4 * Copyright (c) 2009 The DragonFly Project. All rights reserved.
6 * This code is derived from software contributed to The DragonFly Project
7 * by Alex Hornung <ahornung@gmail.com>
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
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
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.
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
36 #include <sys/param.h>
37 #include <sys/systm.h>
39 #include <sys/kernel.h>
41 #include <sys/fcntl.h>
44 #include <sys/signalvar.h>
45 #include <sys/vnode.h>
47 #include <sys/mount.h>
49 #include <sys/namei.h>
50 #include <sys/dirent.h>
51 #include <sys/malloc.h>
54 #include <vm/vm_pager.h>
55 #include <vm/vm_zone.h>
56 #include <vm/vm_object.h>
57 #include <sys/filio.h>
58 #include <sys/ttycom.h>
60 #include <sys/diskslice.h>
61 #include <sys/sysctl.h>
62 #include <sys/devfs.h>
63 #include <sys/pioctl.h>
64 #include <vfs/fifofs/fifo.h>
66 #include <machine/limits.h>
69 #include <sys/sysref2.h>
70 #include <vm/vm_page2.h>
72 #ifndef SPEC_CHAIN_DEBUG
73 #define SPEC_CHAIN_DEBUG 0
76 MALLOC_DECLARE(M_DEVFS);
77 #define DEVFS_BADOP (void *)devfs_vop_badop
79 static int devfs_vop_badop(struct vop_generic_args *);
80 static int devfs_vop_access(struct vop_access_args *);
81 static int devfs_vop_inactive(struct vop_inactive_args *);
82 static int devfs_vop_reclaim(struct vop_reclaim_args *);
83 static int devfs_vop_readdir(struct vop_readdir_args *);
84 static int devfs_vop_getattr(struct vop_getattr_args *);
85 static int devfs_vop_setattr(struct vop_setattr_args *);
86 static int devfs_vop_readlink(struct vop_readlink_args *);
87 static int devfs_vop_print(struct vop_print_args *);
89 static int devfs_vop_nresolve(struct vop_nresolve_args *);
90 static int devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
91 static int devfs_vop_nmkdir(struct vop_nmkdir_args *);
92 static int devfs_vop_nsymlink(struct vop_nsymlink_args *);
93 static int devfs_vop_nrmdir(struct vop_nrmdir_args *);
94 static int devfs_vop_nremove(struct vop_nremove_args *);
96 static int devfs_spec_open(struct vop_open_args *);
97 static int devfs_spec_close(struct vop_close_args *);
98 static int devfs_spec_fsync(struct vop_fsync_args *);
100 static int devfs_spec_read(struct vop_read_args *);
101 static int devfs_spec_write(struct vop_write_args *);
102 static int devfs_spec_ioctl(struct vop_ioctl_args *);
103 static int devfs_spec_kqfilter(struct vop_kqfilter_args *);
104 static int devfs_spec_strategy(struct vop_strategy_args *);
105 static void devfs_spec_strategy_done(struct bio *);
106 static int devfs_spec_freeblks(struct vop_freeblks_args *);
107 static int devfs_spec_bmap(struct vop_bmap_args *);
108 static int devfs_spec_advlock(struct vop_advlock_args *);
109 static void devfs_spec_getpages_iodone(struct bio *);
110 static int devfs_spec_getpages(struct vop_getpages_args *);
112 static int devfs_fo_close(struct file *);
113 static int devfs_fo_read(struct file *, struct uio *, struct ucred *, int);
114 static int devfs_fo_write(struct file *, struct uio *, struct ucred *, int);
115 static int devfs_fo_stat(struct file *, struct stat *, struct ucred *);
116 static int devfs_fo_kqfilter(struct file *, struct knote *);
117 static int devfs_fo_ioctl(struct file *, u_long, caddr_t,
118 struct ucred *, struct sysmsg *);
119 static __inline int sequential_heuristic(struct uio *, struct file *);
121 extern struct lock devfs_lock;
124 * devfs vnode operations for regular files. All vnode ops are MPSAFE.
126 struct vop_ops devfs_vnode_norm_vops = {
127 .vop_default = vop_defaultop,
128 .vop_access = devfs_vop_access,
129 .vop_advlock = DEVFS_BADOP,
130 .vop_bmap = DEVFS_BADOP,
131 .vop_close = vop_stdclose,
132 .vop_getattr = devfs_vop_getattr,
133 .vop_inactive = devfs_vop_inactive,
134 .vop_ncreate = DEVFS_BADOP,
135 .vop_nresolve = devfs_vop_nresolve,
136 .vop_nlookupdotdot = devfs_vop_nlookupdotdot,
137 .vop_nlink = DEVFS_BADOP,
138 .vop_nmkdir = devfs_vop_nmkdir,
139 .vop_nmknod = DEVFS_BADOP,
140 .vop_nremove = devfs_vop_nremove,
141 .vop_nrename = DEVFS_BADOP,
142 .vop_nrmdir = devfs_vop_nrmdir,
143 .vop_nsymlink = devfs_vop_nsymlink,
144 .vop_open = vop_stdopen,
145 .vop_pathconf = vop_stdpathconf,
146 .vop_print = devfs_vop_print,
147 .vop_read = DEVFS_BADOP,
148 .vop_readdir = devfs_vop_readdir,
149 .vop_readlink = devfs_vop_readlink,
150 .vop_reallocblks = DEVFS_BADOP,
151 .vop_reclaim = devfs_vop_reclaim,
152 .vop_setattr = devfs_vop_setattr,
153 .vop_write = DEVFS_BADOP,
154 .vop_ioctl = DEVFS_BADOP
158 * devfs vnode operations for character devices. All vnode ops are MPSAFE.
160 struct vop_ops devfs_vnode_dev_vops = {
161 .vop_default = vop_defaultop,
162 .vop_access = devfs_vop_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_vop_getattr,
169 .vop_getpages = devfs_spec_getpages,
170 .vop_inactive = devfs_vop_inactive,
171 .vop_open = devfs_spec_open,
172 .vop_pathconf = vop_stdpathconf,
173 .vop_print = devfs_vop_print,
174 .vop_kqfilter = devfs_spec_kqfilter,
175 .vop_read = devfs_spec_read,
176 .vop_readdir = DEVFS_BADOP,
177 .vop_readlink = DEVFS_BADOP,
178 .vop_reallocblks = DEVFS_BADOP,
179 .vop_reclaim = devfs_vop_reclaim,
180 .vop_setattr = devfs_vop_setattr,
181 .vop_strategy = devfs_spec_strategy,
182 .vop_write = devfs_spec_write,
183 .vop_ioctl = devfs_spec_ioctl
187 * devfs file pointer operations. All fileops are MPSAFE.
