Skip to contentSkip to conversation

Inode slimming

RFC PATCH In review Theodore Tso 14 participants 45 messages Jun 19, 2006 – Jun 22, 2006

Subsystems (47)
lore
Share & tools
lore
#0 lore raw

cover Inode slimming

The following patches reduce the size of struct inode. Unfortunately, since these structures are used by a large amount of kernel code, some of the patches are quite involved, and/or will require a lot of auditing and code review, for "only" 4 or 8 bytes at a time (maybe more on 64-bit platforms). However, since there are many, many copies of struct inode all over the kernel, even a small reduction in size can have a large beneficial result, and as the old Chinese saying goes, a journey of thousand miles begins with a single step....

What else remains to be done? There are a large number of fields in struct inode which are never populated unless the inode is open, and those should get moved into another structure which is populated only when needed. There are a large number of inodes which are read into memory only because stat(2) was called on them (thanks to things like color ls, et. al).

Linus has suggested moving the i_data structure out to a separate structure, again because there are many inodes which do not have any pages cached in the page cache. The challenge with this a huge number of codepaths assume that i_mapping is always non-NULL. But, i_data is *huge* so the benefits of not having it taking up memory would make this a high-return activity.

Another possibility is moving i_size into the file-specific area of the union, which would save 8 bytes. However there are a largish number block device drivers that seem to have hijacked i_size to store the blocksize(!?!) of the device, and that should really be done in a bdev-specific structure. Untangling this will be somewhat challenging, but should be doable.

--

#1 lore raw

RFC PATCH 1/8 inode_diet: Replace inode.u.generic_ip with inode.i_private

The filesystem or device-specific pointer in the inode is inside a union, which is pretty pointless given that all 30+ users of this field have been using the void pointer. Get rid of the union and rename it to i_private, with a comment to explain who is allowed to use the void pointer. This is just a cleanup, but it allows us to reuse the union 'u' for something something where the union will actually be used.

include/linux/fs.h +1 -3
Index: linux-2.6.17/include/linux/fs.h
===================================================================
@@ -534,9 +534,7 @@
 
 	atomic_t		i_writecount;
 	void			*i_security;
-	union {
-		void		*generic_ip;
-	} u;
+	void			*i_private; /* fs or device private pointer */
 #ifdef __NEED_I_SIZE_ORDERED
 	seqcount_t		i_size_seqcount;
 #endif
arch/powerpc/platforms/cell/spufs/inode.c +1 -1
Index: linux-2.6.17/arch/powerpc/platforms/cell/spufs/inode.c
===================================================================
@@ -120,7 +120,7 @@
 	ret = 0;
 	inode->i_op = &spufs_file_iops;
 	inode->i_fop = fops;
-	inode->u.generic_ip = SPUFS_I(inode)->i_ctx = get_spu_context(ctx);
+	inode->i_private = SPUFS_I(inode)->i_ctx = get_spu_context(ctx);
 	d_add(dentry, inode);
 out:
 	return ret;
arch/s390/kernel/debug.c +1 -1
Index: linux-2.6.17/arch/s390/kernel/debug.c
===================================================================
@@ -604,7 +604,7 @@
 	debug_info_t *debug_info, *debug_info_snapshot;
 
 	down(&debug_lock);
-	debug_info = (struct debug_info*)file->f_dentry->d_inode->u.generic_ip;
+	debug_info = (struct debug_info*)file->f_dentry->d_inode->i_private;
 	/* find debug view */
 	for (i = 0; i < DEBUG_MAX_VIEWS; i++) {
 		if (!debug_info->views[i])
drivers/i2c/chips/tps65010.c +1 -1
Index: linux-2.6.17/drivers/i2c/chips/tps65010.c
===================================================================
@@ -307,7 +307,7 @@
 
 static int dbg_tps_open(struct inode *inode, struct file *file)
 {
-	return single_open(file, dbg_show, inode->u.generic_ip);
+	return single_open(file, dbg_show, inode->i_private);
 }
 
 static struct file_operations debug_fops = {
drivers/infiniband/ulp/ipoib/ipoib_fs.c +2 -2
Index: linux-2.6.17/drivers/infiniband/ulp/ipoib/ipoib_fs.c
===================================================================
@@ -141,7 +141,7 @@
 		return ret;
 
 	seq = file->private_data;
-	seq->private = inode->u.generic_ip;
+	seq->private = inode->i_private;
 
 	return 0;
 }
@@ -247,7 +247,7 @@
 		return ret;
 
 	seq = file->private_data;
-	seq->private = inode->u.generic_ip;
+	seq->private = inode->i_private;
 
 	return 0;
 }
drivers/misc/ibmasm/ibmasmfs.c +8 -8
Index: linux-2.6.17/drivers/misc/ibmasm/ibmasmfs.c
===================================================================
@@ -174,7 +174,7 @@
 	}
 
 	inode->i_fop = fops;
-	inode->u.generic_ip = data;
+	inode->i_private = data;
 
 	d_add(dentry, inode);
 	return dentry;
@@ -243,7 +243,7 @@
 {
 	struct ibmasmfs_command_data *command_data;
 
-	if (!inode->u.generic_ip)
+	if (!inode->i_private)
 		return -ENODEV;
 
 	command_data = kmalloc(sizeof(struct ibmasmfs_command_data), GFP_KERNEL);
@@ -251,7 +251,7 @@
 		return -ENOMEM;
 
 	command_data->command = NULL;
-	command_data->sp = inode->u.generic_ip;
+	command_data->sp = inode->i_private;
 	file->private_data = command_data;
 	return 0;
 }
@@ -350,10 +350,10 @@
 	struct ibmasmfs_event_data *event_data;
 	struct service_processor *sp; 
 
-	if (!inode->u.generic_ip)
+	if (!inode->i_private)
 		return -ENODEV;
 
-	sp = inode->u.generic_ip;
+	sp = inode->i_private;
 
 	event_data = kmalloc(sizeof(struct ibmasmfs_event_data), GFP_KERNEL);
 	if (!event_data)
@@ -438,14 +438,14 @@
 {
 	struct ibmasmfs_heartbeat_data *rhbeat;
 
-	if (!inode->u.generic_ip)
+	if (!inode->i_private)
 		return -ENODEV;
 
 	rhbeat = kmalloc(sizeof(struct ibmasmfs_heartbeat_data), GFP_KERNEL);
 	if (!rhbeat)
 		return -ENOMEM;
 
-	rhbeat->sp = (struct service_processor *)inode->u.generic_ip;
+	rhbeat->sp = (struct service_processor *)inode->i_private;
 	rhbeat->active = 0;
 	ibmasm_init_reverse_heartbeat(rhbeat->sp, &rhbeat->heartbeat);
 	file->private_data = rhbeat;
@@ -507,7 +507,7 @@
 
 static int remote_settings_file_open(struct inode *inode, struct file *file)
 {
-	file->private_data = inode->u.generic_ip;
+	file->private_data = inode->i_private;
 	return 0;
 }
 
drivers/net/irda/vlsi_ir.h +1 -1
Index: linux-2.6.17/drivers/net/irda/vlsi_ir.h
===================================================================
@@ -58,7 +58,7 @@
 
 /* PDE() introduced in 2.5.4 */
 #ifdef CONFIG_PROC_FS
-#define PDE(inode) ((inode)->u.generic_ip)
+#define PDE(inode) ((inode)->i_private)
 #endif
 
 /* irda crc16 calculation exported in 2.5.42 */
drivers/oprofile/oprofilefs.c +5 -5
Index: linux-2.6.17/drivers/oprofile/oprofilefs.c
===================================================================
@@ -110,8 +110,8 @@
 
 static int default_open(struct inode * inode, struct file * filp)
 {
-	if (inode->u.generic_ip)
-		filp->private_data = inode->u.generic_ip;
+	if (inode->i_private)
+		filp->private_data = inode->i_private;
 	return 0;
 }
 
@@ -158,7 +158,7 @@
 	if (!d)
 		return -EFAULT;
 
-	d->d_inode->u.generic_ip = val;
+	d->d_inode->i_private = val;
 	return 0;
 }
 
@@ -171,7 +171,7 @@
 	if (!d)
 		return -EFAULT;
 
-	d->d_inode->u.generic_ip = val;
+	d->d_inode->i_private = val;
 	return 0;
 }
 
@@ -197,7 +197,7 @@
 	if (!d)
 		return -EFAULT;
 
-	d->d_inode->u.generic_ip = val;
+	d->d_inode->i_private = val;
 	return 0;
 }
 
drivers/pci/hotplug/cpqphp_sysfs.c +1 -1
Index: linux-2.6.17/drivers/pci/hotplug/cpqphp_sysfs.c
===================================================================
@@ -141,7 +141,7 @@
 
 static int open(struct inode *inode, struct file *file)
 {
-	struct controller *ctrl = inode->u.generic_ip;
+	struct controller *ctrl = inode->i_private;
 	struct ctrl_dbg *dbg;
 	int retval = -ENOMEM;
 
drivers/usb/core/devio.c +1 -1
Index: linux-2.6.17/drivers/usb/core/devio.c
===================================================================
@@ -553,7 +553,7 @@
 	if (imajor(inode) == USB_DEVICE_MAJOR)
 		dev = usbdev_lookup_minor(iminor(inode));
 	if (!dev)
-		dev = inode->u.generic_ip;
+		dev = inode->i_private;
 	if (!dev) {
 		kfree(ps);
 		goto out;
drivers/usb/core/inode.c +3 -3
Index: linux-2.6.17/drivers/usb/core/inode.c
===================================================================
@@ -403,8 +403,8 @@
 
 static int default_open (struct inode *inode, struct file *file)
 {
-	if (inode->u.generic_ip)
-		file->private_data = inode->u.generic_ip;
+	if (inode->i_private)
+		file->private_data = inode->i_private;
 
 	return 0;
 }
@@ -510,7 +510,7 @@
 	} else {
 		if (dentry->d_inode) {
 			if (data)
-				dentry->d_inode->u.generic_ip = data;
+				dentry->d_inode->i_private = data;
 			if (fops)
 				dentry->d_inode->i_fop = fops;
 			dentry->d_inode->i_uid = uid;
drivers/usb/gadget/inode.c +3 -3
Index: linux-2.6.17/drivers/usb/gadget/inode.c
===================================================================
@@ -845,7 +845,7 @@
 static int
 ep_open (struct inode *inode, struct file *fd)
 {
-	struct ep_data		*data = inode->u.generic_ip;
+	struct ep_data		*data = inode->i_private;
 	int			value = -EBUSY;
 
 	if (down_interruptible (&data->lock) != 0)
@@ -1908,7 +1908,7 @@
 static int
 dev_open (struct inode *inode, struct file *fd)
 {
-	struct dev_data		*dev = inode->u.generic_ip;
+	struct dev_data		*dev = inode->i_private;
 	int			value = -EBUSY;
 
 	if (dev->state == STATE_DEV_DISABLED) {
@@ -1969,7 +1969,7 @@
 		inode->i_blocks = 0;
 		inode->i_atime = inode->i_mtime = inode->i_ctime
 				= CURRENT_TIME;
-		inode->u.generic_ip = data;
+		inode->i_private = data;
 		inode->i_fop = fops;
 	}
 	return inode;
drivers/usb/host/isp116x-hcd.c +1 -1
Index: linux-2.6.17/drivers/usb/host/isp116x-hcd.c
===================================================================
@@ -1204,7 +1204,7 @@
 
 static int isp116x_open_seq(struct inode *inode, struct file *file)
 {
-	return single_open(file, isp116x_show_dbg, inode->u.generic_ip);
+	return single_open(file, isp116x_show_dbg, inode->i_private);
 }
 
 static struct file_operations isp116x_debug_fops = {
drivers/usb/host/uhci-debug.c +1 -1
Index: linux-2.6.17/drivers/usb/host/uhci-debug.c
===================================================================
@@ -406,7 +406,7 @@
 
 static int uhci_debug_open(struct inode *inode, struct file *file)
 {
-	struct uhci_hcd *uhci = inode->u.generic_ip;
+	struct uhci_hcd *uhci = inode->i_private;
 	struct uhci_debug *up;
 	int ret = -ENOMEM;
 	unsigned long flags;
drivers/usb/mon/mon_stat.c +1 -1
Index: linux-2.6.17/drivers/usb/mon/mon_stat.c
===================================================================
@@ -28,7 +28,7 @@
 	if ((sp = kmalloc(sizeof(struct snap), GFP_KERNEL)) == NULL)
 		return -ENOMEM;
 
-	mbus = inode->u.generic_ip;
+	mbus = inode->i_private;
 
 	sp->slen = snprintf(sp->str, STAT_BUF_SIZE,
 	    "nreaders %d text_lost %u\n",
drivers/usb/mon/mon_text.c +2 -2
Index: linux-2.6.17/drivers/usb/mon/mon_text.c
===================================================================
@@ -210,7 +210,7 @@
 	int rc;
 
 	mutex_lock(&mon_lock);
-	mbus = inode->u.generic_ip;
+	mbus = inode->i_private;
 	ubus = mbus->u_bus;
 
 	rp = kzalloc(sizeof(struct mon_reader_text), GFP_KERNEL);
@@ -372,7 +372,7 @@
 	struct mon_event_text *ep;
 
 	mutex_lock(&mon_lock);
-	mbus = inode->u.generic_ip;
+	mbus = inode->i_private;
 
 	if (mbus->nreaders <= 0) {
 		printk(KERN_ERR TAG ": consistency error on close\n");
fs/autofs/inode.c +1 -1
Index: linux-2.6.17/fs/autofs/inode.c
===================================================================
@@ -241,7 +241,7 @@
 		
 		inode->i_op = &autofs_symlink_inode_operations;
 		sl = &sbi->symlink[n];
-		inode->u.generic_ip = sl;
+		inode->i_private = sl;
 		inode->i_mode = S_IFLNK | S_IRWXUGO;
 		inode->i_mtime.tv_sec = inode->i_ctime.tv_sec = sl->mtime;
 		inode->i_mtime.tv_nsec = inode->i_ctime.tv_nsec = 0;
fs/autofs/symlink.c +1 -1
Index: linux-2.6.17/fs/autofs/symlink.c
===================================================================
@@ -15,7 +15,7 @@
 /* Nothing to release.. */
 static void *autofs_follow_link(struct dentry *dentry, struct nameidata *nd)
 {
-	char *s=((struct autofs_symlink *)dentry->d_inode->u.generic_ip)->data;
+	char *s=((struct autofs_symlink *)dentry->d_inode->i_private)->data;
 	nd_set_link(nd, s);
 	return NULL;
 }
fs/binfmt_misc.c +4 -4
Index: linux-2.6.17/fs/binfmt_misc.c
===================================================================
@@ -517,7 +517,7 @@
 
 static void bm_clear_inode(struct inode *inode)
 {
-	kfree(inode->u.generic_ip);
+	kfree(inode->i_private);
 }
 
 static void kill_node(Node *e)
@@ -545,7 +545,7 @@
 static ssize_t
 bm_entry_read(struct file * file, char __user * buf, size_t nbytes, loff_t *ppos)
 {
-	Node *e = file->f_dentry->d_inode->u.generic_ip;
+	Node *e = file->f_dentry->d_inode->i_private;
 	loff_t pos = *ppos;
 	ssize_t res;
 	char *page;
@@ -579,7 +579,7 @@
 				size_t count, loff_t *ppos)
 {
 	struct dentry *root;
-	Node *e = file->f_dentry->d_inode->u.generic_ip;
+	Node *e = file->f_dentry->d_inode->i_private;
 	int res = parse_command(buffer, count);
 
 	switch (res) {
@@ -646,7 +646,7 @@
 	}
 
 	e->dentry = dget(dentry);
-	inode->u.generic_ip = e;
+	inode->i_private = e;
 	inode->i_fop = &bm_entry_operations;
 
 	d_instantiate(dentry, inode);
fs/debugfs/file.c +2 -2
Index: linux-2.6.17/fs/debugfs/file.c
===================================================================
@@ -33,8 +33,8 @@
 
 static int default_open(struct inode *inode, struct file *file)
 {
-	if (inode->u.generic_ip)
-		file->private_data = inode->u.generic_ip;
+	if (inode->i_private)
+		file->private_data = inode->i_private;
 
 	return 0;
 }
fs/debugfs/inode.c +2 -2
Index: linux-2.6.17/fs/debugfs/inode.c
===================================================================
@@ -170,7 +170,7 @@
  *          directory dentry if set.  If this paramater is NULL, then the
  *          file will be created in the root of the debugfs filesystem.
  * @data: a pointer to something that the caller will want to get to later
- *        on.  The inode.u.generic_ip pointer will point to this value on
+ *        on.  The inode.i_private pointer will point to this value on
  *        the open() call.
  * @fops: a pointer to a struct file_operations that should be used for
  *        this file.
@@ -211,7 +211,7 @@
 
 	if (dentry->d_inode) {
 		if (data)
-			dentry->d_inode->u.generic_ip = data;
+			dentry->d_inode->i_private = data;
 		if (fops)
 			dentry->d_inode->i_fop = fops;
 	}
fs/devfs/base.c +5 -5
Index: linux-2.6.17/fs/devfs/base.c
===================================================================
@@ -26,7 +26,7 @@
                Original version.
   v0.1
     19980111   Richard Gooch <[email protected]>
-               Created per-fs inode table rather than using inode->u.generic_ip
+               Created per-fs inode table rather than using inode->i_private
   v0.2
     19980111   Richard Gooch <[email protected]>
                Created .epoch inode which has a ctime of 0.
@@ -521,7 +521,7 @@
   v0.109
     20010807   Richard Gooch <[email protected]>
 	       Fixed inode table races by removing it and using
-	       inode->u.generic_ip instead.
+	       inode->i_private instead.
 	       Moved <devfs_read_inode> into <get_vfs_inode>.
 	       Moved <devfs_write_inode> into <devfs_notify_change>.
   v0.110
@@ -1266,8 +1266,8 @@
 {
 	if (inode == NULL)
 		return NULL;
-	VERIFY_ENTRY((struct devfs_entry *)inode->u.generic_ip);
-	return inode->u.generic_ip;
+	VERIFY_ENTRY((struct devfs_entry *)inode->i_private);
+	return inode->i_private;
 }				/*  End Function get_devfs_entry_from_vfs_inode  */
 
