[lxcfs.git] / debian / patches / extra / 0001-make-meminfo-and-swaps-cgroupv2-aware.patch

From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
From: Wolfgang Bumiller <w.bumiller@proxmox.com>
Date: Thu, 8 Jul 2021 12:04:05 +0200
Subject: [PATCH lxcfs] make meminfo and swaps cgroupv2 aware

and deduplicate the corresponding code

Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
---
 src/bindings.c  | 10 ++++++
 src/bindings.h  |  1 +
 src/proc_fuse.c | 89 +++++++++++++++++++++++++------------------------
 3 files changed, 57 insertions(+), 43 deletions(-)

diff --git a/src/bindings.c b/src/bindings.c
index eb62ddb..e417a13 100644
--- a/src/bindings.c
+++ b/src/bindings.c
@@ -44,6 +44,7 @@ static bool can_use_pidfd;
 static bool can_use_swap;
 static bool can_use_sys_cpu;
 static bool has_versioned_opts;
+static bool memory_is_cgroupv2;
 
 static volatile sig_atomic_t reload_successful;
 
@@ -67,6 +68,11 @@ bool liblxcfs_has_versioned_opts(void)
 	return has_versioned_opts;
 }
 
+bool liblxcfs_memory_is_cgroupv2(void)
+{
+	return memory_is_cgroupv2;
+}
+
 /* Define pivot_root() if missing from the C library */
 #ifndef HAVE_PIVOT_ROOT
 static int pivot_root(const char *new_root, const char *put_old)
@@ -837,6 +843,7 @@ static void __attribute__((constructor)) lxcfs_init(void)
 				  pidfd = -EBADF;
 	int i = 0;
 	pid_t pid;
+	struct hierarchy *hierarchy;
 
 	lxcfs_info("Running constructor %s to reload liblxcfs", __func__);
 
@@ -894,6 +901,9 @@ static void __attribute__((constructor)) lxcfs_init(void)
 	else
 		lxcfs_info("Kernel does not support swap accounting");
 
+	hierarchy = cgroup_ops->get_hierarchy(cgroup_ops, "memory");
+	memory_is_cgroupv2 = hierarchy && is_unified_hierarchy(hierarchy);
+
 	lxcfs_info("api_extensions:");
 	for (size_t nr = 0; nr < nr_api_extensions; nr++)
 		lxcfs_info("- %s", api_extensions[nr]);
diff --git a/src/bindings.h b/src/bindings.h
index cd7d466..2b7c8f4 100644
--- a/src/bindings.h
+++ b/src/bindings.h
@@ -98,6 +98,7 @@ extern void prune_init_slice(char *cg);
 extern bool supports_pidfd(void);
 extern bool liblxcfs_functional(void);
 extern bool liblxcfs_can_use_swap(void);
+extern bool liblxcfs_memory_is_cgroupv2(void);
 extern bool liblxcfs_can_use_sys_cpu(void);
 extern bool liblxcfs_has_versioned_opts(void);
 
diff --git a/src/proc_fuse.c b/src/proc_fuse.c
index 94168c9..a07e8d6 100644
--- a/src/proc_fuse.c
+++ b/src/proc_fuse.c
@@ -299,6 +299,43 @@ static inline bool startswith(const char *line, const char *pref)
 	return strncmp(line, pref, strlen(pref)) == 0;
 }
 
