Skip to contentSkip to conversation

ipc: Save the ipc namespace while reading proc files.

PATCH In review Eric W. Biederman 2 participants 3 messages Feb 6, 2007 – Feb 10, 2007

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ipc: Save the ipc namespace while reading proc files.

The problem we were assuming that current->nsproxy->ipc_ns would never change while someone has our file in /proc/sysvipc/ file open. Given that this can change with both unshare and by passing the file descriptor to another process that assumption is occasionally wrong.

Therefore this patch causes /proc/sysvipc/* to cache the namespace and increment it's count when we open the file and to decrement the count when we close the file, ensuring consistent operation with no surprises.

---
 ipc/util.c |   58 +++++++++++++++++++++++++++++++++++++++++++++-------------
 1 files changed, 45 insertions(+), 13 deletions(-)
ipc/util.c +45 -13
index a9b7a22..7fd10f1 100644
@@ -739,14 +739,20 @@ int ipc_parse_version (int *cmd)
 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
 
 #ifdef CONFIG_PROC_FS
+struct ipc_proc_iter {
+	struct ipc_namespace *ns;
+	struct ipc_proc_iface *iface;
+};
+
 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
 {
-	struct ipc_proc_iface *iface = s->private;
+	struct ipc_proc_iter *iter = s->private;
+	struct ipc_proc_iface *iface = iter->iface;
 	struct kern_ipc_perm *ipc = it;
 	loff_t p;
 	struct ipc_ids *ids;
 
-	ids = current->nsproxy->ipc_ns->ids[iface->ids];
+	ids = iter->ns->ids[iface->ids];
 
 	/* If we had an ipc id locked before, unlock it */
 	if (ipc && ipc != SEQ_START_TOKEN)
@@ -773,12 +779,13 @@ static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
  */
 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
 {
-	struct ipc_proc_iface *iface = s->private;
+	struct ipc_proc_iter *iter = s->private;
+	struct ipc_proc_iface *iface = iter->iface;
 	struct kern_ipc_perm *ipc;
 	loff_t p;
 	struct ipc_ids *ids;
 
-	ids = current->nsproxy->ipc_ns->ids[iface->ids];
+	ids = iter->ns->ids[iface->ids];
 
 	/*
 	 * Take the lock - this will be released by the corresponding
@@ -807,21 +814,23 @@ static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
 static void sysvipc_proc_stop(struct seq_file *s, void *it)
 {
 	struct kern_ipc_perm *ipc = it;
-	struct ipc_proc_iface *iface = s->private;
+	struct ipc_proc_iter *iter = s->private;
+	struct ipc_proc_iface *iface = iter->iface;
 	struct ipc_ids *ids;
 
 	/* If we had a locked segment, release it */
 	if (ipc && ipc != SEQ_START_TOKEN)
 		ipc_unlock(ipc);
 
-	ids = current->nsproxy->ipc_ns->ids[iface->ids];
+	ids = iter->ns->ids[iface->ids];
 	/* Release the lock we took in start() */
 	mutex_unlock(&ids->mutex);
 }
 
 static int sysvipc_proc_show(struct seq_file *s, void *it)
 {
-	struct ipc_proc_iface *iface = s->private;
+	struct ipc_proc_iter *iter = s->private;
+	struct ipc_proc_iface *iface = iter->iface;
 
 	if (it == SEQ_START_TOKEN)
 		return seq_puts(s, iface->header);
@@ -836,22 +845,45 @@ static struct seq_operations sysvipc_proc_seqops = {
 	.show  = sysvipc_proc_show,
 };
 
-static int sysvipc_proc_open(struct inode *inode, struct file *file) {
+static int sysvipc_proc_open(struct inode *inode, struct file *file)
+{
 	int ret;
 	struct seq_file *seq;
+	struct ipc_proc_iter *iter;
+
+	ret = -ENOMEM;
+	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
+	if (!iter)
+		goto out;
 
 	ret = seq_open(file, &sysvipc_proc_seqops);
-	if (!ret) {
-		seq = file->private_data;
-		seq->private = PDE(inode)->data;
-	}
+	if (ret)
+		goto out_kfree;
+
+	seq = file->private_data;
+	seq->private = iter;
+
+	iter->iface = PDE(inode)->data;
+	iter->ns    = get_ipc_ns(current->nsproxy->ipc_ns);
+out:
 	return ret;
+out_kfree:
+	kfree(iter);
+	goto out;
+}
+
+static int sysvipc_proc_release(struct inode *inode, struct file *file)
+{
+	struct seq_file *seq = file->private_data;
+	struct ipc_proc_iter *iter = seq->private;
+	put_ipc_ns(iter->ns);
+	return seq_release_private(inode, file);
 }
 
 static struct file_operations sysvipc_proc_fops = {
 	.open    = sysvipc_proc_open,
 	.read    = seq_read,
 	.llseek  = seq_lseek,
-	.release = seq_release,
+	.release = sysvipc_proc_release,
 };
 #endif /* CONFIG_PROC_FS */
Signature · 1 line
1.4.4.1.g278f
#1 lore raw
Quoted 10 lines · Eric W. Biederman

The problem we were assuming that current->nsproxy->ipc_ns would never change while someone has our file in /proc/sysvipc/ file open. Given that this can change with both unshare and by passing the file descriptor to another process that assumption is occasionally wrong.

Therefore this patch causes /proc/sysvipc/* to cache the namespace and increment it's count when we open the file and to decrement the count when we close the file, ensuring consistent operation with no surprises.