 /**
@@ -1920,7 +1920,7 @@
 		return NULL;
 	}
 	/* FIXME where is devfs_put? */
-	inode->u.generic_ip = devfs_get(de);
+	inode->i_private = devfs_get(de);
 	inode->i_ino = de->inode.ino;
 	DPRINTK(DEBUG_I_GET, "(%d): VFS inode: %p  devfs_entry: %p\n",
 		(int)inode->i_ino, inode, de);
fs/devpts/inode.c +2 -2
Index: linux-2.6.17/fs/devpts/inode.c
===================================================================
@@ -177,7 +177,7 @@
 	inode->i_gid = config.setgid ? config.gid : current->fsgid;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 	init_special_inode(inode, S_IFCHR|config.mode, device);
-	inode->u.generic_ip = tty;
+	inode->i_private = tty;
 
 	dentry = get_node(number);
 	if (!IS_ERR(dentry) && !dentry->d_inode)
@@ -196,7 +196,7 @@
 	tty = NULL;
 	if (!IS_ERR(dentry)) {
 		if (dentry->d_inode)
-			tty = dentry->d_inode->u.generic_ip;
+			tty = dentry->d_inode->i_private;
 		dput(dentry);
 	}
 
fs/freevxfs/vxfs.h +1 -1
Index: linux-2.6.17/fs/freevxfs/vxfs.h
===================================================================
@@ -252,7 +252,7 @@
  * Get filesystem private data from VFS inode.
  */
 #define VXFS_INO(ip) \
-	((struct vxfs_inode_info *)(ip)->u.generic_ip)
+	((struct vxfs_inode_info *)(ip)->i_private)
 
 /*
  * Get filesystem private data from VFS superblock.
fs/freevxfs/vxfs_inode.c +2 -2
Index: linux-2.6.17/fs/freevxfs/vxfs_inode.c
===================================================================
@@ -243,7 +243,7 @@
 	ip->i_blocks = vip->vii_blocks;
 	ip->i_generation = vip->vii_gen;
 
-	ip->u.generic_ip = (void *)vip;
+	ip->i_private = (void *)vip;
 	
 }
 
@@ -338,5 +338,5 @@
 void
 vxfs_clear_inode(struct inode *ip)
 {
-	kmem_cache_free(vxfs_inode_cachep, ip->u.generic_ip);
+	kmem_cache_free(vxfs_inode_cachep, ip->i_private);
 }
fs/jffs/inode-v23.c +17 -17
Index: linux-2.6.17/fs/jffs/inode-v23.c
===================================================================
@@ -369,7 +369,7 @@
 
 	f = jffs_find_file(c, raw_inode->ino);
 
-	inode->u.generic_ip = (void *)f;
+	inode->i_private = (void *)f;
 	insert_inode_hash(inode);
 
 	return inode;
@@ -442,7 +442,7 @@
 	});
 
 	result = -ENOTDIR;
-	if (!(old_dir_f = (struct jffs_file *)old_dir->u.generic_ip)) {
+	if (!(old_dir_f = (struct jffs_file *)old_dir->i_private)) {
 		D(printk("jffs_rename(): Old dir invalid.\n"));
 		goto jffs_rename_end;
 	}
@@ -456,7 +456,7 @@
 
 	/* Find the new directory.  */
 	result = -ENOTDIR;
-	if (!(new_dir_f = (struct jffs_file *)new_dir->u.generic_ip)) {
+	if (!(new_dir_f = (struct jffs_file *)new_dir->i_private)) {
 		D(printk("jffs_rename(): New dir invalid.\n"));
 		goto jffs_rename_end;
 	}
@@ -593,7 +593,7 @@
 		}
 		else {
 			ddino = ((struct jffs_file *)
-				 inode->u.generic_ip)->pino;
+				 inode->i_private)->pino;
 		}
 		D3(printk("jffs_readdir(): \"..\" %u\n", ddino));
 		if (filldir(dirent, "..", 2, filp->f_pos, ddino, DT_DIR) < 0) {
@@ -604,7 +604,7 @@
 		}
 		filp->f_pos++;
 	}
-	f = ((struct jffs_file *)inode->u.generic_ip)->children;
+	f = ((struct jffs_file *)inode->i_private)->children;
 
 	j = 2;
 	while(f && (f->deleted || j++ < filp->f_pos )) {
@@ -668,7 +668,7 @@
 	}
 
 	r = -EACCES;
-	if (!(d = (struct jffs_file *)dir->u.generic_ip)) {
+	if (!(d = (struct jffs_file *)dir->i_private)) {
 		D(printk("jffs_lookup(): No such inode! (%lu)\n",
 			 dir->i_ino));
 		goto jffs_lookup_end;
@@ -739,7 +739,7 @@
 	unsigned long read_len;
 	int result;
 	struct inode *inode = (struct inode*)page->mapping->host;
-	struct jffs_file *f = (struct jffs_file *)inode->u.generic_ip;
+	struct jffs_file *f = (struct jffs_file *)inode->i_private;
 	struct jffs_control *c = (struct jffs_control *)inode->i_sb->s_fs_info;
 	int r;
 	loff_t offset;
@@ -828,7 +828,7 @@
 	});
 
 	lock_kernel();
-	dir_f = (struct jffs_file *)dir->u.generic_ip;
+	dir_f = (struct jffs_file *)dir->i_private;
 
 	ASSERT(if (!dir_f) {
 		printk(KERN_ERR "jffs_mkdir(): No reference to a "
@@ -972,7 +972,7 @@
 		kfree(_name);
 	});
 
-	dir_f = (struct jffs_file *) dir->u.generic_ip;
+	dir_f = (struct jffs_file *) dir->i_private;
 	c = dir_f->c;
 
 	result = -ENOENT;
@@ -1082,7 +1082,7 @@
 	if (!old_valid_dev(rdev))
 		return -EINVAL;
 	lock_kernel();
-	dir_f = (struct jffs_file *)dir->u.generic_ip;
+	dir_f = (struct jffs_file *)dir->i_private;
 	c = dir_f->c;
 
 	D3(printk (KERN_NOTICE "mknod(): down biglock\n"));
@@ -1186,7 +1186,7 @@
 		kfree(_symname);
 	});
 
-	dir_f = (struct jffs_file *)dir->u.generic_ip;
+	dir_f = (struct jffs_file *)dir->i_private;
 	ASSERT(if (!dir_f) {
 		printk(KERN_ERR "jffs_symlink(): No reference to a "
 		       "jffs_file struct in inode.\n");
@@ -1289,7 +1289,7 @@
 		kfree(s);
 	});
 
-	dir_f = (struct jffs_file *)dir->u.generic_ip;
+	dir_f = (struct jffs_file *)dir->i_private;
 	ASSERT(if (!dir_f) {
 		printk(KERN_ERR "jffs_create(): No reference to a "
 		       "jffs_file struct in inode.\n");
@@ -1403,9 +1403,9 @@
 		goto out_isem;
 	}
 
-	if (!(f = (struct jffs_file *)inode->u.generic_ip)) {
-		D(printk("jffs_file_write(): inode->u.generic_ip = 0x%p\n",
-				inode->u.generic_ip));
+	if (!(f = (struct jffs_file *)inode->i_private)) {
+		D(printk("jffs_file_write(): inode->i_private = 0x%p\n",
+				inode->i_private));
 		goto out_isem;
 	}
 
@@ -1693,7 +1693,7 @@
 		mutex_unlock(&c->fmc->biglock);
 		return;
 	}
-	inode->u.generic_ip = (void *)f;
+	inode->i_private = (void *)f;
 	inode->i_mode = f->mode;
 	inode->i_nlink = f->nlink;
 	inode->i_uid = f->uid;
@@ -1748,7 +1748,7 @@
 	lock_kernel();
 	inode->i_size = 0;
 	inode->i_blocks = 0;
-	inode->u.generic_ip = NULL;
+	inode->i_private = NULL;
 	clear_inode(inode);
 	if (inode->i_nlink == 0) {
 		c = (struct jffs_control *) inode->i_sb->s_fs_info;
fs/libfs.c +1 -1
Index: linux-2.6.17/fs/libfs.c
===================================================================
@@ -549,7 +549,7 @@
 
 	attr->get = get;
 	attr->set = set;
-	attr->data = inode->u.generic_ip;
+	attr->data = inode->i_private;
 	attr->fmt = fmt;
 	mutex_init(&attr->mutex);
 
fs/ocfs2/dlmglue.c +1 -1
Index: linux-2.6.17/fs/ocfs2/dlmglue.c
===================================================================
@@ -1995,7 +1995,7 @@
 		mlog_errno(ret);
 		goto out;
 	}
-	osb = (struct ocfs2_super *) inode->u.generic_ip;
+	osb = (struct ocfs2_super *) inode->i_private;
 	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
 	priv->p_dlm_debug = osb->osb_dlm_debug;
 	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
fs/openpromfs/inode.c +11 -11
Index: linux-2.6.17/fs/openpromfs/inode.c
===================================================================
@@ -70,9 +70,9 @@
 	struct inode *inode = file->f_dentry->d_inode;
 	char buffer[10];
 	
-	if (count < 0 || !inode->u.generic_ip)
+	if (count < 0 || !inode->i_private)
 		return -EINVAL;
-	sprintf (buffer, "%8.8x\n", (u32)(long)(inode->u.generic_ip));
+	sprintf (buffer, "%8.8x\n", (u32)(long)(inode->i_private));
 	if (file->f_pos >= 9)
 		return 0;
 	if (count > 9 - file->f_pos)
@@ -95,9 +95,9 @@
 	char buffer[64];
 	
 	if (!filp->private_data) {
-		node = nodes[(u16)((long)inode->u.generic_ip)].node;
-		i = ((u32)(long)inode->u.generic_ip) >> 16;
-		if ((u16)((long)inode->u.generic_ip) == aliases) {
+		node = nodes[(u16)((long)inode->i_private)].node;
+		i = ((u32)(long)inode->i_private) >> 16;
+		if ((u16)((long)inode->i_private) == aliases) {
 			if (i >= aliases_nodes)
 				p = NULL;
 			else
@@ -111,7 +111,7 @@
 			return -EIO;
 		i = prom_getproplen (node, p);
 		if (i < 0) {
-			if ((u16)((long)inode->u.generic_ip) == aliases)
+			if ((u16)((long)inode->i_private) == aliases)
 				i = 0;
 			else
 				return -EIO;
@@ -539,8 +539,8 @@
 	if (!op)
 		return 0;
 	lock_kernel();
-	node = nodes[(u16)((long)inode->u.generic_ip)].node;
-	if ((u16)((long)inode->u.generic_ip) == aliases) {
+	node = nodes[(u16)((long)inode->i_private)].node;
+	if ((u16)((long)inode->i_private) == aliases) {
 		if ((op->flag & OPP_DIRTY) && (op->flag & OPP_STRING)) {
 			char *p = op->name;
 			int i = (op->value - op->name) - strlen (op->name) - 1;
@@ -743,7 +743,7 @@
 		inode->i_mode = S_IFREG | S_IRUGO;
 		inode->i_fop = &openpromfs_nodenum_ops;
 		inode->i_nlink = 1;
-		inode->u.generic_ip = (void *)(long)(n);
+		inode->i_private = (void *)(long)(n);
 		break;
 	case OPFSL_PROPERTY:
 		if ((dirnode == options) && (len == 17)
@@ -760,7 +760,7 @@
 		inode->i_nlink = 1;
 		if (inode->i_size < 0)
 			inode->i_size = 0;
-		inode->u.generic_ip = (void *)(long)(((u16)dirnode) | 
+		inode->i_private = (void *)(long)(((u16)dirnode) |
 			(((u16)(ino - NODEP2INO(NODE(dir->i_ino).first_prop) - 1)) << 16));
 		break;
 	}
@@ -886,7 +886,7 @@
 	inode->i_fop = &openpromfs_prop_ops;
 	inode->i_nlink = 1;
 	if (inode->i_size < 0) inode->i_size = 0;
-	inode->u.generic_ip = (void *)(long)(((u16)aliases) | 
+	inode->i_private = (void *)(long)(((u16)aliases) |
 			(((u16)(aliases_nodes - 1)) << 16));
 	unlock_kernel();
 	d_instantiate(dentry, inode);
security/inode.c +4 -4
Index: linux-2.6.17/security/inode.c
===================================================================
@@ -45,8 +45,8 @@
 
 static int default_open(struct inode *inode, struct file *file)
 {
-	if (inode->u.generic_ip)
-		file->private_data = inode->u.generic_ip;
+	if (inode->i_private)
+		file->private_data = inode->i_private;
 
 	return 0;
 }
@@ -195,7 +195,7 @@
  *          directory dentry if set.  If this paramater is NULL, then the
  *          file will be created in the root of the securityfs filesystem.
  * @data: a pointer to something that the caller will want to get to later
- *        on.  The inode.u.generic_ip pointer will point to this value on
+ *        on.  The inode.i_private pointer will point to this value on
  *        the open() call.
  * @fops: a pointer to a struct file_operations that should be used for
  *        this file.
@@ -241,7 +241,7 @@
 		if (fops)
 			dentry->d_inode->i_fop = fops;
 		if (data)
-			dentry->d_inode->u.generic_ip = data;
+			dentry->d_inode->i_private = data;
 	}
 exit:
 	return dentry;
fs/inode.c +1 -1
Index: linux-2.6.17/fs/inode.c
===================================================================
@@ -164,7 +164,7 @@
 				bdi = sb->s_bdev->bd_inode->i_mapping->backing_dev_info;
 			mapping->backing_dev_info = bdi;
 		}
-		memset(&inode->u, 0, sizeof(inode->u));
+		inode->i_private = 0;
 		inode->i_mapping = mapping;
 	}
 	return inode;
block/blktrace.c +1 -1
Index: linux-2.6.17/block/blktrace.c
===================================================================
@@ -215,7 +215,7 @@
 
 static int blk_dropped_open(struct inode *inode, struct file *filp)
 {
-	filp->private_data = inode->u.generic_ip;
+	filp->private_data = inode->i_private;
 
 	return 0;
 }
drivers/infiniband/hw/ipath/ipath_fs.c +6 -6
Index: linux-2.6.17/drivers/infiniband/hw/ipath/ipath_fs.c
===================================================================
@@ -64,7 +64,7 @@
 	inode->i_blksize = PAGE_CACHE_SIZE;
 	inode->i_blocks = 0;
 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
-	inode->u.generic_ip = data;
+	inode->i_private = data;
 	if ((mode & S_IFMT) == S_IFDIR) {
 		inode->i_op = &simple_dir_inode_operations;
 		inode->i_nlink++;
@@ -119,7 +119,7 @@
 	u16 i;
 	struct ipath_devdata *dd;
 