+static void get_swap_info(const char *cgroup, uint64_t memlimit,
+			  uint64_t memusage, uint64_t *swtotal,
+			  uint64_t *swusage, uint64_t *memswpriority)
+{
+	__do_free char *memswusage_str = NULL, *memswpriority_str = NULL;
+	size_t memswlimit = 0, memswusage = 0;
+	int ret;
+
+	*swtotal = *swusage = 0;
+	*memswpriority = 1;
+
+	memswlimit = get_min_memlimit(cgroup, true);
+	if (memswlimit > 0) {
+		ret = cgroup_ops->get_memory_swap_current(cgroup_ops, cgroup, &memswusage_str);
+		if (ret < 0 || safe_uint64(memswusage_str, &memswusage, 10) != 0)
+			return;
+
+		if (liblxcfs_memory_is_cgroupv2()) {
+			*swtotal = memswlimit / 1024;
+			*swusage = memswusage / 1024;
+		} else {
+			if (memlimit > memswlimit)
+				*swtotal = 0;
+			else
+				*swtotal = (memswlimit - memlimit) / 1024;
+			if (memusage > memswusage || swtotal == 0)
+				*swusage = 0;
+			else
+				*swusage = (memswusage - memusage) / 1024;
+		}
+
+		ret = cgroup_ops->get_memory_swappiness(cgroup_ops, cgroup, &memswpriority_str);
+		if (ret >= 0)
+			safe_uint64(memswpriority_str, memswpriority, 10);
+	}
+}
+
 static int proc_swaps_read(char *buf, size_t size, off_t offset,
 			   struct fuse_file_info *fi)
 {
@@ -307,7 +344,7 @@ static int proc_swaps_read(char *buf, size_t size, off_t offset,
 	struct fuse_context *fc = fuse_get_context();
 	bool wants_swap = lxcfs_has_opt(fuse_get_context()->private_data, LXCFS_SWAP_ON);
 	struct file_info *d = INTTYPE_TO_PTR(fi->fh);
-	uint64_t memswlimit = 0, memlimit = 0, memusage = 0, memswusage = 0,
+	uint64_t memlimit = 0, memusage = 0,
 		 swtotal = 0, swusage = 0, memswpriority = 1,
 		 hostswtotal = 0, hostswfree = 0;
 	ssize_t total_len = 0;
@@ -353,26 +390,8 @@ static int proc_swaps_read(char *buf, size_t size, off_t offset,
 	if (safe_uint64(memusage_str, &memusage, 10) < 0)
 		lxcfs_error("Failed to convert memusage %s", memusage_str);
 
-	if (wants_swap) {
-		memswlimit = get_min_memlimit(cgroup, true);
-		if (memswlimit > 0) {
-			ret = cgroup_ops->get_memory_swap_current(cgroup_ops, cgroup, &memswusage_str);
-			if (ret >= 0 && safe_uint64(memswusage_str, &memswusage, 10) == 0) {
-				if (memlimit > memswlimit)
-					swtotal = 0;
-				else
-					swtotal = (memswlimit - memlimit) / 1024;
-				if (memusage > memswusage || swtotal == 0)
-					swusage = 0;
-				else
-					swusage = (memswusage - memusage) / 1024;
-			}
-
-			ret = cgroup_ops->get_memory_swappiness(cgroup_ops, cgroup, &memswpriority_str);
-			if (ret >= 0)
-				safe_uint64(memswpriority_str, &memswpriority, 10);
-		}
-	}
+	if (wants_swap)
+		get_swap_info(cgroup, memlimit, memusage, &swtotal, &swusage, &memswpriority);
 
 	total_len = snprintf(d->buf, d->size, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
 
@@ -1141,7 +1160,7 @@ static int proc_meminfo_read(char *buf, size_t size, off_t offset,
 	struct fuse_context *fc = fuse_get_context();
 	bool wants_swap = lxcfs_has_opt(fuse_get_context()->private_data, LXCFS_SWAP_ON);
 	struct file_info *d = INTTYPE_TO_PTR(fi->fh);
-	uint64_t memlimit = 0, memusage = 0, memswlimit = 0, memswusage = 0,
+	uint64_t memlimit = 0, memusage = 0,
 		 hosttotal = 0, swfree = 0, swusage = 0, swtotal = 0,
 		 memswpriority = 1;
 	struct memory_stat mstat = {};
@@ -1193,26 +1212,8 @@ static int proc_meminfo_read(char *buf, size_t size, off_t offset,
 	 * Following values are allowed to fail, because swapaccount might be
 	 * turned off for current kernel.
 	 */
-	if (wants_swap) {
-		memswlimit = get_min_memlimit(cgroup, true);
-		if (memswlimit > 0) {
-			ret = cgroup_ops->get_memory_swap_current(cgroup_ops, cgroup, &memswusage_str);
-			if (ret >= 0 && safe_uint64(memswusage_str, &memswusage, 10) == 0) {
-				if (memlimit > memswlimit)
-					swtotal = 0;
-				else
-					swtotal = (memswlimit - memlimit) / 1024;
-				if (memusage > memswusage || swtotal == 0)
-					swusage = 0;
-				else
-					swusage = (memswusage - memusage) / 1024;
-			}
-		}
-
-		ret = cgroup_ops->get_memory_swappiness(cgroup_ops, cgroup, &memswpriority_str);
-		if (ret >= 0)
-			safe_uint64(memswpriority_str, &memswpriority, 10);
-	}
+	if (wants_swap)
+		get_swap_info(cgroup, memlimit, memusage, &swtotal, &swusage, &memswpriority);
 
