |||||  TIMESTAMP  FUNCTION
#              | |         |   |||||     |         |
            less-2527    [004] ..... 61949.896458: mm_filemap_add_to_page_cache: pfn=0x144625 (1328677) i_ino=0x335c6 (210374) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)
            less-2527    [004] d..2. 61949.896926: mm_filemap_delete_from_page_cache: pfn=0x152b07 (1387271) i_ino=0x2d73a (186170) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)
     jbd2/vda3-8-268     [005] ..... 61954.461964: mm_filemap_add_to_page_cache: pfn=0x152b70 (1387376) i_ino=0xfe00003 (266338307) index=0x30bd33 (3194163) s_dev=0x3 (3) order=(0)
     jbd2/vda3-8-268     [005] ..... 61954.462669: mm_filemap_add_to_page_cache: pfn=0x15335b (1389403) i_ino=0xfe00003 (266338307) index=0x30bd40 (3194176) s_dev=0x3 (3) order=(0)
     jbd2/vda3-8-268     [005] ..... 62001.565391: mm_filemap_add_to_page_cache: pfn=0x13a996 (1288598) i_ino=0xfe00003 (266338307) index=0x30bd41 (3194177) s_dev=0x3 (3) order=(0)
     jbd2/vda3-8-268     [005] ..... 62001.566081: mm_filemap_add_to_page_cache: pfn=0x1446b5 (1328821) i_ino=0xfe00003 (266338307) index=0x30bd43 (3194179) s_dev=0x3 (3) order=(0)
            less-2530    [004] ..... 62033.182309: mm_filemap_add_to_page_cache: pfn=0x13d755 (1300309) i_ino=0x2d73a (186170) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)
            less-2530    [004] d..2. 62033.182801: mm_filemap_delete_from_page_cache: pfn=0x144625 (1328677) i_ino=0x335c6 (210374) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)


Just an FYI,

-- Steve

Re: [PATCH v2 1/2] hugetlbfs: support tracepoint

[Dave Chinner]

On Thu, Jul 04, 2024 at 10:13:22AM -0400, Steven Rostedt wrote:
> On Thu, 4 Jul 2024 22:56:29 +1000
> Dave Chinner <david@fromorbit.com> wrote:
> 
> > Having to do this is additional work when writing use-once scripts
> > that get thrown away when the tracepoint output analysis is done
> > is painful, and it's completely unnecessary if the tracepoint output
> > is completely space separated from the start.
> 
> If you are using scripts to parse the output, then you could just
> enable the "fields" options, which will just ignore the TP_printk() and
> print the fields in both their hex and decimal values:
> 
>  # trace-cmd start -e filemap -O fields
> 
> // the above fields change can also be done with:
> //  echo 1 > /sys/kernel/tracing/options/fields
> 
>  # trace-cmd show
> # tracer: nop
> #
> # entries-in-buffer/entries-written: 8/8   #P:8
> #
> #                                _-----=> irqs-off/BH-disabled
> #                               / _----=> need-resched
> #                              | / _---=> hardirq/softirq
> #                              || / _--=> preempt-depth
> #                              ||| / _-=> migrate-disable
> #                              |||| /     delay
> #           TASK-PID     CPU#  |||||  TIMESTAMP  FUNCTION
> #              | |         |   |||||     |         |
>             less-2527    [004] ..... 61949.896458: mm_filemap_add_to_page_cache: pfn=0x144625 (1328677) i_ino=0x335c6 (210374) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)
>             less-2527    [004] d..2. 61949.896926: mm_filemap_delete_from_page_cache: pfn=0x152b07 (1387271) i_ino=0x2d73a (186170) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)
>      jbd2/vda3-8-268     [005] ..... 61954.461964: mm_filemap_add_to_page_cache: pfn=0x152b70 (1387376) i_ino=0xfe00003 (266338307) index=0x30bd33 (3194163) s_dev=0x3 (3) order=(0)
>      jbd2/vda3-8-268     [005] ..... 61954.462669: mm_filemap_add_to_page_cache: pfn=0x15335b (1389403) i_ino=0xfe00003 (266338307) index=0x30bd40 (3194176) s_dev=0x3 (3) order=(0)
>      jbd2/vda3-8-268     [005] ..... 62001.565391: mm_filemap_add_to_page_cache: pfn=0x13a996 (1288598) i_ino=0xfe00003 (266338307) index=0x30bd41 (3194177) s_dev=0x3 (3) order=(0)
>      jbd2/vda3-8-268     [005] ..... 62001.566081: mm_filemap_add_to_page_cache: pfn=0x1446b5 (1328821) i_ino=0xfe00003 (266338307) index=0x30bd43 (3194179) s_dev=0x3 (3) order=(0)
>             less-2530    [004] ..... 62033.182309: mm_filemap_add_to_page_cache: pfn=0x13d755 (1300309) i_ino=0x2d73a (186170) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)
>             less-2530    [004] d..2. 62033.182801: mm_filemap_delete_from_page_cache: pfn=0x144625 (1328677) i_ino=0x335c6 (210374) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)

