kernel - Fix lost lock in devfs
[dragonfly.git] / sys / vfs / devfs / devfs_vnops.c
CommitLineData
21864bc5 1/*
9f889dc4
MD
2 * (MPSAFE)
3 *
21864bc5
MD
4 * Copyright (c) 2009 The DragonFly Project. All rights reserved.
5 *
6 * This code is derived from software contributed to The DragonFly Project
7 * by Alex Hornung <ahornung@gmail.com>
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 *
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in
17 * the documentation and/or other materials provided with the
18 * distribution.
19 * 3. Neither the name of The DragonFly Project nor the names of its
20 * contributors may be used to endorse or promote products derived
21 * from this software without specific, prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
27 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 */
36#include <sys/param.h>
37#include <sys/systm.h>
38#include <sys/time.h>
39#include <sys/kernel.h>
40#include <sys/lock.h>
41#include <sys/fcntl.h>
42#include <sys/proc.h>
43#include <sys/priv.h>
44#include <sys/signalvar.h>
45#include <sys/vnode.h>
46#include <sys/uio.h>
47#include <sys/mount.h>
48#include <sys/file.h>
49#include <sys/fcntl.h>
50#include <sys/namei.h>
51#include <sys/dirent.h>
52#include <sys/malloc.h>
53#include <sys/stat.h>
54#include <sys/reg.h>
21864bc5
MD
55#include <vm/vm_pager.h>
56#include <vm/vm_zone.h>
57#include <vm/vm_object.h>
58#include <sys/filio.h>
59#include <sys/ttycom.h>
21864bc5 60#include <sys/tty.h>
2d076755 61#include <sys/diskslice.h>
3a1032a6 62#include <sys/sysctl.h>
2c1e28dd 63#include <sys/devfs.h>
21864bc5 64#include <sys/pioctl.h>
c705e298 65#include <vfs/fifofs/fifo.h>
21864bc5
MD
66
67#include <machine/limits.h>
684a93c4 68
1a54183b
MD
69#include <sys/buf2.h>
70#include <sys/sysref2.h>
684a93c4
MD
71#include <sys/mplock2.h>
72#include <vm/vm_page2.h>
21864bc5
MD
73
74MALLOC_DECLARE(M_DEVFS);
9f889dc4
MD
75#define DEVFS_BADOP (void *)devfs_vop_badop
76
77static int devfs_vop_badop(struct vop_generic_args *);
78static int devfs_vop_access(struct vop_access_args *);
79static int devfs_vop_inactive(struct vop_inactive_args *);
80static int devfs_vop_reclaim(struct vop_reclaim_args *);
81static int devfs_vop_readdir(struct vop_readdir_args *);
82static int devfs_vop_getattr(struct vop_getattr_args *);
83static int devfs_vop_setattr(struct vop_setattr_args *);
84static int devfs_vop_readlink(struct vop_readlink_args *);
85static int devfs_vop_print(struct vop_print_args *);
86
87static int devfs_vop_nresolve(struct vop_nresolve_args *);
88static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
89static int devfs_vop_nmkdir(struct vop_nmkdir_args *);
90static int devfs_vop_nsymlink(struct vop_nsymlink_args *);
91static int devfs_vop_nrmdir(struct vop_nrmdir_args *);
92static int devfs_vop_nremove(struct vop_nremove_args *);
21864bc5
MD
93
94static int devfs_spec_open(struct vop_open_args *);
95static int devfs_spec_close(struct vop_close_args *);
96static int devfs_spec_fsync(struct vop_fsync_args *);
97
98static int devfs_spec_read(struct vop_read_args *);
99static int devfs_spec_write(struct vop_write_args *);
100static int devfs_spec_ioctl(struct vop_ioctl_args *);
21864bc5
MD
101static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
102static int devfs_spec_strategy(struct vop_strategy_args *);
103static void devfs_spec_strategy_done(struct bio *);
104static int devfs_spec_freeblks(struct vop_freeblks_args *);
105static int devfs_spec_bmap(struct vop_bmap_args *);
106static int devfs_spec_advlock(struct vop_advlock_args *);
107static void devfs_spec_getpages_iodone(struct bio *);
108static int devfs_spec_getpages(struct vop_getpages_args *);
109
9f889dc4
MD
110static int devfs_fo_close(struct file *);
111static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int);
112static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int);
113static int devfs_fo_stat(struct file *, struct stat *, struct ucred *);
114static int devfs_fo_kqfilter(struct file *, struct knote *);
115static int devfs_fo_ioctl(struct file *, u_long, caddr_t,
87baaf0c 116 struct ucred *, struct sysmsg *);
21864bc5 117static __inline int sequential_heuristic(struct uio *, struct file *);
87baaf0c
MD
118
119extern struct lock devfs_lock;
21864bc5
MD
120
121/*
9f889dc4 122 * devfs vnode operations for regular files. All vnode ops are MPSAFE.
21864bc5
MD
123 */
124struct vop_ops devfs_vnode_norm_vops = {
125 .vop_default = vop_defaultop,
9f889dc4 126 .vop_access = devfs_vop_access,
21864bc5 127 .vop_advlock = DEVFS_BADOP,
9f889dc4 128 .vop_bmap = DEVFS_BADOP,
21864bc5 129 .vop_close = vop_stdclose,
9f889dc4
MD
130 .vop_getattr = devfs_vop_getattr,
131 .vop_inactive = devfs_vop_inactive,
21864bc5 132 .vop_ncreate = DEVFS_BADOP,
9f889dc4
MD
133 .vop_nresolve = devfs_vop_nresolve,
134 .vop_nlookupdotdot = devfs_vop_nlookupdotdot,
21864bc5 135 .vop_nlink = DEVFS_BADOP,
9f889dc4 136 .vop_nmkdir = devfs_vop_nmkdir,
21864bc5 137 .vop_nmknod = DEVFS_BADOP,
9f889dc4 138 .vop_nremove = devfs_vop_nremove,
21864bc5 139 .vop_nrename = DEVFS_BADOP,
9f889dc4
MD
140 .vop_nrmdir = devfs_vop_nrmdir,
141 .vop_nsymlink = devfs_vop_nsymlink,
142 .vop_open = vop_stdopen,
21864bc5 143 .vop_pathconf = vop_stdpathconf,
9f889dc4
MD
144 .vop_print = devfs_vop_print,
145 .vop_read = DEVFS_BADOP,
146 .vop_readdir = devfs_vop_readdir,
147 .vop_readlink = devfs_vop_readlink,
148 .vop_reclaim = devfs_vop_reclaim,
149 .vop_setattr = devfs_vop_setattr,
21864bc5
MD
150 .vop_write = DEVFS_BADOP,
151 .vop_ioctl = DEVFS_BADOP
152};
153
154/*
9f889dc4 155 * devfs vnode operations for character devices. All vnode ops are MPSAFE.
21864bc5
MD
156 */
157struct vop_ops devfs_vnode_dev_vops = {
158 .vop_default = vop_defaultop,
9f889dc4 159 .vop_access = devfs_vop_access,
21864bc5 160 .vop_advlock = devfs_spec_advlock,
9f889dc4 161 .vop_bmap = devfs_spec_bmap,
21864bc5
MD
162 .vop_close = devfs_spec_close,
163 .vop_freeblks = devfs_spec_freeblks,
164 .vop_fsync = devfs_spec_fsync,
9f889dc4 165 .vop_getattr = devfs_vop_getattr,
21864bc5 166 .vop_getpages = devfs_spec_getpages,
9f889dc4
MD
167 .vop_inactive = devfs_vop_inactive,
168 .vop_open = devfs_spec_open,
21864bc5 169 .vop_pathconf = vop_stdpathconf,
9f889dc4 170 .vop_print = devfs_vop_print,
21864bc5 171 .vop_kqfilter = devfs_spec_kqfilter,
9f889dc4 172 .vop_read = devfs_spec_read,
21864bc5
MD
173 .vop_readdir = DEVFS_BADOP,
174 .vop_readlink = DEVFS_BADOP,
9f889dc4
MD
175 .vop_reclaim = devfs_vop_reclaim,
176 .vop_setattr = devfs_vop_setattr,
21864bc5
MD
177 .vop_strategy = devfs_spec_strategy,
178 .vop_write = devfs_spec_write,
179 .vop_ioctl = devfs_spec_ioctl
180};
181
9f889dc4
MD
182/*
183 * devfs file pointer operations. All fileops are MPSAFE.
184 */
21864bc5
MD
185struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops;
186
187struct fileops devfs_dev_fileops = {
9f889dc4
MD
188 .fo_read = devfs_fo_read,
189 .fo_write = devfs_fo_write,
190 .fo_ioctl = devfs_fo_ioctl,
191 .fo_kqfilter = devfs_fo_kqfilter,
192 .fo_stat = devfs_fo_stat,
193 .fo_close = devfs_fo_close,
194 .fo_shutdown = nofo_shutdown
21864bc5
MD
195};
196
4062d050 197/*
9f889dc4
MD
198 * These two functions are possibly temporary hacks for devices (aka
199 * the pty code) which want to control the node attributes themselves.
4062d050
MD
200 *
201 * XXX we may ultimately desire to simply remove the uid/gid/mode
202 * from the node entirely.
9f889dc4
MD
203 *
204 * MPSAFE - sorta. Theoretically the overwrite can compete since they
205 * are loading from the same fields.
