take fs_pin stuff to fs/*

Add a new field to fs_pin - kill(pin). That's what umount and r/o remount will be calling for all pins attached to vfsmount and superblock resp. Called after bumping the refcount, so it won't go away under us. Dropping the refcount is responsibility of the instance. All generic stuff moved to fs/fs_pin.c; the next step will rip all the knowledge of kernel/acct.c from fs/super.c and fs/namespace.c. After that - death to mnt_pin(); it was intended to be usable as generic mechanism for code that wants to attach objects to vfsmount, so that they would not make the sucker busy and would get killed on umount. Never got it right; it remained acct.c-specific all along. Now it's very close to being killable. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

take fs_pin stuff to fs/*
Add a new field to fs_pin - kill(pin). That's what umount and r/o remount will be calling for all pins attached to vfsmount and superblock resp. Called after bumping the refcount, so it won't go away under us. Dropping the refcount is responsibility of the instance. All generic stuff moved to fs/fs_pin.c; the next step will rip all the knowledge of kernel/acct.c from fs/super.c and fs/namespace.c. After that - death to mnt_pin(); it was intended to be usable as generic mechanism for code that wants to attach objects to vfsmount, so that they would not make the sucker busy and would get killed on umount. Never got it right; it remained acct.c-specific all along. Now it's very close to being killable. Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Al Viro
1 parent 1629d0eb3e
Showing 5 changed files with 129 additions and 100 deletions Side-by-side Diff
fs/Makefile
fs/fs_pin.c
include/linux/acct.h
include/linux/fs_pin.h
kernel/acct.c
@@ -11,7 +11,7 @@
 		attr.o bad_inode.o file.o filesystems.o namespace.o \
 		seq_file.o xattr.o libfs.o fs-writeback.o \
 		pnode.o splice.o sync.o utimes.o \
-		stack.o fs_struct.o statfs.o
+		stack.o fs_struct.o statfs.o fs_pin.o
  
 ifeq ($(CONFIG_BLOCK),y)
 obj-y +=	buffer.o block_dev.o direct-io.o mpage.o
+#include <linux/fs.h>
+#include <linux/slab.h>
+#include <linux/fs_pin.h>
+#include "mount.h"
+
+static void pin_free_rcu(struct rcu_head *head)
+{
+	kfree(container_of(head, struct fs_pin, rcu));
+}
+
+static DEFINE_SPINLOCK(pin_lock);
+
+void pin_put(struct fs_pin *p)
+{
+	if (atomic_long_dec_and_test(&p->count))
+		call_rcu(&p->rcu, pin_free_rcu);
+}
+
+void pin_remove(struct fs_pin *pin)
+{
+	spin_lock(&pin_lock);
+	hlist_del(&pin->m_list);
+	hlist_del(&pin->s_list);
+	spin_unlock(&pin_lock);
+}
+
+void pin_insert(struct fs_pin *pin, struct vfsmount *m)
+{
+	spin_lock(&pin_lock);
+	hlist_add_head(&pin->s_list, &m->mnt_sb->s_pins);
+	hlist_add_head(&pin->m_list, &real_mount(m)->mnt_pins);
+	spin_unlock(&pin_lock);
+}
+
+void acct_auto_close_mnt(struct hlist_head *list)
+{
+	while (1) {
+		struct hlist_node *p;
+		struct fs_pin *pin;
+		rcu_read_lock();
+		p = ACCESS_ONCE(list->first);
+		if (!p) {
+			rcu_read_unlock();
+			break;
+		}
+		pin = hlist_entry(p, struct fs_pin, m_list);
+		if (!atomic_long_inc_not_zero(&pin->count)) {
+			rcu_read_unlock();
+			cpu_relax();
+			continue;
+		}
+		rcu_read_unlock();
+		pin->kill(pin);
+	}
+}
+
+void acct_auto_close(struct hlist_head *list)
+{
+	while (1) {
+		struct hlist_node *p;
+		struct fs_pin *pin;
+		rcu_read_lock();
+		p = ACCESS_ONCE(list->first);
+		if (!p) {
+			rcu_read_unlock();
+			break;
+		}
+		pin = hlist_entry(p, struct fs_pin, s_list);
+		if (!atomic_long_inc_not_zero(&pin->count)) {
+			rcu_read_unlock();
+			cpu_relax();
+			continue;
+		}
+		rcu_read_unlock();
+		pin->kill(pin);
+	}
+}
@@ -24,18 +24,16 @@
 struct pacct_struct;
 struct pid_namespace;
 extern int acct_parm[]; /* for sysctl */
-extern void acct_auto_close(struct hlist_head *);
-extern void acct_auto_close_mnt(struct hlist_head *);
 extern void acct_collect(long exitcode, int group_dead);
 extern void acct_process(void);
 extern void acct_exit_ns(struct pid_namespace *);
 #else
-#define acct_auto_close(x)	do { } while (0)
-#define acct_auto_close_mnt(x)	do { } while (0)
 #define acct_collect(x,y)	do { } while (0)
 #define acct_process()		do { } while (0)
 #define acct_exit_ns(ns)	do { } while (0)
 #endif
+extern void acct_auto_close(struct hlist_head *);
+extern void acct_auto_close_mnt(struct hlist_head *);
  