Yes, I know about that. But this just makes things harder, because
now there are *3* different formats that have to be handled (i.e.
now we also have to strip "()" around numbers).

-Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH v2 1/2] hugetlbfs: support tracepoint

[Hongbo Li]

On 2024/7/5 7:34, Dave Chinner wrote:
> On Thu, Jul 04, 2024 at 10:13:22AM -0400, Steven Rostedt wrote:
>> On Thu, 4 Jul 2024 22:56:29 +1000
>> Dave Chinner <david@fromorbit.com> wrote:
>>
>>> Having to do this is additional work when writing use-once scripts
>>> that get thrown away when the tracepoint output analysis is done
>>> is painful, and it's completely unnecessary if the tracepoint output
>>> is completely space separated from the start.
>>
>> If you are using scripts to parse the output, then you could just
>> enable the "fields" options, which will just ignore the TP_printk() and
>> print the fields in both their hex and decimal values:
>>
>>   # trace-cmd start -e filemap -O fields
>>
>> // the above fields change can also be done with:
>> //  echo 1 > /sys/kernel/tracing/options/fields
>>
>>   # trace-cmd show
>> # tracer: nop
>> #
>> # entries-in-buffer/entries-written: 8/8   #P:8
>> #
>> #                                _-----=> irqs-off/BH-disabled
>> #                               / _----=> need-resched
>> #                              | / _---=> hardirq/softirq
>> #                              || / _--=> preempt-depth
>> #                              ||| / _-=> migrate-disable
>> #                              |||| /     delay
>> #           TASK-PID     CPU#  |||||  TIMESTAMP  FUNCTION
>> #              | |         |   |||||     |         |
>>              less-2527    [004] ..... 61949.896458: mm_filemap_add_to_page_cache: pfn=0x144625 (1328677) i_ino=0x335c6 (210374) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)
>>              less-2527    [004] d..2. 61949.896926: mm_filemap_delete_from_page_cache: pfn=0x152b07 (1387271) i_ino=0x2d73a (186170) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)
>>       jbd2/vda3-8-268     [005] ..... 61954.461964: mm_filemap_add_to_page_cache: pfn=0x152b70 (1387376) i_ino=0xfe00003 (266338307) index=0x30bd33 (3194163) s_dev=0x3 (3) order=(0)
>>       jbd2/vda3-8-268     [005] ..... 61954.462669: mm_filemap_add_to_page_cache: pfn=0x15335b (1389403) i_ino=0xfe00003 (266338307) index=0x30bd40 (3194176) s_dev=0x3 (3) order=(0)
>>       jbd2/vda3-8-268     [005] ..... 62001.565391: mm_filemap_add_to_page_cache: pfn=0x13a996 (1288598) i_ino=0xfe00003 (266338307) index=0x30bd41 (3194177) s_dev=0x3 (3) order=(0)
>>       jbd2/vda3-8-268     [005] ..... 62001.566081: mm_filemap_add_to_page_cache: pfn=0x1446b5 (1328821) i_ino=0xfe00003 (266338307) index=0x30bd43 (3194179) s_dev=0x3 (3) order=(0)
>>              less-2530    [004] ..... 62033.182309: mm_filemap_add_to_page_cache: pfn=0x13d755 (1300309) i_ino=0x2d73a (186170) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)
>>              less-2530    [004] d..2. 62033.182801: mm_filemap_delete_from_page_cache: pfn=0x144625 (1328677) i_ino=0x335c6 (210374) index=0x0 (0) s_dev=0xfe00003 (266338307) order=(0)
> 
> Yes, I know about that. But this just makes things harder, because
> now there are *3* different formats that have to be handled (i.e.
> now we also have to strip "()" around numbers).
Perhaps if users want to filter the format, they could enable the 
"fields" option in a unified manner. As for TP_printk(), it depends on 
how to better display the data used for debugging.