4062d050
MD
206 */
207static __inline void
208node_sync_dev_get(struct devfs_node *node)
209{
210 cdev_t dev;
211
212 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
213 node->uid = dev->si_uid;
214 node->gid = dev->si_gid;
215 node->mode = dev->si_perms;
216 }
217}
218
219static __inline void
220node_sync_dev_set(struct devfs_node *node)
221{
222 cdev_t dev;
223
224 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
225 dev->si_uid = node->uid;
226 dev->si_gid = node->gid;
227 dev->si_perms = node->mode;
228 }
229}
21864bc5
MD
230
231/*
232 * generic entry point for unsupported operations
233 */
234static int
9f889dc4 235devfs_vop_badop(struct vop_generic_args *ap)
21864bc5 236{
21864bc5
MD
237 return (EIO);
238}
239
240
241static int
9f889dc4 242devfs_vop_access(struct vop_access_args *ap)
21864bc5
MD
243{
244 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
898c91ee 245 int error;
21864bc5 246
894bbb25
AH
247 if (!devfs_node_is_accessible(node))
248 return ENOENT;
4062d050 249 node_sync_dev_get(node);
21864bc5 250 error = vop_helper_access(ap, node->uid, node->gid,
4062d050 251 node->mode, node->flags);
21864bc5 252
21864bc5 253 return error;
21864bc5
MD
254}
255
256
257static int
9f889dc4 258devfs_vop_inactive(struct vop_inactive_args *ap)
21864bc5 259{
ca8d7677 260 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
21864bc5 261
ca8d7677
MD
262 if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
263 vrecycle(ap->a_vp);
21864bc5
MD
264 return 0;
265}
266
267
268static int
9f889dc4 269devfs_vop_reclaim(struct vop_reclaim_args *ap)
21864bc5 270{
be6f2e86
MD
271 struct devfs_node *node;
272 struct vnode *vp;
273 int locked;
274
be6f2e86
MD
275 /*
276 * Check if it is locked already. if not, we acquire the devfs lock
277 */
21864bc5
MD
278 if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) {
279 lockmgr(&devfs_lock, LK_EXCLUSIVE);
280 locked = 1;
be6f2e86
MD
281 } else {
282 locked = 0;
21864bc5
MD
283 }
284
be6f2e86
MD
285 /*
286 * Get rid of the devfs_node if it is no longer linked into the
287 * topology.
288 */
289 vp = ap->a_vp;
290 if ((node = DEVFS_NODE(vp)) != NULL) {
4062d050
MD
291 node->v_node = NULL;
292 if ((node->flags & DEVFS_NODE_LINKED) == 0)
be6f2e86 293 devfs_freep(node);
21864bc5
MD
294 }
295
21864bc5
MD
296 if (locked)
297 lockmgr(&devfs_lock, LK_RELEASE);
298
be6f2e86 299 /*
9b823501
AH
300 * v_rdev needs to be properly released using v_release_rdev
301 * Make sure v_data is NULL as well.
be6f2e86
MD
302 */
303 vp->v_data = NULL;
9b823501 304 v_release_rdev(vp);
21864bc5
MD
305 return 0;
306}
307
308
309static int
9f889dc4 310devfs_vop_readdir(struct vop_readdir_args *ap)
21864bc5 311{
898c91ee 312 struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
21864bc5 313 struct devfs_node *node;
21864bc5
MD
314 int cookie_index;
315 int ncookies;
898c91ee
MD
316 int error2;
317 int error;
318 int r;
21864bc5
MD
319 off_t *cookies;
320 off_t saveoff;
321
322 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
323
324 if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
325 return (EINVAL);
326 if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
327 return (error);
328
c512ab96
MD
329 if (!devfs_node_is_accessible(dnode)) {
330 vn_unlock(ap->a_vp);
ca8d7677 331 return ENOENT;
c512ab96 332 }
ca8d7677
MD
333
334 lockmgr(&devfs_lock, LK_EXCLUSIVE);
335
21864bc5
MD
336 saveoff = ap->a_uio->uio_offset;
337
338 if (ap->a_ncookies) {
339 ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
340 if (ncookies > 256)
341 ncookies = 256;
342 cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
343 cookie_index = 0;
344 } else {
345 ncookies = -1;
346 cookies = NULL;
347 cookie_index = 0;
348 }
349
898c91ee 350 nanotime(&dnode->atime);
21864bc5
MD
351
352 if (saveoff == 0) {
898c91ee
MD
353 r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
354 DT_DIR, 1, ".");
21864bc5
MD
355 if (r)
356 goto done;
357 if (cookies)
358 cookies[cookie_index] = saveoff;
359 saveoff++;
360 cookie_index++;
361 if (cookie_index == ncookies)
362 goto done;
363 }
364
365 if (saveoff == 1) {
898c91ee 366 if (dnode->parent) {
21864bc5 367 r = vop_write_dirent(&error, ap->a_uio,
898c91ee 368 dnode->parent->d_dir.d_ino,
21864bc5
MD
369 DT_DIR, 2, "..");
370 } else {
371 r = vop_write_dirent(&error, ap->a_uio,
898c91ee
MD
372 dnode->d_dir.d_ino,
373 DT_DIR, 2, "..");
21864bc5
MD
374 }
375 if (r)
376 goto done;
377 if (cookies)
378 cookies[cookie_index] = saveoff;
379 saveoff++;
380 cookie_index++;
381 if (cookie_index == ncookies)
382 goto done;
383 }
384
898c91ee
MD
385 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
386 if ((node->flags & DEVFS_HIDDEN) ||
387 (node->flags & DEVFS_INVISIBLE)) {
21864bc5 388 continue;
898c91ee 389 }
21864bc5 390
f7e8960c 391 /*
9f889dc4
MD
392 * If the node type is a valid devfs alias, then we make
393 * sure that the target isn't hidden. If it is, we don't
394 * show the link in the directory listing.
f7e8960c
AH
395 */
396 if ((node->node_type == Plink) && (node->link_target != NULL) &&
397 (node->link_target->flags & DEVFS_HIDDEN))
21864bc5 398 continue;
f7e8960c
AH
399
400 if (node->cookie < saveoff)
21864bc5 401 continue;
f7e8960c 402
21864bc5
MD
403 saveoff = node->cookie;
404
898c91ee
MD
405 error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
406 node->d_dir.d_type,
407 node->d_dir.d_namlen,
408 node->d_dir.d_name);
21864bc5 409
898c91ee 410 if (error2)
21864bc5
MD
411 break;
412
413 saveoff++;
414
415 if (cookies)
416 cookies[cookie_index] = node->cookie;
417 ++cookie_index;
418 if (cookie_index == ncookies)
419 break;
21864bc5
MD
420 }
421
422done:
ca8d7677 423 lockmgr(&devfs_lock, LK_RELEASE);
21864bc5
MD
424 vn_unlock(ap->a_vp);
425
426 ap->a_uio->uio_offset = saveoff;
427 if (error && cookie_index == 0) {
428 if (cookies) {
429 kfree(cookies, M_TEMP);
430 *ap->a_ncookies = 0;
431 *ap->a_cookies = NULL;
432 }
433 } else {
434 if (cookies) {
435 *ap->a_ncookies = cookie_index;
436 *ap->a_cookies = cookies;
437 }
438 }
439 return (error);
440}
441
442
443static int
9f889dc4 444devfs_vop_nresolve(struct vop_nresolve_args *ap)
21864bc5 445{
898c91ee 446 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
21864bc5
MD
447 struct devfs_node *node, *found = NULL;
448 struct namecache *ncp;
449 struct vnode *vp = NULL;
21864bc5
MD
450 int error = 0;
451 int len;
260e4e8b 452 int depth;
21864bc5 453
21864bc5
MD
454 ncp = ap->a_nch->ncp;
455 len = ncp->nc_nlen;
456
898c91ee 457 if (!devfs_node_is_accessible(dnode))
ca8d7677
MD
458 return ENOENT;
459
21864bc5
MD
460 lockmgr(&devfs_lock, LK_EXCLUSIVE);
461
898c91ee 462 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) {
e23485a5 463 error = ENOENT;
21864bc5
MD
464 cache_setvp(ap->a_nch, NULL);
465 goto out;
466 }
467
898c91ee 468 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
21864bc5
MD
469 if (len == node->d_dir.d_namlen) {
470 if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
21864bc5
MD
471 found = node;
472 break;
473 }
474 }
475 }
476
477 if (found) {
260e4e8b
AH
478 depth = 0;
479 while ((found->node_type == Plink) && (found->link_target)) {
480 if (depth >= 8) {
481 devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
482 break;
483 }
484
21864bc5 485 found = found->link_target;
260e4e8b
AH
486 ++depth;
487 }
21864bc5
MD
488
489 if (!(found->flags & DEVFS_HIDDEN))
490 devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
21864bc5
MD
491 }
492
21864bc5 493 if (vp == NULL) {
21864bc5
MD
494 error = ENOENT;
495 cache_setvp(ap->a_nch, NULL);
21864bc5
MD
496 goto out;
497
498 }
21864bc5
MD
499 KKASSERT(vp);
500 vn_unlock(vp);
501 cache_setvp(ap->a_nch, vp);
502 vrele(vp);
21864bc5 503out:
21864bc5 504 lockmgr(&devfs_lock, LK_RELEASE);
898c91ee 505
21864bc5
MD
506 return error;
507}
508
509
510static int
9f889dc4 511devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
21864bc5 512{
898c91ee 513 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
21864bc5 514
898c91ee
MD
515 *ap->a_vpp = NULL;
516 if (!devfs_node_is_accessible(dnode))
894bbb25
AH
517 return ENOENT;
518
21864bc5 519 lockmgr(&devfs_lock, LK_EXCLUSIVE);
898c91ee
MD
520 if (dnode->parent != NULL) {
521 devfs_allocv(ap->a_vpp, dnode->parent);
21864bc5
MD
522 vn_unlock(*ap->a_vpp);
523 }
524 lockmgr(&devfs_lock, LK_RELEASE);
525
526 return ((*ap->a_vpp == NULL) ? ENOENT : 0);
527}
528
529
530static int
9f889dc4 531devfs_vop_getattr(struct vop_getattr_args *ap)
21864bc5 532{
21864bc5 533 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
898c91ee 534 struct vattr *vap = ap->a_vap;
2d076755 535 struct partinfo pinfo;
21864bc5
MD
536 int error = 0;
537
952f0188 538#if 0
894bbb25 539 if (!devfs_node_is_accessible(node))
ca8d7677 540 return ENOENT;
952f0188 541#endif
4062d050 542 node_sync_dev_get(node);
ca8d7677 543
ca8d7677 544 lockmgr(&devfs_lock, LK_EXCLUSIVE);
21864bc5
MD
545
546 /* start by zeroing out the attributes */
547 VATTR_NULL(vap);
548
549 /* next do all the common fields */
550 vap->va_type = ap->a_vp->v_type;
551 vap->va_mode = node->mode;
552 vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
a3c7ebb9 553 vap->va_flags = 0;
21864bc5 554 vap->va_blocksize = DEV_BSIZE;
a3c7ebb9 555 vap->va_bytes = vap->va_size = 0;
21864bc5 556
21864bc5
MD
557 vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
558
21864bc5
MD
559 vap->va_atime = node->atime;
560 vap->va_mtime = node->mtime;
561 vap->va_ctime = node->ctime;
562
563 vap->va_nlink = 1; /* number of references to file */
564
565 vap->va_uid = node->uid;
566 vap->va_gid = node->gid;
567
568 vap->va_rmajor = 0;
569 vap->va_rminor = 0;
570
898c91ee
MD
571 if ((node->node_type == Pdev) && node->d_dev) {
572 reference_dev(node->d_dev);
573 vap->va_rminor = node->d_dev->si_uminor;
574 release_dev(node->d_dev);
21864bc5
MD
575 }
576
577 /* For a softlink the va_size is the length of the softlink */
898c91ee 578 if (node->symlink_name != 0) {
2d076755 579 vap->va_bytes = vap->va_size = node->symlink_namelen;
21864bc5 580 }
2d076755
AH
581
582 /*
583 * For a disk-type device, va_size is the size of the underlying
584 * device, so that lseek() works properly.