-	dd = file->f_dentry->d_inode->u.generic_ip;
+	dd = file->f_dentry->d_inode->i_private;
 
 	for (i = 0; i < NUM_COUNTERS; i++)
 		counters[i] = ipath_snap_cntr(dd, i);
@@ -139,7 +139,7 @@
 	struct ipath_devdata *dd;
 	u64 guid;
 
-	dd = file->f_dentry->d_inode->u.generic_ip;
+	dd = file->f_dentry->d_inode->i_private;
 
 	guid = be64_to_cpu(dd->ipath_guid);
 
@@ -178,7 +178,7 @@
 	u32 tmp, tmp2;
 	struct ipath_devdata *dd;
 
-	dd = file->f_dentry->d_inode->u.generic_ip;
+	dd = file->f_dentry->d_inode->i_private;
 
 	/* so we only initialize non-zero fields. */
 	memset(portinfo, 0, sizeof portinfo);
@@ -325,7 +325,7 @@
 		goto bail;
 	}
 
-	dd = file->f_dentry->d_inode->u.generic_ip;
+	dd = file->f_dentry->d_inode->i_private;
 	if (ipath_eeprom_read(dd, pos, tmp, count)) {
 		ipath_dev_err(dd, "failed to read from flash\n");
 		ret = -ENXIO;
@@ -381,7 +381,7 @@
 		goto bail_tmp;
 	}
 
-	dd = file->f_dentry->d_inode->u.generic_ip;
+	dd = file->f_dentry->d_inode->i_private;
 	if (ipath_eeprom_write(dd, pos, tmp, count)) {
 		ret = -ENXIO;
 		ipath_dev_err(dd, "failed to write to flash\n");
drivers/net/wireless/bcm43xx/bcm43xx_debugfs.c +1 -1
Index: linux-2.6.17/drivers/net/wireless/bcm43xx/bcm43xx_debugfs.c
===================================================================
@@ -54,7 +54,7 @@
 
 static int open_file_generic(struct inode *inode, struct file *file)
 {
-	file->private_data = inode->u.generic_ip;
+	file->private_data = inode->i_private;
 	return 0;
 }
 
kernel/relay.c +1 -1
Index: linux-2.6.17/kernel/relay.c
===================================================================
@@ -669,7 +669,7 @@
  */
 static int relay_file_open(struct inode *inode, struct file *filp)
 {
-	struct rchan_buf *buf = inode->u.generic_ip;
+	struct rchan_buf *buf = inode->i_private;
 	kref_get(&buf->kref);
 	filp->private_data = buf;
 

--

#2 lore raw

Re: [RFC] [PATCH 1/8] inode_diet: Replace inode.u.generic_ip with inode.i_private

arch/s390/kernel/debug.c line 607

Quoted 9 lines
arch/s390/kernel/debug.c +1 -1
@@ -604,7 +604,7 @@
 	debug_info_t *debug_info, *debug_info_snapshot;
 
 	down(&debug_lock);
-	debug_info = (struct debug_info*)file->f_dentry->d_inode->u.generic_ip;
+	debug_info = (struct debug_info*)file->f_dentry->d_inode->i_private;

All these casts can be removed at the same time,
since i_private is a void* anyway and therefore does not require casting.

Quoted 13 lines
arch/s390/kernel/debug.c +4 -4
-	rhbeat->sp = (struct service_processor *)inode->u.generic_ip;
+	rhbeat->sp = (struct service_processor *)inode->i_private;
 
 
-	char *s=((struct autofs_symlink *)dentry->d_inode->u.generic_ip)->data;
+	char *s=((struct autofs_symlink *)dentry->d_inode->i_private)->data;
 
-	VERIFY_ENTRY((struct devfs_entry *)inode->u.generic_ip);
-	return inode->u.generic_ip;
+	VERIFY_ENTRY((struct devfs_entry *)inode->i_private);
+	return inode->i_private;

#define VXFS_INO(ip) \

arch/s390/kernel/debug.c

Quoted 2 lines
arch/s390/kernel/debug.c +1 -1
-	((struct vxfs_inode_info *)(ip)->u.generic_ip)
+	((struct vxfs_inode_info *)(ip)->i_private)

Depends on use of VXFS_INO here.

arch/s390/kernel/debug.c

Quoted 8 lines
arch/s390/kernel/debug.c +3 -3
-	ip->u.generic_ip = (void *)vip;
+	ip->i_private = (void *)vip;
 
-	inode->u.generic_ip = (void *)f;
+	inode->i_private = (void *)f;
 
-	if (!(old_dir_f = (struct jffs_file *)old_dir->u.generic_ip)) {
+	if (!(old_dir_f = (struct jffs_file *)old_dir->i_private)) {

Not sure.

Quoted 24 lines
arch/s390/kernel/debug.c +3 -3
-	if (!(new_dir_f = (struct jffs_file *)new_dir->u.generic_ip)) {
+	if (!(new_dir_f = (struct jffs_file *)new_dir->i_private)) {
 
-	if (!(d = (struct jffs_file *)dir->u.generic_ip)) {
+	if (!(d = (struct jffs_file *)dir->i_private)) {
 
-	struct jffs_file *f = (struct jffs_file *)inode->u.generic_ip;
+	struct jffs_file *f = (struct jffs_file *)inode->i_private;
	struct jffs_control *c = (struct jffs_control *)inode->i_sb->s_fs_info;
arch/s390/kernel/debug.c +5 -5
-	dir_f = (struct jffs_file *)dir->u.generic_ip;
+	dir_f = (struct jffs_file *)dir->i_private;
 
-	dir_f = (struct jffs_file *) dir->u.generic_ip;
+	dir_f = (struct jffs_file *) dir->i_private;
 
-	dir_f = (struct jffs_file *)dir->u.generic_ip;
+	dir_f = (struct jffs_file *)dir->i_private;
 
-	dir_f = (struct jffs_file *)dir->u.generic_ip;
+	dir_f = (struct jffs_file *)dir->i_private;
 
-	dir_f = (struct jffs_file *)dir->u.generic_ip;
+	dir_f = (struct jffs_file *)dir->i_private;

arch/s390/kernel/debug.c

Quoted 10 lines
arch/s390/kernel/debug.c +4 -4
 
-	if (!(f = (struct jffs_file *)inode->u.generic_ip)) {
-		D(printk("jffs_file_write(): inode->u.generic_ip = 0x%p\n",
-				inode->u.generic_ip));
+	if (!(f = (struct jffs_file *)inode->i_private)) {
+		D(printk("jffs_file_write(): inode->i_private = 0x%p\n",
+				inode->i_private));
 
-	inode->u.generic_ip = (void *)f;
+	inode->i_private = (void *)f;

lots more.

Jan Engelhardt

#3 lore raw

Re: [RFC] [PATCH 1/8] inode_diet: Replace inode.u.generic_ip with inode.i_private

In reply to Theodore Tso

arch/s390/kernel/debug.c line 607

Quoted 10 lines

Index: linux-2.6.17/arch/s390/kernel/debug.c
===================================================================

arch/s390/kernel/debug.c +1 -1
@@ -604,7 +604,7 @@
 	debug_info_t *debug_info, *debug_info_snapshot;
 
 	down(&debug_lock);
-	debug_info = (struct debug_info*)file->f_dentry->d_inode->u.generic_ip;
+	debug_info = (struct debug_info*)file->f_dentry->d_inode->i_private;

is this actually okay? who owns the private data in debugfs inodes?

#4 lore raw

Re: [RFC] [PATCH 1/8] inode_diet: Replace inode.u.generic_ip with inode.i_private

Quoted 7 lines · Christoph Hellwig
Quoted 7 lines

Index: linux-2.6.17/arch/s390/kernel/debug.c
===================================================================

arch/s390/kernel/debug.c +0 -0
@@ -604,7 +604,7 @@
 	debug_info_t *debug_info, *debug_info_snapshot;
 
Quoted 5 lines

arch/s390/kernel/debug.c line 607

Quoted 3 lines
arch/s390/kernel/debug.c +1 -1
 	down(&debug_lock);
-	debug_info = (struct debug_info*)file->f_dentry->d_inode->u.generic_ip;
+	debug_info = (struct debug_info*)file->f_dentry->d_inode->i_private;

is this actually okay? who owns the private data in debugfs inodes?

debugfs allows the caller of debugfs_create_file() to pass in private data. So this in fact the intended way it's supposed to work AFAIK. The s/390 debug facility passed in a struct debug_info * in debug_register_view()'s call to debugfs_create_file, and it pulls it out again in debug_open() (in the above referenced patch).

- Ted

#5 lore raw

Re: [RFC] [PATCH 1/8] inode_diet: Replace inode.u.generic_ip with inode.i_private

In reply to Theodore Tso

Hi,

Quoted 16 lines · Theodore Tso

plain text document attachment (inode-slim-1)
The filesystem or device-specific pointer in the inode is inside a union, which is pretty pointless given that all 30+ users of this field have been using the void pointer. Get rid of the union and rename it to i_private, with a comment to explain who is allowed to use the void pointer. This is just a cleanup, but it allows us to reuse the union 'u' for something something where the union will actually be used.

Index: linux-2.6.17/include/linux/fs.h
===================================================================

include/linux/fs.h line 540

Quoted 10 lines
include/linux/fs.h +1 -3
@@ -534,9 +534,7 @@
 
 	atomic_t		i_writecount;
 	void			*i_security;
-	union {
-		void		*generic_ip;
-	} u;
+	void			*i_private; /* fs or device private pointer */
 #ifdef __NEED_I_SIZE_ORDERED
 	seqcount_t		i_size_seqcount;
 #endif

As a further suggestion, I wonder do we really need i_private at all? Since we have sb->s_op->alloc_inode and inode->i_sb->s_op->destroy_inode if all filesystems did something along the following lines:

struct myfs_inode {
	struct inode i_inode;
	...
};

#define MYFS_I(inode) container_of((inode), struct myfs_inode, i_inode)

then it would seem that i_private is redundant. If there is a file system which does genuinely need a pointer here (if indeed such a filesystem does exist, I haven't actually checked that) then a pointer can just be added as the one single other member of (in my example) struct myfs_inode.

Steve.

#6 lore raw

Re: [RFC] [PATCH 1/8] inode_diet: Replace inode.u.generic_ip with inode.i_private

Quoted 9 lines · Steven Whitehouse

As a further suggestion, I wonder do we really need i_private at all? Since we have sb->s_op->alloc_inode and inode->i_sb->s_op->destroy_inode if all filesystems did something along the following lines:

struct myfs_inode {
        struct inode i_inode;
        ...
};

Quoted 8 lines

#define MYFS_I(inode) container_of((inode), struct myfs_inode, i_inode)

then it would seem that i_private is redundant. If there is a file system which does genuinely need a pointer here (if indeed such a filesystem does exist, I haven't actually checked that) then a pointer can just be added as the one single other member of (in my example) struct myfs_inode.

That would mean that all file systems need to implement ->alloc_inode, which in turn need slab caches that eat consume memory even when the file system is not mounted.

Something as simple as nfsctl or devpts should not need that.

	Arnd <><
#7 lore raw

Re: [RFC] [PATCH 1/8] inode_diet: Replace inode.u.generic_ip with inode.i_private

Hi,

Quoted 18 lines · Arnd Bergmann
Quoted 9 lines · Steven Whitehouse

As a further suggestion, I wonder do we really need i_private at all? Since we have sb->s_op->alloc_inode and inode->i_sb->s_op->destroy_inode if all filesystems did something along the following lines:

struct myfs_inode {
        struct inode i_inode;
        ...
};
Quoted 8 lines

#define MYFS_I(inode) container_of((inode), struct myfs_inode, i_inode)

then it would seem that i_private is redundant. If there is a file system which does genuinely need a pointer here (if indeed such a filesystem does exist, I haven't actually checked that) then a pointer can just be added as the one single other member of (in my example) struct myfs_inode.

Quoted 4 lines

That would mean that all file systems need to implement ->alloc_inode, which in turn need slab caches that eat consume memory even when the file system is not mounted.

Yes, although I'm not sure that it would be as significant as the memory saved by removing the pointer bearing in mind the relative numbers of the structures that you'd expect to see in a "normal" working system. We could also try and reduce this by creating a special inode cache which would be shared by all filesystems which did still need just struct inode + private pointer for example.

What you do gain though is (on umount of a filesystem) a much greater likelihood of being able to reclaim the memory which was being used by the inodes of that particular filesystem (particularly so if you only have a single mounted filesystem of a particular type). So hopefully having a separate slab cache per fstype would help reduce memory fragmentation, and more than compensate for the difference.

In fact a number of filesystems already have slab caches for their own private part of the inode anyway... I count 10 of those on my current development box.

Quoted 3 lines

Something as simple as nfsctl or devpts should not need that.

	Arnd <><

Steve.

#8 lore raw

Re: [RFC] [PATCH 1/8] inode_diet: Replace inode.u.generic_ip with inode.i_private

Quoted 7 lines · Steven Whitehouse

Yes, although I'm not sure that it would be as significant as the memory saved by removing the pointer bearing in mind the relative numbers of the structures that you'd expect to see in a "normal" working system. We could also try and reduce this by creating a special inode cache which would be shared by all filesystems which did still need just struct inode + private pointer for example.

To take this further, you could indeed split struct inode into a smaller struct that has all the important parts and a derived struct that has i_private as well as other members that are used only by a minority of file systems.

Alternatively, it might be possible to stuff i_private into the same union as i_pipe, i_cdev and i_bdev. The rationale here being that a file system implementing different file types already is complex enough that you would normally want your own alloc_inode for a derived struct. The simple file systems OTOH normally only support regular files, and sometimes directories.

	Arnd <><
#9 lore raw

Re: [RFC] [PATCH 1/8] inode_diet: Replace inode.u.generic_ip with inode.i_private

Arnd,

Quoted 7 lines · Arnd Bergmann

Alternatively, it might be possible to stuff i_private into the same union as i_pipe, i_cdev and i_bdev. The rationale here being that a file system implementing different file types already is complex enough that you would normally want your own alloc_inode for a derived struct. The simple file systems OTOH normally only support regular files, and sometimes directories.

Placing i_private and i_pipe in the same union will break FIFOs.

Also, i_pipe should not be combined with i_cdev and i_bdev.

#10 lore raw

Re: [RFC] [PATCH 1/8] inode_diet: Replace inode.u.generic_ip with inode.i_private

Arnd,

Quoted 5 lines · Brian F. G. Bidulock

Placing i_private and i_pipe in the same union will break FIFOs.

Also, i_pipe should not be combined with i_cdev and i_bdev.

Also, both character and block devices need separate use of i_private.

#11 lore raw

Re: [RFC] [PATCH 1/8] inode_diet: Replace inode.u.generic_ip with inode.i_private

In reply to Steven Whitehouse
Quoted 10 lines · Steven Whitehouse

As a further suggestion, I wonder do we really need i_private at all? Since we have sb->s_op->alloc_inode and inode->i_sb->s_op->destroy_inode if all filesystems did something along the following lines:

struct myfs_inode {
	struct inode i_inode;
	...
};

#define MYFS_I(inode) container_of((inode), struct myfs_inode, i_inode)

That would work for filesystems but we would also need some solution for device inodes. (And at that point, yes, we could move it into device-specific union, but as I've already noted, that doesn't buy us anything currently.)

It's worth thinking about, but for that amount of effort it might be have better ROI to work on moving the address space out of the inode given that many/most inodes in memory are caching stat information, not pages.

