1/*-
2 * Copyright (c) 1982, 1986, 1990, 1993, 1995
3 *	The Regents of the University of California.  All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * Robert Elz at The University of Melbourne.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 *    may be used to endorse or promote products derived from this software
18 *    without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 *
32 *	@(#)ufs_quota.c	8.5 (Berkeley) 5/20/95
33 */
34
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD$");
37
38#include "opt_ffs.h"
39
40#include <sys/param.h>
41#include <sys/systm.h>
42#include <sys/endian.h>
43#include <sys/fcntl.h>
44#include <sys/kernel.h>
45#include <sys/lock.h>
46#include <sys/malloc.h>
47#include <sys/mount.h>
48#include <sys/mutex.h>
49#include <sys/namei.h>
50#include <sys/priv.h>
51#include <sys/proc.h>
52#include <sys/socket.h>
53#include <sys/stat.h>
54#include <sys/sysctl.h>
55#include <sys/vnode.h>
56
57#include <ufs/ufs/extattr.h>
58#include <ufs/ufs/quota.h>
59#include <ufs/ufs/inode.h>
60#include <ufs/ufs/ufsmount.h>
61#include <ufs/ufs/ufs_extern.h>
62
63CTASSERT(sizeof(struct dqblk64) == sizeof(struct dqhdr64));
64
65static int unprivileged_get_quota = 0;
66SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_get_quota, CTLFLAG_RW,
67    &unprivileged_get_quota, 0,
68    "Unprivileged processes may retrieve quotas for other uids and gids");
69
70static MALLOC_DEFINE(M_DQUOT, "ufs_quota", "UFS quota entries");
71
72/*
73 * Quota name to error message mapping.
74 */
75static char *quotatypes[] = INITQFNAMES;
76
77static int chkdqchg(struct inode *, ufs2_daddr_t, struct ucred *, int, int *);
78static int chkiqchg(struct inode *, int, struct ucred *, int, int *);
79static int dqopen(struct vnode *, struct ufsmount *, int);
80static int dqget(struct vnode *,
81	u_long, struct ufsmount *, int, struct dquot **);
82static int dqsync(struct vnode *, struct dquot *);
83static int dqflush(struct vnode *);
84static int quotaoff1(struct thread *td, struct mount *mp, int type);
85static int quotaoff_inchange(struct thread *td, struct mount *mp, int type);
86
87/* conversion functions - from_to() */
88static void dqb32_dq(const struct dqblk32 *, struct dquot *);
89static void dqb64_dq(const struct dqblk64 *, struct dquot *);
90static void dq_dqb32(const struct dquot *, struct dqblk32 *);
91static void dq_dqb64(const struct dquot *, struct dqblk64 *);
92static void dqb32_dqb64(const struct dqblk32 *, struct dqblk64 *);
93static void dqb64_dqb32(const struct dqblk64 *, struct dqblk32 *);
94
95#ifdef DIAGNOSTIC
96static void dqref(struct dquot *);
97static void chkdquot(struct inode *);
98#endif
99
100/*
101 * Set up the quotas for an inode.
102 *
103 * This routine completely defines the semantics of quotas.
104 * If other criterion want to be used to establish quotas, the
105 * MAXQUOTAS value in quota.h should be increased, and the
106 * additional dquots set up here.
107 */
108int
109getinoquota(struct inode *ip)
110{
111	struct ufsmount *ump;
112	struct vnode *vp;
113	int error;
114
115	vp = ITOV(ip);
116
117	/*
118	 * Disk quotas must be turned off for system files.  Currently
119	 * snapshot and quota files.
120	 */
121	if ((vp->v_vflag & VV_SYSTEM) != 0)
122		return (0);
123	/*
124	 * XXX: Turn off quotas for files with a negative UID or GID.
125	 * This prevents the creation of 100GB+ quota files.
126	 */
127	if ((int)ip->i_uid < 0 || (int)ip->i_gid < 0)
128		return (0);
129	ump = VFSTOUFS(vp->v_mount);
130	/*
131	 * Set up the user quota based on file uid.
132	 * EINVAL means that quotas are not enabled.
133	 */
134	if ((error =
135		dqget(vp, ip->i_uid, ump, USRQUOTA, &ip->i_dquot[USRQUOTA])) &&
136	    error != EINVAL)
137		return (error);
138	/*
139	 * Set up the group quota based on file gid.
140	 * EINVAL means that quotas are not enabled.
141	 */
142	if ((error =
143		dqget(vp, ip->i_gid, ump, GRPQUOTA, &ip->i_dquot[GRPQUOTA])) &&
144	    error != EINVAL)
145		return (error);
146	return (0);
147}
148
149/*
150 * Update disk usage, and take corrective action.
151 */
152int
153chkdq(struct inode *ip, ufs2_daddr_t change, struct ucred *cred, int flags)
154{
155	struct dquot *dq;
156	ufs2_daddr_t ncurblocks;
157	struct vnode *vp = ITOV(ip);
158	int i, error, warn, do_check;
159
160	/*
161	 * Disk quotas must be turned off for system files.  Currently
162	 * snapshot and quota files.
163	 */
164	if ((vp->v_vflag & VV_SYSTEM) != 0)
165		return (0);
166	/*
167	 * XXX: Turn off quotas for files with a negative UID or GID.
168	 * This prevents the creation of 100GB+ quota files.
169	 */
170	if ((int)ip->i_uid < 0 || (int)ip->i_gid < 0)
171		return (0);
172#ifdef DIAGNOSTIC
173	if ((flags & CHOWN) == 0)
174		chkdquot(ip);
175#endif
176	if (change == 0)
177		return (0);
178	if (change < 0) {
179		for (i = 0; i < MAXQUOTAS; i++) {
180			if ((dq = ip->i_dquot[i]) == NODQUOT)
181				continue;
182			DQI_LOCK(dq);
183			DQI_WAIT(dq, PINOD+1, "chkdq1");
184			ncurblocks = dq->dq_curblocks + change;
185			if (ncurblocks >= 0)
186				dq->dq_curblocks = ncurblocks;
187			else
188				dq->dq_curblocks = 0;
189			dq->dq_flags &= ~DQ_BLKS;
190			dq->dq_flags |= DQ_MOD;
191			DQI_UNLOCK(dq);
192		}
193		return (0);
194	}
195	if ((flags & FORCE) == 0 &&
196	    priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA, 0))
197		do_check = 1;
198	else
199		do_check = 0;
200	for (i = 0; i < MAXQUOTAS; i++) {
201		if ((dq = ip->i_dquot[i]) == NODQUOT)
202			continue;
203		warn = 0;
204		DQI_LOCK(dq);
205		DQI_WAIT(dq, PINOD+1, "chkdq2");
206		if (do_check) {
207			error = chkdqchg(ip, change, cred, i, &warn);
208			if (error) {
209				/*
210				 * Roll back user quota changes when
211				 * group quota failed.
212				 */
213				while (i > 0) {
214					--i;
215					dq = ip->i_dquot[i];
216					if (dq == NODQUOT)
217						continue;
218					DQI_LOCK(dq);
219					DQI_WAIT(dq, PINOD+1, "chkdq3");
220					ncurblocks = dq->dq_curblocks - change;
221					if (ncurblocks >= 0)
222						dq->dq_curblocks = ncurblocks;
223					else
224						dq->dq_curblocks = 0;
225					dq->dq_flags &= ~DQ_BLKS;
226					dq->dq_flags |= DQ_MOD;
227					DQI_UNLOCK(dq);
228				}
229				return (error);
230			}
231		}
232		/* Reset timer when crossing soft limit */
233		if (dq->dq_curblocks + change >= dq->dq_bsoftlimit &&
234		    dq->dq_curblocks < dq->dq_bsoftlimit)
235			dq->dq_btime = time_second + ip->i_ump->um_btime[i];
236		dq->dq_curblocks += change;
237		dq->dq_flags |= DQ_MOD;
238		DQI_UNLOCK(dq);
239		if (warn)
240			uprintf("\n%s: warning, %s disk quota exceeded\n",
241			    ITOV(ip)->v_mount->mnt_stat.f_mntonname,
242			    quotatypes[i]);
243	}
244	return (0);
245}
246
247/*
248 * Check for a valid change to a users allocation.
249 * Issue an error message if appropriate.
250 */
251static int
252chkdqchg(struct inode *ip, ufs2_daddr_t change, struct ucred *cred,
253    int type, int *warn)
254{
255	struct dquot *dq = ip->i_dquot[type];
256	ufs2_daddr_t ncurblocks = dq->dq_curblocks + change;
257
258	/*
259	 * If user would exceed their hard limit, disallow space allocation.
