1/*	$NetBSD: coda_vnops.c,v 1.119 2024/05/14 19:00:44 andvar Exp $	*/
2
3/*
4 *
5 *             Coda: an Experimental Distributed File System
6 *                              Release 3.1
7 *
8 *           Copyright (c) 1987-1998 Carnegie Mellon University
9 *                          All Rights Reserved
10 *
11 * Permission  to  use, copy, modify and distribute this software and its
12 * documentation is hereby granted,  provided  that  both  the  copyright
13 * notice  and  this  permission  notice  appear  in  all  copies  of the
14 * software, derivative works or  modified  versions,  and  any  portions
15 * thereof, and that both notices appear in supporting documentation, and
16 * that credit is given to Carnegie Mellon University  in  all  documents
17 * and publicity pertaining to direct or indirect use of this code or its
18 * derivatives.
19 *
20 * CODA IS AN EXPERIMENTAL SOFTWARE SYSTEM AND IS  KNOWN  TO  HAVE  BUGS,
21 * SOME  OF  WHICH MAY HAVE SERIOUS CONSEQUENCES.  CARNEGIE MELLON ALLOWS
22 * FREE USE OF THIS SOFTWARE IN ITS "AS IS" CONDITION.   CARNEGIE  MELLON
23 * DISCLAIMS  ANY  LIABILITY  OF  ANY  KIND  FOR  ANY  DAMAGES WHATSOEVER
24 * RESULTING DIRECTLY OR INDIRECTLY FROM THE USE OF THIS SOFTWARE  OR  OF
25 * ANY DERIVATIVE WORK.
26 *
27 * Carnegie  Mellon  encourages  users  of  this  software  to return any
28 * improvements or extensions that  they  make,  and  to  grant  Carnegie
29 * Mellon the rights to redistribute these changes without encumbrance.
30 *
31 * 	@(#) coda/coda_vnops.c,v 1.1.1.1 1998/08/29 21:26:46 rvb Exp $
32 */
33
34/*
35 * Mach Operating System
36 * Copyright (c) 1990 Carnegie-Mellon University
37 * Copyright (c) 1989 Carnegie-Mellon University
38 * All rights reserved.  The CMU software License Agreement specifies
39 * the terms and conditions for use and redistribution.
40 */
41
42/*
43 * This code was written for the Coda file system at Carnegie Mellon
44 * University.  Contributers include David Steere, James Kistler, and
45 * M. Satyanarayanan.
46 */
47
48#include <sys/cdefs.h>
49__KERNEL_RCSID(0, "$NetBSD: coda_vnops.c,v 1.119 2024/05/14 19:00:44 andvar Exp $");
50
51#include <sys/param.h>
52#include <sys/systm.h>
53#include <sys/malloc.h>
54#include <sys/errno.h>
55#include <sys/acct.h>
56#include <sys/file.h>
57#include <sys/uio.h>
58#include <sys/namei.h>
59#include <sys/ioctl.h>
60#include <sys/mount.h>
61#include <sys/proc.h>
62#include <sys/select.h>
63#include <sys/vnode.h>
64#include <sys/kauth.h>
65#include <sys/dirent.h>
66
67#include <miscfs/genfs/genfs.h>
68#include <miscfs/specfs/specdev.h>
69
70#include <coda/coda.h>
71#include <coda/cnode.h>
72#include <coda/coda_vnops.h>
73#include <coda/coda_venus.h>
74#include <coda/coda_opstats.h>
75#include <coda/coda_subr.h>
76#include <coda/coda_namecache.h>
77#include <coda/coda_pioctl.h>
78
79/*
80 * These flags select various performance enhancements.
81 */
82int coda_attr_cache  = 1;       /* Set to cache attributes in the kernel */
83int coda_symlink_cache = 1;     /* Set to cache symbolic link information */
84int coda_access_cache = 1;      /* Set to handle some access checks directly */
85
86/* structure to keep track of vfs calls */
87
88struct coda_op_stats coda_vnodeopstats[CODA_VNODEOPS_SIZE];
89
90#define MARK_ENTRY(op) (coda_vnodeopstats[op].entries++)
91#define MARK_INT_SAT(op) (coda_vnodeopstats[op].sat_intrn++)
92#define MARK_INT_FAIL(op) (coda_vnodeopstats[op].unsat_intrn++)
93#define MARK_INT_GEN(op) (coda_vnodeopstats[op].gen_intrn++)
94
95/* What we are delaying for in printf */
96static int coda_lockdebug = 0;
97
98#define ENTRY if(coda_vnop_print_entry) myprintf(("Entered %s\n",__func__))
99
100/* Definition of the vnode operation vector */
101
102const struct vnodeopv_entry_desc coda_vnodeop_entries[] = {
103    { &vop_default_desc, coda_vop_error },
104    { &vop_parsepath_desc, genfs_parsepath },	/* parsepath */
105    { &vop_lookup_desc, coda_lookup },		/* lookup */
106    { &vop_create_desc, coda_create },		/* create */
107    { &vop_mknod_desc, coda_vop_error },	/* mknod */
108    { &vop_open_desc, coda_open },		/* open */
109    { &vop_close_desc, coda_close },		/* close */
110    { &vop_access_desc, coda_access },		/* access */
111    { &vop_accessx_desc, genfs_accessx },	/* access */
112    { &vop_getattr_desc, coda_getattr },	/* getattr */
113    { &vop_setattr_desc, coda_setattr },	/* setattr */
114    { &vop_read_desc, coda_read },		/* read */
115    { &vop_write_desc, coda_write },		/* write */
116    { &vop_fallocate_desc, genfs_eopnotsupp },	/* fallocate */
117    { &vop_fdiscard_desc, genfs_eopnotsupp },	/* fdiscard */
118    { &vop_fcntl_desc, genfs_fcntl },		/* fcntl */
119    { &vop_ioctl_desc, coda_ioctl },		/* ioctl */
120    { &vop_mmap_desc, genfs_mmap },		/* mmap */
121    { &vop_fsync_desc, coda_fsync },		/* fsync */
122    { &vop_remove_desc, coda_remove },		/* remove */
123    { &vop_link_desc, coda_link },		/* link */
124    { &vop_rename_desc, coda_rename },		/* rename */
125    { &vop_mkdir_desc, coda_mkdir },		/* mkdir */
126    { &vop_rmdir_desc, coda_rmdir },		/* rmdir */
127    { &vop_symlink_desc, coda_symlink },	/* symlink */
128    { &vop_readdir_desc, coda_readdir },	/* readdir */
129    { &vop_readlink_desc, coda_readlink },	/* readlink */
130    { &vop_abortop_desc, coda_abortop },	/* abortop */
131    { &vop_inactive_desc, coda_inactive },	/* inactive */
132    { &vop_reclaim_desc, coda_reclaim },	/* reclaim */
133    { &vop_lock_desc, coda_lock },		/* lock */
134    { &vop_unlock_desc, coda_unlock },		/* unlock */
135    { &vop_bmap_desc, coda_bmap },		/* bmap */
136    { &vop_strategy_desc, coda_strategy },	/* strategy */
137    { &vop_print_desc, coda_vop_error },	/* print */
138    { &vop_islocked_desc, coda_islocked },	/* islocked */
139    { &vop_pathconf_desc, coda_pathconf },	/* pathconf */
140    { &vop_advlock_desc, coda_vop_nop },	/* advlock */
141    { &vop_bwrite_desc, coda_vop_error },	/* bwrite */
142    { &vop_seek_desc, genfs_seek },		/* seek */
143    { &vop_poll_desc, genfs_poll },		/* poll */
144    { &vop_getpages_desc, coda_getpages },	/* getpages */
145    { &vop_putpages_desc, coda_putpages },	/* putpages */
146    { NULL, NULL }
147};
148
149static void coda_print_vattr(struct vattr *);
150
151int (**coda_vnodeop_p)(void *);
152const struct vnodeopv_desc coda_vnodeop_opv_desc =
153        { &coda_vnodeop_p, coda_vnodeop_entries };
154
155/* Definitions of NetBSD vnodeop interfaces */
156
157/*
158 * A generic error routine.  Return EIO without looking at arguments.
159 */
160int
161coda_vop_error(void *anon) {
162    struct vnodeop_desc **desc = (struct vnodeop_desc **)anon;
163
164    if (codadebug) {
165	myprintf(("%s: Vnode operation %s called (error).\n",
166	    __func__, (*desc)->vdesc_name));
167    }
168
169    return EIO;
170}
171
172/* A generic do-nothing. */
173int
174coda_vop_nop(void *anon) {
175    struct vnodeop_desc **desc = (struct vnodeop_desc **)anon;
176
177    if (codadebug) {
178	myprintf(("Vnode operation %s called, but unsupported\n",
179		  (*desc)->vdesc_name));
180    }
181   return (0);
182}
183
184int
185coda_vnodeopstats_init(void)
186{
187	int i;
188
189	for(i=0;i<CODA_VNODEOPS_SIZE;i++) {
190		coda_vnodeopstats[i].opcode = i;
191		coda_vnodeopstats[i].entries = 0;
192		coda_vnodeopstats[i].sat_intrn = 0;
193		coda_vnodeopstats[i].unsat_intrn = 0;
194		coda_vnodeopstats[i].gen_intrn = 0;
195	}
196
197	return 0;
198}
199
200/*
201 * XXX The entire relationship between VOP_OPEN and having a container
202 * file (via venus_open) needs to be reexamined.  In particular, it's
203 * valid to open/mmap/close and then reference.  Instead of doing
204 * VOP_OPEN when getpages needs a container, we should do the
205 * venus_open part, and record that the vnode has opened the container
206 * for getpages, and do the matching logical close on coda_inactive.
207 * Further, coda_rdwr needs a container file, and sometimes needs to
208 * do the equivalent of open (core dumps).
209 */
210/*
211 * coda_open calls Venus to return the device and inode of the
212 * container file, and then obtains a vnode for that file.  The
213 * container vnode is stored in the coda vnode, and a reference is
214 * added for each open file.
215 */
216int
217coda_open(void *v)
218{
219    /*
220     * NetBSD can pass the O_EXCL flag in mode, even though the check
221     * has already happened.  Venus defensively assumes that if open
222     * is passed the EXCL, it must be a bug.  We strip the flag here.
