BIOCLONE(9F) | Kernel Functions for Drivers | BIOCLONE(9F) |
bioclone - clone another buffer
#include <sys/ddi.h> #include <sys/sunddi.h> struct buf *bioclone(struct buf *bp, off_t off, size_t len, dev_t dev,
daddr_t blkno, int (*iodone) (struct buf *), struct buf *bp_mem,
int sleepflag);
illumos DDI specific (illumos DDI).
bp
off
len
dev
blkno
iodone
bp_mem
sleepflag
The bioclone() function returns an initialized buffer to perform I/O to a portion of another buffer. The new buffer will be set up to perform I/O to the range within the original I/O request specified by the parameters off and len. An offset 0 starts the new I/O request at the same address as the original request. off + len must not exceed b_bcount, the length of the original request. The device number dev specifies the device to which the buffer is to perform I/O. blkno is the block number on device. It will be assigned to the b_blkno field of the cloned buffer structure. iodone lets the driver identify a specific biodone(9F) routine to be called by the driver when the I/O is complete. bp_mem determines from where the space for the buffer should be allocated. If bp_mem is NULL, bioclone() will allocate a new buffer using getrbuf(9F). If sleepflag is set to KM_SLEEP, the driver may sleep until space is freed up. If sleepflag is set to KM_NOSLEEP, the driver will not sleep. In either case, a pointer to the allocated space is returned or NULL to indicate that no space was available. After the transfer is completed, the buffer has to be freed using freerbuf(9F). If bp_mem is not NULL, it will be used as the space for the buffer structure. The driver has to ensure that bp_mem is initialized properly either using getrbuf(9F) or bioinit(9F).
If the original buffer is mapped into the kernel virtual address space using bp_mapin(9F) before calling bioclone(), a clone buffer will share the kernel mapping of the original buffer. An additional bp_mapin() to get a kernel mapping for the clone buffer is not necessary.
The driver has to ensure that the original buffer is not freed while any of the clone buffers is still performing I/O. The biodone() function has to be called on all clone buffers before it is called on the original buffer.
The bioclone() function returns a pointer to the initialized buffer header, or NULL if no space is available.
The bioclone() function can be called from user, interrup, or interrupt context. Drivers must not allow bioclone() to sleep if called from an interrupt routine.
Example 1 Using bioclone() for Disk Striping
A device driver can use bioclone() for disk striping. For each disk in the stripe, a clone buffer is created which performs I/O to a portion of the original buffer.
static int stripe_strategy(struct buf *bp) {
...
bp_orig = bp;
bp_1 = bioclone(bp_orig, 0, size_1, dev_1, blkno_1,
stripe_done, NULL, KM_SLEEP);
fragment++;
...
bp_n = bioclone(bp_orig, offset_n, size_n, dev_n,
blkno_n, stripe_done, NULL, KM_SLEEP);
fragment++;
/* submit bp_1 ... bp_n to device */
xxstrategy(bp_x);
return (0); } static uint_t xxintr(caddr_t arg) {
...
/*
* get bp of completed subrequest. biodone(9F) will
* call stripe_done()
*/
biodone(bp);
return (0); } static int stripe_done(struct buf *bp) {
...
freerbuf(bp);
fragment--;
if (fragment == 0) {
/* get bp_orig */
biodone(bp_orig);
}
return (0); }
biodone(9F), bp_mapin(9F), freerbuf(9F), getrbuf(9F), buf(9S)
Writing Device Drivers
January 16, 2006 | OmniOS |