260	 */
261	if (ncurblocks >= dq->dq_bhardlimit && dq->dq_bhardlimit) {
262		if ((dq->dq_flags & DQ_BLKS) == 0 &&
263		    ip->i_uid == cred->cr_uid) {
264			dq->dq_flags |= DQ_BLKS;
265			DQI_UNLOCK(dq);
266			uprintf("\n%s: write failed, %s disk limit reached\n",
267			    ITOV(ip)->v_mount->mnt_stat.f_mntonname,
268			    quotatypes[type]);
269			return (EDQUOT);
270		}
271		DQI_UNLOCK(dq);
272		return (EDQUOT);
273	}
274	/*
275	 * If user is over their soft limit for too long, disallow space
276	 * allocation. Reset time limit as they cross their soft limit.
277	 */
278	if (ncurblocks >= dq->dq_bsoftlimit && dq->dq_bsoftlimit) {
279		if (dq->dq_curblocks < dq->dq_bsoftlimit) {
280			dq->dq_btime = time_second + ip->i_ump->um_btime[type];
281			if (ip->i_uid == cred->cr_uid)
282				*warn = 1;
283			return (0);
284		}
285		if (time_second > dq->dq_btime) {
286			if ((dq->dq_flags & DQ_BLKS) == 0 &&
287			    ip->i_uid == cred->cr_uid) {
288				dq->dq_flags |= DQ_BLKS;
289				DQI_UNLOCK(dq);
290				uprintf("\n%s: write failed, %s "
291				    "disk quota exceeded for too long\n",
292				    ITOV(ip)->v_mount->mnt_stat.f_mntonname,
293				    quotatypes[type]);
294				return (EDQUOT);
295			}
296			DQI_UNLOCK(dq);
297			return (EDQUOT);
298		}
299	}
300	return (0);
301}
302
303/*
304 * Check the inode limit, applying corrective action.
305 */
306int
307chkiq(struct inode *ip, int change, struct ucred *cred, int flags)
308{
309	struct dquot *dq;
310	ino_t ncurinodes;
311	int i, error, warn, do_check;
312
313#ifdef DIAGNOSTIC
314	if ((flags & CHOWN) == 0)
315		chkdquot(ip);
316#endif
317	if (change == 0)
318		return (0);
319	if (change < 0) {
320		for (i = 0; i < MAXQUOTAS; i++) {
321			if ((dq = ip->i_dquot[i]) == NODQUOT)
322				continue;
323			DQI_LOCK(dq);
324			DQI_WAIT(dq, PINOD+1, "chkiq1");
325			ncurinodes = dq->dq_curinodes + change;
326			/* XXX: ncurinodes is unsigned */
327			if (dq->dq_curinodes != 0 && ncurinodes >= 0)
328				dq->dq_curinodes = ncurinodes;
329			else
330				dq->dq_curinodes = 0;
331			dq->dq_flags &= ~DQ_INODS;
332			dq->dq_flags |= DQ_MOD;
333			DQI_UNLOCK(dq);
334		}
335		return (0);
336	}
337	if ((flags & FORCE) == 0 &&
338	    priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA, 0))
339		do_check = 1;
340	else
341		do_check = 0;
342	for (i = 0; i < MAXQUOTAS; i++) {
343		if ((dq = ip->i_dquot[i]) == NODQUOT)
344			continue;
345		warn = 0;
346		DQI_LOCK(dq);
347		DQI_WAIT(dq, PINOD+1, "chkiq2");
348		if (do_check) {
349			error = chkiqchg(ip, change, cred, i, &warn);
350			if (error) {
351				/*
352				 * Roll back user quota changes when
353				 * group quota failed.
354				 */
355				while (i > 0) {
356					--i;
357					dq = ip->i_dquot[i];
358					if (dq == NODQUOT)
359						continue;
360					DQI_LOCK(dq);
361					DQI_WAIT(dq, PINOD+1, "chkiq3");
362					ncurinodes = dq->dq_curinodes - change;
363					/* XXX: ncurinodes is unsigned */
364					if (dq->dq_curinodes != 0 &&
365					    ncurinodes >= 0)
366						dq->dq_curinodes = ncurinodes;
367					else
368						dq->dq_curinodes = 0;
369					dq->dq_flags &= ~DQ_INODS;
370					dq->dq_flags |= DQ_MOD;
371					DQI_UNLOCK(dq);
372				}
373				return (error);
374			}
375		}
376		/* Reset timer when crossing soft limit */
377		if (dq->dq_curinodes + change >= dq->dq_isoftlimit &&
378		    dq->dq_curinodes < dq->dq_isoftlimit)
379			dq->dq_itime = time_second + ip->i_ump->um_itime[i];
380		dq->dq_curinodes += change;
381		dq->dq_flags |= DQ_MOD;
382		DQI_UNLOCK(dq);
383		if (warn)
384			uprintf("\n%s: warning, %s inode quota exceeded\n",
385			    ITOV(ip)->v_mount->mnt_stat.f_mntonname,
386			    quotatypes[i]);
387	}
388	return (0);
389}
390
391/*
392 * Check for a valid change to a users allocation.
393 * Issue an error message if appropriate.
394 */
395static int
396chkiqchg(struct inode *ip, int change, struct ucred *cred, int type, int *warn)
397{
398	struct dquot *dq = ip->i_dquot[type];
399	ino_t ncurinodes = dq->dq_curinodes + change;
400
401	/*
402	 * If user would exceed their hard limit, disallow inode allocation.
403	 */
404	if (ncurinodes >= dq->dq_ihardlimit && dq->dq_ihardlimit) {
405		if ((dq->dq_flags & DQ_INODS) == 0 &&
406		    ip->i_uid == cred->cr_uid) {
407			dq->dq_flags |= DQ_INODS;
408			DQI_UNLOCK(dq);
409			uprintf("\n%s: write failed, %s inode limit reached\n",
410			    ITOV(ip)->v_mount->mnt_stat.f_mntonname,
411			    quotatypes[type]);
412			return (EDQUOT);
413		}
414		DQI_UNLOCK(dq);
415		return (EDQUOT);
416	}
417	/*
418	 * If user is over their soft limit for too long, disallow inode
419	 * allocation. Reset time limit as they cross their soft limit.
420	 */
421	if (ncurinodes >= dq->dq_isoftlimit && dq->dq_isoftlimit) {
422		if (dq->dq_curinodes < dq->dq_isoftlimit) {
423			dq->dq_itime = time_second + ip->i_ump->um_itime[type];
424			if (ip->i_uid == cred->cr_uid)
425				*warn = 1;
426			return (0);
427		}
428		if (time_second > dq->dq_itime) {
429			if ((dq->dq_flags & DQ_INODS) == 0 &&
430			    ip->i_uid == cred->cr_uid) {
431				dq->dq_flags |= DQ_INODS;
432				DQI_UNLOCK(dq);
433				uprintf("\n%s: write failed, %s "
434				    "inode quota exceeded for too long\n",
435				    ITOV(ip)->v_mount->mnt_stat.f_mntonname,
436				    quotatypes[type]);
437				return (EDQUOT);
438			}
439			DQI_UNLOCK(dq);
440			return (EDQUOT);
441		}
442	}
443	return (0);
444}
445
446#ifdef DIAGNOSTIC
447/*
448 * On filesystems with quotas enabled, it is an error for a file to change
449 * size and not to have a dquot structure associated with it.
450 */
451static void
452chkdquot(struct inode *ip)
453{
454	struct ufsmount *ump = ip->i_ump;
455	struct vnode *vp = ITOV(ip);
456	int i;
457
458	/*
459	 * Disk quotas must be turned off for system files.  Currently
460	 * these are snapshots and quota files.
461	 */
462	if ((vp->v_vflag & VV_SYSTEM) != 0)
463		return;
464	/*
465	 * XXX: Turn off quotas for files with a negative UID or GID.
466	 * This prevents the creation of 100GB+ quota files.
467	 */
468	if ((int)ip->i_uid < 0 || (int)ip->i_gid < 0)
469		return;
470
471	UFS_LOCK(ump);
472	for (i = 0; i < MAXQUOTAS; i++) {
473		if (ump->um_quotas[i] == NULLVP ||
474		    (ump->um_qflags[i] & (QTF_OPENING|QTF_CLOSING)))
475			continue;
476		if (ip->i_dquot[i] == NODQUOT) {
477			UFS_UNLOCK(ump);
478			vprint("chkdquot: missing dquot", ITOV(ip));
479			panic("chkdquot: missing dquot");
480		}
481	}
482	UFS_UNLOCK(ump);
483}
484#endif
485
486/*
487 * Code to process quotactl commands.
