2 * Copyright (c) 2009 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Alex Hornung <ahornung@gmail.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 #include <sys/param.h>
35 #include <sys/systm.h>
37 #include <sys/kernel.h>
39 #include <sys/fcntl.h>
42 #include <sys/signalvar.h>
43 #include <sys/vnode.h>
45 #include <sys/mount.h>
47 #include <sys/fcntl.h>
48 #include <sys/namei.h>
49 #include <sys/dirent.h>
50 #include <sys/malloc.h>
53 #include <vm/vm_pager.h>
54 #include <vm/vm_zone.h>
55 #include <vm/vm_object.h>
56 #include <sys/filio.h>
57 #include <sys/ttycom.h>
59 #include <sys/diskslice.h>
60 #include <sys/sysctl.h>
61 #include <sys/devfs.h>
62 #include <sys/pioctl.h>
63 #include <vfs/fifofs/fifo.h>
65 #include <machine/limits.h>
68 #include <sys/sysref2.h>
69 #include <sys/mplock2.h>
70 #include <vm/vm_page2.h>
72 MALLOC_DECLARE(M_DEVFS);
73 #define DEVFS_BADOP (void *)devfs_badop
75 static int devfs_badop(struct vop_generic_args *);
76 static int devfs_access(struct vop_access_args *);
77 static int devfs_inactive(struct vop_inactive_args *);
78 static int devfs_reclaim(struct vop_reclaim_args *);
79 static int devfs_readdir(struct vop_readdir_args *);
80 static int devfs_getattr(struct vop_getattr_args *);
81 static int devfs_setattr(struct vop_setattr_args *);
82 static int devfs_readlink(struct vop_readlink_args *);
83 static int devfs_print(struct vop_print_args *);
85 static int devfs_nresolve(struct vop_nresolve_args *);
86 static int devfs_nlookupdotdot(struct vop_nlookupdotdot_args *);
87 static int devfs_nmkdir(struct vop_nmkdir_args *);
88 static int devfs_nsymlink(struct vop_nsymlink_args *);
89 static int devfs_nrmdir(struct vop_nrmdir_args *);
90 static int devfs_nremove(struct vop_nremove_args *);
92 static int devfs_spec_open(struct vop_open_args *);
93 static int devfs_spec_close(struct vop_close_args *);
94 static int devfs_spec_fsync(struct vop_fsync_args *);
96 static int devfs_spec_read(struct vop_read_args *);
97 static int devfs_spec_write(struct vop_write_args *);
98 static int devfs_spec_ioctl(struct vop_ioctl_args *);
99 static int devfs_spec_poll(struct vop_poll_args *);
100 static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
101 static int devfs_spec_strategy(struct vop_strategy_args *);
102 static void devfs_spec_strategy_done(struct bio *);
103 static int devfs_spec_freeblks(struct vop_freeblks_args *);
104 static int devfs_spec_bmap(struct vop_bmap_args *);
105 static int devfs_spec_advlock(struct vop_advlock_args *);
106 static void devfs_spec_getpages_iodone(struct bio *);
107 static int devfs_spec_getpages(struct vop_getpages_args *);
110 static int devfs_specf_close(struct file *);
111 static int devfs_specf_read(struct file *, struct uio *, struct ucred *, int);
112 static int devfs_specf_write(struct file *, struct uio *, struct ucred *, int);
113 static int devfs_specf_stat(struct file *, struct stat *, struct ucred *);
114 static int devfs_specf_kqfilter(struct file *, struct knote *);
115 static int devfs_specf_poll(struct file *, int, struct ucred *);
116 static int devfs_specf_ioctl(struct file *, u_long, caddr_t,
117 struct ucred *, struct sysmsg *);
118 static __inline int sequential_heuristic(struct uio *, struct file *);
120 extern struct lock devfs_lock;
122 static int mpsafe_reads, mpsafe_writes, mplock_reads, mplock_writes;
125 * devfs vnode operations for regular files
127 struct vop_ops devfs_vnode_norm_vops = {
128 .vop_default = vop_defaultop,
129 .vop_access = devfs_access,
130 .vop_advlock = DEVFS_BADOP,
131 .vop_bmap = DEVFS_BADOP,
132 .vop_close = vop_stdclose,
133 .vop_getattr = devfs_getattr,
134 .vop_inactive = devfs_inactive,
135 .vop_ncreate = DEVFS_BADOP,
136 .vop_nresolve = devfs_nresolve,
137 .vop_nlookupdotdot = devfs_nlookupdotdot,
138 .vop_nlink = DEVFS_BADOP,
139 .vop_nmkdir = devfs_nmkdir,
140 .vop_nmknod = DEVFS_BADOP,
141 .vop_nremove = devfs_nremove,
142 .vop_nrename = DEVFS_BADOP,
143 .vop_nrmdir = devfs_nrmdir,
144 .vop_nsymlink = devfs_nsymlink,
145 .vop_open = vop_stdopen,
146 .vop_pathconf = vop_stdpathconf,
147 .vop_print = devfs_print,
148 .vop_read = DEVFS_BADOP,
149 .vop_readdir = devfs_readdir,
150 .vop_readlink = devfs_readlink,
151 .vop_reclaim = devfs_reclaim,
152 .vop_setattr = devfs_setattr,
153 .vop_write = DEVFS_BADOP,
154 .vop_ioctl = DEVFS_BADOP
158 * devfs vnode operations for character devices
160 struct vop_ops devfs_vnode_dev_vops = {
161 .vop_default = vop_defaultop,
162 .vop_access = devfs_access,
163 .vop_advlock = devfs_spec_advlock,
164 .vop_bmap = devfs_spec_bmap,
165 .vop_close = devfs_spec_close,
166 .vop_freeblks = devfs_spec_freeblks,
167 .vop_fsync = devfs_spec_fsync,
168 .vop_getattr = devfs_getattr,
169 .vop_getpages = devfs_spec_getpages,
170 .vop_inactive = devfs_inactive,
171 .vop_open = devfs_spec_open,
172 .vop_pathconf = vop_stdpathconf,
173 .vop_print = devfs_print,
174 .vop_poll = devfs_spec_poll,
175 .vop_kqfilter = devfs_spec_kqfilter,
176 .vop_read = devfs_spec_read,
177 .vop_readdir = DEVFS_BADOP,
178 .vop_readlink = DEVFS_BADOP,
179 .vop_reclaim = devfs_reclaim,
180 .vop_setattr = devfs_setattr,
181 .vop_strategy = devfs_spec_strategy,
182 .vop_write = devfs_spec_write,
183 .vop_ioctl = devfs_spec_ioctl
186 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops;
188 struct fileops devfs_dev_fileops = {
189 .fo_read = devfs_specf_read,
190 .fo_write = devfs_specf_write,
191 .fo_ioctl = devfs_specf_ioctl,
192 .fo_poll = devfs_specf_poll,
193 .fo_kqfilter = devfs_specf_kqfilter,
194 .fo_stat = devfs_specf_stat,
195 .fo_close = devfs_specf_close,
196 .fo_shutdown = nofo_shutdown
200 * These two functions are possibly temporary hacks for
201 * devices (aka the pty code) which want to control the
202 * node attributes themselves.
