proc->thread stage 4: rework the VFS and DEVICE subsystems to take thread
[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 $
dadab5e9 35 * $DragonFly: src/sys/vfs/specfs/spec_vnops.c,v 1.5 2003/06/25 03:56:00 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;
dadab5e9 137 struct thread *a_td;
984263bc
MD
138 } */ *ap;
139{
984263bc
MD
140 struct vnode *vp = ap->a_vp;
141 dev_t dev = vp->v_rdev;
142 int error;
143 struct cdevsw *dsw;
144 const char *cp;
145
146 /*
147 * Don't allow open if fs is mounted -nodev.
148 */
149 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_NODEV))
150 return (ENXIO);
151
152 dsw = devsw(dev);
153 if ( (dsw == NULL) || (dsw->d_open == NULL))
154 return ENXIO;
155
156 /* Make this field valid before any I/O in ->d_open */
157 if (!dev->si_iosize_max)
158 dev->si_iosize_max = DFLTPHYS;
159
160 /*
161 * XXX: Disks get special billing here, but it is mostly wrong.
162 * XXX: diskpartitions can overlap and the real checks should
163 * XXX: take this into account, and consequently they need to
164 * XXX: live in the diskslicing code. Some checks do.
165 */
166 if (vn_isdisk(vp, NULL) && ap->a_cred != FSCRED &&
167 (ap->a_mode & FWRITE)) {
168 /*
169 * Never allow opens for write if the device is mounted R/W
170 */
171 if (vp->v_specmountpoint != NULL &&
172 !(vp->v_specmountpoint->mnt_flag & MNT_RDONLY))
173 return (EBUSY);
174
175 /*
176 * When running in secure mode, do not allow opens
177 * for writing if the device is mounted
178 */
179 if (securelevel >= 1 && vp->v_specmountpoint != NULL)
180 return (EPERM);
181
182 /*
183 * When running in very secure mode, do not allow
184 * opens for writing of any devices.
185 */
186 if (securelevel >= 2)
187 return (EPERM);
188 }
189
190 /* XXX: Special casing of ttys for deadfs. Probably redundant */
191 if (dsw->d_flags & D_TTY)
192 vp->v_flag |= VISTTY;
193
dadab5e9
MD
194 VOP_UNLOCK(vp, 0, ap->a_td);
195 error = (*dsw->d_open)(dev, ap->a_mode, S_IFCHR, ap->a_td);
196 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
984263bc
MD
197
198 if (error)
199 return (error);
200
201 if (dsw->d_flags & D_TTY) {
202 if (dev->si_tty) {
203 struct tty *tp;
204 tp = dev->si_tty;
205 if (!tp->t_stop) {
206 printf("Warning:%s: no t_stop, using nottystop\n", devtoname(dev));
207 tp->t_stop = nottystop;
208 }
209 }
210 }
211
212 if (vn_isdisk(vp, NULL)) {
213 if (!dev->si_bsize_phys)
214 dev->si_bsize_phys = DEV_BSIZE;
215 }
216 if ((dsw->d_flags & D_DISK) == 0) {
217 cp = devtoname(dev);
218 if (*cp == '#' && (dsw->d_flags & D_NAGGED) == 0) {
219 printf("WARNING: driver %s should register devices with make_dev() (dev_t = \"%s\")\n",
220 dsw->d_name, cp);
221 dsw->d_flags |= D_NAGGED;
222 }
223 }
224 return (error);
225}
226
227/*
228 * Vnode op for read
229 */
230/* ARGSUSED */
231static int
232spec_read(ap)
233 struct vop_read_args /* {
234 struct vnode *a_vp;
235 struct uio *a_uio;
236 int a_ioflag;
237 struct ucred *a_cred;
238 } */ *ap;
239{
240 struct vnode *vp;
dadab5e9 241 struct thread *td;
984263bc
MD
242 struct uio *uio;
243 dev_t dev;
244 int error;
245
246 vp = ap->a_vp;
247 dev = vp->v_rdev;
248 uio = ap->a_uio;
dadab5e9 249 td = uio->uio_td;
984263bc
MD
250
251 if (uio->uio_resid == 0)
252 return (0);
253
dadab5e9 254 VOP_UNLOCK(vp, 0, td);
984263bc 255 error = (*devsw(dev)->d_read) (dev, uio, ap->a_ioflag);
dadab5e9 256 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
984263bc
MD
257 return (error);
258}
259
260/*
261 * Vnode op for write
262 */
263/* ARGSUSED */
264static int
265spec_write(ap)
266 struct vop_write_args /* {
267 struct vnode *a_vp;
268 struct uio *a_uio;
269 int a_ioflag;
270 struct ucred *a_cred;
271 } */ *ap;
272{
273 struct vnode *vp;
dadab5e9 274 struct thread *td;
984263bc
MD
275 struct uio *uio;
276 dev_t dev;
277 int error;
278
279 vp = ap->a_vp;
280 dev = vp->v_rdev;
281 uio = ap->a_uio;
dadab5e9 282 td = uio->uio_td;
984263bc 283
dadab5e9 284 VOP_UNLOCK(vp, 0, td);
984263bc 285 error = (*devsw(dev)->d_write) (dev, uio, ap->a_ioflag);
dadab5e9 286 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
984263bc
MD
287 return (error);
288}
289
290/*
291 * Device ioctl operation.