488 */
489
490/*
491 * Q_QUOTAON - set up a quota file for a particular filesystem.
492 */
493int
494quotaon(struct thread *td, struct mount *mp, int type, void *fname)
495{
496	struct ufsmount *ump;
497	struct vnode *vp, **vpp;
498	struct vnode *mvp;
499	struct dquot *dq;
500	int error, flags, vfslocked;
501	struct nameidata nd;
502
503	error = priv_check(td, PRIV_UFS_QUOTAON);
504	if (error)
505		return (error);
506
507	if (mp->mnt_flag & MNT_RDONLY)
508		return (EROFS);
509
510	ump = VFSTOUFS(mp);
511	dq = NODQUOT;
512
513	NDINIT(&nd, LOOKUP, FOLLOW | MPSAFE, UIO_USERSPACE, fname, td);
514	flags = FREAD | FWRITE;
515	vfs_ref(mp);
516	vfs_unbusy(mp);
517	error = vn_open(&nd, &flags, 0, NULL);
518	if (error != 0) {
519		vfs_rel(mp);
520		return (error);
521	}
522	vfslocked = NDHASGIANT(&nd);
523	NDFREE(&nd, NDF_ONLY_PNBUF);
524	vp = nd.ni_vp;
525	error = vfs_busy(mp, MBF_NOWAIT);
526	vfs_rel(mp);
527	if (error == 0) {
528		if (vp->v_type != VREG) {
529			error = EACCES;
530			vfs_unbusy(mp);
531		}
532	}
533	if (error != 0) {
534		VOP_UNLOCK(vp, 0);
535		(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
536		VFS_UNLOCK_GIANT(vfslocked);
537		return (error);
538	}
539
540	UFS_LOCK(ump);
541	if ((ump->um_qflags[type] & (QTF_OPENING|QTF_CLOSING)) != 0) {
542		UFS_UNLOCK(ump);
543		VOP_UNLOCK(vp, 0);
544		(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
545		VFS_UNLOCK_GIANT(vfslocked);
546		vfs_unbusy(mp);
547		return (EALREADY);
548	}
549	ump->um_qflags[type] |= QTF_OPENING|QTF_CLOSING;
550	UFS_UNLOCK(ump);
551	if ((error = dqopen(vp, ump, type)) != 0) {
552		VOP_UNLOCK(vp, 0);
553		UFS_LOCK(ump);
554		ump->um_qflags[type] &= ~(QTF_OPENING|QTF_CLOSING);
555		UFS_UNLOCK(ump);
556		(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
557		VFS_UNLOCK_GIANT(vfslocked);
558		vfs_unbusy(mp);
559		return (error);
560	}
561	VOP_UNLOCK(vp, 0);
562	MNT_ILOCK(mp);
563	mp->mnt_flag |= MNT_QUOTA;
564	MNT_IUNLOCK(mp);
565
566	vpp = &ump->um_quotas[type];
567	if (*vpp != vp)
568		quotaoff1(td, mp, type);
569
570	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
571	vp->v_vflag |= VV_SYSTEM;
572	VOP_UNLOCK(vp, 0);
573	*vpp = vp;
574	VFS_UNLOCK_GIANT(vfslocked);
575	/*
576	 * Save the credential of the process that turned on quotas.
577	 * Set up the time limits for this quota.
578	 */
579	ump->um_cred[type] = crhold(td->td_ucred);
580	ump->um_btime[type] = MAX_DQ_TIME;
581	ump->um_itime[type] = MAX_IQ_TIME;
582	if (dqget(NULLVP, 0, ump, type, &dq) == 0) {
583		if (dq->dq_btime > 0)
584			ump->um_btime[type] = dq->dq_btime;
585		if (dq->dq_itime > 0)
586			ump->um_itime[type] = dq->dq_itime;
587		dqrele(NULLVP, dq);
588	}
589	/*
590	 * Allow the getdq from getinoquota below to read the quota
591	 * from file.
592	 */
593	UFS_LOCK(ump);
594	ump->um_qflags[type] &= ~QTF_CLOSING;
595	UFS_UNLOCK(ump);
596	/*
597	 * Search vnodes associated with this mount point,
598	 * adding references to quota file being opened.
599	 * NB: only need to add dquot's for inodes being modified.
600	 */
601again:
602	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
603		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
604			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
605			goto again;
606		}
607		if (vp->v_type == VNON || vp->v_writecount == 0) {
608			VOP_UNLOCK(vp, 0);
609			vrele(vp);
610			continue;
611		}
612		error = getinoquota(VTOI(vp));
613		VOP_UNLOCK(vp, 0);
614		vrele(vp);
615		if (error) {
616			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
617			break;
618		}
619	}
620
621        if (error)
622		quotaoff_inchange(td, mp, type);
623	UFS_LOCK(ump);
624	ump->um_qflags[type] &= ~QTF_OPENING;
625	KASSERT((ump->um_qflags[type] & QTF_CLOSING) == 0,
626		("quotaon: leaking flags"));
627	UFS_UNLOCK(ump);
628
629	vfs_unbusy(mp);
630	return (error);
631}
632
633/*
634 * Main code to turn off disk quotas for a filesystem. Does not change
635 * flags.
636 */
637static int
638quotaoff1(struct thread *td, struct mount *mp, int type)
639{
640	struct vnode *vp;
641	struct vnode *qvp, *mvp;
642	struct ufsmount *ump;
643	struct dquot *dq;
644	struct inode *ip;
645	struct ucred *cr;
646	int vfslocked;
647	int error;
648
649	ump = VFSTOUFS(mp);
650
651	UFS_LOCK(ump);
652	KASSERT((ump->um_qflags[type] & QTF_CLOSING) != 0,
653		("quotaoff1: flags are invalid"));
654	if ((qvp = ump->um_quotas[type]) == NULLVP) {
655		UFS_UNLOCK(ump);
656		return (0);
657	}
658	cr = ump->um_cred[type];
659	UFS_UNLOCK(ump);
660
661	/*
662	 * Search vnodes associated with this mount point,
663	 * deleting any references to quota file being closed.
664	 */
665again:
666	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
667		if (vp->v_type == VNON) {
668			VI_UNLOCK(vp);
669			continue;
670		}
671		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
672			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
673			goto again;
674		}
675		ip = VTOI(vp);
676		dq = ip->i_dquot[type];
677		ip->i_dquot[type] = NODQUOT;
678		dqrele(vp, dq);
679		VOP_UNLOCK(vp, 0);
680		vrele(vp);
681	}
682
683	error = dqflush(qvp);
684	if (error != 0)
685		return (error);
686
687	/*
688	 * Clear um_quotas before closing the quota vnode to prevent
689	 * access to the closed vnode from dqget/dqsync
690	 */
691	UFS_LOCK(ump);
692	ump->um_quotas[type] = NULLVP;
693	ump->um_cred[type] = NOCRED;
694	UFS_UNLOCK(ump);
695
696	vfslocked = VFS_LOCK_GIANT(qvp->v_mount);
697	vn_lock(qvp, LK_EXCLUSIVE | LK_RETRY);
698	qvp->v_vflag &= ~VV_SYSTEM;
699	VOP_UNLOCK(qvp, 0);
700	error = vn_close(qvp, FREAD|FWRITE, td->td_ucred, td);
701	VFS_UNLOCK_GIANT(vfslocked);
702	crfree(cr);
703
704	return (error);
705}
706
707/*
708 * Turns off quotas, assumes that ump->um_qflags are already checked
709 * and QTF_CLOSING is set to indicate operation in progress. Fixes
710 * ump->um_qflags and mp->mnt_flag after.
711 */
712int
713quotaoff_inchange(struct thread *td, struct mount *mp, int type)
714{
715	struct ufsmount *ump;
716	int i;
717	int error;
718
719	error = quotaoff1(td, mp, type);
720
721	ump = VFSTOUFS(mp);
722	UFS_LOCK(ump);
723	ump->um_qflags[type] &= ~QTF_CLOSING;
724	for (i = 0; i < MAXQUOTAS; i++)
725		if (ump->um_quotas[i] != NULLVP)
726			break;
727	if (i == MAXQUOTAS) {
728		MNT_ILOCK(mp);
729		mp->mnt_flag &= ~MNT_QUOTA;
730		MNT_IUNLOCK(mp);
731	}
732	UFS_UNLOCK(ump);
733	return (error);
734}
735
736/*
737 * Q_QUOTAOFF - turn off disk quotas for a filesystem.