- Ted

#12 lore raw

RFC PATCH 2/8 inode-diet: Move i_pipe into a union

In reply to Theodore Tso

Move the i_pipe pointer into a union that will be shared with i_bdev and i_cdev.

include/linux/fs.h +3 -2
Index: linux-2.6.17/include/linux/fs.h
===================================================================
@@ -508,9 +508,10 @@
 #ifdef CONFIG_QUOTA
 	struct dquot		*i_dquot[MAXQUOTAS];
 #endif
-	/* These three should probably be a union */
 	struct list_head	i_devices;
-	struct pipe_inode_info	*i_pipe;
+	union {
+		struct pipe_inode_info	*i_pipe;
+	};
 	struct block_device	*i_bdev;
 	struct cdev		*i_cdev;
 	int			i_cindex;

--

#13 lore raw

Re: [RFC] [PATCH 2/8] inode-diet: Move i_pipe into a union

include/linux/fs.h line 514

Quoted 7 lines

Move the i_pipe pointer into a union that will be shared with i_bdev and i_cdev.

include/linux/fs.h +3 -0
+	union {
+		struct pipe_inode_info	*i_pipe;
+	};

Since in the next patch you did

- if (inode->i_bdev)
+ if (S_ISBLK(inode->i_mode) && inode->i_bdev)

I am just asking, for clarity, if there were any similar lines for pipes that should now read S_IFIFO(inode->i_mode) too, like for bdevs.

Jan Engelhardt

#14 lore raw

Re: [RFC] [PATCH 2/8] inode-diet: Move i_pipe into a union

Quoted 13 lines · Jan Engelhardt

include/linux/fs.h line 514

Quoted 7 lines

Move the i_pipe pointer into a union that will be shared with i_bdev and i_cdev.

include/linux/fs.h +3 -0
+	union {
+		struct pipe_inode_info	*i_pipe;
+	};

Since in the next patch you did

(quoted diff) +1 -1
-       if (inode->i_bdev)
+       if (S_ISBLK(inode->i_mode) && inode->i_bdev)
Quoted 2 lines

I am just asking, for clarity, if there were any similar lines for pipes that should now read S_IFIFO(inode->i_mode) too, like for bdevs.

Nope, I didn't see any, and I did audit all of the places that used i_pipe. At least in the mainline kernel, all of the places which used i_pipe were in places were we knew already that we were dealing with a pipe.

As was mentioned in earlier comment, this will be problematic for the out-of-tree System V Streams code, which hijacks i_pipe as another place to store 4 bytes of random data needed for the Streams code (I believe they needed a pointer to the stream head -- the v_str pointer in a legacy Unix system's inode). But, that is an out-of-tree kernel module, and it's a clear abuse of the i_pipe element in any case.

- Ted

#15 lore raw

Re: [RFC] [PATCH 2/8] inode-diet: Move i_pipe into a union

Theodore,

Quoted 7 lines · Theodore Tso

As was mentioned in earlier comment, this will be problematic for the out-of-tree System V Streams code, which hijacks i_pipe as another place to store 4 bytes of random data needed for the Streams code (I believe they needed a pointer to the stream head -- the v_str pointer in a legacy Unix system's inode). But, that is an out-of-tree kernel module, and it's a clear abuse of the i_pipe element in any case.

It's used for implementing STREAMS-based FIFOs. Which is a proper use of i_pipe (which is for FIFOs). Pipes (both mainline and STREAMS-based pipes) can use i_private instead of i_pipe.

For the same reason why i_pipe cannot be combined with i_private.

#16 lore raw

Re: [RFC] [PATCH 2/8] inode-diet: Move i_pipe into a union

Quoted 4 lines · Brian F. G. Bidulock

It's used for implementing STREAMS-based FIFOs. Which is a proper use of i_pipe (which is for FIFOs). Pipes (both mainline and STREAMS-based pipes) can use i_private instead of i_pipe.

It's is an abuse of i_pipe. You are using something which is supposed to hold a struct pipe_info, and storing the head of the STREAM stack, which is some other type.

In any case, when you state authoratively what "can" and "can not" be combined, please specify when your justification is for a particular out of tree modules.

- Ted

#17 lore raw

Re: [RFC] [PATCH 2/8] inode-diet: Move i_pipe into a union

Theodore,

Quoted 4 lines · Theodore Tso

In any case, when you state authoratively what "can" and "can not" be combined, please specify when your justification is for a particular out of tree modules.

In that case, be my guest: combine i_pipe with i_private: see if what you break is worth the gain.

#18 lore raw

RFC PATCH 3/8 inode-diet: Move i_bdev into a union

In reply to Theodore Tso

Move the i_bdev pointer in struct inode into a union.

include/linux/fs.h +1 -1
Index: linux-2.6.17/include/linux/fs.h
===================================================================
@@ -511,8 +511,8 @@
 	struct list_head	i_devices;
 	union {
 		struct pipe_inode_info	*i_pipe;
+		struct block_device	*i_bdev;
 	};
-	struct block_device	*i_bdev;
 	struct cdev		*i_cdev;
 	int			i_cindex;
 
fs/block_dev.c +1 -1
Index: linux-2.6.17/fs/block_dev.c
===================================================================
@@ -439,7 +439,7 @@
 void bd_forget(struct inode *inode)
 {
 	spin_lock(&bdev_lock);
-	if (inode->i_bdev)
+	if (S_ISBLK(inode->i_mode) && inode->i_bdev)
 		__bd_forget(inode);
 	spin_unlock(&bdev_lock);
 }
fs/inode.c +1 -1
Index: linux-2.6.17/fs/inode.c
===================================================================
@@ -255,7 +255,7 @@
 	DQUOT_DROP(inode);
 	if (inode->i_sb && inode->i_sb->s_op->clear_inode)
 		inode->i_sb->s_op->clear_inode(inode);
-	if (inode->i_bdev)
+	if (S_ISBLK(inode->i_mode) && inode->i_bdev)
 		bd_forget(inode);
 	if (inode->i_cdev)
 		cd_forget(inode);

--

#19 lore raw

RFC PATCH 4/8 inode-diet: Move i_cdev into a union

In reply to Theodore Tso

Move the i_cdev pointer in struct inode into a union.

include/linux/fs.h +1 -1
Index: linux-2.6.17/include/linux/fs.h
===================================================================
@@ -512,8 +512,8 @@
 	union {
 		struct pipe_inode_info	*i_pipe;
 		struct block_device	*i_bdev;
+		struct cdev		*i_cdev;
 	};
-	struct cdev		*i_cdev;
 	int			i_cindex;
 
 	__u32			i_generation;
fs/file_table.c +1 -1
Index: linux-2.6.17/fs/file_table.c
===================================================================
@@ -170,7 +170,7 @@
 	if (file->f_op && file->f_op->release)
 		file->f_op->release(inode, file);
 	security_file_free(file);
-	if (unlikely(inode->i_cdev != NULL))
+	if (unlikely(S_ISCHR(inode->i_mode) && (inode->i_cdev != NULL)))
 		cdev_put(inode->i_cdev);
 	fops_put(file->f_op);
 	if (file->f_mode & FMODE_WRITE)
fs/inode.c +1 -1
Index: linux-2.6.17/fs/inode.c
===================================================================
@@ -257,7 +257,7 @@
 		inode->i_sb->s_op->clear_inode(inode);
 	if (S_ISBLK(inode->i_mode) && inode->i_bdev)
 		bd_forget(inode);
-	if (inode->i_cdev)
+	if (S_ISCHR(inode->i_mode) && inode->i_cdev)
 		cd_forget(inode);
 	inode->i_state = I_CLEAR;
 }

--

#20 lore raw

Re: [RFC] [PATCH 4/8] inode-diet: Move i_cdev into a union

fs/file_table.c line 173

Quoted 10 lines

===================================================================

fs/file_table.c +1 -1
@@ -170,7 +170,7 @@
 	if (file->f_op && file->f_op->release)
 		file->f_op->release(inode, file);
 	security_file_free(file);
-	if (unlikely(inode->i_cdev != NULL))
+	if (unlikely(S_ISCHR(inode->i_mode) && (inode->i_cdev != NULL)))

Am I allowed to be nitpicky? If so, drop the () around inode->i_cdev != NULL. :)

Jan Engelhardt

#21 lore raw

RFC PATCH 5/8 inode-diet: Eliminate i_blksize and use a per-superblock default

In reply to Theodore Tso

This eliminates the i_blksize field from struct inode and uses default value in super->s_blksize. Filesystems that want to provide a per-inode st_blksize can do so by providing their own getattr routine instead of using the generic_fillattr() function.

Note that some filesystems were providing pretty much random (and incorrect) values for i_blksize.

include/linux/fs.h +2 -1
Index: linux-2.6.17/include/linux/fs.h
===================================================================
@@ -492,7 +492,6 @@
 	struct timespec		i_mtime;
 	struct timespec		i_ctime;
 	unsigned int		i_blkbits;
-	unsigned long		i_blksize;
 	unsigned long		i_version;
 	blkcnt_t		i_blocks;
 	unsigned short          i_bytes;
@@ -867,6 +866,8 @@
 	/* Granularity of c/m/atime in ns.
 	   Cannot be worse than a second */
 	u32		   s_time_gran;
+	unsigned long	   s_blksize; /* Optimal i/o size for stat(2) */
+
 };
 
 extern struct timespec current_fs_time(struct super_block *sb);
fs/stat.c +2 -1
Index: linux-2.6.17/fs/stat.c
===================================================================
@@ -33,7 +33,8 @@
 	stat->ctime = inode->i_ctime;
 	stat->size = i_size_read(inode);
 	stat->blocks = inode->i_blocks;
-	stat->blksize = inode->i_blksize;
+	if (inode->i_sb->s_blksize)
+		stat->blksize = inode->i_sb->s_blksize;
 }
 
 EXPORT_SYMBOL(generic_fillattr);
fs/binfmt_misc.c +0 -1
Index: linux-2.6.17/fs/binfmt_misc.c
===================================================================
@@ -507,7 +507,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = 0;
 		inode->i_gid = 0;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_atime = inode->i_mtime = inode->i_ctime =
 			current_fs_time(inode->i_sb);
fs/eventpoll.c +0 -1
Index: linux-2.6.17/fs/eventpoll.c
===================================================================
@@ -1587,7 +1587,6 @@
 	inode->i_uid = current->fsuid;
 	inode->i_gid = current->fsgid;
 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
-	inode->i_blksize = PAGE_SIZE;
 	return inode;
 
 eexit_1:
fs/libfs.c +0 -2
Index: linux-2.6.17/fs/libfs.c
===================================================================
@@ -385,7 +385,6 @@
 		return -ENOMEM;
 	inode->i_mode = S_IFDIR | 0755;
 	inode->i_uid = inode->i_gid = 0;
-	inode->i_blksize = PAGE_CACHE_SIZE;
 	inode->i_blocks = 0;
 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 	inode->i_op = &simple_dir_inode_operations;
@@ -407,7 +406,6 @@
 			goto out;
 		inode->i_mode = S_IFREG | files->mode;
 		inode->i_uid = inode->i_gid = 0;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 		inode->i_fop = files->ops;
fs/pipe.c +0 -1
Index: linux-2.6.17/fs/pipe.c
===================================================================
@@ -879,7 +879,6 @@
 	inode->i_uid = current->fsuid;
 	inode->i_gid = current->fsgid;
 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
-	inode->i_blksize = PAGE_SIZE;
 
 	return inode;
 
fs/super.c +1 -0
Index: linux-2.6.17/fs/super.c
===================================================================
@@ -86,6 +86,7 @@
 		s->s_qcop = sb_quotactl_ops;
 		s->s_op = &default_op;
 		s->s_time_gran = 1000000000;
+		s->s_blksize = PAGE_CACHE_SIZE;
 	}
 out:
 	return s;
arch/powerpc/platforms/cell/spufs/inode.c +0 -1
Index: linux-2.6.17/arch/powerpc/platforms/cell/spufs/inode.c
===================================================================
@@ -82,7 +82,6 @@
 	inode->i_mode = mode;
 	inode->i_uid = current->fsuid;
 	inode->i_gid = current->fsgid;
-	inode->i_blksize = PAGE_CACHE_SIZE;
 	inode->i_blocks = 0;
 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 out:
drivers/isdn/capi/capifs.c +0 -2
Index: linux-2.6.17/drivers/isdn/capi/capifs.c
===================================================================
@@ -104,7 +104,6 @@
 	inode->i_ino = 1;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 	inode->i_blocks = 0;
-	inode->i_blksize = 1024;
 	inode->i_uid = inode->i_gid = 0;
 	inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 	inode->i_op = &simple_dir_inode_operations;
@@ -149,7 +148,6 @@
 	if (!inode)
 		return;
 	inode->i_ino = number+2;
-	inode->i_blksize = 1024;
 	inode->i_uid = config.setuid ? config.uid : current->fsuid;
 	inode->i_gid = config.setgid ? config.gid : current->fsgid;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
drivers/misc/ibmasm/ibmasmfs.c +0 -1
Index: linux-2.6.17/drivers/misc/ibmasm/ibmasmfs.c
===================================================================
@@ -146,7 +146,6 @@
 	if (ret) {
 		ret->i_mode = mode;
 		ret->i_uid = ret->i_gid = 0;
-		ret->i_blksize = PAGE_CACHE_SIZE;
 		ret->i_blocks = 0;
 		ret->i_atime = ret->i_mtime = ret->i_ctime = CURRENT_TIME;
 	}
drivers/oprofile/oprofilefs.c +0 -1
Index: linux-2.6.17/drivers/oprofile/oprofilefs.c
===================================================================
@@ -31,7 +31,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = 0;
 		inode->i_gid = 0;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 	}
drivers/usb/core/inode.c +0 -1
Index: linux-2.6.17/drivers/usb/core/inode.c
===================================================================
@@ -250,7 +250,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = current->fsuid;
 		inode->i_gid = current->fsgid;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 		switch (mode & S_IFMT) {
drivers/usb/gadget/inode.c +0 -1
Index: linux-2.6.17/drivers/usb/gadget/inode.c
===================================================================
@@ -1965,7 +1965,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = default_uid;
 		inode->i_gid = default_gid;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_atime = inode->i_mtime = inode->i_ctime
 				= CURRENT_TIME;
fs/9p/vfs_inode.c +1 -3
Index: linux-2.6.17/fs/9p/vfs_inode.c
===================================================================
@@ -204,7 +204,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = current->fsuid;
 		inode->i_gid = current->fsgid;
-		inode->i_blksize = sb->s_blocksize;
 		inode->i_blocks = 0;
 		inode->i_rdev = 0;
 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
@@ -953,9 +952,8 @@
 
 	inode->i_size = stat->length;
 
-	inode->i_blksize = sb->s_blocksize;
 	inode->i_blocks =
-	    (inode->i_size + inode->i_blksize - 1) >> sb->s_blocksize_bits;
+	    (inode->i_size + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
 }
 
