VM: readd support for forgetting cached FS blocks

Not all services involved in block I/O go through VM to access the
blocks they need.  As a result, the blocks in VM may become stale,
possibly causing corruption when the stale copy is restored by a
service that does go through VM later on.  This patch restores support
for forgetting cached blocks that belong to a particular device, and
makes the relevant file systems use this functionality 1) when
requested by VFS through REQ_FLUSH, and 2) upon unmount.

Change-Id: I0758c5ed8fe4b5ba81d432595d2113175776aff8

commands/service/parse.c

@@ -734,6 +734,7 @@ struct
 	{ "PROCCTL",		VM_PROCCTL },
 	{ "MAPCACHEPAGE",	VM_MAPCACHEPAGE },
 	{ "SETCACHEPAGE",	VM_SETCACHEPAGE },
+	{ "CLEARCACHE",		VM_CLEARCACHE },
 	{ "VFS_MMAP",		VM_VFS_MMAP },
 	{ "VFS_REPLY",		VM_VFS_REPLY },
 	{ NULL,			0 },

etc/system.conf

@@ -110,7 +110,7 @@ service mfs
 {
 	ipc	ALL_SYS;	# All system ipc targets allowed
 	system	BASIC;		# Only basic kernel calls allowed
-	vm	MAPCACHEPAGE SETCACHEPAGE;
+	vm	MAPCACHEPAGE SETCACHEPAGE CLEARCACHE;
 	io	NONE;		# No I/O range allowed
 	irq	NONE;		# No IRQ allowed
 	sigmgr          rs;	# Signal manager is RS
@@ -137,7 +137,7 @@ service ext2
 {
 	ipc	ALL_SYS;	# All system ipc targets allowed
 	system	BASIC;		# Only basic kernel calls allowed
-	vm	MAPCACHEPAGE SETCACHEPAGE;
+	vm	MAPCACHEPAGE SETCACHEPAGE CLEARCACHE;
 	io	NONE;		# No I/O range allowed
 	irq	NONE;		# No IRQ allowed
 	sigmgr          rs;	# Signal manager is RS
@@ -150,7 +150,7 @@ service pfs
 {
 	ipc	ALL_SYS;	# All system ipc targets allowed
 	system	BASIC;		# Only basic kernel calls allowed
-	vm	MAPCACHEPAGE SETCACHEPAGE;
+	vm	MAPCACHEPAGE SETCACHEPAGE CLEARCACHE;
 	io	NONE;		# No I/O range allowed
 	irq	NONE;		# No IRQ allowed
 	sigmgr          rs;	# Signal manager is RS

include/minix/com.h

@@ -978,6 +978,9 @@
 /* To VM: identify cache block in FS */
 #define VM_SETCACHEPAGE		(VM_RQ_BASE+27)
 
+/* To VM: clear all cache blocks for a device */
+#define VM_CLEARCACHE		(VM_RQ_BASE+28)
+
 /* To VFS: fields for request from VM. */
 #	define VFS_VMCALL_REQ		m10_i1
 #	define VFS_VMCALL_FD		m10_i2

include/minix/vm.h

@@ -72,6 +72,8 @@ int vm_set_cacheblock(void *block, u32_t dev, u64_t dev_offset,
 void *vm_map_cacheblock(u32_t dev, u64_t dev_offset,
         u64_t ino, u64_t ino_offset, u32_t *flags, int blocksize);
 
+int vm_clear_cache(u32_t dev);
+
 /* flags for vm cache functions */
 #define VMMC_FLAGS_LOCKED	0x01	/* someone is updating the flags; don't read/write */
 #define VMMC_DIRTY		0x02	/* dirty buffer and it may not be evicted */

lib/libminixfs/cache.c

@@ -577,6 +577,8 @@ void lmfs_invalidate(
 		bp->data = NULL;
 	}
   }
+
+  vm_clear_cache(device);
 }
 
 /*===========================================================================*

lib/libsys/vm_cache.c

@@ -60,3 +60,17 @@ int vm_set_cacheblock(void *block, u32_t dev, u64_t dev_offset,
 	return vm_cachecall(&m, VM_SETCACHEPAGE, block, dev, dev_offset,
 		ino, ino_offset, flags, blocksize);
 }
+
+int
+vm_clear_cache(u32_t dev)
+{
+	message m;
+
+	assert(dev != NO_DEV);
+
+	memset(&m, 0, sizeof(m));
+
+	m.m_u.m_vmmcp.dev = dev;
+
+	return _taskcall(VM_PROC_NR, VM_CLEARCACHE, &m);
+}

servers/ext2/mount.c

@@ -241,6 +241,9 @@ int fs_unmount()
   /* Close the device the file system lives on. */
   bdev_close(fs_dev);
 