738 */
739int
740quotaoff(struct thread *td, struct mount *mp, int type)
741{
742	struct ufsmount *ump;
743	int error;
744
745	error = priv_check(td, PRIV_UFS_QUOTAOFF);
746	if (error)
747		return (error);
748
749	ump = VFSTOUFS(mp);
750	UFS_LOCK(ump);
751	if ((ump->um_qflags[type] & (QTF_OPENING|QTF_CLOSING)) != 0) {
752		UFS_UNLOCK(ump);
753		return (EALREADY);
754	}
755	ump->um_qflags[type] |= QTF_CLOSING;
756	UFS_UNLOCK(ump);
757
758	return (quotaoff_inchange(td, mp, type));
759}
760
761/*
762 * Q_GETQUOTA - return current values in a dqblk structure.
763 */
764static int
765_getquota(struct thread *td, struct mount *mp, u_long id, int type,
766    struct dqblk64 *dqb)
767{
768	struct dquot *dq;
769	int error;
770
771	switch (type) {
772	case USRQUOTA:
773		if ((td->td_ucred->cr_uid != id) && !unprivileged_get_quota) {
774			error = priv_check(td, PRIV_VFS_GETQUOTA);
775			if (error)
776				return (error);
777		}
778		break;
779
780	case GRPQUOTA:
781		if (!groupmember(id, td->td_ucred) &&
782		    !unprivileged_get_quota) {
783			error = priv_check(td, PRIV_VFS_GETQUOTA);
784			if (error)
785				return (error);
786		}
787		break;
788
789	default:
790		return (EINVAL);
791	}
792
793	dq = NODQUOT;
794	error = dqget(NULLVP, id, VFSTOUFS(mp), type, &dq);
795	if (error)
796		return (error);
797	*dqb = dq->dq_dqb;
798	dqrele(NULLVP, dq);
799	return (error);
800}
801
802/*
803 * Q_SETQUOTA - assign an entire dqblk structure.
804 */
805static int
806_setquota(struct thread *td, struct mount *mp, u_long id, int type,
807    struct dqblk64 *dqb)
808{
809	struct dquot *dq;
810	struct dquot *ndq;
811	struct ufsmount *ump;
812	struct dqblk64 newlim;
813	int error;
814
815	error = priv_check(td, PRIV_VFS_SETQUOTA);
816	if (error)
817		return (error);
818
819	newlim = *dqb;
820
821	ndq = NODQUOT;
822	ump = VFSTOUFS(mp);
823
824	error = dqget(NULLVP, id, ump, type, &ndq);
825	if (error)
826		return (error);
827	dq = ndq;
828	DQI_LOCK(dq);
829	DQI_WAIT(dq, PINOD+1, "setqta");
830	/*
831	 * Copy all but the current values.
832	 * Reset time limit if previously had no soft limit or were
833	 * under it, but now have a soft limit and are over it.
834	 */
835	newlim.dqb_curblocks = dq->dq_curblocks;
836	newlim.dqb_curinodes = dq->dq_curinodes;
837	if (dq->dq_id != 0) {
838		newlim.dqb_btime = dq->dq_btime;
839		newlim.dqb_itime = dq->dq_itime;
840	}
841	if (newlim.dqb_bsoftlimit &&
842	    dq->dq_curblocks >= newlim.dqb_bsoftlimit &&
843	    (dq->dq_bsoftlimit == 0 || dq->dq_curblocks < dq->dq_bsoftlimit))
844		newlim.dqb_btime = time_second + ump->um_btime[type];
845	if (newlim.dqb_isoftlimit &&
846	    dq->dq_curinodes >= newlim.dqb_isoftlimit &&
847	    (dq->dq_isoftlimit == 0 || dq->dq_curinodes < dq->dq_isoftlimit))
848		newlim.dqb_itime = time_second + ump->um_itime[type];
849	dq->dq_dqb = newlim;
850	if (dq->dq_curblocks < dq->dq_bsoftlimit)
851		dq->dq_flags &= ~DQ_BLKS;
852	if (dq->dq_curinodes < dq->dq_isoftlimit)
853		dq->dq_flags &= ~DQ_INODS;
854	if (dq->dq_isoftlimit == 0 && dq->dq_bsoftlimit == 0 &&
855	    dq->dq_ihardlimit == 0 && dq->dq_bhardlimit == 0)
856		dq->dq_flags |= DQ_FAKE;
857	else
858		dq->dq_flags &= ~DQ_FAKE;
859	dq->dq_flags |= DQ_MOD;
860	DQI_UNLOCK(dq);
861	dqrele(NULLVP, dq);
862	return (0);
863}
864
865/*
866 * Q_SETUSE - set current inode and block usage.
867 */
868static int
869_setuse(struct thread *td, struct mount *mp, u_long id, int type,
870    struct dqblk64 *dqb)
871{
872	struct dquot *dq;
873	struct ufsmount *ump;
874	struct dquot *ndq;
875	struct dqblk64 usage;
876	int error;
877
878	error = priv_check(td, PRIV_UFS_SETUSE);
879	if (error)
880		return (error);
881
882	usage = *dqb;
883
884	ump = VFSTOUFS(mp);
885	ndq = NODQUOT;
886
887	error = dqget(NULLVP, id, ump, type, &ndq);
888	if (error)
889		return (error);
890	dq = ndq;
891	DQI_LOCK(dq);
892	DQI_WAIT(dq, PINOD+1, "setuse");
893	/*
894	 * Reset time limit if have a soft limit and were
895	 * previously under it, but are now over it.
896	 */
897	if (dq->dq_bsoftlimit && dq->dq_curblocks < dq->dq_bsoftlimit &&
898	    usage.dqb_curblocks >= dq->dq_bsoftlimit)
899		dq->dq_btime = time_second + ump->um_btime[type];
900	if (dq->dq_isoftlimit && dq->dq_curinodes < dq->dq_isoftlimit &&
901	    usage.dqb_curinodes >= dq->dq_isoftlimit)
902		dq->dq_itime = time_second + ump->um_itime[type];
903	dq->dq_curblocks = usage.dqb_curblocks;
904	dq->dq_curinodes = usage.dqb_curinodes;
905	if (dq->dq_curblocks < dq->dq_bsoftlimit)
906		dq->dq_flags &= ~DQ_BLKS;
907	if (dq->dq_curinodes < dq->dq_isoftlimit)
908		dq->dq_flags &= ~DQ_INODS;
909	dq->dq_flags |= DQ_MOD;
910	DQI_UNLOCK(dq);
911	dqrele(NULLVP, dq);
912	return (0);
913}
914
915int
916getquota32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
917{
918	struct dqblk32 dqb32;
919	struct dqblk64 dqb64;
920	int error;
921
922	error = _getquota(td, mp, id, type, &dqb64);
923	if (error)
924		return (error);
925	dqb64_dqb32(&dqb64, &dqb32);
926	error = copyout(&dqb32, addr, sizeof(dqb32));
927	return (error);
928}
929
930int
931setquota32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
932{
933	struct dqblk32 dqb32;
934	struct dqblk64 dqb64;
935	int error;
936
937	error = copyin(addr, &dqb32, sizeof(dqb32));
938	if (error)
939		return (error);
940	dqb32_dqb64(&dqb32, &dqb64);
941	error = _setquota(td, mp, id, type, &dqb64);
942	return (error);
943}
944
945int
946setuse32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
947{
948	struct dqblk32 dqb32;
949	struct dqblk64 dqb64;
950	int error;
951
952	error = copyin(addr, &dqb32, sizeof(dqb32));
953	if (error)
954		return (error);
955	dqb32_dqb64(&dqb32, &dqb64);
956	error = _setuse(td, mp, id, type, &dqb64);
957	return (error);
958}
959
960int
961getquota(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
962{
963	struct dqblk64 dqb64;
964	int error;
965
966	error = _getquota(td, mp, id, type, &dqb64);
967	if (error)
968		return (error);
969	error = copyout(&dqb64, addr, sizeof(dqb64));
970	return (error);
971}
972
973int
974setquota(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
975{
976	struct dqblk64 dqb64;
977	int error;
978
979	error = copyin(addr, &dqb64, sizeof(dqb64));
980	if (error)
981		return (error);
982	error = _setquota(td, mp, id, type, &dqb64);
983	return (error);
984}
985
986int
987setuse(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
988{
989	struct dqblk64 dqb64;
990	int error;
991
992	error = copyin(addr, &dqb64, sizeof(dqb64));
993	if (error)
994		return (error);
995	error = _setuse(td, mp, id, type, &dqb64);
996	return (error);
997}
998
999/*
1000 * Q_GETQUOTASIZE - get bit-size of quota file fields
1001 */
1002int
1003getquotasize(struct thread *td, struct mount *mp, u_long id, int type,
1004    void *sizep)
1005{
1006	struct ufsmount *ump = VFSTOUFS(mp);
1007	int bitsize;
1008
1009	UFS_LOCK(ump);
1010	if (ump->um_quotas[type] == NULLVP ||
1011	    (ump->um_qflags[type] & QTF_CLOSING)) {
1012		UFS_UNLOCK(ump);
1013		return (EINVAL);
1014	}
1015	if ((ump->um_qflags[type] & QTF_64BIT) != 0)
1016		bitsize = 64;
1017	else
1018		bitsize = 32;
1019	UFS_UNLOCK(ump);
1020	return (copyout(&bitsize, sizep, sizeof(int)));
1021}
1022
1023/*
1024 * Q_SYNC - sync quota files to disk.