189 struct vop_ops *devfs_vnode_dev_vops_p = &devfs_vnode_dev_vops;
191 struct fileops devfs_dev_fileops = {
192 .fo_read = devfs_fo_read,
193 .fo_write = devfs_fo_write,
194 .fo_ioctl = devfs_fo_ioctl,
195 .fo_kqfilter = devfs_fo_kqfilter,
196 .fo_stat = devfs_fo_stat,
197 .fo_close = devfs_fo_close,
198 .fo_shutdown = nofo_shutdown
202 * These two functions are possibly temporary hacks for devices (aka
203 * the pty code) which want to control the node attributes themselves.
205 * XXX we may ultimately desire to simply remove the uid/gid/mode
206 * from the node entirely.
208 * MPSAFE - sorta. Theoretically the overwrite can compete since they
209 * are loading from the same fields.
212 node_sync_dev_get(struct devfs_node *node)
216 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
217 node->uid = dev->si_uid;
218 node->gid = dev->si_gid;
219 node->mode = dev->si_perms;
224 node_sync_dev_set(struct devfs_node *node)
228 if ((dev = node->d_dev) && (dev->si_flags & SI_OVERRIDE)) {
229 dev->si_uid = node->uid;
230 dev->si_gid = node->gid;
231 dev->si_perms = node->mode;
236 * generic entry point for unsupported operations
239 devfs_vop_badop(struct vop_generic_args *ap)
246 devfs_vop_access(struct vop_access_args *ap)
248 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
251 if (!devfs_node_is_accessible(node))
253 node_sync_dev_get(node);
254 error = vop_helper_access(ap, node->uid, node->gid,
255 node->mode, node->flags);
262 devfs_vop_inactive(struct vop_inactive_args *ap)
264 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
266 if (node == NULL || (node->flags & DEVFS_NODE_LINKED) == 0)
273 devfs_vop_reclaim(struct vop_reclaim_args *ap)
275 struct devfs_node *node;
280 * Check if it is locked already. if not, we acquire the devfs lock
282 if ((lockstatus(&devfs_lock, curthread)) != LK_EXCLUSIVE) {
283 lockmgr(&devfs_lock, LK_EXCLUSIVE);
290 * Get rid of the devfs_node if it is no longer linked into the
294 if ((node = DEVFS_NODE(vp)) != NULL) {
296 if ((node->flags & DEVFS_NODE_LINKED) == 0)
301 lockmgr(&devfs_lock, LK_RELEASE);
304 * v_rdev needs to be properly released using v_release_rdev
305 * Make sure v_data is NULL as well.
314 devfs_vop_readdir(struct vop_readdir_args *ap)
316 struct devfs_node *dnode = DEVFS_NODE(ap->a_vp);
317 struct devfs_node *node;
326 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_readdir() called!\n");
328 if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX)
330 error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM);
334 if (!devfs_node_is_accessible(dnode)) {
339 lockmgr(&devfs_lock, LK_EXCLUSIVE);
341 saveoff = ap->a_uio->uio_offset;
343 if (ap->a_ncookies) {
344 ncookies = ap->a_uio->uio_resid / 16 + 1; /* Why / 16 ?? */
347 cookies = kmalloc(256 * sizeof(off_t), M_TEMP, M_WAITOK);
355 nanotime(&dnode->atime);
358 r = vop_write_dirent(&error, ap->a_uio, dnode->d_dir.d_ino,
363 cookies[cookie_index] = saveoff;
366 if (cookie_index == ncookies)
372 r = vop_write_dirent(&error, ap->a_uio,
373 dnode->parent->d_dir.d_ino,
376 r = vop_write_dirent(&error, ap->a_uio,
383 cookies[cookie_index] = saveoff;
386 if (cookie_index == ncookies)
390 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
391 if ((node->flags & DEVFS_HIDDEN) ||
392 (node->flags & DEVFS_INVISIBLE)) {
397 * If the node type is a valid devfs alias, then we make
398 * sure that the target isn't hidden. If it is, we don't
399 * show the link in the directory listing.
401 if ((node->node_type == Nlink) && (node->link_target != NULL) &&
402 (node->link_target->flags & DEVFS_HIDDEN))
405 if (node->cookie < saveoff)
408 saveoff = node->cookie;
410 error2 = vop_write_dirent(&error, ap->a_uio, node->d_dir.d_ino,
412 node->d_dir.d_namlen,
421 cookies[cookie_index] = node->cookie;
423 if (cookie_index == ncookies)
428 lockmgr(&devfs_lock, LK_RELEASE);
431 ap->a_uio->uio_offset = saveoff;
432 if (error && cookie_index == 0) {
434 kfree(cookies, M_TEMP);
436 *ap->a_cookies = NULL;
440 *ap->a_ncookies = cookie_index;
441 *ap->a_cookies = cookies;
449 devfs_vop_nresolve(struct vop_nresolve_args *ap)
451 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
452 struct devfs_node *node, *found = NULL;
453 struct namecache *ncp;
454 struct vnode *vp = NULL;
459 ncp = ap->a_nch->ncp;
462 if (!devfs_node_is_accessible(dnode))
465 lockmgr(&devfs_lock, LK_EXCLUSIVE);
467 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir)) {
469 cache_setvp(ap->a_nch, NULL);
473 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
474 if (len == node->d_dir.d_namlen) {
475 if (!memcmp(ncp->nc_name, node->d_dir.d_name, len)) {
484 while ((found->node_type == Nlink) && (found->link_target)) {
486 devfs_debug(DEVFS_DEBUG_SHOW, "Recursive link or depth >= 8");
490 found = found->link_target;
494 if (!(found->flags & DEVFS_HIDDEN))
495 devfs_allocv(/*ap->a_dvp->v_mount, */ &vp, found);
500 cache_setvp(ap->a_nch, NULL);
506 cache_setvp(ap->a_nch, vp);
509 lockmgr(&devfs_lock, LK_RELEASE);
516 devfs_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
518 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
521 if (!devfs_node_is_accessible(dnode))
524 lockmgr(&devfs_lock, LK_EXCLUSIVE);
525 if (dnode->parent != NULL) {
526 devfs_allocv(ap->a_vpp, dnode->parent);
527 vn_unlock(*ap->a_vpp);
529 lockmgr(&devfs_lock, LK_RELEASE);
531 return ((*ap->a_vpp == NULL) ? ENOENT : 0);
536 devfs_vop_getattr(struct vop_getattr_args *ap)
538 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
539 struct vattr *vap = ap->a_vap;
540 struct partinfo pinfo;
544 if (!devfs_node_is_accessible(node))
547 node_sync_dev_get(node);
549 lockmgr(&devfs_lock, LK_EXCLUSIVE);
551 /* start by zeroing out the attributes */
554 /* next do all the common fields */
555 vap->va_type = ap->a_vp->v_type;
556 vap->va_mode = node->mode;
557 vap->va_fileid = DEVFS_NODE(ap->a_vp)->d_dir.d_ino ;
559 vap->va_blocksize = DEV_BSIZE;
560 vap->va_bytes = vap->va_size = 0;
562 vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
564 vap->va_atime = node->atime;
565 vap->va_mtime = node->mtime;
566 vap->va_ctime = node->ctime;
568 vap->va_nlink = 1; /* number of references to file */
570 vap->va_uid = node->uid;
571 vap->va_gid = node->gid;
576 if ((node->node_type == Ndev) && node->d_dev) {
577 reference_dev(node->d_dev);
578 vap->va_rminor = node->d_dev->si_uminor;
579 release_dev(node->d_dev);
582 /* For a softlink the va_size is the length of the softlink */
583 if (node->symlink_name != 0) {
584 vap->va_bytes = vap->va_size = node->symlink_namelen;
588 * For a disk-type device, va_size is the size of the underlying
589 * device, so that lseek() works properly.