 	f = fopen_cached("/proc/meminfo", "re", &fopen_cache);
 	if (!f)
@@ -1251,7 +1252,9 @@ static int proc_meminfo_read(char *buf, size_t size, off_t offset,
 				   This is because the kernel can swap as much as it
 				   wants and not only up to swtotal. */
 
-				swtotal = memlimit + swtotal;
+				if (!liblxcfs_memory_is_cgroupv2())
+					swtotal += memlimit;
+
 				if (hostswtotal < swtotal) {
 					swtotal = hostswtotal;
 				}
Commit	Line	Data
62c5f3ad WB	1	From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
	2	From: Wolfgang Bumiller <w.bumiller@proxmox.com>
	3	Date: Thu, 8 Jul 2021 12:04:05 +0200
	4	Subject: [PATCH lxcfs] make meminfo and swaps cgroupv2 aware
	5
	6	and deduplicate the corresponding code
	7
	8	Signed-off-by: Wolfgang Bumiller <w.bumiller@proxmox.com>
	9	---
	10	src/bindings.c \| 10 ++++++
	11	src/bindings.h \| 1 +
	12	src/proc_fuse.c \| 89 +++++++++++++++++++++++++------------------------
	13	3 files changed, 57 insertions(+), 43 deletions(-)
	14
	15	diff --git a/src/bindings.c b/src/bindings.c
47819977	16	index eb62ddb..e417a13 100644
62c5f3ad WB	17	--- a/src/bindings.c
62c5f3ad WB	18	+++ b/src/bindings.c
47819977 WB	19	@@ -44,6 +44,7 @@ static bool can_use_pidfd;
	20	static bool can_use_swap;
	21	static bool can_use_sys_cpu;
	22	static bool has_versioned_opts;
62c5f3ad WB	23	+static bool memory_is_cgroupv2;
	24
	25	static volatile sig_atomic_t reload_successful;
	26
47819977 WB	27	@@ -67,6 +68,11 @@ bool liblxcfs_has_versioned_opts(void)
47819977 WB	28	return has_versioned_opts;
62c5f3ad WB	29	}
	30
	31	+bool liblxcfs_memory_is_cgroupv2(void)
	32	+{
	33	+ return memory_is_cgroupv2;
	34	+}
	35	+
	36	/* Define pivot_root() if missing from the C library */
	37	#ifndef HAVE_PIVOT_ROOT
	38	static int pivot_root(const char new_root, const char put_old)
47819977	39	@@ -837,6 +843,7 @@ static void __attribute__((constructor)) lxcfs_init(void)
62c5f3ad WB	40	pidfd = -EBADF;
	41	int i = 0;
	42	pid_t pid;
	43	+ struct hierarchy *hierarchy;
	44
	45	lxcfs_info("Running constructor %s to reload liblxcfs", __func__);
	46
47819977 WB	47	@@ -894,6 +901,9 @@ static void __attribute__((constructor)) lxcfs_init(void)
	48	else
	49	lxcfs_info("Kernel does not support swap accounting");
62c5f3ad WB	50
	51	+ hierarchy = cgroup_ops->get_hierarchy(cgroup_ops, "memory");
	52	+ memory_is_cgroupv2 = hierarchy && is_unified_hierarchy(hierarchy);
	53	+
	54	lxcfs_info("api_extensions:");
47819977 WB	55	for (size_t nr = 0; nr < nr_api_extensions; nr++)
47819977 WB	56	lxcfs_info("- %s", api_extensions[nr]);
62c5f3ad	57	diff --git a/src/bindings.h b/src/bindings.h
47819977	58	index cd7d466..2b7c8f4 100644
62c5f3ad WB	59	--- a/src/bindings.h
62c5f3ad WB	60	+++ b/src/bindings.h
47819977	61	@@ -98,6 +98,7 @@ extern void prune_init_slice(char *cg);
62c5f3ad WB	62	extern bool supports_pidfd(void);
62c5f3ad WB	63	extern bool liblxcfs_functional(void);