204 * XXX we may ultimately desire to simply remove the uid/gid/mode
205 * from the node entirely.
208 node_sync_dev_get(struct devfs_node *node)
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;
220 node_sync_dev_set(struct devfs_node *node)
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;
232 * generic entry point for unsupported operations
235 devfs_badop(struct vop_generic_args *ap)
242 devfs_access(struct vop_access_args *ap)
244 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
247 if (!devfs_node_is_accessible(node))
249 node_sync_dev_get(node);
250 error = vop_helper_access(ap, node->uid, node->gid,
251 node->mode, node->flags);
258 devfs_inactive(struct vop_inactive_args *ap)
260 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
262 if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
269 devfs_reclaim(struct vop_reclaim_args *ap)
271 struct devfs_node *node;
276 * Check if it is locked already. if not, we acquire the devfs lock
278 if (!(lockstatus(&devfs_lock, curthread)) == LK_EXCLUSIVE) {
279 lockmgr(&devfs_lock, LK_EXCLUSIVE);
286 * Get rid of the devfs_node if it is no longer linked into the
290 if ((node = DEVFS_NODE(vp)) != NULL) {
292 if ((node->flags & DEVFS_NODE_LINKED) == 0)
297 lockmgr(&devfs_lock, LK_RELEASE);
300 * v_rdev needs to be properly released using v_release_rdev
301 * Make sure v_data is NULL as well.
310 devfs_readdir(struct vop_readdir_args *ap)
312 struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
313 struct devfs_node *node;
322 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
324 if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
326 if ((error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY)) != 0)
329 if (!devfs_node_is_accessible(dnode)) {
334 lockmgr(&devfs_lock, LK_EXCLUSIVE);
336 saveoff = ap->a_uio->uio_offset;
338 if (ap->a_ncookies) {
339 ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
342 cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
350 nanotime(&dnode->atime);
353 r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
358 cookies[cookie_index] = saveoff;
361 if (cookie_index == ncookies)
367 r = vop_write_dirent(&error, ap->a_uio,
368 dnode->parent->d_dir.d_ino,
371 r = vop_write_dirent(&error, ap->a_uio,
378 cookies[cookie_index] = saveoff;
381 if (cookie_index == ncookies)
385 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
386 if ((node->flags & DEVFS_HIDDEN) ||
387 (node->flags & DEVFS_INVISIBLE)) {
392 * If the node type is a valid devfs alias, then we make sure that the
393 * target isn't hidden. If it is, we don't show the link in the
396 if ((node->node_type == Plink) && (node->link_target != NULL) &&
397 (node->link_target->flags & DEVFS_HIDDEN))
400 if (node->cookie < saveoff)
403 saveoff = node->cookie;
405 error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
407 node->d_dir.d_namlen,
416 cookies[cookie_index] = node->cookie;
418 if (cookie_index == ncookies)
423 lockmgr(&devfs_lock, LK_RELEASE);
426 ap->a_uio->uio_offset = saveoff;
427 if (error && cookie_index == 0) {
429 kfree(cookies, M_TEMP);
431 *ap->a_cookies = NULL;
435 *ap->a_ncookies = cookie_index;
436 *ap->a_cookies = cookies;
444 devfs_nresolve(struct vop_nresolve_args *ap)
446 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
447 struct devfs_node *node, *found = NULL;
448 struct namecache *ncp;
449 struct vnode *vp = NULL;
454 ncp = ap->a_nch->ncp;
457 if (!devfs_node_is_accessible(dnode))
460 lockmgr(&devfs_lock, LK_EXCLUSIVE);
462 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir)) {
464 cache_setvp(ap->a_nch, NULL);
468 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
469 if (len == node->d_dir.d_namlen) {
470 if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
479 while ((found->node_type == Plink) && (found->link_target)) {
481 devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
485 found = found->link_target;
489 if (!(found->flags & DEVFS_HIDDEN))
490 devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
495 cache_setvp(ap->a_nch, NULL);
501 cache_setvp(ap->a_nch, vp);
504 lockmgr(&devfs_lock, LK_RELEASE);
511 devfs_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
513 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
516 if (!devfs_node_is_accessible(dnode))
519 lockmgr(&devfs_lock, LK_EXCLUSIVE);
520 if (dnode->parent != NULL) {
521 devfs_allocv(ap->a_vpp, dnode->parent);
522 vn_unlock(*ap->a_vpp);
524 lockmgr(&devfs_lock, LK_RELEASE);
526 return ((*ap->a_vpp == NULL) ? ENOENT : 0);
531 devfs_getattr(struct vop_getattr_args *ap)
533 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
534 struct vattr *vap = ap->a_vap;
535 struct partinfo pinfo;
539 if (!devfs_node_is_accessible(node))
542 node_sync_dev_get(node);
544 lockmgr(&devfs_lock, LK_EXCLUSIVE);
546 /* start by zeroing out the attributes */
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 ;
554 vap->va_blocksize = DEV_BSIZE;
555 vap->va_bytes = vap->va_size = 0;
557 vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
559 vap->va_atime = node->atime;
560 vap->va_mtime = node->mtime;
561 vap->va_ctime = node->ctime;
563 vap->va_nlink = 1; /* number of references to file */
565 vap->va_uid = node->uid;
566 vap->va_gid = node->gid;
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);
577 /* For a softlink the va_size is the length of the softlink */
578 if (node->symlink_name != 0) {
579 vap->va_bytes = vap->va_size = node->symlink_namelen;
583 * For a disk-type device, va_size is the size of the underlying
584 * device, so that lseek() works properly.
