[RFC] xfs_io: Add cachestat syscall support

[RFC] xfs_io: Add cachestat syscall support

[Ritesh Harjani (IBM)]

This adds -c "cachestat off len" command which uses cachestat() syscall
[1]. This can provide following pagecache detail for a file.

- no. of cached pages,
- no. of dirty pages,
- no. of pages marked for writeback,
- no. of evicted pages,
- no. of recently evicted pages

[1]: https://lore.kernel.org/all/20230503013608.2431726-3-nphamcs@gmail.com/T/#u

Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
---
 configure.ac          |  1 +
 include/builddefs.in  |  1 +
 io/Makefile           |  5 +++
 io/cachestat.c        | 77 +++++++++++++++++++++++++++++++++++++++++++
 io/init.c             |  1 +
 io/io.h               |  6 ++++
 m4/package_libcdev.m4 | 19 +++++++++++
 7 files changed, 110 insertions(+)
 create mode 100644 io/cachestat.c

diff --git a/configure.ac b/configure.ac
index 8c76f398..f039bc91 100644
--- a/configure.ac
+++ b/configure.ac
@@ -154,6 +154,7 @@ AC_PACKAGE_NEED_RCU_INIT
 
 AC_HAVE_PWRITEV2
 AC_HAVE_COPY_FILE_RANGE
+AC_HAVE_CACHESTAT
 AC_NEED_INTERNAL_FSXATTR
 AC_NEED_INTERNAL_FSCRYPT_ADD_KEY_ARG
 AC_NEED_INTERNAL_FSCRYPT_POLICY_V2
diff --git a/include/builddefs.in b/include/builddefs.in
index 82840ec7..fe2a7824 100644
--- a/include/builddefs.in
+++ b/include/builddefs.in
@@ -95,6 +95,7 @@ HAVE_ZIPPED_MANPAGES = @have_zipped_manpages@
 
 HAVE_PWRITEV2 = @have_pwritev2@
 HAVE_COPY_FILE_RANGE = @have_copy_file_range@
+HAVE_CACHESTAT = @have_cachestat@
 NEED_INTERNAL_FSXATTR = @need_internal_fsxattr@
 NEED_INTERNAL_FSCRYPT_ADD_KEY_ARG = @need_internal_fscrypt_add_key_arg@
 NEED_INTERNAL_FSCRYPT_POLICY_V2 = @need_internal_fscrypt_policy_v2@
diff --git a/io/Makefile b/io/Makefile
index 14a3fe20..444e2d6a 100644
--- a/io/Makefile
+++ b/io/Makefile
@@ -61,6 +61,11 @@ CFILES += copy_file_range.c
 LCFLAGS += -DHAVE_COPY_FILE_RANGE
 endif
 
