Add missing commit for the VM load heuristic and page allocation rate
[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 $
81b5c339 35 * $DragonFly: src/sys/vfs/specfs/spec_vnops.c,v 1.30 2006/02/17 19:18:07 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>
335dda38 44#include <sys/device.h>
984263bc
MD
45#include <sys/mount.h>
46#include <sys/vnode.h>
47#include <sys/stat.h>
48#include <sys/fcntl.h>
49#include <sys/vmmeter.h>
50#include <sys/tty.h>
51
52#include <vm/vm.h>
53#include <vm/vm_object.h>
54#include <vm/vm_page.h>
55#include <vm/vm_pager.h>
56
3020e3be
MD
57#include <sys/buf2.h>
58
165dba55
DR
59#include <sys/thread2.h>
60
a6ee311a
RG
61static int spec_advlock (struct vop_advlock_args *);
62static int spec_bmap (struct vop_bmap_args *);
63static int spec_close (struct vop_close_args *);
64static int spec_freeblks (struct vop_freeblks_args *);
65static int spec_fsync (struct vop_fsync_args *);
66static int spec_getpages (struct vop_getpages_args *);
67static int spec_inactive (struct vop_inactive_args *);
68static int spec_ioctl (struct vop_ioctl_args *);
69static int spec_open (struct vop_open_args *);
70static int spec_poll (struct vop_poll_args *);
71static int spec_kqfilter (struct vop_kqfilter_args *);
72static int spec_print (struct vop_print_args *);
73static int spec_read (struct vop_read_args *);
74static int spec_strategy (struct vop_strategy_args *);
75static int spec_write (struct vop_write_args *);
984263bc 76
2d3e977e 77struct vop_ops *spec_vnode_vops;
0961aa92 78struct vnodeopv_entry_desc spec_vnodeop_entries[] = {
2d3e977e
MD
79 { &vop_default_desc, vop_defaultop },
80 { &vop_access_desc, vop_ebadf },
625ddaba
JS
81 { &vop_advlock_desc, (vnodeopv_entry_t) spec_advlock },
82 { &vop_bmap_desc, (vnodeopv_entry_t) spec_bmap },
83 { &vop_close_desc, (vnodeopv_entry_t) spec_close },
e62afb5f 84 { &vop_old_create_desc, vop_panic },
625ddaba
JS
85 { &vop_freeblks_desc, (vnodeopv_entry_t) spec_freeblks },
86 { &vop_fsync_desc, (vnodeopv_entry_t) spec_fsync },
87 { &vop_getpages_desc, (vnodeopv_entry_t) spec_getpages },
88 { &vop_inactive_desc, (vnodeopv_entry_t) spec_inactive },
89 { &vop_ioctl_desc, (vnodeopv_entry_t) spec_ioctl },
2d3e977e 90 { &vop_lease_desc, vop_null },
e62afb5f
MD
91 { &vop_old_link_desc, vop_panic },
92 { &vop_old_mkdir_desc, vop_panic },
93 { &vop_old_mknod_desc, vop_panic },
625ddaba
JS
94 { &vop_open_desc, (vnodeopv_entry_t) spec_open },
95 { &vop_pathconf_desc, (vnodeopv_entry_t) vop_stdpathconf },
96 { &vop_poll_desc, (vnodeopv_entry_t) spec_poll },
97 { &vop_kqfilter_desc, (vnodeopv_entry_t) spec_kqfilter },
98 { &vop_print_desc, (vnodeopv_entry_t) spec_print },
99 { &vop_read_desc, (vnodeopv_entry_t) spec_read },
2d3e977e
MD
100 { &vop_readdir_desc, vop_panic },
101 { &vop_readlink_desc, vop_panic },
102 { &vop_reallocblks_desc, vop_panic },
103 { &vop_reclaim_desc, vop_null },
e62afb5f
MD
104 { &vop_old_remove_desc, vop_panic },
105 { &vop_old_rename_desc, vop_panic },
106 { &vop_old_rmdir_desc, vop_panic },
2d3e977e 107 { &vop_setattr_desc, vop_ebadf },
625ddaba 108 { &vop_strategy_desc, (vnodeopv_entry_t) spec_strategy },
e62afb5f 109 { &vop_old_symlink_desc, vop_panic },
625ddaba 110 { &vop_write_desc, (vnodeopv_entry_t) spec_write },
984263bc
MD
111 { NULL, NULL }
112};
113static struct vnodeopv_desc spec_vnodeop_opv_desc =
dc1be39c 114 { &spec_vnode_vops, spec_vnodeop_entries, 0 };
984263bc
MD
115VNODEOP_SET(spec_vnodeop_opv_desc);
116
e4c9c0c8
MD
117extern int dev_ref_debug;
118
50775720
CP
119/*
120 * spec_vnoperate(struct vnodeop_desc *a_desc, ...)