 /*
  * ACCT_VERSION numbers as yet defined:
+#include <linux/fs.h>
+
+struct fs_pin {
+	atomic_long_t		count;
+	union {
+		struct {
+			struct hlist_node	s_list;
+			struct hlist_node	m_list;
+		};
+		struct rcu_head rcu;
+	};
+	void (*kill)(struct fs_pin *);
+};
+
+void pin_put(struct fs_pin *);
+void pin_remove(struct fs_pin *);
+void pin_insert(struct fs_pin *, struct vfsmount *);
@@ -59,7 +59,7 @@
 #include <asm/div64.h>
 #include <linux/blkdev.h> /* sector_div */
 #include <linux/pid_namespace.h>
-#include <../fs/mount.h>	/* will go away when we refactor */
+#include <linux/fs_pin.h>
  
 /*
  * These constants control the amount of freespace that suspend and
@@ -78,17 +78,6 @@
  */
 static void do_acct_process(struct bsd_acct_struct *acct);
  
-struct fs_pin {
-	atomic_long_t		count;
-	union {
-		struct {
-			struct hlist_node	s_list;
-			struct hlist_node	m_list;
-		};
-		struct rcu_head rcu;
-	};
-};
-
 struct bsd_acct_struct {
 	struct fs_pin		pin;
 	struct mutex		lock;
@@ -100,13 +89,6 @@
 	struct completion	done;
 };
  
-static void pin_free_rcu(struct rcu_head *head)
-{
-	kfree(container_of(head, struct fs_pin, rcu));
-}
-
-static DEFINE_SPINLOCK(acct_lock);
-
 /*
  * Check the amount of free space and suspend/resume accordingly.
  */
@@ -142,29 +124,6 @@
 	return acct->active;
 }
  
-static void pin_put(struct fs_pin *p)
-{
-	if (atomic_long_dec_and_test(&p->count))
-		call_rcu(&p->rcu, pin_free_rcu);
-}
-
-static struct bsd_acct_struct *__acct_get(struct bsd_acct_struct *res)
-{
-	if (!atomic_long_inc_not_zero(&res->pin.count)) {
-		rcu_read_unlock();
-		cpu_relax();
-		return NULL;
-	}
-	rcu_read_unlock();
-	mutex_lock(&res->lock);
-	if (!res->ns) {
-		mutex_unlock(&res->lock);
-		pin_put(&res->pin);
-		return NULL;
-	}
-	return res;
-}
-
 static struct bsd_acct_struct *acct_get(struct pid_namespace *ns)
 {
 	struct bsd_acct_struct *res;
  
@@ -176,9 +135,18 @@
 		rcu_read_unlock();
 		return NULL;
 	}
-	res = __acct_get(res);
-	if (!res)
+	if (!atomic_long_inc_not_zero(&res->pin.count)) {
+		rcu_read_unlock();
+		cpu_relax();
 		goto again;
+	}
+	rcu_read_unlock();
+	mutex_lock(&res->lock);
+	if (!res->ns) {
+		mutex_unlock(&res->lock);
+		pin_put(&res->pin);
+		goto again;
+	}
 	return res;
 }
  