+ifeq ($(HAVE_CACHESTAT),yes)
+CFILES += cachestat.c
+LCFLAGS += -DHAVE_CACHESTAT
+endif
+
 ifeq ($(ENABLE_EDITLINE),yes)
 LLDLIBS += $(LIBEDITLINE) $(LIBTERMCAP)
 endif
diff --git a/io/cachestat.c b/io/cachestat.c
new file mode 100644
index 00000000..9edf3f9a
--- /dev/null
+++ b/io/cachestat.c
@@ -0,0 +1,77 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "command.h"
+#include "input.h"
+#include "init.h"
+#include "io.h"
+#include <unistd.h>
+#include <linux/mman.h>
+#include <asm/unistd.h>
+
+static cmdinfo_t cachestat_cmd;
+
+static void print_cachestat(struct cachestat *cs)
+{
+	printf(_("Cached: %llu, Dirty: %llu, Writeback: %llu, Evicted: %llu, Recently Evicted: %llu\n"),
+			cs->nr_cache, cs->nr_dirty, cs->nr_writeback,
+			cs->nr_evicted, cs->nr_recently_evicted);
+}
+
+static int
+cachestat_f(int argc, char **argv)
+{
+	off_t offset = 0, length = 0;
+	size_t blocksize, sectsize;
+	struct cachestat_range cs_range;
+	struct cachestat cs;
+
+	if (argc != 3) {
+		exitcode = 1;
+		return command_usage(&cachestat_cmd);
+	}
+
+	init_cvtnum(&blocksize, &sectsize);
+	offset = cvtnum(blocksize, sectsize, argv[1]);
+	if (offset < 0) {
+		printf(_("invalid offset argument -- %s\n"), argv[1]);
+		exitcode = 1;
+		return 0;
+	}
+
+	length = cvtnum(blocksize, sectsize, argv[2]);
+	if (length < 0) {
+		printf(_("invalid length argument -- %s\n"), argv[2]);
+		exitcode = 1;
+		return 0;
+	}
+
+	cs_range.off = offset;
+	cs_range.len = length;
+
+	if (syscall(__NR_cachestat, file->fd, &cs_range, &cs, 0)) {
+		perror("cachestat");
+		exitcode = 1;
+		return 0;
+	}
+
+	print_cachestat(&cs);
+
+	return 0;
+}
+
+static cmdinfo_t cachestat_cmd = {
+	.name		= "cachestat",
+	.altname	= "cs",
+	.cfunc		= cachestat_f,
+	.argmin		= 2,
+	.argmax		= 2,
+	.flags		= CMD_NOMAP_OK | CMD_FOREIGN_OK,
+	.args		= "[off len]",
+	.oneline	= "find page cache pages for a given file",
+};
+
+void cachestat_init(void)
+{
+	add_command(&cachestat_cmd);
+}
+
diff --git a/io/init.c b/io/init.c
index 4831deae..49e9e7cb 100644
--- a/io/init.c
+++ b/io/init.c
@@ -49,6 +49,7 @@ init_commands(void)
 	bmap_init();
 	bulkstat_init();
 	copy_range_init();
+	cachestat_init();
 	cowextsize_init();
 	encrypt_init();
 	fadvise_init();
diff --git a/io/io.h b/io/io.h
index d9906558..259c0349 100644
--- a/io/io.h
+++ b/io/io.h
@@ -132,6 +132,12 @@ extern void		copy_range_init(void);
 #define copy_range_init()	do { } while (0)
 #endif
 
+#ifdef HAVE_CACHESTAT
+extern void cachestat_init(void);
+#else
+#define cachestat_init() do { } while (0)
+#endif
+
 extern void		sync_range_init(void);
 extern void		readdir_init(void);
 extern void		reflink_init(void);
diff --git a/m4/package_libcdev.m4 b/m4/package_libcdev.m4
index 4ef7e8f6..af9da812 100644
--- a/m4/package_libcdev.m4
+++ b/m4/package_libcdev.m4
@@ -35,6 +35,25 @@ syscall(__NR_copy_file_range, 0, 0, 0, 0, 0, 0);
     AC_SUBST(have_copy_file_range)
   ])
 
+#
+# Check if we have a cachestat system call (Linux)
+#
+AC_DEFUN([AC_HAVE_CACHESTAT],
+  [ AC_MSG_CHECKING([for cachestat])
+    AC_LINK_IFELSE(
+    [	AC_LANG_PROGRAM([[
+#include <unistd.h>
+#include <linux/mman.h>
+#include <asm/unistd.h>
+	]], [[
+syscall(__NR_cachestat, 0, 0, 0, 0);
+	]])
+    ], have_cachestat=yes
+       AC_MSG_RESULT(yes),
+       AC_MSG_RESULT(no))
+    AC_SUBST(have_cachestat)
+  ])
+
 #
 # Check if we need to override the system struct fsxattr with
 # the internal definition.  This /only/ happens if the system
-- 
2.48.1

Re: [RFC] xfs_io: Add cachestat syscall support

[Andrey Albershteyn]