121 */
984263bc 122int
50775720 123spec_vnoperate(struct vop_generic_args *ap)
984263bc 124{
0961aa92 125 return (VOCALL(spec_vnode_vops, ap));
984263bc
MD
126}
127
81b5c339 128static void spec_getpages_iodone (struct bio *bio);
984263bc
MD
129
130/*
131 * Open a special file.
50775720
CP
132 *
133 * spec_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred,
134 * struct thread *a_td)
984263bc
MD
135 */
136/* ARGSUSED */
137static int
50775720 138spec_open(struct vop_open_args *ap)
984263bc 139{
984263bc 140 struct vnode *vp = ap->a_vp;
e4c9c0c8 141 dev_t dev;
984263bc 142 int error;
e4c9c0c8 143 int isblk = (vp->v_type == VBLK) ? 1 : 0;
984263bc
MD
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
e4c9c0c8
MD
152 /*
153 * Resolve the device. If the vnode is already open v_rdev may
154 * already be resolved. However, if the device changes out from
1fbb5fc0
MD
155 * under us we report it (and, for now, we allow it). Since
156 * v_release_rdev() zero's v_opencount, we have to save and restore
157 * it when replacing the rdev reference.
e4c9c0c8
MD
158 */
159 if (vp->v_rdev != NULL) {
160 dev = udev2dev(vp->v_udev, isblk);
161 if (dev != vp->v_rdev) {
1fbb5fc0 162 int oc = vp->v_opencount;
e4c9c0c8
MD
163 printf(
164 "Warning: spec_open: dev %s was lost",
165 vp->v_rdev->si_name);
166 v_release_rdev(vp);
167 error = v_associate_rdev(vp,
168 udev2dev(vp->v_udev, isblk));
1fbb5fc0 169 if (error) {
e4c9c0c8 170 printf(", reacquisition failed\n");
1fbb5fc0
MD
171 } else {
172 vp->v_opencount = oc;
e4c9c0c8 173 printf(", reacquisition successful\n");
1fbb5fc0 174 }
e4c9c0c8
MD
175 } else {
176 error = 0;
177 }
178 } else {
179 error = v_associate_rdev(vp, udev2dev(vp->v_udev, isblk));
180 }
181 if (error)
182 return(error);
183
500df1da
MD
184 /*
185 * Prevent degenerate open/close sequences from nulling out rdev.
186 */
187 ++vp->v_opencount;
e4c9c0c8 188 dev = vp->v_rdev;
500df1da 189 KKASSERT(dev != NULL);
984263bc 190
e4c9c0c8
MD
191 /*
192 * Make this field valid before any I/O in ->d_open. XXX the
193 * device itself should probably be required to initialize
194 * this field in d_open.
195 */
984263bc
MD
196 if (!dev->si_iosize_max)
197 dev->si_iosize_max = DFLTPHYS;
198
199 /*
200 * XXX: Disks get special billing here, but it is mostly wrong.
201 * XXX: diskpartitions can overlap and the real checks should
202 * XXX: take this into account, and consequently they need to
203 * XXX: live in the diskslicing code. Some checks do.
204 */
205 if (vn_isdisk(vp, NULL) && ap->a_cred != FSCRED &&
206 (ap->a_mode & FWRITE)) {
207 /*
208 * Never allow opens for write if the device is mounted R/W
209 */
e4c9c0c8
MD
210 if (vp->v_rdev && vp->v_rdev->si_mountpoint &&
211 !(vp->v_rdev->si_mountpoint->mnt_flag & MNT_RDONLY)) {
212 error = EBUSY;
213 goto done;
214 }
984263bc
MD
215
216 /*
217 * When running in secure mode, do not allow opens
218 * for writing if the device is mounted
219 */
e4c9c0c8
MD
220 if (securelevel >= 1 && vfs_mountedon(vp)) {
221 error = EPERM;
222 goto done;
223 }
984263bc
MD
224
225 /*
226 * When running in very secure mode, do not allow
227 * opens for writing of any devices.
