2 * Copyright (c) 1989, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * @(#)spec_vnops.c 8.14 (Berkeley) 5/21/95
34 * $FreeBSD: src/sys/miscfs/specfs/spec_vnops.c,v 1.131.2.4 2001/02/26 04:23:20 jlemon Exp $
35 * $DragonFly: src/sys/vfs/specfs/spec_vnops.c,v 1.56 2007/08/21 17:26:48 dillon Exp $
38 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
44 #include <sys/device.h>
45 #include <sys/mount.h>
46 #include <sys/vnode.h>
48 #include <sys/fcntl.h>
49 #include <sys/vmmeter.h>
54 #include <vm/vm_object.h>
55 #include <vm/vm_page.h>
56 #include <vm/vm_pager.h>
58 #include <machine/limits.h>
62 #include <sys/thread2.h>
65 * Specfs chained debugging (bitmask)
67 * 0 - disable debugging
68 * 1 - report chained I/Os
69 * 2 - force 4K chained I/Os
71 #define SPEC_CHAIN_DEBUG 0
73 static int spec_advlock (struct vop_advlock_args
*);
74 static int spec_bmap (struct vop_bmap_args
*);
75 static int spec_close (struct vop_close_args
*);
76 static int spec_freeblks (struct vop_freeblks_args
*);
77 static int spec_fsync (struct vop_fsync_args
*);
78 static int spec_getpages (struct vop_getpages_args
*);
79 static int spec_inactive (struct vop_inactive_args
*);
80 static int spec_ioctl (struct vop_ioctl_args
*);
81 static int spec_open (struct vop_open_args
*);
82 static int spec_poll (struct vop_poll_args
*);
83 static int spec_kqfilter (struct vop_kqfilter_args
*);
84 static int spec_print (struct vop_print_args
*);
85 static int spec_read (struct vop_read_args
*);
86 static int spec_strategy (struct vop_strategy_args
*);
87 static int spec_write (struct vop_write_args
*);
88 static void spec_strategy_done(struct bio
*nbio
);
90 struct vop_ops spec_vnode_vops
= {
91 .vop_default
= vop_defaultop
,
92 .vop_access
= (void *)vop_ebadf
,
93 .vop_advlock
= spec_advlock
,
94 .vop_bmap
= spec_bmap
,
95 .vop_close
= spec_close
,
96 .vop_old_create
= (void *)vop_panic
,
97 .vop_freeblks
= spec_freeblks
,
98 .vop_fsync
= spec_fsync
,
99 .vop_getpages
= spec_getpages
,
100 .vop_inactive
= spec_inactive
,
101 .vop_ioctl
= spec_ioctl
,
102 .vop_old_link
= (void *)vop_panic
,
103 .vop_old_mkdir
= (void *)vop_panic
,
104 .vop_old_mknod
= (void *)vop_panic
,
105 .vop_open
= spec_open
,
106 .vop_pathconf
= vop_stdpathconf
,
107 .vop_poll
= spec_poll
,
108 .vop_kqfilter
= spec_kqfilter
,
109 .vop_print
= spec_print
,
110 .vop_read
= spec_read
,
111 .vop_readdir
= (void *)vop_panic
,
112 .vop_readlink
= (void *)vop_panic
,
113 .vop_reallocblks
= (void *)vop_panic
,
114 .vop_reclaim
= (void *)vop_null
,
115 .vop_old_remove
= (void *)vop_panic
,
116 .vop_old_rename
= (void *)vop_panic
,
117 .vop_old_rmdir
= (void *)vop_panic
,
118 .vop_setattr
= (void *)vop_ebadf
,
119 .vop_strategy
= spec_strategy
,
120 .vop_old_symlink
= (void *)vop_panic
,
121 .vop_write
= spec_write
124 struct vop_ops
*spec_vnode_vops_p
= &spec_vnode_vops
;
126 VNODEOP_SET(spec_vnode_vops
);
128 extern int dev_ref_debug
;
134 spec_vnoperate(struct vop_generic_args
*ap
)
136 return (VOCALL(&spec_vnode_vops
, ap
));
139 static void spec_getpages_iodone (struct bio
*bio
);
142 * Open a special file.
144 * spec_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
149 spec_open(struct vop_open_args
*ap
)
151 struct vnode
*vp
= ap
->a_vp
;
157 * Don't allow open if fs is mounted -nodev.