591 if ((node->d_dev) && (dev_dflags(node->d_dev) & D_DISK)) {
592 bzero(&pinfo, sizeof(pinfo));
593 error = dev_dioctl(node->d_dev, DIOCGPART, (void *)&pinfo,
594 0, proc0.p_ucred, NULL, NULL);
595 if ((error == 0) && (pinfo.media_blksize != 0)) {
596 vap->va_size = pinfo.media_size;
603 lockmgr(&devfs_lock, LK_RELEASE);
610 devfs_vop_setattr(struct vop_setattr_args *ap)
612 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
619 if (!devfs_node_is_accessible(node))
621 node_sync_dev_get(node);
623 lockmgr(&devfs_lock, LK_EXCLUSIVE);
627 if ((vap->va_uid != (uid_t)VNOVAL) || (vap->va_gid != (gid_t)VNOVAL)) {
630 cur_mode = node->mode;
631 error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
632 ap->a_cred, &cur_uid, &cur_gid, &cur_mode);
636 if (node->uid != cur_uid || node->gid != cur_gid) {
639 node->mode = cur_mode;
643 if (vap->va_mode != (mode_t)VNOVAL) {
644 cur_mode = node->mode;
645 error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
646 node->uid, node->gid, &cur_mode);
647 if (error == 0 && node->mode != cur_mode) {
648 node->mode = cur_mode;
653 node_sync_dev_set(node);
654 nanotime(&node->ctime);
655 lockmgr(&devfs_lock, LK_RELEASE);
662 devfs_vop_readlink(struct vop_readlink_args *ap)
664 struct devfs_node *node = DEVFS_NODE(ap->a_vp);
667 if (!devfs_node_is_accessible(node))
670 lockmgr(&devfs_lock, LK_EXCLUSIVE);
671 ret = uiomove(node->symlink_name, node->symlink_namelen, ap->a_uio);
672 lockmgr(&devfs_lock, LK_RELEASE);
679 devfs_vop_print(struct vop_print_args *ap)
685 devfs_vop_nmkdir(struct vop_nmkdir_args *ap)
687 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
688 struct devfs_node *node;
690 if (!devfs_node_is_accessible(dnode))
693 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
696 lockmgr(&devfs_lock, LK_EXCLUSIVE);
697 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Ndir,
698 ap->a_nch->ncp->nc_name, dnode, NULL);
701 node = DEVFS_NODE(*ap->a_vpp);
702 node->flags |= DEVFS_USER_CREATED;
703 cache_setunresolved(ap->a_nch);
704 cache_setvp(ap->a_nch, *ap->a_vpp);
706 lockmgr(&devfs_lock, LK_RELEASE);
708 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
712 devfs_vop_nsymlink(struct vop_nsymlink_args *ap)
714 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
715 struct devfs_node *node;
718 if (!devfs_node_is_accessible(dnode))
721 ap->a_vap->va_type = VLNK;
723 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
726 lockmgr(&devfs_lock, LK_EXCLUSIVE);
727 devfs_allocvp(ap->a_dvp->v_mount, ap->a_vpp, Nlink,
728 ap->a_nch->ncp->nc_name, dnode, NULL);
730 targetlen = strlen(ap->a_target);
732 node = DEVFS_NODE(*ap->a_vpp);
733 node->flags |= DEVFS_USER_CREATED;
734 node->symlink_namelen = targetlen;
735 node->symlink_name = kmalloc(targetlen + 1, M_DEVFS, M_WAITOK);
736 memcpy(node->symlink_name, ap->a_target, targetlen);
737 node->symlink_name[targetlen] = '\0';
738 cache_setunresolved(ap->a_nch);
739 cache_setvp(ap->a_nch, *ap->a_vpp);
741 lockmgr(&devfs_lock, LK_RELEASE);
743 return ((*ap->a_vpp == NULL) ? ENOTDIR : 0);
747 devfs_vop_nrmdir(struct vop_nrmdir_args *ap)
749 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
750 struct devfs_node *node;
751 struct namecache *ncp;
754 ncp = ap->a_nch->ncp;
756 if (!devfs_node_is_accessible(dnode))
759 lockmgr(&devfs_lock, LK_EXCLUSIVE);
761 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
764 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
765 if (ncp->nc_nlen != node->d_dir.d_namlen)
767 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
771 * only allow removal of user created dirs
773 if ((node->flags & DEVFS_USER_CREATED) == 0) {
776 } else if (node->node_type != Ndir) {
779 } else if (node->nchildren > 2) {
784 cache_inval_vp(node->v_node, CINV_DESTROY);
791 cache_unlink(ap->a_nch);
793 lockmgr(&devfs_lock, LK_RELEASE);
798 devfs_vop_nremove(struct vop_nremove_args *ap)
800 struct devfs_node *dnode = DEVFS_NODE(ap->a_dvp);
801 struct devfs_node *node;
802 struct namecache *ncp;
805 ncp = ap->a_nch->ncp;
807 if (!devfs_node_is_accessible(dnode))
810 lockmgr(&devfs_lock, LK_EXCLUSIVE);
812 if ((dnode->node_type != Nroot) && (dnode->node_type != Ndir))
815 TAILQ_FOREACH(node, DEVFS_DENODE_HEAD(dnode), link) {
816 if (ncp->nc_nlen != node->d_dir.d_namlen)
818 if (memcmp(ncp->nc_name, node->d_dir.d_name, ncp->nc_nlen))
822 * only allow removal of user created stuff (e.g. symlinks)
824 if ((node->flags & DEVFS_USER_CREATED) == 0) {
827 } else if (node->node_type == Ndir) {
832 cache_inval_vp(node->v_node, CINV_DESTROY);
839 cache_unlink(ap->a_nch);
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 vn_lock(vp, LK_UPGRADE | LK_RETRY);
868 if (node && ap->a_fp) {
871 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_open: -1.1-\n");
872 lockmgr(&devfs_lock, LK_EXCLUSIVE);
874 ndev = devfs_clone(dev, node->d_dir.d_name,
875 node->d_dir.d_namlen,
876 ap->a_mode, ap->a_cred);
878 newnode = devfs_create_device_node(
879 DEVFS_MNTDATA(vp->v_mount)->root_node,
880 ndev, &exists, NULL, NULL);
881 /* XXX: possibly destroy device if this happens */
883 if (newnode != NULL) {
888 devfs_debug(DEVFS_DEBUG_DEBUG,
889 "parent here is: %s, node is: |%s|\n",
890 ((node->parent->node_type == Nroot) ?