228 */
e4c9c0c8
MD
229 if (securelevel >= 2) {
230 error = EPERM;
231 goto done;
232 }
984263bc
MD
233 }
234
235 /* XXX: Special casing of ttys for deadfs. Probably redundant */
335dda38 236 if (dev_dflags(dev) & D_TTY)
984263bc
MD
237 vp->v_flag |= VISTTY;
238
e4c9c0c8
MD
239 /*
240 * dev_dopen() is always called for each open. dev_dclose() is
241 * only called for the last close unless D_TRACKCLOSE is set.
242 */
5fd012e0 243 VOP_UNLOCK(vp, 0, ap->a_td);
335dda38 244 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_td);
5fd012e0 245 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
984263bc
MD
246
247 if (error)
e4c9c0c8 248 goto done;
984263bc 249
335dda38 250 if (dev_dflags(dev) & D_TTY) {
984263bc
MD
251 if (dev->si_tty) {
252 struct tty *tp;
253 tp = dev->si_tty;
254 if (!tp->t_stop) {
255 printf("Warning:%s: no t_stop, using nottystop\n", devtoname(dev));
256 tp->t_stop = nottystop;
257 }
258 }
259 }
260
261 if (vn_isdisk(vp, NULL)) {
262 if (!dev->si_bsize_phys)
263 dev->si_bsize_phys = DEV_BSIZE;
264 }
335dda38 265 if ((dev_dflags(dev) & D_DISK) == 0) {
984263bc 266 cp = devtoname(dev);
335dda38 267 if (*cp == '#') {
984263bc 268 printf("WARNING: driver %s should register devices with make_dev() (dev_t = \"%s\")\n",
335dda38 269 dev_dname(dev), cp);
984263bc
MD
270 }
271 }
fad57d0e
MD
272
273 /*
274 * If we were handed a file pointer we may be able to install a
275 * shortcut which issues device read and write operations directly
276 * from the fileops rather then having to go through spec_read()
277 * and spec_write().
278 */
279 if (ap->a_fp)
280 vn_setspecops(ap->a_fp);
281
e4c9c0c8
MD
282 if (dev_ref_debug)
283 printf("spec_open: %s %d\n", dev->si_name, vp->v_opencount);
284done:
285 if (error) {
500df1da
MD
286 KKASSERT(vp->v_opencount > 0);
287 if (--vp->v_opencount == 0)
e4c9c0c8
MD
288 v_release_rdev(vp);
289 }
984263bc
MD
290 return (error);
291}
292
293/*
294 * Vnode op for read
50775720
CP
295 *
296 * spec_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
297 * struct ucred *a_cred)
984263bc
MD
298 */
299/* ARGSUSED */
300static int
50775720 301spec_read(struct vop_read_args *ap)
984263bc
MD
302{
303 struct vnode *vp;
dadab5e9 304 struct thread *td;
984263bc
MD
305 struct uio *uio;
306 dev_t dev;
307 int error;
308
309 vp = ap->a_vp;
310 dev = vp->v_rdev;
311 uio = ap->a_uio;
dadab5e9 312 td = uio->uio_td;
984263bc 313
72402f22
MD
314 if (dev == NULL) /* device was revoked */
315 return (EBADF);
984263bc
MD
316 if (uio->uio_resid == 0)
317 return (0);
318
5fd012e0 319 VOP_UNLOCK(vp, 0, td);
335dda38 320 error = dev_dread(dev, uio, ap->a_ioflag);
5fd012e0 321 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
984263bc
MD
322 return (error);
323}
324
325/*
326 * Vnode op for write
50775720
CP
327 *
328 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
329 * struct ucred *a_cred)
984263bc
MD
330 */
331/* ARGSUSED */
332static int
50775720 333spec_write(struct vop_write_args *ap)
984263bc
MD
334{
335 struct vnode *vp;
dadab5e9 336 struct thread *td;
984263bc
MD
337 struct uio *uio;
338 dev_t dev;
339 int error;
340
341 vp = ap->a_vp;
342 dev = vp->v_rdev;
343 uio = ap->a_uio;
dadab5e9 344 td = uio->uio_td;
984263bc 345
72402f22
MD
346 if (dev == NULL) /* device was revoked */
347 return (EBADF);
348
5fd012e0 349 VOP_UNLOCK(vp, 0, td);
335dda38 350 error = dev_dwrite(dev, uio, ap->a_ioflag);
5fd012e0 351 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
984263bc
MD
352 return (error);
353}
354
355/*
356 * Device ioctl operation.