223     */
224/* true args */
225    struct vop_open_args *ap = v;
226    vnode_t *vp = ap->a_vp;
227    struct cnode *cp = VTOC(vp);
228    int flag = ap->a_mode & (~O_EXCL);
229    kauth_cred_t cred = ap->a_cred;
230/* locals */
231    int error;
232    dev_t dev;			/* container file device, inode, vnode */
233    ino_t inode;
234    vnode_t *container_vp;
235
236    MARK_ENTRY(CODA_OPEN_STATS);
237
238    KASSERT(VOP_ISLOCKED(vp));
239    /* Check for open of control file. */
240    if (IS_CTL_VP(vp)) {
241	/* if (WRITABLE(flag)) */
242	if (flag & (FWRITE | O_TRUNC | O_CREAT | O_EXCL)) {
243	    MARK_INT_FAIL(CODA_OPEN_STATS);
244	    return(EACCES);
245	}
246	MARK_INT_SAT(CODA_OPEN_STATS);
247	return(0);
248    }
249
250    error = venus_open(vtomi(vp), &cp->c_fid, flag, cred, curlwp, &dev, &inode);
251    if (error)
252	return (error);
253    if (!error) {
254	    CODADEBUG(CODA_OPEN, myprintf((
255		"%s: dev 0x%llx inode %llu result %d\n", __func__,
256		(unsigned long long)dev, (unsigned long long)inode, error));)
257    }
258
259    /*
260     * Obtain locked and referenced container vnode from container
261     * device/inode.
262     */
263    error = coda_grab_vnode(vp, dev, inode, &container_vp);
264    if (error)
265	return (error);
266
267    /* Save the vnode pointer for the container file. */
268    if (cp->c_ovp == NULL) {
269	cp->c_ovp = container_vp;
270    } else {
271	if (cp->c_ovp != container_vp)
272	    /*
273	     * Perhaps venus returned a different container, or
274	     * something else went wrong.
275	     */
276	    panic("%s: cp->c_ovp != container_vp", __func__);
277    }
278    cp->c_ocount++;
279
280    /* Flush the attribute cache if writing the file. */
281    if (flag & FWRITE) {
282	cp->c_owrite++;
283	cp->c_flags &= ~C_VATTR;
284    }
285
286    /*
287     * Save the <device, inode> pair for the container file to speed
288     * up subsequent reads while closed (mmap, program execution).
289     * This is perhaps safe because venus will invalidate the node
290     * before changing the container file mapping.
291     */
292    cp->c_device = dev;
293    cp->c_inode = inode;
294
295    /* Open the container file. */
296    error = VOP_OPEN(container_vp, flag, cred);
297    /*
298     * Drop the lock on the container, after we have done VOP_OPEN
299     * (which requires a locked vnode).
300     */
301    VOP_UNLOCK(container_vp);
302    return(error);
303}
304
305/*
306 * Close the cache file used for I/O and notify Venus.
307 */
308int
309coda_close(void *v)
310{
311/* true args */
312    struct vop_close_args *ap = v;
313    vnode_t *vp = ap->a_vp;
314    struct cnode *cp = VTOC(vp);
315    int flag = ap->a_fflag;
316    kauth_cred_t cred = ap->a_cred;
317/* locals */
318    int error;
319
320    MARK_ENTRY(CODA_CLOSE_STATS);
321
322    /* Check for close of control file. */
323    if (IS_CTL_VP(vp)) {
324	MARK_INT_SAT(CODA_CLOSE_STATS);
325	return(0);
326    }
327
328    /*
329     * XXX The IS_UNMOUNTING part of this is very suspect.
330     */
331    if (IS_UNMOUNTING(cp)) {
332	if (cp->c_ovp) {
333#ifdef	CODA_VERBOSE
334	    printf("%s: destroying container %d, ufs vp %p of vp %p/cp %p\n",
335		__func__, vrefcnt(vp), cp->c_ovp, vp, cp);
336#endif
337#ifdef	hmm
338	    vgone(cp->c_ovp);
339#else
340	    vn_lock(cp->c_ovp, LK_EXCLUSIVE | LK_RETRY);
341	    VOP_CLOSE(cp->c_ovp, flag, cred); /* Do errors matter here? */
342	    vput(cp->c_ovp);
343#endif
344	} else {
345#ifdef	CODA_VERBOSE
346	    printf("%s: NO container vp %p/cp %p\n", __func__, vp, cp);
347#endif
348	}
349	return ENODEV;
350    }
351
352    /* Lock the container node, and VOP_CLOSE it. */
353    vn_lock(cp->c_ovp, LK_EXCLUSIVE | LK_RETRY);
354    VOP_CLOSE(cp->c_ovp, flag, cred); /* Do errors matter here? */
355    /*
356     * Drop the lock we just obtained, and vrele the container vnode.
357     * Decrement reference counts, and clear container vnode pointer on
358     * last close.
359     */
360    vput(cp->c_ovp);
361    if (flag & FWRITE)
362	--cp->c_owrite;
363    if (--cp->c_ocount == 0)
364	cp->c_ovp = NULL;
365
366    error = venus_close(vtomi(vp), &cp->c_fid, flag, cred, curlwp);
367
368    CODADEBUG(CODA_CLOSE, myprintf(("%s: result %d\n", __func__, error)); )
369    return(error);
370}
371
372int
373coda_read(void *v)
374{
375    struct vop_read_args *ap = v;
376
377    ENTRY;
378    return(coda_rdwr(ap->a_vp, ap->a_uio, UIO_READ,
379		    ap->a_ioflag, ap->a_cred, curlwp));
380}
381
382int
383coda_write(void *v)
384{
385    struct vop_write_args *ap = v;
386
387    ENTRY;
388    return(coda_rdwr(ap->a_vp, ap->a_uio, UIO_WRITE,
389		    ap->a_ioflag, ap->a_cred, curlwp));
390}
391
392int
393coda_rdwr(vnode_t *vp, struct uio *uiop, enum uio_rw rw, int ioflag,
394	kauth_cred_t cred, struct lwp *l)
395{
396/* upcall decl */
397  /* NOTE: container file operation!!! */
398/* locals */
399    struct cnode *cp = VTOC(vp);
400    vnode_t *cfvp = cp->c_ovp;
401    struct proc *p = l->l_proc;
402    int opened_internally = 0;
403    int error = 0;
404
405    MARK_ENTRY(CODA_RDWR_STATS);
406
407    CODADEBUG(CODA_RDWR, myprintf(("coda_rdwr(%d, %p, %lu, %lld)\n", rw,
408	uiop->uio_iov->iov_base, (unsigned long) uiop->uio_resid,
409	(long long) uiop->uio_offset)); )
410
411    /* Check for rdwr of control object. */
412    if (IS_CTL_VP(vp)) {
413	MARK_INT_FAIL(CODA_RDWR_STATS);
414	return(EINVAL);
415    }
416
417    /* Redirect the request to UFS. */
418
419    /*
420     * If file is not already open this must be a page
421     * {read,write} request.  Iget the cache file's inode
422     * pointer if we still have its <device, inode> pair.
423     * Otherwise, we must do an internal open to derive the
424     * pair.
425     * XXX Integrate this into a coherent strategy for container
426     * file acquisition.
427     */
428    if (cfvp == NULL) {
429	/*
430	 * If we're dumping core, do the internal open. Otherwise
431	 * venus won't have the correct size of the core when
432	 * it's completely written.
433	 */
434	if (cp->c_inode != 0 && !(p && (p->p_acflag & ACORE))) {
435#ifdef CODA_VERBOSE
436	    printf("%s: grabbing container vnode, losing reference\n",
437		__func__);
438#endif
439	    /* Get locked and refed vnode. */
440	    error = coda_grab_vnode(vp, cp->c_device, cp->c_inode, &cfvp);
441	    if (error) {
442		MARK_INT_FAIL(CODA_RDWR_STATS);
443		return(error);
444	    }
445	    /*
446	     * Drop lock.
447	     * XXX Where is reference released.