891 "ROOT!" : node->parent->d_dir.d_name),
892 newnode->d_dir.d_name);
893 devfs_debug(DEVFS_DEBUG_DEBUG,
895 ((struct devfs_node *)(TAILQ_LAST(DEVFS_DENODE_HEAD(node->parent), devfs_node_head)))->d_dir.d_name);
898 * orig_vp is set to the original vp if we
901 /* node->flags |= DEVFS_CLONED; */
902 devfs_allocv(&vp, newnode);
907 lockmgr(&devfs_lock, LK_RELEASE);
909 * Synchronize devfs here to make sure that, if the cloned
910 * device creates other device nodes in addition to the
911 * cloned one, all of them are created by the time we return
912 * from opening the cloned one.
918 devfs_debug(DEVFS_DEBUG_DEBUG,
919 "devfs_spec_open() called on %s! \n",
923 * Make this field valid before any I/O in ->d_open
925 if (!dev->si_iosize_max)
926 /* XXX: old DFLTPHYS == 64KB dependency */
927 dev->si_iosize_max = min(MAXPHYS,64*1024);
929 if (dev_dflags(dev) & D_TTY)
930 vsetflags(vp, VISTTY);
933 * Open underlying device
936 error = dev_dopen(dev, ap->a_mode, S_IFCHR, ap->a_cred, ap->a_fp);
937 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
940 * Clean up any cloned vp if we error out.
946 /* orig_vp = NULL; */
952 * This checks if the disk device is going to be opened for writing.
953 * It will be only allowed in the cases where securelevel permits it
954 * and it's not mounted R/W.
956 if ((dev_dflags(dev) & D_DISK) && (ap->a_mode & FWRITE) &&
957 (ap->a_cred != FSCRED)) {
959 /* Very secure mode. No open for writing allowed */
960 if (securelevel >= 2)
964 * If it is mounted R/W, do not allow to open for writing.
965 * In the case it's mounted read-only but securelevel
966 * is >= 1, then do not allow opening for writing either.
968 if (vfs_mountedon(vp)) {
969 if (!(dev->si_mountpoint->mnt_flag & MNT_RDONLY))
971 else if (securelevel >= 1)
976 if (dev_dflags(dev) & D_TTY) {
981 devfs_debug(DEVFS_DEBUG_DEBUG,
982 "devfs: no t_stop\n");
983 tp->t_stop = nottystop;
989 if (vn_isdisk(vp, NULL)) {
990 if (!dev->si_bsize_phys)
991 dev->si_bsize_phys = DEV_BSIZE;
992 vinitvmio(vp, IDX_TO_OFF(INT_MAX), PAGE_SIZE, -1);
998 nanotime(&node->atime);
1002 * If we replaced the vp the vop_stdopen() call will have loaded
1003 * it into fp->f_data and vref()d the vp, giving us two refs. So
1004 * instead of just unlocking it here we have to vput() it.
1009 /* Ugly pty magic, to make pty devices appear once they are opened */
1010 if (node && (node->flags & DEVFS_PTY) == DEVFS_PTY)
1011 node->flags &= ~DEVFS_INVISIBLE;
1014 KKASSERT(ap->a_fp->f_type == DTYPE_VNODE);
1015 KKASSERT((ap->a_fp->f_flag & FMASK) == (ap->a_mode & FMASK));
1016 ap->a_fp->f_ops = &devfs_dev_fileops;
1017 KKASSERT(ap->a_fp->f_data == (void *)vp);
1024 devfs_spec_close(struct vop_close_args *ap)
1026 struct devfs_node *node;
1027 struct proc *p = curproc;
1028 struct vnode *vp = ap->a_vp;
1029 cdev_t dev = vp->v_rdev;
1035 * We do special tests on the opencount so unfortunately we need
1036 * an exclusive lock.
1038 vn_lock(vp, LK_UPGRADE | LK_RETRY);
1041 devfs_debug(DEVFS_DEBUG_DEBUG,
1042 "devfs_spec_close() called on %s! \n",
1045 devfs_debug(DEVFS_DEBUG_DEBUG,
1046 "devfs_spec_close() called, null vode!\n");
1049 * A couple of hacks for devices and tty devices. The
1050 * vnode ref count cannot be used to figure out the
1051 * last close, but we can use v_opencount now that
1052 * revoke works properly.
1054 * Detect the last close on a controlling terminal and clear
1055 * the session (half-close).
1057 * XXX opencount is not SMP safe. The vnode is locked but there
1058 * may be multiple vnodes referencing the same device.
1062 * NOTE: Try to avoid global tokens when testing opencount
1063 * XXX hack, fixme. needs a struct lock and opencount in
1064 * struct cdev itself.
1067 opencount = vp->v_opencount;
1069 opencount = count_dev(dev); /* XXX NOT SMP SAFE */
1074 if (p && vp->v_opencount <= 1 && vp == p->p_session->s_ttyvp) {
1075 p->p_session->s_ttyvp = NULL;
1080 * Vnodes can be opened and closed multiple times. Do not really
1081 * close the device unless (1) it is being closed forcibly,
1082 * (2) the device wants to track closes, or (3) this is the last
1083 * vnode doing its last close on the device.
1085 * XXX the VXLOCK (force close) case can leave vnodes referencing
1086 * a closed device. This might not occur now that our revoke is
1089 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -1- \n");
1090 if (dev && ((vp->v_flag & VRECLAIMED) ||
1091 (dev_dflags(dev) & D_TRACKCLOSE) ||
1092 (opencount == 1))) {
1094 * Ugly pty magic, to make pty devices disappear again once
1097 node = DEVFS_NODE(ap->a_vp);
1098 if (node && (node->flags & DEVFS_PTY))
1099 node->flags |= DEVFS_INVISIBLE;
1102 * Unlock around dev_dclose(), unless the vnode is
1103 * undergoing a vgone/reclaim (during umount).
1106 if ((vp->v_flag & VRECLAIMED) == 0 && vn_islocked(vp)) {
1112 * WARNING! If the device destroys itself the devfs node
1113 * can disappear here.
1115 * WARNING! vn_lock() will fail if the vp is in a VRECLAIM,
1116 * which can occur during umount.