1025 */
1026int
1027qsync(struct mount *mp)
1028{
1029	struct ufsmount *ump = VFSTOUFS(mp);
1030	struct thread *td = curthread;		/* XXX */
1031	struct vnode *vp, *mvp;
1032	struct dquot *dq;
1033	int i, error;
1034
1035	/*
1036	 * Check if the mount point has any quotas.
1037	 * If not, simply return.
1038	 */
1039	UFS_LOCK(ump);
1040	for (i = 0; i < MAXQUOTAS; i++)
1041		if (ump->um_quotas[i] != NULLVP)
1042			break;
1043	UFS_UNLOCK(ump);
1044	if (i == MAXQUOTAS)
1045		return (0);
1046	/*
1047	 * Search vnodes associated with this mount point,
1048	 * synchronizing any modified dquot structures.
1049	 */
1050again:
1051	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
1052		if (vp->v_type == VNON) {
1053			VI_UNLOCK(vp);
1054			continue;
1055		}
1056		error = vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td);
1057		if (error) {
1058			if (error == ENOENT) {
1059				MNT_VNODE_FOREACH_ACTIVE_ABORT(mp, mvp);
1060				goto again;
1061			}
1062			continue;
1063		}
1064		for (i = 0; i < MAXQUOTAS; i++) {
1065			dq = VTOI(vp)->i_dquot[i];
1066			if (dq != NODQUOT)
1067				dqsync(vp, dq);
1068		}
1069		vput(vp);
1070	}
1071	return (0);
1072}
1073
1074/*
1075 * Sync quota file for given vnode to disk.
1076 */
1077int
1078qsyncvp(struct vnode *vp)
1079{
1080	struct ufsmount *ump = VFSTOUFS(vp->v_mount);
1081	struct dquot *dq;
1082	int i;
1083
1084	/*
1085	 * Check if the mount point has any quotas.
1086	 * If not, simply return.
1087	 */
1088	UFS_LOCK(ump);
1089	for (i = 0; i < MAXQUOTAS; i++)
1090		if (ump->um_quotas[i] != NULLVP)
1091			break;
1092	UFS_UNLOCK(ump);
1093	if (i == MAXQUOTAS)
1094		return (0);
1095	/*
1096	 * Search quotas associated with this vnode
1097	 * synchronizing any modified dquot structures.
1098	 */
1099	for (i = 0; i < MAXQUOTAS; i++) {
1100		dq = VTOI(vp)->i_dquot[i];
1101		if (dq != NODQUOT)
1102			dqsync(vp, dq);
1103	}
1104	return (0);
1105}
1106
1107/*
1108 * Code pertaining to management of the in-core dquot data structures.
1109 */
1110#define DQHASH(dqvp, id) \
1111	(&dqhashtbl[((((intptr_t)(dqvp)) >> 8) + id) & dqhash])
1112static LIST_HEAD(dqhash, dquot) *dqhashtbl;
1113static u_long dqhash;
1114
1115/*
1116 * Dquot free list.
1117 */
1118#define	DQUOTINC	5	/* minimum free dquots desired */
1119static TAILQ_HEAD(dqfreelist, dquot) dqfreelist;
1120static long numdquot, desireddquot = DQUOTINC;
1121
1122/*
1123 * Lock to protect quota hash, dq free list and dq_cnt ref counters of
1124 * _all_ dqs.
1125 */
1126struct mtx dqhlock;
1127
1128#define	DQH_LOCK()	mtx_lock(&dqhlock)
1129#define	DQH_UNLOCK()	mtx_unlock(&dqhlock)
1130
1131static struct dquot *dqhashfind(struct dqhash *dqh, u_long id,
1132	struct vnode *dqvp);
1133
1134/*
1135 * Initialize the quota system.
1136 */
1137void
1138dqinit(void)
1139{
1140
1141	mtx_init(&dqhlock, "dqhlock", NULL, MTX_DEF);
1142	dqhashtbl = hashinit(desiredvnodes, M_DQUOT, &dqhash);
1143	TAILQ_INIT(&dqfreelist);
1144}
1145
1146/*
1147 * Shut down the quota system.
1148 */
1149void
1150dquninit(void)
1151{
1152	struct dquot *dq;
1153
1154	hashdestroy(dqhashtbl, M_DQUOT, dqhash);
1155	while ((dq = TAILQ_FIRST(&dqfreelist)) != NULL) {
1156		TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
1157		mtx_destroy(&dq->dq_lock);
1158		free(dq, M_DQUOT);
1159	}
1160	mtx_destroy(&dqhlock);
1161}
1162
1163static struct dquot *
1164dqhashfind(struct dqhash *dqh, u_long id, struct vnode *dqvp)
1165{
1166	struct dquot *dq;
1167
1168	mtx_assert(&dqhlock, MA_OWNED);
1169	LIST_FOREACH(dq, dqh, dq_hash) {
1170		if (dq->dq_id != id ||
1171		    dq->dq_ump->um_quotas[dq->dq_type] != dqvp)
1172			continue;
1173		/*
1174		 * Cache hit with no references.  Take
1175		 * the structure off the free list.
1176		 */
1177		if (dq->dq_cnt == 0)
1178			TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
1179		DQREF(dq);
1180		return (dq);
1181	}
1182	return (NODQUOT);
1183}
1184
1185/*
1186 * Determine the quota file type.
1187 *
1188 * A 32-bit quota file is simply an array of struct dqblk32.
1189 *
1190 * A 64-bit quota file is a struct dqhdr64 followed by an array of struct
1191 * dqblk64.  The header contains various magic bits which allow us to be
1192 * reasonably confident that it is indeeda 64-bit quota file and not just
1193 * a 32-bit quota file that just happens to "look right".
1194 *
1195 */
1196static int
1197dqopen(struct vnode *vp, struct ufsmount *ump, int type)
1198{
1199	struct dqhdr64 dqh;
1200	struct iovec aiov;
1201	struct uio auio;
1202	int error;
1203
1204	ASSERT_VOP_LOCKED(vp, "dqopen");
1205	auio.uio_iov = &aiov;
1206	auio.uio_iovcnt = 1;
1207	aiov.iov_base = &dqh;
1208	aiov.iov_len = sizeof(dqh);
1209	auio.uio_resid = sizeof(dqh);
1210	auio.uio_offset = 0;
1211	auio.uio_segflg = UIO_SYSSPACE;
1212	auio.uio_rw = UIO_READ;
1213	auio.uio_td = (struct thread *)0;
1214	error = VOP_READ(vp, &auio, 0, ump->um_cred[type]);
1215
1216	if (error != 0)
1217		return (error);
1218	if (auio.uio_resid > 0) {
1219		/* assume 32 bits */
1220		return (0);
1221	}
1222
1223	UFS_LOCK(ump);
1224	if (strcmp(dqh.dqh_magic, Q_DQHDR64_MAGIC) == 0 &&
1225	    be32toh(dqh.dqh_version) == Q_DQHDR64_VERSION &&
1226	    be32toh(dqh.dqh_hdrlen) == (uint32_t)sizeof(struct dqhdr64) &&
1227	    be32toh(dqh.dqh_reclen) == (uint32_t)sizeof(struct dqblk64)) {
1228		/* XXX: what if the magic matches, but the sizes are wrong? */
1229		ump->um_qflags[type] |= QTF_64BIT;
1230	} else {
1231		ump->um_qflags[type] &= ~QTF_64BIT;
1232	}
1233	UFS_UNLOCK(ump);
1234
1235	return (0);
1236}
1237
1238/*
1239 * Obtain a dquot structure for the specified identifier and quota file
1240 * reading the information from the file if necessary.