50775720
CP
357 *
358 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
359 * int a_fflag, struct ucred *a_cred, struct thread *a_td)
984263bc
MD
360 */
361/* ARGSUSED */
362static int
50775720 363spec_ioctl(struct vop_ioctl_args *ap)
984263bc
MD
364{
365 dev_t dev;
366
72402f22
MD
367 if ((dev = ap->a_vp->v_rdev) == NULL)
368 return (EBADF); /* device was revoked */
369
335dda38
MD
370 return (dev_dioctl(dev, ap->a_command, ap->a_data,
371 ap->a_fflag, ap->a_td));
984263bc
MD
372}
373
50775720
CP
374/*
375 * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred,
376 * struct thread *a_td)
377 */
984263bc
MD
378/* ARGSUSED */
379static int
50775720 380spec_poll(struct vop_poll_args *ap)
984263bc
MD
381{
382 dev_t dev;
383
72402f22
MD
384 if ((dev = ap->a_vp->v_rdev) == NULL)
385 return (EBADF); /* device was revoked */
335dda38 386 return (dev_dpoll(dev, ap->a_events, ap->a_td));
984263bc
MD
387}
388
50775720
CP
389/*
390 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
391 */
984263bc
MD
392/* ARGSUSED */
393static int
50775720 394spec_kqfilter(struct vop_kqfilter_args *ap)
984263bc
MD
395{
396 dev_t dev;
397
72402f22
MD
398 if ((dev = ap->a_vp->v_rdev) == NULL)
399 return (EBADF); /* device was revoked */
335dda38 400 return (dev_dkqfilter(dev, ap->a_kn));
984263bc
MD
401}
402
403/*
404 * Synch buffers associated with a block device
50775720
CP
405 *
406 * spec_fsync(struct vnode *a_vp, struct ucred *a_cred,
407 * int a_waitfor, struct thread *a_td)
984263bc
MD
408 */
409/* ARGSUSED */
410static int
50775720 411spec_fsync(struct vop_fsync_args *ap)
984263bc
MD
412{
413 struct vnode *vp = ap->a_vp;
6bae6177 414 int error;
984263bc
MD
415
416 if (!vn_isdisk(vp, NULL))
417 return (0);
418
984263bc
MD
419 /*
420 * Flush all dirty buffers associated with a block device.
421 */
6bae6177
MD
422 error = vfsync(vp, ap->a_waitfor, 10000, (daddr_t)-1, NULL, NULL);
423 return (error);
984263bc
MD
424}
425
50775720
CP
426/*
427 * spec_inactive(struct vnode *a_vp, struct thread *a_td)
428 */
984263bc 429static int
50775720 430spec_inactive(struct vop_inactive_args *ap)
984263bc 431{
984263bc
MD
432 return (0);
433}
434
435/*
436 * Just call the device strategy routine
50775720 437 *
81b5c339 438 * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
984263bc
MD
439 */
440static int
50775720 441spec_strategy(struct vop_strategy_args *ap)
984263bc 442{
81b5c339
MD
443 struct bio *bio = ap->a_bio;
444 struct buf *bp = bio->bio_buf;
984263bc
MD
445 struct vnode *vp;
446 struct mount *mp;
447
984263bc
MD
448 if (((bp->b_flags & B_READ) == 0) &&
449 (LIST_FIRST(&bp->b_dep)) != NULL && bioops.io_start)
450 (*bioops.io_start)(bp);
451
452 /*
453 * Collect statistics on synchronous and asynchronous read
454 * and write counts for disks that have associated filesystems.
455 */
456 vp = ap->a_vp;
72402f22 457 KKASSERT(vp->v_rdev != NULL); /* XXX */
e4c9c0c8 458 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
984263bc 459 if ((bp->b_flags & B_READ) == 0) {
dadab5e9 460 if (bp->b_lock.lk_lockholder == LK_KERNTHREAD)
984263bc
MD
461 mp->mnt_stat.f_asyncwrites++;
462 else
463 mp->mnt_stat.f_syncwrites++;
464 } else {
dadab5e9 465 if (bp->b_lock.lk_lockholder == LK_KERNTHREAD)
984263bc
MD
466 mp->mnt_stat.f_asyncreads++;
467 else
468 mp->mnt_stat.f_syncreads++;
469 }
470 }
81b5c339 471 dev_dstrategy_chain(vp->v_rdev, bio);
984263bc
MD
472 return (0);
473}
474
50775720
CP
475/*
476 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
477 */
984263bc 478static int
50775720 479spec_freeblks(struct vop_freeblks_args *ap)
984263bc 480{
984263bc
MD
481 struct buf *bp;
482
483 /*
484 * XXX: This assumes that strategy does the deed right away.
