proc->thread stage 3.5: Add an IO_CORE flag so coda doesn't have to dig
[dragonfly.git] / sys / vfs / specfs / spec_vnops.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1989, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
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.
20 *
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
31 * SUCH DAMAGE.
32 *
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 $
41c20dac 35 * $DragonFly: src/sys/vfs/specfs/spec_vnops.c,v 1.4 2003/06/23 17:55:44 dillon Exp $
984263bc
MD
36 */
37
38#include <sys/param.h>
39#include <sys/proc.h>
40#include <sys/systm.h>
41#include <sys/kernel.h>
42#include <sys/conf.h>
43#include <sys/buf.h>
44#include <sys/mount.h>
45#include <sys/vnode.h>
46#include <sys/stat.h>
47#include <sys/fcntl.h>
48#include <sys/vmmeter.h>
49#include <sys/tty.h>
50
51#include <vm/vm.h>
52#include <vm/vm_object.h>
53#include <vm/vm_page.h>
54#include <vm/vm_pager.h>
55
3020e3be
MD
56#include <sys/buf2.h>
57
984263bc
MD
58static int spec_advlock __P((struct vop_advlock_args *));
59static int spec_bmap __P((struct vop_bmap_args *));
60static int spec_close __P((struct vop_close_args *));
61static int spec_freeblks __P((struct vop_freeblks_args *));
62static int spec_fsync __P((struct vop_fsync_args *));
63static int spec_getpages __P((struct vop_getpages_args *));
64static int spec_inactive __P((struct vop_inactive_args *));
65static int spec_ioctl __P((struct vop_ioctl_args *));
66static int spec_open __P((struct vop_open_args *));
67static int spec_poll __P((struct vop_poll_args *));
68static int spec_kqfilter __P((struct vop_kqfilter_args *));
69static int spec_print __P((struct vop_print_args *));
70static int spec_read __P((struct vop_read_args *));
71static int spec_strategy __P((struct vop_strategy_args *));
72static int spec_write __P((struct vop_write_args *));
73
74vop_t **spec_vnodeop_p;
75static struct vnodeopv_entry_desc spec_vnodeop_entries[] = {
76 { &vop_default_desc, (vop_t *) vop_defaultop },
77 { &vop_access_desc, (vop_t *) vop_ebadf },
78 { &vop_advlock_desc, (vop_t *) spec_advlock },
79 { &vop_bmap_desc, (vop_t *) spec_bmap },
80 { &vop_close_desc, (vop_t *) spec_close },
81 { &vop_create_desc, (vop_t *) vop_panic },
82 { &vop_freeblks_desc, (vop_t *) spec_freeblks },
83 { &vop_fsync_desc, (vop_t *) spec_fsync },
84 { &vop_getpages_desc, (vop_t *) spec_getpages },
85 { &vop_inactive_desc, (vop_t *) spec_inactive },
86 { &vop_ioctl_desc, (vop_t *) spec_ioctl },
87 { &vop_lease_desc, (vop_t *) vop_null },
88 { &vop_link_desc, (vop_t *) vop_panic },
89 { &vop_mkdir_desc, (vop_t *) vop_panic },
90 { &vop_mknod_desc, (vop_t *) vop_panic },
91 { &vop_open_desc, (vop_t *) spec_open },
92 { &vop_pathconf_desc, (vop_t *) vop_stdpathconf },
93 { &vop_poll_desc, (vop_t *) spec_poll },
94 { &vop_kqfilter_desc, (vop_t *) spec_kqfilter },
95 { &vop_print_desc, (vop_t *) spec_print },
96 { &vop_read_desc, (vop_t *) spec_read },
97 { &vop_readdir_desc, (vop_t *) vop_panic },
98 { &vop_readlink_desc, (vop_t *) vop_panic },
99 { &vop_reallocblks_desc, (vop_t *) vop_panic },
100 { &vop_reclaim_desc, (vop_t *) vop_null },
101 { &vop_remove_desc, (vop_t *) vop_panic },
102 { &vop_rename_desc, (vop_t *) vop_panic },
103 { &vop_rmdir_desc, (vop_t *) vop_panic },
104 { &vop_setattr_desc, (vop_t *) vop_ebadf },
105 { &vop_strategy_desc, (vop_t *) spec_strategy },
106 { &vop_symlink_desc, (vop_t *) vop_panic },
107 { &vop_write_desc, (vop_t *) spec_write },
108 { NULL, NULL }
109};
110static struct vnodeopv_desc spec_vnodeop_opv_desc =
111 { &spec_vnodeop_p, spec_vnodeop_entries };
112
113VNODEOP_SET(spec_vnodeop_opv_desc);
114
115int
116spec_vnoperate(ap)
117 struct vop_generic_args /* {
118 struct vnodeop_desc *a_desc;
119 <other random data follows, presumably>
120 } */ *ap;
121{
122 return (VOCALL(spec_vnodeop_p, ap->a_desc->vdesc_offset, ap));
123}
124
125static void spec_getpages_iodone __P((struct buf *bp));
126
127/*
128 * Open a special file.