159 if (vp
->v_mount
&& (vp
->v_mount
->mnt_flag
& MNT_NODEV
))
161 if (vp
->v_type
== VBLK
)
165 * Resolve the device. If the vnode is already open v_rdev may
166 * already be resolved. However, if the device changes out from
167 * under us we report it (and, for now, we allow it). Since
168 * v_release_rdev() zero's v_opencount, we have to save and restore
169 * it when replacing the rdev reference.
171 if (vp
->v_rdev
!= NULL
) {
172 dev
= get_dev(vp
->v_umajor
, vp
->v_uminor
);
173 if (dev
!= vp
->v_rdev
) {
174 int oc
= vp
->v_opencount
;
176 "Warning: spec_open: dev %s was lost",
177 vp
->v_rdev
->si_name
);
179 error
= v_associate_rdev(vp
,
180 get_dev(vp
->v_umajor
, vp
->v_uminor
));
182 kprintf(", reacquisition failed\n");
184 vp
->v_opencount
= oc
;
185 kprintf(", reacquisition successful\n");
191 error
= v_associate_rdev(vp
, get_dev(vp
->v_umajor
, vp
->v_uminor
));
197 * Prevent degenerate open/close sequences from nulling out rdev.
200 KKASSERT(dev
!= NULL
);
203 * Make this field valid before any I/O in ->d_open. XXX the
204 * device itself should probably be required to initialize
205 * this field in d_open.
207 if (!dev
->si_iosize_max
)
208 dev
->si_iosize_max
= DFLTPHYS
;
211 * XXX: Disks get special billing here, but it is mostly wrong.
212 * XXX: diskpartitions can overlap and the real checks should
213 * XXX: take this into account, and consequently they need to
214 * XXX: live in the diskslicing code. Some checks do.
216 if (vn_isdisk(vp
, NULL
) && ap
->a_cred
!= FSCRED
&&
217 (ap
->a_mode
& FWRITE
)) {
219 * Never allow opens for write if the device is mounted R/W
221 if (vp
->v_rdev
&& vp
->v_rdev
->si_mountpoint
&&
222 !(vp
->v_rdev
->si_mountpoint
->mnt_flag
& MNT_RDONLY
)) {
228 * When running in secure mode, do not allow opens
229 * for writing if the device is mounted
231 if (securelevel
>= 1 && vfs_mountedon(vp
)) {
237 * When running in very secure mode, do not allow
238 * opens for writing of any devices.
240 if (securelevel
>= 2) {
246 /* XXX: Special casing of ttys for deadfs. Probably redundant */
247 if (dev_dflags(dev
) & D_TTY
)
248 vp
->v_flag
|= VISTTY
;
251 * dev_dopen() is always called for each open. dev_dclose() is
252 * only called for the last close unless D_TRACKCLOSE is set.
255 error
= dev_dopen(dev
, ap
->a_mode
, S_IFCHR
, ap
->a_cred
);
256 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
261 if (dev_dflags(dev
) & D_TTY
) {
266 kprintf("Warning:%s: no t_stop, using nottystop\n", devtoname(dev
));
267 tp
->t_stop
= nottystop
;
273 * If this is 'disk' or disk-like device, associate a VM object
276 if (vn_isdisk(vp
, NULL
)) {
277 if (!dev
->si_bsize_phys
)
278 dev
->si_bsize_phys
= DEV_BSIZE
;
279 vinitvmio(vp
, IDX_TO_OFF(INT_MAX
));
281 if ((dev_dflags(dev
) & D_DISK
) == 0) {
284 kprintf("WARNING: driver %s should register devices with make_dev() (cdev_t = \"%s\")\n",
290 * If we were handed a file pointer we may be able to install a
291 * shortcut which issues device read and write operations directly
292 * from the fileops rather then having to go through spec_read()
296 vn_setspecops(ap
->a_fp
);
299 kprintf("spec_open: %s %d\n", dev
->si_name
, vp
->v_opencount
);
302 if (vp
->v_opencount
== 0)
313 * spec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
314 * struct ucred *a_cred)
318 spec_read(struct vop_read_args
*ap
)
331 if (dev
== NULL
) /* device was revoked */
333 if (uio
->uio_resid
== 0)
337 error
= dev_dread(dev
, uio
, ap
->a_ioflag
);
338 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
345 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
346 * struct ucred *a_cred)
350 spec_write(struct vop_write_args
*ap
)
363 KKASSERT(uio
->uio_segflg
!= UIO_NOCOPY
);
365 if (dev
== NULL
) /* device was revoked */
369 error
= dev_dwrite(dev
, uio
, ap
->a_ioflag
);
370 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
375 * Device ioctl operation.