585 */
586 if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
587 bzero(&pinfo, sizeof(pinfo));
588 error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
589 0, proc0.p_ucred, NULL);
590 if ((error == 0) && (pinfo.media_blksize != 0)) {
591 vap->va_size = pinfo.media_size;
592 } else {
593 vap->va_size = 0;
594 error = 0;
595 }
596 }
597
ca8d7677 598 lockmgr(&devfs_lock, LK_RELEASE);
898c91ee 599
894bbb25 600 return (error);
21864bc5
MD
601}
602
603
604static int
9f889dc4 605devfs_vop_setattr(struct vop_setattr_args *ap)
21864bc5 606{
898c91ee 607 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
21864bc5 608 struct vattr *vap;
dffaed1b
AH
609 uid_t cur_uid;
610 gid_t cur_gid;
611 mode_t cur_mode;
21864bc5
MD
612 int error = 0;
613
894bbb25 614 if (!devfs_node_is_accessible(node))
ca8d7677 615 return ENOENT;
4062d050 616 node_sync_dev_get(node);
21864bc5 617
21864bc5
MD
618 lockmgr(&devfs_lock, LK_EXCLUSIVE);
619
620 vap = ap->a_vap;
21864bc5 621
dffaed1b
AH
622 if ((vap->va_uid != (uid_t)VNOVAL) || (vap->va_gid != (gid_t)VNOVAL)) {
623 cur_uid = node->uid;
624 cur_gid = node->gid;
625 cur_mode = node->mode;
626 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
627 ap->a_cred, &cur_uid, &cur_gid, &cur_mode);
628 if (error)
629 goto out;
630
631 if (node->uid != cur_uid || node->gid != cur_gid) {
632 node->uid = cur_uid;
633 node->gid = cur_gid;
634 node->mode = cur_mode;
21864bc5 635 }
21864bc5
MD
636 }
637
638 if (vap->va_mode != (mode_t)VNOVAL) {
dffaed1b
AH
639 cur_mode = node->mode;
640 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
641 node->uid, node->gid, &cur_mode);
642 if (error == 0 && node->mode != cur_mode) {
643 node->mode = cur_mode;
21864bc5 644 }
21864bc5
MD
645 }
646
647out:
4062d050 648 node_sync_dev_set(node);
07dfa375 649 nanotime(&node->ctime);
21864bc5 650 lockmgr(&devfs_lock, LK_RELEASE);
898c91ee 651
21864bc5
MD
652 return error;
653}
654
655
656static int
9f889dc4 657devfs_vop_readlink(struct vop_readlink_args *ap)
21864bc5
MD
658{
659 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
ca8d7677
MD
660 int ret;
661
894bbb25 662 if (!devfs_node_is_accessible(node))
ca8d7677 663 return ENOENT;
21864bc5 664
ca8d7677
MD
665 lockmgr(&devfs_lock, LK_EXCLUSIVE);
666 ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
667 lockmgr(&devfs_lock, LK_RELEASE);
668
669 return ret;
21864bc5
MD
670}
671
672
673static int
9f889dc4 674devfs_vop_print(struct vop_print_args *ap)
21864bc5 675{
21864bc5
MD
676 return (0);
677}
678
39a08947 679static int
9f889dc4 680devfs_vop_nmkdir(struct vop_nmkdir_args *ap)
39a08947
AH
681{
682 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
683 struct devfs_node *node;
684
685 if (!devfs_node_is_accessible(dnode))
686 return ENOENT;
687
688 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
689 goto out;
690
691 lockmgr(&devfs_lock, LK_EXCLUSIVE);
692 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Pdir,
693 ap->a_nch->ncp->nc_name, dnode, NULL);
694
695 if (*ap->a_vpp) {
696 node = DEVFS_NODE(*ap->a_vpp);
697 node->flags |= DEVFS_USER_CREATED;
698 cache_setunresolved(ap->a_nch);
699 cache_setvp(ap->a_nch, *ap->a_vpp);
700 }
701 lockmgr(&devfs_lock, LK_RELEASE);
702out:
703 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
704}
21864bc5
MD
705
706static int
9f889dc4 707devfs_vop_nsymlink(struct vop_nsymlink_args *ap)
21864bc5 708{
898c91ee
MD
709 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
710 struct devfs_node *node;
711 size_t targetlen;
21864bc5 712
898c91ee 713 if (!devfs_node_is_accessible(dnode))
ca8d7677
MD
714 return ENOENT;
715
894bbb25
AH
716 ap->a_vap->va_type = VLNK;
717
898c91ee 718 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
21864bc5 719 goto out;
898c91ee 720
21864bc5
MD
721 lockmgr(&devfs_lock, LK_EXCLUSIVE);
722 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Plink,
898c91ee 723 ap->a_nch->ncp->nc_name, dnode, NULL);
21864bc5 724
898c91ee 725 targetlen = strlen(ap->a_target);
21864bc5 726 if (*ap->a_vpp) {
898c91ee
MD
727 node = DEVFS_NODE(*ap->a_vpp);
728 node->flags |= DEVFS_USER_CREATED;
729 node->symlink_namelen = targetlen;
730 node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
731 memcpy(node->symlink_name, ap->a_target, targetlen);
732 node->symlink_name[targetlen] = '\0';
21864bc5 733 cache_setunresolved(ap->a_nch);
21864bc5
MD
734 cache_setvp(ap->a_nch, *ap->a_vpp);
735 }
736 lockmgr(&devfs_lock, LK_RELEASE);
737out:
738 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
21864bc5
MD
739}
740
39a08947 741static int
9f889dc4 742devfs_vop_nrmdir(struct vop_nrmdir_args *ap)
39a08947
AH
743{
744 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
745 struct devfs_node *node;
746 struct namecache *ncp;
747 int error = ENOENT;
748
749 ncp = ap->a_nch->ncp;
750
751 if (!devfs_node_is_accessible(dnode))
752 return ENOENT;
753
754 lockmgr(&devfs_lock, LK_EXCLUSIVE);
755
756 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
757 goto out;
758
759 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
760 if (ncp->nc_nlen != node->d_dir.d_namlen)
761 continue;
762 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
763 continue;
764
765 /*
766 * only allow removal of user created dirs
767 */
768 if ((node->flags & DEVFS_USER_CREATED) == 0) {
769 error = EPERM;
770 goto out;
771 } else if (node->node_type != Pdir) {
772 error = ENOTDIR;
773 goto out;
774 } else if (node->nchildren > 2) {
775 error = ENOTEMPTY;
776 goto out;
777 } else {
778 if (node->v_node)
779 cache_inval_vp(node->v_node, CINV_DESTROY);
780 devfs_unlinkp(node);
781 error = 0;
782 break;
783 }
784 }
785
786 cache_setunresolved(ap->a_nch);
787 cache_setvp(ap->a_nch, NULL);
788
789out:
790 lockmgr(&devfs_lock, LK_RELEASE);
791 return error;
792}
21864bc5
MD
793
794static int
9f889dc4 795devfs_vop_nremove(struct vop_nremove_args *ap)
21864bc5 796{
898c91ee 797 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
21864bc5
MD
798 struct devfs_node *node;
799 struct namecache *ncp;
21864bc5
MD
800 int error = ENOENT;
801
21864bc5
MD
802 ncp = ap->a_nch->ncp;
803
898c91ee 804 if (!devfs_node_is_accessible(dnode))
ca8d7677
MD
805 return ENOENT;
806
21864bc5
MD
807 lockmgr(&devfs_lock, LK_EXCLUSIVE);
808
898c91ee 809 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
21864bc5 810 goto out;
21864bc5 811
898c91ee
MD
812 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
813 if (ncp->nc_nlen != node->d_dir.d_namlen)
814 continue;
815 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
816 continue;
21864bc5 817
898c91ee
MD
818 /*
819 * only allow removal of user created stuff (e.g. symlinks)
820 */
821 if ((node->flags & DEVFS_USER_CREATED) == 0) {
822 error = EPERM;
823 goto out;
39a08947
AH
824 } else if (node->node_type == Pdir) {
825 error = EISDIR;