448	     */
449	    VOP_UNLOCK(cfvp);
450	}
451	else {
452#ifdef CODA_VERBOSE
453	    printf("%s: internal VOP_OPEN\n", __func__);
454#endif
455	    opened_internally = 1;
456	    MARK_INT_GEN(CODA_OPEN_STATS);
457	    error = VOP_OPEN(vp, (rw == UIO_READ ? FREAD : FWRITE), cred);
458#ifdef	CODA_VERBOSE
459	    printf("%s: Internally Opening %p\n", __func__, vp);
460#endif
461	    if (error) {
462		MARK_INT_FAIL(CODA_RDWR_STATS);
463		return(error);
464	    }
465	    cfvp = cp->c_ovp;
466	}
467    }
468
469    /* Have UFS handle the call. */
470    CODADEBUG(CODA_RDWR, myprintf(("%s: fid = %s, refcnt = %d\n", __func__,
471	coda_f2s(&cp->c_fid), vrefcnt(CTOV(cp)))); )
472
473    if (rw == UIO_READ) {
474	error = VOP_READ(cfvp, uiop, ioflag, cred);
475    } else {
476	error = VOP_WRITE(cfvp, uiop, ioflag, cred);
477    }
478
479    if (error)
480	MARK_INT_FAIL(CODA_RDWR_STATS);
481    else
482	MARK_INT_SAT(CODA_RDWR_STATS);
483
484    /* Do an internal close if necessary. */
485    if (opened_internally) {
486	MARK_INT_GEN(CODA_CLOSE_STATS);
487	(void)VOP_CLOSE(vp, (rw == UIO_READ ? FREAD : FWRITE), cred);
488    }
489
490    /* Invalidate cached attributes if writing. */
491    if (rw == UIO_WRITE)
492	cp->c_flags &= ~C_VATTR;
493    return(error);
494}
495
496int
497coda_ioctl(void *v)
498{
499/* true args */
500    struct vop_ioctl_args *ap = v;
501    vnode_t *vp = ap->a_vp;
502    int com = ap->a_command;
503    void *data = ap->a_data;
504    int flag = ap->a_fflag;
505    kauth_cred_t cred = ap->a_cred;
506/* locals */
507    int error;
508    vnode_t *tvp;
509    struct PioctlData *iap = (struct PioctlData *)data;
510    namei_simple_flags_t sflags;
511
512    MARK_ENTRY(CODA_IOCTL_STATS);
513
514    CODADEBUG(CODA_IOCTL, myprintf(("in coda_ioctl on %s\n", iap->path));)
515
516    /* Don't check for operation on a dying object, for ctlvp it
517       shouldn't matter */
518
519    /* Must be control object to succeed. */
520    if (!IS_CTL_VP(vp)) {
521	MARK_INT_FAIL(CODA_IOCTL_STATS);
522	CODADEBUG(CODA_IOCTL, myprintf(("%s error: vp != ctlvp", __func__));)
523	return (EOPNOTSUPP);
524    }
525    /* Look up the pathname. */
526
527    /* Should we use the name cache here? It would get it from
528       lookupname sooner or later anyway, right? */
529
530    sflags = iap->follow ? NSM_FOLLOW_NOEMULROOT : NSM_NOFOLLOW_NOEMULROOT;
531    error = namei_simple_user(iap->path, sflags, &tvp);
532
533    if (error) {
534	MARK_INT_FAIL(CODA_IOCTL_STATS);
535	CODADEBUG(CODA_IOCTL, myprintf(("%s error: lookup returns %d\n",
536	    __func__, error));)
537	return(error);
538    }
539
540    /*
541     * Make sure this is a coda style cnode, but it may be a
542     * different vfsp
543     */
544    /* XXX: this totally violates the comment about vtagtype in vnode.h */
545    if (tvp->v_tag != VT_CODA) {
546	vrele(tvp);
547	MARK_INT_FAIL(CODA_IOCTL_STATS);
548	CODADEBUG(CODA_IOCTL, myprintf(("%s error: %s not a coda object\n",
549	    __func__, iap->path));)
550	return(EINVAL);
551    }
552
553    if (iap->vi.in_size > VC_MAXDATASIZE || iap->vi.out_size > VC_MAXDATASIZE) {
554	vrele(tvp);
555	return(EINVAL);
556    }
557    error = venus_ioctl(vtomi(tvp), &((VTOC(tvp))->c_fid), com, flag, data,
558	cred, curlwp);
559
560    if (error)
561	MARK_INT_FAIL(CODA_IOCTL_STATS);
562    else
563	CODADEBUG(CODA_IOCTL, myprintf(("Ioctl returns %d \n", error)); )
564
565    vrele(tvp);
566    return(error);
567}
568
569/*
570 * To reduce the cost of a user-level venus;we cache attributes in
571 * the kernel.  Each cnode has storage allocated for an attribute. If
572 * c_vattr is valid, return a reference to it. Otherwise, get the
573 * attributes from venus and store them in the cnode.  There is some
574 * question if this method is a security leak. But I think that in
575 * order to make this call, the user must have done a lookup and
576 * opened the file, and therefore should already have access.
577 */
578int
579coda_getattr(void *v)
580{
581/* true args */
582    struct vop_getattr_args *ap = v;
583    vnode_t *vp = ap->a_vp;
584    struct cnode *cp = VTOC(vp);
585    struct vattr *vap = ap->a_vap;
586    kauth_cred_t cred = ap->a_cred;
587/* locals */
588    int error;
589
590    MARK_ENTRY(CODA_GETATTR_STATS);
591
592    /* Check for getattr of control object. */
593    if (IS_CTL_VP(vp)) {
594	MARK_INT_FAIL(CODA_GETATTR_STATS);
595	return(ENOENT);
596    }
597
598    /* Check to see if the attributes have already been cached */
599    if (VALID_VATTR(cp)) {
600	CODADEBUG(CODA_GETATTR, { myprintf(("%s: attr cache hit: %s\n",
601	    __func__, coda_f2s(&cp->c_fid)));})
602	CODADEBUG(CODA_GETATTR, if (!(codadebug & ~CODA_GETATTR))
603	    coda_print_vattr(&cp->c_vattr); )
604
605	*vap = cp->c_vattr;
606	MARK_INT_SAT(CODA_GETATTR_STATS);
607	return(0);
608    }
609
610    error = venus_getattr(vtomi(vp), &cp->c_fid, cred, curlwp, vap);
611
612    if (!error) {
613	CODADEBUG(CODA_GETATTR, myprintf(("%s miss %s: result %d\n",
614	    __func__, coda_f2s(&cp->c_fid), error)); )
615
616	CODADEBUG(CODA_GETATTR, if (!(codadebug & ~CODA_GETATTR))
617	    coda_print_vattr(vap);	)
618
619	/* If not open for write, store attributes in cnode */
620	if ((cp->c_owrite == 0) && (coda_attr_cache)) {
621	    cp->c_vattr = *vap;
622	    cp->c_flags |= C_VATTR;
623	}
624
625    }
626    return(error);
627}
628
629int
630coda_setattr(void *v)
631{
632/* true args */
633    struct vop_setattr_args *ap = v;
634    vnode_t *vp = ap->a_vp;
635    struct cnode *cp = VTOC(vp);
636    struct vattr *vap = ap->a_vap;
637    kauth_cred_t cred = ap->a_cred;
638/* locals */
639    int error;
640
641    MARK_ENTRY(CODA_SETATTR_STATS);
642
643    /* Check for setattr of control object. */
644    if (IS_CTL_VP(vp)) {
645	MARK_INT_FAIL(CODA_SETATTR_STATS);
646	return(ENOENT);
647    }
648
649    if (codadebug & CODADBGMSK(CODA_SETATTR)) {
650	coda_print_vattr(vap);
651    }
652    error = venus_setattr(vtomi(vp), &cp->c_fid, vap, cred, curlwp);
653
654    if (!error)
655	cp->c_flags &= ~C_VATTR;
656
657    CODADEBUG(CODA_SETATTR,	myprintf(("setattr %d\n", error)); )
658    return(error);
659}
660
661int
662coda_access(void *v)
663{
664/* true args */
665    struct vop_access_args *ap = v;
666    vnode_t *vp = ap->a_vp;
667    struct cnode *cp = VTOC(vp);
668    accmode_t accmode = ap->a_accmode;
669    kauth_cred_t cred = ap->a_cred;
670/* locals */
671    int error;
672
673    MARK_ENTRY(CODA_ACCESS_STATS);
674
675    KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0);
676    /* Check for access of control object.  Only read access is
677       allowed on it. */
678    if (IS_CTL_VP(vp)) {
679	/* bogus hack - all will be marked as successes */
680	MARK_INT_SAT(CODA_ACCESS_STATS);
681	return(((accmode & VREAD) && !(accmode & (VWRITE | VEXEC)))
682	       ? 0 : EACCES);
683    }
684
685    /*
686     * if the file is a directory, and we are checking exec (eg lookup)
687     * access, and the file is in the namecache, then the user must have
688     * lookup access to it.
689     */
690    if (coda_access_cache) {
691	if ((vp->v_type == VDIR) && (accmode & VEXEC)) {
692	    if (coda_nc_lookup(cp, ".", 1, cred)) {
693		MARK_INT_SAT(CODA_ACCESS_STATS);
694		return(0);                     /* it was in the cache */
695	    }
696	}
697    }
698
699    error = venus_access(vtomi(vp), &cp->c_fid, accmode, cred, curlwp);
700
701    return(error);
702}
703
704/*
705 * CODA abort op, called after namei() when a CREATE/DELETE isn't actually
706 * done. If a buffer has been saved in anticipation of a coda_create or
707 * a coda_remove, delete it.
708 */
709/* ARGSUSED */
710int
711coda_abortop(void *v)
712{
713/* true args */
714    struct vop_abortop_args /* {
715	vnode_t *a_dvp;
716	struct componentname *a_cnp;
717    } */ *ap = v;
718
719    (void)ap;
720/* upcall decl */
721/* locals */
722
723    return (0);
724}
725
726int
727coda_readlink(void *v)
728{
729/* true args */
730    struct vop_readlink_args *ap = v;
731    vnode_t *vp = ap->a_vp;
732    struct cnode *cp = VTOC(vp);
733    struct uio *uiop = ap->a_uio;
734    kauth_cred_t cred = ap->a_cred;
735/* locals */
736    struct lwp *l = curlwp;
737    int error;
738    char *str;
739    int len;
740
741    MARK_ENTRY(CODA_READLINK_STATS);
742
743    /* Check for readlink of control object. */
744    if (IS_CTL_VP(vp)) {
745	MARK_INT_FAIL(CODA_READLINK_STATS);
746	return(ENOENT);
747    }
748
749    if ((coda_symlink_cache) && (VALID_SYMLINK(cp))) { /* symlink was cached */
750	uiop->uio_rw = UIO_READ;
751	error = uiomove(cp->c_symlink, (int)cp->c_symlen, uiop);
752	if (error)
753	    MARK_INT_FAIL(CODA_READLINK_STATS);
754	else
755	    MARK_INT_SAT(CODA_READLINK_STATS);
756	return(error);
757    }
758
759    error = venus_readlink(vtomi(vp), &cp->c_fid, cred, l, &str, &len);
760
761    if (!error) {
762	uiop->uio_rw = UIO_READ;
763	error = uiomove(str, len, uiop);
764
765	if (coda_symlink_cache) {
766	    cp->c_symlink = str;
767	    cp->c_symlen = len;
768	    cp->c_flags |= C_SYMLINK;
769	} else
770	    CODA_FREE(str, len);
771    }
772
773    CODADEBUG(CODA_READLINK, myprintf(("in readlink result %d\n",error));)
774    return(error);
775}
776
777int
778coda_fsync(void *v)
779{
780/* true args */
781    struct vop_fsync_args *ap = v;
782    vnode_t *vp = ap->a_vp;
783    struct cnode *cp = VTOC(vp);
784    kauth_cred_t cred = ap->a_cred;
785/* locals */
786    vnode_t *convp = cp->c_ovp;
787    int error;
788
789    MARK_ENTRY(CODA_FSYNC_STATS);
790
791    /* Check for fsync on an unmounting object */
792    /* The NetBSD kernel, in its infinite wisdom, can try to fsync
793     * after an unmount has been initiated.  This is a Bad Thing,
794     * which we have to avoid.  Not a legitimate failure for stats.
795     */
796    if (IS_UNMOUNTING(cp)) {
797	return(ENODEV);
798    }
799
800    /* Check for fsync of control object or uninitialized cnode. */
801    if (IS_CTL_VP(vp) || vp->v_type == VNON) {
802	MARK_INT_SAT(CODA_FSYNC_STATS);
803	return(0);
804    }
805
806    if (convp)
807    	VOP_FSYNC(convp, cred, MNT_WAIT, 0, 0);
808
809    /*
810     * We can expect fsync on any vnode at all if venus is pruging it.
