xref: /freebsd/sys/ufs/ufs/ufs_quota.c (revision 9768746b)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1990, 1993, 1995
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * Robert Elz at The University of Melbourne.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)ufs_quota.c	8.5 (Berkeley) 5/20/95
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_ffs.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/endian.h>
45 #include <sys/fcntl.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mutex.h>
51 #include <sys/namei.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/stat.h>
56 #include <sys/sysctl.h>
57 #include <sys/vnode.h>
58 
59 #include <ufs/ufs/extattr.h>
60 #include <ufs/ufs/quota.h>
61 #include <ufs/ufs/inode.h>
62 #include <ufs/ufs/ufsmount.h>
63 #include <ufs/ufs/ufs_extern.h>
64 
65 CTASSERT(sizeof(struct dqblk64) == sizeof(struct dqhdr64));
66 
67 static int unprivileged_get_quota = 0;
68 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_get_quota, CTLFLAG_RW,
69     &unprivileged_get_quota, 0,
70     "Unprivileged processes may retrieve quotas for other uids and gids");
71 
72 static MALLOC_DEFINE(M_DQUOT, "ufs_quota", "UFS quota entries");
73 
74 /*
75  * Quota name to error message mapping.
76  */
77 static char *quotatypes[] = INITQFNAMES;
78 
79 static int chkdqchg(struct inode *, ufs2_daddr_t, struct ucred *, int, int *);
80 static int chkiqchg(struct inode *, int, struct ucred *, int, int *);
81 static int dqopen(struct vnode *, struct ufsmount *, int);
82 static int dqget(struct vnode *,
83 	u_long, struct ufsmount *, int, struct dquot **);
84 static int dqsync(struct vnode *, struct dquot *);
85 static int dqflush(struct vnode *);
86 static int quotaoff1(struct thread *td, struct mount *mp, int type);
87 static int quotaoff_inchange(struct thread *td, struct mount *mp, int type);
88 
89 /* conversion functions - from_to() */
90 static void dqb32_dq(const struct dqblk32 *, struct dquot *);
91 static void dqb64_dq(const struct dqblk64 *, struct dquot *);
92 static void dq_dqb32(const struct dquot *, struct dqblk32 *);
93 static void dq_dqb64(const struct dquot *, struct dqblk64 *);
94 static void dqb32_dqb64(const struct dqblk32 *, struct dqblk64 *);
95 static void dqb64_dqb32(const struct dqblk64 *, struct dqblk32 *);
96 
97 #ifdef DIAGNOSTIC
98 static void dqref(struct dquot *);
99 static void chkdquot(struct inode *);
100 #endif
101 
102 /*
103  * Set up the quotas for an inode.
104  *
105  * This routine completely defines the semantics of quotas.
106  * If other criterion want to be used to establish quotas, the
107  * MAXQUOTAS value in quota.h should be increased, and the
108  * additional dquots set up here.
109  */
110 int
111 getinoquota(struct inode *ip)
112 {
113 	struct ufsmount *ump;
114 	struct vnode *vp;
115 	int error;
116 
117 	vp = ITOV(ip);
118 
119 	/*
120 	 * Disk quotas must be turned off for system files.  Currently
121 	 * snapshot and quota files.
122 	 */
123 	if ((vp->v_vflag & VV_SYSTEM) != 0)
124 		return (0);
125 	/*
126 	 * XXX: Turn off quotas for files with a negative UID or GID.
127 	 * This prevents the creation of 100GB+ quota files.
128 	 */
129 	if ((int)ip->i_uid < 0 || (int)ip->i_gid < 0)
130 		return (0);
131 	ump = VFSTOUFS(vp->v_mount);
132 	/*
133 	 * Set up the user quota based on file uid.
134 	 * EINVAL means that quotas are not enabled.
135 	 */
136 	if ((error =
137 		dqget(vp, ip->i_uid, ump, USRQUOTA, &ip->i_dquot[USRQUOTA])) &&
138 	    error != EINVAL)
139 		return (error);
140 	/*
141 	 * Set up the group quota based on file gid.
142 	 * EINVAL means that quotas are not enabled.