826 goto out;
898c91ee
MD
827 } else {
828 if (node->v_node)
829 cache_inval_vp(node->v_node, CINV_DESTROY);
830 devfs_unlinkp(node);
831 error = 0;
832 break;
21864bc5
MD
833 }
834 }
835
836 cache_setunresolved(ap->a_nch);
837 cache_setvp(ap->a_nch, NULL);
21864bc5
MD
838
839out:
840 lockmgr(&devfs_lock, LK_RELEASE);
21864bc5
MD
841 return error;
842}
843
844
845static int
846devfs_spec_open(struct vop_open_args *ap)
847{
848 struct vnode *vp = ap->a_vp;
ca8d7677 849 struct vnode *orig_vp = NULL;
898c91ee
MD
850 struct devfs_node *node = DEVFS_NODE(vp);
851 struct devfs_node *newnode;
21864bc5 852 cdev_t dev, ndev = NULL;
21864bc5 853 int error = 0;
21864bc5 854
898c91ee
MD
855 if (node) {
856 if (node->d_dev == NULL)
21864bc5 857 return ENXIO;
898c91ee 858 if (!devfs_node_is_accessible(node))
894bbb25 859 return ENOENT;
21864bc5
MD
860 }
861
21864bc5
MD
862 if ((dev = vp->v_rdev) == NULL)
863 return ENXIO;
864
898c91ee 865 if (node && ap->a_fp) {
21864bc5
MD
866 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
867 lockmgr(&devfs_lock, LK_EXCLUSIVE);
07dfa375
AH
868
869 ndev = devfs_clone(dev, node->d_dir.d_name, node->d_dir.d_namlen,
870 ap->a_mode, ap->a_cred);
871 if (ndev != NULL) {
898c91ee
MD
872 newnode = devfs_create_device_node(
873 DEVFS_MNTDATA(vp->v_mount)->root_node,
07dfa375
AH
874 ndev, NULL, NULL);
875 /* XXX: possibly destroy device if this happens */
898c91ee 876
07dfa375
AH
877 if (newnode != NULL) {
878 dev = ndev;
879 devfs_link_dev(dev);
21864bc5 880
07dfa375
AH
881 devfs_debug(DEVFS_DEBUG_DEBUG,
882 "parent here is: %s, node is: |%s|\n",
883 ((node->parent->node_type == Proot) ?
884 "ROOT!" : node->parent->d_dir.d_name),
885 newnode->d_dir.d_name);
886 devfs_debug(DEVFS_DEBUG_DEBUG,
887 "test: %s\n",
888 ((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
889
890 /*
891 * orig_vp is set to the original vp if we cloned.
892 */
893 /* node->flags |= DEVFS_CLONED; */
894 devfs_allocv(&vp, newnode);
895 orig_vp = ap->a_vp;
896 ap->a_vp = vp;
897 }
21864bc5
MD
898 }
899 lockmgr(&devfs_lock, LK_RELEASE);
900 }
901
898c91ee
MD
902 devfs_debug(DEVFS_DEBUG_DEBUG,
903 "devfs_spec_open() called on %s! \n",
904 dev->si_name);
905
21864bc5
MD
906 /*
907 * Make this field valid before any I/O in ->d_open
908 */
909 if (!dev->si_iosize_max)
910 dev->si_iosize_max = DFLTPHYS;
911
912 if (dev_dflags(dev) & D_TTY)
2247fe02 913 vsetflags(vp, VISTTY);
21864bc5 914
9f889dc4
MD
915 /*
916 * Open underlying device
917 */
21864bc5
MD
918 vn_unlock(vp);
919 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred);
920 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
921
ca8d7677
MD
922 /*
923 * Clean up any cloned vp if we error out.
924 */
21864bc5 925 if (error) {
ca8d7677 926 if (orig_vp) {
21864bc5 927 vput(vp);
ca8d7677
MD
928 ap->a_vp = orig_vp;
929 /* orig_vp = NULL; */
930 }
21864bc5
MD
931 return error;
932 }
933
d894b0eb 934 /*
ddd7de82
AH
935 * This checks if the disk device is going to be opened for writing.
936 * It will be only allowed in the cases where securelevel permits it
937 * and it's not mounted R/W.
d894b0eb 938 */
ddd7de82
AH
939 if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
940 (ap->a_cred != FSCRED)) {
941
942 /* Very secure mode. No open for writing allowed */
943 if (securelevel >= 2)
944 return EPERM;
945
946 /*
947 * If it is mounted R/W, do not allow to open for writing.
948 * In the case it's mounted read-only but securelevel
949 * is >= 1, then do not allow opening for writing either.
950 */
951 if (vfs_mountedon(vp)) {
952 if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
953 return EBUSY;
954 else if (securelevel >= 1)
d894b0eb
AH
955 return EPERM;
956 }
957 }
21864bc5
MD
958
959 if (dev_dflags(dev) & D_TTY) {
960 if (dev->si_tty) {
961 struct tty *tp;
962 tp = dev->si_tty;
963 if (!tp->t_stop) {
898c91ee
MD
964 devfs_debug(DEVFS_DEBUG_DEBUG,
965 "devfs: no t_stop\n");
21864bc5
MD
966 tp->t_stop = nottystop;
967 }
968 }
969 }
970
971
972 if (vn_isdisk(vp, NULL)) {
973 if (!dev->si_bsize_phys)
974 dev->si_bsize_phys = DEV_BSIZE;
b0d18f7d 975 vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
21864bc5
MD
976 }
977
978 vop_stdopen(ap);
07dfa375 979#if 0
898c91ee
MD
980 if (node)
981 nanotime(&node->atime);
07dfa375 982#endif
21864bc5 983
b80b4c32
MD
984 /*
985 * If we replaced the vp the vop_stdopen() call will have loaded
986 * it into fp->f_data and vref()d the vp, giving us two refs. So
987 * instead of just unlocking it here we have to vput() it.
988 */
ca8d7677 989 if (orig_vp)
b80b4c32 990 vput(vp);
21864bc5
MD
991
992 /* Ugly pty magic, to make pty devices appear once they are opened */
898c91ee
MD
993 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
994 node->flags &= ~DEVFS_INVISIBLE;
21864bc5
MD
995
996 if (ap->a_fp) {
b80b4c32 997 KKASSERT(ap->a_fp->f_type == DTYPE_VNODE);
fb12e9a5 998 KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK));
21864bc5 999 ap->a_fp->f_ops = &devfs_dev_fileops;
b80b4c32 1000 KKASSERT(ap->a_fp->f_data == (void *)vp);
21864bc5
MD
1001 }
1002
21864bc5
MD
1003 return 0;
1004}
1005
1006
1007static int
1008devfs_spec_close(struct vop_close_args *ap)
1009{
92fb0c6a 1010 struct devfs_node *node;
21864bc5
MD
1011 struct proc *p = curproc;
1012 struct vnode *vp = ap->a_vp;
1013 cdev_t dev = vp->v_rdev;
1014 int error = 0;
1015 int needrelock;
1016
52f98df9
SG
1017 if (dev)
1018 devfs_debug(DEVFS_DEBUG_DEBUG,
1019 "devfs_spec_close() called on %s! \n",
1020 dev->si_name);
1021 else
1022 devfs_debug(DEVFS_DEBUG_DEBUG,
1023 "devfs_spec_close() called, null vode!\n");
21864bc5
MD
1024
1025 /*
1026 * A couple of hacks for devices and tty devices. The
1027 * vnode ref count cannot be used to figure out the
1028 * last close, but we can use v_opencount now that
1029 * revoke works properly.
1030 *
1031 * Detect the last close on a controlling terminal and clear
1032 * the session (half-close).
1033 */
1034 if (dev)
1035 reference_dev(dev);
1036
1037 if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1038 p->p_session->s_ttyvp = NULL;
1039 vrele(vp);
1040 }
1041
1042 /*
1043 * Vnodes can be opened and closed multiple times. Do not really
1044 * close the device unless (1) it is being closed forcibly,
1045 * (2) the device wants to track closes, or (3) this is the last
1046 * vnode doing its last close on the device.
1047 *
1048 * XXX the VXLOCK (force close) case can leave vnodes referencing
1049 * a closed device. This might not occur now that our revoke is
1050 * fixed.
1051 */
1052 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1053 if (dev && ((vp->v_flag & VRECLAIMED) ||
1054 (dev_dflags(dev) & D_TRACKCLOSE) ||
1055 (vp->v_opencount == 1))) {
898c91ee 1056 /*
92fb0c6a
MD
1057 * Ugly pty magic, to make pty devices disappear again once
1058 * they are closed.