485 * XXX: this may not be TRTTD.
486 */
72402f22 487 KKASSERT(ap->a_vp->v_rdev != NULL);
335dda38 488 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
984263bc
MD
489 return (0);
490 bp = geteblk(ap->a_length);
491 bp->b_flags |= B_FREEBUF;
81b5c339
MD
492 bp->b_bio1.bio_blkno = ap->a_addr;
493 bp->b_bio1.bio_offset = dbtob(ap->a_addr);
984263bc 494 bp->b_bcount = ap->a_length;
81b5c339 495 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
984263bc
MD
496 return (0);
497}
498
499/*
500 * Implement degenerate case where the block requested is the block
501 * returned, and assume that the entire device is contiguous in regards
502 * to the contiguous block range (runp and runb).
50775720
CP
503 *
504 * spec_bmap(struct vnode *a_vp, daddr_t a_bn, struct vnode **a_vpp,
505 * daddr_t *a_bnp, int *a_runp, int *a_runb)
984263bc
MD
506 */
507static int
50775720 508spec_bmap(struct vop_bmap_args *ap)
984263bc
MD
509{
510 struct vnode *vp = ap->a_vp;
511 int runp = 0;
512 int runb = 0;
513
514 if (ap->a_vpp != NULL)
515 *ap->a_vpp = vp;
516 if (ap->a_bnp != NULL)
517 *ap->a_bnp = ap->a_bn;
518 if (vp->v_mount != NULL)
519 runp = runb = MAXBSIZE / vp->v_mount->mnt_stat.f_iosize;
520 if (ap->a_runp != NULL)
521 *ap->a_runp = runp;
522 if (ap->a_runb != NULL)
523 *ap->a_runb = runb;
524 return (0);
525}
526
527/*
528 * Device close routine
50775720
CP
529 *
530 * spec_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
531 * struct thread *a_td)
53994a38
MD
532 *
533 * NOTE: the vnode may or may not be locked on call.
984263bc
MD
534 */
535/* ARGSUSED */
536static int
50775720 537spec_close(struct vop_close_args *ap)
984263bc 538{
dadab5e9 539 struct proc *p = ap->a_td->td_proc;
984263bc 540 struct vnode *vp = ap->a_vp;
984263bc 541 dev_t dev = vp->v_rdev;
e4c9c0c8 542 int error;
53994a38 543 int needrelock;
984263bc
MD
544
545 /*
546 * Hack: a tty device that is a controlling terminal
547 * has a reference from the session structure.
548 * We cannot easily tell that a character device is
549 * a controlling terminal, unless it is the closing
550 * process' controlling terminal. In that case,
551 * if the reference count is 2 (this last descriptor
552 * plus the session), release the reference from the session.
1fbb5fc0
MD
553 *
554 * It is possible for v_opencount to be 0 or 1 in this case, 0
555 * because the tty might have been revoked.
984263bc 556 */
72402f22
MD
557 if (dev)
558 reference_dev(dev);
1fbb5fc0 559 if (vcount(vp) == 2 && vp->v_opencount <= 1 &&
5fd012e0 560 p && vp == p->p_session->s_ttyvp) {
984263bc 561 p->p_session->s_ttyvp = NULL;
1fbb5fc0 562 vrele(vp);
984263bc 563 }
e4c9c0c8 564
984263bc 565 /*
e4c9c0c8
MD
566 * Vnodes can be opened and close multiple times. Do not really
567 * close the device unless (1) it is being closed forcibly,
568 * (2) the device wants to track closes, or (3) this is the last
569 * vnode doing its last close on the device.
570 *
571 * XXX the VXLOCK (force close) case can leave vnodes referencing
572 * a closed device.
