[PATCH] partition handling locking cleanups

Horrors with open/reread_partition exclusion are starting to get fixed.

It's not the final variant, but at least we are getting the logics into
one place; switch to final variant will happen once we get per-disk
analog of gendisks.  New fields - ->bd_part_sem and ->bd_part_count.

The latter counts the amount of opened partitions.  The former protects
said count _and_ is held while we are rereading partition tables.
Helpers - dev_part_lock()/dev_part_unlock() (currently taking kdev_t; that
will change pretty soon).  No more ->open() and ->release() for partitions,
all that logics went to generic code.  Lock hierachy is currently messy:

  ->bd_sem for partitions -> ->bd_part_sem -> ->bd_sem for entire disks

Ugly, but that'll go away and to get the final variant of locking right
now would take _really_ big patch - with a lot of steps glued together.
The damn thing is large as it is...

drivers/acorn/block/mfmhd.c

@@ -184,8 +184,6 @@ struct mfm_info {
 #define NEED_1_RECAL -2
 #define NEED_2_RECAL -3
 		 int cylinder;
-	unsigned int access_count;
-	unsigned int busy;
 	struct {
 		char recal;
 		char report;
@@ -197,7 +195,6 @@ struct mfm_info {
 
 static struct hd_struct mfm[MFM_MAXDRIVES << 6];
 static int mfm_sizes[MFM_MAXDRIVES << 6];
-static DECLARE_WAIT_QUEUE_HEAD(mfm_wait_open);
 
 /* Stuff from the assembly routines */
 extern unsigned int hdc63463_baseaddress;	/* Controller base address */
@@ -1219,24 +1216,8 @@ static int mfm_ioctl(struct inode *inode, struct file *file, u_int cmd, u_long a
 static int mfm_open(struct inode *inode, struct file *file)
 {
 	int dev = DEVICE_NR(minor(inode->i_rdev));
-
 	if (dev >= mfm_drives)
 		return -ENODEV;
-
-	while (mfm_info[dev].busy)
-		sleep_on (&mfm_wait_open);
-
-	mfm_info[dev].access_count++;
-	return 0;
-}
-
-/*
- * Releasing a block device means we sync() it, so that it can safely
- * be forgotten about...
- */
-static int mfm_release(struct inode *inode, struct file *file)
-{
-	mfm_info[DEVICE_NR(minor(inode->i_rdev))].access_count--;
 	return 0;
 }
 
@@ -1296,7 +1277,6 @@ static struct block_device_operations mfm_fops =
 {
 	owner:		THIS_MODULE,
 	open:		mfm_open,
-	release:	mfm_release,
 	ioctl:		mfm_ioctl,
 };
 
@@ -1425,33 +1405,19 @@ int mfm_init (void)
 /*
  * This routine is called to flush all partitions and partition tables
  * for a changed MFM disk, and then re-read the new partition table.
- * If we are revalidating due to an ioctl, we have USAGE == 1.
  */
 static int mfm_reread_partitions(kdev_t dev)
 {
-	unsigned int start, i, maxp, target = DEVICE_NR(minor(dev));
-	unsigned long flags;
-
-	local_irq_save(flags);
-	if (mfm_info[target].busy || mfm_info[target].access_count > 1) {
-		local_irq_restore (flags);
-		return -EBUSY;
-	}
-	mfm_info[target].busy = 1;
-	local_irq_restore (flags);
-
-	maxp = 1 << mfm_gendisk.minor_shift;
-	start = target << mfm_gendisk.minor_shift;
-
-	wipe_partitions(mk_kdev(MAJOR_NR, start));
-
+	unsigned int unit = DEVICE_NR(minor(dev));
+	kdev_t device = mk_kdev(MAJOR_NR, unit << mfm_gendisk.minor_shift);
+	int err = dev_lock_part(device);
+	if (err)
+		return err;
+	wipe_partitions(device);
 	/* Divide by 2, since sectors are 2 times smaller than usual ;-) */
-
-	grok_partitions(&mfm_gendisk, target, 1<<6, mfm_info[target].heads *
-		    mfm_info[target].cylinders * mfm_info[target].sectors / 2);
-
-	mfm_info[target].busy = 0;
-	wake_up (&mfm_wait_open);
+	grok_partitions(device, mfm_info[unit].heads *
+		    mfm_info[unit].cylinders * mfm_info[unit].sectors / 2);
+	dev_unlock_part(device);
 	return 0;
 }
 

drivers/block/acsi.c

@@ -248,8 +248,6 @@ static int				NDevices = 0;
 static int				acsi_sizes[MAX_DEV<<4] = { 0, };
 static struct hd_struct	acsi_part[MAX_DEV<<4] = { {0,0}, };
 static int 				access_count[MAX_DEV] = { 0, };
-static char 			busy[MAX_DEV] = { 0, };
-static DECLARE_WAIT_QUEUE_HEAD(busy_wait);
 
 static int				CurrentNReq;
 static int				CurrentNSect;
@@ -1161,8 +1159,6 @@ static int acsi_open( struct inode * inode, struct file * filp )
 	if (device >= NDevices)
 		return -ENXIO;
 	aip = &acsi_info[device];
-	while (busy[device])
-		sleep_on(&busy_wait);
 
 	if (access_count[device] == 0 && aip->removable) {
 #if 0
@@ -1803,10 +1799,6 @@ void cleanup_module(void)
 }
 #endif
 