143 	 */
144 	if ((error =
145 		dqget(vp, ip->i_gid, ump, GRPQUOTA, &ip->i_dquot[GRPQUOTA])) &&
146 	    error != EINVAL)
147 		return (error);
148 	return (0);
149 }
150 
151 /*
152  * Update disk usage, and take corrective action.
153  */
154 int
155 chkdq(struct inode *ip, ufs2_daddr_t change, struct ucred *cred, int flags)
156 {
157 	struct dquot *dq;
158 	ufs2_daddr_t ncurblocks;
159 	struct vnode *vp = ITOV(ip);
160 	int i, error, warn, do_check;
161 
162 	MPASS(cred != NOCRED || (flags & FORCE) != 0);
163 	/*
164 	 * Disk quotas must be turned off for system files.  Currently
165 	 * snapshot and quota files.
166 	 */
167 	if ((vp->v_vflag & VV_SYSTEM) != 0)
168 		return (0);
169 	/*
170 	 * XXX: Turn off quotas for files with a negative UID or GID.
171 	 * This prevents the creation of 100GB+ quota files.
172 	 */
173 	if ((int)ip->i_uid < 0 || (int)ip->i_gid < 0)
174 		return (0);
175 #ifdef DIAGNOSTIC
176 	if ((flags & CHOWN) == 0)
177 		chkdquot(ip);
178 #endif
179 	if (change == 0)
180 		return (0);
181 	if (change < 0) {
182 		for (i = 0; i < MAXQUOTAS; i++) {
183 			if ((dq = ip->i_dquot[i]) == NODQUOT)
184 				continue;
185 			DQI_LOCK(dq);
186 			DQI_WAIT(dq, PINOD+1, "chkdq1");
187 			ncurblocks = dq->dq_curblocks + change;
188 			if (ncurblocks >= 0)
189 				dq->dq_curblocks = ncurblocks;
190 			else
191 				dq->dq_curblocks = 0;
192 			dq->dq_flags &= ~DQ_BLKS;
193 			dq->dq_flags |= DQ_MOD;
194 			DQI_UNLOCK(dq);
195 		}
196 		return (0);
197 	}
198 	if ((flags & FORCE) == 0 &&
199 	    priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA))
200 		do_check = 1;
201 	else
202 		do_check = 0;
203 	for (i = 0; i < MAXQUOTAS; i++) {
204 		if ((dq = ip->i_dquot[i]) == NODQUOT)
205 			continue;
206 		warn = 0;
207 		DQI_LOCK(dq);
208 		DQI_WAIT(dq, PINOD+1, "chkdq2");
209 		if (do_check) {
210 			error = chkdqchg(ip, change, cred, i, &warn);
211 			if (error) {
212 				/*
213 				 * Roll back user quota changes when
214 				 * group quota failed.
215 				 */
216 				while (i > 0) {
217 					--i;
218 					dq = ip->i_dquot[i];
219 					if (dq == NODQUOT)
220 						continue;
221 					DQI_LOCK(dq);
222 					DQI_WAIT(dq, PINOD+1, "chkdq3");
223 					ncurblocks = dq->dq_curblocks - change;
224 					if (ncurblocks >= 0)
225 						dq->dq_curblocks = ncurblocks;
226 					else
227 						dq->dq_curblocks = 0;
228 					dq->dq_flags &= ~DQ_BLKS;
229 					dq->dq_flags |= DQ_MOD;
230 					DQI_UNLOCK(dq);
231 				}
232 				return (error);
233 			}
234 		}
235 		/* Reset timer when crossing soft limit */
236 		if (dq->dq_curblocks + change >= dq->dq_bsoftlimit &&
237 		    dq->dq_curblocks < dq->dq_bsoftlimit)
238 			dq->dq_btime = time_second + ITOUMP(ip)->um_btime[i];
239 		dq->dq_curblocks += change;
240 		dq->dq_flags |= DQ_MOD;
241 		DQI_UNLOCK(dq);
242 		if (warn)
243 			uprintf("\n%s: warning, %s disk quota exceeded\n",
244 			    ITOVFS(ip)->mnt_stat.f_mntonname,
245 			    quotatypes[i]);
246 	}
247 	return (0);
248 }
249 
250 /*
251  * Check for a valid change to a users allocation.
