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 1.58 2008/05/06 00:14:12 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);
89 static int spec_getattr (struct vop_getattr_args *);
91 struct vop_ops spec_vnode_vops = {
92 .vop_default = vop_defaultop,
93 .vop_access = (void *)vop_ebadf,
94 .vop_advlock = spec_advlock,
95 .vop_bmap = spec_bmap,
96 .vop_close = spec_close,
97 .vop_old_create = (void *)vop_panic,
98 .vop_freeblks = spec_freeblks,
99 .vop_fsync = spec_fsync,
100 .vop_getpages = spec_getpages,
101 .vop_inactive = spec_inactive,
102 .vop_ioctl = spec_ioctl,
103 .vop_old_link = (void *)vop_panic,
104 .vop_old_mkdir = (void *)vop_panic,
105 .vop_old_mknod = (void *)vop_panic,
106 .vop_open = spec_open,
107 .vop_pathconf = vop_stdpathconf,
108 .vop_poll = spec_poll,
109 .vop_kqfilter = spec_kqfilter,
110 .vop_print = spec_print,
111 .vop_read = spec_read,
112 .vop_readdir = (void *)vop_panic,
113 .vop_readlink = (void *)vop_panic,
114 .vop_reallocblks = (void *)vop_panic,
115 .vop_reclaim = (void *)vop_null,
116 .vop_old_remove = (void *)vop_panic,
117 .vop_old_rename = (void *)vop_panic,
118 .vop_old_rmdir = (void *)vop_panic,
119 .vop_setattr = (void *)vop_ebadf,
120 .vop_getattr = spec_getattr,
121 .vop_strategy = spec_strategy,
122 .vop_old_symlink = (void *)vop_panic,
123 .vop_write = spec_write
126 struct vop_ops *spec_vnode_vops_p = &spec_vnode_vops;
128 VNODEOP_SET(spec_vnode_vops);
130 extern int dev_ref_debug;
136 spec_vnoperate(struct vop_generic_args *ap)
138 return (VOCALL(&spec_vnode_vops, ap));
141 static void spec_getpages_iodone (struct bio *bio);
144 * Open a special file.
146 * spec_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
151 spec_open(struct vop_open_args *ap)
153 struct vnode *vp = ap->a_vp;
159 * Don't allow open if fs is mounted -nodev.
161 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
163 if (vp->v_type == VBLK)
167 * Resolve the device. If the vnode is already open v_rdev may
168 * already be resolved. However, if the device changes out from
169 * under us we report it (and, for now, we allow it). Since
170 * v_release_rdev() zero's v_opencount, we have to save and restore
171 * it when replacing the rdev reference.
173 if (vp->v_rdev != NULL) {
174 dev = get_dev(vp->v_umajor, vp->v_uminor);
175 if (dev != vp->v_rdev) {
176 int oc = vp->v_opencount;
178 "Warning: spec_open: dev %s was lost",
179 vp->v_rdev->si_name);
181 error = v_associate_rdev(vp,
182 get_dev(vp->v_umajor, vp->v_uminor));
184 kprintf(", reacquisition failed\n");
186 vp->v_opencount = oc;
187 kprintf(", reacquisition successful\n");
193 error = v_associate_rdev(vp, get_dev(vp->v_umajor, vp->v_uminor));
199 * Prevent degenerate open/close sequences from nulling out rdev.
202 KKASSERT(dev != NULL);
205 * Make this field valid before any I/O in ->d_open. XXX the
206 * device itself should probably be required to initialize
207 * this field in d_open.
209 if (!dev->si_iosize_max)
210 dev->si_iosize_max = DFLTPHYS;
213 * XXX: Disks get special billing here, but it is mostly wrong.
214 * XXX: diskpartitions can overlap and the real checks should
215 * XXX: take this into account, and consequently they need to
216 * XXX: live in the diskslicing code. Some checks do.
218 if (vn_isdisk(vp, NULL) && ap->a_cred != FSCRED &&
219 (ap->a_mode & FWRITE)) {
221 * Never allow opens for write if the device is mounted R/W
223 if (vp->v_rdev && vp->v_rdev->si_mountpoint &&
224 !(vp->v_rdev->si_mountpoint->mnt_flag & MNT_RDONLY)) {
230 * When running in secure mode, do not allow opens
231 * for writing if the device is mounted
233 if (securelevel >= 1 && vfs_mountedon(vp)) {
239 * When running in very secure mode, do not allow
240 * opens for writing of any devices.