  
@@ -203,19 +171,8 @@
 		init_completion(&acct->done);
 		schedule_work(&acct->work);
 		wait_for_completion(&acct->done);
-		spin_lock(&acct_lock);
-		hlist_del(&acct->pin.m_list);
-		hlist_del(&acct->pin.s_list);
-		spin_unlock(&acct_lock);
+		pin_remove(&acct->pin);
 		ns->bacct = new;
-		if (new) {
-			struct vfsmount *m = new->file->f_path.mnt;
-			spin_lock(&acct_lock);
-			hlist_add_head(&new->pin.s_list, &m->mnt_sb->s_pins);
-			hlist_add_head(&new->pin.m_list, &real_mount(m)->mnt_pins);
-			spin_unlock(&acct_lock);
-			mutex_unlock(&new->lock);
-		}
 		acct->ns = NULL;
 		atomic_long_dec(&acct->pin.count);
 		mutex_unlock(&acct->lock);
@@ -223,6 +180,19 @@
 	}
 }
  
+static void acct_pin_kill(struct fs_pin *pin)
+{
+	struct bsd_acct_struct *acct;
+	acct = container_of(pin, struct bsd_acct_struct, pin);
+	mutex_lock(&acct->lock);
+	if (!acct->ns) {
+		mutex_unlock(&acct->lock);
+		pin_put(pin);
+		acct = NULL;
+	}
+	acct_kill(acct, NULL);
+}
+
 static int acct_on(struct filename *pathname)
 {
 	struct file *file;
  
  
  
  
@@ -254,25 +224,22 @@
 	}
  
 	atomic_long_set(&acct->pin.count, 1);
+	acct->pin.kill = acct_pin_kill;
 	acct->file = file;
 	acct->needcheck = jiffies;
 	acct->ns = ns;
 	mutex_init(&acct->lock);
 	mnt = file->f_path.mnt;
 	mnt_pin(mnt);
+	mutex_lock_nested(&acct->lock, 1);	/* nobody has seen it yet */
+	pin_insert(&acct->pin, mnt);
  
 	old = acct_get(ns);
-	mutex_lock_nested(&acct->lock, 1);	/* nobody has seen it yet */
-	if (old) {
+	if (old)
 		acct_kill(old, acct);
-	} else {
+	else
 		ns->bacct = acct;
-		spin_lock(&acct_lock);
-		hlist_add_head(&acct->pin.s_list, &mnt->mnt_sb->s_pins);
-		hlist_add_head(&acct->pin.m_list, &real_mount(mnt)->mnt_pins);
-		spin_unlock(&acct_lock);
-		mutex_unlock(&acct->lock);
-	}
+	mutex_unlock(&acct->lock);
 	mntput(mnt); /* it's pinned, now give up active reference */
 	return 0;
 }
@@ -310,36 +277,6 @@
 	}
  
 	return error;
-}
-
-void acct_auto_close_mnt(struct hlist_head *list)
-{
-	rcu_read_lock();
-	while (1) {
-		struct hlist_node *p = ACCESS_ONCE(list->first);
-		if (!p)
-			break;
-		acct_kill(__acct_get(hlist_entry(p,
-						 struct bsd_acct_struct,
-						 pin.m_list)), NULL);
-		rcu_read_lock();
-	}
-	rcu_read_unlock();
-}
-
-void acct_auto_close(struct hlist_head *list)
-{
-	rcu_read_lock();
-	while (1) {
-		struct hlist_node *p = ACCESS_ONCE(list->first);
-		if (!p)
-			break;
-		acct_kill(__acct_get(hlist_entry(p,
-						 struct bsd_acct_struct,
-						 pin.s_list)), NULL);
-		rcu_read_lock();
-	}
-	rcu_read_unlock();
 }
  