-#define DEVICE_BUSY busy[device]
-#define USAGE access_count[device]
-#define GENDISK_STRUCT acsi_gendisk
-
 /*
  * This routine is called to flush all partitions and partition tables
  * for a changed scsi disk, and then re-read the new partition table.
@@ -1828,24 +1820,15 @@ void cleanup_module(void)
 
 static int revalidate_acsidisk( int dev, int maxusage )
 {
-	int device;
-	struct gendisk * gdev;
-	int res;
-	struct acsi_info_struct *aip;
-	
-	device = DEVICE_NR(minor(dev));
-	aip = &acsi_info[device];
-	gdev = &GENDISK_STRUCT;
+	int unit = DEVICE_NR(minor(dev));
+	struct acsi_info_struct *aip = &acsi_info[unit];
+	kdev_t device = mk_kdev(MAJOR_NR, unit<<4);
+	int res = dev_lock_part(device);
 
-	cli();
-	if (DEVICE_BUSY || USAGE > maxusage) {
-		sti();
-		return -EBUSY;
-	};
-	DEVICE_BUSY = 1;
-	sti();
+	if (res < 0)
+		return res;
 
-	res = wipe_partitions(dev);
+	res = wipe_partitions(device);
 
 	stdma_lock( NULL, NULL );
 
@@ -1862,10 +1845,9 @@ static int revalidate_acsidisk( int dev, int maxusage )
 	stdma_release();
 
 	if (!res)
-		grok_partitions(dev, aip->size);
+		grok_partitions(device, aip->size);
 
-	DEVICE_BUSY = 0;
-	wake_up(&busy_wait);
+	dev_unlock_part(device);
 	return res;
 }
 

drivers/block/cciss.c

@@ -735,6 +735,10 @@ static int cciss_ioctl(struct inode *inode, struct file *filep,
 }
 
 /* Borrowed and adapted from sd.c */
+/*
+ * FIXME: we are missing the exclusion with ->open() here - it can happen
+ * just as we are rereading partition tables.
+ */
 static int revalidate_logvol(kdev_t dev, int maxusage)
 {
         int ctlr, target;
@@ -1304,13 +1308,12 @@ static int register_new_disk(kdev_t dev, int ctlr)
 	hba[ctlr]->drv[logvol].usage_count = 0;
 	max_p = 1 << gdev->minor_shift;
         start = logvol<< gdev->minor_shift;
+	kdev = mk_kdev(MAJOR_NR + ctlr, logvol<< gdev->minor_shift);
 
-	wipe_partitions(MAJOR_NR + ctlr, start);
-
+	wipe_partitions(kdev);
 	++hba[ctlr]->num_luns;
 	gdev->nr_real = hba[ctlr]->highest_lun + 1;
 	/* setup partitions per disk */
-	kdev = mk_kdev(MAJOR_NR + ctlr, logvol<< gdev->minor_shift);
 	grok_partitions(kdev, hba[ctlr]->drv[logvol].nr_blocks);
 	
 	kfree(ld_buff);

drivers/block/cpqarray.c

@@ -1533,6 +1533,9 @@ static int revalidate_allvol(kdev_t dev)
 }
 
 /* Borrowed and adapted from sd.c */
+/*
+ * FIXME: exclusion with ->open()
+ */
 static int revalidate_logvol(kdev_t dev, int maxusage)
 {
 	int ctlr, target;

drivers/block/paride/pd.c

@@ -414,12 +414,11 @@ int pd_init (void)
 }
 
 static int pd_open (struct inode *inode, struct file *file)
+{
+	int unit = DEVICE_NR(inode->i_rdev);
 
-{       int unit = DEVICE_NR(inode->i_rdev);
-
-        if ((unit >= PD_UNITS) || (!PD.present)) return -ENODEV;
-
-	wait_event (pd_wait_open, pd_valid);
+        if ((unit >= PD_UNITS) || (!PD.present))
+		return -ENODEV;
 
         PD.access++;
 
@@ -480,12 +479,8 @@ static int pd_ioctl(struct inode *inode,struct file *file,
 }
 
 static int pd_release (struct inode *inode, struct file *file)
-
-{       kdev_t devp;
-	int	unit;
-
-        devp = inode->i_rdev;
-	unit = DEVICE_NR(devp);
+{
+	int	unit = DEVICE_NR(inode->i_rdev);
 
 	if ((unit >= PD_UNITS) || (PD.access <= 0)) 
 		return -EINVAL;
@@ -513,29 +508,22 @@ static int pd_check_media( kdev_t dev)
 
 static int pd_revalidate(kdev_t dev)
 {
-	int unit, res;
-	long flags;
+	int unit = DEVICE_NR(dev);
+	kdev_t device = mk_kdev(MAJOR_NR, unit << PD_BITS);
+	int res;
 
-	unit = DEVICE_NR(dev);
 	if ((unit >= PD_UNITS) || !PD.present)
 		return -ENODEV;
 