242 if (securelevel >= 2) {
248 /* XXX: Special casing of ttys for deadfs. Probably redundant */
249 if (dev_dflags(dev) & D_TTY)
250 vp->v_flag |= VISTTY;
253 * dev_dopen() is always called for each open. dev_dclose() is
254 * only called for the last close unless D_TRACKCLOSE is set.
257 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred);
258 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
263 if (dev_dflags(dev) & D_TTY) {
268 kprintf("Warning:%s: no t_stop, using nottystop\n", devtoname(dev));
269 tp->t_stop = nottystop;
275 * If this is 'disk' or disk-like device, associate a VM object
278 if (vn_isdisk(vp, NULL)) {
279 if (!dev->si_bsize_phys)
280 dev->si_bsize_phys = DEV_BSIZE;
281 vinitvmio(vp, IDX_TO_OFF(INT_MAX));
283 if ((dev_dflags(dev) & D_DISK) == 0) {
286 kprintf("WARNING: driver %s should register devices with make_dev() (cdev_t = \"%s\")\n",
292 * If we were handed a file pointer we may be able to install a
293 * shortcut which issues device read and write operations directly
294 * from the fileops rather then having to go through spec_read()
298 vn_setspecops(ap->a_fp);
301 kprintf("spec_open: %s %d\n", dev->si_name, vp->v_opencount);
304 if (vp->v_opencount == 0)
315 * spec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
316 * struct ucred *a_cred)
320 spec_read(struct vop_read_args *ap)
333 if (dev == NULL) /* device was revoked */
335 if (uio->uio_resid == 0)
339 error = dev_dread(dev, uio, ap->a_ioflag);
340 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
347 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
348 * struct ucred *a_cred)
352 spec_write(struct vop_write_args *ap)
365 KKASSERT(uio->uio_segflg != UIO_NOCOPY);
367 if (dev == NULL) /* device was revoked */
371 error = dev_dwrite(dev, uio, ap->a_ioflag);
372 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
377 * This call modifying an upper layer filesystem's idea of the owner only.
378 * It does NOT replace the upper layer filesystem's getattr.
380 * XXX spec_getattr() - TEMPORARY - remove when devfs is fully installed.
381 * This is a horrible hack. The code will get confused
382 * if root chown's a tty device and then fails to chown
383 * it back after finishing with it. Non-root chowns are
384 * strictly stored in the device and don't have the issue.
387 spec_getattr (struct vop_getattr_args *ap)
389 struct vattr *vap = ap->a_vap;
390 struct vnode *vp = ap->a_vp;
393 if (vp->v_type != VCHR)
395 if ((dev = vp->v_rdev) == NULL) {
396 if ((dev = vp->v_rdev) == NULL)
397 dev = get_dev(vp->v_umajor, vp->v_uminor);
401 if (vap->va_uid == 0) {
402 vap->va_uid = dev->si_uid;
408 * Device ioctl operation.
410 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
411 * int a_fflag, struct ucred *a_cred)
415 spec_ioctl(struct vop_ioctl_args *ap)
419 if ((dev = ap->a_vp->v_rdev) == NULL)
420 return (EBADF); /* device was revoked */
422 return (dev_dioctl(dev, ap->a_command, ap->a_data,
423 ap->a_fflag, ap->a_cred));
427 * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred)
431 spec_poll(struct vop_poll_args *ap)
435 if ((dev = ap->a_vp->v_rdev) == NULL)
436 return (EBADF); /* device was revoked */
437 return (dev_dpoll(dev, ap->a_events));
441 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
445 spec_kqfilter(struct vop_kqfilter_args *ap)
449 if ((dev = ap->a_vp->v_rdev) == NULL)
450 return (EBADF); /* device was revoked */
451 return (dev_dkqfilter(dev, ap->a_kn));
455 * Synch buffers associated with a block device
457 * spec_fsync(struct vnode *a_vp, int a_waitfor)
461 spec_fsync(struct vop_fsync_args *ap)
463 struct vnode *vp = ap->a_vp;
466 if (!vn_isdisk(vp, NULL))
470 * Flush all dirty buffers associated with a block device.