129 */
130/* ARGSUSED */
131static int
132spec_open(ap)
133 struct vop_open_args /* {
134 struct vnode *a_vp;
135 int a_mode;
136 struct ucred *a_cred;
137 struct proc *a_p;
138 } */ *ap;
139{
140 struct proc *p = ap->a_p;
141 struct vnode *vp = ap->a_vp;
142 dev_t dev = vp->v_rdev;
143 int error;
144 struct cdevsw *dsw;
145 const char *cp;
146
147 /*
148 * Don't allow open if fs is mounted -nodev.
149 */
150 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
151 return (ENXIO);
152
153 dsw = devsw(dev);
154 if ( (dsw == NULL) || (dsw->d_open == NULL))
155 return ENXIO;
156
157 /* Make this field valid before any I/O in ->d_open */
158 if (!dev->si_iosize_max)
159 dev->si_iosize_max = DFLTPHYS;
160
161 /*
162 * XXX: Disks get special billing here, but it is mostly wrong.
163 * XXX: diskpartitions can overlap and the real checks should
164 * XXX: take this into account, and consequently they need to
165 * XXX: live in the diskslicing code. Some checks do.
166 */
167 if (vn_isdisk(vp, NULL) && ap->a_cred != FSCRED &&
168 (ap->a_mode & FWRITE)) {
169 /*
170 * Never allow opens for write if the device is mounted R/W
171 */
172 if (vp->v_specmountpoint != NULL &&
173 !(vp->v_specmountpoint->mnt_flag & MNT_RDONLY))
174 return (EBUSY);
175
176 /*
177 * When running in secure mode, do not allow opens
178 * for writing if the device is mounted
179 */
180 if (securelevel >= 1 && vp->v_specmountpoint != NULL)
181 return (EPERM);
182
183 /*
184 * When running in very secure mode, do not allow
185 * opens for writing of any devices.
186 */
187 if (securelevel >= 2)
188 return (EPERM);
189 }
190
191 /* XXX: Special casing of ttys for deadfs. Probably redundant */
192 if (dsw->d_flags & D_TTY)
193 vp->v_flag |= VISTTY;
194
195 VOP_UNLOCK(vp, 0, p);
41c20dac 196 error = (*dsw->d_open)(dev, ap->a_mode, S_IFCHR, p->p_thread);
984263bc
MD
197 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
198
199 if (error)
200 return (error);
201
202 if (dsw->d_flags & D_TTY) {
203 if (dev->si_tty) {
204 struct tty *tp;
205 tp = dev->si_tty;
206 if (!tp->t_stop) {
207 printf("Warning:%s: no t_stop, using nottystop\n", devtoname(dev));
208 tp->t_stop = nottystop;
209 }
210 }
211 }
212
213 if (vn_isdisk(vp, NULL)) {
214 if (!dev->si_bsize_phys)
215 dev->si_bsize_phys = DEV_BSIZE;
216 }
217 if ((dsw->d_flags & D_DISK) == 0) {
218 cp = devtoname(dev);
219 if (*cp == '#' && (dsw->d_flags & D_NAGGED) == 0) {
220 printf("WARNING: driver %s should register devices with make_dev() (dev_t = \"%s\")\n",
221 dsw->d_name, cp);
222 dsw->d_flags |= D_NAGGED;
223 }
224 }
225 return (error);
226}
227
228/*
229 * Vnode op for read
230 */
231/* ARGSUSED */
232static int
233spec_read(ap)
234 struct vop_read_args /* {
235 struct vnode *a_vp;
236 struct uio *a_uio;
237 int a_ioflag;
238 struct ucred *a_cred;
239 } */ *ap;
240{
241 struct vnode *vp;
242 struct proc *p;
243 struct uio *uio;
244 dev_t dev;
245 int error;
246
247 vp = ap->a_vp;
248 dev = vp->v_rdev;
249 uio = ap->a_uio;
250 p = uio->uio_procp;
251
252 if (uio->uio_resid == 0)
253 return (0);
254
255 VOP_UNLOCK(vp, 0, p);
256 error = (*devsw(dev)->d_read) (dev, uio, ap->a_ioflag);
257 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
258 return (error);
259}
260
261/*
262 * Vnode op for write
263 */
264/* ARGSUSED */
265static int
266spec_write(ap)
267 struct vop_write_args /* {
268 struct vnode *a_vp;
269 struct uio *a_uio;
270 int a_ioflag;
271 struct ucred *a_cred;
272 } */ *ap;
273{
274 struct vnode *vp;
275 struct proc *p;
276 struct uio *uio;
277 dev_t dev;
278 int error;
279
280 vp = ap->a_vp;
281 dev = vp->v_rdev;
282 uio = ap->a_uio;
283 p = uio->uio_procp;
284
285 VOP_UNLOCK(vp, 0, p);
286 error = (*devsw(dev)->d_write) (dev, uio, ap->a_ioflag);
287 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
288 return (error);
289}
290
291/*
292 * Device ioctl operation.