-	save_flags(flags);
-	cli(); 
-	if (PD.access > 1) {
-		restore_flags(flags);
-		return -EBUSY;
-	}
-	pd_valid = 0;
-	restore_flags(flags);   
-
-	res = wipe_partitions(dev);
+	res = dev_part_lock(device);
+	if (res < 0)
+		return res;
+	res = wipe_partitions(device);
 
 	if (res == 0 && pd_identify(unit))
-		grok_partitions(dev, PD.capacity);
+		grok_partitions(device, PD.capacity);
 
-	pd_valid = 1;
-	wake_up(&pd_wait_open);
+	dev_unlock_part(device);
  
         return res;
 }

drivers/block/ps2esdi.c

@@ -93,8 +93,6 @@ static void ps2esdi_geometry_int_handler(u_int);
 
 static int ps2esdi_open(struct inode *inode, struct file *file);
 
-static int ps2esdi_release(struct inode *inode, struct file *file);
-
 static int ps2esdi_ioctl(struct inode *inode, struct file *file,
 			 u_int cmd, u_long arg);
 
@@ -111,11 +109,8 @@ static void ps2esdi_reset_timer(unsigned long unused);
 static u_int dma_arb_level;		/* DMA arbitration level */
 
 static DECLARE_WAIT_QUEUE_HEAD(ps2esdi_int);
-static DECLARE_WAIT_QUEUE_HEAD(ps2esdi_wait_open);
 
 static int no_int_yet;
-static int access_count[MAX_HD];
-static char ps2esdi_valid[MAX_HD];
 static int ps2esdi_sizes[MAX_HD << 6];
 static int ps2esdi_drives;
 static struct hd_struct ps2esdi[MAX_HD << 6];
@@ -152,7 +147,6 @@ static struct block_device_operations ps2esdi_fops =
 {
 	owner:		THIS_MODULE,
 	open:		ps2esdi_open,
-	release:	ps2esdi_release,
 	ioctl:		ps2esdi_ioctl,
 };
 
@@ -441,13 +435,11 @@ static int __init ps2esdi_geninit(void)
 	}
 	blk_queue_max_sectors(BLK_DEFAULT_QUEUE(MAJOR_NR), 128);
 
-	for (i = 0; i < ps2esdi_drives; i++) {
+	for (i = 0; i < ps2esdi_drives; i++)
 		register_disk(&ps2esdi_gendisk,mk_kdev(MAJOR_NR,i<<6),1<<6,
 				&ps2esdi_fops,
 				ps2esdi_info[i].head * ps2esdi_info[i].sect *
 				ps2esdi_info[i].cyl);
-		ps2esdi_valid[i] = 1;
-	}
 	return 0;
 
 err_out3:
@@ -1083,32 +1075,11 @@ static void dump_cmd_complete_status(u_int int_ret_code)
 static int ps2esdi_open(struct inode *inode, struct file *file)
 {
 	int dev = DEVICE_NR(inode->i_rdev);
-
-	if (dev < ps2esdi_drives) {
-		while (!ps2esdi_valid[dev])
-			sleep_on(&ps2esdi_wait_open);
-
-		access_count[dev]++;
-
-		return (0);
-	} else
-		return (-ENODEV);
-}
-
-
-
-static int ps2esdi_release(struct inode *inode, struct file *file)
-{
-	int dev = DEVICE_NR(inode->i_rdev);
-
-	if (dev < ps2esdi_drives) {
-		access_count[dev]--;
-	}
+	if (dev >= ps2esdi_drives)
+		return -ENODEV;
 	return 0;
 }
 
-
-
 static int ps2esdi_ioctl(struct inode *inode,
 			 struct file *file, u_int cmd, u_long arg)
 {
@@ -1155,24 +1126,20 @@ static int ps2esdi_ioctl(struct inode *inode,
 static int ps2esdi_reread_partitions(kdev_t dev)
 {
 	int target = DEVICE_NR(dev);
-	int res;
+	kdev_t device = mk_kdev(MAJOR_NR, target << 6);
+	int res = dev_lock_part(device);
 
-	cli();
-	ps2esdi_valid[target] = (access_count[target] != 1);
-	sti();
-	if (ps2esdi_valid[target])
-		return (-EBUSY);
+	if (res < 0)
+		return res;
 
-	res = wipe_partitions(dev);
+	res = wipe_partitions(device);
 	if (res == 0)
-		grok_partitions(dev, ps2esdi_info[target].head
+		grok_partitions(device, ps2esdi_info[target].head
 				* ps2esdi_info[target].cyl
 				* ps2esdi_info[target].sect);
  
-	ps2esdi_valid[target] = 1;
-	wake_up(&ps2esdi_wait_open);
-
-	return (res);
+	dev_unlock_part(device);
+	return res;
 }
 
 static void ps2esdi_reset_timer(unsigned long unused)