811     * Venus can't very well answer the fsync request, now can it?
812     * Hopefully, it won't have to, because hopefully, venus preserves
813     * the (possibly untrue) invariant that it never purges an open
814     * vnode.  Hopefully.
815     */
816    if (cp->c_flags & C_PURGING) {
817	return(0);
818    }
819
820    error = venus_fsync(vtomi(vp), &cp->c_fid, cred, curlwp);
821
822    CODADEBUG(CODA_FSYNC, myprintf(("in fsync result %d\n",error)); )
823    return(error);
824}
825
826/*
827 * vp is locked on entry, and we must unlock it.
828 * XXX This routine is suspect and probably needs rewriting.
829 */
830int
831coda_inactive(void *v)
832{
833/* true args */
834    struct vop_inactive_v2_args *ap = v;
835    vnode_t *vp = ap->a_vp;
836    struct cnode *cp = VTOC(vp);
837    kauth_cred_t cred __unused = NULL;
838
839    /* We don't need to send inactive to venus - DCS */
840    MARK_ENTRY(CODA_INACTIVE_STATS);
841
842    if (IS_CTL_VP(vp)) {
843	MARK_INT_SAT(CODA_INACTIVE_STATS);
844	return 0;
845    }
846
847    CODADEBUG(CODA_INACTIVE, myprintf(("in inactive, %s, vfsp %p\n",
848				  coda_f2s(&cp->c_fid), vp->v_mount));)
849
850    if (vp->v_mount->mnt_data == NULL) {
851	myprintf(("Help! vfsp->vfs_data was NULL, but vnode %p wasn't dying\n", vp));
852	panic("badness in coda_inactive");
853    }
854
855#ifdef CODA_VERBOSE
856    /* Sanity checks that perhaps should be panic. */
857    if (vrefcnt(vp) > 1)
858	printf("%s: %p usecount %d\n", __func__, vp, vrefcnt(vp));
859    if (cp->c_ovp != NULL)
860	printf("%s: %p ovp != NULL\n", __func__, vp);
861#endif
862    /* XXX Do we need to VOP_CLOSE container vnodes? */
863    if (!IS_UNMOUNTING(cp))
864	*ap->a_recycle = true;
865
866    MARK_INT_SAT(CODA_INACTIVE_STATS);
867    return(0);
868}
869
870/*
871 * Coda does not use the normal namecache, but a private version.
872 * Consider how to use the standard facility instead.
873 */
874int
875coda_lookup(void *v)
876{
877/* true args */
878    struct vop_lookup_v2_args *ap = v;
879    /* (locked) vnode of dir in which to do lookup */
880    vnode_t *dvp = ap->a_dvp;
881    struct cnode *dcp = VTOC(dvp);
882    /* output variable for result */
883    vnode_t **vpp = ap->a_vpp;
884    /* name to lookup */
885    struct componentname *cnp = ap->a_cnp;
886    kauth_cred_t cred = cnp->cn_cred;
887    struct lwp *l = curlwp;
888/* locals */
889    struct cnode *cp;
890    const char *nm = cnp->cn_nameptr;
891    int len = cnp->cn_namelen;
892    CodaFid VFid;
893    int	vtype;
894    int error = 0;
895
896    MARK_ENTRY(CODA_LOOKUP_STATS);
897
898    CODADEBUG(CODA_LOOKUP, myprintf(("%s: %s in %s\n", __func__,
899	nm, coda_f2s(&dcp->c_fid)));)
900
901    /*
902     * XXX componentname flags in MODMASK are not handled at all
903     */
904
905    /*
906     * The overall strategy is to switch on the lookup type and get a
907     * result vnode that is vref'd but not locked.
908     */
909
910    /* Check for lookup of control object. */
911    if (IS_CTL_NAME(dvp, nm, len)) {
912	*vpp = coda_ctlvp;
913	vref(*vpp);
914	MARK_INT_SAT(CODA_LOOKUP_STATS);
915	goto exit;
916    }
917
918    /* Avoid trying to hand venus an unreasonably long name. */
919    if (len+1 > CODA_MAXNAMLEN) {
920	MARK_INT_FAIL(CODA_LOOKUP_STATS);
921	CODADEBUG(CODA_LOOKUP, myprintf(("%s: name too long:, %s (%s)\n",
922	    __func__, coda_f2s(&dcp->c_fid), nm));)
923	*vpp = (vnode_t *)0;
924	error = EINVAL;
925	goto exit;
926    }
927
928    /*
929     * Try to resolve the lookup in the minicache.  If that fails, ask
930     * venus to do the lookup.  XXX The interaction between vnode
931     * locking and any locking that coda does is not clear.
932     */
933    cp = coda_nc_lookup(dcp, nm, len, cred);
934    if (cp) {
935	*vpp = CTOV(cp);
936	vref(*vpp);
937	CODADEBUG(CODA_LOOKUP,
938		 myprintf(("lookup result %d vpp %p\n",error,*vpp));)
939    } else {
940	/* The name wasn't cached, so ask Venus. */
941	error = venus_lookup(vtomi(dvp), &dcp->c_fid, nm, len, cred, l, &VFid,
942	    &vtype);
943
944	if (error) {
945	    MARK_INT_FAIL(CODA_LOOKUP_STATS);
946	    CODADEBUG(CODA_LOOKUP, myprintf(("%s: lookup error on %s (%s)%d\n",
947		__func__, coda_f2s(&dcp->c_fid), nm, error));)
948	    *vpp = (vnode_t *)0;
949	} else {
950	    MARK_INT_SAT(CODA_LOOKUP_STATS);
951	    CODADEBUG(CODA_LOOKUP, myprintf(("%s: %s type %o result %d\n",
952		__func__, coda_f2s(&VFid), vtype, error)); )
953
954	    cp = make_coda_node(&VFid, dvp->v_mount, vtype);
955	    *vpp = CTOV(cp);
956	    /* vpp is now vrefed. */
957
958	    /*
959	     * Unless this vnode is marked CODA_NOCACHE, enter it into
960	     * the coda name cache to avoid a future venus round-trip.
961	     * XXX Interaction with componentname NOCACHE is unclear.
962	     */
963	    if (!(vtype & CODA_NOCACHE))
964		coda_nc_enter(VTOC(dvp), nm, len, cred, VTOC(*vpp));
965	}
966    }
967
968 exit:
969    /*
970     * If we are creating, and this was the last name to be looked up,
971     * and the error was ENOENT, then make the leaf NULL and return
972     * success.
973     * XXX Check against new lookup rules.
974     */
975    if (((cnp->cn_nameiop == CREATE) || (cnp->cn_nameiop == RENAME))
976	&& (cnp->cn_flags & ISLASTCN)
977	&& (error == ENOENT))
978    {
979	error = EJUSTRETURN;
980	*ap->a_vpp = NULL;
981    }
982
983    return(error);
984}
985
986/*ARGSUSED*/
987int
988coda_create(void *v)
989{
990/* true args */
991    struct vop_create_v3_args *ap = v;
992    vnode_t *dvp = ap->a_dvp;
993    struct cnode *dcp = VTOC(dvp);
994    struct vattr *va = ap->a_vap;
995    int exclusive = 1;
996    int mode = ap->a_vap->va_mode;
997    vnode_t **vpp = ap->a_vpp;
998    struct componentname  *cnp = ap->a_cnp;
999    kauth_cred_t cred = cnp->cn_cred;
1000    struct lwp *l = curlwp;
1001/* locals */
1002    int error;
1003    struct cnode *cp;
1004    const char *nm = cnp->cn_nameptr;
1005    int len = cnp->cn_namelen;
1006    CodaFid VFid;
1007    struct vattr attr;
1008
1009    MARK_ENTRY(CODA_CREATE_STATS);
1010
1011    /* All creates are exclusive XXX */
1012    /* I'm assuming the 'mode' argument is the file mode bits XXX */
1013
1014    /* Check for create of control object. */
1015    if (IS_CTL_NAME(dvp, nm, len)) {
1016	*vpp = (vnode_t *)0;
1017	MARK_INT_FAIL(CODA_CREATE_STATS);
1018	return(EACCES);
1019    }
1020
1021    error = venus_create(vtomi(dvp), &dcp->c_fid, nm, len, exclusive, mode, va, cred, l, &VFid, &attr);
1022
1023    if (!error) {
1024
1025        /*
1026	 * XXX Violation of venus/kernel invariants is a difficult case,
1027	 * but venus should not be able to cause a panic.