47819977	64	extern bool liblxcfs_can_use_swap(void);
62c5f3ad	65	+extern bool liblxcfs_memory_is_cgroupv2(void);
47819977 WB	66	extern bool liblxcfs_can_use_sys_cpu(void);
47819977 WB	67	extern bool liblxcfs_has_versioned_opts(void);
62c5f3ad	68
62c5f3ad	69	diff --git a/src/proc_fuse.c b/src/proc_fuse.c
47819977	70	index 94168c9..a07e8d6 100644
62c5f3ad WB	71	--- a/src/proc_fuse.c
62c5f3ad WB	72	+++ b/src/proc_fuse.c
47819977	73	@@ -299,6 +299,43 @@ static inline bool startswith(const char line, const char pref)
62c5f3ad WB	74	return strncmp(line, pref, strlen(pref)) == 0;
	75	}
	76
	77	+static void get_swap_info(const char *cgroup, uint64_t memlimit,
	78	+ uint64_t memusage, uint64_t *swtotal,
	79	+ uint64_t swusage, uint64_t memswpriority)
	80	+{
	81	+ __do_free char memswusage_str = NULL, memswpriority_str = NULL;
	82	+ size_t memswlimit = 0, memswusage = 0;
	83	+ int ret;
	84	+
	85	+ swtotal = swusage = 0;
	86	+ *memswpriority = 1;
	87	+
	88	+ memswlimit = get_min_memlimit(cgroup, true);
	89	+ if (memswlimit > 0) {
	90	+ ret = cgroup_ops->get_memory_swap_current(cgroup_ops, cgroup, &memswusage_str);
	91	+ if (ret < 0 \|\| safe_uint64(memswusage_str, &memswusage, 10) != 0)
	92	+ return;
	93	+
	94	+ if (liblxcfs_memory_is_cgroupv2()) {
	95	+ *swtotal = memswlimit / 1024;
	96	+ *swusage = memswusage / 1024;
	97	+ } else {
	98	+ if (memlimit > memswlimit)
	99	+ *swtotal = 0;
	100	+ else
	101	+ *swtotal = (memswlimit - memlimit) / 1024;
	102	+ if (memusage > memswusage \|\| swtotal == 0)
	103	+ *swusage = 0;
	104	+ else
	105	+ *swusage = (memswusage - memusage) / 1024;
	106	+ }
	107	+
	108	+ ret = cgroup_ops->get_memory_swappiness(cgroup_ops, cgroup, &memswpriority_str);
	109	+ if (ret >= 0)
	110	+ safe_uint64(memswpriority_str, memswpriority, 10);
	111	+ }
	112	+}
	113	+
	114	static int proc_swaps_read(char *buf, size_t size, off_t offset,
	115	struct fuse_file_info *fi)
	116	{
47819977	117	@@ -307,7 +344,7 @@ static int proc_swaps_read(char *buf, size_t size, off_t offset,
62c5f3ad	118	struct fuse_context *fc = fuse_get_context();
47819977	119	bool wants_swap = lxcfs_has_opt(fuse_get_context()->private_data, LXCFS_SWAP_ON);
62c5f3ad WB	120	struct file_info *d = INTTYPE_TO_PTR(fi->fh);
	121	- uint64_t memswlimit = 0, memlimit = 0, memusage = 0, memswusage = 0,
	122	+ uint64_t memlimit = 0, memusage = 0,
	123	swtotal = 0, swusage = 0, memswpriority = 1,
	124	hostswtotal = 0, hostswfree = 0;
	125	ssize_t total_len = 0;
47819977	126	@@ -353,26 +390,8 @@ static int proc_swaps_read(char *buf, size_t size, off_t offset,
62c5f3ad WB	127	if (safe_uint64(memusage_str, &memusage, 10) < 0)
	128	lxcfs_error("Failed to convert memusage %s", memusage_str);
	129
	130	- if (wants_swap) {
	131	- memswlimit = get_min_memlimit(cgroup, true);
	132	- if (memswlimit > 0) {
	133	- ret = cgroup_ops->get_memory_swap_current(cgroup_ops, cgroup, &memswusage_str);