472 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
477 * spec_inactive(struct vnode *a_vp)
480 spec_inactive(struct vop_inactive_args *ap)
486 * Convert a vnode strategy call into a device strategy call. Vnode strategy
487 * calls are not limited to device DMA limits so we have to deal with the
490 * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
493 spec_strategy(struct vop_strategy_args *ap)
495 struct bio *bio = ap->a_bio;
496 struct buf *bp = bio->bio_buf;
503 if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
507 * Collect statistics on synchronous and asynchronous read
508 * and write counts for disks that have associated filesystems.
511 KKASSERT(vp->v_rdev != NULL); /* XXX */
512 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
513 if (bp->b_cmd == BUF_CMD_READ) {
514 if (bp->b_flags & B_ASYNC)
515 mp->mnt_stat.f_asyncreads++;
517 mp->mnt_stat.f_syncreads++;
519 if (bp->b_flags & B_ASYNC)
520 mp->mnt_stat.f_asyncwrites++;
522 mp->mnt_stat.f_syncwrites++;
527 * Device iosize limitations only apply to read and write. Shortcut
528 * the I/O if it fits.
530 if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
531 kprintf("%s: si_iosize_max not set!\n", dev_dname(vp->v_rdev));
534 #if SPEC_CHAIN_DEBUG & 2
537 if (bp->b_bcount <= maxiosize ||
538 (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
539 dev_dstrategy_chain(vp->v_rdev, bio);
544 * Clone the buffer and set up an I/O chain to chunk up the I/O.
546 nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
550 BUF_LOCK(nbp, LK_EXCLUSIVE);
553 nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
554 nbp->b_data = bp->b_data;
555 nbp->b_bio1.bio_done = spec_strategy_done;
556 nbp->b_bio1.bio_offset = bio->bio_offset;
557 nbp->b_bio1.bio_caller_info1.ptr = bio;
560 * Start the first transfer
562 if (vn_isdisk(vp, NULL))
563 chunksize = vp->v_rdev->si_bsize_phys;
565 chunksize = DEV_BSIZE;
566 chunksize = maxiosize / chunksize * chunksize;
567 #if SPEC_CHAIN_DEBUG & 1
568 kprintf("spec_strategy chained I/O chunksize=%d\n", chunksize);
570 nbp->b_cmd = bp->b_cmd;
571 nbp->b_bcount = chunksize;
572 nbp->b_bufsize = chunksize; /* used to detect a short I/O */
573 nbp->b_bio1.bio_caller_info2.index = chunksize;
575 #if SPEC_CHAIN_DEBUG & 1
576 kprintf("spec_strategy: chain %p offset %d/%d bcount %d\n",
577 bp, 0, bp->b_bcount, nbp->b_bcount);
580 dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
585 * Chunked up transfer completion routine - chain transfers until done
589 spec_strategy_done(struct bio *nbio)
591 struct buf *nbp = nbio->bio_buf;
592 struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */
593 struct buf *bp = bio->bio_buf; /* original bp */
594 int chunksize = nbio->bio_caller_info2.index; /* chunking */
595 int boffset = nbp->b_data - bp->b_data;
597 if (nbp->b_flags & B_ERROR) {
599 * An error terminates the chain, propogate the error back
602 bp->b_flags |= B_ERROR;
603 bp->b_error = nbp->b_error;
604 bp->b_resid = bp->b_bcount - boffset +
605 (nbp->b_bcount - nbp->b_resid);
606 #if SPEC_CHAIN_DEBUG & 1
607 kprintf("spec_strategy: chain %p error %d bcount %d/%d\n",
608 bp, bp->b_error, bp->b_bcount,
609 bp->b_bcount - bp->b_resid);
611 kfree(nbp, M_DEVBUF);
613 } else if (nbp->b_resid) {
615 * A short read or write terminates the chain
617 bp->b_error = nbp->b_error;
618 bp->b_resid = bp->b_bcount - boffset +
619 (nbp->b_bcount - nbp->b_resid);
620 #if SPEC_CHAIN_DEBUG & 1
621 kprintf("spec_strategy: chain %p short read(1) bcount %d/%d\n",
622 bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
624 kfree(nbp, M_DEVBUF);
626 } else if (nbp->b_bcount != nbp->b_bufsize) {
628 * A short read or write can also occur by truncating b_bcount
630 #if SPEC_CHAIN_DEBUG & 1
