Skip to contentSkip to conversation

cgroup map files: Add a key/value map file type to cgroups

PATCH In review [email protected] 3 participants 8 messages Feb 21, 2008 – Feb 23, 2008

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover cgroup map files: Add a key/value map file type to cgroups

[ Updated from the previous version to remove the colon from the map output ]

These patches add a new cgroup control file output type - a map from strings to u64 values - and make use of it for the memory controller "stat" file.

It is intended for use when the subsystem wants to return a collection of values that are related in some way, for which a separate control file for each value would make the reporting unwieldy.

The advantages of this are:

- more standardized output from control files that report similarly-structured data that needs to be parsed programmatically

- less boilerplate required in cgroup subsystems

- simplifies transition to a future efficient cgroups binary API

--

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#1 lore raw

PATCH 1/2 cgroup map files: Add cgroup map data type

Adds a new type of supported control file representation, a map from strings to u64 values.

The map type is printed in a similar format to /proc/meminfo or /proc/<pid>/status, i.e. "$key: $value\n"

---
 include/linux/cgroup.h |   19 +++++++++++++++
 kernel/cgroup.c        |   59 ++++++++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 77 insertions(+), 1 deletion(-)
include/linux/cgroup.h +19 -0
Index: cgroupmap-2.6.25-rc2-mm1/include/linux/cgroup.h
===================================================================
@@ -166,6 +166,16 @@ struct css_set {
 
 };
 
+/*
+ * cgroup_map_cb is an abstract callback API for reporting map-valued
+ * control files
+ */
+
+struct cgroup_map_cb {
+	int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
+	void *state;
+};
+
 /* struct cftype:
  *
  * The files in the cgroup filesystem mostly have a very simple read/write
@@ -194,6 +204,15 @@ struct cftype {
 	 * single integer. Use it in place of read()
 	 */
 	u64 (*read_uint) (struct cgroup *cont, struct cftype *cft);
+	/*
+	 * read_map() is used for defining a map of key/value
+	 * pairs. It should call cb->fill(cb, key, value) for each
+	 * entry. The key/value pairs (and their ordering) should not
+	 * change between reboots.
+	 */
+	int (*read_map) (struct cgroup *cont, struct cftype *cft,
+			 struct cgroup_map_cb *cb);
+
 	ssize_t (*write) (struct cgroup *cont, struct cftype *cft,
 			  struct file *file,
 			  const char __user *buf, size_t nbytes, loff_t *ppos);
kernel/cgroup.c +58 -1

77 lines not on this page · raw message

--

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#2 lore raw

Re: [PATCH 1/2] cgroup map files: Add cgroup map data type

Quoted 2 lines

The map type is printed in a similar format to /proc/meminfo or /proc/<pid>/status, i.e. "$key: $value\n"

this description doesn't seem to match with the code.

YAMAMOTO Takashi

kernel/cgroup.c static ssize_t cgroup_file_read(struct f · line 1504

Quoted 5 lines
kernel/cgroup.c +5 -0
+static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
+{
+	struct seq_file *sf = cb->state;
+	return seq_printf(sf, "%s %llu\n", key, value);
+}

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

maintainer
#3 lore raw

Re: [PATCH 1/2] cgroup map files: Add cgroup map data type

In reply to [email protected]
Quoted 63 lines · On Thu, 21 Feb 2008 13:28:55 -0800 [email protected] wrote:

Adds a new type of supported control file representation, a map from strings to u64 values.

The map type is printed in a similar format to /proc/meminfo or /proc/<pid>/status, i.e. "$key: $value\n"

---
 include/linux/cgroup.h |   19 +++++++++++++++
 kernel/cgroup.c        |   59 ++++++++++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 77 insertions(+), 1 deletion(-)

Index: cgroupmap-2.6.25-rc2-mm1/include/linux/cgroup.h
===================================================================

include/linux/cgroup.h +19 -0
@@ -166,6 +166,16 @@ struct css_set {
 
 };
 
+/*
+ * cgroup_map_cb is an abstract callback API for reporting map-valued
+ * control files
+ */
+
+struct cgroup_map_cb {
+	int (*fill)(struct cgroup_map_cb *cb, const char *key, u64 value);
+	void *state;
+};
+
 /* struct cftype:
  *
  * The files in the cgroup filesystem mostly have a very simple read/write
@@ -194,6 +204,15 @@ struct cftype {
 	 * single integer. Use it in place of read()
 	 */
 	u64 (*read_uint) (struct cgroup *cont, struct cftype *cft);
+	/*
+	 * read_map() is used for defining a map of key/value
+	 * pairs. It should call cb->fill(cb, key, value) for each
+	 * entry. The key/value pairs (and their ordering) should not
+	 * change between reboots.
+	 */
+	int (*read_map) (struct cgroup *cont, struct cftype *cft,
+			 struct cgroup_map_cb *cb);
+
 	ssize_t (*write) (struct cgroup *cont, struct cftype *cft,
 			  struct file *file,
 			  const char __user *buf, size_t nbytes, loff_t *ppos);