1059 */
1060 node = DEVFS_NODE(ap->a_vp);
1061 if (node && (node->flags & DEVFS_PTY))
1062 node->flags |= DEVFS_INVISIBLE;
1063
1064 /*
495d3a1e
MD
1065 * Unlock around dev_dclose(), unless the vnode is
1066 * undergoing a vgone/reclaim (during umount).
898c91ee 1067 */
21864bc5 1068 needrelock = 0;
495d3a1e 1069 if ((vp->v_flag & VRECLAIMED) == 0 && vn_islocked(vp)) {
21864bc5
MD
1070 needrelock = 1;
1071 vn_unlock(vp);
1072 }
898c91ee
MD
1073
1074 /*
92fb0c6a
MD
1075 * WARNING! If the device destroys itself the devfs node
1076 * can disappear here.
495d3a1e
MD
1077 *
1078 * WARNING! vn_lock() will fail if the vp is in a VRECLAIM,
1079 * which can occur during umount.
898c91ee 1080 */
92fb0c6a
MD
1081 error = dev_dclose(dev, ap->a_fflag, S_IFCHR);
1082 /* node is now stale */
21864bc5 1083
495d3a1e
MD
1084 if (needrelock) {
1085 if (vn_lock(vp, LK_EXCLUSIVE | LK_RETRY) != 0) {
1086 panic("devfs_spec_close: vnode %p "
1087 "unexpectedly could not be relocked",
1088 vp);
1089 }
1090 }
21864bc5
MD
1091 } else {
1092 error = 0;
1093 }
1094 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
898c91ee 1095
21864bc5
MD
1096 /*
1097 * Track the actual opens and closes on the vnode. The last close
898c91ee
MD
1098 * disassociates the rdev. If the rdev is already disassociated or
1099 * the opencount is already 0, the vnode might have been revoked
1100 * and no further opencount tracking occurs.
21864bc5 1101 */
898c91ee 1102 if (dev)
21864bc5 1103 release_dev(dev);
898c91ee 1104 if (vp->v_opencount > 0)
21864bc5 1105 vop_stdclose(ap);
21864bc5
MD
1106 return(error);
1107
1108}
1109
1110
1111static int
9f889dc4 1112devfs_fo_close(struct file *fp)
21864bc5 1113{
21864bc5 1114 struct vnode *vp = (struct vnode *)fp->f_data;
898c91ee 1115 int error;
21864bc5 1116
21864bc5 1117 fp->f_ops = &badfileops;
21864bc5 1118 error = vn_close(vp, fp->f_flag);
21864bc5
MD
1119
1120 return (error);
1121}
1122
1123
1124/*
1125 * Device-optimized file table vnode read routine.
1126 *
1127 * This bypasses the VOP table and talks directly to the device. Most
1128 * filesystems just route to specfs and can make this optimization.
1129 *
1130 * MPALMOSTSAFE - acquires mplock
1131 */
1132static int
9f889dc4 1133devfs_fo_read(struct file *fp, struct uio *uio,
898c91ee 1134 struct ucred *cred, int flags)
21864bc5 1135{
898c91ee 1136 struct devfs_node *node;
21864bc5
MD
1137 struct vnode *vp;
1138 int ioflag;
1139 int error;
1140 cdev_t dev;
1141
21864bc5
MD
1142 KASSERT(uio->uio_td == curthread,
1143 ("uio_td %p is not td %p", uio->uio_td, curthread));
1144
3a1032a6
AH
1145 if (uio->uio_resid == 0)
1146 return 0;
1147
21864bc5 1148 vp = (struct vnode *)fp->f_data;
3a1032a6
AH
1149 if (vp == NULL || vp->v_type == VBAD)
1150 return EBADF;
1151
898c91ee 1152 node = DEVFS_NODE(vp);
21864bc5 1153
3a1032a6
AH
1154 if ((dev = vp->v_rdev) == NULL)
1155 return EBADF;
1156
21864bc5
MD
1157 reference_dev(dev);
1158
21864bc5
MD
1159 if ((flags & O_FOFFSET) == 0)
1160 uio->uio_offset = fp->f_offset;
1161
1162 ioflag = 0;
1163 if (flags & O_FBLOCKING) {
1164 /* ioflag &= ~IO_NDELAY; */
1165 } else if (flags & O_FNONBLOCKING) {
1166 ioflag |= IO_NDELAY;
1167 } else if (fp->f_flag & FNONBLOCK) {
1168 ioflag |= IO_NDELAY;
1169 }
1170 if (flags & O_FBUFFERED) {
1171 /* ioflag &= ~IO_DIRECT; */
1172 } else if (flags & O_FUNBUFFERED) {
1173 ioflag |= IO_DIRECT;
1174 } else if (fp->f_flag & O_DIRECT) {
1175 ioflag |= IO_DIRECT;
1176 }
1177 ioflag |= sequential_heuristic(uio, fp);
1178
1179 error = dev_dread(dev, uio, ioflag);
1180
1181 release_dev(dev);
898c91ee
MD
1182 if (node)
1183 nanotime(&node->atime);
21864bc5
MD
1184 if ((flags & O_FOFFSET) == 0)
1185 fp->f_offset = uio->uio_offset;
1186 fp->f_nextoff = uio->uio_offset;
3a1032a6 1187
21864bc5
MD
1188 return (error);
1189}
1190
1191
1192static int
9f889dc4 1193devfs_fo_write(struct file *fp, struct uio *uio,
898c91ee 1194 struct ucred *cred, int flags)
21864bc5 1195{
898c91ee 1196 struct devfs_node *node;
21864bc5
MD
1197 struct vnode *vp;
1198 int ioflag;
1199 int error;
1200 cdev_t dev;
1201
21864bc5
MD
1202 KASSERT(uio->uio_td == curthread,
1203 ("uio_td %p is not p %p", uio->uio_td, curthread));
1204
1205 vp = (struct vnode *)fp->f_data;
3a1032a6
AH
1206 if (vp == NULL || vp->v_type == VBAD)
1207 return EBADF;
1208
898c91ee 1209 node = DEVFS_NODE(vp);
3a1032a6 1210
21864bc5
MD
1211 if (vp->v_type == VREG)
1212 bwillwrite(uio->uio_resid);
3a1032a6 1213
21864bc5
MD
1214 vp = (struct vnode *)fp->f_data;
1215
3a1032a6
AH
1216 if ((dev = vp->v_rdev) == NULL)
1217 return EBADF;
1218
21864bc5
MD
1219 reference_dev(dev);
1220
1221 if ((flags & O_FOFFSET) == 0)
1222 uio->uio_offset = fp->f_offset;
1223
1224 ioflag = IO_UNIT;
1225 if (vp->v_type == VREG &&
1226 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1227 ioflag |= IO_APPEND;
1228 }
1229
1230 if (flags & O_FBLOCKING) {
1231 /* ioflag &= ~IO_NDELAY; */
1232 } else if (flags & O_FNONBLOCKING) {
1233 ioflag |= IO_NDELAY;
1234 } else if (fp->f_flag & FNONBLOCK) {
1235 ioflag |= IO_NDELAY;
1236 }
1237 if (flags & O_FBUFFERED) {
1238 /* ioflag &= ~IO_DIRECT; */
1239 } else if (flags & O_FUNBUFFERED) {
1240 ioflag |= IO_DIRECT;
1241 } else if (fp->f_flag & O_DIRECT) {
1242 ioflag |= IO_DIRECT;
1243 }
1244 if (flags & O_FASYNCWRITE) {
1245 /* ioflag &= ~IO_SYNC; */
1246 } else if (flags & O_FSYNCWRITE) {
1247 ioflag |= IO_SYNC;
1248 } else if (fp->f_flag & O_FSYNC) {
1249 ioflag |= IO_SYNC;
1250 }
1251
1252 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1253 ioflag |= IO_SYNC;
1254 ioflag |= sequential_heuristic(uio, fp);
1255
1256 error = dev_dwrite(dev, uio, ioflag);
1257
1258 release_dev(dev);
07dfa375
AH
1259 if (node) {
1260 nanotime(&node->atime);
898c91ee 1261 nanotime(&node->mtime);
07dfa375 1262 }
21864bc5
MD
1263
1264 if ((flags & O_FOFFSET) == 0)
1265 fp->f_offset = uio->uio_offset;
1266 fp->f_nextoff = uio->uio_offset;
3a1032a6 1267
21864bc5
MD
1268 return (error);
1269}
1270
1271
1272static int
9f889dc4 1273devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
21864bc5
MD
1274{
1275 struct vnode *vp;
3a1032a6
AH
1276 struct vattr vattr;
1277 struct vattr *vap;
1278 u_short mode;
1279 cdev_t dev;
21864bc5
MD
1280 int error;
1281
21864bc5 1282 vp = (struct vnode *)fp->f_data;
3a1032a6
AH
1283 if (vp == NULL || vp->v_type == VBAD)
1284 return EBADF;
1285
21864bc5 1286 error = vn_stat(vp, sb, cred);
3a1032a6 1287 if (error)
21864bc5 1288 return (error);
21864bc5
MD
1289
1290 vap = &vattr;
1291 error = VOP_GETATTR(vp, vap);
3a1032a6 1292 if (error)
21864bc5 1293 return (error);
21864bc5
MD
1294
1295 /*
1296 * Zero the spare stat fields
1297 */
1298 sb->st_lspare = 0;
d98152a8
MD
1299 sb->st_qspare1 = 0;
1300 sb->st_qspare2 = 0;
21864bc5
MD
1301
1302 /*
1303 * Copy from vattr table ... or not in case it's a cloned device
1304 */
1305 if (vap->va_fsid != VNOVAL)
1306 sb->st_dev = vap->va_fsid;
1307 else
1308 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1309
1310 sb->st_ino = vap->va_fileid;
1311
1312 mode = vap->va_mode;
1313 mode |= S_IFCHR;
1314 sb->st_mode = mode;
1315
1316 if (vap->va_nlink > (nlink_t)-1)
1317 sb->st_nlink = (nlink_t)-1;
1318 else
1319 sb->st_nlink = vap->va_nlink;
3a1032a6 1320
21864bc5
MD
1321 sb->st_uid = vap->va_uid;
1322 sb->st_gid = vap->va_gid;
ca8d7677 1323 sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev);
2d076755 1324 sb->st_size = vap->va_bytes;
21864bc5
MD
1325 sb->st_atimespec = vap->va_atime;
1326 sb->st_mtimespec = vap->va_mtime;
1327 sb->st_ctimespec = vap->va_ctime;
1328
1329 /*
1330 * A VCHR and VBLK device may track the last access and last modified
1331 * time independantly of the filesystem. This is particularly true
1332 * because device read and write calls may bypass the filesystem.