1028	 */
1029	/* If this is an exclusive create, panic if the file already exists. */
1030	/* Venus should have detected the file and reported EEXIST. */
1031
1032	if ((exclusive == 1) &&
1033	    (coda_find(&VFid) != NULL))
1034	    panic("cnode existed for newly created file!");
1035
1036	cp = make_coda_node(&VFid, dvp->v_mount, attr.va_type);
1037	*vpp = CTOV(cp);
1038
1039	/* XXX vnodeops doesn't say this argument can be changed. */
1040	/* Update va to reflect the new attributes. */
1041	(*va) = attr;
1042
1043	/* Update the attribute cache and mark it as valid */
1044	if (coda_attr_cache) {
1045	    VTOC(*vpp)->c_vattr = attr;
1046	    VTOC(*vpp)->c_flags |= C_VATTR;
1047	}
1048
1049	/* Invalidate parent's attr cache (modification time has changed). */
1050	VTOC(dvp)->c_flags &= ~C_VATTR;
1051
1052	/* enter the new vnode in the Name Cache */
1053	coda_nc_enter(VTOC(dvp), nm, len, cred, VTOC(*vpp));
1054
1055	CODADEBUG(CODA_CREATE, myprintf(("%s: %s, result %d\n", __func__,
1056	    coda_f2s(&VFid), error)); )
1057    } else {
1058	*vpp = (vnode_t *)0;
1059	CODADEBUG(CODA_CREATE, myprintf(("%s: create error %d\n", __func__,
1060	    error));)
1061    }
1062
1063    if (!error) {
1064#ifdef CODA_VERBOSE
1065	if ((cnp->cn_flags & LOCKLEAF) == 0)
1066	    /* This should not happen; flags are for lookup only. */
1067	    printf("%s: LOCKLEAF not set!\n", __func__);
1068#endif
1069    }
1070
1071    return(error);
1072}
1073
1074int
1075coda_remove(void *v)
1076{
1077/* true args */
1078    struct vop_remove_v3_args *ap = v;
1079    vnode_t *dvp = ap->a_dvp;
1080    struct cnode *cp = VTOC(dvp);
1081    vnode_t *vp = ap->a_vp;
1082    struct componentname  *cnp = ap->a_cnp;
1083    kauth_cred_t cred = cnp->cn_cred;
1084    struct lwp *l = curlwp;
1085/* locals */
1086    int error;
1087    const char *nm = cnp->cn_nameptr;
1088    int len = cnp->cn_namelen;
1089    struct cnode *tp;
1090
1091    MARK_ENTRY(CODA_REMOVE_STATS);
1092
1093    CODADEBUG(CODA_REMOVE, myprintf(("%s: %s in %s\n", __func__,
1094	nm, coda_f2s(&cp->c_fid)));)
1095
1096    /* Remove the file's entry from the CODA Name Cache */
1097    /* We're being conservative here, it might be that this person
1098     * doesn't really have sufficient access to delete the file
1099     * but we feel zapping the entry won't really hurt anyone -- dcs
1100     */
1101    /* I'm gonna go out on a limb here. If a file and a hardlink to it
1102     * exist, and one is removed, the link count on the other will be
1103     * off by 1. We could either invalidate the attrs if cached, or
1104     * fix them. I'll try to fix them. DCS 11/8/94
1105     */
1106    tp = coda_nc_lookup(VTOC(dvp), nm, len, cred);
1107    if (tp) {
1108	if (VALID_VATTR(tp)) {	/* If attrs are cached */
1109	    if (tp->c_vattr.va_nlink > 1) {	/* If it's a hard link */
1110		tp->c_vattr.va_nlink--;
1111	    }
1112	}
1113
1114	coda_nc_zapfile(VTOC(dvp), nm, len);
1115	/* No need to flush it if it doesn't exist! */
1116    }
1117    /* Invalidate the parent's attr cache, the modification time has changed */
1118    VTOC(dvp)->c_flags &= ~C_VATTR;
1119
1120    /* Check for remove of control object. */
1121    if (IS_CTL_NAME(dvp, nm, len)) {
1122	MARK_INT_FAIL(CODA_REMOVE_STATS);
1123	return(ENOENT);
1124    }
1125
1126    error = venus_remove(vtomi(dvp), &cp->c_fid, nm, len, cred, l);
1127
1128    CODADEBUG(CODA_REMOVE, myprintf(("in remove result %d\n",error)); )
1129
1130    /*
1131     * Unlock and release child (avoiding double if ".").
1132     */
1133    if (dvp == vp) {
1134	vrele(vp);
1135    } else {
1136	vput(vp);
1137    }
1138
1139    return(error);
1140}
1141
1142/*
1143 * dvp is the directory where the link is to go, and is locked.
1144 * vp is the object to be linked to, and is unlocked.
1145 * At exit, we must unlock dvp, and vput dvp.
1146 */
1147int
1148coda_link(void *v)
1149{
1150/* true args */
1151    struct vop_link_v2_args *ap = v;
1152    vnode_t *vp = ap->a_vp;
1153    struct cnode *cp = VTOC(vp);
1154    vnode_t *dvp = ap->a_dvp;
1155    struct cnode *dcp = VTOC(dvp);
1156    struct componentname *cnp = ap->a_cnp;
1157    kauth_cred_t cred = cnp->cn_cred;
1158    struct lwp *l = curlwp;
1159/* locals */
1160    int error;
1161    const char *nm = cnp->cn_nameptr;
1162    int len = cnp->cn_namelen;
1163
1164    MARK_ENTRY(CODA_LINK_STATS);
1165
1166    if (codadebug & CODADBGMSK(CODA_LINK)) {
1167
1168	myprintf(("%s: vp fid: %s\n", __func__, coda_f2s(&cp->c_fid)));
1169	myprintf(("%s: dvp fid: %s)\n", __func__, coda_f2s(&dcp->c_fid)));
1170
1171    }
1172    if (codadebug & CODADBGMSK(CODA_LINK)) {
1173	myprintf(("%s: vp fid: %s\n", __func__, coda_f2s(&cp->c_fid)));
1174	myprintf(("%s: dvp fid: %s\n", __func__, coda_f2s(&dcp->c_fid)));
1175
1176    }
1177
1178    /* Check for link to/from control object. */
1179    if (IS_CTL_NAME(dvp, nm, len) || IS_CTL_VP(vp)) {
1180	MARK_INT_FAIL(CODA_LINK_STATS);
1181	return(EACCES);
1182    }
1183
1184    /* If linking . to a name, error out earlier. */
1185    if (vp == dvp) {
1186#ifdef CODA_VERBOSE
1187        printf("%s coda_link vp==dvp\n", __func__);
1188#endif
1189	error = EISDIR;
1190	goto exit;
1191    }
1192
1193    /* XXX Why does venus_link need the vnode to be locked?*/
1194    if ((error = vn_lock(vp, LK_EXCLUSIVE)) != 0) {
1195#ifdef CODA_VERBOSE
1196	printf("%s: couldn't lock vnode %p\n", __func__, vp);
1197#endif
1198	error = EFAULT;		/* XXX better value */
1199	goto exit;
1200    }
1201    error = kauth_authorize_vnode(cnp->cn_cred, KAUTH_VNODE_ADD_LINK, vp,
1202	dvp, 0);
1203    if (error)
1204	    goto exit;
1205    error = venus_link(vtomi(vp), &cp->c_fid, &dcp->c_fid, nm, len, cred, l);
1206    VOP_UNLOCK(vp);
1207
1208    /* Invalidate parent's attr cache (the modification time has changed). */
1209    VTOC(dvp)->c_flags &= ~C_VATTR;
1210    /* Invalidate child's attr cache (XXX why). */
1211    VTOC(vp)->c_flags &= ~C_VATTR;
1212
1213    CODADEBUG(CODA_LINK,	myprintf(("in link result %d\n",error)); )
1214
1215exit:
1216    return(error);
1217}
1218
1219int
1220coda_rename(void *v)
1221{
1222/* true args */
1223    struct vop_rename_args *ap = v;
1224    vnode_t *odvp = ap->a_fdvp;
1225    struct cnode *odcp = VTOC(odvp);
1226    struct componentname  *fcnp = ap->a_fcnp;
1227    vnode_t *ndvp = ap->a_tdvp;
1228    struct cnode *ndcp = VTOC(ndvp);
1229    struct componentname  *tcnp = ap->a_tcnp;
1230    kauth_cred_t cred = fcnp->cn_cred;
1231    struct lwp *l = curlwp;
1232/* true args */
1233    int error;
1234    const char *fnm = fcnp->cn_nameptr;
1235    int flen = fcnp->cn_namelen;
1236    const char *tnm = tcnp->cn_nameptr;
1237    int tlen = tcnp->cn_namelen;
1238
1239    MARK_ENTRY(CODA_RENAME_STATS);
1240
1241    /* Hmmm.  The vnodes are already looked up.  Perhaps they are locked?
1242       This could be Bad. XXX */
1243#ifdef OLD_DIAGNOSTIC
1244    if ((fcnp->cn_cred != tcnp->cn_cred)
1245	|| (fcnp->cn_lwp != tcnp->cn_lwp))
1246    {
1247	panic("%s: component names don't agree", __func__);
1248    }
1249#endif
1250
1251    /* Check for rename involving control object. */
1252    if (IS_CTL_NAME(odvp, fnm, flen) || IS_CTL_NAME(ndvp, tnm, tlen)) {
1253	MARK_INT_FAIL(CODA_RENAME_STATS);
1254	return(EACCES);
1255    }
1256
1257    /* Problem with moving directories -- need to flush entry for .. */
1258    if (odvp != ndvp) {
1259	struct cnode *ovcp = coda_nc_lookup(VTOC(odvp), fnm, flen, cred);
1260	if (ovcp) {
1261	    vnode_t *ovp = CTOV(ovcp);
1262	    if ((ovp) &&
1263		(ovp->v_type == VDIR)) /* If it's a directory */
1264		coda_nc_zapfile(VTOC(ovp),"..", 2);
1265	}
1266    }
1267
1268    /* Remove the entries for both source and target files */
1269    coda_nc_zapfile(VTOC(odvp), fnm, flen);
1270    coda_nc_zapfile(VTOC(ndvp), tnm, tlen);
1271
1272    /* Invalidate the parent's attr cache, the modification time has changed */