Index: cgroupmap-2.6.25-rc2-mm1/kernel/cgroup.c
===================================================================

kernel/cgroup.c +5 -0
@@ -1487,6 +1487,46 @@ static ssize_t cgroup_file_read(struct f
 	return -EINVAL;
 }
 
+/*
+ * seqfile ops/methods for returning structured data. Currently just
+ * supports string->u64 maps, but can be extended in future.
+ */
+

kernel/cgroup.c static ssize_t cgroup_file_read(struct f · line 1504

Quoted 10 lines
kernel/cgroup.c +10 -0
+struct cgroup_seqfile_state {
+	struct cftype *cft;
+	struct cgroup *cgroup;
+};
+
+static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
+{
+	struct seq_file *sf = cb->state;
+	return seq_printf(sf, "%s %llu\n", key, value);
+}

We don't know what type the architecture uses to implement u64. This will warn on powerpc, sparc64, maybe others.

kernel/cgroup.c static ssize_t cgroup_file_read(struct f · line 1517

Quoted 12 lines
kernel/cgroup.c +12 -0
+static int cgroup_seqfile_show(struct seq_file *m, void *arg)
+{
+	struct cgroup_seqfile_state *state = m->private;
+	struct cftype *cft = state->cft;
+	if (cft->read_map) {
+		struct cgroup_map_cb cb = {
+			.fill = cgroup_map_add,
+			.state = m,
+		};
+		return cft->read_map(state->cgroup, cft, &cb);
+	} else {
+		BUG();

That's not really needed. Just call cft->read_map unconditionally. if it's zero we'll get a null-pointer deref which will have just the same effect as a BUG.

kernel/cgroup.c static ssize_t cgroup_file_read(struct f · line 1528

Quoted 11 lines
kernel/cgroup.c +11 -0
+	}
+}
+
+int cgroup_seqfile_release(struct inode *inode, struct file *file)
+{
+	struct seq_file *seq = file->private_data;
+	kfree(seq->private);
+	return single_release(inode, file);
+}
+
+static struct file_operations cgroup_seqfile_operations;

afaict you can just move the definition of cgroup_seqfile_operations here and avoid the forward decl.

kernel/cgroup.c static int cgroup_file_open(struct inode · line 1542

Quoted 9 lines
kernel/cgroup.c +1 -1
 static int cgroup_file_open(struct inode *inode, struct file *file)
 {
 	int err;
@@ -1499,7 +1539,18 @@ static int cgroup_file_open(struct inode
 	cft = __d_cft(file->f_dentry);
 	if (!cft)
 		return -ENODEV;
-	if (cft->open)
+	if (cft->read_map) {

But above a NULL value is illegal. Why are we testing it here?

Quoted 19 lines
kernel/cgroup.c +12 -0
+		struct cgroup_seqfile_state *state =
+			kzalloc(sizeof(*state), GFP_USER);
+		if (!state)
+			return -ENOMEM;
+		state->cft = cft;
+		state->cgroup = __d_cgrp(file->f_dentry->d_parent);
+		file->f_op = &cgroup_seqfile_operations;
+		err = single_open(file, cgroup_seqfile_show, state);
+		if (err < 0)
+			kfree(state);
+	} else if (cft->open)
 		err = cft->open(inode, file);
 	else
 		err = 0;
@@ -1538,6 +1589,12 @@ static struct file_operations cgroup_fil
 	.release = cgroup_file_release,
 };
 
+static struct file_operations cgroup_seqfile_operations = {

kernel/cgroup.c static struct file_operations cgroup_fil

Quoted 10 lines
kernel/cgroup.c +5 -1
+	.read = seq_read,
+	.llseek = seq_lseek,
+	.release = cgroup_seqfile_release,
+};
+
 static struct inode_operations cgroup_dir_inode_operations = {
 	.lookup = simple_lookup,
 	.mkdir = cgroup_mkdir,
 
--

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#4 lore raw

Re: [PATCH 1/2] cgroup map files: Add cgroup map data type

Quoted 8 lines · Andrew Morton

kernel/cgroup.c static ssize_t cgroup_file_read(struct f · line 1504

Quoted 5 lines
kernel/cgroup.c +5 -0
+static int cgroup_map_add(struct cgroup_map_cb *cb, const char *key, u64 value)
+{
+     struct seq_file *sf = cb->state;
+     return seq_printf(sf, "%s %llu\n", key, value);
+}

We don't know what type the architecture uses to implement u64. This will warn on powerpc, sparc64, maybe others.