 void acct_exit_ns(struct pid_namespace *ns)
...	...	@@ -11,7 +11,7 @@
11	11	attr.o bad_inode.o file.o filesystems.o namespace.o \
12	12	seq_file.o xattr.o libfs.o fs-writeback.o \
13	13	pnode.o splice.o sync.o utimes.o \
14		- stack.o fs_struct.o statfs.o
	14	+ stack.o fs_struct.o statfs.o fs_pin.o
15	15
16	16	ifeq ($(CONFIG_BLOCK),y)
17	17	obj-y += buffer.o block_dev.o direct-io.o mpage.o
	1	+#include <linux/fs.h>
	2	+#include <linux/slab.h>
	3	+#include <linux/fs_pin.h>
	4	+#include "mount.h"
	5	+
	6	+static void pin_free_rcu(struct rcu_head *head)
	7	+{
	8	+ kfree(container_of(head, struct fs_pin, rcu));
	9	+}
	10	+
	11	+static DEFINE_SPINLOCK(pin_lock);
	12	+
	13	+void pin_put(struct fs_pin *p)
	14	+{
	15	+ if (atomic_long_dec_and_test(&p->count))
	16	+ call_rcu(&p->rcu, pin_free_rcu);
	17	+}
	18	+
	19	+void pin_remove(struct fs_pin *pin)
	20	+{
	21	+ spin_lock(&pin_lock);
	22	+ hlist_del(&pin->m_list);
	23	+ hlist_del(&pin->s_list);
	24	+ spin_unlock(&pin_lock);
	25	+}
	26	+
	27	+void pin_insert(struct fs_pin pin, struct vfsmount m)
	28	+{
	29	+ spin_lock(&pin_lock);
	30	+ hlist_add_head(&pin->s_list, &m->mnt_sb->s_pins);
	31	+ hlist_add_head(&pin->m_list, &real_mount(m)->mnt_pins);
	32	+ spin_unlock(&pin_lock);
	33	+}
	34	+
	35	+void acct_auto_close_mnt(struct hlist_head *list)
	36	+{
	37	+ while (1) {
	38	+ struct hlist_node *p;
	39	+ struct fs_pin *pin;
	40	+ rcu_read_lock();
	41	+ p = ACCESS_ONCE(list->first);
	42	+ if (!p) {
	43	+ rcu_read_unlock();
	44	+ break;
	45	+ }
	46	+ pin = hlist_entry(p, struct fs_pin, m_list);
	47	+ if (!atomic_long_inc_not_zero(&pin->count)) {
	48	+ rcu_read_unlock();
	49	+ cpu_relax();
	50	+ continue;
	51	+ }
	52	+ rcu_read_unlock();
	53	+ pin->kill(pin);
	54	+ }
	55	+}
	56	+
	57	+void acct_auto_close(struct hlist_head *list)
	58	+{
	59	+ while (1) {
	60	+ struct hlist_node *p;
	61	+ struct fs_pin *pin;
	62	+ rcu_read_lock();
	63	+ p = ACCESS_ONCE(list->first);
	64	+ if (!p) {
	65	+ rcu_read_unlock();
	66	+ break;
	67	+ }
	68	+ pin = hlist_entry(p, struct fs_pin, s_list);
	69	+ if (!atomic_long_inc_not_zero(&pin->count)) {
	70	+ rcu_read_unlock();
	71	+ cpu_relax();
	72	+ continue;
	73	+ }
	74	+ rcu_read_unlock();
	75	+ pin->kill(pin);
	76	+ }
	77	+}
...	...	@@ -24,18 +24,16 @@
24	24	struct pacct_struct;
25	25	struct pid_namespace;
26	26	extern int acct_parm[]; /* for sysctl */
27		-extern void acct_auto_close(struct hlist_head *);
28		-extern void acct_auto_close_mnt(struct hlist_head *);
29	27	extern void acct_collect(long exitcode, int group_dead);
30	28	extern void acct_process(void);
31	29	extern void acct_exit_ns(struct pid_namespace *);
32	30	#else
33		-#define acct_auto_close(x) do { } while (0)
34		-#define acct_auto_close_mnt(x) do { } while (0)
35	31	#define acct_collect(x,y) do { } while (0)
36	32	#define acct_process() do { } while (0)
37	33	#define acct_exit_ns(ns) do { } while (0)
38	34	#endif
	35	+extern void acct_auto_close(struct hlist_head *);