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;
598 lockmgr(&devfs_lock, LK_RELEASE);
605 devfs_setattr(struct vop_setattr_args *ap)
607 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
611 if (!devfs_node_is_accessible(node))
613 node_sync_dev_get(node);
615 lockmgr(&devfs_lock, LK_EXCLUSIVE);
619 if (vap->va_uid != (uid_t)VNOVAL) {
620 if ((ap->a_cred->cr_uid != node->uid) &&
621 (!groupmember(node->gid, ap->a_cred))) {
622 error = priv_check(curthread, PRIV_VFS_CHOWN);
626 node->uid = vap->va_uid;
629 if (vap->va_gid != (uid_t)VNOVAL) {
630 if ((ap->a_cred->cr_uid != node->uid) &&
631 (!groupmember(node->gid, ap->a_cred))) {
632 error = priv_check(curthread, PRIV_VFS_CHOWN);
636 node->gid = vap->va_gid;
639 if (vap->va_mode != (mode_t)VNOVAL) {
640 if (ap->a_cred->cr_uid != node->uid) {
641 error = priv_check(curthread, PRIV_VFS_ADMIN);
645 node->mode = vap->va_mode;
649 node_sync_dev_set(node);
650 nanotime(&node->ctime);
651 lockmgr(&devfs_lock, LK_RELEASE);
658 devfs_readlink(struct vop_readlink_args *ap)
660 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
663 if (!devfs_node_is_accessible(node))
666 lockmgr(&devfs_lock, LK_EXCLUSIVE);
667 ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
668 lockmgr(&devfs_lock, LK_RELEASE);
675 devfs_print(struct vop_print_args *ap)
681 devfs_nmkdir(struct vop_nmkdir_args *ap)
683 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
684 struct devfs_node *node;
686 if (!devfs_node_is_accessible(dnode))
689 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
692 lockmgr(&devfs_lock, LK_EXCLUSIVE);
693 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Pdir,
694 ap->a_nch->ncp->nc_name, dnode, NULL);
697 node = DEVFS_NODE(*ap->a_vpp);
698 node->flags |= DEVFS_USER_CREATED;
699 cache_setunresolved(ap->a_nch);
700 cache_setvp(ap->a_nch, *ap->a_vpp);
702 lockmgr(&devfs_lock, LK_RELEASE);
704 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
708 devfs_nsymlink(struct vop_nsymlink_args *ap)
710 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
711 struct devfs_node *node;
714 if (!devfs_node_is_accessible(dnode))
717 ap->a_vap->va_type = VLNK;
719 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
722 lockmgr(&devfs_lock, LK_EXCLUSIVE);
723 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Plink,
724 ap->a_nch->ncp->nc_name, dnode, NULL);
726 targetlen = strlen(ap->a_target);
728 node = DEVFS_NODE(*ap->a_vpp);
729 node->flags |= DEVFS_USER_CREATED;
730 node->symlink_namelen = targetlen;
731 node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
732 memcpy(node->symlink_name, ap->a_target, targetlen);
733 node->symlink_name[targetlen] = '\0';
734 cache_setunresolved(ap->a_nch);
735 cache_setvp(ap->a_nch, *ap->a_vpp);
737 lockmgr(&devfs_lock, LK_RELEASE);
739 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
743 devfs_nrmdir(struct vop_nrmdir_args *ap)
745 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
746 struct devfs_node *node;
747 struct namecache *ncp;
750 ncp = ap->a_nch->ncp;
752 if (!devfs_node_is_accessible(dnode))
755 lockmgr(&devfs_lock, LK_EXCLUSIVE);
757 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
760 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
761 if (ncp->nc_nlen != node->d_dir.d_namlen)
763 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
767 * only allow removal of user created dirs
769 if ((node->flags & DEVFS_USER_CREATED) == 0) {
772 } else if (node->node_type != Pdir) {
775 } else if (node->nchildren > 2) {
780 cache_inval_vp(node->v_node, CINV_DESTROY);
787 cache_setunresolved(ap->a_nch);
788 cache_setvp(ap->a_nch, NULL);
791 lockmgr(&devfs_lock, LK_RELEASE);
796 devfs_nremove(struct vop_nremove_args *ap)
798 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
799 struct devfs_node *node;
800 struct namecache *ncp;
803 ncp = ap->a_nch->ncp;
805 if (!devfs_node_is_accessible(dnode))
808 lockmgr(&devfs_lock, LK_EXCLUSIVE);
810 if ((dnode->node_type != Proot) && (dnode->node_type != Pdir))
813 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
814 if (ncp->nc_nlen != node->d_dir.d_namlen)
816 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
820 * only allow removal of user created stuff (e.g. symlinks)
822 if ((node->flags & DEVFS_USER_CREATED) == 0) {
825 } else if (node->node_type == Pdir) {
830 cache_inval_vp(node->v_node, CINV_DESTROY);
837 cache_setunresolved(ap->a_nch);
838 cache_setvp(ap->a_nch, NULL);
841 lockmgr(&devfs_lock, LK_RELEASE);
847 devfs_spec_open(struct vop_open_args *ap)
849 struct vnode *vp = ap->a_vp;
850 struct vnode *orig_vp = NULL;
851 struct devfs_node *node = DEVFS_NODE(vp);
852 struct devfs_node *newnode;
853 cdev_t dev, ndev = NULL;
857 if (node->d_dev == NULL)
859 if (!devfs_node_is_accessible(node))
863 if ((dev = vp->v_rdev) == NULL)
866 if (node && ap->a_fp) {
867 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
868 lockmgr(&devfs_lock, LK_EXCLUSIVE);
870 ndev = devfs_clone(dev, node->d_dir.d_name, node->d_dir.d_namlen,
871 ap->a_mode, ap->a_cred);
873 newnode = devfs_create_device_node(
874 DEVFS_MNTDATA(vp->v_mount)->root_node,
876 /* XXX: possibly destroy device if this happens */
878 if (newnode != NULL) {
882 devfs_debug(DEVFS_DEBUG_DEBUG,
883 "parent here is: %s, node is: |%s|\n",
884 ((node->parent->node_type == Proot) ?