+  /* Throw all blocks out of the VM cache, to prevent corruption later. */
+  lmfs_invalidate(fs_dev);
+
   /* Finish off the unmount. */
   superblock->s_dev = NO_DEV;
   unmountdone = TRUE;

servers/mfs/mount.c

@@ -187,6 +187,9 @@ int fs_unmount()
   /* Close the device the file system lives on. */
   bdev_close(fs_dev);
 
+  /* Throw out blocks out of the VM cache, to prevent corruption later. */
+  lmfs_invalidate(fs_dev);
+
   /* Finish off the unmount. */
   superblock.s_dev = NO_DEV;
   unmountdone = TRUE;

servers/vm/cache.c

@@ -304,6 +304,25 @@ int cache_freepages(int pages)
 	return freed;
 }
 
+/*
+ * Remove all pages that are associated with the given device.
+ */
+void
+clear_cache_bydev(dev_t dev)
+{
+	struct cached_page *cp, *ncp;
+	int h;
+
+	for (h = 0; h < HASHSIZE; h++) {
+		for (cp = cache_hash_bydev[h]; cp != NULL; cp = ncp) {
+			ncp = cp->hash_next_dev;
+
+			if (cp->dev == dev)
+				rmcache(cp);
+		}
+	}
+}
+
 void get_stats_info(struct vm_stats_info *vsi)
 {
         vsi->vsi_cached = cached_pages;

servers/vm/main.c

@@ -485,6 +485,7 @@ void init_vm(void)
 	/* Cache blocks. */
 	CALLMAP(VM_MAPCACHEPAGE, do_mapcache);
 	CALLMAP(VM_SETCACHEPAGE, do_setcache);
+	CALLMAP(VM_CLEARCACHE, do_clearcache);
 
 	/* getrusage */
 	CALLMAP(VM_GETRUSAGE, do_getrusage);

servers/vm/mem_cache.c

@@ -238,3 +238,17 @@ do_setcache(message *msg)
 	return OK;
 }
 
+/*
+ * A file system wants to invalidate all pages belonging to a certain device.
+ */
+int
+do_clearcache(message *msg)
+{
+	dev_t dev;
+
+	dev = msg->m_u.m_vmmcp.dev;
+
+	clear_cache_bydev(dev);
+
+	return OK;
+}

servers/vm/proto.h

@@ -218,6 +218,7 @@ void shared_setsource(struct vir_region *vr, endpoint_t ep, struct vir_region *s
 /* mem_cache.c */
 int do_mapcache(message *m);
 int do_setcache(message *m);
+int do_clearcache(message *m);
 
 /* cache.c */
 struct cached_page *find_cached_page_bydev(dev_t dev, u64_t dev_off,
@@ -229,6 +230,7 @@ int cache_freepages(int pages);
 void get_stats_info(struct vm_stats_info *vsi);
 void cache_lru_touch(struct cached_page *hb);
 void rmcache(struct cached_page *cp);
+void clear_cache_bydev(dev_t dev);
 
 /* vfs.c */
 int vfs_request(int reqno, int fd, struct vmproc *vmp, u64_t offset,

test/test72.c

@@ -224,16 +224,6 @@ panic(const char *fmt, ...)
 	exit(1);
 }
 
-int vm_forgetblock(u64_t id)
-{
-	return ENOSYS;
-}
-
-void vm_forgetblocks(void)
-{
-	return;
-}
-
 int
 vm_info_stats(struct vm_stats_info *vsi)
 {
@@ -274,6 +264,11 @@ void *vm_map_cacheblock(u32_t dev, u64_t dev_offset,
 	return MAP_FAILED;
 }
 
+int vm_clear_cache(u32_t dev)
+{
+	return 0;
+}
+
 int
 main(int argc, char *argv[])
 {

test/testvm.c

@@ -177,6 +177,8 @@ main(int argc, char *argv[])
 
 	writepipe(&info);
 
+	vm_clear_cache(MYDEV);
+
 	return 0;
 }
 

test/testvm.conf

@@ -1,7 +1,7 @@
 service testvm {
 	ipc	ALL;	# All system ipc targets allowed
 	system	BASIC;		# Only basic kernel calls allowed
-	vm      MAPCACHEPAGE SETCACHEPAGE;
+	vm      MAPCACHEPAGE SETCACHEPAGE CLEARCACHE;
 	io	NONE;		# No I/O range allowed
 	irq	NONE;		# No IRQ allowed
 	sigmgr          rs;	# Signal manager is RS