252  * Issue an error message if appropriate.
253  */
254 static int
255 chkdqchg(struct inode *ip, ufs2_daddr_t change, struct ucred *cred,
256     int type, int *warn)
257 {
258 	struct dquot *dq = ip->i_dquot[type];
259 	ufs2_daddr_t ncurblocks = dq->dq_curblocks + change;
260 
261 	/*
262 	 * If user would exceed their hard limit, disallow space allocation.
263 	 */
264 	if (ncurblocks >= dq->dq_bhardlimit && dq->dq_bhardlimit) {
265 		if ((dq->dq_flags & DQ_BLKS) == 0 &&
266 		    ip->i_uid == cred->cr_uid) {
267 			dq->dq_flags |= DQ_BLKS;
268 			DQI_UNLOCK(dq);
269 			uprintf("\n%s: write failed, %s disk limit reached\n",
270 			    ITOVFS(ip)->mnt_stat.f_mntonname,
271 			    quotatypes[type]);
272 			return (EDQUOT);
273 		}
274 		DQI_UNLOCK(dq);
275 		return (EDQUOT);
276 	}
277 	/*
278 	 * If user is over their soft limit for too long, disallow space
279 	 * allocation. Reset time limit as they cross their soft limit.
280 	 */
281 	if (ncurblocks >= dq->dq_bsoftlimit && dq->dq_bsoftlimit) {
282 		if (dq->dq_curblocks < dq->dq_bsoftlimit) {
283 			dq->dq_btime = time_second + ITOUMP(ip)->um_btime[type];
284 			if (ip->i_uid == cred->cr_uid)
285 				*warn = 1;
286 			return (0);
287 		}
288 		if (time_second > dq->dq_btime) {
289 			if ((dq->dq_flags & DQ_BLKS) == 0 &&
290 			    ip->i_uid == cred->cr_uid) {
291 				dq->dq_flags |= DQ_BLKS;
292 				DQI_UNLOCK(dq);
293 				uprintf("\n%s: write failed, %s "
294 				    "disk quota exceeded for too long\n",
295 				    ITOVFS(ip)->mnt_stat.f_mntonname,
296 				    quotatypes[type]);
297 				return (EDQUOT);
298 			}
299 			DQI_UNLOCK(dq);
300 			return (EDQUOT);
301 		}
302 	}
303 	return (0);
304 }
305 
306 /*
307  * Check the inode limit, applying corrective action.
308  */
309 int
310 chkiq(struct inode *ip, int change, struct ucred *cred, int flags)
311 {
312 	struct dquot *dq;
313 	int i, error, warn, do_check;
314 
315 	MPASS(cred != NOCRED || (flags & FORCE) != 0);
316 #ifdef DIAGNOSTIC
317 	if ((flags & CHOWN) == 0)
318 		chkdquot(ip);
319 #endif
320 	if (change == 0)
321 		return (0);
322 	if (change < 0) {
323 		for (i = 0; i < MAXQUOTAS; i++) {
324 			if ((dq = ip->i_dquot[i]) == NODQUOT)
325 				continue;
326 			DQI_LOCK(dq);
327 			DQI_WAIT(dq, PINOD+1, "chkiq1");
328 			if (dq->dq_curinodes >= -change)
329 				dq->dq_curinodes += change;
330 			else
331 				dq->dq_curinodes = 0;
332 			dq->dq_flags &= ~DQ_INODS;
333 			dq->dq_flags |= DQ_MOD;
334 			DQI_UNLOCK(dq);
335 		}
336 		return (0);
337 	}
338 	if ((flags & FORCE) == 0 &&
339 	    priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA))
340 		do_check = 1;
341 	else
342 		do_check = 0;
343 	for (i = 0; i < MAXQUOTAS; i++) {
344 		if ((dq = ip->i_dquot[i]) == NODQUOT)
345 			continue;
346 		warn = 0;
347 		DQI_LOCK(dq);
348 		DQI_WAIT(dq, PINOD+1, "chkiq2");
349 		if (do_check) {
350 			error = chkiqchg(ip, change, cred, i, &warn);
351 			if (error) {
352 				/*
353 				 * Roll back user quota changes when
354 				 * group quota failed.