885 "ROOT!" : node->parent->d_dir.d_name),
886 newnode->d_dir.d_name);
887 devfs_debug(DEVFS_DEBUG_DEBUG,
889 ((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
892 * orig_vp is set to the original vp if we cloned.
894 /* node->flags |= DEVFS_CLONED; */
895 devfs_allocv(&vp, newnode);
900 lockmgr(&devfs_lock, LK_RELEASE);
903 devfs_debug(DEVFS_DEBUG_DEBUG,
904 "devfs_spec_open() called on %s! \n",
908 * Make this field valid before any I/O in ->d_open
910 if (!dev->si_iosize_max)
911 dev->si_iosize_max = DFLTPHYS;
913 if (dev_dflags(dev) & D_TTY)
914 vsetflags(vp, VISTTY);
917 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred);
918 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
921 * Clean up any cloned vp if we error out.
927 /* orig_vp = NULL; */
933 * This checks if the disk device is going to be opened for writing.
934 * It will be only allowed in the cases where securelevel permits it
935 * and it's not mounted R/W.
937 if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
938 (ap->a_cred != FSCRED)) {
940 /* Very secure mode. No open for writing allowed */
941 if (securelevel >= 2)
945 * If it is mounted R/W, do not allow to open for writing.
946 * In the case it's mounted read-only but securelevel
947 * is >= 1, then do not allow opening for writing either.
949 if (vfs_mountedon(vp)) {
950 if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
952 else if (securelevel >= 1)
957 if (dev_dflags(dev) & D_TTY) {
962 devfs_debug(DEVFS_DEBUG_DEBUG,
963 "devfs: no t_stop\n");
964 tp->t_stop = nottystop;
970 if (vn_isdisk(vp, NULL)) {
971 if (!dev->si_bsize_phys)
972 dev->si_bsize_phys = DEV_BSIZE;
973 vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
979 nanotime(&node->atime);
985 /* Ugly pty magic, to make pty devices appear once they are opened */
986 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
987 node->flags &= ~DEVFS_INVISIBLE;
990 ap->a_fp->f_type = DTYPE_VNODE;
991 ap->a_fp->f_flag = ap->a_mode & FMASK;
992 ap->a_fp->f_ops = &devfs_dev_fileops;
993 ap->a_fp->f_data = vp;
1001 devfs_spec_close(struct vop_close_args *ap)
1003 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
1004 struct proc *p = curproc;
1005 struct vnode *vp = ap->a_vp;
1006 cdev_t dev = vp->v_rdev;
1010 devfs_debug(DEVFS_DEBUG_DEBUG,
1011 "devfs_spec_close() called on %s! \n",
1015 * A couple of hacks for devices and tty devices. The
1016 * vnode ref count cannot be used to figure out the
1017 * last close, but we can use v_opencount now that
1018 * revoke works properly.
1020 * Detect the last close on a controlling terminal and clear
1021 * the session (half-close).
1026 if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1027 p->p_session->s_ttyvp = NULL;
1032 * Vnodes can be opened and closed multiple times. Do not really
1033 * close the device unless (1) it is being closed forcibly,
1034 * (2) the device wants to track closes, or (3) this is the last
1035 * vnode doing its last close on the device.
1037 * XXX the VXLOCK (force close) case can leave vnodes referencing
1038 * a closed device. This might not occur now that our revoke is
1041 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1042 if (dev && ((vp->v_flag & VRECLAIMED) ||
1043 (dev_dflags(dev) & D_TRACKCLOSE) ||
1044 (vp->v_opencount == 1))) {
1046 * Unlock around dev_dclose()
1049 if (vn_islocked(vp)) {
1053 error = dev_dclose(dev, ap->a_fflag, S_IFCHR);
1056 * Ugly pty magic, to make pty devices disappear again once
1059 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
1060 node->flags |= DEVFS_INVISIBLE;
1063 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1067 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
1070 * Track the actual opens and closes on the vnode. The last close
1071 * disassociates the rdev. If the rdev is already disassociated or
1072 * the opencount is already 0, the vnode might have been revoked
1073 * and no further opencount tracking occurs.
1077 if (vp->v_opencount > 0)
1085 devfs_specf_close(struct file *fp)
1087 struct vnode *vp = (struct vnode *)fp->f_data;
1091 fp->f_ops = &badfileops;
1092 error = vn_close(vp, fp->f_flag);
1100 * Device-optimized file table vnode read routine.
1102 * This bypasses the VOP table and talks directly to the device. Most
1103 * filesystems just route to specfs and can make this optimization.