 /**
fs/adfs/inode.c +0 -1
Index: linux-2.6.17/fs/adfs/inode.c
===================================================================
@@ -269,7 +269,6 @@
 	inode->i_ino	 = obj->file_id;
 	inode->i_size	 = obj->size;
 	inode->i_nlink	 = 2;
-	inode->i_blksize = PAGE_SIZE;
 	inode->i_blocks	 = (inode->i_size + sb->s_blocksize - 1) >>
 			    sb->s_blocksize_bits;
 
fs/afs/inode.c +0 -1
Index: linux-2.6.17/fs/afs/inode.c
===================================================================
@@ -72,7 +72,6 @@
 	inode->i_ctime.tv_sec	= vnode->status.mtime_server;
 	inode->i_ctime.tv_nsec	= 0;
 	inode->i_atime		= inode->i_mtime = inode->i_ctime;
-	inode->i_blksize	= PAGE_CACHE_SIZE;
 	inode->i_blocks		= 0;
 	inode->i_version	= vnode->fid.unique;
 	inode->i_mapping->a_ops	= &afs_fs_aops;
fs/autofs/inode.c +0 -1
Index: linux-2.6.17/fs/autofs/inode.c
===================================================================
@@ -216,7 +216,6 @@
 	inode->i_nlink = 2;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 	inode->i_blocks = 0;
-	inode->i_blksize = 1024;
 
 	if ( ino == AUTOFS_ROOT_INO ) {
 		inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
fs/autofs4/inode.c +0 -1
Index: linux-2.6.17/fs/autofs4/inode.c
===================================================================
@@ -446,7 +446,6 @@
 		inode->i_uid = 0;
 		inode->i_gid = 0;
 	}
-	inode->i_blksize = PAGE_CACHE_SIZE;
 	inode->i_blocks = 0;
 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 
fs/befs/linuxvfs.c +0 -1
Index: linux-2.6.17/fs/befs/linuxvfs.c
===================================================================
@@ -365,7 +365,6 @@
 	inode->i_mtime.tv_nsec = 0;   /* lower 16 bits are not a time */	
 	inode->i_ctime = inode->i_mtime;
 	inode->i_atime = inode->i_mtime;
-	inode->i_blksize = befs_sb->block_size;
 
 	befs_ino->i_inode_num = fsrun_to_cpu(sb, raw_inode->inode_num);
 	befs_ino->i_parent = fsrun_to_cpu(sb, raw_inode->parent);
fs/bfs/dir.c +1 -1
Index: linux-2.6.17/fs/bfs/dir.c
===================================================================
@@ -102,7 +102,7 @@
 	inode->i_uid = current->fsuid;
 	inode->i_gid = (dir->i_mode & S_ISGID) ? dir->i_gid : current->fsgid;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
-	inode->i_blocks = inode->i_blksize = 0;
+	inode->i_blocks = 0;
 	inode->i_op = &bfs_file_inops;
 	inode->i_fop = &bfs_file_operations;
 	inode->i_mapping->a_ops = &bfs_aops;
fs/bfs/inode.c +0 -1
Index: linux-2.6.17/fs/bfs/inode.c
===================================================================
@@ -76,7 +76,6 @@
 	inode->i_size = BFS_FILESIZE(di);
 	inode->i_blocks = BFS_FILEBLOCKS(di);
         if (inode->i_size || inode->i_blocks) dprintf("Registered inode with %lld size, %ld blocks\n", inode->i_size, inode->i_blocks);
-	inode->i_blksize = PAGE_SIZE;
 	inode->i_atime.tv_sec =  le32_to_cpu(di->i_atime);
 	inode->i_mtime.tv_sec =  le32_to_cpu(di->i_mtime);
 	inode->i_ctime.tv_sec =  le32_to_cpu(di->i_ctime);
fs/configfs/inode.c +0 -1
Index: linux-2.6.17/fs/configfs/inode.c
===================================================================
@@ -136,7 +136,6 @@
 {
 	struct inode * inode = new_inode(configfs_sb);
 	if (inode) {
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_mapping->a_ops = &configfs_aops;
 		inode->i_mapping->backing_dev_info = &configfs_backing_dev_info;
fs/cramfs/inode.c +0 -1
Index: linux-2.6.17/fs/cramfs/inode.c
===================================================================
@@ -73,7 +73,6 @@
 	inode->i_uid = cramfs_inode->uid;
 	inode->i_size = cramfs_inode->size;
 	inode->i_blocks = (cramfs_inode->size - 1) / 512 + 1;
-	inode->i_blksize = PAGE_CACHE_SIZE;
 	inode->i_gid = cramfs_inode->gid;
 	/* Struct copy intentional */
 	inode->i_mtime = inode->i_atime = inode->i_ctime = zerotime;
fs/debugfs/inode.c +0 -1
Index: linux-2.6.17/fs/debugfs/inode.c
===================================================================
@@ -41,7 +41,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = 0;
 		inode->i_gid = 0;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 		switch (mode & S_IFMT) {
fs/devpts/inode.c +0 -2
Index: linux-2.6.17/fs/devpts/inode.c
===================================================================
@@ -113,7 +113,6 @@
 	inode->i_ino = 1;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 	inode->i_blocks = 0;
-	inode->i_blksize = 1024;
 	inode->i_uid = inode->i_gid = 0;
 	inode->i_mode = S_IFDIR | S_IRUGO | S_IXUGO | S_IWUSR;
 	inode->i_op = &simple_dir_inode_operations;
@@ -172,7 +171,6 @@
 		return -ENOMEM;
 
 	inode->i_ino = number+2;
-	inode->i_blksize = 1024;
 	inode->i_uid = config.setuid ? config.uid : current->fsuid;
 	inode->i_gid = config.setgid ? config.gid : current->fsgid;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
fs/ext2/ialloc.c +0 -1
Index: linux-2.6.17/fs/ext2/ialloc.c
===================================================================
@@ -575,7 +575,6 @@
 	inode->i_mode = mode;
 
 	inode->i_ino = ino;
-	inode->i_blksize = PAGE_SIZE;	/* This is the optimal IO size (for stat), not the fs block size */
 	inode->i_blocks = 0;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
 	memset(ei->i_data, 0, sizeof(ei->i_data));
fs/ext2/inode.c +0 -1
Index: linux-2.6.17/fs/ext2/inode.c
===================================================================
@@ -1094,7 +1094,6 @@
 		brelse (bh);
 		goto bad_inode;
 	}
-	inode->i_blksize = PAGE_SIZE;	/* This is the optimal IO size (for stat), not the fs block size */
 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
 	ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
fs/ext3/ialloc.c +0 -1
Index: linux-2.6.17/fs/ext3/ialloc.c
===================================================================
@@ -557,7 +557,6 @@
 
 	inode->i_ino = ino;
 	/* This is the optimal IO size (for stat), not the fs block size */
-	inode->i_blksize = PAGE_SIZE;
 	inode->i_blocks = 0;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
 
fs/ext3/inode.c +0 -3
Index: linux-2.6.17/fs/ext3/inode.c
===================================================================
@@ -2627,9 +2627,6 @@
 		 * recovery code: that's fine, we're about to complete
 		 * the process of deleting those. */
 	}
-	inode->i_blksize = PAGE_SIZE;	/* This is the optimal IO size
-					 * (for stat), not the fs block
-					 * size */  
 	inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
 	ei->i_flags = le32_to_cpu(raw_inode->i_flags);
 #ifdef EXT3_FRAGMENTS
fs/fat/inode.c +0 -3
Index: linux-2.6.17/fs/fat/inode.c
===================================================================
@@ -375,8 +375,6 @@
 			inode->i_flags |= S_IMMUTABLE;
 	}
 	MSDOS_I(inode)->i_attrs = de->attr & ATTR_UNUSED;
-	/* this is as close to the truth as we can get ... */
-	inode->i_blksize = sbi->cluster_size;
 	inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
 			   & ~((loff_t)sbi->cluster_size - 1)) >> 9;
 	inode->i_mtime.tv_sec =
@@ -1137,7 +1135,6 @@
 		MSDOS_I(inode)->i_start = 0;
 		inode->i_size = sbi->dir_entries * sizeof(struct msdos_dir_entry);
 	}
-	inode->i_blksize = sbi->cluster_size;
 	inode->i_blocks = ((inode->i_size + (sbi->cluster_size - 1))
 			   & ~((loff_t)sbi->cluster_size - 1)) >> 9;
 	MSDOS_I(inode)->i_logstart = 0;
fs/freevxfs/vxfs_inode.c +0 -1
Index: linux-2.6.17/fs/freevxfs/vxfs_inode.c
===================================================================
@@ -239,7 +239,6 @@
 	ip->i_ctime.tv_nsec = 0;
 	ip->i_mtime.tv_nsec = 0;
 
-	ip->i_blksize = PAGE_SIZE;
 	ip->i_blocks = vip->vii_blocks;
 	ip->i_generation = vip->vii_gen;
 
fs/fuse/inode.c +0 -1
Index: linux-2.6.17/fs/fuse/inode.c
===================================================================
@@ -115,7 +115,6 @@
 	inode->i_uid     = attr->uid;
 	inode->i_gid     = attr->gid;
 	i_size_write(inode, attr->size);
-	inode->i_blksize = PAGE_CACHE_SIZE;
 	inode->i_blocks  = attr->blocks;
 	inode->i_atime.tv_sec   = attr->atime;
 	inode->i_atime.tv_nsec  = attr->atimensec;
fs/hfs/inode.c +0 -2
Index: linux-2.6.17/fs/hfs/inode.c
===================================================================
@@ -154,7 +154,6 @@
 	inode->i_gid = current->fsgid;
 	inode->i_nlink = 1;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
-	inode->i_blksize = HFS_SB(sb)->alloc_blksz;
 	HFS_I(inode)->flags = 0;
 	HFS_I(inode)->rsrc_inode = NULL;
 	HFS_I(inode)->fs_blocks = 0;
@@ -284,7 +283,6 @@
 	inode->i_uid = hsb->s_uid;
 	inode->i_gid = hsb->s_gid;
 	inode->i_nlink = 1;
-	inode->i_blksize = HFS_SB(inode->i_sb)->alloc_blksz;
 
 	if (idata->key)
 		HFS_I(inode)->cat_key = *idata->key;
fs/hfsplus/inode.c +0 -2
Index: linux-2.6.17/fs/hfsplus/inode.c
===================================================================
@@ -304,7 +304,6 @@
 	inode->i_gid = current->fsgid;
 	inode->i_nlink = 1;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
-	inode->i_blksize = HFSPLUS_SB(sb).alloc_blksz;
 	INIT_LIST_HEAD(&HFSPLUS_I(inode).open_dir_list);
 	init_MUTEX(&HFSPLUS_I(inode).extents_lock);
 	atomic_set(&HFSPLUS_I(inode).opencnt, 0);
@@ -407,7 +406,6 @@
 	type = hfs_bnode_read_u16(fd->bnode, fd->entryoffset);
 
 	HFSPLUS_I(inode).dev = 0;
-	inode->i_blksize = HFSPLUS_SB(inode->i_sb).alloc_blksz;
 	if (type == HFSPLUS_FOLDER) {
 		struct hfsplus_cat_folder *folder = &entry.folder;
 
fs/hpfs/inode.c +0 -1
Index: linux-2.6.17/fs/hpfs/inode.c
===================================================================
@@ -17,7 +17,6 @@
 	i->i_gid = hpfs_sb(sb)->sb_gid;
 	i->i_mode = hpfs_sb(sb)->sb_mode;
 	hpfs_inode->i_conv = hpfs_sb(sb)->sb_conv;
-	i->i_blksize = 512;
 	i->i_size = -1;
 	i->i_blocks = -1;
 	
fs/hppfs/hppfs_kern.c +0 -1
Index: linux-2.6.17/fs/hppfs/hppfs_kern.c
===================================================================
@@ -152,7 +152,6 @@
 	ino->i_mode = proc_ino->i_mode;
 	ino->i_nlink = proc_ino->i_nlink;
 	ino->i_size = proc_ino->i_size;
-	ino->i_blksize = proc_ino->i_blksize;
 	ino->i_blocks = proc_ino->i_blocks;
 }
 
fs/isofs/inode.c +1 -2
Index: linux-2.6.17/fs/isofs/inode.c
===================================================================
@@ -1237,7 +1237,7 @@
 	}
 	inode->i_uid = sbi->s_uid;
 	inode->i_gid = sbi->s_gid;
-	inode->i_blocks = inode->i_blksize = 0;
+	inode->i_blocks = 0;
 
 	ei->i_format_parm[0] = 0;
 	ei->i_format_parm[1] = 0;
@@ -1293,7 +1293,6 @@
 			      isonum_711 (de->ext_attr_length));
 
 	/* Set the number of blocks for stat() - should be done before RR */
-	inode->i_blksize = PAGE_CACHE_SIZE; /* For stat() only */
 	inode->i_blocks  = (inode->i_size + 511) >> 9;
 
 	/*
fs/jffs/inode-v23.c +0 -2
Index: linux-2.6.17/fs/jffs/inode-v23.c
===================================================================
@@ -364,7 +364,6 @@
 	inode->i_ctime.tv_nsec = 0;
 	inode->i_mtime.tv_nsec = 0;
 	inode->i_atime.tv_nsec = 0;
-	inode->i_blksize = PAGE_SIZE;
 	inode->i_blocks = (inode->i_size + 511) >> 9;
 
 	f = jffs_find_file(c, raw_inode->ino);
@@ -1706,7 +1705,6 @@
 	inode->i_mtime.tv_nsec = 
 	inode->i_ctime.tv_nsec = 0;
 
-	inode->i_blksize = PAGE_SIZE;
 	inode->i_blocks = (inode->i_size + 511) >> 9;
 	if (S_ISREG(inode->i_mode)) {
 		inode->i_op = &jffs_file_inode_operations;
fs/jffs2/fs.c +0 -2
Index: linux-2.6.17/fs/jffs2/fs.c
===================================================================
@@ -254,7 +254,6 @@
 
 	inode->i_nlink = f->inocache->nlink;
 
-	inode->i_blksize = PAGE_SIZE;
 	inode->i_blocks = (inode->i_size + 511) >> 9;
 
 	switch (inode->i_mode & S_IFMT) {
@@ -430,7 +429,6 @@
 	inode->i_atime = inode->i_ctime = inode->i_mtime = CURRENT_TIME_SEC;
 	ri->atime = ri->mtime = ri->ctime = cpu_to_je32(I_SEC(inode->i_mtime));
 
-	inode->i_blksize = PAGE_SIZE;
 	inode->i_blocks = 0;
 	inode->i_size = 0;
 
fs/minix/bitmap.c +1 -1
Index: linux-2.6.17/fs/minix/bitmap.c
===================================================================
@@ -254,7 +254,7 @@
 	inode->i_gid = (dir->i_mode & S_ISGID) ? dir->i_gid : current->fsgid;
 	inode->i_ino = j;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
-	inode->i_blocks = inode->i_blksize = 0;
+	inode->i_blocks = 0;
 	memset(&minix_i(inode)->u, 0, sizeof(minix_i(inode)->u));
 	insert_inode_hash(inode);
 	mark_inode_dirty(inode);
fs/minix/inode.c +2 -2
Index: linux-2.6.17/fs/minix/inode.c
===================================================================
@@ -392,7 +392,7 @@
 	inode->i_mtime.tv_nsec = 0;
 	inode->i_atime.tv_nsec = 0;
 	inode->i_ctime.tv_nsec = 0;
-	inode->i_blocks = inode->i_blksize = 0;
+	inode->i_blocks = 0;
 	for (i = 0; i < 9; i++)
 		minix_inode->u.i1_data[i] = raw_inode->i_zone[i];
 	minix_set_inode(inode, old_decode_dev(raw_inode->i_zone[0]));
@@ -425,7 +425,7 @@
 	inode->i_mtime.tv_nsec = 0;
 	inode->i_atime.tv_nsec = 0;
 	inode->i_ctime.tv_nsec = 0;
-	inode->i_blocks = inode->i_blksize = 0;
+	inode->i_blocks = 0;
 	for (i = 0; i < 10; i++)
 		minix_inode->u.i2_data[i] = raw_inode->i_zone[i];
 	minix_set_inode(inode, old_decode_dev(raw_inode->i_zone[0]));
fs/ntfs/inode.c +0 -4
Index: linux-2.6.17/fs/ntfs/inode.c
===================================================================
@@ -540,8 +540,6 @@
 
 	/* Setup the generic vfs inode parts now. */
 
-	/* This is the optimal IO size (for stat), not the fs block size. */
-	vi->i_blksize = PAGE_CACHE_SIZE;
 	/*
 	 * This is for checking whether an inode has changed w.r.t. a file so
 	 * that the file can be updated if necessary (compare with f_version).
@@ -1218,7 +1216,6 @@
 	base_ni = NTFS_I(base_vi);
 
 	/* Just mirror the values from the base inode. */
-	vi->i_blksize	= base_vi->i_blksize;
 	vi->i_version	= base_vi->i_version;
 	vi->i_uid	= base_vi->i_uid;
 	vi->i_gid	= base_vi->i_gid;
@@ -1488,7 +1485,6 @@
 	ni	= NTFS_I(vi);
 	base_ni = NTFS_I(base_vi);
 	/* Just mirror the values from the base inode. */
-	vi->i_blksize	= base_vi->i_blksize;
 	vi->i_version	= base_vi->i_version;
 	vi->i_uid	= base_vi->i_uid;
 	vi->i_gid	= base_vi->i_gid;
fs/ntfs/mft.c +0 -5
Index: linux-2.6.17/fs/ntfs/mft.c
===================================================================
@@ -2638,11 +2638,6 @@
 		}
 		vi->i_ino = bit;
 		/*
-		 * This is the optimal IO size (for stat), not the fs block
-		 * size.
-		 */
-		vi->i_blksize = PAGE_CACHE_SIZE;
-		/*
 		 * This is for checking whether an inode has changed w.r.t. a
 		 * file so that the file can be updated if necessary (compare
 		 * with f_version).
fs/ocfs2/dlm/dlmfs.c +0 -2
Index: linux-2.6.17/fs/ocfs2/dlm/dlmfs.c
===================================================================
@@ -335,7 +335,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = current->fsuid;
 		inode->i_gid = current->fsgid;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
@@ -362,7 +361,6 @@
 	inode->i_mode = mode;
 	inode->i_uid = current->fsuid;
 	inode->i_gid = current->fsgid;
-	inode->i_blksize = PAGE_CACHE_SIZE;
 	inode->i_blocks = 0;
 	inode->i_mapping->backing_dev_info = &dlmfs_backing_dev_info;
 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
fs/ocfs2/inode.c +0 -2
Index: linux-2.6.17/fs/ocfs2/inode.c
===================================================================
@@ -251,7 +251,6 @@
 	inode->i_mode = le16_to_cpu(fe->i_mode);
 	inode->i_uid = le32_to_cpu(fe->i_uid);
 	inode->i_gid = le32_to_cpu(fe->i_gid);
-	inode->i_blksize = (u32)osb->s_clustersize;
 