1333 */
1334 if (vp->v_type == VCHR || vp->v_type == VBLK) {
1335 dev = vp->v_rdev;
1336 if (dev != NULL) {
1337 if (dev->si_lastread) {
1338 sb->st_atimespec.tv_sec = dev->si_lastread;
1339 sb->st_atimespec.tv_nsec = 0;
1340 }
1341 if (dev->si_lastwrite) {
1342 sb->st_atimespec.tv_sec = dev->si_lastwrite;
1343 sb->st_atimespec.tv_nsec = 0;
1344 }
1345 }
1346 }
1347
1348 /*
1349 * According to www.opengroup.org, the meaning of st_blksize is
1350 * "a filesystem-specific preferred I/O block size for this
1351 * object. In some filesystem types, this may vary from file
1352 * to file"
1353 * Default to PAGE_SIZE after much discussion.
1354 */
1355
1356 sb->st_blksize = PAGE_SIZE;
1357
1358 sb->st_flags = vap->va_flags;
1359
1360 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1361 if (error)
1362 sb->st_gen = 0;
1363 else
1364 sb->st_gen = (u_int32_t)vap->va_gen;
1365
1366 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
21864bc5 1367
21864bc5
MD
1368 return (0);
1369}
1370
1371
1372static int
9f889dc4 1373devfs_fo_kqfilter(struct file *fp, struct knote *kn)
21864bc5
MD
1374{
1375 struct vnode *vp;
21864bc5
MD
1376 int error;
1377 cdev_t dev;
1378
21864bc5
MD
1379 vp = (struct vnode *)fp->f_data;
1380 if (vp == NULL || vp->v_type == VBAD) {
1381 error = EBADF;
1382 goto done;
1383 }
21864bc5
MD
1384 if ((dev = vp->v_rdev) == NULL) {
1385 error = EBADF;
1386 goto done;
1387 }
1388 reference_dev(dev);
1389
1390 error = dev_dkqfilter(dev, kn);
1391
1392 release_dev(dev);
1393
21864bc5 1394done:
b287d649 1395 return (error);
21864bc5
MD
1396}
1397
21864bc5
MD
1398/*
1399 * MPALMOSTSAFE - acquires mplock
1400 */
1401static int
9f889dc4 1402devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data,
87baaf0c 1403 struct ucred *ucred, struct sysmsg *msg)
21864bc5 1404{
898c91ee
MD
1405 struct devfs_node *node;
1406 struct vnode *vp;
21864bc5 1407 struct vnode *ovp;
21864bc5
MD
1408 cdev_t dev;
1409 int error;
1410 struct fiodname_args *name_args;
1411 size_t namlen;
1412 const char *name;
1413
898c91ee 1414 vp = ((struct vnode *)fp->f_data);
3a1032a6
AH
1415
1416 if ((dev = vp->v_rdev) == NULL)
1417 return EBADF; /* device was revoked */
1418
1419 reference_dev(dev);
21864bc5 1420
898c91ee
MD
1421 node = DEVFS_NODE(vp);
1422
1423 devfs_debug(DEVFS_DEBUG_DEBUG,
9f889dc4 1424 "devfs_fo_ioctl() called! for dev %s\n",
898c91ee 1425 dev->si_name);
21864bc5
MD
1426
1427 if (com == FIODTYPE) {
1428 *(int *)data = dev_dflags(dev) & D_TYPEMASK;
1429 error = 0;
1430 goto out;
1431 } else if (com == FIODNAME) {
1432 name_args = (struct fiodname_args *)data;
1433 name = dev->si_name;
1434 namlen = strlen(name) + 1;
1435
898c91ee
MD
1436 devfs_debug(DEVFS_DEBUG_DEBUG,
1437 "ioctl, got: FIODNAME for %s\n", name);
21864bc5
MD
1438
1439 if (namlen <= name_args->len)
1440 error = copyout(dev->si_name, name_args->name, namlen);
1441 else
1442 error = EINVAL;
1443
898c91ee
MD
1444 devfs_debug(DEVFS_DEBUG_DEBUG,
1445 "ioctl stuff: error: %d\n", error);
21864bc5
MD
1446 goto out;
1447 }
3a1032a6 1448
87baaf0c 1449 error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg);
3a1032a6 1450
07dfa375 1451#if 0
898c91ee
MD
1452 if (node) {
1453 nanotime(&node->atime);
1454 nanotime(&node->mtime);
21864bc5 1455 }
07dfa375 1456#endif
898c91ee
MD
1457 if (com == TIOCSCTTY) {
1458 devfs_debug(DEVFS_DEBUG_DEBUG,
9f889dc4 1459 "devfs_fo_ioctl: got TIOCSCTTY on %s\n",
898c91ee
MD
1460 dev->si_name);
1461 }
21864bc5 1462 if (error == 0 && com == TIOCSCTTY) {
21864bc5
MD
1463 struct proc *p = curthread->td_proc;
1464 struct session *sess;
898c91ee
MD
1465
1466 devfs_debug(DEVFS_DEBUG_DEBUG,
9f889dc4 1467 "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n",
898c91ee
MD
1468 dev->si_name);
1469 if (p == NULL) {
21864bc5
MD
1470 error = ENOTTY;
1471 goto out;
1472 }
1473 sess = p->p_session;
898c91ee
MD
1474
1475 /*
1476 * Do nothing if reassigning same control tty
1477 */
21864bc5
MD
1478 if (sess->s_ttyvp == vp) {
1479 error = 0;
1480 goto out;
1481 }
898c91ee
MD
1482
1483 /*
1484 * Get rid of reference to old control tty
1485 */
21864bc5
MD
1486 ovp = sess->s_ttyvp;
1487 vref(vp);
1488 sess->s_ttyvp = vp;
1489 if (ovp)
1490 vrele(ovp);
1491 }
1492
1493out:
3a1032a6 1494 release_dev(dev);
9f889dc4 1495 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n");
21864bc5
MD
1496 return (error);
1497}
1498
1499
1500static int
1501devfs_spec_fsync(struct vop_fsync_args *ap)
1502{
1503 struct vnode *vp = ap->a_vp;
1504 int error;
1505
1506 if (!vn_isdisk(vp, NULL))
1507 return (0);
1508
1509 /*
1510 * Flush all dirty buffers associated with a block device.
1511 */
1512 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1513 return (error);
1514}
1515
21864bc5
MD
1516static int
1517devfs_spec_read(struct vop_read_args *ap)
1518{
898c91ee 1519 struct devfs_node *node;
21864bc5
MD
1520 struct vnode *vp;
1521 struct uio *uio;
1522 cdev_t dev;
1523 int error;
1524
1525 vp = ap->a_vp;
1526 dev = vp->v_rdev;
1527 uio = ap->a_uio;
898c91ee 1528 node = DEVFS_NODE(vp);
21864bc5
MD
1529
1530 if (dev == NULL) /* device was revoked */
1531 return (EBADF);
1532 if (uio->uio_resid == 0)
1533 return (0);
1534
1535 vn_unlock(vp);
1536 error = dev_dread(dev, uio, ap->a_ioflag);
1537 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1538
898c91ee
MD
1539 if (node)
1540 nanotime(&node->atime);
21864bc5
MD
1541
1542 return (error);
1543}
1544
1545/*
1546 * Vnode op for write
1547 *
1548 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1549 * struct ucred *a_cred)
1550 */
21864bc5
MD
1551static int
1552devfs_spec_write(struct vop_write_args *ap)
1553{
898c91ee 1554 struct devfs_node *node;
21864bc5
MD
1555 struct vnode *vp;
1556 struct uio *uio;
1557 cdev_t dev;
1558 int error;
1559
1560 vp = ap->a_vp;
1561 dev = vp->v_rdev;
1562 uio = ap->a_uio;
898c91ee 1563 node = DEVFS_NODE(vp);
21864bc5
MD
1564
1565 KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1566
1567 if (dev == NULL) /* device was revoked */
1568 return (EBADF);
1569
1570 vn_unlock(vp);
1571 error = dev_dwrite(dev, uio, ap->a_ioflag);
1572 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1573
07dfa375
AH
1574 if (node) {
1575 nanotime(&node->atime);
898c91ee 1576 nanotime(&node->mtime);
07dfa375 1577 }
21864bc5
MD
1578
1579 return (error);
1580}
1581
1582/*
1583 * Device ioctl operation.