1105 * MPALMOSTSAFE - acquires mplock
1108 devfs_specf_read(struct file *fp, struct uio *uio,
1109 struct ucred *cred, int flags)
1111 struct devfs_node *node;
1117 KASSERT(uio->uio_td == curthread,
1118 ("uio_td %p is not td %p", uio->uio_td, curthread));
1120 if (uio->uio_resid == 0)
1123 vp = (struct vnode *)fp->f_data;
1124 if (vp == NULL || vp->v_type == VBAD)
1127 node = DEVFS_NODE(vp);
1129 if ((dev = vp->v_rdev) == NULL)
1132 /* only acquire mplock for devices that require it */
1133 if (!(dev_dflags(dev) & D_MPSAFE_READ)) {
1134 atomic_add_int(&mplock_reads, 1);
1137 atomic_add_int(&mpsafe_reads, 1);
1142 if ((flags & O_FOFFSET) == 0)
1143 uio->uio_offset = fp->f_offset;
1146 if (flags & O_FBLOCKING) {
1147 /* ioflag &= ~IO_NDELAY; */
1148 } else if (flags & O_FNONBLOCKING) {
1149 ioflag |= IO_NDELAY;
1150 } else if (fp->f_flag & FNONBLOCK) {
1151 ioflag |= IO_NDELAY;
1153 if (flags & O_FBUFFERED) {
1154 /* ioflag &= ~IO_DIRECT; */
1155 } else if (flags & O_FUNBUFFERED) {
1156 ioflag |= IO_DIRECT;
1157 } else if (fp->f_flag & O_DIRECT) {
1158 ioflag |= IO_DIRECT;
1160 ioflag |= sequential_heuristic(uio, fp);
1162 error = dev_dread(dev, uio, ioflag);
1166 nanotime(&node->atime);
1167 if ((flags & O_FOFFSET) == 0)
1168 fp->f_offset = uio->uio_offset;
1169 fp->f_nextoff = uio->uio_offset;
1171 if (!(dev_dflags(dev) & D_MPSAFE_READ))
1179 devfs_specf_write(struct file *fp, struct uio *uio,
1180 struct ucred *cred, int flags)
1182 struct devfs_node *node;
1188 KASSERT(uio->uio_td == curthread,
1189 ("uio_td %p is not p %p", uio->uio_td, curthread));
1191 vp = (struct vnode *)fp->f_data;
1192 if (vp == NULL || vp->v_type == VBAD)
1195 node = DEVFS_NODE(vp);
1197 if (vp->v_type == VREG)
1198 bwillwrite(uio->uio_resid);
1200 vp = (struct vnode *)fp->f_data;
1202 if ((dev = vp->v_rdev) == NULL)
1205 /* only acquire mplock for devices that require it */
1206 if (!(dev_dflags(dev) & D_MPSAFE_WRITE)) {
1207 atomic_add_int(&mplock_writes, 1);
1210 atomic_add_int(&mpsafe_writes, 1);
1215 if ((flags & O_FOFFSET) == 0)
1216 uio->uio_offset = fp->f_offset;
1219 if (vp->v_type == VREG &&
1220 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1221 ioflag |= IO_APPEND;
1224 if (flags & O_FBLOCKING) {
1225 /* ioflag &= ~IO_NDELAY; */
1226 } else if (flags & O_FNONBLOCKING) {
1227 ioflag |= IO_NDELAY;
1228 } else if (fp->f_flag & FNONBLOCK) {
1229 ioflag |= IO_NDELAY;
1231 if (flags & O_FBUFFERED) {
1232 /* ioflag &= ~IO_DIRECT; */
1233 } else if (flags & O_FUNBUFFERED) {
1234 ioflag |= IO_DIRECT;
1235 } else if (fp->f_flag & O_DIRECT) {
1236 ioflag |= IO_DIRECT;
1238 if (flags & O_FASYNCWRITE) {
1239 /* ioflag &= ~IO_SYNC; */
1240 } else if (flags & O_FSYNCWRITE) {
1242 } else if (fp->f_flag & O_FSYNC) {
1246 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1248 ioflag |= sequential_heuristic(uio, fp);
1250 error = dev_dwrite(dev, uio, ioflag);
1254 nanotime(&node->atime);
1255 nanotime(&node->mtime);
1258 if ((flags & O_FOFFSET) == 0)
1259 fp->f_offset = uio->uio_offset;
1260 fp->f_nextoff = uio->uio_offset;
1262 if (!(dev_dflags(dev) & D_MPSAFE_WRITE))
1269 devfs_specf_stat(struct file *fp, struct stat *sb, struct ucred *cred)
1278 vp = (struct vnode *)fp->f_data;
1279 if (vp == NULL || vp->v_type == VBAD)
1282 error = vn_stat(vp, sb, cred);
1287 error = VOP_GETATTR(vp, vap);
1292 * Zero the spare stat fields
1299 * Copy from vattr table ... or not in case it's a cloned device
1301 if (vap->va_fsid != VNOVAL)
1302 sb->st_dev = vap->va_fsid;
1304 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1306 sb->st_ino = vap->va_fileid;
1308 mode = vap->va_mode;
1312 if (vap->va_nlink > (nlink_t)-1)
1313 sb->st_nlink = (nlink_t)-1;
1315 sb->st_nlink = vap->va_nlink;
1317 sb->st_uid = vap->va_uid;
1318 sb->st_gid = vap->va_gid;
1319 sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev);
1320 sb->st_size = vap->va_bytes;
1321 sb->st_atimespec = vap->va_atime;
1322 sb->st_mtimespec = vap->va_mtime;
1323 sb->st_ctimespec = vap->va_ctime;
1326 * A VCHR and VBLK device may track the last access and last modified
1327 * time independantly of the filesystem. This is particularly true
1328 * because device read and write calls may bypass the filesystem.
1330 if (vp->v_type == VCHR || vp->v_type == VBLK) {
1333 if (dev->si_lastread) {
1334 sb->st_atimespec.tv_sec = dev->si_lastread;
1335 sb->st_atimespec.tv_nsec = 0;
1337 if (dev->si_lastwrite) {
1338 sb->st_atimespec.tv_sec = dev->si_lastwrite;
1339 sb->st_atimespec.tv_nsec = 0;
1345 * According to www.opengroup.org, the meaning of st_blksize is
1346 * "a filesystem-specific preferred I/O block size for this
1347 * object. In some filesystem types, this may vary from file
1349 * Default to PAGE_SIZE after much discussion.