631 kprintf("spec_strategy: chain %p short read(2) bcount %d/%d\n",
632 bp, nbp->b_bcount + boffset, bp->b_bcount);
635 bp->b_bcount = nbp->b_bcount + boffset;
636 bp->b_resid = nbp->b_resid;
637 kfree(nbp, M_DEVBUF);
639 } else if (nbp->b_bcount + boffset == bp->b_bcount) {
641 * No more data terminates the chain
643 #if SPEC_CHAIN_DEBUG & 1
644 kprintf("spec_strategy: chain %p finished bcount %d\n",
649 kfree(nbp, M_DEVBUF);
655 boffset += nbp->b_bcount;
656 nbp->b_data = bp->b_data + boffset;
657 nbp->b_bcount = bp->b_bcount - boffset;
658 if (nbp->b_bcount > chunksize)
659 nbp->b_bcount = chunksize;
660 nbp->b_bio1.bio_done = spec_strategy_done;
661 nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
663 #if SPEC_CHAIN_DEBUG & 1
664 kprintf("spec_strategy: chain %p offset %d/%d bcount %d\n",
665 bp, boffset, bp->b_bcount, nbp->b_bcount);
668 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
673 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
676 spec_freeblks(struct vop_freeblks_args *ap)
681 * XXX: This assumes that strategy does the deed right away.
682 * XXX: this may not be TRTTD.
684 KKASSERT(ap->a_vp->v_rdev != NULL);
685 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
687 bp = geteblk(ap->a_length);
688 bp->b_cmd = BUF_CMD_FREEBLKS;
689 bp->b_bio1.bio_offset = ap->a_offset;
690 bp->b_bcount = ap->a_length;
691 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
696 * Implement degenerate case where the block requested is the block
697 * returned, and assume that the entire device is contiguous in regards
698 * to the contiguous block range (runp and runb).
700 * spec_bmap(struct vnode *a_vp, off_t a_loffset,
701 * off_t *a_doffsetp, int *a_runp, int *a_runb)
704 spec_bmap(struct vop_bmap_args *ap)
706 if (ap->a_doffsetp != NULL)
707 *ap->a_doffsetp = ap->a_loffset;
708 if (ap->a_runp != NULL)
709 *ap->a_runp = MAXBSIZE;
710 if (ap->a_runb != NULL) {
711 if (ap->a_loffset < MAXBSIZE)
712 *ap->a_runb = (int)ap->a_loffset;
714 *ap->a_runb = MAXBSIZE;
720 * Device close routine
722 * spec_close(struct vnode *a_vp, int a_fflag)
724 * NOTE: The vnode may or may not be locked on call.
728 spec_close(struct vop_close_args *ap)
730 struct proc *p = curproc;
731 struct vnode *vp = ap->a_vp;
732 cdev_t dev = vp->v_rdev;
737 * A couple of hacks for devices and tty devices. The
738 * vnode ref count cannot be used to figure out the
739 * last close, but we can use v_opencount now that
740 * revoke works properly.
742 * Detect the last close on a controlling terminal and clear
743 * the session (half-close).
748 if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
749 p->p_session->s_ttyvp = NULL;
754 * Vnodes can be opened and closed multiple times. Do not really
755 * close the device unless (1) it is being closed forcibly,
756 * (2) the device wants to track closes, or (3) this is the last
757 * vnode doing its last close on the device.
759 * XXX the VXLOCK (force close) case can leave vnodes referencing
760 * a closed device. This might not occur now that our revoke is
763 if (dev && ((vp->v_flag & VRECLAIMED) ||
764 (dev_dflags(dev) & D_TRACKCLOSE) ||
765 (vp->v_opencount == 1))) {
767 if (vn_islocked(vp)) {
771 error = dev_dclose(dev, ap->a_fflag, S_IFCHR);
773 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
779 * Track the actual opens and closes on the vnode. The last close
780 * disassociates the rdev. If the rdev is already disassociated or the
781 * opencount is already 0, the vnode might have been revoked and no
782 * further opencount tracking occurs.
786 kprintf("spec_close: %s %d\n",
787 dev->si_name, vp->v_opencount - 1);
789 if (vp->v_opencount == 1)
793 if (vp->v_opencount > 0)
799 * Print out the contents of a special device vnode.