377 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
378 * int a_fflag, struct ucred *a_cred)
382 spec_ioctl(struct vop_ioctl_args
*ap
)
386 if ((dev
= ap
->a_vp
->v_rdev
) == NULL
)
387 return (EBADF
); /* device was revoked */
389 return (dev_dioctl(dev
, ap
->a_command
, ap
->a_data
,
390 ap
->a_fflag
, ap
->a_cred
));
394 * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred)
398 spec_poll(struct vop_poll_args
*ap
)
402 if ((dev
= ap
->a_vp
->v_rdev
) == NULL
)
403 return (EBADF
); /* device was revoked */
404 return (dev_dpoll(dev
, ap
->a_events
));
408 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
412 spec_kqfilter(struct vop_kqfilter_args
*ap
)
416 if ((dev
= ap
->a_vp
->v_rdev
) == NULL
)
417 return (EBADF
); /* device was revoked */
418 return (dev_dkqfilter(dev
, ap
->a_kn
));
422 * Synch buffers associated with a block device
424 * spec_fsync(struct vnode *a_vp, int a_waitfor)
428 spec_fsync(struct vop_fsync_args
*ap
)
430 struct vnode
*vp
= ap
->a_vp
;
433 if (!vn_isdisk(vp
, NULL
))
437 * Flush all dirty buffers associated with a block device.
439 error
= vfsync(vp
, ap
->a_waitfor
, 10000, NULL
, NULL
);
444 * spec_inactive(struct vnode *a_vp)
447 spec_inactive(struct vop_inactive_args
*ap
)
453 * Convert a vnode strategy call into a device strategy call. Vnode strategy
454 * calls are not limited to device DMA limits so we have to deal with the
457 * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
460 spec_strategy(struct vop_strategy_args
*ap
)
462 struct bio
*bio
= ap
->a_bio
;
463 struct buf
*bp
= bio
->bio_buf
;
470 if (bp
->b_cmd
!= BUF_CMD_READ
&&
471 (LIST_FIRST(&bp
->b_dep
)) != NULL
&& bioops
.io_start
) {
472 (*bioops
.io_start
)(bp
);
476 * Collect statistics on synchronous and asynchronous read
477 * and write counts for disks that have associated filesystems.
480 KKASSERT(vp
->v_rdev
!= NULL
); /* XXX */
481 if (vn_isdisk(vp
, NULL
) && (mp
= vp
->v_rdev
->si_mountpoint
) != NULL
) {
482 if (bp
->b_cmd
== BUF_CMD_READ
) {
483 if (bp
->b_flags
& B_ASYNC
)
484 mp
->mnt_stat
.f_asyncreads
++;
486 mp
->mnt_stat
.f_syncreads
++;
488 if (bp
->b_flags
& B_ASYNC
)
489 mp
->mnt_stat
.f_asyncwrites
++;
491 mp
->mnt_stat
.f_syncwrites
++;
496 * Device iosize limitations only apply to read and write. Shortcut
497 * the I/O if it fits.
499 if ((maxiosize
= vp
->v_rdev
->si_iosize_max
) == 0) {
500 kprintf("%s: si_iosize_max not set!\n", dev_dname(vp
->v_rdev
));
503 #if SPEC_CHAIN_DEBUG & 2
506 if (bp
->b_bcount
<= maxiosize
||
507 (bp
->b_cmd
!= BUF_CMD_READ
&& bp
->b_cmd
!= BUF_CMD_WRITE
)) {
508 dev_dstrategy_chain(vp
->v_rdev
, bio
);
513 * Clone the buffer and set up an I/O chain to chunk up the I/O.