1584 *
1585 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
87baaf0c 1586 * int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
21864bc5 1587 */
21864bc5
MD
1588static int
1589devfs_spec_ioctl(struct vop_ioctl_args *ap)
1590{
21864bc5 1591 struct vnode *vp = ap->a_vp;
898c91ee
MD
1592 struct devfs_node *node;
1593 cdev_t dev;
21864bc5
MD
1594
1595 if ((dev = vp->v_rdev) == NULL)
1596 return (EBADF); /* device was revoked */
898c91ee 1597 node = DEVFS_NODE(vp);
21864bc5 1598
07dfa375 1599#if 0
898c91ee
MD
1600 if (node) {
1601 nanotime(&node->atime);
1602 nanotime(&node->mtime);
21864bc5 1603 }
07dfa375 1604#endif
21864bc5 1605
87baaf0c
MD
1606 return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1607 ap->a_cred, ap->a_sysmsg));
21864bc5
MD
1608}
1609
1610/*
21864bc5
MD
1611 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1612 */
1613/* ARGSUSED */
1614static int
1615devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1616{
21864bc5 1617 struct vnode *vp = ap->a_vp;
898c91ee
MD
1618 struct devfs_node *node;
1619 cdev_t dev;
21864bc5
MD
1620
1621 if ((dev = vp->v_rdev) == NULL)
b287d649 1622 return (EBADF); /* device was revoked (EBADF) */
898c91ee 1623 node = DEVFS_NODE(vp);
21864bc5 1624
07dfa375 1625#if 0
898c91ee
MD
1626 if (node)
1627 nanotime(&node->atime);
07dfa375 1628#endif
21864bc5 1629
b287d649 1630 return (dev_dkqfilter(dev, ap->a_kn));
21864bc5
MD
1631}
1632
21864bc5
MD
1633/*
1634 * Convert a vnode strategy call into a device strategy call. Vnode strategy
1635 * calls are not limited to device DMA limits so we have to deal with the
1636 * case.
1637 *
1638 * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1639 */
1640static int
1641devfs_spec_strategy(struct vop_strategy_args *ap)
1642{
1643 struct bio *bio = ap->a_bio;
1644 struct buf *bp = bio->bio_buf;
1645 struct buf *nbp;
1646 struct vnode *vp;
1647 struct mount *mp;
1648 int chunksize;
1649 int maxiosize;
1650
1651 if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1652 buf_start(bp);
1653
1654 /*
1655 * Collect statistics on synchronous and asynchronous read
1656 * and write counts for disks that have associated filesystems.
1657 */
1658 vp = ap->a_vp;
1659 KKASSERT(vp->v_rdev != NULL); /* XXX */
1660 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1661 if (bp->b_cmd == BUF_CMD_READ) {
21864bc5
MD
1662 if (bp->b_flags & BIO_SYNC)
1663 mp->mnt_stat.f_syncreads++;
1664 else
1665 mp->mnt_stat.f_asyncreads++;
1666 } else {
1667 if (bp->b_flags & BIO_SYNC)
1668 mp->mnt_stat.f_syncwrites++;
1669 else
1670 mp->mnt_stat.f_asyncwrites++;
1671 }
1672 }
1673
1674 /*
1675 * Device iosize limitations only apply to read and write. Shortcut
1676 * the I/O if it fits.
1677 */
1678 if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
898c91ee
MD
1679 devfs_debug(DEVFS_DEBUG_DEBUG,
1680 "%s: si_iosize_max not set!\n",
1681 dev_dname(vp->v_rdev));
21864bc5
MD
1682 maxiosize = MAXPHYS;
1683 }
1684#if SPEC_CHAIN_DEBUG & 2
1685 maxiosize = 4096;
1686#endif
1687 if (bp->b_bcount <= maxiosize ||
1688 (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1689 dev_dstrategy_chain(vp->v_rdev, bio);
1690 return (0);
1691 }
1692
1693 /*
1694 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1695 */
1696 nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1697 initbufbio(nbp);
1698 buf_dep_init(nbp);
21864bc5
MD
1699 BUF_LOCK(nbp, LK_EXCLUSIVE);
1700 BUF_KERNPROC(nbp);
1701 nbp->b_vp = vp;
1702 nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
1703 nbp->b_data = bp->b_data;
1704 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1705 nbp->b_bio1.bio_offset = bio->bio_offset;
1706 nbp->b_bio1.bio_caller_info1.ptr = bio;
1707
1708 /*
1709 * Start the first transfer
1710 */
1711 if (vn_isdisk(vp, NULL))
1712 chunksize = vp->v_rdev->si_bsize_phys;
1713 else
1714 chunksize = DEV_BSIZE;
1715 chunksize = maxiosize / chunksize * chunksize;
1716#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1717 devfs_debug(DEVFS_DEBUG_DEBUG,
1718 "spec_strategy chained I/O chunksize=%d\n",
1719 chunksize);
21864bc5
MD
1720#endif
1721 nbp->b_cmd = bp->b_cmd;
1722 nbp->b_bcount = chunksize;
1723 nbp->b_bufsize = chunksize; /* used to detect a short I/O */
1724 nbp->b_bio1.bio_caller_info2.index = chunksize;
1725
1726#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1727 devfs_debug(DEVFS_DEBUG_DEBUG,
1728 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1729 bp, 0, bp->b_bcount, nbp->b_bcount);
21864bc5
MD
1730#endif
1731
1732 dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1733
1734 if (DEVFS_NODE(vp)) {
1735 nanotime(&DEVFS_NODE(vp)->atime);
1736 nanotime(&DEVFS_NODE(vp)->mtime);
1737 }
1738
1739 return (0);
1740}
1741
1742/*
1743 * Chunked up transfer completion routine - chain transfers until done
77912481
MD
1744 *
1745 * NOTE: MPSAFE callback.
21864bc5
MD
1746 */
1747static
1748void
1749devfs_spec_strategy_done(struct bio *nbio)
1750{
1751 struct buf *nbp = nbio->bio_buf;
1752 struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */
1753 struct buf *bp = bio->bio_buf; /* original bp */
1754 int chunksize = nbio->bio_caller_info2.index; /* chunking */
1755 int boffset = nbp->b_data - bp->b_data;
1756
1757 if (nbp->b_flags & B_ERROR) {
1758 /*
1759 * An error terminates the chain, propogate the error back
1760 * to the original bp
1761 */
1762 bp->b_flags |= B_ERROR;
1763 bp->b_error = nbp->b_error;
1764 bp->b_resid = bp->b_bcount - boffset +
1765 (nbp->b_bcount - nbp->b_resid);
1766#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1767 devfs_debug(DEVFS_DEBUG_DEBUG,
1768 "spec_strategy: chain %p error %d bcount %d/%d\n",
1769 bp, bp->b_error, bp->b_bcount,
1770 bp->b_bcount - bp->b_resid);
21864bc5 1771#endif
21864bc5
MD
1772 } else if (nbp->b_resid) {
1773 /*
1774 * A short read or write terminates the chain
1775 */
1776 bp->b_error = nbp->b_error;
1777 bp->b_resid = bp->b_bcount - boffset +
1778 (nbp->b_bcount - nbp->b_resid);
1779#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1780 devfs_debug(DEVFS_DEBUG_DEBUG,
1781 "spec_strategy: chain %p short read(1) "
1782 "bcount %d/%d\n",
1783 bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
21864bc5 1784#endif
21864bc5
MD
1785 } else if (nbp->b_bcount != nbp->b_bufsize) {
1786 /*
1787 * A short read or write can also occur by truncating b_bcount
1788 */
1789#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1790 devfs_debug(DEVFS_DEBUG_DEBUG,
1791 "spec_strategy: chain %p short read(2) "
1792 "bcount %d/%d\n",
1793 bp, nbp->b_bcount + boffset, bp->b_bcount);
21864bc5
MD
1794#endif
1795 bp->b_error = 0;
1796 bp->b_bcount = nbp->b_bcount + boffset;
1797 bp->b_resid = nbp->b_resid;
21864bc5
MD
1798 } else if (nbp->b_bcount + boffset == bp->b_bcount) {
1799 /*
1800 * No more data terminates the chain
1801 */
1802#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1803 devfs_debug(DEVFS_DEBUG_DEBUG,
1804 "spec_strategy: chain %p finished bcount %d\n",
1805 bp, bp->b_bcount);
21864bc5
MD
1806#endif
1807 bp->b_error = 0;
1808 bp->b_resid = 0;
21864bc5
MD
1809 } else {
1810 /*
1811 * Continue the chain
1812 */
1813 boffset += nbp->b_bcount;
1814 nbp->b_data = bp->b_data + boffset;
1815 nbp->b_bcount = bp->b_bcount - boffset;
1816 if (nbp->b_bcount > chunksize)
1817 nbp->b_bcount = chunksize;
1818 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1819 nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1820
1821#if SPEC_CHAIN_DEBUG & 1
898c91ee
MD
1822 devfs_debug(DEVFS_DEBUG_DEBUG,
1823 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1824 bp, boffset, bp->b_bcount, nbp->b_bcount);
21864bc5
MD
1825#endif
1826
1827 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
b5d7061d 1828 return;
21864bc5 1829 }
b5d7061d
MD
1830
1831 /*
1832 * Fall through to here on termination. biodone(bp) and
1833 * clean up and free nbp.
1834 */
1835 biodone(bio);
1836 BUF_UNLOCK(nbp);
1837 uninitbufbio(nbp);
1838 kfree(nbp, M_DEVBUF);
21864bc5
MD
1839}
1840
1841/*
1842 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1843 */
1844static int
1845devfs_spec_freeblks(struct vop_freeblks_args *ap)
1846{
1847 struct buf *bp;
1848
1849 /*
1850 * XXX: This assumes that strategy does the deed right away.