293 */
294/* ARGSUSED */
295static int
296spec_ioctl(ap)
297 struct vop_ioctl_args /* {
298 struct vnode *a_vp;
299 int a_command;
300 caddr_t a_data;
301 int a_fflag;
302 struct ucred *a_cred;
303 struct proc *a_p;
304 } */ *ap;
305{
306 dev_t dev;
307
308 dev = ap->a_vp->v_rdev;
309 return ((*devsw(dev)->d_ioctl)(dev, ap->a_command,
41c20dac 310 ap->a_data, ap->a_fflag, ap->a_p->p_thread));
984263bc
MD
311}
312
313/* ARGSUSED */
314static int
315spec_poll(ap)
316 struct vop_poll_args /* {
317 struct vnode *a_vp;
318 int a_events;
319 struct ucred *a_cred;
320 struct proc *a_p;
321 } */ *ap;
322{
323 dev_t dev;
324
325 dev = ap->a_vp->v_rdev;
41c20dac 326 return (*devsw(dev)->d_poll)(dev, ap->a_events, ap->a_p->p_thread);
984263bc
MD
327}
328
329/* ARGSUSED */
330static int
331spec_kqfilter(ap)
332 struct vop_kqfilter_args /* {
333 struct vnode *a_vp;
334 struct knote *a_kn;
335 } */ *ap;
336{
337 dev_t dev;
338
339 dev = ap->a_vp->v_rdev;
340 if (devsw(dev)->d_flags & D_KQFILTER)
341 return (*devsw(dev)->d_kqfilter)(dev, ap->a_kn);
342 return (1);
343}
344
345/*
346 * Synch buffers associated with a block device
347 */
348/* ARGSUSED */
349static int
350spec_fsync(ap)
351 struct vop_fsync_args /* {
352 struct vnode *a_vp;
353 struct ucred *a_cred;
354 int a_waitfor;
355 struct proc *a_p;
356 } */ *ap;
357{
358 struct vnode *vp = ap->a_vp;
359 struct buf *bp;
360 struct buf *nbp;
361 int s;
362 int maxretry = 10000; /* large, arbitrarily chosen */
363
364 if (!vn_isdisk(vp, NULL))
365 return (0);
366
367loop1:
368 /*
369 * MARK/SCAN initialization to avoid infinite loops
370 */
371 s = splbio();
372 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp;
373 bp = TAILQ_NEXT(bp, b_vnbufs)) {
374 bp->b_flags &= ~B_SCANNED;
375 }
376 splx(s);
377
378 /*
379 * Flush all dirty buffers associated with a block device.
380 */
381loop2:
382 s = splbio();
383 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
384 nbp = TAILQ_NEXT(bp, b_vnbufs);
385 if ((bp->b_flags & B_SCANNED) != 0)
386 continue;
387 bp->b_flags |= B_SCANNED;
388 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
389 continue;
390 if ((bp->b_flags & B_DELWRI) == 0)
391 panic("spec_fsync: not dirty");
392 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
393 BUF_UNLOCK(bp);
394 vfs_bio_awrite(bp);
395 splx(s);
396 } else {
397 bremfree(bp);
398 splx(s);
399 bawrite(bp);
400 }
401 goto loop2;
402 }
403
404 /*
405 * If synchronous the caller expects us to completely resolve all
406 * dirty buffers in the system. Wait for in-progress I/O to
407 * complete (which could include background bitmap writes), then
408 * retry if dirty blocks still exist.