1352 sb->st_blksize = PAGE_SIZE;
1354 sb->st_flags = vap->va_flags;
1356 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1360 sb->st_gen = (u_int32_t)vap->va_gen;
1362 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1369 devfs_specf_kqfilter(struct file *fp, struct knote *kn)
1377 vp = (struct vnode *)fp->f_data;
1378 if (vp == NULL || vp->v_type == VBAD) {
1382 if ((dev = vp->v_rdev) == NULL) {
1388 error = dev_dkqfilter(dev, kn);
1399 devfs_specf_poll(struct file *fp, int events, struct ucred *cred)
1401 struct devfs_node *node;
1408 vp = (struct vnode *)fp->f_data;
1409 if (vp == NULL || vp->v_type == VBAD) {
1413 node = DEVFS_NODE(vp);
1415 if ((dev = vp->v_rdev) == NULL) {
1420 error = dev_dpoll(dev, events);
1426 nanotime(&node->atime);
1435 * MPALMOSTSAFE - acquires mplock
1438 devfs_specf_ioctl(struct file *fp, u_long com, caddr_t data,
1439 struct ucred *ucred, struct sysmsg *msg)
1441 struct devfs_node *node;
1446 struct fiodname_args *name_args;
1450 vp = ((struct vnode *)fp->f_data);
1452 if ((dev = vp->v_rdev) == NULL)
1453 return EBADF; /* device was revoked */
1457 node = DEVFS_NODE(vp);
1459 devfs_debug(DEVFS_DEBUG_DEBUG,
1460 "devfs_specf_ioctl() called! for dev %s\n",
1463 if (com == FIODTYPE) {
1464 *(int *)data = dev_dflags(dev) & D_TYPEMASK;
1467 } else if (com == FIODNAME) {
1468 name_args = (struct fiodname_args *)data;
1469 name = dev->si_name;
1470 namlen = strlen(name) + 1;
1472 devfs_debug(DEVFS_DEBUG_DEBUG,
1473 "ioctl, got: FIODNAME for %s\n", name);
1475 if (namlen <= name_args->len)
1476 error = copyout(dev->si_name, name_args->name, namlen);
1480 devfs_debug(DEVFS_DEBUG_DEBUG,
1481 "ioctl stuff: error: %d\n", error);
1485 /* only acquire mplock for devices that require it */
1486 if (!(dev_dflags(dev) & D_MPSAFE_IOCTL))
1489 error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg);
1493 nanotime(&node->atime);
1494 nanotime(&node->mtime);
1498 if (!(dev_dflags(dev) & D_MPSAFE_IOCTL))
1501 if (com == TIOCSCTTY) {
1502 devfs_debug(DEVFS_DEBUG_DEBUG,
1503 "devfs_specf_ioctl: got TIOCSCTTY on %s\n",
1506 if (error == 0 && com == TIOCSCTTY) {
1507 struct proc *p = curthread->td_proc;
1508 struct session *sess;
1510 devfs_debug(DEVFS_DEBUG_DEBUG,
1511 "devfs_specf_ioctl: dealing with TIOCSCTTY on %s\n",
1517 sess = p->p_session;
1520 * Do nothing if reassigning same control tty
1522 if (sess->s_ttyvp == vp) {
1528 * Get rid of reference to old control tty
1530 ovp = sess->s_ttyvp;
1539 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_specf_ioctl() finished! \n");
1545 devfs_spec_fsync(struct vop_fsync_args *ap)
1547 struct vnode *vp = ap->a_vp;
1550 if (!vn_isdisk(vp, NULL))
1554 * Flush all dirty buffers associated with a block device.
1556 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1561 devfs_spec_read(struct vop_read_args *ap)
1563 struct devfs_node *node;
1572 node = DEVFS_NODE(vp);
1574 if (dev == NULL) /* device was revoked */
1576 if (uio->uio_resid == 0)
1580 error = dev_dread(dev, uio, ap->a_ioflag);
1581 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1584 nanotime(&node->atime);
1590 * Vnode op for write
1592 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1593 * struct ucred *a_cred)
1596 devfs_spec_write(struct vop_write_args *ap)
1598 struct devfs_node *node;
1607 node = DEVFS_NODE(vp);
1609 KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1611 if (dev == NULL) /* device was revoked */
1615 error = dev_dwrite(dev, uio, ap->a_ioflag);
1616 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1619 nanotime(&node->atime);
1620 nanotime(&node->mtime);
1627 * Device ioctl operation.
1629 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
1630 * int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
1633 devfs_spec_ioctl(struct vop_ioctl_args *ap)
1635 struct vnode *vp = ap->a_vp;
1636 struct devfs_node *node;
1639 if ((dev = vp->v_rdev) == NULL)
1640 return (EBADF); /* device was revoked */
1641 node = DEVFS_NODE(vp);
1645 nanotime(&node->atime);
1646 nanotime(&node->mtime);
1650 return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1651 ap->a_cred, ap->a_sysmsg));
1655 * spec_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred)
1659 devfs_spec_poll(struct vop_poll_args *ap)
1661 struct vnode *vp = ap->a_vp;
1662 struct devfs_node *node;
1665 if ((dev = vp->v_rdev) == NULL)
1666 return (EBADF); /* device was revoked */
1667 node = DEVFS_NODE(vp);
1671 nanotime(&node->atime);
1674 return (dev_dpoll(dev, ap->a_events));
1678 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1682 devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1684 struct vnode *vp = ap->a_vp;
1685 struct devfs_node *node;
1688 if ((dev = vp->v_rdev) == NULL)
1689 return (EBADF); /* device was revoked (EBADF) */
1690 node = DEVFS_NODE(vp);
1694 nanotime(&node->atime);
1697 return (dev_dkqfilter(dev, ap->a_kn));
1701 * Convert a vnode strategy call into a device strategy call. Vnode strategy
1702 * calls are not limited to device DMA limits so we have to deal with the
1705 * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1708 devfs_spec_strategy(struct vop_strategy_args *ap)
1710 struct bio *bio = ap->a_bio;
1711 struct buf *bp = bio->bio_buf;
1718 if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1722 * Collect statistics on synchronous and asynchronous read
1723 * and write counts for disks that have associated filesystems.
1726 KKASSERT(vp->v_rdev != NULL); /* XXX */
1727 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1728 if (bp->b_cmd == BUF_CMD_READ) {
1729 if (bp->b_flags & BIO_SYNC)
1730 mp->mnt_stat.f_syncreads++;
1732 mp->mnt_stat.f_asyncreads++;
1734 if (bp->b_flags & BIO_SYNC)
1735 mp->mnt_stat.f_syncwrites++;
1737 mp->mnt_stat.f_asyncwrites++;
1742 * Device iosize limitations only apply to read and write. Shortcut
1743 * the I/O if it fits.