On 2025-03-17 00:15:29, Ritesh Harjani (IBM) wrote:
> This adds -c "cachestat off len" command which uses cachestat() syscall
> [1]. This can provide following pagecache detail for a file.
> 
> - no. of cached pages,
> - no. of dirty pages,
> - no. of pages marked for writeback,
> - no. of evicted pages,
> - no. of recently evicted pages
> 
> [1]: https://lore.kernel.org/all/20230503013608.2431726-3-nphamcs@gmail.com/T/#u
> 
> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
> ---
>  configure.ac          |  1 +
>  include/builddefs.in  |  1 +
>  io/Makefile           |  5 +++
>  io/cachestat.c        | 77 +++++++++++++++++++++++++++++++++++++++++++
>  io/init.c             |  1 +
>  io/io.h               |  6 ++++
>  m4/package_libcdev.m4 | 19 +++++++++++
>  7 files changed, 110 insertions(+)
>  create mode 100644 io/cachestat.c
> 
> diff --git a/configure.ac b/configure.ac
> index 8c76f398..f039bc91 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -154,6 +154,7 @@ AC_PACKAGE_NEED_RCU_INIT
>  
>  AC_HAVE_PWRITEV2
>  AC_HAVE_COPY_FILE_RANGE
> +AC_HAVE_CACHESTAT
>  AC_NEED_INTERNAL_FSXATTR
>  AC_NEED_INTERNAL_FSCRYPT_ADD_KEY_ARG
>  AC_NEED_INTERNAL_FSCRYPT_POLICY_V2
> diff --git a/include/builddefs.in b/include/builddefs.in
> index 82840ec7..fe2a7824 100644
> --- a/include/builddefs.in
> +++ b/include/builddefs.in
> @@ -95,6 +95,7 @@ HAVE_ZIPPED_MANPAGES = @have_zipped_manpages@
>  
>  HAVE_PWRITEV2 = @have_pwritev2@
>  HAVE_COPY_FILE_RANGE = @have_copy_file_range@
> +HAVE_CACHESTAT = @have_cachestat@
>  NEED_INTERNAL_FSXATTR = @need_internal_fsxattr@
>  NEED_INTERNAL_FSCRYPT_ADD_KEY_ARG = @need_internal_fscrypt_add_key_arg@
>  NEED_INTERNAL_FSCRYPT_POLICY_V2 = @need_internal_fscrypt_policy_v2@
> diff --git a/io/Makefile b/io/Makefile
> index 14a3fe20..444e2d6a 100644
> --- a/io/Makefile
> +++ b/io/Makefile
> @@ -61,6 +61,11 @@ CFILES += copy_file_range.c
>  LCFLAGS += -DHAVE_COPY_FILE_RANGE
>  endif
>  
> +ifeq ($(HAVE_CACHESTAT),yes)
> +CFILES += cachestat.c
> +LCFLAGS += -DHAVE_CACHESTAT
> +endif
> +
>  ifeq ($(ENABLE_EDITLINE),yes)
>  LLDLIBS += $(LIBEDITLINE) $(LIBTERMCAP)
>  endif
> diff --git a/io/cachestat.c b/io/cachestat.c
> new file mode 100644
> index 00000000..9edf3f9a
> --- /dev/null
> +++ b/io/cachestat.c
> @@ -0,0 +1,77 @@
> +// SPDX-License-Identifier: GPL-2.0
> +
> +#include "command.h"
> +#include "input.h"
> +#include "init.h"
> +#include "io.h"
> +#include <unistd.h>
> +#include <linux/mman.h>
> +#include <asm/unistd.h>
> +
> +static cmdinfo_t cachestat_cmd;
> +
> +static void print_cachestat(struct cachestat *cs)
> +{
> +	printf(_("Cached: %llu, Dirty: %llu, Writeback: %llu, Evicted: %llu, Recently Evicted: %llu\n"),
> +			cs->nr_cache, cs->nr_dirty, cs->nr_writeback,
> +			cs->nr_evicted, cs->nr_recently_evicted);
> +}
> +
> +static int
> +cachestat_f(int argc, char **argv)
> +{
> +	off_t offset = 0, length = 0;
> +	size_t blocksize, sectsize;
> +	struct cachestat_range cs_range;
> +	struct cachestat cs;
> +
> +	if (argc != 3) {
> +		exitcode = 1;
> +		return command_usage(&cachestat_cmd);
> +	}
> +
> +	init_cvtnum(&blocksize, &sectsize);
> +	offset = cvtnum(blocksize, sectsize, argv[1]);
> +	if (offset < 0) {
> +		printf(_("invalid offset argument -- %s\n"), argv[1]);
> +		exitcode = 1;
> +		return 0;
> +	}
> +
> +	length = cvtnum(blocksize, sectsize, argv[2]);
> +	if (length < 0) {
> +		printf(_("invalid length argument -- %s\n"), argv[2]);
> +		exitcode = 1;
> +		return 0;
> +	}
> +
> +	cs_range.off = offset;
> +	cs_range.len = length;
> +
> +	if (syscall(__NR_cachestat, file->fd, &cs_range, &cs, 0)) {
> +		perror("cachestat");
> +		exitcode = 1;
> +		return 0;
> +	}
> +
> +	print_cachestat(&cs);
> +
> +	return 0;
> +}
> +
> +static cmdinfo_t cachestat_cmd = {
> +	.name		= "cachestat",
> +	.altname	= "cs",
> +	.cfunc		= cachestat_f,
> +	.argmin		= 2,
> +	.argmax		= 2,
> +	.flags		= CMD_NOMAP_OK | CMD_FOREIGN_OK,
> +	.args		= "[off len]",
I suppose [] aren't needed, as those arguments are mandatory. I can
fix while pulling if no v2