	36	+extern void acct_auto_close_mnt(struct hlist_head *);
39	37
40	38	/*
41	39	* ACCT_VERSION numbers as yet defined:
	1	+#include <linux/fs.h>
	2	+
	3	+struct fs_pin {
	4	+ atomic_long_t count;
	5	+ union {
	6	+ struct {
	7	+ struct hlist_node s_list;
	8	+ struct hlist_node m_list;
	9	+ };
	10	+ struct rcu_head rcu;
	11	+ };
	12	+ void (kill)(struct fs_pin );
	13	+};
	14	+
	15	+void pin_put(struct fs_pin *);
	16	+void pin_remove(struct fs_pin *);
	17	+void pin_insert(struct fs_pin , struct vfsmount );
...	...	@@ -59,7 +59,7 @@
59	59	#include <asm/div64.h>
60	60	#include <linux/blkdev.h> /* sector_div */
61	61	#include <linux/pid_namespace.h>
62		-#include <../fs/mount.h> /* will go away when we refactor */
	62	+#include <linux/fs_pin.h>
63	63
64	64	/*
65	65	* These constants control the amount of freespace that suspend and
...	...	@@ -78,17 +78,6 @@
78	78	*/
79	79	static void do_acct_process(struct bsd_acct_struct *acct);
80	80
81		-struct fs_pin {
82		- atomic_long_t count;
83		- union {
84		- struct {
85		- struct hlist_node s_list;
86		- struct hlist_node m_list;
87		- };
88		- struct rcu_head rcu;
89		- };
90		-};
91		-
92	81	struct bsd_acct_struct {
93	82	struct fs_pin pin;
94	83	struct mutex lock;
...	...	@@ -100,13 +89,6 @@
100	89	struct completion done;
101	90	};
102	91
103		-static void pin_free_rcu(struct rcu_head *head)
104		-{
105		- kfree(container_of(head, struct fs_pin, rcu));
106		-}
107		-
108		-static DEFINE_SPINLOCK(acct_lock);
109		-
110	92	/*
111	93	* Check the amount of free space and suspend/resume accordingly.
112	94	*/
...	...	@@ -142,29 +124,6 @@
142	124	return acct->active;
143	125	}
144	126
145		-static void pin_put(struct fs_pin *p)
146		-{
147		- if (atomic_long_dec_and_test(&p->count))
148		- call_rcu(&p->rcu, pin_free_rcu);
149		-}
150		-
151		-static struct bsd_acct_struct __acct_get(struct bsd_acct_struct res)
152		-{
153		- if (!atomic_long_inc_not_zero(&res->pin.count)) {
154		- rcu_read_unlock();
155		- cpu_relax();
156		- return NULL;
157		- }
158		- rcu_read_unlock();
159		- mutex_lock(&res->lock);
160		- if (!res->ns) {
161		- mutex_unlock(&res->lock);
162		- pin_put(&res->pin);
163		- return NULL;
164		- }
165		- return res;
166		-}
167		-
168	127	static struct bsd_acct_struct acct_get(struct pid_namespace ns)
169	128	{
170	129	struct bsd_acct_struct *res;
171	130
...	...	@@ -176,9 +135,18 @@
176	135	rcu_read_unlock();
177	136	return NULL;
178	137	}
179		- res = __acct_get(res);
180		- if (!res)
	138	+ if (!atomic_long_inc_not_zero(&res->pin.count)) {
	139	+ rcu_read_unlock();
	140	+ cpu_relax();
181	141	goto again;
	142	+ }
	143	+ rcu_read_unlock();
	144	+ mutex_lock(&res->lock);
	145	+ if (!res->ns) {
	146	+ mutex_unlock(&res->lock);
	147	+ pin_put(&res->pin);
	148	+ goto again;
	149	+ }
182	150	return res;
183	151	}
184	152
185	153
...	...	@@ -203,19 +171,8 @@
203	171	init_completion(&acct->done);
204	172	schedule_work(&acct->work);
205	173	wait_for_completion(&acct->done);
206		- spin_lock(&acct_lock);