Looks good, thanks.

Looks like mounts already does this with the mounts ns, and we don't need it for uts. Something to watch in any upcoming network patches though.

thanks
-serge

Quoted 136 lines
---
 ipc/util.c |   58 +++++++++++++++++++++++++++++++++++++++++++++-------------
 1 files changed, 45 insertions(+), 13 deletions(-)
ipc/util.c +45 -13
index a9b7a22..7fd10f1 100644
@@ -739,14 +739,20 @@ int ipc_parse_version (int *cmd)
 #endif /* __ARCH_WANT_IPC_PARSE_VERSION */
 
 #ifdef CONFIG_PROC_FS
+struct ipc_proc_iter {
+	struct ipc_namespace *ns;
+	struct ipc_proc_iface *iface;
+};
+
 static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
 {
-	struct ipc_proc_iface *iface = s->private;
+	struct ipc_proc_iter *iter = s->private;
+	struct ipc_proc_iface *iface = iter->iface;
 	struct kern_ipc_perm *ipc = it;
 	loff_t p;
 	struct ipc_ids *ids;
 
-	ids = current->nsproxy->ipc_ns->ids[iface->ids];
+	ids = iter->ns->ids[iface->ids];
 
 	/* If we had an ipc id locked before, unlock it */
 	if (ipc && ipc != SEQ_START_TOKEN)
@@ -773,12 +779,13 @@ static void *sysvipc_proc_next(struct seq_file *s, void *it, loff_t *pos)
  */
 static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
 {
-	struct ipc_proc_iface *iface = s->private;
+	struct ipc_proc_iter *iter = s->private;
+	struct ipc_proc_iface *iface = iter->iface;
 	struct kern_ipc_perm *ipc;
 	loff_t p;
 	struct ipc_ids *ids;
 
-	ids = current->nsproxy->ipc_ns->ids[iface->ids];
+	ids = iter->ns->ids[iface->ids];
 
 	/*
 	 * Take the lock - this will be released by the corresponding
@@ -807,21 +814,23 @@ static void *sysvipc_proc_start(struct seq_file *s, loff_t *pos)
 static void sysvipc_proc_stop(struct seq_file *s, void *it)
 {
 	struct kern_ipc_perm *ipc = it;
-	struct ipc_proc_iface *iface = s->private;
+	struct ipc_proc_iter *iter = s->private;
+	struct ipc_proc_iface *iface = iter->iface;
 	struct ipc_ids *ids;
 
 	/* If we had a locked segment, release it */
 	if (ipc && ipc != SEQ_START_TOKEN)
 		ipc_unlock(ipc);
 
-	ids = current->nsproxy->ipc_ns->ids[iface->ids];
+	ids = iter->ns->ids[iface->ids];
 	/* Release the lock we took in start() */
 	mutex_unlock(&ids->mutex);
 }
 
 static int sysvipc_proc_show(struct seq_file *s, void *it)
 {
-	struct ipc_proc_iface *iface = s->private;
+	struct ipc_proc_iter *iter = s->private;
+	struct ipc_proc_iface *iface = iter->iface;
 
 	if (it == SEQ_START_TOKEN)
 		return seq_puts(s, iface->header);
@@ -836,22 +845,45 @@ static struct seq_operations sysvipc_proc_seqops = {
 	.show  = sysvipc_proc_show,
 };
 
-static int sysvipc_proc_open(struct inode *inode, struct file *file) {
+static int sysvipc_proc_open(struct inode *inode, struct file *file)
+{
 	int ret;
 	struct seq_file *seq;
+	struct ipc_proc_iter *iter;
+
+	ret = -ENOMEM;
+	iter = kmalloc(sizeof(*iter), GFP_KERNEL);
+	if (!iter)
+		goto out;
 
 	ret = seq_open(file, &sysvipc_proc_seqops);
-	if (!ret) {
-		seq = file->private_data;
-		seq->private = PDE(inode)->data;
-	}
+	if (ret)
+		goto out_kfree;
+
+	seq = file->private_data;
+	seq->private = iter;
+
+	iter->iface = PDE(inode)->data;
+	iter->ns    = get_ipc_ns(current->nsproxy->ipc_ns);
+out:
 	return ret;
+out_kfree:
+	kfree(iter);
+	goto out;
+}
+
+static int sysvipc_proc_release(struct inode *inode, struct file *file)
+{
+	struct seq_file *seq = file->private_data;
+	struct ipc_proc_iter *iter = seq->private;
+	put_ipc_ns(iter->ns);
+	return seq_release_private(inode, file);
 }
 
 static struct file_operations sysvipc_proc_fops = {
 	.open    = sysvipc_proc_open,
 	.read    = seq_read,
 	.llseek  = seq_lseek,
-	.release = seq_release,
+	.release = sysvipc_proc_release,
 };
 #endif /* CONFIG_PROC_FS */
Signature · 7 lines
1.4.4.1.g278f

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/
#2 lore raw
Quoted 6 lines · Serge E. Hallyn

Looks good, thanks.

Looks like mounts already does this with the mounts ns, and we don't need it for uts. Something to watch in any upcoming network patches though.

Which is where I got the idea. :)

I did this so many times in my network namespace patches when I looked at util.c it was blinding obvious...

Thanks for the review of the other obvious candidates. I fixed this one because I tripped over it, when fixing the ref counting bug.

Eric