292 */
293/* ARGSUSED */
294static int
295spec_ioctl(ap)
296 struct vop_ioctl_args /* {
297 struct vnode *a_vp;
298 int a_command;
299 caddr_t a_data;
300 int a_fflag;
301 struct ucred *a_cred;
dadab5e9 302 struct thread *a_td;
984263bc
MD
303 } */ *ap;
304{
305 dev_t dev;
306
307 dev = ap->a_vp->v_rdev;
308 return ((*devsw(dev)->d_ioctl)(dev, ap->a_command,
dadab5e9 309 ap->a_data, ap->a_fflag, ap->a_td));
984263bc
MD
310}
311
312/* ARGSUSED */
313static int
314spec_poll(ap)
315 struct vop_poll_args /* {
316 struct vnode *a_vp;
317 int a_events;
318 struct ucred *a_cred;
dadab5e9 319 struct thread *a_td;
984263bc
MD
320 } */ *ap;
321{
322 dev_t dev;
323
324 dev = ap->a_vp->v_rdev;
dadab5e9 325 return (*devsw(dev)->d_poll)(dev, ap->a_events, ap->a_td);
984263bc
MD
326}
327
328/* ARGSUSED */
329static int
330spec_kqfilter(ap)
331 struct vop_kqfilter_args /* {
332 struct vnode *a_vp;
333 struct knote *a_kn;
334 } */ *ap;
335{
336 dev_t dev;
337
338 dev = ap->a_vp->v_rdev;
339 if (devsw(dev)->d_flags & D_KQFILTER)
340 return (*devsw(dev)->d_kqfilter)(dev, ap->a_kn);
341 return (1);
342}
343
344/*
345 * Synch buffers associated with a block device
346 */
347/* ARGSUSED */
348static int
349spec_fsync(ap)
350 struct vop_fsync_args /* {
351 struct vnode *a_vp;
352 struct ucred *a_cred;
353 int a_waitfor;
dadab5e9 354 struct thread *a_td;
984263bc
MD
355 } */ *ap;
356{
357 struct vnode *vp = ap->a_vp;
358 struct buf *bp;
359 struct buf *nbp;
360 int s;
361 int maxretry = 10000; /* large, arbitrarily chosen */
362
363 if (!vn_isdisk(vp, NULL))
364 return (0);
365
366loop1:
367 /*
368 * MARK/SCAN initialization to avoid infinite loops
369 */
370 s = splbio();
371 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp;
372 bp = TAILQ_NEXT(bp, b_vnbufs)) {
373 bp->b_flags &= ~B_SCANNED;
374 }
375 splx(s);
376
377 /*
378 * Flush all dirty buffers associated with a block device.
379 */
380loop2:
381 s = splbio();
382 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
383 nbp = TAILQ_NEXT(bp, b_vnbufs);
384 if ((bp->b_flags & B_SCANNED) != 0)
385 continue;
386 bp->b_flags |= B_SCANNED;
387 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT))
388 continue;
389 if ((bp->b_flags & B_DELWRI) == 0)
390 panic("spec_fsync: not dirty");
391 if ((vp->v_flag & VOBJBUF) && (bp->b_flags & B_CLUSTEROK)) {
392 BUF_UNLOCK(bp);
393 vfs_bio_awrite(bp);
394 splx(s);
395 } else {
396 bremfree(bp);
397 splx(s);
398 bawrite(bp);
399 }
400 goto loop2;
401 }
402
403 /*
404 * If synchronous the caller expects us to completely resolve all
405 * dirty buffers in the system. Wait for in-progress I/O to
406 * complete (which could include background bitmap writes), then
407 * retry if dirty blocks still exist.