207		- hlist_del(&acct->pin.m_list);
208		- hlist_del(&acct->pin.s_list);
209		- spin_unlock(&acct_lock);
	174	+ pin_remove(&acct->pin);
210	175	ns->bacct = new;
211		- if (new) {
212		- struct vfsmount *m = new->file->f_path.mnt;
213		- spin_lock(&acct_lock);
214		- hlist_add_head(&new->pin.s_list, &m->mnt_sb->s_pins);
215		- hlist_add_head(&new->pin.m_list, &real_mount(m)->mnt_pins);
216		- spin_unlock(&acct_lock);
217		- mutex_unlock(&new->lock);
218		- }
219	176	acct->ns = NULL;
220	177	atomic_long_dec(&acct->pin.count);
221	178	mutex_unlock(&acct->lock);
...	...	@@ -223,6 +180,19 @@
223	180	}
224	181	}
225	182
	183	+static void acct_pin_kill(struct fs_pin *pin)
	184	+{
	185	+ struct bsd_acct_struct *acct;
	186	+ acct = container_of(pin, struct bsd_acct_struct, pin);
	187	+ mutex_lock(&acct->lock);
	188	+ if (!acct->ns) {
	189	+ mutex_unlock(&acct->lock);
	190	+ pin_put(pin);
	191	+ acct = NULL;
	192	+ }
	193	+ acct_kill(acct, NULL);
	194	+}
	195	+
226	196	static int acct_on(struct filename *pathname)
227	197	{
228	198	struct file *file;
229	199
230	200
231	201
232	202
...	...	@@ -254,25 +224,22 @@
254	224	}
255	225
256	226	atomic_long_set(&acct->pin.count, 1);
	227	+ acct->pin.kill = acct_pin_kill;
257	228	acct->file = file;
258	229	acct->needcheck = jiffies;
259	230	acct->ns = ns;
260	231	mutex_init(&acct->lock);
261	232	mnt = file->f_path.mnt;
262	233	mnt_pin(mnt);
	234	+ mutex_lock_nested(&acct->lock, 1); /* nobody has seen it yet */
	235	+ pin_insert(&acct->pin, mnt);
263	236
264	237	old = acct_get(ns);
265		- mutex_lock_nested(&acct->lock, 1); /* nobody has seen it yet */
266		- if (old) {
	238	+ if (old)
267	239	acct_kill(old, acct);
268		- } else {
	240	+ else
269	241	ns->bacct = acct;
270		- spin_lock(&acct_lock);
271		- hlist_add_head(&acct->pin.s_list, &mnt->mnt_sb->s_pins);
272		- hlist_add_head(&acct->pin.m_list, &real_mount(mnt)->mnt_pins);
273		- spin_unlock(&acct_lock);
274		- mutex_unlock(&acct->lock);
275		- }
	242	+ mutex_unlock(&acct->lock);
276	243	mntput(mnt); /* it's pinned, now give up active reference */
277	244	return 0;
278	245	}
...	...	@@ -310,36 +277,6 @@
310	277	}
311	278
312	279	return error;
313		-}
314		-
315		-void acct_auto_close_mnt(struct hlist_head *list)
316		-{
317		- rcu_read_lock();
318		- while (1) {
319		- struct hlist_node *p = ACCESS_ONCE(list->first);
320		- if (!p)
321		- break;
322		- acct_kill(__acct_get(hlist_entry(p,
323		- struct bsd_acct_struct,
324		- pin.m_list)), NULL);
325		- rcu_read_lock();
326		- }
327		- rcu_read_unlock();
328		-}
329		-
330		-void acct_auto_close(struct hlist_head *list)
331		-{
332		- rcu_read_lock();
333		- while (1) {
334		- struct hlist_node *p = ACCESS_ONCE(list->first);
335		- if (!p)
336		- break;
337		- acct_kill(__acct_get(hlist_entry(p,
338		- struct bsd_acct_struct,
339		- pin.s_list)), NULL);
340		- rcu_read_lock();
341		- }
342		- rcu_read_unlock();
343	280	}
344	281
345	282	void acct_exit_ns(struct pid_namespace *ns)