984263bc 573 */
5fd012e0 574 if (dev && ((vp->v_flag & VRECLAIMED) ||
e4c9c0c8 575 (dev_dflags(dev) & D_TRACKCLOSE) ||
72402f22 576 (vcount(vp) <= 1 && vp->v_opencount == 1))) {
53994a38
MD
577 needrelock = 0;
578 if (VOP_ISLOCKED(vp, ap->a_td)) {
579 needrelock = 1;
580 VOP_UNLOCK(vp, 0, ap->a_td);
581 }
e4c9c0c8 582 error = dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_td);
53994a38
MD
583 if (needrelock)
584 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, ap->a_td);
e4c9c0c8
MD
585 } else {
586 error = 0;
984263bc 587 }
e4c9c0c8
MD
588
589 /*
590 * Track the actual opens and closes on the vnode. The last close
1fbb5fc0
MD
591 * disassociates the rdev. If the rdev is already disassociated
592 * the vnode might have been revoked and no further opencount
593 * tracking occurs.
e4c9c0c8 594 */
1fbb5fc0
MD
595 if (dev) {
596 KKASSERT(vp->v_opencount > 0);
597 if (dev_ref_debug) {
72402f22
MD
598 printf("spec_close: %s %d\n",
599 dev->si_name, vp->v_opencount - 1);
72402f22 600 }
1fbb5fc0
MD
601 if (--vp->v_opencount == 0)
602 v_release_rdev(vp);
72402f22 603 release_dev(dev);
1fbb5fc0 604 }
e4c9c0c8 605 return(error);
984263bc
MD
606}
607
608/*
609 * Print out the contents of a special device vnode.
50775720
CP
610 *
611 * spec_print(struct vnode *a_vp)
984263bc
MD
612 */
613static int
50775720 614spec_print(struct vop_print_args *ap)
984263bc 615{
984263bc
MD
616 printf("tag VT_NON, dev %s\n", devtoname(ap->a_vp->v_rdev));
617 return (0);
618}
619
620/*
621 * Special device advisory byte-level locks.
50775720
CP
622 *
623 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
624 * struct flock *a_fl, int a_flags)
984263bc
MD
625 */
626/* ARGSUSED */
627static int
50775720 628spec_advlock(struct vop_advlock_args *ap)
984263bc 629{
984263bc
MD
630 return (ap->a_flags & F_FLOCK ? EOPNOTSUPP : EINVAL);
631}
632
633static void
81b5c339 634spec_getpages_iodone(struct bio *bio)
984263bc 635{
81b5c339
MD
636 bio->bio_buf->b_flags |= B_DONE;
637 wakeup(bio->bio_buf);
984263bc
MD
638}
639
640static int
50775720 641spec_getpages(struct vop_getpages_args *ap)
984263bc
MD
642{
643 vm_offset_t kva;
644 int error;
165dba55 645 int i, pcount, size;
984263bc
MD
646 daddr_t blkno;
647 struct buf *bp;
648 vm_page_t m;
649 vm_ooffset_t offset;
650 int toff, nextoff, nread;
651 struct vnode *vp = ap->a_vp;
652 int blksiz;
653 int gotreqpage;
654
655 error = 0;
656 pcount = round_page(ap->a_count) / PAGE_SIZE;
657
658 /*
659 * Calculate the offset of the transfer and do sanity check.
81b5c339 660 * FreeBSD currently only supports an 8 TB range due to bio_blkno
984263bc 661 * being in DEV_BSIZE ( usually 512 ) byte chunks on call to
81b5c339 662 * vn_strategy(). XXX
984263bc
MD
663 */
664 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
665
666#define DADDR_T_BIT (sizeof(daddr_t)*8)
667#define OFFSET_MAX ((1LL << (DADDR_T_BIT + DEV_BSHIFT)) - 1)
668
669 if (offset < 0 || offset > OFFSET_MAX) {
670 /* XXX still no %q in kernel. */
671 printf("spec_getpages: preposterous offset 0x%x%08x\n",
672 (u_int)((u_quad_t)offset >> 32),
673 (u_int)(offset & 0xffffffff));
674 return (VM_PAGER_ERROR);
675 }
676
677 blkno = btodb(offset);
678
679 /*
680 * Round up physical size for real devices. We cannot round using
681 * v_mount's block size data because v_mount has nothing to do with
682 * the device. i.e. it's usually '/dev'. We need the physical block
683 * size for the device itself.
684 *
e4c9c0c8 685 * We can't use v_rdev->si_mountpoint because it only exists when the
984263bc
MD
686 * block device is mounted. However, we can use v_rdev.