1241 */
1242static int
1243dqget(struct vnode *vp, u_long id, struct ufsmount *ump, int type,
1244    struct dquot **dqp)
1245{
1246	uint8_t buf[sizeof(struct dqblk64)];
1247	off_t base, recsize;
1248	struct dquot *dq, *dq1;
1249	struct dqhash *dqh;
1250	struct vnode *dqvp;
1251	struct iovec aiov;
1252	struct uio auio;
1253	int vfslocked, dqvplocked, error;
1254
1255#ifdef DEBUG_VFS_LOCKS
1256	if (vp != NULLVP)
1257		ASSERT_VOP_ELOCKED(vp, "dqget");
1258#endif
1259
1260	if (vp != NULLVP && *dqp != NODQUOT) {
1261		return (0);
1262	}
1263
1264	/* XXX: Disallow negative id values to prevent the
1265	* creation of 100GB+ quota data files.
1266	*/
1267	if ((int)id < 0)
1268		return (EINVAL);
1269
1270	UFS_LOCK(ump);
1271	dqvp = ump->um_quotas[type];
1272	if (dqvp == NULLVP || (ump->um_qflags[type] & QTF_CLOSING)) {
1273		*dqp = NODQUOT;
1274		UFS_UNLOCK(ump);
1275		return (EINVAL);
1276	}
1277	vref(dqvp);
1278	UFS_UNLOCK(ump);
1279	error = 0;
1280	dqvplocked = 0;
1281
1282	/*
1283	 * Check the cache first.
1284	 */
1285	dqh = DQHASH(dqvp, id);
1286	DQH_LOCK();
1287	dq = dqhashfind(dqh, id, dqvp);
1288	if (dq != NULL) {
1289		DQH_UNLOCK();
1290hfound:		DQI_LOCK(dq);
1291		DQI_WAIT(dq, PINOD+1, "dqget");
1292		DQI_UNLOCK(dq);
1293		if (dq->dq_ump == NULL) {
1294			dqrele(vp, dq);
1295			dq = NODQUOT;
1296			error = EIO;
1297		}
1298		*dqp = dq;
1299		vfslocked = VFS_LOCK_GIANT(dqvp->v_mount);
1300		if (dqvplocked)
1301			vput(dqvp);
1302		else
1303			vrele(dqvp);
1304		VFS_UNLOCK_GIANT(vfslocked);
1305		return (error);
1306	}
1307
1308	/*
1309	 * Quota vnode lock is before DQ_LOCK. Acquire dqvp lock there
1310	 * since new dq will appear on the hash chain DQ_LOCKed.
1311	 */
1312	if (vp != dqvp) {
1313		DQH_UNLOCK();
1314		vn_lock(dqvp, LK_SHARED | LK_RETRY);
1315		dqvplocked = 1;
1316		DQH_LOCK();
1317		/*
1318		 * Recheck the cache after sleep for quota vnode lock.
1319		 */
1320		dq = dqhashfind(dqh, id, dqvp);
1321		if (dq != NULL) {
1322			DQH_UNLOCK();
1323			goto hfound;
1324		}
1325	}
1326
1327	/*
1328	 * Not in cache, allocate a new one or take it from the
1329	 * free list.
1330	 */
1331	if (TAILQ_FIRST(&dqfreelist) == NODQUOT &&
1332	    numdquot < MAXQUOTAS * desiredvnodes)
1333		desireddquot += DQUOTINC;
1334	if (numdquot < desireddquot) {
1335		numdquot++;
1336		DQH_UNLOCK();
1337		dq1 = malloc(sizeof *dq1, M_DQUOT, M_WAITOK | M_ZERO);
1338		mtx_init(&dq1->dq_lock, "dqlock", NULL, MTX_DEF);
1339		DQH_LOCK();
1340		/*
1341		 * Recheck the cache after sleep for memory.
1342		 */
1343		dq = dqhashfind(dqh, id, dqvp);
1344		if (dq != NULL) {
1345			numdquot--;
1346			DQH_UNLOCK();
1347			mtx_destroy(&dq1->dq_lock);
1348			free(dq1, M_DQUOT);
1349			goto hfound;
1350		}
1351		dq = dq1;
1352	} else {
1353		if ((dq = TAILQ_FIRST(&dqfreelist)) == NULL) {
1354			DQH_UNLOCK();
1355			tablefull("dquot");
1356			*dqp = NODQUOT;
1357			vfslocked = VFS_LOCK_GIANT(dqvp->v_mount);
1358			if (dqvplocked)
1359				vput(dqvp);
1360			else
1361				vrele(dqvp);
1362			VFS_UNLOCK_GIANT(vfslocked);
1363			return (EUSERS);
1364		}
1365		if (dq->dq_cnt || (dq->dq_flags & DQ_MOD))
1366			panic("dqget: free dquot isn't %p", dq);
1367		TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
1368		if (dq->dq_ump != NULL)
1369			LIST_REMOVE(dq, dq_hash);
1370	}
1371
1372	/*
1373	 * Dq is put into hash already locked to prevent parallel
1374	 * usage while it is being read from file.
1375	 */
1376	dq->dq_flags = DQ_LOCK;
1377	dq->dq_id = id;
1378	dq->dq_type = type;
1379	dq->dq_ump = ump;
1380	LIST_INSERT_HEAD(dqh, dq, dq_hash);
1381	DQREF(dq);
1382	DQH_UNLOCK();
1383
1384	/*
1385	 * Read the requested quota record from the quota file, performing
1386	 * any necessary conversions.
1387	 */
1388	if (ump->um_qflags[type] & QTF_64BIT) {
1389		recsize = sizeof(struct dqblk64);
1390		base = sizeof(struct dqhdr64);
1391	} else {
1392		recsize = sizeof(struct dqblk32);
1393		base = 0;
1394	}
1395	auio.uio_iov = &aiov;
1396	auio.uio_iovcnt = 1;
1397	aiov.iov_base = buf;
1398	aiov.iov_len = recsize;
1399	auio.uio_resid = recsize;
1400	auio.uio_offset = base + id * recsize;
1401	auio.uio_segflg = UIO_SYSSPACE;
1402	auio.uio_rw = UIO_READ;
1403	auio.uio_td = (struct thread *)0;
1404
1405	vfslocked = VFS_LOCK_GIANT(dqvp->v_mount);
1406	error = VOP_READ(dqvp, &auio, 0, ump->um_cred[type]);
1407	if (auio.uio_resid == recsize && error == 0) {
1408		bzero(&dq->dq_dqb, sizeof(dq->dq_dqb));
1409	} else {
1410		if (ump->um_qflags[type] & QTF_64BIT)
1411			dqb64_dq((struct dqblk64 *)buf, dq);
1412		else
1413			dqb32_dq((struct dqblk32 *)buf, dq);
1414	}
1415	if (dqvplocked)
1416		vput(dqvp);
1417	else
1418		vrele(dqvp);
1419	VFS_UNLOCK_GIANT(vfslocked);
1420	/*
1421	 * I/O error in reading quota file, release
1422	 * quota structure and reflect problem to caller.
1423	 */
1424	if (error) {
1425		DQH_LOCK();
1426		dq->dq_ump = NULL;
1427		LIST_REMOVE(dq, dq_hash);
1428		DQH_UNLOCK();
1429		DQI_LOCK(dq);
1430		if (dq->dq_flags & DQ_WANT)
1431			wakeup(dq);
1432		dq->dq_flags = 0;
1433		DQI_UNLOCK(dq);
1434		dqrele(vp, dq);
1435		*dqp = NODQUOT;
1436		return (error);
1437	}
1438	DQI_LOCK(dq);
1439	/*
1440	 * Check for no limit to enforce.
1441	 * Initialize time values if necessary.