drivers/block/umem.c

@@ -812,31 +812,21 @@ static void del_battery_timer(void)
  * Note no locks taken out here.  In a worst case scenario, we could drop
  * a chunk of system memory.  But that should never happen, since validation
  * happens at open or mount time, when locks are held.
+ *
+ *	That's crap, since doing that while some partitions are opened
+ * or mounted will give you really nasty results.
  */
 static int mm_revalidate(kdev_t i_rdev)
 {
-	int i;
-
 	int card_number = DEVICE_NR(i_rdev);
-	/* first partition, # of partitions */
-	int part1 = (card_number << MM_SHIFT) + 1;
-	int npart = (1 << MM_SHIFT) -1;
-
-	/* first clear old partition information */
-	for (i=0; i<npart ;i++) {
-		mm_gendisk.sizes[part1+i]=0;
-		mm_gendisk.part[part1+i].start_sect = 0;
-		mm_gendisk.part[part1+i].nr_sects = 0;
-	}
-
-	mm_gendisk.part[card_number << MM_SHIFT].nr_sects =
-		cards[card_number].mm_size << 1;
-
-
-	/* then fill new info */
+	kdev_t device = mk_mdev(MAJOR_NR, card_number << MM_SHIFT);
+	int res = dev_lock_part(device);
+	if (res < 0)
+		return res;
+	wipe_partitions(device);
 	printk(KERN_INFO "mm partition check: (%d)\n", card_number);
-	grok_partitions(mk_kdev(major_nr,part1-1),
-			mm_gendisk.sizes[card_number<<MM_SHIFT]);
+	grok_partitions(device, cards[card_number].mm_size << 1);
+	dev_unlock_part(device);
 	return 0;
 }
 /*
@@ -925,22 +915,12 @@ static int mm_open(struct inode *i, struct file *filp)
 }
 /*
 -----------------------------------------------------------------------------------
---                              mm_do_release
------------------------------------------------------------------------------------
-*/
-static int mm_do_release(struct inode *i, struct file *filp)
-{
-	return 0;
-}
-/*
------------------------------------------------------------------------------------
 --                             mm_fops
 -----------------------------------------------------------------------------------
 */
 static struct block_device_operations mm_fops = {
 	owner:		THIS_MODULE,
 	open:		mm_open,
-	release:	mm_do_release,
 	ioctl:		mm_ioctl,
 	revalidate:	mm_revalidate,
 	check_media_change: mm_check_change,

drivers/block/xd.c

@@ -122,7 +122,7 @@ static unsigned int xd_bases[] __initdata =
 };
 
 static struct hd_struct xd_struct[XD_MAXDRIVES << 6];
-static int xd_sizes[XD_MAXDRIVES << 6], xd_access[XD_MAXDRIVES];
+static int xd_sizes[XD_MAXDRIVES << 6];
 
 static spinlock_t xd_lock = SPIN_LOCK_UNLOCKED;
 
@@ -140,12 +140,9 @@ static struct gendisk xd_gendisk = {
 static struct block_device_operations xd_fops = {
 	owner:		THIS_MODULE,
 	open:		xd_open,
-	release:	xd_release,
 	ioctl:		xd_ioctl,
 };
 static DECLARE_WAIT_QUEUE_HEAD(xd_wait_int);
-static DECLARE_WAIT_QUEUE_HEAD(xd_wait_open);
-static u_char xd_valid[XD_MAXDRIVES] = { 0,0 };
 static u_char xd_drives, xd_irq = 5, xd_dma = 3, xd_maxsectors;
 static u_char xd_override __initdata = 0, xd_type __initdata = 0;
 static u_short xd_iobase = 0x320;
@@ -244,14 +241,11 @@ static void __init xd_geninit (void)
 	/* xd_maxsectors depends on controller - so set after detection */
 	blk_queue_max_sectors(BLK_DEFAULT_QUEUE(MAJOR_NR), xd_maxsectors);
 
-	for (i = 0; i < xd_drives; i++) {
-		xd_valid[i] = 1;
+	for (i = 0; i < xd_drives; i++)
 		register_disk(&xd_gendisk, mk_kdev(MAJOR_NR,i<<6), 1<<6,
 				&xd_fops,
 				xd_info[i].heads * xd_info[i].cylinders *
 				xd_info[i].sectors);
-	}
-
 	xd_gendisk.nr_real = xd_drives;
 }
 
@@ -259,17 +253,9 @@ static void __init xd_geninit (void)
 static int xd_open (struct inode *inode,struct file *file)
 {
 	int dev = DEVICE_NR(inode->i_rdev);
-
-	if (dev < xd_drives) {
-		while (!xd_valid[dev])
-			sleep_on(&xd_wait_open);
-
-		xd_access[dev]++;
-
-		return (0);
-	}
-
+	if (dev >= xd_drives)
 		return -ENXIO;
+	return 0;
 }
 
 /* do_xd_request: handle an incoming request */
@@ -364,37 +350,23 @@ static int xd_ioctl (struct inode *inode,struct file *file,u_int cmd,u_long arg)
 	}
 }
 
-/* xd_release: release the device */
-static int xd_release (struct inode *inode, struct file *file)
-{
-	int target = DEVICE_NR(inode->i_rdev);
-	if (target < xd_drives)
-		xd_access[target]--;
-	return 0;
-}
-
 /* xd_reread_partitions: rereads the partition table from a drive */
 static int xd_reread_partitions(kdev_t dev)
 {
-	int target;
-	int res;
+	int target = DEVICE_NR(dev);
+	kdev_t device = mk_kdev(MAJOR_NR, target << 6);
+	int res = dev_lock_part(device);
 	
-	target = DEVICE_NR(dev);
-
-	cli();
-	xd_valid[target] = (xd_access[target] != 1);
-        sti();
-	if (xd_valid[target])
-		return -EBUSY;
+	if (res < 0)
+		return 0;
 