Thanks,
Hongbo
> 
> -Dave.

[PATCH v2 2/2] hugetlbfs: use tracepoints in hugetlbfs functions.

[Hongbo Li]

Here we use the hugetlbfs tracepoint to track the call stack. And
the output in trace is as follows:

```
touch-5250    [002] .....   123.557640: hugetlbfs_alloc_inode: dev = (0,49), ino = 29897, dir = 24079, mode = 0100644
touch-5250    [002] .....   123.557650: hugetlbfs_setattr: dev = (0,49), ino = 29897, name = test1, ia_valid = 131184, ia_mode = 036720, ia_uid = 2544251907, ia_gid = 65534, old_size = 0, ia_size = 4064
truncate-5251    [002] .....   142.939424: hugetlbfs_setattr: dev = (0,49), ino = 29897, name = test1, ia_valid = 8296, ia_mode = 00, ia_uid = 0, ia_gid = 0, old_size = 0, ia_size = 2097152
rm-5273    [002] .....   412.618383: hugetlbfs_evict_inode: dev = (0,49), ino = 29897, i_mode = 0100644, i_size = 2097152, i_nlink = 0, seals = 1, i_blocks = 0
<idle>-0       [002] ..s1.   412.634518: hugetlbfs_free_inode: dev = (0,49), ino = 29897, i_mode = 0100644, i_size = 2097152, i_nlink = 0, seals = 1, i_blocks = 0
```

Signed-off-by: Hongbo Li <lihongbo22@huawei.com>
---
 fs/hugetlbfs/inode.c | 17 +++++++++++++++--
 1 file changed, 15 insertions(+), 2 deletions(-)

diff --git a/fs/hugetlbfs/inode.c b/fs/hugetlbfs/inode.c
index 412f295acebe..2e826bbcb6ed 100644
--- a/fs/hugetlbfs/inode.c
+++ b/fs/hugetlbfs/inode.c
@@ -39,6 +39,9 @@
 #include <linux/uaccess.h>
 #include <linux/sched/mm.h>
 
+#define CREATE_TRACE_POINTS
+#include <trace/events/hugetlbfs.h>
+
 static const struct address_space_operations hugetlbfs_aops;
 static const struct file_operations hugetlbfs_file_operations;
 static const struct inode_operations hugetlbfs_dir_inode_operations;
@@ -686,6 +689,7 @@ static void hugetlbfs_evict_inode(struct inode *inode)
 {
 	struct resv_map *resv_map;
 
+	trace_hugetlbfs_evict_inode(inode);
 	remove_inode_hugepages(inode, 0, LLONG_MAX);
 
 	/*
@@ -813,8 +817,10 @@ static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
 	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
 		return -EOPNOTSUPP;
 
-	if (mode & FALLOC_FL_PUNCH_HOLE)
-		return hugetlbfs_punch_hole(inode, offset, len);
+	if (mode & FALLOC_FL_PUNCH_HOLE) {
+		error = hugetlbfs_punch_hole(inode, offset, len);
+		goto out_nolock;
+	}
 
 	/*
 	 * Default preallocate case.
@@ -918,6 +924,9 @@ static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
 	inode_set_ctime_current(inode);
 out:
 	inode_unlock(inode);
+
+out_nolock:
+	trace_hugetlbfs_fallocate(inode, mode, offset, len, error);
 	return error;
 }
 
@@ -934,6 +943,8 @@ static int hugetlbfs_setattr(struct mnt_idmap *idmap,
 	if (error)
 		return error;
 
+	trace_hugetlbfs_setattr(inode, dentry, attr);
+
 	if (ia_valid & ATTR_SIZE) {
 		loff_t oldsize = inode->i_size;
 		loff_t newsize = attr->ia_size;
@@ -1032,6 +1043,7 @@ static struct inode *hugetlbfs_get_inode(struct super_block *sb,
 			break;
 		}
 		lockdep_annotate_inode_mutex_key(inode);
+		trace_hugetlbfs_alloc_inode(inode, dir, mode);
 	} else {
 		if (resv_map)
 			kref_put(&resv_map->refs, resv_map_release);
@@ -1274,6 +1286,7 @@ static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
 
 static void hugetlbfs_free_inode(struct inode *inode)
 {
+	trace_hugetlbfs_free_inode(inode);
 	kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
 }
 
-- 
2.34.1