1118 error = dev_dclose(dev, ap->a_fflag, S_IFCHR, ap->a_fp);
1119 /* node is now stale */
1122 if (vn_lock(vp, LK_EXCLUSIVE |
1124 LK_FAILRECLAIM) != 0) {
1125 panic("devfs_spec_close: vnode %p "
1126 "unexpectedly could not be relocked",
1133 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_spec_close() -2- \n");
1136 * Track the actual opens and closes on the vnode. The last close
1137 * disassociates the rdev. If the rdev is already disassociated or
1138 * the opencount is already 0, the vnode might have been revoked
1139 * and no further opencount tracking occurs.
1143 if (vp->v_opencount > 0)
1151 devfs_fo_close(struct file *fp)
1153 struct vnode *vp = (struct vnode *)fp->f_data;
1156 fp->f_ops = &badfileops;
1157 error = vn_close(vp, fp->f_flag, fp);
1158 devfs_clear_cdevpriv(fp);
1165 * Device-optimized file table vnode read routine.
1167 * This bypasses the VOP table and talks directly to the device. Most
1168 * filesystems just route to specfs and can make this optimization.
1171 devfs_fo_read(struct file *fp, struct uio *uio,
1172 struct ucred *cred, int flags)
1174 struct devfs_node *node;
1180 KASSERT(uio->uio_td == curthread,
1181 ("uio_td %p is not td %p", uio->uio_td, curthread));
1183 if (uio->uio_resid == 0)
1186 vp = (struct vnode *)fp->f_data;
1187 if (vp == NULL || vp->v_type == VBAD)
1190 node = DEVFS_NODE(vp);
1192 if ((dev = vp->v_rdev) == NULL)
1197 if ((flags & O_FOFFSET) == 0)
1198 uio->uio_offset = fp->f_offset;
1201 if (flags & O_FBLOCKING) {
1202 /* ioflag &= ~IO_NDELAY; */
1203 } else if (flags & O_FNONBLOCKING) {
1204 ioflag |= IO_NDELAY;
1205 } else if (fp->f_flag & FNONBLOCK) {
1206 ioflag |= IO_NDELAY;
1208 if (fp->f_flag & O_DIRECT) {
1209 ioflag |= IO_DIRECT;
1211 ioflag |= sequential_heuristic(uio, fp);
1213 error = dev_dread(dev, uio, ioflag, fp);
1217 nanotime(&node->atime);
1218 if ((flags & O_FOFFSET) == 0)
1219 fp->f_offset = uio->uio_offset;
1220 fp->f_nextoff = uio->uio_offset;
1227 devfs_fo_write(struct file *fp, struct uio *uio,
1228 struct ucred *cred, int flags)
1230 struct devfs_node *node;
1236 KASSERT(uio->uio_td == curthread,
1237 ("uio_td %p is not p %p", uio->uio_td, curthread));
1239 vp = (struct vnode *)fp->f_data;
1240 if (vp == NULL || vp->v_type == VBAD)
1243 node = DEVFS_NODE(vp);
1245 if (vp->v_type == VREG)
1246 bwillwrite(uio->uio_resid);
1248 vp = (struct vnode *)fp->f_data;
1250 if ((dev = vp->v_rdev) == NULL)
1255 if ((flags & O_FOFFSET) == 0)
1256 uio->uio_offset = fp->f_offset;
1259 if (vp->v_type == VREG &&
1260 ((fp->f_flag & O_APPEND) || (flags & O_FAPPEND))) {
1261 ioflag |= IO_APPEND;
1264 if (flags & O_FBLOCKING) {
1265 /* ioflag &= ~IO_NDELAY; */
1266 } else if (flags & O_FNONBLOCKING) {
1267 ioflag |= IO_NDELAY;
1268 } else if (fp->f_flag & FNONBLOCK) {
1269 ioflag |= IO_NDELAY;
1271 if (fp->f_flag & O_DIRECT) {
1272 ioflag |= IO_DIRECT;
1274 if (flags & O_FASYNCWRITE) {
1275 /* ioflag &= ~IO_SYNC; */
1276 } else if (flags & O_FSYNCWRITE) {
1278 } else if (fp->f_flag & O_FSYNC) {
1282 if (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))
1284 ioflag |= sequential_heuristic(uio, fp);
1286 error = dev_dwrite(dev, uio, ioflag, fp);
1290 nanotime(&node->atime);
1291 nanotime(&node->mtime);
1294 if ((flags & O_FOFFSET) == 0)
1295 fp->f_offset = uio->uio_offset;
1296 fp->f_nextoff = uio->uio_offset;
1303 devfs_fo_stat(struct file *fp, struct stat *sb, struct ucred *cred)
1312 vp = (struct vnode *)fp->f_data;
1313 if (vp == NULL || vp->v_type == VBAD)
1316 error = vn_stat(vp, sb, cred);
1321 error = VOP_GETATTR(vp, vap);
1326 * Zero the spare stat fields
1333 * Copy from vattr table ... or not in case it's a cloned device
1335 if (vap->va_fsid != VNOVAL)
1336 sb->st_dev = vap->va_fsid;
1338 sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0];
1340 sb->st_ino = vap->va_fileid;
1342 mode = vap->va_mode;
1346 if (vap->va_nlink > (nlink_t)-1)
1347 sb->st_nlink = (nlink_t)-1;
1349 sb->st_nlink = vap->va_nlink;
1351 sb->st_uid = vap->va_uid;
1352 sb->st_gid = vap->va_gid;
1353 sb->st_rdev = dev2udev(DEVFS_NODE(vp)->d_dev);
1354 sb->st_size = vap->va_bytes;
1355 sb->st_atimespec = vap->va_atime;
1356 sb->st_mtimespec = vap->va_mtime;
1357 sb->st_ctimespec = vap->va_ctime;
1360 * A VCHR and VBLK device may track the last access and last modified
1361 * time independantly of the filesystem. This is particularly true
1362 * because device read and write calls may bypass the filesystem.
1364 if (vp->v_type == VCHR || vp->v_type == VBLK) {
1367 if (dev->si_lastread) {
1368 sb->st_atimespec.tv_sec = time_second +
1371 sb->st_atimespec.tv_nsec = 0;
1373 if (dev->si_lastwrite) {
1374 sb->st_atimespec.tv_sec = time_second +
1377 sb->st_atimespec.tv_nsec = 0;
1383 * According to www.opengroup.org, the meaning of st_blksize is
1384 * "a filesystem-specific preferred I/O block size for this
1385 * object. In some filesystem types, this may vary from file
1387 * Default to PAGE_SIZE after much discussion.