> +	.oneline	= "find page cache pages for a given file",
> +};

Otherwise, looks good to me
Reviewed-by: Andrey Albershteyn <aalbersh@kernel.org>

-- 
- Andrey

Re: [RFC] xfs_io: Add cachestat syscall support

[Darrick J. Wong]

On Mon, Mar 17, 2025 at 12:15:29AM +0530, Ritesh Harjani (IBM) wrote:
> This adds -c "cachestat off len" command which uses cachestat() syscall
> [1]. This can provide following pagecache detail for a file.
> 
> - no. of cached pages,
> - no. of dirty pages,
> - no. of pages marked for writeback,
> - no. of evicted pages,
> - no. of recently evicted pages
> 
> [1]: https://lore.kernel.org/all/20230503013608.2431726-3-nphamcs@gmail.com/T/#u
> 
> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>

Just out of curiosity, is there a manpage change for xfs_io.8 that goes
with this new subcommand?

--D

> ---
>  configure.ac          |  1 +
>  include/builddefs.in  |  1 +
>  io/Makefile           |  5 +++
>  io/cachestat.c        | 77 +++++++++++++++++++++++++++++++++++++++++++
>  io/init.c             |  1 +
>  io/io.h               |  6 ++++
>  m4/package_libcdev.m4 | 19 +++++++++++
>  7 files changed, 110 insertions(+)
>  create mode 100644 io/cachestat.c
> 
> diff --git a/configure.ac b/configure.ac
> index 8c76f398..f039bc91 100644
> --- a/configure.ac
> +++ b/configure.ac
> @@ -154,6 +154,7 @@ AC_PACKAGE_NEED_RCU_INIT
>  
>  AC_HAVE_PWRITEV2
>  AC_HAVE_COPY_FILE_RANGE
> +AC_HAVE_CACHESTAT
>  AC_NEED_INTERNAL_FSXATTR
>  AC_NEED_INTERNAL_FSCRYPT_ADD_KEY_ARG
>  AC_NEED_INTERNAL_FSCRYPT_POLICY_V2
> diff --git a/include/builddefs.in b/include/builddefs.in
> index 82840ec7..fe2a7824 100644
> --- a/include/builddefs.in
> +++ b/include/builddefs.in
> @@ -95,6 +95,7 @@ HAVE_ZIPPED_MANPAGES = @have_zipped_manpages@
>  
>  HAVE_PWRITEV2 = @have_pwritev2@
>  HAVE_COPY_FILE_RANGE = @have_copy_file_range@
> +HAVE_CACHESTAT = @have_cachestat@
>  NEED_INTERNAL_FSXATTR = @need_internal_fsxattr@
>  NEED_INTERNAL_FSCRYPT_ADD_KEY_ARG = @need_internal_fscrypt_add_key_arg@
>  NEED_INTERNAL_FSCRYPT_POLICY_V2 = @need_internal_fscrypt_policy_v2@
> diff --git a/io/Makefile b/io/Makefile
> index 14a3fe20..444e2d6a 100644
> --- a/io/Makefile
> +++ b/io/Makefile
> @@ -61,6 +61,11 @@ CFILES += copy_file_range.c
>  LCFLAGS += -DHAVE_COPY_FILE_RANGE
>  endif
>  
> +ifeq ($(HAVE_CACHESTAT),yes)
> +CFILES += cachestat.c
> +LCFLAGS += -DHAVE_CACHESTAT
> +endif
> +
>  ifeq ($(ENABLE_EDITLINE),yes)
>  LLDLIBS += $(LIBEDITLINE) $(LIBTERMCAP)
>  endif
> diff --git a/io/cachestat.c b/io/cachestat.c
> new file mode 100644
> index 00000000..9edf3f9a
> --- /dev/null
> +++ b/io/cachestat.c
> @@ -0,0 +1,77 @@
> +// SPDX-License-Identifier: GPL-2.0
> +
> +#include "command.h"
> +#include "input.h"
> +#include "init.h"
> +#include "io.h"
> +#include <unistd.h>
> +#include <linux/mman.h>
> +#include <asm/unistd.h>
> +
> +static cmdinfo_t cachestat_cmd;
> +
> +static void print_cachestat(struct cachestat *cs)
> +{
> +	printf(_("Cached: %llu, Dirty: %llu, Writeback: %llu, Evicted: %llu, Recently Evicted: %llu\n"),