 	/* Fast symlinks will have i_size but no allocated clusters. */
 	if (S_ISLNK(inode->i_mode) && !fe->i_clusters)
@@ -1174,7 +1173,6 @@
 	inode->i_uid = le32_to_cpu(fe->i_uid);
 	inode->i_gid = le32_to_cpu(fe->i_gid);
 	inode->i_mode = le16_to_cpu(fe->i_mode);
-	inode->i_blksize = (u32) osb->s_clustersize;
 	if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
 		inode->i_blocks = 0;
 	else
fs/ocfs2/super.c +1 -1
Index: linux-2.6.17/fs/ocfs2/super.c
===================================================================
@@ -1390,7 +1390,7 @@
 	/* get some pseudo constants for clustersize bits */
 	osb->s_clustersize_bits =
 		le32_to_cpu(di->id2.i_super.s_clustersize_bits);
-	osb->s_clustersize = 1 << osb->s_clustersize_bits;
+	osb->s_blksize = osb->s_clustersize = 1 << osb->s_clustersize_bits;
 	mlog(0, "clusterbits=%d\n", osb->s_clustersize_bits);
 
 	if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
fs/qnx4/inode.c +0 -1
Index: linux-2.6.17/fs/qnx4/inode.c
===================================================================
@@ -496,7 +496,6 @@
 	inode->i_ctime.tv_sec   = le32_to_cpu(raw_inode->di_ctime);
 	inode->i_ctime.tv_nsec = 0;
 	inode->i_blocks  = le32_to_cpu(raw_inode->di_first_xtnt.xtnt_size);
-	inode->i_blksize = QNX4_DIR_ENTRY_SIZE;
 
 	memcpy(qnx4_inode, raw_inode, QNX4_DIR_ENTRY_SIZE);
 	if (S_ISREG(inode->i_mode)) {
fs/ramfs/inode.c +0 -1
Index: linux-2.6.17/fs/ramfs/inode.c
===================================================================
@@ -58,7 +58,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = current->fsuid;
 		inode->i_gid = current->fsgid;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_mapping->a_ops = &ramfs_aops;
 		inode->i_mapping->backing_dev_info = &ramfs_backing_dev_info;
fs/reiserfs/inode.c +0 -4
Index: linux-2.6.17/fs/reiserfs/inode.c
===================================================================
@@ -18,8 +18,6 @@
 #include <linux/writeback.h>
 #include <linux/quotaops.h>
 
-extern int reiserfs_default_io_size;	/* default io size devuned in super.c */
-
 static int reiserfs_commit_write(struct file *f, struct page *page,
 				 unsigned from, unsigned to);
 static int reiserfs_prepare_write(struct file *f, struct page *page,
@@ -1131,7 +1129,6 @@
 	ih = PATH_PITEM_HEAD(path);
 
 	copy_key(INODE_PKEY(inode), &(ih->ih_key));
-	inode->i_blksize = reiserfs_default_io_size;
 
 	INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list));
 	REISERFS_I(inode)->i_flags = 0;
@@ -1886,7 +1883,6 @@
 	}
 	// these do not go to on-disk stat data
 	inode->i_ino = le32_to_cpu(ih.ih_key.k_objectid);
-	inode->i_blksize = reiserfs_default_io_size;
 
 	// store in in-core inode the key of stat data and version all
 	// object items will have (directory items will have old offset
fs/reiserfs/super.c +8 -8
Index: linux-2.6.17/fs/reiserfs/super.c
===================================================================
@@ -726,12 +726,6 @@
 	{NULL, 0, 0},
 };
 
-int reiserfs_default_io_size = 128 * 1024;	/* Default recommended I/O size is 128k.
-						   There might be broken applications that are
-						   confused by this. Use nolargeio mount option
-						   to get usual i/o size = PAGE_SIZE.
-						 */
-
 /* proceed only one option from a list *cur - string containing of mount options
    opts - array of options which are accepted
    opt_arg - if option is found and requires an argument and if it is specifed
@@ -959,6 +953,12 @@
 			*commit_max_age = (unsigned int)val;
 		}
 
+		/*
+		 * Default recommended I/O size is 128k.
+		 * There might be broken applications that are
+		 * confused by this. Use nolargeio mount option
+		 * to get usual i/o size = PAGE_SIZE.
+		 */
 		if (c == 'w') {
 			char *p = NULL;
 			int val = simple_strtoul(arg, &p, 0);
@@ -970,9 +970,9 @@
 				return 0;
 			}
 			if (val)
-				reiserfs_default_io_size = PAGE_SIZE;
+				s->s_blksize = PAGE_SIZE;
 			else
-				reiserfs_default_io_size = 128 * 1024;
+				s->s_blksize = 128 * 1024;
 		}
 
 		if (c == 'j') {
fs/sysfs/inode.c +0 -1
Index: linux-2.6.17/fs/sysfs/inode.c
===================================================================
@@ -113,7 +113,6 @@
 {
 	struct inode * inode = new_inode(sysfs_sb);
 	if (inode) {
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_mapping->a_ops = &sysfs_aops;
 		inode->i_mapping->backing_dev_info = &sysfs_backing_dev_info;
fs/sysv/ialloc.c +1 -1
Index: linux-2.6.17/fs/sysv/ialloc.c
===================================================================
@@ -170,7 +170,7 @@
 	inode->i_uid = current->fsuid;
 	inode->i_ino = fs16_to_cpu(sbi, ino);
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
-	inode->i_blocks = inode->i_blksize = 0;
+	inode->i_blocks = 0;
 	memset(SYSV_I(inode)->i_data, 0, sizeof(SYSV_I(inode)->i_data));
 	SYSV_I(inode)->i_dir_start_lookup = 0;
 	insert_inode_hash(inode);
fs/sysv/inode.c +1 -1
Index: linux-2.6.17/fs/sysv/inode.c
===================================================================
@@ -200,7 +200,7 @@
 	inode->i_ctime.tv_nsec = 0;
 	inode->i_atime.tv_nsec = 0;
 	inode->i_mtime.tv_nsec = 0;
-	inode->i_blocks = inode->i_blksize = 0;
+	inode->i_blocks = 0;
 
 	si = SYSV_I(inode);
 	for (block = 0; block < 10+1+1+1; block++)
fs/udf/ialloc.c +0 -1
Index: linux-2.6.17/fs/udf/ialloc.c
===================================================================
@@ -120,7 +120,6 @@
 	UDF_I_LOCATION(inode).logicalBlockNum = block;
 	UDF_I_LOCATION(inode).partitionReferenceNum = UDF_I_LOCATION(dir).partitionReferenceNum;
 	inode->i_ino = udf_get_lb_pblock(sb, UDF_I_LOCATION(inode), 0);
-	inode->i_blksize = PAGE_SIZE;
 	inode->i_blocks = 0;
 	UDF_I_LENEATTR(inode) = 0;
 	UDF_I_LENALLOC(inode) = 0;
fs/udf/inode.c +0 -2
Index: linux-2.6.17/fs/udf/inode.c
===================================================================
@@ -916,8 +916,6 @@
 	 *      i_nlink = 1
 	 *      i_op = NULL;
 	 */
-	inode->i_blksize = PAGE_SIZE;
-
 	bh = udf_read_ptagged(inode->i_sb, UDF_I_LOCATION(inode), 0, &ident);
 
 	if (!bh)
fs/ufs/ialloc.c +0 -1
Index: linux-2.6.17/fs/ufs/ialloc.c
===================================================================
@@ -263,7 +263,6 @@
 		inode->i_gid = current->fsgid;
 
 	inode->i_ino = cg * uspi->s_ipg + bit;
-	inode->i_blksize = PAGE_SIZE;	/* This is the optimal IO size (for stat), not the fs block size */
 	inode->i_blocks = 0;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
 	ufsi->i_flags = UFS_I(dir)->i_flags;
fs/ufs/inode.c +0 -2
Index: linux-2.6.17/fs/ufs/inode.c
===================================================================
@@ -596,7 +596,6 @@
 	inode->i_atime.tv_nsec = 0;
 	inode->i_ctime.tv_nsec = 0;
 	inode->i_blocks = fs32_to_cpu(sb, ufs_inode->ui_blocks);
-	inode->i_blksize = PAGE_SIZE;   /* This is the optimal IO size (for stat) */
 	inode->i_version++;
 	ufsi->i_flags = fs32_to_cpu(sb, ufs_inode->ui_flags);
 	ufsi->i_gen = fs32_to_cpu(sb, ufs_inode->ui_gen);
@@ -668,7 +667,6 @@
 	inode->i_atime.tv_nsec = 0;
 	inode->i_ctime.tv_nsec = 0;
 	inode->i_blocks = fs64_to_cpu(sb, ufs2_inode->ui_blocks);
-	inode->i_blksize = PAGE_SIZE; /*This is the optimal IO size(for stat)*/
 
 	inode->i_version++;
 	ufsi->i_flags = fs32_to_cpu(sb, ufs2_inode->ui_flags);
include/linux/nfsd/nfsfh.h +2 -2
Index: linux-2.6.17/include/linux/nfsd/nfsfh.h
===================================================================
@@ -270,8 +270,8 @@
 	fhp->fh_post_uid	= inode->i_uid;
 	fhp->fh_post_gid	= inode->i_gid;
 	fhp->fh_post_size       = inode->i_size;
-	if (inode->i_blksize) {
-		fhp->fh_post_blksize    = inode->i_blksize;
+	if (inode->i_sb->s_blksize) {
+		fhp->fh_post_blksize    = inode->i_sb->s_blksize;
 		fhp->fh_post_blocks     = inode->i_blocks;
 	} else {
 		fhp->fh_post_blksize    = BLOCK_SIZE;
ipc/mqueue.c +0 -1
Index: linux-2.6.17/ipc/mqueue.c
===================================================================
@@ -112,7 +112,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = current->fsuid;
 		inode->i_gid = current->fsgid;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_mtime = inode->i_ctime = inode->i_atime =
 				CURRENT_TIME;
kernel/cpuset.c +0 -1
Index: linux-2.6.17/kernel/cpuset.c
===================================================================
@@ -289,7 +289,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = current->fsuid;
 		inode->i_gid = current->fsgid;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 		inode->i_mapping->backing_dev_info = &cpuset_backing_dev_info;
mm/shmem.c +0 -1
Index: linux-2.6.17/mm/shmem.c
===================================================================
@@ -1360,7 +1360,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = current->fsuid;
 		inode->i_gid = current->fsgid;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_mapping->a_ops = &shmem_aops;
 		inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
net/sunrpc/rpc_pipe.c +0 -1
Index: linux-2.6.17/net/sunrpc/rpc_pipe.c
===================================================================
@@ -491,7 +491,6 @@
 		return NULL;
 	inode->i_mode = mode;
 	inode->i_uid = inode->i_gid = 0;
-	inode->i_blksize = PAGE_CACHE_SIZE;
 	inode->i_blocks = 0;
 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 	switch(mode & S_IFMT) {
security/inode.c +0 -1
Index: linux-2.6.17/security/inode.c
===================================================================
@@ -65,7 +65,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = 0;
 		inode->i_gid = 0;
-		inode->i_blksize = PAGE_CACHE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 		switch (mode & S_IFMT) {
security/selinux/selinuxfs.c +0 -1
Index: linux-2.6.17/security/selinux/selinuxfs.c
===================================================================
@@ -707,7 +707,6 @@
 	if (ret) {
 		ret->i_mode = mode;
 		ret->i_uid = ret->i_gid = 0;
-		ret->i_blksize = PAGE_CACHE_SIZE;
 		ret->i_blocks = 0;
 		ret->i_atime = ret->i_mtime = ret->i_ctime = CURRENT_TIME;
 	}
fs/cifs/cifsfs.c +1 -2
Index: linux-2.6.17/fs/cifs/cifsfs.c
===================================================================
@@ -112,7 +112,7 @@
 #ifdef CONFIG_CIFS_QUOTA
 	sb->s_qcop = &cifs_quotactl_ops;
 #endif
-	sb->s_blocksize = CIFS_MAX_MSGSIZE;
+	sb->s_blksize = sb->s_blocksize = CIFS_MAX_MSGSIZE;
 	sb->s_blocksize_bits = 14;	/* default 2**14 = CIFS_MAX_MSGSIZE */
 	inode = iget(sb, ROOT_I);
 
@@ -254,7 +254,6 @@
 	file data or metadata */
 	cifs_inode->clientCanCacheRead = FALSE;
 	cifs_inode->clientCanCacheAll = FALSE;
-	cifs_inode->vfs_inode.i_blksize = CIFS_MAX_MSGSIZE;
 	cifs_inode->vfs_inode.i_blkbits = 14;  /* 2**14 = CIFS_MAX_MSGSIZE */
 	cifs_inode->vfs_inode.i_flags = S_NOATIME | S_NOCMTIME;
 	INIT_LIST_HEAD(&cifs_inode->openFileList);
fs/cifs/readdir.c +2 -3
Index: linux-2.6.17/fs/cifs/readdir.c
===================================================================
@@ -197,10 +197,9 @@
 
 	if (allocation_size < end_of_file)
 		cFYI(1, ("May be sparse file, allocation less than file size"));
-	cFYI(1, ("File Size %ld and blocks %llu and blocksize %ld",
+	cFYI(1, ("File Size %ld and blocks %llu",
 		(unsigned long)tmp_inode->i_size,
-		(unsigned long long)tmp_inode->i_blocks,
-		tmp_inode->i_blksize));
+		(unsigned long long)tmp_inode->i_blocks));
 	if (S_ISREG(tmp_inode->i_mode)) {
 		cFYI(1, ("File inode"));
 		tmp_inode->i_op = &cifs_file_inode_ops;
fs/devfs/base.c +0 -2
Index: linux-2.6.17/fs/devfs/base.c
===================================================================
@@ -693,7 +693,6 @@
 #define FIRST_INODE 1
 
 #define STRING_LENGTH 256
-#define FAKE_BLOCK_SIZE 1024
 #define POISON_PTR ( *(void **) poison_array )
 #define MAGIC_VALUE 0x327db823
 
@@ -1925,7 +1924,6 @@
 	DPRINTK(DEBUG_I_GET, "(%d): VFS inode: %p  devfs_entry: %p\n",
 		(int)inode->i_ino, inode, de);
 	inode->i_blocks = 0;
-	inode->i_blksize = FAKE_BLOCK_SIZE;
 	inode->i_op = &devfs_iops;
 	inode->i_mode = de->mode;
 	if (S_ISDIR(de->mode)) {
fs/hugetlbfs/inode.c +1 -1
Index: linux-2.6.17/fs/hugetlbfs/inode.c
===================================================================
@@ -361,7 +361,6 @@
 		inode->i_mode = mode;
 		inode->i_uid = uid;
 		inode->i_gid = gid;
-		inode->i_blksize = HPAGE_SIZE;
 		inode->i_blocks = 0;
 		inode->i_mapping->a_ops = &hugetlbfs_aops;
 		inode->i_mapping->backing_dev_info =&hugetlbfs_backing_dev_info;
@@ -678,6 +677,7 @@
 	sb->s_magic = HUGETLBFS_MAGIC;
 	sb->s_op = &hugetlbfs_ops;
 	sb->s_time_gran = 1;
+	sb->s_blksize = HPAGE_SIZE;
 	inode = hugetlbfs_get_inode(sb, config.uid, config.gid,
 					S_IFDIR | config.mode, 0);
 	if (!inode)
fs/jfs/jfs_extent.c +1 -1
Index: linux-2.6.17/fs/jfs/jfs_extent.c
===================================================================
@@ -468,7 +468,7 @@
 int extFill(struct inode *ip, xad_t * xp)
 {
 	int rc, nbperpage = JFS_SBI(ip->i_sb)->nbperpage;
-	s64 blkno = offsetXAD(xp) >> ip->i_blksize;
+	s64 blkno = offsetXAD(xp) >> ip->i_blkbits;
 