515 nbp
= kmalloc(sizeof(*bp
), M_DEVBUF
, M_INTWAIT
|M_ZERO
);
517 LIST_INIT(&nbp
->b_dep
);
519 BUF_LOCK(nbp
, LK_EXCLUSIVE
);
522 nbp
->b_flags
= B_PAGING
| (bp
->b_flags
& B_BNOCLIP
);
523 nbp
->b_data
= bp
->b_data
;
524 nbp
->b_bio1
.bio_done
= spec_strategy_done
;
525 nbp
->b_bio1
.bio_offset
= bio
->bio_offset
;
526 nbp
->b_bio1
.bio_caller_info1
.ptr
= bio
;
529 * Start the first transfer
531 if (vn_isdisk(vp
, NULL
))
532 chunksize
= vp
->v_rdev
->si_bsize_phys
;
534 chunksize
= DEV_BSIZE
;
535 chunksize
= maxiosize
/ chunksize
* chunksize
;
536 #if SPEC_CHAIN_DEBUG & 1
537 kprintf("spec_strategy chained I/O chunksize=%d\n", chunksize
);
539 nbp
->b_cmd
= bp
->b_cmd
;
540 nbp
->b_bcount
= chunksize
;
541 nbp
->b_bufsize
= chunksize
; /* used to detect a short I/O */
542 nbp
->b_bio1
.bio_caller_info2
.index
= chunksize
;
544 #if SPEC_CHAIN_DEBUG & 1
545 kprintf("spec_strategy: chain %p offset %d/%d bcount %d\n",
546 bp
, 0, bp
->b_bcount
, nbp
->b_bcount
);
549 dev_dstrategy(vp
->v_rdev
, &nbp
->b_bio1
);
554 * Chunked up transfer completion routine - chain transfers until done
558 spec_strategy_done(struct bio
*nbio
)
560 struct buf
*nbp
= nbio
->bio_buf
;
561 struct bio
*bio
= nbio
->bio_caller_info1
.ptr
; /* original bio */
562 struct buf
*bp
= bio
->bio_buf
; /* original bp */
563 int chunksize
= nbio
->bio_caller_info2
.index
; /* chunking */
564 int boffset
= nbp
->b_data
- bp
->b_data
;
566 if (nbp
->b_flags
& B_ERROR
) {
568 * An error terminates the chain, propogate the error back
571 bp
->b_flags
|= B_ERROR
;
572 bp
->b_error
= nbp
->b_error
;
573 bp
->b_resid
= bp
->b_bcount
- boffset
+
574 (nbp
->b_bcount
- nbp
->b_resid
);
575 #if SPEC_CHAIN_DEBUG & 1
576 kprintf("spec_strategy: chain %p error %d bcount %d/%d\n",
577 bp
, bp
->b_error
, bp
->b_bcount
,
578 bp
->b_bcount
- bp
->b_resid
);
580 kfree(nbp
, M_DEVBUF
);
582 } else if (nbp
->b_resid
) {
584 * A short read or write terminates the chain
586 bp
->b_error
= nbp
->b_error
;
587 bp
->b_resid
= bp
->b_bcount
- boffset
+
588 (nbp
->b_bcount
- nbp
->b_resid
);
589 #if SPEC_CHAIN_DEBUG & 1
590 kprintf("spec_strategy: chain %p short read(1) bcount %d/%d\n",
591 bp
, bp
->b_bcount
- bp
->b_resid
, bp
->b_bcount
);
593 kfree(nbp
, M_DEVBUF
);
595 } else if (nbp
->b_bcount
!= nbp
->b_bufsize
) {
597 * A short read or write can also occur by truncating b_bcount
599 #if SPEC_CHAIN_DEBUG & 1
600 kprintf("spec_strategy: chain %p short read(2) bcount %d/%d\n",
601 bp
, nbp
->b_bcount
+ boffset
, bp
->b_bcount
);
604 bp
->b_bcount
= nbp
->b_bcount
+ boffset
;
605 bp
->b_resid
= nbp
->b_resid
;
606 kfree(nbp
, M_DEVBUF
);
608 } else if (nbp
->b_bcount
+ boffset
== bp
->b_bcount
) {
610 * No more data terminates the chain
612 #if SPEC_CHAIN_DEBUG & 1
613 kprintf("spec_strategy: chain %p finished bcount %d\n",
618 kfree(nbp
, M_DEVBUF
);
624 boffset
+= nbp
->b_bcount
;
625 nbp
->b_data
= bp
->b_data
+ boffset
;
626 nbp
->b_bcount
= bp
->b_bcount
- boffset
;
627 if (nbp
->b_bcount
> chunksize
)
628 nbp
->b_bcount
= chunksize
;
629 nbp
->b_bio1
.bio_done
= spec_strategy_done
;
630 nbp
->b_bio1
.bio_offset
= bio
->bio_offset
+ boffset
;
632 #if SPEC_CHAIN_DEBUG & 1
633 kprintf("spec_strategy: chain %p offset %d/%d bcount %d\n",
634 bp
, boffset
, bp
->b_bcount
, nbp
->b_bcount
);
637 dev_dstrategy(nbp
->b_vp
->v_rdev
, &nbp
->b_bio1
);
642 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
645 spec_freeblks(struct vop_freeblks_args
*ap
)
650 * XXX: This assumes that strategy does the deed right away.