1442	 */
1443	if (dq->dq_isoftlimit == 0 && dq->dq_bsoftlimit == 0 &&
1444	    dq->dq_ihardlimit == 0 && dq->dq_bhardlimit == 0)
1445		dq->dq_flags |= DQ_FAKE;
1446	if (dq->dq_id != 0) {
1447		if (dq->dq_btime == 0) {
1448			dq->dq_btime = time_second + ump->um_btime[type];
1449			if (dq->dq_bsoftlimit &&
1450			    dq->dq_curblocks >= dq->dq_bsoftlimit)
1451				dq->dq_flags |= DQ_MOD;
1452		}
1453		if (dq->dq_itime == 0) {
1454			dq->dq_itime = time_second + ump->um_itime[type];
1455			if (dq->dq_isoftlimit &&
1456			    dq->dq_curinodes >= dq->dq_isoftlimit)
1457				dq->dq_flags |= DQ_MOD;
1458		}
1459	}
1460	DQI_WAKEUP(dq);
1461	DQI_UNLOCK(dq);
1462	*dqp = dq;
1463	return (0);
1464}
1465
1466#ifdef DIAGNOSTIC
1467/*
1468 * Obtain a reference to a dquot.
1469 */
1470static void
1471dqref(struct dquot *dq)
1472{
1473
1474	dq->dq_cnt++;
1475}
1476#endif
1477
1478/*
1479 * Release a reference to a dquot.
1480 */
1481void
1482dqrele(struct vnode *vp, struct dquot *dq)
1483{
1484
1485	if (dq == NODQUOT)
1486		return;
1487	DQH_LOCK();
1488	KASSERT(dq->dq_cnt > 0, ("Lost dq %p reference 1", dq));
1489	if (dq->dq_cnt > 1) {
1490		dq->dq_cnt--;
1491		DQH_UNLOCK();
1492		return;
1493	}
1494	DQH_UNLOCK();
1495sync:
1496	(void) dqsync(vp, dq);
1497
1498	DQH_LOCK();
1499	KASSERT(dq->dq_cnt > 0, ("Lost dq %p reference 2", dq));
1500	if (--dq->dq_cnt > 0)
1501	{
1502		DQH_UNLOCK();
1503		return;
1504	}
1505
1506	/*
1507	 * The dq may become dirty after it is synced but before it is
1508	 * put to the free list. Checking the DQ_MOD there without
1509	 * locking dq should be safe since no other references to the
1510	 * dq exist.
1511	 */
1512	if ((dq->dq_flags & DQ_MOD) != 0) {
1513		dq->dq_cnt++;
1514		DQH_UNLOCK();
1515		goto sync;
1516	}
1517	TAILQ_INSERT_TAIL(&dqfreelist, dq, dq_freelist);
1518	DQH_UNLOCK();
1519}
1520
1521/*
1522 * Update the disk quota in the quota file.
1523 */
1524static int
1525dqsync(struct vnode *vp, struct dquot *dq)
1526{
1527	uint8_t buf[sizeof(struct dqblk64)];
1528	off_t base, recsize;
1529	struct vnode *dqvp;
1530	struct iovec aiov;
1531	struct uio auio;
1532	int vfslocked, error;
1533	struct mount *mp;
1534	struct ufsmount *ump;
1535
1536#ifdef DEBUG_VFS_LOCKS
1537	if (vp != NULL)
1538		ASSERT_VOP_ELOCKED(vp, "dqsync");
1539#endif
1540
1541	mp = NULL;
1542	error = 0;
1543	if (dq == NODQUOT)
1544		panic("dqsync: dquot");
1545	if ((ump = dq->dq_ump) == NULL)
1546		return (0);
1547	UFS_LOCK(ump);
1548	if ((dqvp = ump->um_quotas[dq->dq_type]) == NULLVP)
1549		panic("dqsync: file");
1550	vref(dqvp);
1551	UFS_UNLOCK(ump);
1552
1553	vfslocked = VFS_LOCK_GIANT(dqvp->v_mount);
1554	DQI_LOCK(dq);
1555	if ((dq->dq_flags & DQ_MOD) == 0) {
1556		DQI_UNLOCK(dq);
1557		vrele(dqvp);
1558		VFS_UNLOCK_GIANT(vfslocked);
1559		return (0);
1560	}
1561	DQI_UNLOCK(dq);
1562
1563	(void) vn_start_secondary_write(dqvp, &mp, V_WAIT);
1564	if (vp != dqvp)
1565		vn_lock(dqvp, LK_EXCLUSIVE | LK_RETRY);
1566
1567	VFS_UNLOCK_GIANT(vfslocked);
1568	DQI_LOCK(dq);
1569	DQI_WAIT(dq, PINOD+2, "dqsync");
1570	if ((dq->dq_flags & DQ_MOD) == 0)
1571		goto out;
1572	dq->dq_flags |= DQ_LOCK;
1573	DQI_UNLOCK(dq);
1574
1575	/*
1576	 * Write the quota record to the quota file, performing any
1577	 * necessary conversions.  See dqget() for additional details.
1578	 */
1579	if (ump->um_qflags[dq->dq_type] & QTF_64BIT) {
1580		dq_dqb64(dq, (struct dqblk64 *)buf);
1581		recsize = sizeof(struct dqblk64);
1582		base = sizeof(struct dqhdr64);
1583	} else {
1584		dq_dqb32(dq, (struct dqblk32 *)buf);
1585		recsize = sizeof(struct dqblk32);
1586		base = 0;
1587	}
1588
1589	auio.uio_iov = &aiov;
1590	auio.uio_iovcnt = 1;
1591	aiov.iov_base = buf;
1592	aiov.iov_len = recsize;
1593	auio.uio_resid = recsize;
1594	auio.uio_offset = base + dq->dq_id * recsize;
1595	auio.uio_segflg = UIO_SYSSPACE;
1596	auio.uio_rw = UIO_WRITE;
1597	auio.uio_td = (struct thread *)0;
1598	vfslocked = VFS_LOCK_GIANT(dqvp->v_mount);
1599	error = VOP_WRITE(dqvp, &auio, 0, dq->dq_ump->um_cred[dq->dq_type]);
1600	VFS_UNLOCK_GIANT(vfslocked);
1601	if (auio.uio_resid && error == 0)
1602		error = EIO;
1603
1604	DQI_LOCK(dq);
1605	DQI_WAKEUP(dq);
1606	dq->dq_flags &= ~DQ_MOD;
1607out:
1608	DQI_UNLOCK(dq);
1609	vfslocked = VFS_LOCK_GIANT(dqvp->v_mount);
1610	if (vp != dqvp)
1611		vput(dqvp);
1612	else
1613		vrele(dqvp);
1614	vn_finished_secondary_write(mp);
1615	VFS_UNLOCK_GIANT(vfslocked);
1616	return (error);
1617}
1618
1619/*
1620 * Flush all entries from the cache for a particular vnode.
1621 */
1622static int
1623dqflush(struct vnode *vp)
1624{
1625	struct dquot *dq, *nextdq;
1626	struct dqhash *dqh;
1627	int error;
1628
1629	/*
1630	 * Move all dquot's that used to refer to this quota
1631	 * file off their hash chains (they will eventually
1632	 * fall off the head of the free list and be re-used).
1633	 */
1634	error = 0;
1635	DQH_LOCK();
1636	for (dqh = &dqhashtbl[dqhash]; dqh >= dqhashtbl; dqh--) {
1637		for (dq = LIST_FIRST(dqh); dq; dq = nextdq) {
1638			nextdq = LIST_NEXT(dq, dq_hash);
1639			if (dq->dq_ump->um_quotas[dq->dq_type] != vp)
1640				continue;
1641			if (dq->dq_cnt)
1642				error = EBUSY;
1643			else {
1644				LIST_REMOVE(dq, dq_hash);
1645				dq->dq_ump = NULL;
1646			}
1647		}
1648	}
1649	DQH_UNLOCK();
1650	return (error);
1651}
1652
1653/*
1654 * The following three functions are provided for the adjustment of
1655 * quotas by the soft updates code.
1656 */
1657#ifdef SOFTUPDATES
1658/*
1659 * Acquire a reference to the quota structures associated with a vnode.
1660 * Return count of number of quota structures found.
1661 */
1662int
1663quotaref(vp, qrp)
1664	struct vnode *vp;
1665	struct dquot **qrp;
1666{
1667	struct inode *ip;
1668	struct dquot *dq;
1669	int i, found;
1670
1671	for (i = 0; i < MAXQUOTAS; i++)
1672		qrp[i] = NODQUOT;
1673	/*
1674	 * Disk quotas must be turned off for system files.  Currently
1675	 * snapshot and quota files.