OK, I'll add an (unsigned long long) cast.

Quoted 18 lines
Quoted 4 lines
kernel/cgroup.c +2 -0
+static int cgroup_seqfile_show(struct seq_file *m, void *arg)
+{

kernel/cgroup.c static ssize_t cgroup_file_read(struct f · line 1517

Quoted 10 lines
kernel/cgroup.c +10 -0
+     struct cgroup_seqfile_state *state = m->private;
+     struct cftype *cft = state->cft;
+     if (cft->read_map) {
+             struct cgroup_map_cb cb = {
+                     .fill = cgroup_map_add,
+                     .state = m,
+             };
+             return cft->read_map(state->cgroup, cft, &cb);
+     } else {
+             BUG();

That's not really needed. Just call cft->read_map unconditionally. if it's zero we'll get a null-pointer deref which will have just the same effect as a BUG.

OK. The long-term plan is to have other kinds of files also handled by this function, so eventually it would look something like:

if (cft->read_map) {
...
} else if (cft->read_something_else) {
...
}
...
} else {
BUG();
}

But I guess I can save that for the future.

Quoted 13 lines

kernel/cgroup.c static int cgroup_file_open(struct inode · line 1542

Quoted 9 lines

static int cgroup_file_open(struct inode *inode, struct file *file)

 {
      int err;
kernel/cgroup.c +1 -1
@@ -1499,7 +1539,18 @@ static int cgroup_file_open(struct inode
      cft = __d_cft(file->f_dentry);
      if (!cft)
              return -ENODEV;
-     if (cft->open)
+     if (cft->read_map) {

But above a NULL value is illegal. Why are we testing it here?

The existence of cft->read_map causes us to open a seq_file. Otherwise we do nothing special and carry on down the normal open path.

Paul

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#5 lore raw

PATCH 2/2 cgroup map files: Use cgroup map for memcontrol stats file

In reply to [email protected]

Remove the seq_file boilerplate used to construct the memcontrol stats map, and instead use the new map representation for cgroup control files

---
 mm/memcontrol.c |   30 ++++++------------------------
 1 file changed, 6 insertions(+), 24 deletions(-)
mm/memcontrol.c +6 -24
Index: cgroupmap-2.6.25-rc2-mm1/mm/memcontrol.c
===================================================================
@@ -974,9 +974,9 @@ static const struct mem_cgroup_stat_desc
 	[MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
 };
 
-static int mem_control_stat_show(struct seq_file *m, void *arg)
+static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
+				 struct cgroup_map_cb *cb)
 {
-	struct cgroup *cont = m->private;
 	struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
 	struct mem_cgroup_stat *stat = &mem_cont->stat;
 	int i;
@@ -986,8 +986,7 @@ static int mem_control_stat_show(struct 
 
 		val = mem_cgroup_read_stat(stat, i);
 		val *= mem_cgroup_stat_desc[i].unit;
-		seq_printf(m, "%s %lld\n", mem_cgroup_stat_desc[i].msg,
-				(long long)val);
+		cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
 	}
 	/* showing # of active pages */
 	{
@@ -997,29 +996,12 @@ static int mem_control_stat_show(struct 
 						MEM_CGROUP_ZSTAT_INACTIVE);
 		active = mem_cgroup_get_all_zonestat(mem_cont,
 						MEM_CGROUP_ZSTAT_ACTIVE);
-		seq_printf(m, "active %ld\n", (active) * PAGE_SIZE);
-		seq_printf(m, "inactive %ld\n", (inactive) * PAGE_SIZE);
+		cb->fill(cb, "active", (active) * PAGE_SIZE);
+		cb->fill(cb, "inactive", (inactive) * PAGE_SIZE);
 	}
 	return 0;
 }
 