-	res = wipe_partitions(dev);
+	res = wipe_partitions(device);
 	if (!res)
-		grok_partitions(dev, xd_info[target].heads
+		grok_partitions(device, xd_info[target].heads
 				* xd_info[target].cylinders
 				* xd_info[target].sectors);
 
-	xd_valid[target] = 1;
-	wake_up(&xd_wait_open);
+	dev_unlock_part(device);
 
 	return res;
 }

drivers/block/xd.h

@@ -113,7 +113,6 @@ static void xd_geninit (void);
 static int xd_open (struct inode *inode,struct file *file);
 static void do_xd_request (request_queue_t * q);
 static int xd_ioctl (struct inode *inode,struct file *file,unsigned int cmd,unsigned long arg);
-static int xd_release (struct inode *inode,struct file *file);
 static int xd_reread_partitions (kdev_t dev);
 static int xd_readwrite (u_char operation,u_char drive,char *buffer,u_int block,u_int count);
 static void xd_recalibrate (u_char drive);

drivers/ide/hd.c

@@ -83,9 +83,6 @@ static void bad_rw_intr(void);
 
 static char recalibrate[MAX_HD];
 static char special_op[MAX_HD];
-static int access_count[MAX_HD];
-static char busy[MAX_HD];
-static DECLARE_WAIT_QUEUE_HEAD(busy_wait);
 
 static int reset;
 static int hd_error;
@@ -668,14 +665,9 @@ static int hd_ioctl(struct inode * inode, struct file * file,
 
 static int hd_open(struct inode * inode, struct file * filp)
 {
-	int target;
-	target =  DEVICE_NR(inode->i_rdev);
-
+	int target =  DEVICE_NR(inode->i_rdev);
 	if (target >= NR_HD)
 		return -ENODEV;
-	while (busy[target])
-		sleep_on(&busy_wait);
-	access_count[target]++;
 	return 0;
 }
 
@@ -683,12 +675,6 @@ static int hd_open(struct inode * inode, struct file * filp)
  * Releasing a block device means we sync() it, so that it can safely
  * be forgotten about...
  */
-static int hd_release(struct inode * inode, struct file * file)
-{
-        int target =  DEVICE_NR(inode->i_rdev);
-	access_count[target]--;
-	return 0;
-}
 
 extern struct block_device_operations hd_fops;
 
@@ -715,7 +701,6 @@ static void hd_interrupt(int irq, void *dev_id, struct pt_regs *regs)
 
 static struct block_device_operations hd_fops = {
 	.open =		hd_open,
-	.release =	hd_release,
 	.ioctl =	hd_ioctl,
 };
 
@@ -854,13 +839,9 @@ int __init hd_init(void)
 	return 0;
 }
 
-#define DEVICE_BUSY busy[target]
-#define USAGE access_count[target]
 #define CAPACITY (hd_info[target].head*hd_info[target].sect*hd_info[target].cyl)
 /* We assume that the BIOS parameters do not change, so the disk capacity
    will not change */
-#undef MAYBE_REINIT
-#define GENDISK_STRUCT hd_gendisk
 
 /*
  * This routine is called to flush all partitions and partition tables
@@ -872,36 +853,15 @@ int __init hd_init(void)
  */
 static int revalidate_hddisk(kdev_t dev, int maxusage)
 {
-	int target;
-	struct gendisk * gdev;
-	int res;
-	long flags;
-
-	target = DEVICE_NR(dev);
-	gdev = &GENDISK_STRUCT;
-
-	save_flags(flags);
-	cli();
-	if (DEVICE_BUSY || USAGE > maxusage) {
-		restore_flags(flags);
-		return -EBUSY;
-	}
-	DEVICE_BUSY = 1;
-	restore_flags(flags);
-
-	res = wipe_partitions(dev);
-	if (res)
-		goto leave;
-
-#ifdef MAYBE_REINIT
-	MAYBE_REINIT;
-#endif
-
-	grok_partitions(dev, CAPACITY);
-
-leave:
-	DEVICE_BUSY = 0;
-	wake_up(&busy_wait);
+	int target = DEVICE_NR(dev);
+	kdev_t device = mk_kdev(MAJOR_NR, target << 6);
+	int res = dev_lock_part(device);
+	if (res < 0)
+		return res;
+	res = wipe_partitions(device);
+	if (!res)
+		grok_partitions(device, CAPACITY);
+	dev_unlock_part(device);
 	return res;
 }
 

drivers/ide/ide.c

@@ -1105,9 +1105,6 @@ static int ide_open(struct inode * inode, struct file * filp)
 	if (drive->driver == NULL)
 		ide_driver_module();
 
-	while (drive->busy)
-		sleep_on(&drive->wqueue);
-
 	++drive->usage;
 	if (ata_ops(drive) && ata_ops(drive)->open)
 		return ata_ops(drive)->open(inode, filp, drive);

drivers/ide/main.c

@@ -249,28 +249,22 @@ struct ata_device *get_info_ptr(kdev_t i_rdev)
  */
 int ata_revalidate(kdev_t i_rdev)
 {
+	kdev_t device = mk_kdev(major(i_rdev), minor(i_rdev) & ~PARTN_MASK);
 	struct ata_device *drive;
-	unsigned long flags;
 	int res;
 