1676	 */
1677	if ((vp->v_vflag & VV_SYSTEM) != 0)
1678		return (0);
1679	/*
1680	 * Iterate through and copy active quotas.
1681	 */
1682	found = 0;
1683	ip = VTOI(vp);
1684	mtx_lock(&dqhlock);
1685	for (i = 0; i < MAXQUOTAS; i++) {
1686		if ((dq = ip->i_dquot[i]) == NODQUOT)
1687			continue;
1688		DQREF(dq);
1689		qrp[i] = dq;
1690		found++;
1691	}
1692	mtx_unlock(&dqhlock);
1693	return (found);
1694}
1695
1696/*
1697 * Release a set of quota structures obtained from a vnode.
1698 */
1699void
1700quotarele(qrp)
1701	struct dquot **qrp;
1702{
1703	struct dquot *dq;
1704	int i;
1705
1706	for (i = 0; i < MAXQUOTAS; i++) {
1707		if ((dq = qrp[i]) == NODQUOT)
1708			continue;
1709		dqrele(NULL, dq);
1710	}
1711}
1712
1713/*
1714 * Adjust the number of blocks associated with a quota.
1715 * Positive numbers when adding blocks; negative numbers when freeing blocks.
1716 */
1717void
1718quotaadj(qrp, ump, blkcount)
1719	struct dquot **qrp;
1720	struct ufsmount *ump;
1721	int64_t blkcount;
1722{
1723	struct dquot *dq;
1724	ufs2_daddr_t ncurblocks;
1725	int i;
1726
1727	if (blkcount == 0)
1728		return;
1729	for (i = 0; i < MAXQUOTAS; i++) {
1730		if ((dq = qrp[i]) == NODQUOT)
1731			continue;
1732		DQI_LOCK(dq);
1733		DQI_WAIT(dq, PINOD+1, "adjqta");
1734		ncurblocks = dq->dq_curblocks + blkcount;
1735		if (ncurblocks >= 0)
1736			dq->dq_curblocks = ncurblocks;
1737		else
1738			dq->dq_curblocks = 0;
1739		if (blkcount < 0)
1740			dq->dq_flags &= ~DQ_BLKS;
1741		else if (dq->dq_curblocks + blkcount >= dq->dq_bsoftlimit &&
1742			 dq->dq_curblocks < dq->dq_bsoftlimit)
1743			dq->dq_btime = time_second + ump->um_btime[i];
1744		dq->dq_flags |= DQ_MOD;
1745		DQI_UNLOCK(dq);
1746	}
1747}
1748#endif /* SOFTUPDATES */
1749
1750/*
1751 * 32-bit / 64-bit conversion functions.
1752 *
1753 * 32-bit quota records are stored in native byte order.  Attention must
1754 * be paid to overflow issues.
1755 *
1756 * 64-bit quota records are stored in network byte order.
1757 */
1758
1759#define CLIP32(u64) (u64 > UINT32_MAX ? UINT32_MAX : (uint32_t)u64)
1760
1761/*
1762 * Convert 32-bit host-order structure to dquot.
1763 */
1764static void
1765dqb32_dq(const struct dqblk32 *dqb32, struct dquot *dq)
1766{
1767
1768	dq->dq_bhardlimit = dqb32->dqb_bhardlimit;
1769	dq->dq_bsoftlimit = dqb32->dqb_bsoftlimit;
1770	dq->dq_curblocks = dqb32->dqb_curblocks;
1771	dq->dq_ihardlimit = dqb32->dqb_ihardlimit;
1772	dq->dq_isoftlimit = dqb32->dqb_isoftlimit;
1773	dq->dq_curinodes = dqb32->dqb_curinodes;
1774	dq->dq_btime = dqb32->dqb_btime;
1775	dq->dq_itime = dqb32->dqb_itime;
1776}
1777
1778/*
1779 * Convert 64-bit network-order structure to dquot.
1780 */
1781static void
1782dqb64_dq(const struct dqblk64 *dqb64, struct dquot *dq)
1783{
1784
1785	dq->dq_bhardlimit = be64toh(dqb64->dqb_bhardlimit);
1786	dq->dq_bsoftlimit = be64toh(dqb64->dqb_bsoftlimit);
1787	dq->dq_curblocks = be64toh(dqb64->dqb_curblocks);
1788	dq->dq_ihardlimit = be64toh(dqb64->dqb_ihardlimit);
1789	dq->dq_isoftlimit = be64toh(dqb64->dqb_isoftlimit);
1790	dq->dq_curinodes = be64toh(dqb64->dqb_curinodes);
1791	dq->dq_btime = be64toh(dqb64->dqb_btime);
1792	dq->dq_itime = be64toh(dqb64->dqb_itime);
1793}
1794
1795/*
1796 * Convert dquot to 32-bit host-order structure.
1797 */
1798static void
1799dq_dqb32(const struct dquot *dq, struct dqblk32 *dqb32)
1800{
1801
1802	dqb32->dqb_bhardlimit = CLIP32(dq->dq_bhardlimit);
1803	dqb32->dqb_bsoftlimit = CLIP32(dq->dq_bsoftlimit);
1804	dqb32->dqb_curblocks = CLIP32(dq->dq_curblocks);
1805	dqb32->dqb_ihardlimit = CLIP32(dq->dq_ihardlimit);
1806	dqb32->dqb_isoftlimit = CLIP32(dq->dq_isoftlimit);
1807	dqb32->dqb_curinodes = CLIP32(dq->dq_curinodes);
1808	dqb32->dqb_btime = CLIP32(dq->dq_btime);
1809	dqb32->dqb_itime = CLIP32(dq->dq_itime);
1810}
1811
1812/*
1813 * Convert dquot to 64-bit network-order structure.
1814 */
1815static void
1816dq_dqb64(const struct dquot *dq, struct dqblk64 *dqb64)
1817{
1818
1819	dqb64->dqb_bhardlimit = htobe64(dq->dq_bhardlimit);
1820	dqb64->dqb_bsoftlimit = htobe64(dq->dq_bsoftlimit);
1821	dqb64->dqb_curblocks = htobe64(dq->dq_curblocks);
1822	dqb64->dqb_ihardlimit = htobe64(dq->dq_ihardlimit);
1823	dqb64->dqb_isoftlimit = htobe64(dq->dq_isoftlimit);
1824	dqb64->dqb_curinodes = htobe64(dq->dq_curinodes);
1825	dqb64->dqb_btime = htobe64(dq->dq_btime);
1826	dqb64->dqb_itime = htobe64(dq->dq_itime);
1827}
1828
1829/*
1830 * Convert 64-bit host-order structure to 32-bit host-order structure.
1831 */
1832static void
1833dqb64_dqb32(const struct dqblk64 *dqb64, struct dqblk32 *dqb32)
1834{
1835
1836	dqb32->dqb_bhardlimit = CLIP32(dqb64->dqb_bhardlimit);
1837	dqb32->dqb_bsoftlimit = CLIP32(dqb64->dqb_bsoftlimit);
1838	dqb32->dqb_curblocks = CLIP32(dqb64->dqb_curblocks);
1839	dqb32->dqb_ihardlimit = CLIP32(dqb64->dqb_ihardlimit);
1840	dqb32->dqb_isoftlimit = CLIP32(dqb64->dqb_isoftlimit);
1841	dqb32->dqb_curinodes = CLIP32(dqb64->dqb_curinodes);
1842	dqb32->dqb_btime = CLIP32(dqb64->dqb_btime);
1843	dqb32->dqb_itime = CLIP32(dqb64->dqb_itime);
1844}
1845
1846/*
1847 * Convert 32-bit host-order structure to 64-bit host-order structure.
1848 */
1849static void
1850dqb32_dqb64(const struct dqblk32 *dqb32, struct dqblk64 *dqb64)
1851{
1852
1853	dqb64->dqb_bhardlimit = dqb32->dqb_bhardlimit;
1854	dqb64->dqb_bsoftlimit = dqb32->dqb_bsoftlimit;
1855	dqb64->dqb_curblocks = dqb32->dqb_curblocks;
1856	dqb64->dqb_ihardlimit = dqb32->dqb_ihardlimit;
1857	dqb64->dqb_isoftlimit = dqb32->dqb_isoftlimit;
1858	dqb64->dqb_curinodes = dqb32->dqb_curinodes;
1859	dqb64->dqb_btime = dqb32->dqb_btime;
1860	dqb64->dqb_itime = dqb32->dqb_itime;
1861}
1862