-static const struct file_operations mem_control_stat_file_operations = {
-	.read = seq_read,
-	.llseek = seq_lseek,
-	.release = single_release,
-};
-
-static int mem_control_stat_open(struct inode *unused, struct file *file)
-{
-	/* XXX __d_cont */
-	struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
-
-	file->f_op = &mem_control_stat_file_operations;
-	return single_open(file, mem_control_stat_show, cont);
-}
-
-
-
 static struct cftype mem_cgroup_files[] = {
 	{
 		.name = "usage_in_bytes",
@@ -1044,7 +1026,7 @@ static struct cftype mem_cgroup_files[] 
 	},
 	{
 		.name = "stat",
-		.open = mem_control_stat_open,
+		.read_map = mem_control_stat_show,
 	},
 };
 

--

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

maintainer
#6 lore raw

Re: [PATCH 2/2] cgroup map files: Use cgroup map for memcontrol stats file

Quoted 37 lines · On Thu, 21 Feb 2008 13:28:56 -0800 [email protected] wrote:

Remove the seq_file boilerplate used to construct the memcontrol stats map, and instead use the new map representation for cgroup control files

---
 mm/memcontrol.c |   30 ++++++------------------------
 1 file changed, 6 insertions(+), 24 deletions(-)

Index: cgroupmap-2.6.25-rc2-mm1/mm/memcontrol.c
===================================================================

mm/memcontrol.c +3 -4
@@ -974,9 +974,9 @@ static const struct mem_cgroup_stat_desc
 	[MEM_CGROUP_STAT_RSS] = { "rss", PAGE_SIZE, },
 };
 
-static int mem_control_stat_show(struct seq_file *m, void *arg)
+static int mem_control_stat_show(struct cgroup *cont, struct cftype *cft,
+				 struct cgroup_map_cb *cb)
 {
-	struct cgroup *cont = m->private;
 	struct mem_cgroup *mem_cont = mem_cgroup_from_cont(cont);
 	struct mem_cgroup_stat *stat = &mem_cont->stat;
 	int i;
@@ -986,8 +986,7 @@ static int mem_control_stat_show(struct 
 
 		val = mem_cgroup_read_stat(stat, i);
 		val *= mem_cgroup_stat_desc[i].unit;
-		seq_printf(m, "%s %lld\n", mem_cgroup_stat_desc[i].msg,
-				(long long)val);
+		cb->fill(cb, mem_cgroup_stat_desc[i].msg, val);
 	}
 	/* showing # of active pages */
 	{

mm/memcontrol.c static int mem_control_stat_show(struct · line 1000

Quoted 7 lines
mm/memcontrol.c +2 -2
@@ -997,29 +996,12 @@ static int mem_control_stat_show(struct 
 						MEM_CGROUP_ZSTAT_INACTIVE);
 		active = mem_cgroup_get_all_zonestat(mem_cont,
 						MEM_CGROUP_ZSTAT_ACTIVE);
-		seq_printf(m, "active %ld\n", (active) * PAGE_SIZE);
-		seq_printf(m, "inactive %ld\n", (inactive) * PAGE_SIZE);
+		cb->fill(cb, "active", (active) * PAGE_SIZE);
+		cb->fill(cb, "inactive", (inactive) * PAGE_SIZE);

umm, afacit this might be a prior bug in mem_control_stat_show(). On a 32-bit machine can [in]active*PAGE_SIZE overflow 32-bits?

Quoted 25 lines
mm/memcontrol.c +0 -17
 	}
 	return 0;
 }
 
-static const struct file_operations mem_control_stat_file_operations = {
-	.read = seq_read,
-	.llseek = seq_lseek,
-	.release = single_release,
-};
-
-static int mem_control_stat_open(struct inode *unused, struct file *file)
-{
-	/* XXX __d_cont */
-	struct cgroup *cont = file->f_dentry->d_parent->d_fsdata;
-
-	file->f_op = &mem_control_stat_file_operations;
-	return single_open(file, mem_control_stat_show, cont);
-}
-
-
-
 static struct cftype mem_cgroup_files[] = {
 	{
 		.name = "usage_in_bytes",

mm/memcontrol.c static struct cftype mem_cgroup_files[]

Quoted 10 lines
mm/memcontrol.c +1 -2
@@ -1044,7 +1026,7 @@ static struct cftype mem_cgroup_files[] 
 	},
 	{
 		.name = "stat",
-		.open = mem_control_stat_open,
+		.read_map = mem_control_stat_show,
 	},
 };
 
 
--

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

maintainer
#7 lore raw
In reply to [email protected]
Quoted 3 lines · On Thu, 21 Feb 2008 13:28:54 -0800 [email protected] wrote:

These patches add a new cgroup control file output type - a map from strings to u64 values - and make use of it for the memory controller "stat" file.

Can we document the moderately obscure kernel->userspace interface somewhere please?

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>