409 */
410 if (ap->a_waitfor == MNT_WAIT) {
411 while (vp->v_numoutput) {
412 vp->v_flag |= VBWAIT;
413 (void) tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "spfsyn", 0);
414 }
415 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
416 if (--maxretry != 0) {
417 splx(s);
418 goto loop1;
419 }
420 vprint("spec_fsync: giving up on dirty", vp);
421 }
422 }
423 splx(s);
424 return (0);
425}
426
427static int
428spec_inactive(ap)
429 struct vop_inactive_args /* {
430 struct vnode *a_vp;
431 struct proc *a_p;
432 } */ *ap;
433{
434
435 VOP_UNLOCK(ap->a_vp, 0, ap->a_p);
436 return (0);
437}
438
439/*
440 * Just call the device strategy routine
441 */
442static int
443spec_strategy(ap)
444 struct vop_strategy_args /* {
445 struct vnode *a_vp;
446 struct buf *a_bp;
447 } */ *ap;
448{
449 struct buf *bp;
450 struct vnode *vp;
451 struct mount *mp;
452
453 bp = ap->a_bp;
454 if (((bp->b_flags & B_READ) == 0) &&
455 (LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
456 (*bioops.io_start)(bp);
457
458 /*
459 * Collect statistics on synchronous and asynchronous read
460 * and write counts for disks that have associated filesystems.
461 */
462 vp = ap->a_vp;
463 if (vn_isdisk(vp, NULL) && (mp = vp->v_specmountpoint) != NULL) {
464 if ((bp->b_flags & B_READ) == 0) {
465 if (bp->b_lock.lk_lockholder == LK_KERNPROC)
466 mp->mnt_stat.f_asyncwrites++;
467 else
468 mp->mnt_stat.f_syncwrites++;
469 } else {
470 if (bp->b_lock.lk_lockholder == LK_KERNPROC)
471 mp->mnt_stat.f_asyncreads++;
472 else
473 mp->mnt_stat.f_syncreads++;
474 }
475 }
476 KASSERT(devsw(bp->b_dev) != NULL,
477 ("No devsw on dev %s responsible for buffer %p\n",
478 devtoname(bp->b_dev), bp));
479 KASSERT(devsw(bp->b_dev)->d_strategy != NULL,
480 ("No strategy on dev %s responsible for buffer %p\n",
481 devtoname(bp->b_dev), bp));
482 BUF_STRATEGY(bp, 0);
483 return (0);
484}
485
486static int
487spec_freeblks(ap)
488 struct vop_freeblks_args /* {
489 struct vnode *a_vp;
490 daddr_t a_addr;
491 daddr_t a_length;
492 } */ *ap;
493{
494 struct cdevsw *bsw;
495 struct buf *bp;
496
497 /*
498 * XXX: This assumes that strategy does the deed right away.
499 * XXX: this may not be TRTTD.
500 */
501 bsw = devsw(ap->a_vp->v_rdev);
502 if ((bsw->d_flags & D_CANFREE) == 0)
503 return (0);
504 bp = geteblk(ap->a_length);
505 bp->b_flags |= B_FREEBUF;
506 bp->b_dev = ap->a_vp->v_rdev;
507 bp->b_blkno = ap->a_addr;
508 bp->b_offset = dbtob(ap->a_addr);
509 bp->b_bcount = ap->a_length;
510 BUF_STRATEGY(bp, 0);
511 return (0);
512}
513
514/*
515 * Implement degenerate case where the block requested is the block
516 * returned, and assume that the entire device is contiguous in regards
517 * to the contiguous block range (runp and runb).
518 */
519static int
520spec_bmap(ap)
521 struct vop_bmap_args /* {
522 struct vnode *a_vp;
523 daddr_t a_bn;
524 struct vnode **a_vpp;
525 daddr_t *a_bnp;
526 int *a_runp;
527 int *a_runb;
528 } */ *ap;
529{
530 struct vnode *vp = ap->a_vp;
531 int runp = 0;
532 int runb = 0;
533
534 if (ap->a_vpp != NULL)
535 *ap->a_vpp = vp;
536 if (ap->a_bnp != NULL)
537 *ap->a_bnp = ap->a_bn;
538 if (vp->v_mount != NULL)
539 runp = runb = MAXBSIZE / vp->v_mount->mnt_stat.f_iosize;
540 if (ap->a_runp != NULL)
541 *ap->a_runp = runp;
542 if (ap->a_runb != NULL)
543 *ap->a_runb = runb;
544 return (0);
545}
546
547/*
548 * Device close routine
549 */
550/* ARGSUSED */
551static int
552spec_close(ap)
553 struct vop_close_args /* {
554 struct vnode *a_vp;
555 int a_fflag;
556 struct ucred *a_cred;
557 struct proc *a_p;
558 } */ *ap;
559{
560 struct vnode *vp = ap->a_vp;
561 struct proc *p = ap->a_p;
562 dev_t dev = vp->v_rdev;
563
564 /*
565 * Hack: a tty device that is a controlling terminal
566 * has a reference from the session structure.