1390 sb->st_blksize = PAGE_SIZE;
1392 sb->st_flags = vap->va_flags;
1394 error = priv_check_cred(cred, PRIV_VFS_GENERATION, 0);
1398 sb->st_gen = (u_int32_t)vap->va_gen;
1400 sb->st_blocks = vap->va_bytes / S_BLKSIZE;
1407 devfs_fo_kqfilter(struct file *fp, struct knote *kn)
1413 vp = (struct vnode *)fp->f_data;
1414 if (vp == NULL || vp->v_type == VBAD) {
1418 if ((dev = vp->v_rdev) == NULL) {
1424 error = dev_dkqfilter(dev, kn, fp);
1433 devfs_fo_ioctl(struct file *fp, u_long com, caddr_t data,
1434 struct ucred *ucred, struct sysmsg *msg)
1437 struct devfs_node *node;
1443 struct fiodname_args *name_args;
1447 vp = ((struct vnode *)fp->f_data);
1449 if ((dev = vp->v_rdev) == NULL)
1450 return EBADF; /* device was revoked */
1455 node = DEVFS_NODE(vp);
1458 devfs_debug(DEVFS_DEBUG_DEBUG,
1459 "devfs_fo_ioctl() called! for dev %s\n",
1462 if (com == FIODTYPE) {
1463 *(int *)data = dev_dflags(dev) & D_TYPEMASK;
1466 } else if (com == FIODNAME) {
1467 name_args = (struct fiodname_args *)data;
1468 name = dev->si_name;
1469 namlen = strlen(name) + 1;
1471 devfs_debug(DEVFS_DEBUG_DEBUG,
1472 "ioctl, got: FIODNAME for %s\n", name);
1474 if (namlen <= name_args->len)
1475 error = copyout(dev->si_name, name_args->name, namlen);
1479 devfs_debug(DEVFS_DEBUG_DEBUG,
1480 "ioctl stuff: error: %d\n", error);
1484 error = dev_dioctl(dev, com, data, fp->f_flag, ucred, msg, fp);
1488 nanotime(&node->atime);
1489 nanotime(&node->mtime);
1492 if (com == TIOCSCTTY) {
1493 devfs_debug(DEVFS_DEBUG_DEBUG,
1494 "devfs_fo_ioctl: got TIOCSCTTY on %s\n",
1497 if (error == 0 && com == TIOCSCTTY) {
1498 struct proc *p = curthread->td_proc;
1499 struct session *sess;
1501 devfs_debug(DEVFS_DEBUG_DEBUG,
1502 "devfs_fo_ioctl: dealing with TIOCSCTTY on %s\n",
1508 sess = p->p_session;
1511 * Do nothing if reassigning same control tty
1513 if (sess->s_ttyvp == vp) {
1519 * Get rid of reference to old control tty
1521 ovp = sess->s_ttyvp;
1530 devfs_debug(DEVFS_DEBUG_DEBUG, "devfs_fo_ioctl() finished! \n");
1536 devfs_spec_fsync(struct vop_fsync_args *ap)
1538 struct vnode *vp = ap->a_vp;
1541 if (!vn_isdisk(vp, NULL))
1545 * Flush all dirty buffers associated with a block device.
1547 error = vfsync(vp, ap->a_waitfor, 10000, NULL, NULL);
1552 devfs_spec_read(struct vop_read_args *ap)
1554 struct devfs_node *node;
1563 node = DEVFS_NODE(vp);
1565 if (dev == NULL) /* device was revoked */
1567 if (uio->uio_resid == 0)
1571 error = dev_dread(dev, uio, ap->a_ioflag, NULL);
1572 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1575 nanotime(&node->atime);
1581 * Vnode op for write
1583 * spec_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
1584 * struct ucred *a_cred)
1587 devfs_spec_write(struct vop_write_args *ap)
1589 struct devfs_node *node;
1598 node = DEVFS_NODE(vp);
1600 KKASSERT(uio->uio_segflg != UIO_NOCOPY);
1602 if (dev == NULL) /* device was revoked */
1606 error = dev_dwrite(dev, uio, ap->a_ioflag, NULL);
1607 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1610 nanotime(&node->atime);
1611 nanotime(&node->mtime);
1618 * Device ioctl operation.
1620 * spec_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data,
1621 * int a_fflag, struct ucred *a_cred, struct sysmsg *msg)
1624 devfs_spec_ioctl(struct vop_ioctl_args *ap)
1626 struct vnode *vp = ap->a_vp;
1628 struct devfs_node *node;
1632 if ((dev = vp->v_rdev) == NULL)
1633 return (EBADF); /* device was revoked */
1635 node = DEVFS_NODE(vp);
1638 nanotime(&node->atime);
1639 nanotime(&node->mtime);
1643 return (dev_dioctl(dev, ap->a_command, ap->a_data, ap->a_fflag,
1644 ap->a_cred, ap->a_sysmsg, NULL));
1648 * spec_kqfilter(struct vnode *a_vp, struct knote *a_kn)
1652 devfs_spec_kqfilter(struct vop_kqfilter_args *ap)
1654 struct vnode *vp = ap->a_vp;
1656 struct devfs_node *node;
1660 if ((dev = vp->v_rdev) == NULL)
1661 return (EBADF); /* device was revoked (EBADF) */
1663 node = DEVFS_NODE(vp);
1666 nanotime(&node->atime);
1669 return (dev_dkqfilter(dev, ap->a_kn, NULL));
1673 * Convert a vnode strategy call into a device strategy call. Vnode strategy
1674 * calls are not limited to device DMA limits so we have to deal with the
1677 * spec_strategy(struct vnode *a_vp, struct bio *a_bio)
1680 devfs_spec_strategy(struct vop_strategy_args *ap)
1682 struct bio *bio = ap->a_bio;
1683 struct buf *bp = bio->bio_buf;
1690 if (bp->b_cmd != BUF_CMD_READ && LIST_FIRST(&bp->b_dep) != NULL)
1694 * Collect statistics on synchronous and asynchronous read
1695 * and write counts for disks that have associated filesystems.
1698 KKASSERT(vp->v_rdev != NULL); /* XXX */
1699 if (vn_isdisk(vp, NULL) && (mp = vp->v_rdev->si_mountpoint) != NULL) {
1700 if (bp->b_cmd == BUF_CMD_READ) {
1701 if (bp->b_flags & BIO_SYNC)
1702 mp->mnt_stat.f_syncreads++;
1704 mp->mnt_stat.f_asyncreads++;
1706 if (bp->b_flags & BIO_SYNC)
1707 mp->mnt_stat.f_syncwrites++;
1709 mp->mnt_stat.f_asyncwrites++;
1714 * Device iosize limitations only apply to read and write. Shortcut
1715 * the I/O if it fits.
1717 if ((maxiosize = vp->v_rdev->si_iosize_max) == 0) {
1718 devfs_debug(DEVFS_DEBUG_DEBUG,
1719 "%s: si_iosize_max not set!\n",
1720 dev_dname(vp->v_rdev));
1721 maxiosize = MAXPHYS;
1723 #if SPEC_CHAIN_DEBUG & 2
1726 if (bp->b_bcount <= maxiosize ||
1727 (bp->b_cmd != BUF_CMD_READ && bp->b_cmd != BUF_CMD_WRITE)) {
1728 dev_dstrategy_chain(vp->v_rdev, bio);
1733 * Clone the buffer and set up an I/O chain to chunk up the I/O.