355 				 */
356 				while (i > 0) {
357 					--i;
358 					dq = ip->i_dquot[i];
359 					if (dq == NODQUOT)
360 						continue;
361 					DQI_LOCK(dq);
362 					DQI_WAIT(dq, PINOD+1, "chkiq3");
363 					if (dq->dq_curinodes >= change)
364 						dq->dq_curinodes -= change;
365 					else
366 						dq->dq_curinodes = 0;
367 					dq->dq_flags &= ~DQ_INODS;
368 					dq->dq_flags |= DQ_MOD;
369 					DQI_UNLOCK(dq);
370 				}
371 				return (error);
372 			}
373 		}
374 		/* Reset timer when crossing soft limit */
375 		if (dq->dq_curinodes + change >= dq->dq_isoftlimit &&
376 		    dq->dq_curinodes < dq->dq_isoftlimit)
377 			dq->dq_itime = time_second + ITOUMP(ip)->um_itime[i];
378 		dq->dq_curinodes += change;
379 		dq->dq_flags |= DQ_MOD;
380 		DQI_UNLOCK(dq);
381 		if (warn)
382 			uprintf("\n%s: warning, %s inode quota exceeded\n",
383 			    ITOVFS(ip)->mnt_stat.f_mntonname,
384 			    quotatypes[i]);
385 	}
386 	return (0);
387 }
388 
389 /*
390  * Check for a valid change to a users allocation.
391  * Issue an error message if appropriate.
392  */
393 static int
394 chkiqchg(struct inode *ip, int change, struct ucred *cred, int type, int *warn)
395 {
396 	struct dquot *dq = ip->i_dquot[type];
397 	ino_t ncurinodes = dq->dq_curinodes + change;
398 
399 	/*
400 	 * If user would exceed their hard limit, disallow inode allocation.
401 	 */
402 	if (ncurinodes >= dq->dq_ihardlimit && dq->dq_ihardlimit) {
403 		if ((dq->dq_flags & DQ_INODS) == 0 &&
404 		    ip->i_uid == cred->cr_uid) {
405 			dq->dq_flags |= DQ_INODS;
406 			DQI_UNLOCK(dq);
407 			uprintf("\n%s: write failed, %s inode limit reached\n",
408 			    ITOVFS(ip)->mnt_stat.f_mntonname,
409 			    quotatypes[type]);
410 			return (EDQUOT);
411 		}
412 		DQI_UNLOCK(dq);
413 		return (EDQUOT);
414 	}
415 	/*
416 	 * If user is over their soft limit for too long, disallow inode
417 	 * allocation. Reset time limit as they cross their soft limit.
418 	 */
419 	if (ncurinodes >= dq->dq_isoftlimit && dq->dq_isoftlimit) {
420 		if (dq->dq_curinodes < dq->dq_isoftlimit) {
421 			dq->dq_itime = time_second + ITOUMP(ip)->um_itime[type];
422 			if (ip->i_uid == cred->cr_uid)
423 				*warn = 1;
424 			return (0);
425 		}
426 		if (time_second > dq->dq_itime) {
427 			if ((dq->dq_flags & DQ_INODS) == 0 &&
428 			    ip->i_uid == cred->cr_uid) {
429 				dq->dq_flags |= DQ_INODS;
430 				DQI_UNLOCK(dq);
431 				uprintf("\n%s: write failed, %s "
432 				    "inode quota exceeded for too long\n",
433 				    ITOVFS(ip)->mnt_stat.f_mntonname,
434 				    quotatypes[type]);
435 				return (EDQUOT);
436 			}
437 			DQI_UNLOCK(dq);
438 			return (EDQUOT);
439 		}
440 	}
441 	return (0);
442 }
443 
444 #ifdef DIAGNOSTIC
445 /*
446  * On filesystems with quotas enabled, it is an error for a file to change
447  * size and not to have a dquot structure associated with it.