-	if ((drive = get_info_ptr(i_rdev)) == NULL)
+	if ((drive = get_info_ptr(device)) == NULL)
 		return -ENODEV;
 
-	/* FIXME: The locking here doesn't make the slightest sense! */
-	spin_lock_irqsave(&ide_lock, flags);
-
-	if (drive->busy || (drive->usage > 1)) {
-		spin_unlock_irqrestore(&ide_lock, flags);
-
-		return -EBUSY;
-	}
-
-	drive->busy = 1;
 	MOD_INC_USE_COUNT;
 
-	spin_unlock_irqrestore(&ide_lock, flags);
+	res = dev_lock_part(device);
+	if (res < 0) {
+		MOD_DEC_USE_COUNT;
+		return res;
+	}
 
-	res = wipe_partitions(i_rdev);
+	res = wipe_partitions(device);
 	if (!res) {
 		if (ata_ops(drive) && ata_ops(drive)->revalidate) {
 			ata_get(ata_ops(drive));
@@ -281,14 +275,11 @@ int ata_revalidate(kdev_t i_rdev)
 			ata_ops(drive)->revalidate(drive);
 			ata_put(ata_ops(drive));
 		} else
-			grok_partitions(i_rdev, ata_capacity(drive));
+			grok_partitions(device, ata_capacity(drive));
 	}
 
-	drive->busy = 0;
-	wake_up(&drive->wqueue);
-
+	dev_unlock_part(device);
 	MOD_DEC_USE_COUNT;
-
 	return res;
 }
 

drivers/mtd/ftl.c

@@ -1156,22 +1156,17 @@ static int ftl_reread_partitions(kdev_t dev)
 {
 	int minor = minor(dev);
 	partition_t *part = myparts[minor >> 4];
-    int res;
-
-    DEBUG(0, "ftl_cs: ftl_reread_partition(%d)\n", minor);
-    if ((atomic_read(&part->open) > 1)) {
-	    return -EBUSY;
-    }
-
-    res = wipe_partitions(dev);
-    if (res)
-	goto leave;
-
+	kdev_t device = mk_kdev(MAJOR_NR, minor & ~15);
+	int res = dev_lock_part(device);
+	if (rec < 0)
+		return res;
+	res = wipe_partitions(device);
+	if (!res) {
 		scan_header(part);
-
-    register_disk(&ftl_gendisk, whole >> PART_BITS, MAX_PART,
-		  &ftl_blk_fops, le32_to_cpu(part->header.FormattedSize)/SECTOR_SIZE);
-
+		grok_partitions(device,
+			le32_to_cpu(part->header.FormattedSize)/SECTOR_SIZE);
+	}
+	dev_unlock_part(device);
 	return res;
 }
 
@@ -1282,13 +1277,13 @@ static void ftl_notify_add(struct mtd_info *mtd)
 
 	partition->mtd = mtd;
 
-	if ((scan_header(partition) == 0) && 
-	    (build_maps(partition) == 0)) {
-		
+	if ((scan_header(partition) == 0) && (build_maps(partition) == 0)) {
 		partition->state = FTL_FORMATTED;
 		atomic_set(&partition->open, 0);
 		myparts[device] = partition;
-		ftl_reread_partitions(device << 4);
+		register_disk(&ftl_gendisk, mk_kdev(MAJOR_NR, device << 4),
+			      MAX_PART, &ftl_blk_fops,
+			      le32_to_cpu(partition->header.FormattedSize)/SECTOR_SIZE);
 #ifdef PCMCIA_DEBUG
 		printk(KERN_INFO "ftl_cs: opening %d kb FTL partition\n",
 		       le32_to_cpu(partition->header.FormattedSize) >> 10);

drivers/mtd/nftlcore.c

@@ -115,7 +115,6 @@ static void NFTL_setup(struct mtd_info *mtd)
 #endif
 
         /* linux stuff */
-	nftl->usecount = 0;
 	nftl->cylinders = 1024;
 	nftl->heads = 16;
 
@@ -153,8 +152,9 @@ static void NFTL_setup(struct mtd_info *mtd)
 #if LINUX_VERSION_CODE < 0x20328
 	resetup_one_dev(&nftl_gendisk, firstfree);
 #else
-	grok_partitions(mk_kdev(MAJOR_NR,firstfree<<NFTL_PARTN_BITS),
-			nftl->nr_sects);
+	register_disk(&nftl_gendisk,
+		      mk_kdev(MAJOR_NR,firstfree<<NFTL_PARTN_BITS),
+		      1<<NFTL_PARTN_BITS, &nftl_fops, nftl->nr_sects);
 #endif
 }
 
@@ -800,16 +800,17 @@ static int nftl_ioctl(struct inode * inode, struct file * file, unsigned int cmd
 
 	case BLKRRPART:
 		if (!capable(CAP_SYS_ADMIN)) return -EACCES;
-		if (nftl->usecount > 1) return -EBUSY;
-		/* 
-		 * We have to flush all buffers and invalidate caches,
-		 * or we won't be able to re-use the partitions,
-		 * if there was a change and we don't want to reboot
-		 */
-		res = wipe_partitions(inode->i_rdev);
+		{
+		kdev_t device = mk_kdev(MAJOR_NR,
+			minor(inode->i_rdev) & -(1<<NFTL_PARTN_BITS));
+		res = dev_lock_part(device);
+		if (res < 0)
+			return res;
+		res = wipe_partitions(device);
 		if (!res)
-			grok_partitions(inode->i_rdev, nftl->nr_sects);
-
+			grok_partitions(device, nftl->nr_sects);
+		dev_unlock_part(device);
+		}
 		return res;
 