408 */
409 if (ap->a_waitfor == MNT_WAIT) {
410 while (vp->v_numoutput) {
411 vp->v_flag |= VBWAIT;
412 (void) tsleep((caddr_t)&vp->v_numoutput, PRIBIO + 1, "spfsyn", 0);
413 }
414 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
415 if (--maxretry != 0) {
416 splx(s);
417 goto loop1;
418 }
419 vprint("spec_fsync: giving up on dirty", vp);
420 }
421 }
422 splx(s);
423 return (0);
424}
425
426static int
427spec_inactive(ap)
428 struct vop_inactive_args /* {
429 struct vnode *a_vp;
dadab5e9 430 struct thread *a_td;
984263bc
MD
431 } */ *ap;
432{
433
dadab5e9 434 VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
984263bc
MD
435 return (0);
436}
437
438/*
439 * Just call the device strategy routine
440 */
441static int
442spec_strategy(ap)
443 struct vop_strategy_args /* {
444 struct vnode *a_vp;
445 struct buf *a_bp;
446 } */ *ap;
447{
448 struct buf *bp;
449 struct vnode *vp;
450 struct mount *mp;
451
452 bp = ap->a_bp;
453 if (((bp->b_flags & B_READ) == 0) &&
454 (LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
455 (*bioops.io_start)(bp);
456
457 /*
458 * Collect statistics on synchronous and asynchronous read
459 * and write counts for disks that have associated filesystems.
460 */
461 vp = ap->a_vp;
462 if (vn_isdisk(vp, NULL) && (mp = vp->v_specmountpoint) != NULL) {
463 if ((bp->b_flags & B_READ) == 0) {
dadab5e9 464 if (bp->b_lock.lk_lockholder == LK_KERNTHREAD)
984263bc
MD
465 mp->mnt_stat.f_asyncwrites++;
466 else
467 mp->mnt_stat.f_syncwrites++;
468 } else {
dadab5e9 469 if (bp->b_lock.lk_lockholder == LK_KERNTHREAD)
984263bc
MD
470 mp->mnt_stat.f_asyncreads++;
471 else
472 mp->mnt_stat.f_syncreads++;
473 }
474 }
475 KASSERT(devsw(bp->b_dev) != NULL,
476 ("No devsw on dev %s responsible for buffer %p\n",
477 devtoname(bp->b_dev), bp));
478 KASSERT(devsw(bp->b_dev)->d_strategy != NULL,
479 ("No strategy on dev %s responsible for buffer %p\n",
480 devtoname(bp->b_dev), bp));
481 BUF_STRATEGY(bp, 0);
482 return (0);
483}
484
485static int
486spec_freeblks(ap)
487 struct vop_freeblks_args /* {
488 struct vnode *a_vp;
489 daddr_t a_addr;
490 daddr_t a_length;
491 } */ *ap;
492{
493 struct cdevsw *bsw;
494 struct buf *bp;
495
496 /*
497 * XXX: This assumes that strategy does the deed right away.
498 * XXX: this may not be TRTTD.
499 */
500 bsw = devsw(ap->a_vp->v_rdev);
501 if ((bsw->d_flags & D_CANFREE) == 0)
502 return (0);
503 bp = geteblk(ap->a_length);
504 bp->b_flags |= B_FREEBUF;
505 bp->b_dev = ap->a_vp->v_rdev;
506 bp->b_blkno = ap->a_addr;
507 bp->b_offset = dbtob(ap->a_addr);
508 bp->b_bcount = ap->a_length;
509 BUF_STRATEGY(bp, 0);
510 return (0);
511}
512
513/*
514 * Implement degenerate case where the block requested is the block
515 * returned, and assume that the entire device is contiguous in regards
516 * to the contiguous block range (runp and runb).
517 */
518static int
519spec_bmap(ap)
520 struct vop_bmap_args /* {
521 struct vnode *a_vp;
522 daddr_t a_bn;
523 struct vnode **a_vpp;
524 daddr_t *a_bnp;
525 int *a_runp;
526 int *a_runb;
527 } */ *ap;
528{
529 struct vnode *vp = ap->a_vp;
530 int runp = 0;
531 int runb = 0;
532
533 if (ap->a_vpp != NULL)
534 *ap->a_vpp = vp;
535 if (ap->a_bnp != NULL)
536 *ap->a_bnp = ap->a_bn;
537 if (vp->v_mount != NULL)
538 runp = runb = MAXBSIZE / vp->v_mount->mnt_stat.f_iosize;
539 if (ap->a_runp != NULL)
540 *ap->a_runp = runp;
541 if (ap->a_runb != NULL)
542 *ap->a_runb = runb;
543 return (0);
544}
545
546/*
547 * Device close routine
548 */
549/* ARGSUSED */
550static int
551spec_close(ap)
552 struct vop_close_args /* {
553 struct vnode *a_vp;
554 int a_fflag;
555 struct ucred *a_cred;
dadab5e9 556 struct thread *a_td;
984263bc
MD
557 } */ *ap;
558{
dadab5e9 559 struct proc *p = ap->a_td->td_proc;
984263bc 560 struct vnode *vp = ap->a_vp;
984263bc
MD
561 dev_t dev = vp->v_rdev;
562
563 /*
564 * Hack: a tty device that is a controlling terminal
565 * has a reference from the session structure.