1273    VTOC(odvp)->c_flags &= ~C_VATTR;
1274    VTOC(ndvp)->c_flags &= ~C_VATTR;
1275
1276    if (flen+1 > CODA_MAXNAMLEN) {
1277	MARK_INT_FAIL(CODA_RENAME_STATS);
1278	error = EINVAL;
1279	goto exit;
1280    }
1281
1282    if (tlen+1 > CODA_MAXNAMLEN) {
1283	MARK_INT_FAIL(CODA_RENAME_STATS);
1284	error = EINVAL;
1285	goto exit;
1286    }
1287
1288    error = venus_rename(vtomi(odvp), &odcp->c_fid, &ndcp->c_fid, fnm, flen, tnm, tlen, cred, l);
1289
1290 exit:
1291    CODADEBUG(CODA_RENAME, myprintf(("in rename result %d\n",error));)
1292    /* XXX - do we need to call cache pureg on the moved vnode? */
1293    cache_purge(ap->a_fvp);
1294
1295    /* It seems to be incumbent on us to drop locks on all four vnodes */
1296    /* From-vnodes are not locked, only ref'd.  To-vnodes are locked. */
1297
1298    vrele(ap->a_fvp);
1299    vrele(odvp);
1300
1301    if (ap->a_tvp) {
1302	if (ap->a_tvp == ndvp) {
1303	    vrele(ap->a_tvp);
1304	} else {
1305	    vput(ap->a_tvp);
1306	}
1307    }
1308
1309    vput(ndvp);
1310    return(error);
1311}
1312
1313int
1314coda_mkdir(void *v)
1315{
1316/* true args */
1317    struct vop_mkdir_v3_args *ap = v;
1318    vnode_t *dvp = ap->a_dvp;
1319    struct cnode *dcp = VTOC(dvp);
1320    struct componentname  *cnp = ap->a_cnp;
1321    struct vattr *va = ap->a_vap;
1322    vnode_t **vpp = ap->a_vpp;
1323    kauth_cred_t cred = cnp->cn_cred;
1324    struct lwp *l = curlwp;
1325/* locals */
1326    int error;
1327    const char *nm = cnp->cn_nameptr;
1328    int len = cnp->cn_namelen;
1329    struct cnode *cp;
1330    CodaFid VFid;
1331    struct vattr ova;
1332
1333    MARK_ENTRY(CODA_MKDIR_STATS);
1334
1335    /* Check for mkdir of target object. */
1336    if (IS_CTL_NAME(dvp, nm, len)) {
1337	*vpp = (vnode_t *)0;
1338	MARK_INT_FAIL(CODA_MKDIR_STATS);
1339	return(EACCES);
1340    }
1341
1342    if (len+1 > CODA_MAXNAMLEN) {
1343	*vpp = (vnode_t *)0;
1344	MARK_INT_FAIL(CODA_MKDIR_STATS);
1345	return(EACCES);
1346    }
1347
1348    error = venus_mkdir(vtomi(dvp), &dcp->c_fid, nm, len, va, cred, l, &VFid, &ova);
1349
1350    if (!error) {
1351	if (coda_find(&VFid) != NULL)
1352	    panic("cnode existed for newly created directory!");
1353
1354
1355	cp =  make_coda_node(&VFid, dvp->v_mount, va->va_type);
1356	*vpp = CTOV(cp);
1357
1358	/* enter the new vnode in the Name Cache */
1359	coda_nc_enter(VTOC(dvp), nm, len, cred, VTOC(*vpp));
1360
1361	/* as a side effect, enter "." and ".." for the directory */
1362	coda_nc_enter(VTOC(*vpp), ".", 1, cred, VTOC(*vpp));
1363	coda_nc_enter(VTOC(*vpp), "..", 2, cred, VTOC(dvp));
1364
1365	if (coda_attr_cache) {
1366	    VTOC(*vpp)->c_vattr = ova;		/* update the attr cache */
1367	    VTOC(*vpp)->c_flags |= C_VATTR;	/* Valid attributes in cnode */
1368	}
1369
1370	/* Invalidate the parent's attr cache, the modification time has changed */
1371	VTOC(dvp)->c_flags &= ~C_VATTR;
1372
1373	CODADEBUG( CODA_MKDIR, myprintf(("%s: %s result %d\n", __func__,
1374	    coda_f2s(&VFid), error)); )
1375    } else {
1376	*vpp = (vnode_t *)0;
1377	CODADEBUG(CODA_MKDIR, myprintf(("%s error %d\n", __func__, error));)
1378    }
1379
1380    return(error);
1381}
1382
1383int
1384coda_rmdir(void *v)
1385{
1386/* true args */
1387    struct vop_rmdir_v2_args *ap = v;
1388    vnode_t *dvp = ap->a_dvp;
1389    struct cnode *dcp = VTOC(dvp);
1390    vnode_t *vp = ap->a_vp;
1391    struct componentname  *cnp = ap->a_cnp;
1392    kauth_cred_t cred = cnp->cn_cred;
1393    struct lwp *l = curlwp;
1394/* true args */
1395    int error;
1396    const char *nm = cnp->cn_nameptr;
1397    int len = cnp->cn_namelen;
1398    struct cnode *cp;
1399
1400    MARK_ENTRY(CODA_RMDIR_STATS);
1401
1402    /* Check for rmdir of control object. */
1403    if (IS_CTL_NAME(dvp, nm, len)) {
1404	MARK_INT_FAIL(CODA_RMDIR_STATS);
1405	return(ENOENT);
1406    }
1407
1408    /* Can't remove . in self. */
1409    if (dvp == vp) {
1410#ifdef CODA_VERBOSE
1411	printf("%s: dvp == vp\n", __func__);
1412#endif
1413	error = EINVAL;
1414	goto exit;
1415    }
1416
1417    /*
1418     * The caller may not have adequate permissions, and the venus
1419     * operation may fail, but it doesn't hurt from a correctness
1420     * viewpoint to invalidate cache entries.
1421     * XXX Why isn't this done after the venus_rmdir call?
1422     */
1423    /* Look up child in name cache (by name, from parent). */
1424    cp = coda_nc_lookup(dcp, nm, len, cred);
1425    /* If found, remove all children of the child (., ..). */
1426    if (cp) coda_nc_zapParentfid(&(cp->c_fid), NOT_DOWNCALL);
1427
1428    /* Remove child's own entry. */
1429    coda_nc_zapfile(dcp, nm, len);
1430
1431    /* Invalidate parent's attr cache (the modification time has changed). */
1432    dcp->c_flags &= ~C_VATTR;
1433
1434    error = venus_rmdir(vtomi(dvp), &dcp->c_fid, nm, len, cred, l);
1435
1436    CODADEBUG(CODA_RMDIR, myprintf(("in rmdir result %d\n", error)); )
1437
1438exit:
1439    /* unlock and release child */
1440    if (dvp == vp) {
1441	vrele(vp);
1442    } else {
1443	vput(vp);
1444    }
1445
1446    return(error);
1447}
1448
1449int
1450coda_symlink(void *v)
1451{
1452/* true args */
1453    struct vop_symlink_v3_args *ap = v;
1454    vnode_t *dvp = ap->a_dvp;
1455    struct cnode *dcp = VTOC(dvp);
1456    /* a_vpp is used in place below */
1457    struct componentname *cnp = ap->a_cnp;
1458    struct vattr *tva = ap->a_vap;
1459    char *path = ap->a_target;
1460    kauth_cred_t cred = cnp->cn_cred;
1461    struct lwp *l = curlwp;
1462/* locals */
1463    int error;
1464    u_long saved_cn_flags;
1465    const char *nm = cnp->cn_nameptr;
1466    int len = cnp->cn_namelen;
1467    int plen = strlen(path);
1468
1469    /*
1470     * Here's the strategy for the moment: perform the symlink, then
1471     * do a lookup to grab the resulting vnode.  I know this requires
1472     * two communications with Venus for a new symbolic link, but
1473     * that's the way the ball bounces.  I don't yet want to change
1474     * the way the Mach symlink works.  When Mach support is
1475     * deprecated, we should change symlink so that the common case
1476     * returns the resultant vnode in a vpp argument.
1477     */
1478
1479    MARK_ENTRY(CODA_SYMLINK_STATS);
1480
1481    /* Check for symlink of control object. */
1482    if (IS_CTL_NAME(dvp, nm, len)) {
1483	MARK_INT_FAIL(CODA_SYMLINK_STATS);
1484	error = EACCES;
1485	goto exit;
1486    }
1487
1488    if (plen+1 > CODA_MAXPATHLEN) {
1489	MARK_INT_FAIL(CODA_SYMLINK_STATS);
1490	error = EINVAL;
1491	goto exit;
1492    }
1493
1494    if (len+1 > CODA_MAXNAMLEN) {
1495	MARK_INT_FAIL(CODA_SYMLINK_STATS);
1496	error = EINVAL;
1497	goto exit;
1498    }
1499
1500    error = venus_symlink(vtomi(dvp), &dcp->c_fid, path, plen, nm, len, tva, cred, l);
1501
1502    /* Invalidate the parent's attr cache (modification time has changed). */
1503    dcp->c_flags &= ~C_VATTR;
1504
1505    if (!error) {
1506	/*
1507	 * VOP_SYMLINK is not defined to pay attention to cnp->cn_flags;
1508	 * these are defined only for VOP_LOOKUP.   We desire to reuse
1509	 * cnp for a VOP_LOOKUP operation, and must be sure to not pass
1510	 * stray flags passed to us.  Such stray flags can occur because
1511	 * sys_symlink makes a namei call and then reuses the
1512	 * componentname structure.
1513	 */
1514	/*
1515	 * XXX Arguably we should create our own componentname structure
1516	 * and not reuse the one that was passed in.
1517	 */
1518	saved_cn_flags = cnp->cn_flags;
1519	cnp->cn_flags &= ~(MODMASK | OPMASK);
1520	cnp->cn_flags |= LOOKUP;
1521	error = VOP_LOOKUP(dvp, ap->a_vpp, cnp);
1522	cnp->cn_flags = saved_cn_flags;
1523    }
1524
1525 exit:
1526    CODADEBUG(CODA_SYMLINK, myprintf(("in symlink result %d\n",error)); )
1527    return(error);
1528}
1529
1530/*
1531 * Read directory entries.