 #if (LINUX_VERSION_CODE < 0x20303)		
@@ -955,7 +956,6 @@ static int nftl_open(struct inode *ip, struct file *fp)
 		return -EROFS;
 #endif /* !CONFIG_NFTL_RW */
 
-	thisNFTL->usecount++;
 	if (!get_mtd_device(thisNFTL->mtd, -1))
 		return /* -E'SBUGGEREDOFF */ -ENXIO;
 
@@ -972,7 +972,6 @@ static int nftl_release(struct inode *inode, struct file *fp)
 
 	if (thisNFTL->mtd->sync)
 		thisNFTL->mtd->sync(thisNFTL->mtd);
-	thisNFTL->usecount--;
 
 	put_mtd_device(thisNFTL->mtd);
 

drivers/scsi/sd.c

@@ -506,16 +506,6 @@ static int sd_open(struct inode *inode, struct file *filp)
 	 */
 	if (!scsi_block_when_processing_errors(sdp))
 		return -ENXIO;
-	/*
-	 * Make sure that only one process can do a check_change_disk at 
-	 * one time.  This is also used to lock out further access when 
-	 * the partition table is being re-read.
-	 */
-
-	while (sdp->busy) {
-		barrier();
-		cpu_relax();
-	}
 	/*
 	 * The following code can sleep.
 	 * Module unloading must be prevented
@@ -527,9 +517,7 @@ static int sd_open(struct inode *inode, struct file *filp)
 	sdp->access_count++;
 
 	if (sdp->removable) {
-		sdp->allow_revalidate = 1;
 		check_disk_change(inode->i_rdev);
-		sdp->allow_revalidate = 0;
 
 		/*
 		 * If the drive is empty, just let the open fail.
@@ -1464,9 +1452,10 @@ static int sd_attach(Scsi_Device * sdp)
 int revalidate_scsidisk(kdev_t dev, int maxusage)
 {
 	int dsk_nr = DEVICE_NR(dev);
-	int res;
 	Scsi_Disk * sdkp;
 	Scsi_Device * sdp;
+	kdev_t device = mk_kdev(major(dev), minor(dev) & ~15);
+	int res;
 
 	SCSI_LOG_HLQUEUE(3, printk("revalidate_scsidisk: dsk_nr=%d\n", 
 				   DEVICE_NR(dev)));
@@ -1474,24 +1463,20 @@ int revalidate_scsidisk(kdev_t dev, int maxusage)
 	if ((NULL == sdkp) || (NULL == (sdp = sdkp->device)))
 		return -ENODEV;
 
-	if (sdp->busy || ((sdp->allow_revalidate == 0) && 
-			  (sdp->access_count > maxusage))) {
-		printk(KERN_WARNING "Device busy for revalidation "
-		       "(access_count=%d)\n", sdp->access_count);
-		return -EBUSY;
-	}
-	sdp->busy = 1;
+	res = dev_lock_part(device);
+	if (res < 0)
+		return res;
 
-	res = wipe_partitions(dev);
+	res = wipe_partitions(device);
 	if (res)
 		goto leave;
 
 	sd_init_onedisk(sdkp, dsk_nr);
 
-	grok_partitions(dev, sdkp->capacity);
+	grok_partitions(device, sdkp->capacity);
 
 leave:
-	sdp->busy = 0;
+	dev_unlock_part(device);
 	return res;
 }
 

fs/block_dev.c

@@ -293,6 +293,8 @@ struct block_device *bdget(dev_t dev)
 			new_bdev->bd_queue = NULL;
 			new_bdev->bd_contains = NULL;
 			new_bdev->bd_inode = inode;
+			new_bdev->bd_part_count = 0;
+			sema_init(&new_bdev->bd_part_sem, 1);
 			inode->i_mode = S_IFBLK;
 			inode->i_rdev = kdev;
 			inode->i_bdev = new_bdev;
@@ -415,9 +417,9 @@ int get_blkdev_list(char * p)
 	Return the function table of a device.
 	Load the driver if needed.
 */
-const struct block_device_operations * get_blkfops(unsigned int major)
+struct block_device_operations * get_blkfops(unsigned int major)
 {
-	const struct block_device_operations *ret = NULL;
+	struct block_device_operations *ret = NULL;
 