 //      assert(ISSPARSE(ip));
 
fs/jfs/jfs_imap.c +0 -1
Index: linux-2.6.17/fs/jfs/jfs_imap.c
===================================================================
@@ -3115,7 +3115,6 @@
 	ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec);
 	ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec);
 	ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec);
-	ip->i_blksize = ip->i_sb->s_blocksize;
 	ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks));
 	ip->i_generation = le32_to_cpu(dip->di_gen);
 
fs/jfs/jfs_inode.c +0 -1
Index: linux-2.6.17/fs/jfs/jfs_inode.c
===================================================================
@@ -115,7 +115,6 @@
 	}
 	jfs_inode->mode2 |= mode;
 
-	inode->i_blksize = sb->s_blocksize;
 	inode->i_blocks = 0;
 	inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 	jfs_inode->otime = inode->i_ctime.tv_sec;
fs/jfs/jfs_metapage.c +1 -1
Index: linux-2.6.17/fs/jfs/jfs_metapage.c
===================================================================
@@ -257,7 +257,7 @@
 	int rc = 0;
 	int xflag;
 	s64 xaddr;
-	sector_t file_blocks = (inode->i_size + inode->i_blksize - 1) >>
+	sector_t file_blocks = (inode->i_size + inode->i_sb->s_blocksize - 1) >>
 			       inode->i_blkbits;
 
 	if (lblock >= file_blocks)
fs/ncpfs/inode.c +1 -1
Index: linux-2.6.17/fs/ncpfs/inode.c
===================================================================
@@ -225,7 +225,6 @@
 	inode->i_nlink = 1;
 	inode->i_uid = server->m.uid;
 	inode->i_gid = server->m.gid;
-	inode->i_blksize = NCP_BLOCK_SIZE;
 
 	ncp_update_dates(inode, &nwinfo->i);
 	ncp_update_inode(inode, nwinfo);
@@ -493,6 +492,7 @@
 	sb->s_blocksize_bits = 10;
 	sb->s_magic = NCP_SUPER_MAGIC;
 	sb->s_op = &ncp_sops;
+	sb->s_blksize = NCP_BLOCK_SIZE;
 
 	server = NCP_SBP(sb);
 	memset(server, 0, sizeof(*server));
drivers/block/loop.c +5 -2
Index: linux-2.6.17/drivers/block/loop.c
===================================================================
@@ -664,7 +664,8 @@
 
 	mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
 	lo->lo_backing_file = file;
-	lo->lo_blocksize = mapping->host->i_blksize;
+	lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
+		mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
 	lo->old_gfp_mask = mapping_gfp_mask(mapping);
 	mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
 	complete(&p->wait);
@@ -796,7 +797,9 @@
 		if (!(lo_flags & LO_FLAGS_USE_AOPS) && !file->f_op->write)
 			lo_flags |= LO_FLAGS_READ_ONLY;
 
-		lo_blocksize = inode->i_blksize;
+		lo_blocksize = S_ISBLK(inode->i_mode) ?
+			inode->i_bdev->bd_block_size : PAGE_SIZE;
+
 		error = 0;
 	} else {
 		goto out_putf;
fs/coda/coda_linux.c +0 -2
Index: linux-2.6.17/fs/coda/coda_linux.c
===================================================================
@@ -110,8 +110,6 @@
 	        inode->i_nlink = attr->va_nlink;
 	if (attr->va_size != -1)
 	        inode->i_size = attr->va_size;
-	if (attr->va_blocksize != -1)
-		inode->i_blksize = attr->va_blocksize;
 	if (attr->va_size != -1)
 		inode->i_blocks = (attr->va_size + 511) >> 9;
 	if (attr->va_atime.tv_sec != -1) 
fs/smbfs/inode.c +0 -2
Index: linux-2.6.17/fs/smbfs/inode.c
===================================================================
@@ -168,7 +168,6 @@
 	fattr->f_mtime	= inode->i_mtime;
 	fattr->f_ctime	= inode->i_ctime;
 	fattr->f_atime	= inode->i_atime;
-	fattr->f_blksize= inode->i_blksize;
 	fattr->f_blocks	= inode->i_blocks;
 
 	fattr->attr	= SMB_I(inode)->attr;
@@ -202,7 +201,6 @@
 	inode->i_uid	= fattr->f_uid;
 	inode->i_gid	= fattr->f_gid;
 	inode->i_ctime	= fattr->f_ctime;
-	inode->i_blksize= fattr->f_blksize;
 	inode->i_blocks = fattr->f_blocks;
 	inode->i_size	= fattr->f_size;
 	inode->i_mtime	= fattr->f_mtime;
fs/smbfs/proc.c +0 -1
Index: linux-2.6.17/fs/smbfs/proc.c
===================================================================
@@ -1826,7 +1826,6 @@
 	fattr->f_nlink = 1;
 	fattr->f_uid = server->mnt->uid;
 	fattr->f_gid = server->mnt->gid;
-	fattr->f_blksize = SMB_ST_BLKSIZE;
 	fattr->f_unix = 0;
 }
 
include/linux/smb.h +0 -1
Index: linux-2.6.17/include/linux/smb.h
===================================================================
@@ -88,7 +88,6 @@
 	struct timespec	f_atime;
 	struct timespec f_mtime;
 	struct timespec f_ctime;
-	unsigned long	f_blksize;
 	unsigned long	f_blocks;
 	int		f_unix;
 };
fs/hostfs/hostfs_kern.c +0 -1
Index: linux-2.6.17/fs/hostfs/hostfs_kern.c
===================================================================
@@ -156,7 +156,6 @@
 	ino->i_mode = i_mode;
 	ino->i_nlink = i_nlink;
 	ino->i_size = i_size;
-	ino->i_blksize = i_blksize;
 	ino->i_blocks = i_blocks;
 	return(0);
 }
fs/nfs/inode.c +0 -4
Index: linux-2.6.17/fs/nfs/inode.c
===================================================================
@@ -910,10 +910,8 @@
 			 * report the blocks in 512byte units
 			 */
 			inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
-			inode->i_blksize = inode->i_sb->s_blocksize;
 		} else {
 			inode->i_blocks = fattr->du.nfs2.blocks;
-			inode->i_blksize = fattr->du.nfs2.blocksize;
 		}
 		nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
 		nfsi->attrtimeo_timestamp = jiffies;
@@ -1606,10 +1604,8 @@
 		 * report the blocks in 512byte units
 		 */
 		inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
-		inode->i_blksize = inode->i_sb->s_blocksize;
  	} else {
  		inode->i_blocks = fattr->du.nfs2.blocks;
- 		inode->i_blksize = fattr->du.nfs2.blocksize;
  	}
 
 	if ((fattr->valid & NFS_ATTR_FATTR_V4)) {
fs/xfs/linux-2.6/xfs_super.c +0 -1
Index: linux-2.6.17/fs/xfs/linux-2.6/xfs_super.c
===================================================================
@@ -173,7 +173,6 @@
 		break;
 	}
 
-	inode->i_blksize = xfs_preferred_iosize(mp);
 	inode->i_generation = ip->i_d.di_gen;
 	i_size_write(inode, ip->i_d.di_size);
 	inode->i_blocks =
fs/xfs/linux-2.6/xfs_vnode.c +0 -1
Index: linux-2.6.17/fs/xfs/linux-2.6/xfs_vnode.c
===================================================================
@@ -106,7 +106,6 @@
 	inode->i_blocks	    = vap->va_nblocks;
 	inode->i_mtime	    = vap->va_mtime;
 	inode->i_ctime	    = vap->va_ctime;
-	inode->i_blksize    = vap->va_blocksize;
 	if (vap->va_xflags & XFS_XFLAG_IMMUTABLE)
 		inode->i_flags |= S_IMMUTABLE;
 	else
drivers/infiniband/hw/ipath/ipath_fs.c +0 -1
Index: linux-2.6.17/drivers/infiniband/hw/ipath/ipath_fs.c
===================================================================
@@ -61,7 +61,6 @@
 	inode->i_mode = mode;
 	inode->i_uid = 0;
 	inode->i_gid = 0;
-	inode->i_blksize = PAGE_CACHE_SIZE;
 	inode->i_blocks = 0;
 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 	inode->i_private = data;

--

#22 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

Quoted 15 lines · Theodore Tso

This eliminates the i_blksize field from struct inode and uses default value in super->s_blksize. Filesystems that want to provide a per-inode st_blksize can do so by providing their own getattr routine instead of using the generic_fillattr() function.

Index: linux-2.6.17/include/linux/nfsd/nfsfh.h
===================================================================
--- linux-2.6.17.orig/include/linux/nfsd/nfsfh.h 2006-06-18
19:37:14.000000000 -0400
+++ linux-2.6.17/include/linux/nfsd/nfsfh.h 2006-06-18
19:53:54.000000000 -0400

include/linux/nfsd/nfsfh.h line 277

Quoted 10 lines
include/linux/nfsd/nfsfh.h +2 -2
@@ -270,8 +270,8 @@
        fhp->fh_post_uid        = inode->i_uid;
        fhp->fh_post_gid        = inode->i_gid;
        fhp->fh_post_size       = inode->i_size;
-       if (inode->i_blksize) {
-               fhp->fh_post_blksize    = inode->i_blksize;
+       if (inode->i_sb->s_blksize) {
+               fhp->fh_post_blksize    = inode->i_sb->s_blksize;
                fhp->fh_post_blocks     = inode->i_blocks;
        } else {
                fhp->fh_post_blksize    = BLOCK_SIZE;

Isn't this a behavioral change? If a filesystem chooses to provide i_blksize via a getattr method, it will not show on nfs mounts.

Signature · 1 line
error compiling committee.c: too many arguments to function
#23 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

Quoted 3 lines · Avi Kivity

Isn't this a behavioral change? If a filesystem chooses to provide i_blksize via a getattr method, it will not show on nfs mounts.

There's actually a philosophical question hiding here over what's the right thing to do with how st_blksize should be handled over NFS --- st_blksize is supposed to be the "optimum I/O size"; the question is what is the right answer that we report in both directions. For example, it doesn't matter if we're exporting a Ultra Fast whiz-bang cluster filesystem with a stripsize of 100 gigabytes. If wsize is 1024, it might be that we shouldn't be sending back an st_blksize that's really big.

So I'm not pretending that what we have in the patch is the right thing, only that I'm not entirely sure we were ever doing the right thing here.

- Ted

#24 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

In reply to Theodore Tso
Quoted 7 lines · Theodore Tso

This eliminates the i_blksize field from struct inode and uses default value in super->s_blksize. Filesystems that want to provide a per-inode st_blksize can do so by providing their own getattr routine instead of using the generic_fillattr() function.

Note that some filesystems were providing pretty much random (and incorrect) values for i_blksize.

Blease don't add a field to the superblock for the optimal I/O size. Any filesystem that wants to override it can do so directly in ->getattr.

#25 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

Quoted 2 lines · Christoph Hellwig

Blease don't add a field to the superblock for the optimal I/O size. Any filesystem that wants to override it can do so directly in ->getattr.

I don't disagree with you, but the idea of everyone implementing
->getattr where they just let it work before scares me. It's a ton of cut-n-paste error waiting to happen. Especially if we make something stale.
Perhaps add generic_fillattr_blksize()?

myfs_fillattr(inode, stat)
{
generic_fillattr_size(inode, myfavoriteblksize, stat)
}

Joel

Signature · 9 lines
Life's Little Instruction Book #407

	"Every once in a while, take the scenic route."

Joel Becker
Principal Software Developer
Oracle
E-mail: [email protected]
Phone: (650) 506-8127
#26 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

Quoted 9 lines · Joel Becker
Quoted 2 lines · Christoph Hellwig

Blease don't add a field to the superblock for the optimal I/O size. Any filesystem that wants to override it can do so directly in ->getattr.

I don't disagree with you, but the idea of everyone implementing
->getattr where they just let it work before scares me. It's a ton of cut-n-paste error waiting to happen. Especially if we make something stale.
Perhaps add generic_fillattr_blksize()?

Well, as far as I know the only filesystems today that would need to do something different are xfs, ocfs2, and reiserfs, and IMHO only the first two have any kind of justification for doing it. Part of the problem is what st_blksize actually means was never well-defined; it was never in POSIX, and in SuSv3 all that is stated is, "A file system-specific preferred I/O block size for this object." This is why Reiserfs got away with specifying 128 megs (I assume it helped on some benchmark), and why being ill-defined, using such a large value might cause some applications (like /bin/cp) to core dump.

Given that most filesystems use the generic page cache read/write functions, using PAGE_CACHE_SIZE as the default seems to make a huge amount of sense. I really wonder how useful setting st_blksize really is, actually, at least in the real-world, as opposed to just for benchmarks.

- Ted

#27 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

Quoted 5 lines · Theodore Tso
Quoted 4 lines · Joel Becker
Quoted 2 lines · Christoph Hellwig

Blease don't add a field to the superblock for the optimal I/O size. Any filesystem that wants to override it can do so directly in ->getattr.

Quoted 8 lines
Quoted 5 lines

I don't disagree with you, but the idea of everyone implementing
->getattr where they just let it work before scares me. It's a ton of cut-n-paste error waiting to happen. Especially if we make something stale.
Perhaps add generic_fillattr_blksize()?

Well, as far as I know the only filesystems today that would need to do something different are xfs, ocfs2, and reiserfs,

And to answer Joel's statment of these three two already implement their own ->getattr. Also it doesn't mean a filesystem has to completely reimplement it, they just have to override it by reusing generic_fillattr, e.g.

static int my_getattr(struct vfsmount *mnt, struct dentry *dentry,
struct kstat *stat)

{
	int error = generic_fillattr(dentry->d_inode, stat);
	if (!error)
		stat->blksize = something_useful;
	return error;
}		
#28 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

Quoted 4 lines · Christoph Hellwig

And to answer Joel's statment of these three two already implement their own ->getattr. Also it doesn't mean a filesystem has to completely reimplement it, they just have to override it by reusing generic_fillattr, e.g.

And actually, that's what we do in ocfs2_getattr() today already:

	generic_fillattr(inode, stat);

	/* We set the blksize from the cluster size for performance */
	stat->blksize = osb->s_clustersize;

Thanks,

	--Mark

--
Mark Fasheh
Senior Software Developer, Oracle
[email protected]

#29 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

In reply to Theodore Tso
Quoted 21 lines · Theodore Tso
Quoted 9 lines · Joel Becker
Quoted 2 lines · Christoph Hellwig

Blease don't add a field to the superblock for the optimal I/O size. Any filesystem that wants to override it can do so directly in ->getattr.

I don't disagree with you, but the idea of everyone implementing
->getattr where they just let it work before scares me. It's a ton of cut-n-paste error waiting to happen. Especially if we make something stale.
Perhaps add generic_fillattr_blksize()?

Well, as far as I know the only filesystems today that would need to do something different are xfs, ocfs2, and reiserfs, and IMHO only the first two have any kind of justification for doing it. Part of the problem is what st_blksize actually means was never well-defined; it was never in POSIX, and in SuSv3 all that is stated is, "A file system-specific preferred I/O block size for this object." This is why Reiserfs got away with specifying 128 megs (I assume it helped on some benchmark), and why being ill-defined, using such a large value might cause some applications (like /bin/cp) to core dump.

Quoted 5 lines

Given that most filesystems use the generic page cache read/write functions, using PAGE_CACHE_SIZE as the default seems to make a huge amount of sense. I really wonder how useful setting st_blksize really is, actually, at least in the real-world, as opposed to just for benchmarks.