1735 nbp = kmalloc(sizeof(*bp), M_DEVBUF, M_INTWAIT|M_ZERO);
1738 BUF_LOCK(nbp, LK_EXCLUSIVE);
1741 nbp->b_flags = B_PAGING | (bp->b_flags & B_BNOCLIP);
1742 nbp->b_data = bp->b_data;
1743 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1744 nbp->b_bio1.bio_offset = bio->bio_offset;
1745 nbp->b_bio1.bio_caller_info1.ptr = bio;
1748 * Start the first transfer
1750 if (vn_isdisk(vp, NULL))
1751 chunksize = vp->v_rdev->si_bsize_phys;
1753 chunksize = DEV_BSIZE;
1754 chunksize = maxiosize / chunksize * chunksize;
1755 #if SPEC_CHAIN_DEBUG & 1
1756 devfs_debug(DEVFS_DEBUG_DEBUG,
1757 "spec_strategy chained I/O chunksize=%d\n",
1760 nbp->b_cmd = bp->b_cmd;
1761 nbp->b_bcount = chunksize;
1762 nbp->b_bufsize = chunksize; /* used to detect a short I/O */
1763 nbp->b_bio1.bio_caller_info2.index = chunksize;
1765 #if SPEC_CHAIN_DEBUG & 1
1766 devfs_debug(DEVFS_DEBUG_DEBUG,
1767 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1768 bp, 0, bp->b_bcount, nbp->b_bcount);
1771 dev_dstrategy(vp->v_rdev, &nbp->b_bio1);
1773 if (DEVFS_NODE(vp)) {
1774 nanotime(&DEVFS_NODE(vp)->atime);
1775 nanotime(&DEVFS_NODE(vp)->mtime);
1782 * Chunked up transfer completion routine - chain transfers until done
1784 * NOTE: MPSAFE callback.
1788 devfs_spec_strategy_done(struct bio *nbio)
1790 struct buf *nbp = nbio->bio_buf;
1791 struct bio *bio = nbio->bio_caller_info1.ptr; /* original bio */
1792 struct buf *bp = bio->bio_buf; /* original bp */
1793 int chunksize = nbio->bio_caller_info2.index; /* chunking */
1794 int boffset = nbp->b_data - bp->b_data;
1796 if (nbp->b_flags & B_ERROR) {
1798 * An error terminates the chain, propogate the error back
1799 * to the original bp
1801 bp->b_flags |= B_ERROR;
1802 bp->b_error = nbp->b_error;
1803 bp->b_resid = bp->b_bcount - boffset +
1804 (nbp->b_bcount - nbp->b_resid);
1805 #if SPEC_CHAIN_DEBUG & 1
1806 devfs_debug(DEVFS_DEBUG_DEBUG,
1807 "spec_strategy: chain %p error %d bcount %d/%d\n",
1808 bp, bp->b_error, bp->b_bcount,
1809 bp->b_bcount - bp->b_resid);
1811 } else if (nbp->b_resid) {
1813 * A short read or write terminates the chain
1815 bp->b_error = nbp->b_error;
1816 bp->b_resid = bp->b_bcount - boffset +
1817 (nbp->b_bcount - nbp->b_resid);
1818 #if SPEC_CHAIN_DEBUG & 1
1819 devfs_debug(DEVFS_DEBUG_DEBUG,
1820 "spec_strategy: chain %p short read(1) "
1822 bp, bp->b_bcount - bp->b_resid, bp->b_bcount);
1824 } else if (nbp->b_bcount != nbp->b_bufsize) {
1826 * A short read or write can also occur by truncating b_bcount
1828 #if SPEC_CHAIN_DEBUG & 1
1829 devfs_debug(DEVFS_DEBUG_DEBUG,
1830 "spec_strategy: chain %p short read(2) "
1832 bp, nbp->b_bcount + boffset, bp->b_bcount);
1835 bp->b_bcount = nbp->b_bcount + boffset;
1836 bp->b_resid = nbp->b_resid;
1837 } else if (nbp->b_bcount + boffset == bp->b_bcount) {
1839 * No more data terminates the chain
1841 #if SPEC_CHAIN_DEBUG & 1
1842 devfs_debug(DEVFS_DEBUG_DEBUG,
1843 "spec_strategy: chain %p finished bcount %d\n",
1850 * Continue the chain
1852 boffset += nbp->b_bcount;
1853 nbp->b_data = bp->b_data + boffset;
1854 nbp->b_bcount = bp->b_bcount - boffset;
1855 if (nbp->b_bcount > chunksize)
1856 nbp->b_bcount = chunksize;
1857 nbp->b_bio1.bio_done = devfs_spec_strategy_done;
1858 nbp->b_bio1.bio_offset = bio->bio_offset + boffset;
1860 #if SPEC_CHAIN_DEBUG & 1
1861 devfs_debug(DEVFS_DEBUG_DEBUG,
1862 "spec_strategy: chain %p offset %d/%d bcount %d\n",
1863 bp, boffset, bp->b_bcount, nbp->b_bcount);
1866 dev_dstrategy(nbp->b_vp->v_rdev, &nbp->b_bio1);
1871 * Fall through to here on termination. biodone(bp) and
1872 * clean up and free nbp.
1877 kfree(nbp, M_DEVBUF);
1881 * spec_freeblks(struct vnode *a_vp, daddr_t a_addr, daddr_t a_length)
1884 devfs_spec_freeblks(struct vop_freeblks_args *ap)
1889 * Must be a synchronous operation
1891 KKASSERT(ap->a_vp->v_rdev != NULL);
1892 if ((ap->a_vp->v_rdev->si_flags & SI_CANFREE) == 0)
1894 bp = geteblk(ap->a_length);
1895 bp->b_cmd = BUF_CMD_FREEBLKS;
1896 bp->b_bio1.bio_flags |= BIO_SYNC;
1897 bp->b_bio1.bio_offset = ap->a_offset;
1898 bp->b_bio1.bio_done = biodone_sync;
1899 bp->b_bcount = ap->a_length;
1900 dev_dstrategy(ap->a_vp->v_rdev, &bp->b_bio1);
1901 biowait(&bp->b_bio1, "TRIM");
1908 * Implement degenerate case where the block requested is the block
1909 * returned, and assume that the entire device is contiguous in regards
1910 * to the contiguous block range (runp and runb).
1912 * spec_bmap(struct vnode *a_vp, off_t a_loffset,
1913 * off_t *a_doffsetp, int *a_runp, int *a_runb)
1916 devfs_spec_bmap(struct vop_bmap_args *ap)
1918 if (ap->a_doffsetp != NULL)
1919 *ap->a_doffsetp = ap->a_loffset;
1920 if (ap->a_runp != NULL)
1921 *ap->a_runp = MAXBSIZE;
1922 if (ap->a_runb != NULL) {
1923 if (ap->a_loffset < MAXBSIZE)
1924 *ap->a_runb = (int)ap->a_loffset;
1926 *ap->a_runb = MAXBSIZE;
1933 * Special device advisory byte-level locks.