 	/* major 0 is used for non-device mounts */
 	if (major && major < MAX_BLKDEV) {
@@ -479,7 +481,7 @@ int unregister_blkdev(unsigned int major, const char * name)
 int check_disk_change(kdev_t dev)
 {
 	int i;
-	const struct block_device_operations * bdops = NULL;
+	struct block_device_operations * bdops = NULL;
 
 	i = major(dev);
 	if (i < MAX_BLKDEV)
@@ -567,11 +569,17 @@ static int do_open(struct block_device *bdev, struct inode *inode, struct file *
 			}
 		}
 	}
+	if (bdev->bd_contains == bdev) {
 		if (current_ops->open) {
 			ret = current_ops->open(inode, file);
 			if (ret)
 				goto out2;
 		}
+	} else {
+		down(&bdev->bd_contains->bd_part_sem);
+		bdev->bd_contains->bd_part_count++;
+		up(&bdev->bd_contains->bd_part_sem);
+	}
 	if (!bdev->bd_op)
 		bdev->bd_op = ops;
 	else if (owner)
@@ -688,8 +696,14 @@ int blkdev_put(struct block_device *bdev, int kind)
 	}
 	if (!--bdev->bd_openers)
 		kill_bdev(bdev);
+	if (bdev->bd_contains == bdev) {
 		if (bdev->bd_op->release)
 			ret = bdev->bd_op->release(bd_inode, NULL);
+	} else {
+		down(&bdev->bd_contains->bd_part_sem);
+		bdev->bd_contains->bd_part_count--;
+		up(&bdev->bd_contains->bd_part_sem);
+	}
 	if (!bdev->bd_openers) {
 		if (bdev->bd_op->owner)
 			__MOD_DEC_USE_COUNT(bdev->bd_op->owner);

fs/partitions/check.c

@@ -372,23 +372,8 @@ static void check_partition(struct gendisk *hd, kdev_t dev)
 			sprintf(state->name, "p");
 	}
 	bdev = bdget(kdev_t_to_nr(dev));
-	bdev->bd_contains = bdev;
-	bdev->bd_inode->i_size = (loff_t)hd->part[minor(dev)].nr_sects << 9;
-	if (!bdev->bd_openers) {
-		struct blk_dev_struct *p = blk_dev + major(dev);
-		unsigned bsize = bdev_hardsect_size(bdev);
-		while (bsize < PAGE_CACHE_SIZE) {
-			if (bdev->bd_inode->i_size & bsize)
-				break;
-			bsize <<= 1;
-		}
-		if (p->queue)
-			bdev->bd_queue =  p->queue(dev);
-		else
-			bdev->bd_queue = &p->request_queue;
-		bdev->bd_block_size = bsize;
-		bdev->bd_inode->i_blkbits = blksize_bits(bsize);
-	}
+	if (blkdev_get(bdev, FMODE_READ, 0, BDEV_RAW))
+		goto out;
 	state->limit = 1<<hd->minor_shift;
 	for (i = 0; check_part[i]; i++) {
 		int res, j;
@@ -417,10 +402,8 @@ static void check_partition(struct gendisk *hd, kdev_t dev)
 
 	printk(" unknown partition table\n");
 setup_devfs:
-	invalidate_bdev(bdev, 1);
-	truncate_inode_pages(bdev->bd_inode->i_mapping, 0);
-	bdput(bdev);
-
+	blkdev_put(bdev, BDEV_RAW);
+out:
 	/* Setup driverfs tree */
 	if (hd->sizes)
 		driverfs_create_partitions(hd, minor(dev));

include/linux/fs.h

@@ -342,7 +342,7 @@ struct block_device {
 	struct inode *		bd_inode;
 	dev_t			bd_dev;  /* not a kdev_t - it's a search key */
 	int			bd_openers;
-	const struct block_device_operations *bd_op;
+	struct block_device_operations *bd_op;
 	struct request_queue	*bd_queue;
 	struct semaphore	bd_sem;	/* open/close mutex */
 	struct list_head	bd_inodes;
@@ -351,6 +351,8 @@ struct block_device {
 	struct block_device *	bd_contains;
 	unsigned		bd_block_size;
 	unsigned long		bd_offset;
+	struct semaphore	bd_part_sem;
+	unsigned		bd_part_count;
 };
 
 struct inode {
@@ -1083,7 +1085,7 @@ extern void bd_release(struct block_device *);
 extern void blk_run_queues(void);
 
 /* fs/devices.c */
-extern const struct block_device_operations *get_blkfops(unsigned int);
+extern struct block_device_operations *get_blkfops(unsigned int);
 extern int register_chrdev(unsigned int, const char *, struct file_operations *);
 extern int unregister_chrdev(unsigned int, const char *);
 extern int chrdev_open(struct inode *, struct file *);
@@ -1293,5 +1295,31 @@ static inline ino_t parent_ino(struct dentry *dentry)
 	return res;
 }
 
+/* NOTE NOTE NOTE: this interface _will_ change in a couple of patches */
+
+static inline int dev_lock_part(kdev_t dev)
+{
+	struct block_device *bdev = bdget(kdev_t_to_nr(dev));
+	if (!bdev)
+		return -ENOMEM;
+	if (!down_trylock(&bdev->bd_part_sem)) {
+		if (!bdev->bd_part_count)
+			return 0;
+		up(&bdev->bd_part_sem);
+	}
+	bdput(bdev);
+	return -EBUSY;
+}
+
+static inline void dev_unlock_part(kdev_t dev)
+{
+	struct block_device *bdev = bdget(kdev_t_to_nr(dev));
+	if (!bdev)
+		BUG();
+	up(&bdev->bd_part_sem);
+	bdput(bdev);
+	bdput(bdev);
+}
+
 #endif /* __KERNEL__ */
 #endif /* _LINUX_FS_H */