448  */
449 static void
450 chkdquot(struct inode *ip)
451 {
452 	struct ufsmount *ump;
453 	struct vnode *vp;
454 	int i;
455 
456 	ump = ITOUMP(ip);
457 	vp = ITOV(ip);
458 
459 	/*
460 	 * Disk quotas must be turned off for system files.  Currently
461 	 * these are snapshots and quota files.
462 	 */
463 	if ((vp->v_vflag & VV_SYSTEM) != 0)
464 		return;
465 	/*
466 	 * XXX: Turn off quotas for files with a negative UID or GID.
467 	 * This prevents the creation of 100GB+ quota files.
468 	 */
469 	if ((int)ip->i_uid < 0 || (int)ip->i_gid < 0)
470 		return;
471 
472 	UFS_LOCK(ump);
473 	for (i = 0; i < MAXQUOTAS; i++) {
474 		if (ump->um_quotas[i] == NULLVP ||
475 		    (ump->um_qflags[i] & (QTF_OPENING|QTF_CLOSING)))
476 			continue;
477 		if (ip->i_dquot[i] == NODQUOT) {
478 			UFS_UNLOCK(ump);
479 			vn_printf(ITOV(ip), "chkdquot: missing dquot ");
480 			panic("chkdquot: missing dquot");
481 		}
482 	}
483 	UFS_UNLOCK(ump);
484 }
485 #endif
486 
487 /*
488  * Code to process quotactl commands.
489  */
490 
491 /*
492  * Q_QUOTAON - set up a quota file for a particular filesystem.
493  */
494 int
495 quotaon(struct thread *td, struct mount *mp, int type, void *fname,
496     bool *mp_busy)
497 {
498 	struct ufsmount *ump;
499 	struct vnode *vp, **vpp;
500 	struct vnode *mvp;
501 	struct dquot *dq;
502 	int error, flags;
503 	struct nameidata nd;
504 
505 	error = priv_check(td, PRIV_UFS_QUOTAON);
506 	if (error != 0)
507 		return (error);
508 
509 	if ((mp->mnt_flag & MNT_RDONLY) != 0)
510 		return (EROFS);
511 
512 	ump = VFSTOUFS(mp);
513 	dq = NODQUOT;
514 
515 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fname);
516 	flags = FREAD | FWRITE;
517 	vfs_ref(mp);
518 	KASSERT(*mp_busy, ("%s called without busied mount", __func__));
519 	vfs_unbusy(mp);
520 	*mp_busy = false;
521 	error = vn_open(&nd, &flags, 0, NULL);
522 	if (error != 0) {
523 		vfs_rel(mp);
524 		return (error);
525 	}
526 	NDFREE_PNBUF(&nd);
527 	vp = nd.ni_vp;
528 	error = vfs_busy(mp, MBF_NOWAIT);
529 	vfs_rel(mp);
530 	if (error == 0) {
531 		*mp_busy = true;
532 		if (vp->v_type != VREG)
533 			error = EACCES;
534 	}
535 	if (error != 0) {
536 		VOP_UNLOCK(vp);
537 		(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
538 		return (error);
539 	}
540 
541 	UFS_LOCK(ump);
542 	if ((ump->um_qflags[type] & (QTF_OPENING|QTF_CLOSING)) != 0) {
543 		UFS_UNLOCK(ump);
544 		VOP_UNLOCK(vp);
545 		(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
546 		return (EALREADY);
547 	}
548 	ump->um_qflags[type] |= QTF_OPENING|QTF_CLOSING;
549 	UFS_UNLOCK(ump);
550 	if ((error = dqopen(vp, ump, type)) != 0) {
551 		VOP_UNLOCK(vp);
552 		UFS_LOCK(ump);
553 		ump->um_qflags[type] &= ~(QTF_OPENING|QTF_CLOSING);
554 		UFS_UNLOCK(ump);
555 		(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
556 		return (error);
557 	}
558 	VOP_UNLOCK(vp);
559 	MNT_ILOCK(mp);
560 	mp->mnt_flag |= MNT_QUOTA;
561 	mp->mnt_stat.f_flags |= MNT_QUOTA;
562 	MNT_IUNLOCK(mp);
563 
564 	vpp = &ump->um_quotas[type];
565 	if (*vpp != vp)
566 		quotaoff1(td, mp, type);
567 
568 	/*
569 	 * When the directory vnode containing the quota file is
570 	 * inactivated, due to the shared lookup of the quota file
571 	 * vput()ing the dvp, the qsyncvp() call for the containing
572 	 * directory would try to acquire the quota lock exclusive.