> +			cs->nr_cache, cs->nr_dirty, cs->nr_writeback,
> +			cs->nr_evicted, cs->nr_recently_evicted);
> +}
> +
> +static int
> +cachestat_f(int argc, char **argv)
> +{
> +	off_t offset = 0, length = 0;
> +	size_t blocksize, sectsize;
> +	struct cachestat_range cs_range;
> +	struct cachestat cs;
> +
> +	if (argc != 3) {
> +		exitcode = 1;
> +		return command_usage(&cachestat_cmd);
> +	}
> +
> +	init_cvtnum(&blocksize, &sectsize);
> +	offset = cvtnum(blocksize, sectsize, argv[1]);
> +	if (offset < 0) {
> +		printf(_("invalid offset argument -- %s\n"), argv[1]);
> +		exitcode = 1;
> +		return 0;
> +	}
> +
> +	length = cvtnum(blocksize, sectsize, argv[2]);
> +	if (length < 0) {
> +		printf(_("invalid length argument -- %s\n"), argv[2]);
> +		exitcode = 1;
> +		return 0;
> +	}
> +
> +	cs_range.off = offset;
> +	cs_range.len = length;
> +
> +	if (syscall(__NR_cachestat, file->fd, &cs_range, &cs, 0)) {
> +		perror("cachestat");
> +		exitcode = 1;
> +		return 0;
> +	}
> +
> +	print_cachestat(&cs);
> +
> +	return 0;
> +}
> +
> +static cmdinfo_t cachestat_cmd = {
> +	.name		= "cachestat",
> +	.altname	= "cs",
> +	.cfunc		= cachestat_f,
> +	.argmin		= 2,
> +	.argmax		= 2,
> +	.flags		= CMD_NOMAP_OK | CMD_FOREIGN_OK,
> +	.args		= "[off len]",
> +	.oneline	= "find page cache pages for a given file",
> +};
> +
> +void cachestat_init(void)
> +{
> +	add_command(&cachestat_cmd);
> +}
> +
> diff --git a/io/init.c b/io/init.c
> index 4831deae..49e9e7cb 100644
> --- a/io/init.c
> +++ b/io/init.c
> @@ -49,6 +49,7 @@ init_commands(void)
>  	bmap_init();
>  	bulkstat_init();
>  	copy_range_init();
> +	cachestat_init();
>  	cowextsize_init();
>  	encrypt_init();
>  	fadvise_init();
> diff --git a/io/io.h b/io/io.h
> index d9906558..259c0349 100644
> --- a/io/io.h
> +++ b/io/io.h
> @@ -132,6 +132,12 @@ extern void		copy_range_init(void);
>  #define copy_range_init()	do { } while (0)
>  #endif
>  
> +#ifdef HAVE_CACHESTAT
> +extern void cachestat_init(void);
> +#else
> +#define cachestat_init() do { } while (0)
> +#endif
> +
>  extern void		sync_range_init(void);
>  extern void		readdir_init(void);
>  extern void		reflink_init(void);
> diff --git a/m4/package_libcdev.m4 b/m4/package_libcdev.m4
> index 4ef7e8f6..af9da812 100644
> --- a/m4/package_libcdev.m4
> +++ b/m4/package_libcdev.m4
> @@ -35,6 +35,25 @@ syscall(__NR_copy_file_range, 0, 0, 0, 0, 0, 0);
>      AC_SUBST(have_copy_file_range)
>    ])
>  
> +#
> +# Check if we have a cachestat system call (Linux)
> +#
> +AC_DEFUN([AC_HAVE_CACHESTAT],
> +  [ AC_MSG_CHECKING([for cachestat])
> +    AC_LINK_IFELSE(
> +    [	AC_LANG_PROGRAM([[
> +#include <unistd.h>
> +#include <linux/mman.h>
> +#include <asm/unistd.h>
> +	]], [[
> +syscall(__NR_cachestat, 0, 0, 0, 0);
> +	]])
> +    ], have_cachestat=yes
> +       AC_MSG_RESULT(yes),
> +       AC_MSG_RESULT(no))
> +    AC_SUBST(have_cachestat)
> +  ])
> +
>  #
>  # Check if we need to override the system struct fsxattr with
>  # the internal definition.  This /only/ happens if the system
> -- 
> 2.48.1
> 
> 