687 */
688
689 if (vn_isdisk(vp, NULL))
690 blksiz = vp->v_rdev->si_bsize_phys;
691 else
692 blksiz = DEV_BSIZE;
693
694 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
695
696 bp = getpbuf(NULL);
697 kva = (vm_offset_t)bp->b_data;
698
699 /*
700 * Map the pages to be read into the kva.
701 */
702 pmap_qenter(kva, ap->a_m, pcount);
703
704 /* Build a minimal buffer header. */
5ff4cfb5 705 bp->b_flags = B_READ;
984263bc
MD
706
707 /* B_PHYS is not set, but it is nice to fill this in. */
984263bc
MD
708 pbgetvp(ap->a_vp, bp);
709 bp->b_bcount = size;
710 bp->b_bufsize = size;
711 bp->b_resid = 0;
712 bp->b_runningbufspace = bp->b_bufsize;
713 runningbufspace += bp->b_runningbufspace;
714
81b5c339
MD
715 bp->b_bio1.bio_blkno = blkno;
716 bp->b_bio1.bio_done = spec_getpages_iodone;
717
12e4aaff
MD
718 mycpu->gd_cnt.v_vnodein++;
719 mycpu->gd_cnt.v_vnodepgsin += pcount;
984263bc
MD
720
721 /* Do the input. */
81b5c339 722 vn_strategy(ap->a_vp, &bp->b_bio1);
984263bc 723
165dba55 724 crit_enter();
984263bc
MD
725
726 /* We definitely need to be at splbio here. */
26a0694b 727 while ((bp->b_flags & B_DONE) == 0) {
377d4740 728 tsleep(bp, 0, "spread", 0);
26a0694b 729 }
984263bc 730
165dba55 731 crit_exit();
984263bc
MD
732
733 if ((bp->b_flags & B_ERROR) != 0) {
734 if (bp->b_error)
735 error = bp->b_error;
736 else
737 error = EIO;
738 }
739
740 nread = size - bp->b_resid;
741
742 if (nread < ap->a_count) {
743 bzero((caddr_t)kva + nread,
744 ap->a_count - nread);
745 }
746 pmap_qremove(kva, pcount);
747
748
749 gotreqpage = 0;
750 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
751 nextoff = toff + PAGE_SIZE;
752 m = ap->a_m[i];
753
754 m->flags &= ~PG_ZERO;
755
756 if (nextoff <= nread) {
757 m->valid = VM_PAGE_BITS_ALL;
758 vm_page_undirty(m);
759 } else if (toff < nread) {
760 /*
761 * Since this is a VM request, we have to supply the
762 * unaligned offset to allow vm_page_set_validclean()
763 * to zero sub-DEV_BSIZE'd portions of the page.
764 */
765 vm_page_set_validclean(m, 0, nread - toff);
766 } else {
767 m->valid = 0;
768 vm_page_undirty(m);
769 }
770
771 if (i != ap->a_reqpage) {
772 /*
773 * Just in case someone was asking for this page we
774 * now tell them that it is ok to use.
775 */
776 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
777 if (m->valid) {
778 if (m->flags & PG_WANTED) {
779 vm_page_activate(m);
780 } else {
781 vm_page_deactivate(m);
782 }
783 vm_page_wakeup(m);
784 } else {
785 vm_page_free(m);
786 }
787 } else {
788 vm_page_free(m);
789 }
790 } else if (m->valid) {
791 gotreqpage = 1;
792 /*
793 * Since this is a VM request, we need to make the
794 * entire page presentable by zeroing invalid sections.
795 */
796 if (m->valid != VM_PAGE_BITS_ALL)
797 vm_page_zero_invalid(m, FALSE);
798 }
799 }
800 if (!gotreqpage) {
801 m = ap->a_m[ap->a_reqpage];
802 printf(
803 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
81b5c339 804 devtoname(vp->v_rdev), error, bp, bp->b_vp);
984263bc
MD
805 printf(
806 " size: %d, resid: %ld, a_count: %d, valid: 0x%x\n",
807 size, bp->b_resid, ap->a_count, m->valid);
808 printf(
809 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
810 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
811 /*
812 * Free the buffer header back to the swap buffer pool.
813 */
814 relpbuf(bp, NULL);
815 return VM_PAGER_ERROR;
816 }
817 /*
818 * Free the buffer header back to the swap buffer pool.
819 */
820 relpbuf(bp, NULL);
821 return VM_PAGER_OK;
822}