1745 if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
1746 devfs_debug(DEVFS_DEBUG_DEBUG,
1747 "%s: si_iosize_max not set!\n",
1748 dev_dname(vp->v_rdev));
1749 maxiosize = MAXPHYS;
1751 #if SPEC_CHAIN_DEBUG & 2
1754 if (bp->b_bcount <= maxiosize ||
1755 (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1756 dev_dstrategy_chain(vp->v_rdev, bio);
1761 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1763 nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1767 BUF_LOCK(nbp, LK_EXCLUSIVE);
1770 nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
1771 nbp->b_data = bp->b_data;
1772 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1773 nbp->b_bio1.bio_offset = bio->bio_offset;
1774 nbp->b_bio1.bio_caller_info1.ptr = bio;
1777 * Start the first transfer
1779 if (vn_isdisk(vp, NULL))
1780 chunksize = vp->v_rdev->si_bsize_phys;
1782 chunksize = DEV_BSIZE;
1783 chunksize = maxiosize / chunksize * chunksize;
1784 #if SPEC_CHAIN_DEBUG & 1
1785 devfs_debug(DEVFS_DEBUG_DEBUG,
1786 "spec_strategy chained I/O chunksize=%d\n",
1789 nbp->b_cmd = bp->b_cmd;
1790 nbp->b_bcount = chunksize;
1791 nbp->b_bufsize = chunksize; /* used to detect a short I/O */
1792 nbp->b_bio1.bio_caller_info2.index = chunksize;
1794 #if SPEC_CHAIN_DEBUG & 1
1795 devfs_debug(DEVFS_DEBUG_DEBUG,
1796 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1797 bp, 0, bp->b_bcount, nbp->b_bcount);
1800 dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1802 if (DEVFS_NODE(vp)) {
1803 nanotime(&DEVFS_NODE(vp)->atime);
1804 nanotime(&DEVFS_NODE(vp)->mtime);
1811 * Chunked up transfer completion routine - chain transfers until done
1815 devfs_spec_strategy_done(struct bio *nbio)
1817 struct buf *nbp = nbio->bio_buf;
1818 struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */
1819 struct buf *bp = bio->bio_buf; /* original bp */
1820 int chunksize = nbio->bio_caller_info2.index; /* chunking */
1821 int boffset = nbp->b_data - bp->b_data;
1823 if (nbp->b_flags & B_ERROR) {
1825 * An error terminates the chain, propogate the error back
1826 * to the original bp
1828 bp->b_flags |= B_ERROR;
1829 bp->b_error = nbp->b_error;
1830 bp->b_resid = bp->b_bcount - boffset +
1831 (nbp->b_bcount - nbp->b_resid);
1832 #if SPEC_CHAIN_DEBUG & 1
1833 devfs_debug(DEVFS_DEBUG_DEBUG,
1834 "spec_strategy: chain %p error %d bcount %d/%d\n",
1835 bp, bp->b_error, bp->b_bcount,
1836 bp->b_bcount - bp->b_resid);
1838 kfree(nbp, M_DEVBUF);
1840 } else if (nbp->b_resid) {
1842 * A short read or write terminates the chain
1844 bp->b_error = nbp->b_error;
1845 bp->b_resid = bp->b_bcount - boffset +
1846 (nbp->b_bcount - nbp->b_resid);
1847 #if SPEC_CHAIN_DEBUG & 1
1848 devfs_debug(DEVFS_DEBUG_DEBUG,
1849 "spec_strategy: chain %p short read(1) "
1851 bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
1853 kfree(nbp, M_DEVBUF);
1855 } else if (nbp->b_bcount != nbp->b_bufsize) {
1857 * A short read or write can also occur by truncating b_bcount
1859 #if SPEC_CHAIN_DEBUG & 1
1860 devfs_debug(DEVFS_DEBUG_DEBUG,
1861 "spec_strategy: chain %p short read(2) "
1863 bp, nbp->b_bcount + boffset, bp->b_bcount);
1866 bp->b_bcount = nbp->b_bcount + boffset;
1867 bp->b_resid = nbp->b_resid;
1868 kfree(nbp, M_DEVBUF);
1870 } else if (nbp->b_bcount + boffset == bp->b_bcount) {
1872 * No more data terminates the chain
1874 #if SPEC_CHAIN_DEBUG & 1
1875 devfs_debug(DEVFS_DEBUG_DEBUG,
1876 "spec_strategy: chain %p finished bcount %d\n",
1881 kfree(nbp, M_DEVBUF);
1885 * Continue the chain
1887 boffset += nbp->b_bcount;
1888 nbp->b_data = bp->b_data + boffset;
1889 nbp->b_bcount = bp->b_bcount - boffset;
1890 if (nbp->b_bcount > chunksize)
1891 nbp->b_bcount = chunksize;
1892 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1893 nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1895 #if SPEC_CHAIN_DEBUG & 1
1896 devfs_debug(DEVFS_DEBUG_DEBUG,
1897 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1898 bp, boffset, bp->b_bcount, nbp->b_bcount);
1901 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
1906 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1909 devfs_spec_freeblks(struct vop_freeblks_args *ap)
1914 * XXX: This assumes that strategy does the deed right away.
1915 * XXX: this may not be TRTTD.
1917 KKASSERT(ap->a_vp->v_rdev != NULL);
1918 if ((dev_dflags(ap->a_vp->v_rdev) & D_CANFREE) == 0)
1920 bp = geteblk(ap->a_length);
1921 bp->b_cmd = BUF_CMD_FREEBLKS;
1922 bp->b_bio1.bio_offset = ap->a_offset;
1923 bp->b_bcount = ap->a_length;
1924 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1929 * Implement degenerate case where the block requested is the block
1930 * returned, and assume that the entire device is contiguous in regards
1931 * to the contiguous block range (runp and runb).
1933 * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1934 * off_t *a_doffsetp, int *a_runp, int *a_runb)
1937 devfs_spec_bmap(struct vop_bmap_args *ap)
1939 if (ap->a_doffsetp != NULL)
1940 *ap->a_doffsetp = ap->a_loffset;
1941 if (ap->a_runp != NULL)
1942 *ap->a_runp = MAXBSIZE;
1943 if (ap->a_runb != NULL) {
1944 if (ap->a_loffset < MAXBSIZE)
1945 *ap->a_runb = (int)ap->a_loffset;
1947 *ap->a_runb = MAXBSIZE;
1954 * Special device advisory byte-level locks.
1956 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
1957 * struct flock *a_fl, int a_flags)
1961 devfs_spec_advlock(struct vop_advlock_args *ap)
1963 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
1967 devfs_spec_getpages_iodone(struct bio *bio)
1969 bio->bio_buf->b_cmd = BUF_CMD_DONE;
1970 wakeup(bio->bio_buf);
1974 * spec_getpages() - get pages associated with device vnode.