1851 * XXX: this may not be TRTTD.
1852 */
1853 KKASSERT(ap->a_vp->v_rdev != NULL);
bf390b25 1854 if ((ap->a_vp->v_rdev->si_flags & SI_CANFREE) == 0)
21864bc5
MD
1855 return (0);
1856 bp = geteblk(ap->a_length);
1857 bp->b_cmd = BUF_CMD_FREEBLKS;
1858 bp->b_bio1.bio_offset = ap->a_offset;
1859 bp->b_bcount = ap->a_length;
1860 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1861 return (0);
1862}
1863
1864/*
1865 * Implement degenerate case where the block requested is the block
1866 * returned, and assume that the entire device is contiguous in regards
1867 * to the contiguous block range (runp and runb).
1868 *
1869 * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1870 * off_t *a_doffsetp, int *a_runp, int *a_runb)
1871 */
1872static int
1873devfs_spec_bmap(struct vop_bmap_args *ap)
1874{
1875 if (ap->a_doffsetp != NULL)
1876 *ap->a_doffsetp = ap->a_loffset;
1877 if (ap->a_runp != NULL)
1878 *ap->a_runp = MAXBSIZE;
1879 if (ap->a_runb != NULL) {
1880 if (ap->a_loffset < MAXBSIZE)
1881 *ap->a_runb = (int)ap->a_loffset;
1882 else
1883 *ap->a_runb = MAXBSIZE;
1884 }
1885 return (0);
1886}
1887
1888
1889/*
1890 * Special device advisory byte-level locks.
1891 *
1892 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
1893 * struct flock *a_fl, int a_flags)
1894 */
1895/* ARGSUSED */
1896static int
1897devfs_spec_advlock(struct vop_advlock_args *ap)
1898{
1899 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
1900}
1901
77912481
MD
1902/*
1903 * NOTE: MPSAFE callback.
1904 */
21864bc5
MD
1905static void
1906devfs_spec_getpages_iodone(struct bio *bio)
1907{
1908 bio->bio_buf->b_cmd = BUF_CMD_DONE;
1909 wakeup(bio->bio_buf);
1910}
1911
1912/*
1913 * spec_getpages() - get pages associated with device vnode.
1914 *
1915 * Note that spec_read and spec_write do not use the buffer cache, so we
1916 * must fully implement getpages here.
1917 */
1918static int
1919devfs_spec_getpages(struct vop_getpages_args *ap)
1920{
1921 vm_offset_t kva;
1922 int error;
1923 int i, pcount, size;
1924 struct buf *bp;
1925 vm_page_t m;
1926 vm_ooffset_t offset;
1927 int toff, nextoff, nread;
1928 struct vnode *vp = ap->a_vp;
1929 int blksiz;
1930 int gotreqpage;
1931
1932 error = 0;
1933 pcount = round_page(ap->a_count) / PAGE_SIZE;
1934
1935 /*
1936 * Calculate the offset of the transfer and do sanity check.
1937 */
1938 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
1939
1940 /*
1941 * Round up physical size for real devices. We cannot round using
1942 * v_mount's block size data because v_mount has nothing to do with
1943 * the device. i.e. it's usually '/dev'. We need the physical block
1944 * size for the device itself.
1945 *
1946 * We can't use v_rdev->si_mountpoint because it only exists when the
1947 * block device is mounted. However, we can use v_rdev.
1948 */
21864bc5
MD
1949 if (vn_isdisk(vp, NULL))
1950 blksiz = vp->v_rdev->si_bsize_phys;
1951 else
1952 blksiz = DEV_BSIZE;
1953
1954 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
1955
ad8b1a17 1956 bp = getpbuf_kva(NULL);
21864bc5
MD
1957 kva = (vm_offset_t)bp->b_data;
1958
1959 /*
1960 * Map the pages to be read into the kva.
1961 */
1962 pmap_qenter(kva, ap->a_m, pcount);
1963
1964 /* Build a minimal buffer header. */
1965 bp->b_cmd = BUF_CMD_READ;
1966 bp->b_bcount = size;
1967 bp->b_resid = 0;
77912481 1968 bsetrunningbufspace(bp, size);
21864bc5
MD
1969
1970 bp->b_bio1.bio_offset = offset;
1971 bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
1972
1973 mycpu->gd_cnt.v_vnodein++;
1974 mycpu->gd_cnt.v_vnodepgsin += pcount;
1975
1976 /* Do the input. */
1977 vn_strategy(ap->a_vp, &bp->b_bio1);
1978
1979 crit_enter();
1980
1981 /* We definitely need to be at splbio here. */
1982 while (bp->b_cmd != BUF_CMD_DONE)
1983 tsleep(bp, 0, "spread", 0);
1984
1985 crit_exit();
1986
1987 if (bp->b_flags & B_ERROR) {
1988 if (bp->b_error)
1989 error = bp->b_error;
1990 else
1991 error = EIO;
1992 }
1993
1994 /*
1995 * If EOF is encountered we must zero-extend the result in order
1996 * to ensure that the page does not contain garabge. When no
1997 * error occurs, an early EOF is indicated if b_bcount got truncated.
1998 * b_resid is relative to b_bcount and should be 0, but some devices
1999 * might indicate an EOF with b_resid instead of truncating b_bcount.
2000 */
2001 nread = bp->b_bcount - bp->b_resid;
2002 if (nread < ap->a_count)
2003 bzero((caddr_t)kva + nread, ap->a_count - nread);
2004 pmap_qremove(kva, pcount);
2005
2006 gotreqpage = 0;
2007 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
2008 nextoff = toff + PAGE_SIZE;
2009 m = ap->a_m[i];
2010
2011 m->flags &= ~PG_ZERO;
2012
cb1cf930
MD
2013 /*
2014 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
2015 * pmap modified bit. pmap modified bit should have
2016 * already been cleared.
2017 */
21864bc5
MD
2018 if (nextoff <= nread) {
2019 m->valid = VM_PAGE_BITS_ALL;
2020 vm_page_undirty(m);
2021 } else if (toff < nread) {
2022 /*
2023 * Since this is a VM request, we have to supply the
cb1cf930 2024 * unaligned offset to allow vm_page_set_valid()
21864bc5
MD
2025 * to zero sub-DEV_BSIZE'd portions of the page.
2026 */
1a54183b
MD
2027 vm_page_set_valid(m, 0, nread - toff);
2028 vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
21864bc5
MD
2029 } else {
2030 m->valid = 0;
2031 vm_page_undirty(m);
2032 }
2033
2034 if (i != ap->a_reqpage) {
2035 /*
2036 * Just in case someone was asking for this page we
2037 * now tell them that it is ok to use.
2038 */
2039 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2040 if (m->valid) {
b12defdc 2041 if (m->flags & PG_REFERENCED) {
21864bc5
MD
2042 vm_page_activate(m);
2043 } else {
2044 vm_page_deactivate(m);
2045 }
2046 vm_page_wakeup(m);
2047 } else {
2048 vm_page_free(m);
2049 }
2050 } else {
2051 vm_page_free(m);
2052 }
2053 } else if (m->valid) {
2054 gotreqpage = 1;
2055 /*
2056 * Since this is a VM request, we need to make the
2057 * entire page presentable by zeroing invalid sections.
2058 */
2059 if (m->valid != VM_PAGE_BITS_ALL)
2060 vm_page_zero_invalid(m, FALSE);
2061 }
2062 }
2063 if (!gotreqpage) {
2064 m = ap->a_m[ap->a_reqpage];
2065 devfs_debug(DEVFS_DEBUG_WARNING,
2066 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2067 devtoname(vp->v_rdev), error, bp, bp->b_vp);
2068 devfs_debug(DEVFS_DEBUG_WARNING,
2069 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2070 size, bp->b_resid, ap->a_count, m->valid);
2071 devfs_debug(DEVFS_DEBUG_WARNING,
2072 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2073 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2074 /*
2075 * Free the buffer header back to the swap buffer pool.
2076 */
2077 relpbuf(bp, NULL);
2078 return VM_PAGER_ERROR;
2079 }
2080 /*
2081 * Free the buffer header back to the swap buffer pool.
2082 */
2083 relpbuf(bp, NULL);
07dfa375
AH
2084 if (DEVFS_NODE(ap->a_vp))
2085 nanotime(&DEVFS_NODE(ap->a_vp)->mtime);
21864bc5
MD
2086 return VM_PAGER_OK;
2087}
2088
21864bc5
MD
2089static __inline
2090int
2091sequential_heuristic(struct uio *uio, struct file *fp)
2092{
2093 /*
2094 * Sequential heuristic - detect sequential operation
2095 */
2096 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2097 uio->uio_offset == fp->f_nextoff) {
21864bc5
MD
2098 /*
2099 * XXX we assume that the filesystem block size is
2100 * the default. Not true, but still gives us a pretty
2101 * good indicator of how sequential the read operations
2102 * are.
2103 */
898c91ee
MD
2104 int tmpseq = fp->f_seqcount;
2105
21864bc5
MD
2106 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
2107 if (tmpseq > IO_SEQMAX)
2108 tmpseq = IO_SEQMAX;
2109 fp->f_seqcount = tmpseq;
2110 return(fp->f_seqcount << IO_SEQSHIFT);
2111 }
2112
2113 /*
2114 * Not sequential, quick draw-down of seqcount
2115 */
2116 if (fp->f_seqcount > 1)
2117 fp->f_seqcount = 1;
2118 else
2119 fp->f_seqcount = 0;
2120 return(0);
2121}