1532 */
1533int
1534coda_readdir(void *v)
1535{
1536/* true args */
1537	struct vop_readdir_args *ap = v;
1538	vnode_t *vp = ap->a_vp;
1539	struct cnode *cp = VTOC(vp);
1540	struct uio *uiop = ap->a_uio;
1541	kauth_cred_t cred = ap->a_cred;
1542	int *eofflag = ap->a_eofflag;
1543/* upcall decl */
1544/* locals */
1545	size_t initial_resid = uiop->uio_resid;
1546	int error = 0;
1547	int opened_internally = 0;
1548	int ncookies;
1549	char *buf;
1550	struct vnode *cvp;
1551	struct dirent *dirp;
1552
1553	MARK_ENTRY(CODA_READDIR_STATS);
1554
1555	CODADEBUG(CODA_READDIR, myprintf(("%s: (%p, %lu, %lld)\n", __func__,
1556	    uiop->uio_iov->iov_base, (unsigned long) uiop->uio_resid,
1557	    (long long) uiop->uio_offset)); )
1558
1559	/* Check for readdir of control object. */
1560	if (IS_CTL_VP(vp)) {
1561		MARK_INT_FAIL(CODA_READDIR_STATS);
1562		return ENOENT;
1563	}
1564
1565	/* If directory is not already open do an "internal open" on it. */
1566	if (cp->c_ovp == NULL) {
1567		opened_internally = 1;
1568		MARK_INT_GEN(CODA_OPEN_STATS);
1569		error = VOP_OPEN(vp, FREAD, cred);
1570#ifdef	CODA_VERBOSE
1571		printf("%s: Internally Opening %p\n", __func__, vp);
1572#endif
1573		if (error)
1574			return error;
1575		KASSERT(cp->c_ovp != NULL);
1576	}
1577	cvp = cp->c_ovp;
1578
1579	CODADEBUG(CODA_READDIR, myprintf(("%s: fid = %s, refcnt = %d\n",
1580	    __func__, coda_f2s(&cp->c_fid), vrefcnt(cvp))); )
1581
1582	if (ap->a_ncookies) {
1583		ncookies = ap->a_uio->uio_resid / _DIRENT_RECLEN(dirp, 1);
1584		*ap->a_ncookies = 0;
1585		*ap->a_cookies = malloc(ncookies * sizeof (off_t),
1586		    M_TEMP, M_WAITOK);
1587	}
1588	buf = kmem_alloc(CODA_DIRBLKSIZ, KM_SLEEP);
1589	dirp = kmem_alloc(sizeof(*dirp), KM_SLEEP);
1590	vn_lock(cvp, LK_EXCLUSIVE | LK_RETRY);
1591
1592	while (error == 0) {
1593		size_t resid = 0;
1594		char *dp, *ep;
1595
1596		if (!ALIGNED_POINTER(uiop->uio_offset, uint32_t)) {
1597			error = EINVAL;
1598			break;
1599		}
1600		error = vn_rdwr(UIO_READ, cvp, buf,
1601		    CODA_DIRBLKSIZ, uiop->uio_offset,
1602		    UIO_SYSSPACE, IO_NODELOCKED, cred, &resid, curlwp);
1603		if (error || resid == CODA_DIRBLKSIZ)
1604			break;
1605		for (dp = buf, ep = dp + CODA_DIRBLKSIZ - resid; dp < ep; ) {
1606			off_t off;
1607			struct venus_dirent *vd = (struct venus_dirent *)dp;
1608
1609			if (!ALIGNED_POINTER(vd, uint32_t) ||
1610			    !ALIGNED_POINTER(vd->d_reclen, uint32_t) ||
1611			    vd->d_reclen == 0) {
1612				error = EINVAL;
1613				break;
1614			}
1615			if (dp + vd->d_reclen > ep) {
1616				error = ENAMETOOLONG;
1617				break;
1618			}
1619			if (vd->d_namlen == 0) {
1620				uiop->uio_offset += vd->d_reclen;
1621				dp += vd->d_reclen;
1622				continue;
1623			}
1624
1625			dirp->d_fileno = vd->d_fileno;
1626			dirp->d_type = vd->d_type;
1627			dirp->d_namlen = vd->d_namlen;
1628			dirp->d_reclen = _DIRENT_SIZE(dirp);
1629			strlcpy(dirp->d_name, vd->d_name, dirp->d_namlen + 1);
1630
1631			if (uiop->uio_resid < dirp->d_reclen) {
1632				error = ENAMETOOLONG;
1633				break;
1634			}
1635
1636			off = uiop->uio_offset;
1637			error = uiomove(dirp, dirp->d_reclen, uiop);
1638			uiop->uio_offset = off;
1639			if (error)
1640				break;
1641
1642			uiop->uio_offset += vd->d_reclen;
1643			dp += vd->d_reclen;
1644			if (ap->a_ncookies)
1645				(*ap->a_cookies)[(*ap->a_ncookies)++] =
1646				    uiop->uio_offset;
1647		}
1648	}
1649
1650	VOP_UNLOCK(cvp);
1651	kmem_free(dirp, sizeof(*dirp));
1652	kmem_free(buf, CODA_DIRBLKSIZ);
1653	if (eofflag && error == 0)
1654		*eofflag = 1;
1655	if (uiop->uio_resid < initial_resid && error == ENAMETOOLONG)
1656		error = 0;
1657	if (ap->a_ncookies && error) {
1658		free(*ap->a_cookies, M_TEMP);
1659		*ap->a_ncookies = 0;
1660		*ap->a_cookies = NULL;
1661	}
1662	if (error)
1663		MARK_INT_FAIL(CODA_READDIR_STATS);
1664	else
1665		MARK_INT_SAT(CODA_READDIR_STATS);
1666
1667	/* Do an "internal close" if necessary. */
1668	if (opened_internally) {
1669		MARK_INT_GEN(CODA_CLOSE_STATS);
1670		(void)VOP_CLOSE(vp, FREAD, cred);
1671	}
1672
1673	return error;
1674}
1675
1676/*
1677 * Convert from file system blocks to device blocks
1678 */
1679int
1680coda_bmap(void *v)
1681{
1682    /* XXX on the global proc */
1683/* true args */
1684    struct vop_bmap_args *ap = v;
1685    vnode_t *vp __unused = ap->a_vp;	/* file's vnode */
1686    daddr_t bn __unused = ap->a_bn;	/* fs block number */
1687    vnode_t **vpp = ap->a_vpp;			/* RETURN vp of device */
1688    daddr_t *bnp __unused = ap->a_bnp;	/* RETURN device block number */
1689    struct lwp *l __unused = curlwp;
1690/* upcall decl */
1691/* locals */
1692
1693	*vpp = (vnode_t *)0;
1694	myprintf(("coda_bmap called!\n"));
1695	return(EINVAL);
1696}
1697
1698/*
1699 * I don't think the following two things are used anywhere, so I've
1700 * commented them out
1701 *
1702 * struct buf *async_bufhead;
1703 * int async_daemon_count;
1704 */
1705int
1706coda_strategy(void *v)
1707{
1708/* true args */
1709    struct vop_strategy_args *ap = v;
1710    struct buf *bp __unused = ap->a_bp;
1711    struct lwp *l __unused = curlwp;
1712/* upcall decl */
1713/* locals */
1714
1715	myprintf(("coda_strategy called!  "));
1716	return(EINVAL);
1717}
1718
1719int
1720coda_reclaim(void *v)
1721{
1722/* true args */
1723    struct vop_reclaim_v2_args *ap = v;
1724    vnode_t *vp = ap->a_vp;
1725    struct cnode *cp = VTOC(vp);
1726/* upcall decl */
1727/* locals */
1728
1729    VOP_UNLOCK(vp);
1730
1731/*
1732 * Forced unmount/flush will let vnodes with non zero use be destroyed!
1733 */
1734    ENTRY;
1735
1736    if (IS_UNMOUNTING(cp)) {
1737#ifdef	DEBUG
1738	if (VTOC(vp)->c_ovp) {
1739	    if (IS_UNMOUNTING(cp))
1740		printf("%s: c_ovp not void: vp %p, cp %p\n", __func__, vp, cp);
1741	}
1742#endif
1743    } else {
1744#ifdef OLD_DIAGNOSTIC
1745	if (vrefcnt(vp) != 0)
1746	    print("%s: pushing active %p\n", __func__, vp);
1747	if (VTOC(vp)->c_ovp) {
1748	    panic("%s: c_ovp not void", __func__);
1749	}
1750#endif
1751    }
1752    /* If an array has been allocated to hold the symlink, deallocate it */
1753    if ((coda_symlink_cache) && (VALID_SYMLINK(cp))) {
1754	if (cp->c_symlink == NULL)
1755	    panic("%s: null symlink pointer in cnode", __func__);
1756
1757	CODA_FREE(cp->c_symlink, cp->c_symlen);
1758	cp->c_flags &= ~C_SYMLINK;
1759	cp->c_symlen = 0;
1760    }
1761
1762    mutex_enter(vp->v_interlock);
1763    mutex_enter(&cp->c_lock);
1764    SET_VTOC(vp) = NULL;
1765    mutex_exit(&cp->c_lock);
1766    mutex_exit(vp->v_interlock);
1767    mutex_destroy(&cp->c_lock);
1768    kmem_free(cp, sizeof(*cp));
1769
1770    return (0);
1771}
1772
1773int
1774coda_lock(void *v)
1775{
1776/* true args */
1777    struct vop_lock_args *ap = v;
1778    vnode_t *vp = ap->a_vp;
1779    struct cnode *cp = VTOC(vp);
1780/* upcall decl */
1781/* locals */
1782
1783    ENTRY;
1784
1785    if (coda_lockdebug) {
1786	myprintf(("Attempting lock on %s\n",
1787		  coda_f2s(&cp->c_fid)));
1788    }
1789
1790    return genfs_lock(v);
1791}
1792
1793int
1794coda_unlock(void *v)
1795{
1796/* true args */
1797    struct vop_unlock_args *ap = v;
1798    vnode_t *vp = ap->a_vp;
1799    struct cnode *cp = VTOC(vp);
1800/* upcall decl */
1801/* locals */
1802
1803    ENTRY;
1804    if (coda_lockdebug) {
1805	myprintf(("Attempting unlock on %s\n",
1806		  coda_f2s(&cp->c_fid)));
1807    }
1808
1809    return genfs_unlock(v);
1810}
1811
1812int
1813coda_islocked(void *v)
1814{
1815/* true args */
1816    ENTRY;
1817
1818    return genfs_islocked(v);
1819}
1820
1821int
1822coda_pathconf(void *v)
1823{
1824	struct vop_pathconf_args *ap = v;
1825
1826	switch (ap->a_name) {
1827	default:
1828		return EINVAL;
1829	}
1830	/* NOTREACHED */
1831}
1832
1833/*
1834 * Given a device and inode, obtain a locked vnode.  One reference is
1835 * obtained and passed back to the caller.
1836 */
1837int
1838coda_grab_vnode(vnode_t *uvp, dev_t dev, ino_t ino, vnode_t **vpp)
1839{
1840    int           error;
1841    struct mount *mp;
1842
1843    /* Obtain mount point structure from device. */
1844    if (!(mp = devtomp(dev))) {
1845	myprintf(("%s: devtomp(0x%llx) returns NULL\n", __func__,
1846	    (unsigned long long)dev));
1847	return(ENXIO);
1848    }
1849
1850    /*
1851     * Obtain vnode from mount point and inode.