1935 * spec_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
1936 * struct flock *a_fl, int a_flags)
1940 devfs_spec_advlock(struct vop_advlock_args *ap)
1942 return ((ap->a_flags & F_POSIX) ? EINVAL : EOPNOTSUPP);
1946 * NOTE: MPSAFE callback.
1949 devfs_spec_getpages_iodone(struct bio *bio)
1951 bio->bio_buf->b_cmd = BUF_CMD_DONE;
1952 wakeup(bio->bio_buf);
1956 * spec_getpages() - get pages associated with device vnode.
1958 * Note that spec_read and spec_write do not use the buffer cache, so we
1959 * must fully implement getpages here.
1962 devfs_spec_getpages(struct vop_getpages_args *ap)
1966 int i, pcount, size;
1969 vm_ooffset_t offset;
1970 int toff, nextoff, nread;
1971 struct vnode *vp = ap->a_vp;
1976 pcount = round_page(ap->a_count) / PAGE_SIZE;
1979 * Calculate the offset of the transfer and do sanity check.
1981 offset = IDX_TO_OFF(ap->a_m[0]->pindex) + ap->a_offset;
1984 * Round up physical size for real devices. We cannot round using
1985 * v_mount's block size data because v_mount has nothing to do with
1986 * the device. i.e. it's usually '/dev'. We need the physical block
1987 * size for the device itself.
1989 * We can't use v_rdev->si_mountpoint because it only exists when the
1990 * block device is mounted. However, we can use v_rdev.
1992 if (vn_isdisk(vp, NULL))
1993 blksiz = vp->v_rdev->si_bsize_phys;
1997 size = roundup2(ap->a_count, blksiz);
1999 bp = getpbuf_kva(NULL);
2000 kva = (vm_offset_t)bp->b_data;
2003 * Map the pages to be read into the kva.
2005 pmap_qenter(kva, ap->a_m, pcount);
2007 /* Build a minimal buffer header. */
2008 bp->b_cmd = BUF_CMD_READ;
2009 bp->b_bcount = size;
2011 bsetrunningbufspace(bp, size);
2013 bp->b_bio1.bio_offset = offset;
2014 bp->b_bio1.bio_done = devfs_spec_getpages_iodone;
2016 mycpu->gd_cnt.v_vnodein++;
2017 mycpu->gd_cnt.v_vnodepgsin += pcount;
2020 vn_strategy(ap->a_vp, &bp->b_bio1);
2024 /* We definitely need to be at splbio here. */
2025 while (bp->b_cmd != BUF_CMD_DONE)
2026 tsleep(bp, 0, "spread", 0);
2030 if (bp->b_flags & B_ERROR) {
2032 error = bp->b_error;
2038 * If EOF is encountered we must zero-extend the result in order
2039 * to ensure that the page does not contain garabge. When no
2040 * error occurs, an early EOF is indicated if b_bcount got truncated.
2041 * b_resid is relative to b_bcount and should be 0, but some devices
2042 * might indicate an EOF with b_resid instead of truncating b_bcount.
2044 nread = bp->b_bcount - bp->b_resid;
2045 if (nread < ap->a_count)
2046 bzero((caddr_t)kva + nread, ap->a_count - nread);
2047 pmap_qremove(kva, pcount);
2050 for (i = 0, toff = 0; i < pcount; i++, toff = nextoff) {
2051 nextoff = toff + PAGE_SIZE;
2055 * NOTE: vm_page_undirty/clear_dirty etc do not clear the
2056 * pmap modified bit. pmap modified bit should have
2057 * already been cleared.
2059 if (nextoff <= nread) {
2060 m->valid = VM_PAGE_BITS_ALL;
2062 } else if (toff < nread) {
2064 * Since this is a VM request, we have to supply the
2065 * unaligned offset to allow vm_page_set_valid()
2066 * to zero sub-DEV_BSIZE'd portions of the page.
2068 vm_page_set_valid(m, 0, nread - toff);
2069 vm_page_clear_dirty_end_nonincl(m, 0, nread - toff);
2075 if (i != ap->a_reqpage) {
2077 * Just in case someone was asking for this page we
2078 * now tell them that it is ok to use.
2080 if (!error || (m->valid == VM_PAGE_BITS_ALL)) {
2082 if (m->flags & PG_REFERENCED) {
2083 vm_page_activate(m);
2085 vm_page_deactivate(m);
2094 } else if (m->valid) {
2097 * Since this is a VM request, we need to make the
2098 * entire page presentable by zeroing invalid sections.
2100 if (m->valid != VM_PAGE_BITS_ALL)
2101 vm_page_zero_invalid(m, FALSE);
2105 m = ap->a_m[ap->a_reqpage];
2106 devfs_debug(DEVFS_DEBUG_WARNING,
2107 "spec_getpages:(%s) I/O read failure: (error=%d) bp %p vp %p\n",
2108 devtoname(vp->v_rdev), error, bp, bp->b_vp);
2109 devfs_debug(DEVFS_DEBUG_WARNING,
2110 " size: %d, resid: %d, a_count: %d, valid: 0x%x\n",
2111 size, bp->b_resid, ap->a_count, m->valid);
2112 devfs_debug(DEVFS_DEBUG_WARNING,
2113 " nread: %d, reqpage: %d, pindex: %lu, pcount: %d\n",
2114 nread, ap->a_reqpage, (u_long)m->pindex, pcount);
2116 * Free the buffer header back to the swap buffer pool.
2119 return VM_PAGER_ERROR;
2122 * Free the buffer header back to the swap buffer pool.
2125 if (DEVFS_NODE(ap->a_vp))
2126 nanotime(&DEVFS_NODE(ap->a_vp)->mtime);
2132 sequential_heuristic(struct uio *uio, struct file *fp)
2135 * Sequential heuristic - detect sequential operation
2137 if ((uio->uio_offset == 0 && fp->f_seqcount > 0) ||
2138 uio->uio_offset == fp->f_nextoff) {
2140 * XXX we assume that the filesystem block size is
2141 * the default. Not true, but still gives us a pretty
2142 * good indicator of how sequential the read operations
2145 int tmpseq = fp->f_seqcount;
2147 tmpseq += (uio->uio_resid + MAXBSIZE - 1) / MAXBSIZE;
2148 if (tmpseq > IO_SEQMAX)
2150 fp->f_seqcount = tmpseq;
2151 return(fp->f_seqcount << IO_SEQSHIFT);
2155 * Not sequential, quick draw-down of seqcount
2157 if (fp->f_seqcount > 1)