566 * We cannot easily tell that a character device is
567 * a controlling terminal, unless it is the closing
568 * process' controlling terminal. In that case,
569 * if the reference count is 2 (this last descriptor
570 * plus the session), release the reference from the session.
571 */
572 if (vcount(vp) == 2 && p && (vp->v_flag & VXLOCK) == 0 &&
573 vp == p->p_session->s_ttyvp) {
574 vrele(vp);
575 p->p_session->s_ttyvp = NULL;
576 }
577 /*
578 * We do not want to really close the device if it
579 * is still in use unless we are trying to close it
580 * forcibly. Since every use (buffer, vnode, swap, cmap)
581 * holds a reference to the vnode, and because we mark
582 * any other vnodes that alias this device, when the
583 * sum of the reference counts on all the aliased
584 * vnodes descends to one, we are on last close.
585 */
586 if (vp->v_flag & VXLOCK) {
587 /* Forced close */
588 } else if (devsw(dev)->d_flags & D_TRACKCLOSE) {
589 /* Keep device updated on status */
590 } else if (vcount(vp) > 1) {
591 return (0);
592 }
dadab5e9 593 return (devsw(dev)->d_close(dev, ap->a_fflag, S_IFCHR, ap->a_td));
984263bc
MD
594}
595
596/*
597 * Print out the contents of a special device vnode.
598 */
599static int
600spec_print(ap)
601 struct vop_print_args /* {
602 struct vnode *a_vp;
603 } */ *ap;
604{
605
606 printf("tag VT_NON, dev %s\n", devtoname(ap->a_vp->v_rdev));
607 return (0);
608}
609
610/*
611 * Special device advisory byte-level locks.
612 */
613/* ARGSUSED */
614static int
615spec_advlock(ap)
616 struct vop_advlock_args /* {
617 struct vnode *a_vp;
618 caddr_t a_id;
619 int a_op;
620 struct flock *a_fl;
621 int a_flags;
622 } */ *ap;
623{
624
625 return (ap->a_flags & F_FLOCK ? EOPNOTSUPP : EINVAL);
626}
627
628static void
629spec_getpages_iodone(bp)
630 struct buf *bp;
631{
632
633 bp->b_flags |= B_DONE;
634 wakeup(bp);
635}
636
637static int
638spec_getpages(ap)
639 struct vop_getpages_args *ap;
640{
641 vm_offset_t kva;
642 int error;
643 int i, pcount, size, s;
644 daddr_t blkno;
645 struct buf *bp;
646 vm_page_t m;
647 vm_ooffset_t offset;
648 int toff, nextoff, nread;
649 struct vnode *vp = ap->a_vp;
650 int blksiz;
651 int gotreqpage;
652
653 error = 0;
654 pcount = round_page(ap->a_count) / PAGE_SIZE;
655
656 /*
657 * Calculate the offset of the transfer and do sanity check.
658 * FreeBSD currently only supports an 8 TB range due to b_blkno
659 * being in DEV_BSIZE ( usually 512 ) byte chunks on call to
660 * VOP_STRATEGY. XXX
661 */
662 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
663
664#define DADDR_T_BIT (sizeof(daddr_t)*8)
665#define OFFSET_MAX ((1LL << (DADDR_T_BIT + DEV_BSHIFT)) - 1)
666
667 if (offset < 0 || offset > OFFSET_MAX) {
668 /* XXX still no %q in kernel. */
669 printf("spec_getpages: preposterous offset 0x%x%08x\n",
670 (u_int)((u_quad_t)offset >> 32),
671 (u_int)(offset & 0xffffffff));
672 return (VM_PAGER_ERROR);
673 }
674
675 blkno = btodb(offset);
676
677 /*
678 * Round up physical size for real devices. We cannot round using
679 * v_mount's block size data because v_mount has nothing to do with
680 * the device. i.e. it's usually '/dev'. We need the physical block
681 * size for the device itself.