Re: [RFC] xfs_io: Add cachestat syscall support

[Ritesh Harjani]

"Darrick J. Wong" <djwong@kernel.org> writes:

> On Mon, Mar 17, 2025 at 12:15:29AM +0530, Ritesh Harjani (IBM) wrote:
>> This adds -c "cachestat off len" command which uses cachestat() syscall
>> [1]. This can provide following pagecache detail for a file.
>> 
>> - no. of cached pages,
>> - no. of dirty pages,
>> - no. of pages marked for writeback,
>> - no. of evicted pages,
>> - no. of recently evicted pages
>> 
>> [1]: https://lore.kernel.org/all/20230503013608.2431726-3-nphamcs@gmail.com/T/#u
>> 
>> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
>
> Just out of curiosity, is there a manpage change for xfs_io.8 that goes
> with this new subcommand?
>

Looks like I might have missed the man page entry change,...

>> ---
>>  configure.ac          |  1 +
>>  include/builddefs.in  |  1 +
>>  io/Makefile           |  5 +++
>>  io/cachestat.c        | 77 +++++++++++++++++++++++++++++++++++++++++++
>>  io/init.c             |  1 +
>>  io/io.h               |  6 ++++
>>  m4/package_libcdev.m4 | 19 +++++++++++
>>  7 files changed, 110 insertions(+)
>>  create mode 100644 io/cachestat.c

...else we would have seen an entry in man/man8/xfs_io.8 file. 

Since this is taken in for-next, I can submit a separate patch for the
man page entry update.

Thanks for catching it. 

-ritesh