567 * We cannot easily tell that a character device is
568 * a controlling terminal, unless it is the closing
569 * process' controlling terminal. In that case,
570 * if the reference count is 2 (this last descriptor
571 * plus the session), release the reference from the session.
572 */
573 if (vcount(vp) == 2 && p && (vp->v_flag & VXLOCK) == 0 &&
574 vp == p->p_session->s_ttyvp) {
575 vrele(vp);
576 p->p_session->s_ttyvp = NULL;
577 }
578 /*
579 * We do not want to really close the device if it
580 * is still in use unless we are trying to close it
581 * forcibly. Since every use (buffer, vnode, swap, cmap)
582 * holds a reference to the vnode, and because we mark
583 * any other vnodes that alias this device, when the
584 * sum of the reference counts on all the aliased
585 * vnodes descends to one, we are on last close.
586 */
587 if (vp->v_flag & VXLOCK) {
588 /* Forced close */
589 } else if (devsw(dev)->d_flags & D_TRACKCLOSE) {
590 /* Keep device updated on status */
591 } else if (vcount(vp) > 1) {
592 return (0);
593 }
41c20dac 594 return (devsw(dev)->d_close(dev, ap->a_fflag, S_IFCHR, p->p_thread));
984263bc
MD
595}
596
597/*
598 * Print out the contents of a special device vnode.
599 */
600static int
601spec_print(ap)
602 struct vop_print_args /* {
603 struct vnode *a_vp;
604 } */ *ap;
605{
606
607 printf("tag VT_NON, dev %s\n", devtoname(ap->a_vp->v_rdev));
608 return (0);
609}
610
611/*
612 * Special device advisory byte-level locks.
613 */
614/* ARGSUSED */
615static int
616spec_advlock(ap)
617 struct vop_advlock_args /* {
618 struct vnode *a_vp;
619 caddr_t a_id;
620 int a_op;
621 struct flock *a_fl;
622 int a_flags;
623 } */ *ap;
624{
625
626 return (ap->a_flags & F_FLOCK ? EOPNOTSUPP : EINVAL);
627}
628
629static void
630spec_getpages_iodone(bp)
631 struct buf *bp;
632{
633
634 bp->b_flags |= B_DONE;
635 wakeup(bp);
636}
637
638static int
639spec_getpages(ap)
640 struct vop_getpages_args *ap;
641{
642 vm_offset_t kva;
643 int error;
644 int i, pcount, size, s;
645 daddr_t blkno;
646 struct buf *bp;
647 vm_page_t m;
648 vm_ooffset_t offset;
649 int toff, nextoff, nread;
650 struct vnode *vp = ap->a_vp;
651 int blksiz;
652 int gotreqpage;
653
654 error = 0;
655 pcount = round_page(ap->a_count) / PAGE_SIZE;
656
657 /*
658 * Calculate the offset of the transfer and do sanity check.
659 * FreeBSD currently only supports an 8 TB range due to b_blkno
660 * being in DEV_BSIZE ( usually 512 ) byte chunks on call to
661 * VOP_STRATEGY. XXX
662 */
663 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
664
665#define DADDR_T_BIT (sizeof(daddr_t)*8)
666#define OFFSET_MAX ((1LL << (DADDR_T_BIT + DEV_BSHIFT)) - 1)
667
668 if (offset < 0 || offset > OFFSET_MAX) {
669 /* XXX still no %q in kernel. */
670 printf("spec_getpages: preposterous offset 0x%x%08x\n",
671 (u_int)((u_quad_t)offset >> 32),
672 (u_int)(offset & 0xffffffff));
673 return (VM_PAGER_ERROR);
674 }
675
676 blkno = btodb(offset);
677
678 /*
679 * Round up physical size for real devices. We cannot round using
680 * v_mount's block size data because v_mount has nothing to do with
681 * the device. i.e. it's usually '/dev'. We need the physical block
682 * size for the device itself.