	134	- if (ret >= 0 && safe_uint64(memswusage_str, &memswusage, 10) == 0) {
	135	- if (memlimit > memswlimit)
	136	- swtotal = 0;
	137	- else
	138	- swtotal = (memswlimit - memlimit) / 1024;
	139	- if (memusage > memswusage \|\| swtotal == 0)
	140	- swusage = 0;
	141	- else
	142	- swusage = (memswusage - memusage) / 1024;
	143	- }
	144	-
	145	- ret = cgroup_ops->get_memory_swappiness(cgroup_ops, cgroup, &memswpriority_str);
	146	- if (ret >= 0)
	147	- safe_uint64(memswpriority_str, &memswpriority, 10);
	148	- }
	149	- }
	150	+ if (wants_swap)
	151	+ get_swap_info(cgroup, memlimit, memusage, &swtotal, &swusage, &memswpriority);
	152
	153	total_len = snprintf(d->buf, d->size, "Filename\t\t\t\tType\t\tSize\tUsed\tPriority\n");
	154
47819977	155	@@ -1141,7 +1160,7 @@ static int proc_meminfo_read(char *buf, size_t size, off_t offset,
62c5f3ad	156	struct fuse_context *fc = fuse_get_context();
47819977	157	bool wants_swap = lxcfs_has_opt(fuse_get_context()->private_data, LXCFS_SWAP_ON);
62c5f3ad WB	158	struct file_info *d = INTTYPE_TO_PTR(fi->fh);
	159	- uint64_t memlimit = 0, memusage = 0, memswlimit = 0, memswusage = 0,
	160	+ uint64_t memlimit = 0, memusage = 0,
	161	hosttotal = 0, swfree = 0, swusage = 0, swtotal = 0,
	162	memswpriority = 1;
	163	struct memory_stat mstat = {};
47819977	164	@@ -1193,26 +1212,8 @@ static int proc_meminfo_read(char *buf, size_t size, off_t offset,
62c5f3ad WB	165	* Following values are allowed to fail, because swapaccount might be
	166	* turned off for current kernel.
	167	*/
	168	- if (wants_swap) {
	169	- memswlimit = get_min_memlimit(cgroup, true);
	170	- if (memswlimit > 0) {
	171	- ret = cgroup_ops->get_memory_swap_current(cgroup_ops, cgroup, &memswusage_str);
	172	- if (ret >= 0 && safe_uint64(memswusage_str, &memswusage, 10) == 0) {
	173	- if (memlimit > memswlimit)
	174	- swtotal = 0;
	175	- else
	176	- swtotal = (memswlimit - memlimit) / 1024;
	177	- if (memusage > memswusage \|\| swtotal == 0)
	178	- swusage = 0;
	179	- else
	180	- swusage = (memswusage - memusage) / 1024;
	181	- }
	182	- }
	183	-
	184	- ret = cgroup_ops->get_memory_swappiness(cgroup_ops, cgroup, &memswpriority_str);
	185	- if (ret >= 0)
	186	- safe_uint64(memswpriority_str, &memswpriority, 10);
	187	- }
	188	+ if (wants_swap)
	189	+ get_swap_info(cgroup, memlimit, memusage, &swtotal, &swusage, &memswpriority);
	190
	191	f = fopen_cached("/proc/meminfo", "re", &fopen_cache);
	192	if (!f)
47819977	193	@@ -1251,7 +1252,9 @@ static int proc_meminfo_read(char *buf, size_t size, off_t offset,
62c5f3ad WB	194	This is because the kernel can swap as much as it
	195	wants and not only up to swtotal. */
	196
	197	- swtotal = memlimit + swtotal;
	198	+ if (!liblxcfs_memory_is_cgroupv2())
	199	+ swtotal += memlimit;
	200	+
	201	if (hostswtotal < swtotal) {
	202	swtotal = hostswtotal;
	203	}