1852     */
1853    error = VFS_VGET(mp, ino, LK_EXCLUSIVE, vpp);
1854    if (error) {
1855	myprintf(("%s: iget/vget(0x%llx, %llu) returns %p, err %d\n", __func__,
1856	    (unsigned long long)dev, (unsigned long long)ino, *vpp, error));
1857	return(ENOENT);
1858    }
1859    /* share the underlying vnode lock with the coda vnode */
1860    vshareilock(*vpp, uvp);
1861    KASSERT(VOP_ISLOCKED(*vpp));
1862    return(0);
1863}
1864
1865static void
1866coda_print_vattr(struct vattr *attr)
1867{
1868    const char *typestr;
1869
1870    switch (attr->va_type) {
1871    case VNON:
1872	typestr = "VNON";
1873	break;
1874    case VREG:
1875	typestr = "VREG";
1876	break;
1877    case VDIR:
1878	typestr = "VDIR";
1879	break;
1880    case VBLK:
1881	typestr = "VBLK";
1882	break;
1883    case VCHR:
1884	typestr = "VCHR";
1885	break;
1886    case VLNK:
1887	typestr = "VLNK";
1888	break;
1889    case VSOCK:
1890	typestr = "VSCK";
1891	break;
1892    case VFIFO:
1893	typestr = "VFFO";
1894	break;
1895    case VBAD:
1896	typestr = "VBAD";
1897	break;
1898    default:
1899	typestr = "????";
1900	break;
1901    }
1902
1903
1904    myprintf(("attr: type %s mode %d uid %d gid %d fsid %d rdev %d\n",
1905	      typestr, (int)attr->va_mode, (int)attr->va_uid,
1906	      (int)attr->va_gid, (int)attr->va_fsid, (int)attr->va_rdev));
1907
1908    myprintf(("      fileid %d nlink %d size %d blocksize %d bytes %d\n",
1909	      (int)attr->va_fileid, (int)attr->va_nlink,
1910	      (int)attr->va_size,
1911	      (int)attr->va_blocksize,(int)attr->va_bytes));
1912    myprintf(("      gen %ld flags %ld vaflags %d\n",
1913	      attr->va_gen, attr->va_flags, attr->va_vaflags));
1914    myprintf(("      atime sec %d nsec %d\n",
1915	      (int)attr->va_atime.tv_sec, (int)attr->va_atime.tv_nsec));
1916    myprintf(("      mtime sec %d nsec %d\n",
1917	      (int)attr->va_mtime.tv_sec, (int)attr->va_mtime.tv_nsec));
1918    myprintf(("      ctime sec %d nsec %d\n",
1919	      (int)attr->va_ctime.tv_sec, (int)attr->va_ctime.tv_nsec));
1920}
1921
1922/*
1923 * Return a vnode for the given fid.
1924 * If no cnode exists for this fid create one and put it
1925 * in a table hashed by coda_f2i().  If the cnode for
1926 * this fid is already in the table return it (ref count is
1927 * incremented by coda_find.  The cnode will be flushed from the
1928 * table when coda_inactive calls coda_unsave.
1929 */
1930struct cnode *
1931make_coda_node(CodaFid *fid, struct mount *fvsp, short type)
1932{
1933	int error __diagused;
1934	struct vnode *vp;
1935	struct cnode *cp;
1936
1937	error = vcache_get(fvsp, fid, sizeof(CodaFid), &vp);
1938	KASSERT(error == 0);
1939
1940	mutex_enter(vp->v_interlock);
1941	cp = VTOC(vp);
1942	KASSERT(cp != NULL);
1943	mutex_enter(&cp->c_lock);
1944	mutex_exit(vp->v_interlock);
1945
1946	if (vp->v_type != type) {
1947		if (vp->v_type == VCHR || vp->v_type == VBLK)
1948			spec_node_destroy(vp);
1949		vp->v_type = type;
1950		if (type == VCHR || type == VBLK)
1951			spec_node_init(vp, NODEV);
1952		uvm_vnp_setsize(vp, 0);
1953	}
1954	mutex_exit(&cp->c_lock);
1955
1956	return cp;
1957}
1958
1959/*
1960 * coda_getpages may be called on a vnode which has not been opened,
1961 * e.g. to fault in pages to execute a program.  In that case, we must
1962 * open the file to get the container.  The vnode may or may not be
1963 * locked, and we must leave it in the same state.
1964 */
1965int
1966coda_getpages(void *v)
1967{
1968	struct vop_getpages_args /* {
1969		vnode_t *a_vp;
1970		voff_t a_offset;
1971		struct vm_page **a_m;
1972		int *a_count;
1973		int a_centeridx;
1974		vm_prot_t a_access_type;
1975		int a_advice;
1976		int a_flags;
1977	} */ *ap = v;
1978	vnode_t *vp = ap->a_vp, *cvp;
1979	struct cnode *cp = VTOC(vp);
1980	struct lwp *l = curlwp;
1981	kauth_cred_t cred = l->l_cred;
1982	int error, cerror;
1983	int waslocked;	       /* 1 if vnode lock was held on entry */
1984	int didopen = 0;	/* 1 if we opened container file */
1985	krw_t op;
1986
1987	/*
1988	 * Handle a case that uvm_fault doesn't quite use yet.
1989	 * See layer_vnops.c. for inspiration.
1990	 */
1991	if (ap->a_flags & PGO_LOCKED) {
1992		return EBUSY;
1993	}
1994
1995	KASSERT(rw_lock_held(vp->v_uobj.vmobjlock));
1996
1997	/* Check for control object. */
1998	if (IS_CTL_VP(vp)) {
1999#ifdef CODA_VERBOSE
2000		printf("%s: control object %p\n", __func__, vp);
2001#endif
2002		return(EINVAL);
2003	}
2004
2005	/*
2006	 * XXX It's really not ok to be releasing the lock we get,
2007	 * because we could be overlapping with another call to
2008	 * getpages and drop a lock they are relying on.  We need to
2009	 * figure out whether getpages ever is called holding the
2010	 * lock, and if we should serialize getpages calls by some
2011	 * mechanism.
2012	 */
2013	/* XXX VOP_ISLOCKED() may not be used for lock decisions. */
2014	op = rw_lock_op(vp->v_uobj.vmobjlock);
2015	waslocked = VOP_ISLOCKED(vp);
2016
2017	/* Get container file if not already present. */
2018	cvp = cp->c_ovp;
2019	if (cvp == NULL) {
2020		/*
2021		 * VOP_OPEN requires a locked vnode.  We must avoid
2022		 * locking the vnode if it is already locked, and
2023		 * leave it in the same state on exit.
2024		 */
2025		if (waslocked == 0) {
2026			rw_exit(vp->v_uobj.vmobjlock);
2027			cerror = vn_lock(vp, LK_EXCLUSIVE);
2028			if (cerror) {
2029#ifdef CODA_VERBOSE
2030				printf("%s: can't lock vnode %p\n",
2031				    __func__, vp);
2032#endif
2033				return cerror;
2034			}
2035#ifdef CODA_VERBOSE
2036			printf("%s: locked vnode %p\n", __func__, vp);
2037#endif
2038		}
2039
2040		/*
2041		 * Open file (causes upcall to venus).
2042		 * XXX Perhaps we should not fully open the file, but
2043		 * simply obtain a container file.
2044		 */
2045		/* XXX Is it ok to do this while holding the mutex? */
2046		cerror = VOP_OPEN(vp, FREAD, cred);
2047
2048		if (cerror) {
2049#ifdef CODA_VERBOSE
2050			printf("%s: cannot open vnode %p => %d\n", __func__,
2051			    vp, cerror);
2052#endif
2053			if (waslocked == 0)
2054				VOP_UNLOCK(vp);
2055			return cerror;
2056		}
2057
2058#ifdef CODA_VERBOSE
2059		printf("%s: opened vnode %p\n", __func__, vp);
2060#endif
2061		cvp = cp->c_ovp;
2062		didopen = 1;
2063		if (waslocked == 0)
2064			rw_enter(vp->v_uobj.vmobjlock, op);
2065	}
2066	KASSERT(cvp != NULL);
2067
2068	/* Munge the arg structure to refer to the container vnode. */
2069	KASSERT(cvp->v_uobj.vmobjlock == vp->v_uobj.vmobjlock);
2070	ap->a_vp = cp->c_ovp;
2071
2072	/* Finally, call getpages on it. */
2073	error = VCALL(ap->a_vp, VOFFSET(vop_getpages), ap);
2074
2075	/* If we opened the vnode, we must close it. */
2076	if (didopen) {
2077		/*
2078		 * VOP_CLOSE requires a locked vnode, but we are still
2079		 * holding the lock (or riding a caller's lock).
2080		 */
2081		cerror = VOP_CLOSE(vp, FREAD, cred);
2082#ifdef CODA_VERBOSE
2083		if (cerror != 0)
2084			/* XXX How should we handle this? */
2085			printf("%s: closed vnode %p -> %d\n", __func__,
2086			    vp, cerror);
2087#endif
2088
2089		/* If we obtained a lock, drop it. */
2090		if (waslocked == 0)
2091			VOP_UNLOCK(vp);
2092	}
2093
2094	return error;
2095}
2096
2097/*
2098 * The protocol requires v_interlock to be held by the caller.
2099 */
2100int
2101coda_putpages(void *v)
2102{
2103	struct vop_putpages_args /* {
2104		vnode_t *a_vp;
2105		voff_t a_offlo;
2106		voff_t a_offhi;
2107		int a_flags;
2108	} */ *ap = v;
2109	vnode_t *vp = ap->a_vp, *cvp;
2110	struct cnode *cp = VTOC(vp);
2111	int error;
2112
2113	KASSERT(rw_write_held(vp->v_uobj.vmobjlock));
2114
2115	/* Check for control object. */
2116	if (IS_CTL_VP(vp)) {
2117		rw_exit(vp->v_uobj.vmobjlock);
2118#ifdef CODA_VERBOSE
2119		printf("%s: control object %p\n", __func__, vp);
2120#endif
2121		return 0;
2122	}
2123
2124	/*
2125	 * If container object is not present, then there are no pages
2126	 * to put; just return without error.  This happens all the
2127	 * time, apparently during discard of a closed vnode (which
2128	 * trivially can't have dirty pages).
2129	 */
2130	cvp = cp->c_ovp;
2131	if (cvp == NULL) {
2132		rw_exit(vp->v_uobj.vmobjlock);
2133		return 0;
2134	}
2135
2136	/* Munge the arg structure to refer to the container vnode. */
2137	KASSERT(cvp->v_uobj.vmobjlock == vp->v_uobj.vmobjlock);
2138	ap->a_vp = cvp;
2139
2140	/* Finally, call putpages on it. */
2141	error = VCALL(ap->a_vp, VOFFSET(vop_putpages), ap);
2142
2143	return error;
2144}
2145