I assume "128 megs" is a typo, it's 128k, of course. And certainly it would have helped speed things up, not just benchmarks, because for modern disks, doing 128k vs 4k takes like 25% more time. Wu's adaptive readahead patches might make this outdated, though, since they now support "read 10 pages, seek, read 10 pages, seek, etc" type workloads.

NATE

#30 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

In reply to Christoph Hellwig
Quoted 11 lines · Christoph Hellwig
Quoted 7 lines · Theodore Tso

This eliminates the i_blksize field from struct inode and uses default value in super->s_blksize. Filesystems that want to provide a per-inode st_blksize can do so by providing their own getattr routine instead of using the generic_fillattr() function.

Note that some filesystems were providing pretty much random (and incorrect) values for i_blksize.

Blease don't add a field to the superblock for the optimal I/O size. Any filesystem that wants to override it can do so directly in ->getattr.

I was mainly adding a field to the superblock for the sake of reiserfs; it was the easiest way to support it's current insanity of reporting 128megs as st_blksize, but with a mount option that affected things on a global scale. (Hey, at least my patch made resierfs support the mount option as a per-superblock setting, instead of setting a global variable that changed the behaviour for all mounts.)

I don't mind espceially removing the superblock field, but if we do, I refuse to mess with adding a getattr special for reiserfs; someone who wants to support its current behaviour is free to send me or LKML patches....

- Ted

#31 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

Quoted 11 lines · Theodore Tso

I was mainly adding a field to the superblock for the sake of reiserfs; it was the easiest way to support it's current insanity of reporting 128megs as st_blksize, but with a mount option that affected things on a global scale. (Hey, at least my patch made resierfs support the mount option as a per-superblock setting, instead of setting a global variable that changed the behaviour for all mounts.)

I don't mind espceially removing the superblock field, but if we do, I refuse to mess with adding a getattr special for reiserfs; someone who wants to support its current behaviour is free to send me or LKML patches....

I'd suggest to either keep their current buggy behaviour and let Hans fix it or use a field in the reiserfs specific superblock structure.

#32 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

In reply to Theodore Tso

fs/ocfs2/super.c line 1393

Quoted 10 lines · Theodore Tso

Index: linux-2.6.17/fs/ocfs2/super.c
===================================================================

fs/ocfs2/super.c +1 -1
@@ -1390,7 +1390,7 @@
 	/* get some pseudo constants for clustersize bits */
 	osb->s_clustersize_bits =
 		le32_to_cpu(di->id2.i_super.s_clustersize_bits);
-	osb->s_clustersize = 1 << osb->s_clustersize_bits;
+	osb->s_blksize = osb->s_clustersize = 1 << osb->s_clustersize_bits;
	I assume you meant sb->s_blksize.  osb is a private structure
hanging off of sb->s_fs_info.
	I have to say, having sb->s_blocksize and sb->s_blksize might be

a little confusing.

Joel

Signature · 9 lines
"Heav'n hath no rage like love to hatred turn'd, nor Hell a fury,
 like a woman scorn'd."
        - William Congreve

Joel Becker
Principal Software Developer
Oracle
E-mail: [email protected]
Phone: (650) 506-8127
#33 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

Quoted 3 lines · Joel Becker
	I assume you meant sb->s_blksize.  osb is a private structure

hanging off of sb->s_fs_info.

Yep, mea culpa.

Quoted 2 lines
	I have to say, having sb->s_blocksize and sb->s_blksize might be

a little confusing.

How strongly do you feel about reporting stat.st_blksize out to be the clustersize? Keeping in mind that if you report a value which is too big, /bin/cp will start coredumping....

If I take Christoph's suggestion of simply removing sb->s_blksize altogether, and forcing filesystems that want to return a non-default value for stat.st_blksize to supply their own filesystem-specific getattr, will you mind terribly?

- Ted

#34 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

Quoted 3 lines · Theodore Tso

How strongly do you feel about reporting stat.st_blksize out to be the clustersize? Keeping in mind that if you report a value which is too big, /bin/cp will start coredumping....

What's "too big"? We certainly don't specify 128MB, but we will
specify up to 1MB.
When we have large cluster size filesystems, we've seen backups
with O_DIRECT-enabled versions of cp(1) and dd(1) go way faster than when they read/write 4K at a time. This matters to folks who have the files open O_DIRECT by other processes and cannot trust the page cache. At least, that's is how it was when we originally did this :-)

Quoted 4 lines

If I take Christoph's suggestion of simply removing sb->s_blksize altogether, and forcing filesystems that want to return a non-default value for stat.st_blksize to supply their own filesystem-specific getattr, will you mind terribly?

Well, Christoph points out we already have a valid getattr()
that calls generic_fillattr() and then overloads our stat->blksize. So I think we're safe. I don't think we need sb->s_blksize, but I'll defer to Mark for the final say.

Joel

Signature · 11 lines
"The lawgiver, of all beings, most owes the law allegiance.  He of all
 men should behave as though the law compelled him.  But it is the
 universal weakness of mankind that what we are given to administer we
 presently imagine we own."
        - H.G. Wells

Joel Becker
Principal Software Developer
Oracle
E-mail: [email protected]
Phone: (650) 506-8127
#35 lore raw

Re: [RFC] [PATCH 5/8] inode-diet: Eliminate i_blksize and use a per-superblock default

In reply to Theodore Tso
Quoted 3 lines · Theodore Tso

How strongly do you feel about reporting stat.st_blksize out to be the clustersize? Keeping in mind that if you report a value which is too big, /bin/cp will start coredumping....

Any idea on what constitutes "too big"? We could always cap things at some reasonable maximum. AFAIR, the reason it's set to clustersize these days is because it actually made a performance impact on large file operations - backing up a data base file, for example.

Quoted 4 lines

If I take Christoph's suggestion of simply removing sb->s_blksize altogether, and forcing filesystems that want to return a non-default value for stat.st_blksize to supply their own filesystem-specific getattr, will you mind terribly?

No, I don't mind really - it's a minor bit of code and it'll save us all some space on the super block structure. I assume we can just continue to follow Christoph's suggestion in ocfs2_getattr().

	--Mark

--
Mark Fasheh
Senior Software Developer, Oracle
[email protected]

#36 lore raw

RFC PATCH 6/8 inode-diet: Move i_cindex from struct inode to struct file

In reply to Theodore Tso

inode.i_cindex isn't initialized until the character device is opened anyway, and there are far more struct inodes in memory than there are struct file. So move the cindex field to file.f_cindex, and change the one(!) user of cindex to use file pointer, which is in fact simpler.

include/linux/fs.h +1 -1
Index: linux-2.6.17/include/linux/fs.h
===================================================================
@@ -513,7 +513,6 @@
 		struct block_device	*i_bdev;
 		struct cdev		*i_cdev;
 	};
-	int			i_cindex;
 
 	__u32			i_generation;
 
@@ -659,6 +658,7 @@
 	spinlock_t		f_ep_lock;
 #endif /* #ifdef CONFIG_EPOLL */
 	struct address_space	*f_mapping;
+	int			f_cindex;
 };
 extern spinlock_t files_lock;
 #define file_list_lock() spin_lock(&files_lock);
fs/char_dev.c +1 -1
Index: linux-2.6.17/fs/char_dev.c
===================================================================
@@ -290,7 +290,7 @@
 		p = inode->i_cdev;
 		if (!p) {
 			inode->i_cdev = p = new;
-			inode->i_cindex = idx;
+			filp->f_cindex = idx;
 			list_add(&inode->i_devices, &p->list);
 			new = NULL;
 		} else if (!cdev_get(p))
drivers/ieee1394/ieee1394_core.h +1 -1
Index: linux-2.6.17/drivers/ieee1394/ieee1394_core.h
===================================================================
@@ -212,7 +212,7 @@
 /* return the index (within a minor number block) of a file */
 static inline unsigned char ieee1394_file_to_instance(struct file *file)
 {
-	return file->f_dentry->d_inode->i_cindex;
+	return file->f_cindex;
 }
 
 extern int hpsb_disable_irm;

--

#37 lore raw

Re: [RFC] [PATCH 6/8] inode-diet: Move i_cindex from struct inode to struct file

Quoted 4 lines · Theodore Tso

inode.i_cindex isn't initialized until the character device is opened anyway, and there are far more struct inodes in memory than there are struct file. So move the cindex field to file.f_cindex, and change the one(!) user of cindex to use file pointer, which is in fact simpler.

NAK. Please, take it to the union into cdev part.

#38 lore raw

Re: [RFC] [PATCH 6/8] inode-diet: Move i_cindex from struct inode to struct file

Quoted 7 lines · Al Viro
Quoted 4 lines · Theodore Tso

inode.i_cindex isn't initialized until the character device is opened anyway, and there are far more struct inodes in memory than there are struct file. So move the cindex field to file.f_cindex, and change the one(!) user of cindex to use file pointer, which is in fact simpler.

NAK. Please, take it to the union into cdev part.

Explanation: the whole point of that sucker is to avoid i_rdev use in drivers, switching to i_cdev + i_cindex. _Especially_ in open(). There is a bunch of other drivers that would get cleaner from that, including a lot of tty code.

#39 lore raw

Re: [RFC] [PATCH 6/8] inode-diet: Move i_cindex from struct inode to struct file

Quoted 6 lines · Al Viro
Quoted 1 line

NAK. Please, take it to the union into cdev part.

Explanation: the whole point of that sucker is to avoid i_rdev use in drivers, switching to i_cdev + i_cindex. _Especially_ in open(). There is a bunch of other drivers that would get cleaner from that, including a lot of tty code.

The problem is that we already have more stuff in the cdev union (list_head's are *big*, especially on Opterons), so moving it there doesn't actually save us anything.

Moving it into struct file however should be good enough clean up the character devices drivers that you are concerned about, however. They are passed the struct file pointer in the file_operations->open call. So we can clean up the tty code, et. al, by using file->f_cindex just as easily.

Furthermore, the inode->i_cindex isn't guaranteed to be valid until the high-level vfs open code is called anyway, so you might as well tell people to reference it from filp->f_cindex in the device driver's open() routine.

Does that make sense?

- Ted

#40 lore raw

Re: [RFC] [PATCH 6/8] inode-diet: Move i_cindex from struct inode to struct file

Quoted 18 lines · Theodore Tso
Quoted 6 lines · Al Viro
Quoted 1 line

NAK. Please, take it to the union into cdev part.

Explanation: the whole point of that sucker is to avoid i_rdev use in drivers, switching to i_cdev + i_cindex. _Especially_ in open(). There is a bunch of other drivers that would get cleaner from that, including a lot of tty code.

The problem is that we already have more stuff in the cdev union (list_head's are *big*, especially on Opterons), so moving it there doesn't actually save us anything.

Moving it into struct file however should be good enough clean up the character devices drivers that you are concerned about, however. They are passed the struct file pointer in the file_operations->open call. So we can clean up the tty code, et. al, by using file->f_cindex just as easily.

Quoted 6 lines

Furthermore, the inode->i_cindex isn't guaranteed to be valid until the high-level vfs open code is called anyway, so you might as well tell people to reference it from filp->f_cindex in the device driver's open() routine.

Does that make sense?

The patch is obviously proper for usage by any file_operations->open() users like the single current one (or two actually, dv1394_open() and video1394_open()). I don't know if there are any contrary usages planned in other subsystems.

Signature · 3 lines
Stefan Richter
-=====-=-==- -==- =--==
http://arcgraph.de/sr/
#41 lore raw

RFC PATCH 7/8 inode-diet: Use a union for i_blocks and i_size, i_rdev and i_devices

In reply to Theodore Tso

The i_blocks and i_size fields are only used for regular files. So we move them into the union, along with i_rdev and i_devices, which are only used by block or character devices.

include/linux/fs.h +12 -6
Index: linux-2.6.17/include/linux/fs.h
===================================================================
@@ -486,15 +486,12 @@
 	unsigned int		i_nlink;
 	uid_t			i_uid;
 	gid_t			i_gid;
-	dev_t			i_rdev;
 	loff_t			i_size;
 	struct timespec		i_atime;
 	struct timespec		i_mtime;
 	struct timespec		i_ctime;
 	unsigned int		i_blkbits;
 	unsigned long		i_version;
-	blkcnt_t		i_blocks;
-	unsigned short          i_bytes;
 	spinlock_t		i_lock;	/* i_blocks, i_bytes, maybe i_size */
 	struct mutex		i_mutex;
 	struct rw_semaphore	i_alloc_sem;
@@ -507,11 +504,20 @@
 #ifdef CONFIG_QUOTA
 	struct dquot		*i_dquot[MAXQUOTAS];
 #endif
-	struct list_head	i_devices;
 	union {
 		struct pipe_inode_info	*i_pipe;
-		struct block_device	*i_bdev;
-		struct cdev		*i_cdev;
+		struct {
+			union {
+				struct block_device	*i_bdev;
+				struct cdev		*i_cdev;
+			};
+			dev_t			i_rdev;
+			struct list_head	i_devices;
+		};
+		struct {
+			unsigned short          i_bytes;
+			blkcnt_t		i_blocks;
+		};
 	};
 
 	__u32			i_generation;

--

#42 lore raw

RFC PATCH 8/8 inode-diet: Fix size of i_blkbits, i_version, and i_dnotify_mask

In reply to Theodore Tso

i_blkbits stores the log n of the blocksize; there is no reason for it to take more than 16 bits, so change it to be a short and put it next to i_mode.

i_version and i_dnotify_mask need to be 32 bits, but there is no reason for them to be 64-bit values on 64-bit architectures, so make them be unsigned int's instead of unsigned long's.

include/linux/fs.h +3 -3
Index: linux-2.6.17/include/linux/fs.h
===================================================================
@@ -483,6 +483,7 @@
 	unsigned long		i_ino;
 	atomic_t		i_count;
 	umode_t			i_mode;
+	unsigned short		i_blkbits;
 	unsigned int		i_nlink;
 	uid_t			i_uid;
 	gid_t			i_gid;
@@ -490,8 +491,7 @@
 	struct timespec		i_atime;
 	struct timespec		i_mtime;
 	struct timespec		i_ctime;
-	unsigned int		i_blkbits;
-	unsigned long		i_version;
+	unsigned int		i_version;
 	spinlock_t		i_lock;	/* i_blocks, i_bytes, maybe i_size */
 	struct mutex		i_mutex;
 	struct rw_semaphore	i_alloc_sem;
@@ -523,7 +523,7 @@
 	__u32			i_generation;
 
 #ifdef CONFIG_DNOTIFY
-	unsigned long		i_dnotify_mask; /* Directory notify events */
+	unsigned int		i_dnotify_mask; /* Directory notify events */
 	struct dnotify_struct	*i_dnotify; /* for directory notifications */
 #endif
 

--

#43 lore raw
In reply to Theodore Tso
Quoted 6 lines · Theodore Tso

What else remains to be done? There are a large number of fields in struct inode which are never populated unless the inode is open, and those should get moved into another structure which is populated only when needed. There are a large number of inodes which are read into memory only because stat(2) was called on them (thanks to things like color ls, et. al).

One could remove the reclaim list and use the slab lists of the slab allocator to scan through the inodes and reclaim them in such a way that would maximize the number of pages freed. I will post an RFC on that one later. This may reduce the complexity of inode reclaim.

#44 lore raw
Quoted 9 lines · Christoph Lameter
Quoted 6 lines · Theodore Tso

What else remains to be done? There are a large number of fields in struct inode which are never populated unless the inode is open, and those should get moved into another structure which is populated only when needed. There are a large number of inodes which are read into memory only because stat(2) was called on them (thanks to things like color ls, et. al).

Quoted 4 lines

One could remove the reclaim list and use the slab lists of the slab allocator to scan through the inodes and reclaim them in such a way that would maximize the number of pages freed. I will post an RFC on that one later. This may reduce the complexity of inode reclaim.

That may very well be a good thing to do (although if it is too aggressive we may end up reducing the utility of the dentry cache --- your patch is going to try to free the dentries pinning inodes, right? Otherwise it will probably not have much effect), but I think that's a largely orthogonal issue. It will still be good to be able to cache inodes (and dentries), but it would be desireable if we can do this in less memory than it currently takes, and separating out those fields which are only needed when the inode is opened, or when its pages are cached in the page cache, would be a good way of minimizing the footprint of inodes who only needs to have their stat(2) information cached.

Regards,

- Ted