682 *
683 * We can't use v_specmountpoint because it only exists when the
684 * block device is mounted. However, we can use v_rdev.
685 */
686
687 if (vn_isdisk(vp, NULL))
688 blksiz = vp->v_rdev->si_bsize_phys;
689 else
690 blksiz = DEV_BSIZE;
691
692 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
693
694 bp = getpbuf(NULL);
695 kva = (vm_offset_t)bp->b_data;
696
697 /*
698 * Map the pages to be read into the kva.
699 */
700 pmap_qenter(kva, ap->a_m, pcount);
701
702 /* Build a minimal buffer header. */
703 bp->b_flags = B_READ | B_CALL;
704 bp->b_iodone = spec_getpages_iodone;
705
706 /* B_PHYS is not set, but it is nice to fill this in. */
707 bp->b_rcred = bp->b_wcred = curproc->p_ucred;
708 if (bp->b_rcred != NOCRED)
709 crhold(bp->b_rcred);
710 if (bp->b_wcred != NOCRED)
711 crhold(bp->b_wcred);
712 bp->b_blkno = blkno;
713 bp->b_lblkno = blkno;
714 pbgetvp(ap->a_vp, bp);
715 bp->b_bcount = size;
716 bp->b_bufsize = size;
717 bp->b_resid = 0;
718 bp->b_runningbufspace = bp->b_bufsize;
719 runningbufspace += bp->b_runningbufspace;
720
721 cnt.v_vnodein++;
722 cnt.v_vnodepgsin += pcount;
723
724 /* Do the input. */
725 VOP_STRATEGY(bp->b_vp, bp);
726
727 s = splbio();
728
729 /* We definitely need to be at splbio here. */
730 while ((bp->b_flags & B_DONE) == 0)
731 tsleep(bp, PVM, "spread", 0);
732
733 splx(s);
734
735 if ((bp->b_flags & B_ERROR) != 0) {
736 if (bp->b_error)
737 error = bp->b_error;
738 else
739 error = EIO;
740 }
741
742 nread = size - bp->b_resid;
743
744 if (nread < ap->a_count) {
745 bzero((caddr_t)kva + nread,
746 ap->a_count - nread);
747 }
748 pmap_qremove(kva, pcount);
749
750
751 gotreqpage = 0;
752 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
753 nextoff = toff + PAGE_SIZE;
754 m = ap->a_m[i];
755
756 m->flags &= ~PG_ZERO;
757
758 if (nextoff <= nread) {
759 m->valid = VM_PAGE_BITS_ALL;
760 vm_page_undirty(m);
761 } else if (toff < nread) {
762 /*
763 * Since this is a VM request, we have to supply the
764 * unaligned offset to allow vm_page_set_validclean()
765 * to zero sub-DEV_BSIZE'd portions of the page.
766 */
767 vm_page_set_validclean(m, 0, nread - toff);
768 } else {
769 m->valid = 0;
770 vm_page_undirty(m);
771 }
772
773 if (i != ap->a_reqpage) {
774 /*
775 * Just in case someone was asking for this page we
776 * now tell them that it is ok to use.
777 */
778 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
779 if (m->valid) {
780 if (m->flags & PG_WANTED) {
781 vm_page_activate(m);
782 } else {
783 vm_page_deactivate(m);
784 }
785 vm_page_wakeup(m);
786 } else {
787 vm_page_free(m);
788 }
789 } else {
790 vm_page_free(m);
791 }
792 } else if (m->valid) {
793 gotreqpage = 1;
794 /*
795 * Since this is a VM request, we need to make the
796 * entire page presentable by zeroing invalid sections.
797 */
798 if (m->valid != VM_PAGE_BITS_ALL)
799 vm_page_zero_invalid(m, FALSE);
800 }
801 }
802 if (!gotreqpage) {
803 m = ap->a_m[ap->a_reqpage];
804 printf(
805 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
806 devtoname(bp->b_dev), error, bp, bp->b_vp);
807 printf(
808 " size: %d, resid: %ld, a_count: %d, valid: 0x%x\n",
809 size, bp->b_resid, ap->a_count, m->valid);
810 printf(
811 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
812 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
813 /*
814 * Free the buffer header back to the swap buffer pool.
815 */
816 relpbuf(bp, NULL);
817 return VM_PAGER_ERROR;
818 }
819 /*
820 * Free the buffer header back to the swap buffer pool.
821 */
822 relpbuf(bp, NULL);
823 return VM_PAGER_OK;
824}