683 *
684 * We can't use v_specmountpoint because it only exists when the
685 * block device is mounted. However, we can use v_rdev.
686 */
687
688 if (vn_isdisk(vp, NULL))
689 blksiz = vp->v_rdev->si_bsize_phys;
690 else
691 blksiz = DEV_BSIZE;
692
693 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
694
695 bp = getpbuf(NULL);
696 kva = (vm_offset_t)bp->b_data;
697
698 /*
699 * Map the pages to be read into the kva.
700 */
701 pmap_qenter(kva, ap->a_m, pcount);
702
703 /* Build a minimal buffer header. */
704 bp->b_flags = B_READ | B_CALL;
705 bp->b_iodone = spec_getpages_iodone;
706
707 /* B_PHYS is not set, but it is nice to fill this in. */
708 bp->b_rcred = bp->b_wcred = curproc->p_ucred;
709 if (bp->b_rcred != NOCRED)
710 crhold(bp->b_rcred);
711 if (bp->b_wcred != NOCRED)
712 crhold(bp->b_wcred);
713 bp->b_blkno = blkno;
714 bp->b_lblkno = blkno;
715 pbgetvp(ap->a_vp, bp);
716 bp->b_bcount = size;
717 bp->b_bufsize = size;
718 bp->b_resid = 0;
719 bp->b_runningbufspace = bp->b_bufsize;
720 runningbufspace += bp->b_runningbufspace;
721
722 cnt.v_vnodein++;
723 cnt.v_vnodepgsin += pcount;
724
725 /* Do the input. */
726 VOP_STRATEGY(bp->b_vp, bp);
727
728 s = splbio();
729
730 /* We definitely need to be at splbio here. */
731 while ((bp->b_flags & B_DONE) == 0)
732 tsleep(bp, PVM, "spread", 0);
733
734 splx(s);
735
736 if ((bp->b_flags & B_ERROR) != 0) {
737 if (bp->b_error)
738 error = bp->b_error;
739 else
740 error = EIO;
741 }
742
743 nread = size - bp->b_resid;
744
745 if (nread < ap->a_count) {
746 bzero((caddr_t)kva + nread,
747 ap->a_count - nread);
748 }
749 pmap_qremove(kva, pcount);
750
751
752 gotreqpage = 0;
753 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
754 nextoff = toff + PAGE_SIZE;
755 m = ap->a_m[i];
756
757 m->flags &= ~PG_ZERO;
758
759 if (nextoff <= nread) {
760 m->valid = VM_PAGE_BITS_ALL;
761 vm_page_undirty(m);
762 } else if (toff < nread) {
763 /*
764 * Since this is a VM request, we have to supply the
765 * unaligned offset to allow vm_page_set_validclean()
766 * to zero sub-DEV_BSIZE'd portions of the page.
767 */
768 vm_page_set_validclean(m, 0, nread - toff);
769 } else {
770 m->valid = 0;
771 vm_page_undirty(m);
772 }
773
774 if (i != ap->a_reqpage) {
775 /*
776 * Just in case someone was asking for this page we
777 * now tell them that it is ok to use.
778 */
779 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
780 if (m->valid) {
781 if (m->flags & PG_WANTED) {
782 vm_page_activate(m);
783 } else {
784 vm_page_deactivate(m);
785 }
786 vm_page_wakeup(m);
787 } else {
788 vm_page_free(m);
789 }
790 } else {
791 vm_page_free(m);
792 }
793 } else if (m->valid) {
794 gotreqpage = 1;
795 /*
796 * Since this is a VM request, we need to make the
797 * entire page presentable by zeroing invalid sections.
798 */
799 if (m->valid != VM_PAGE_BITS_ALL)
800 vm_page_zero_invalid(m, FALSE);
801 }
802 }
803 if (!gotreqpage) {
804 m = ap->a_m[ap->a_reqpage];
805 printf(
806 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
807 devtoname(bp->b_dev), error, bp, bp->b_vp);
808 printf(
809 " size: %d, resid: %ld, a_count: %d, valid: 0x%x\n",
810 size, bp->b_resid, ap->a_count, m->valid);
811 printf(
812 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
813 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
814 /*
815 * Free the buffer header back to the swap buffer pool.
816 */
817 relpbuf(bp, NULL);
818 return VM_PAGER_ERROR;
819 }
820 /*
821 * Free the buffer header back to the swap buffer pool.
822 */
823 relpbuf(bp, NULL);
824 return VM_PAGER_OK;
825}