1976 * Note that spec_read and spec_write do not use the buffer cache, so we
1977 * must fully implement getpages here.
1980 devfs_spec_getpages(struct vop_getpages_args *ap)
1984 int i, pcount, size;
1987 vm_ooffset_t offset;
1988 int toff, nextoff, nread;
1989 struct vnode *vp = ap->a_vp;
1994 pcount = round_page(ap->a_count) / PAGE_SIZE;
1997 * Calculate the offset of the transfer and do sanity check.
1999 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
2002 * Round up physical size for real devices. We cannot round using
2003 * v_mount's block size data because v_mount has nothing to do with
2004 * the device. i.e. it's usually '/dev'. We need the physical block
2005 * size for the device itself.
2007 * We can't use v_rdev->si_mountpoint because it only exists when the
2008 * block device is mounted. However, we can use v_rdev.
2010 if (vn_isdisk(vp, NULL))
2011 blksiz = vp->v_rdev->si_bsize_phys;
2015 size = (ap->a_count + blksiz - 1) & ~(blksiz - 1);
2018 kva = (vm_offset_t)bp->b_data;
2021 * Map the pages to be read into the kva.
2023 pmap_qenter(kva, ap->a_m, pcount);
2025 /* Build a minimal buffer header. */
2026 bp->b_cmd = BUF_CMD_READ;
2027 bp->b_bcount = size;
2029 bp->b_runningbufspace = size;
2031 runningbufspace += bp->b_runningbufspace;
2035 bp->b_bio1.bio_offset = offset;
2036 bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
2038 mycpu->gd_cnt.v_vnodein++;
2039 mycpu->gd_cnt.v_vnodepgsin += pcount;
2042 vn_strategy(ap->a_vp, &bp->b_bio1);
2046 /* We definitely need to be at splbio here. */
2047 while (bp->b_cmd != BUF_CMD_DONE)
2048 tsleep(bp, 0, "spread", 0);
2052 if (bp->b_flags & B_ERROR) {
2054 error = bp->b_error;
2060 * If EOF is encountered we must zero-extend the result in order
2061 * to ensure that the page does not contain garabge. When no
2062 * error occurs, an early EOF is indicated if b_bcount got truncated.
2063 * b_resid is relative to b_bcount and should be 0, but some devices
2064 * might indicate an EOF with b_resid instead of truncating b_bcount.
2066 nread = bp->b_bcount - bp->b_resid;
2067 if (nread < ap->a_count)
2068 bzero((caddr_t)kva + nread, ap->a_count - nread);
2069 pmap_qremove(kva, pcount);
2072 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
2073 nextoff = toff + PAGE_SIZE;
2076 m->flags &= ~PG_ZERO;
2079 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
2080 * pmap modified bit. pmap modified bit should have
2081 * already been cleared.
2083 if (nextoff <= nread) {
2084 m->valid = VM_PAGE_BITS_ALL;
2086 } else if (toff < nread) {
2088 * Since this is a VM request, we have to supply the
2089 * unaligned offset to allow vm_page_set_valid()
2090 * to zero sub-DEV_BSIZE'd portions of the page.
2092 vm_page_set_valid(m, 0, nread - toff);
2093 vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
2099 if (i != ap->a_reqpage) {
2101 * Just in case someone was asking for this page we
2102 * now tell them that it is ok to use.
2104 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2106 if (m->flags & PG_WANTED) {
2107 vm_page_activate(m);
2109 vm_page_deactivate(m);
2118 } else if (m->valid) {
2121 * Since this is a VM request, we need to make the
2122 * entire page presentable by zeroing invalid sections.
2124 if (m->valid != VM_PAGE_BITS_ALL)
2125 vm_page_zero_invalid(m, FALSE);
2129 m = ap->a_m[ap->a_reqpage];
2130 devfs_debug(DEVFS_DEBUG_WARNING,
2131 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2132 devtoname(vp->v_rdev), error, bp, bp->b_vp);
2133 devfs_debug(DEVFS_DEBUG_WARNING,
2134 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2135 size, bp->b_resid, ap->a_count, m->valid);
2136 devfs_debug(DEVFS_DEBUG_WARNING,
2137 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2138 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2140 * Free the buffer header back to the swap buffer pool.
2143 return VM_PAGER_ERROR;
2146 * Free the buffer header back to the swap buffer pool.
2149 if (DEVFS_NODE(ap->a_vp))
2150 nanotime(&DEVFS_NODE(ap->a_vp)->mtime);
2156 sequential_heuristic(struct uio *uio, struct file *fp)
2159 * Sequential heuristic - detect sequential operation
2161 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2162 uio->uio_offset == fp->f_nextoff) {
2164 * XXX we assume that the filesystem block size is
2165 * the default. Not true, but still gives us a pretty
2166 * good indicator of how sequential the read operations
2169 int tmpseq = fp->f_seqcount;
2171 tmpseq += (uio->uio_resid + BKVASIZE - 1) / BKVASIZE;
2172 if (tmpseq > IO_SEQMAX)
2174 fp->f_seqcount = tmpseq;
2175 return(fp->f_seqcount << IO_SEQSHIFT);
2179 * Not sequential, quick draw-down of seqcount
2181 if (fp->f_seqcount > 1)
2188 extern SYSCTL_NODE(_vfs, OID_AUTO, devfs, CTLFLAG_RW, 0, "devfs");
2190 SYSCTL_INT(_vfs_devfs, OID_AUTO, mpsafe_writes, CTLFLAG_RD, &mpsafe_writes,
2191 0, "mpsafe writes");
2192 SYSCTL_INT(_vfs_devfs, OID_AUTO, mplock_writes, CTLFLAG_RD, &mplock_writes,
2193 0, "non-mpsafe writes");
2194 SYSCTL_INT(_vfs_devfs, OID_AUTO, mpsafe_reads, CTLFLAG_RD, &mpsafe_reads,
2196 SYSCTL_INT(_vfs_devfs, OID_AUTO, mplock_reads, CTLFLAG_RD, &mplock_reads,
2197 0, "non-mpsafe reads");