651 * XXX: this may not be TRTTD.
653 KKASSERT(ap
->a_vp
->v_rdev
!= NULL
);
654 if ((dev_dflags(ap
->a_vp
->v_rdev
) & D_CANFREE
) == 0)
656 bp
= geteblk(ap
->a_length
);
657 bp
->b_cmd
= BUF_CMD_FREEBLKS
;
658 bp
->b_bio1
.bio_offset
= ap
->a_offset
;
659 bp
->b_bcount
= ap
->a_length
;
660 dev_dstrategy(ap
->a_vp
->v_rdev
, &bp
->b_bio1
);
665 * Implement degenerate case where the block requested is the block
666 * returned, and assume that the entire device is contiguous in regards
667 * to the contiguous block range (runp and runb).
669 * spec_bmap(struct vnode *a_vp, off_t a_loffset,
670 * off_t *a_doffsetp, int *a_runp, int *a_runb)
673 spec_bmap(struct vop_bmap_args
*ap
)
675 if (ap
->a_doffsetp
!= NULL
)
676 *ap
->a_doffsetp
= ap
->a_loffset
;
677 if (ap
->a_runp
!= NULL
)
678 *ap
->a_runp
= MAXBSIZE
;
679 if (ap
->a_runb
!= NULL
) {
680 if (ap
->a_loffset
< MAXBSIZE
)
681 *ap
->a_runb
= (int)ap
->a_loffset
;
683 *ap
->a_runb
= MAXBSIZE
;
689 * Device close routine
691 * spec_close(struct vnode *a_vp, int a_fflag)
693 * NOTE: the vnode may or may not be locked on call.
697 spec_close(struct vop_close_args
*ap
)
699 struct proc
*p
= curproc
;
700 struct vnode
*vp
= ap
->a_vp
;
701 cdev_t dev
= vp
->v_rdev
;
706 * Hack: a tty device that is a controlling terminal
707 * has a reference from the session structure.
708 * We cannot easily tell that a character device is
709 * a controlling terminal, unless it is the closing
710 * process' controlling terminal. In that case,
711 * if the reference count is 2 (this last descriptor
712 * plus the session), release the reference from the session.
714 * It is possible for v_opencount to be 0 or 1 in this case, 0
715 * because the tty might have been revoked.
719 if (vcount(vp
) == 2 && vp
->v_opencount
<= 1 &&
720 p
&& vp
== p
->p_session
->s_ttyvp
) {
721 p
->p_session
->s_ttyvp
= NULL
;
726 * Vnodes can be opened and close multiple times. Do not really
727 * close the device unless (1) it is being closed forcibly,
728 * (2) the device wants to track closes, or (3) this is the last
729 * vnode doing its last close on the device.
731 * XXX the VXLOCK (force close) case can leave vnodes referencing
734 if (dev
&& ((vp
->v_flag
& VRECLAIMED
) ||
735 (dev_dflags(dev
) & D_TRACKCLOSE
) ||
736 (vcount(vp
) <= 1 && vp
->v_opencount
== 1))) {
738 if (vn_islocked(vp
)) {
742 error
= dev_dclose(dev
, ap
->a_fflag
, S_IFCHR
);
744 vn_lock(vp
, LK_EXCLUSIVE
| LK_RETRY
);
750 * Track the actual opens and closes on the vnode. The last close
751 * disassociates the rdev. If the rdev is already disassociated
752 * the vnode might have been revoked and no further opencount
756 /*KKASSERT(vp->v_opencount > 0);*/
758 kprintf("spec_close: %s %d\n",
759 dev
->si_name
, vp
->v_opencount
- 1);
761 if (vp
->v_opencount
== 1)
770 * Print out the contents of a special device vnode.
772 * spec_print(struct vnode *a_vp)
775 spec_print(struct vop_print_args
*ap
)
777 kprintf("tag VT_NON, dev %s\n", devtoname(ap
->a_vp
->v_rdev
));
782 * Special device advisory byte-level locks.
784 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
785 * struct flock *a_fl, int a_flags)
789 spec_advlock(struct vop_advlock_args
*ap
)
791 return ((ap
->a_flags
& F_POSIX
) ? EINVAL
: EOPNOTSUPP
);
795 spec_getpages_iodone(struct bio
*bio
)
797 bio
->bio_buf
->b_cmd
= BUF_CMD_DONE
;
798 wakeup(bio
->bio_buf
);
802 * spec_getpages() - get pages associated with device vnode.