801 * spec_print(struct vnode *a_vp)
804 spec_print(struct vop_print_args *ap)
806 kprintf("tag VT_NON, dev %s\n", devtoname(ap->a_vp->v_rdev));
811 * Special device advisory byte-level locks.
813 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
814 * struct flock *a_fl, int a_flags)
818 spec_advlock(struct vop_advlock_args *ap)
820 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
824 spec_getpages_iodone(struct bio *bio)
826 bio->bio_buf->b_cmd = BUF_CMD_DONE;
827 wakeup(bio->bio_buf);
831 * spec_getpages() - get pages associated with device vnode.
833 * Note that spec_read and spec_write do not use the buffer cache, so we
834 * must fully implement getpages here.
837 spec_getpages(struct vop_getpages_args *ap)
845 int toff, nextoff, nread;
846 struct vnode *vp = ap->a_vp;
851 pcount = round_page(ap->a_count) / PAGE_SIZE;
854 * Calculate the offset of the transfer and do sanity check.
856 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
859 * Round up physical size for real devices. We cannot round using
860 * v_mount's block size data because v_mount has nothing to do with
861 * the device. i.e. it's usually '/dev'. We need the physical block
862 * size for the device itself.
864 * We can't use v_rdev->si_mountpoint because it only exists when the
865 * block device is mounted. However, we can use v_rdev.
868 if (vn_isdisk(vp, NULL))
869 blksiz = vp->v_rdev->si_bsize_phys;
873 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
876 kva = (vm_offset_t)bp->b_data;
879 * Map the pages to be read into the kva.
881 pmap_qenter(kva, ap->a_m, pcount);
883 /* Build a minimal buffer header. */
884 bp->b_cmd = BUF_CMD_READ;
887 bp->b_runningbufspace = size;
889 runningbufspace += bp->b_runningbufspace;
893 bp->b_bio1.bio_offset = offset;
894 bp->b_bio1.bio_done = spec_getpages_iodone;
896 mycpu->gd_cnt.v_vnodein++;
897 mycpu->gd_cnt.v_vnodepgsin += pcount;
900 vn_strategy(ap->a_vp, &bp->b_bio1);
904 /* We definitely need to be at splbio here. */
905 while (bp->b_cmd != BUF_CMD_DONE)
906 tsleep(bp, 0, "spread", 0);
910 if (bp->b_flags & B_ERROR) {
918 * If EOF is encountered we must zero-extend the result in order
919 * to ensure that the page does not contain garabge. When no
920 * error occurs, an early EOF is indicated if b_bcount got truncated.
921 * b_resid is relative to b_bcount and should be 0, but some devices
922 * might indicate an EOF with b_resid instead of truncating b_bcount.
924 nread = bp->b_bcount - bp->b_resid;
925 if (nread < ap->a_count)
926 bzero((caddr_t)kva + nread, ap->a_count - nread);
927 pmap_qremove(kva, pcount);
930 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
931 nextoff = toff + PAGE_SIZE;
934 m->flags &= ~PG_ZERO;
936 if (nextoff <= nread) {
937 m->valid = VM_PAGE_BITS_ALL;
939 } else if (toff < nread) {
941 * Since this is a VM request, we have to supply the
942 * unaligned offset to allow vm_page_set_validclean()
943 * to zero sub-DEV_BSIZE'd portions of the page.
945 vm_page_set_validclean(m, 0, nread - toff);
951 if (i != ap->a_reqpage) {
953 * Just in case someone was asking for this page we
954 * now tell them that it is ok to use.
956 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
958 if (m->flags & PG_WANTED) {
961 vm_page_deactivate(m);
970 } else if (m->valid) {
973 * Since this is a VM request, we need to make the
974 * entire page presentable by zeroing invalid sections.
976 if (m->valid != VM_PAGE_BITS_ALL)
977 vm_page_zero_invalid(m, FALSE);
981 m = ap->a_m[ap->a_reqpage];
983 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
984 devtoname(vp->v_rdev), error, bp, bp->b_vp);
986 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
987 size, bp->b_resid, ap->a_count, m->valid);
989 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
990 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
992 * Free the buffer header back to the swap buffer pool.
995 return VM_PAGER_ERROR;
998 * Free the buffer header back to the swap buffer pool.