804 * Note that spec_read and spec_write do not use the buffer cache, so we
805 * must fully implement getpages here.
808 spec_getpages(struct vop_getpages_args
*ap
)
816 int toff
, nextoff
, nread
;
817 struct vnode
*vp
= ap
->a_vp
;
822 pcount
= round_page(ap
->a_count
) / PAGE_SIZE
;
825 * Calculate the offset of the transfer and do sanity check.
827 offset
= IDX_TO_OFF(ap
->a_m
[0]->pindex
) + ap
->a_offset
;
830 * Round up physical size for real devices. We cannot round using
831 * v_mount's block size data because v_mount has nothing to do with
832 * the device. i.e. it's usually '/dev'. We need the physical block
833 * size for the device itself.
835 * We can't use v_rdev->si_mountpoint because it only exists when the
836 * block device is mounted. However, we can use v_rdev.
839 if (vn_isdisk(vp
, NULL
))
840 blksiz
= vp
->v_rdev
->si_bsize_phys
;
844 size
= (ap
->a_count
+ blksiz
- 1) & ~(blksiz
- 1);
847 kva
= (vm_offset_t
)bp
->b_data
;
850 * Map the pages to be read into the kva.
852 pmap_qenter(kva
, ap
->a_m
, pcount
);
854 /* Build a minimal buffer header. */
855 bp
->b_cmd
= BUF_CMD_READ
;
858 bp
->b_runningbufspace
= size
;
859 runningbufspace
+= bp
->b_runningbufspace
;
861 bp
->b_bio1
.bio_offset
= offset
;
862 bp
->b_bio1
.bio_done
= spec_getpages_iodone
;
864 mycpu
->gd_cnt
.v_vnodein
++;
865 mycpu
->gd_cnt
.v_vnodepgsin
+= pcount
;
868 vn_strategy(ap
->a_vp
, &bp
->b_bio1
);
872 /* We definitely need to be at splbio here. */
873 while (bp
->b_cmd
!= BUF_CMD_DONE
)
874 tsleep(bp
, 0, "spread", 0);
878 if (bp
->b_flags
& B_ERROR
) {
886 * If EOF is encountered we must zero-extend the result in order
887 * to ensure that the page does not contain garabge. When no
888 * error occurs, an early EOF is indicated if b_bcount got truncated.
889 * b_resid is relative to b_bcount and should be 0, but some devices
890 * might indicate an EOF with b_resid instead of truncating b_bcount.
892 nread
= bp
->b_bcount
- bp
->b_resid
;
893 if (nread
< ap
->a_count
)
894 bzero((caddr_t
)kva
+ nread
, ap
->a_count
- nread
);
895 pmap_qremove(kva
, pcount
);
898 for (i
= 0, toff
= 0; i
< pcount
; i
++, toff
= nextoff
) {
899 nextoff
= toff
+ PAGE_SIZE
;
902 m
->flags
&= ~PG_ZERO
;
904 if (nextoff
<= nread
) {
905 m
->valid
= VM_PAGE_BITS_ALL
;
907 } else if (toff
< nread
) {
909 * Since this is a VM request, we have to supply the
910 * unaligned offset to allow vm_page_set_validclean()
911 * to zero sub-DEV_BSIZE'd portions of the page.
913 vm_page_set_validclean(m
, 0, nread
- toff
);
919 if (i
!= ap
->a_reqpage
) {
921 * Just in case someone was asking for this page we
922 * now tell them that it is ok to use.
924 if (!error
|| (m
->valid
== VM_PAGE_BITS_ALL
)) {
926 if (m
->flags
& PG_WANTED
) {
929 vm_page_deactivate(m
);
938 } else if (m
->valid
) {
941 * Since this is a VM request, we need to make the
942 * entire page presentable by zeroing invalid sections.
944 if (m
->valid
!= VM_PAGE_BITS_ALL
)
945 vm_page_zero_invalid(m
, FALSE
);
949 m
= ap
->a_m
[ap
->a_reqpage
];
951 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
952 devtoname(vp
->v_rdev
), error
, bp
, bp
->b_vp
);
954 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
955 size
, bp
->b_resid
, ap
->a_count
, m
->valid
);
957 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
958 nread
, ap
->a_reqpage
, (u_long
)m
->pindex
, pcount
);
960 * Free the buffer header back to the swap buffer pool.
963 return VM_PAGER_ERROR
;
966 * Free the buffer header back to the swap buffer pool.