573 	 * At the same time, lookup already locked the quota vnode
574 	 * shared.  Mark the quota vnode lock as allowing recursion
575 	 * and automatically converting shared locks to exclusive.
576 	 *
577 	 * Also mark quota vnode as system.
578 	 */
579 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
580 	vp->v_vflag |= VV_SYSTEM;
581 	VN_LOCK_AREC(vp);
582 	VN_LOCK_DSHARE(vp);
583 	VOP_UNLOCK(vp);
584 	*vpp = vp;
585 	/*
586 	 * Save the credential of the process that turned on quotas.
587 	 * Set up the time limits for this quota.
588 	 */
589 	ump->um_cred[type] = crhold(td->td_ucred);
590 	ump->um_btime[type] = MAX_DQ_TIME;
591 	ump->um_itime[type] = MAX_IQ_TIME;
592 	if (dqget(NULLVP, 0, ump, type, &dq) == 0) {
593 		if (dq->dq_btime > 0)
594 			ump->um_btime[type] = dq->dq_btime;
595 		if (dq->dq_itime > 0)
596 			ump->um_itime[type] = dq->dq_itime;
597 		dqrele(NULLVP, dq);
598 	}
599 	/*
600 	 * Allow the getdq from getinoquota below to read the quota
601 	 * from file.
602 	 */
603 	UFS_LOCK(ump);
604 	ump->um_qflags[type] &= ~QTF_CLOSING;
605 	UFS_UNLOCK(ump);
606 	/*
607 	 * Search vnodes associated with this mount point,
608 	 * adding references to quota file being opened.
609 	 * NB: only need to add dquot's for inodes being modified.
610 	 */
611 again:
612 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
613 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
614 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
615 			goto again;
616 		}
617 		if (vp->v_type == VNON || vp->v_writecount <= 0) {
618 			vput(vp);
619 			continue;
620 		}
621 		error = getinoquota(VTOI(vp));
622 		vput(vp);
623 		if (error) {
624 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
625 			break;
626 		}
627 	}
628 
629         if (error)
630 		quotaoff_inchange(td, mp, type);
631 	UFS_LOCK(ump);
632 	ump->um_qflags[type] &= ~QTF_OPENING;
633 	KASSERT((ump->um_qflags[type] & QTF_CLOSING) == 0,
634 		("quotaon: leaking flags"));
635 	UFS_UNLOCK(ump);
636 
637 	return (error);
638 }
639 
640 /*
641  * Main code to turn off disk quotas for a filesystem. Does not change
642  * flags.
643  */
644 static int
645 quotaoff1(struct thread *td, struct mount *mp, int type)
646 {
647 	struct vnode *vp;
648 	struct vnode *qvp, *mvp;
649 	struct ufsmount *ump;
650 	struct dquot *dq;
651 	struct inode *ip;
652 	struct ucred *cr;
653 	int error;
654 
655 	ump = VFSTOUFS(mp);
656 
657 	UFS_LOCK(ump);
658 	KASSERT((ump->um_qflags[type] & QTF_CLOSING) != 0,
659 		("quotaoff1: flags are invalid"));
660 	if ((qvp = ump->um_quotas[type]) == NULLVP) {
661 		UFS_UNLOCK(ump);
662 		return (0);
663 	}
664 	cr = ump->um_cred[type];
665 	UFS_UNLOCK(ump);
666 
667 	/*
668 	 * Search vnodes associated with this mount point,
669 	 * deleting any references to quota file being closed.
670 	 */
671 again:
672 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
673 		if (vp->v_type == VNON) {
674 			VI_UNLOCK(vp);
675 			continue;
676 		}
677 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK)) {
678 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
679 			goto again;
680 		}
681 		ip = VTOI(vp);
682 		dq = ip->i_dquot[type];
683 		ip->i_dquot[type] = NODQUOT;
684 		dqrele(vp, dq);
685 		vput(vp);
686 	}
687 
688 	error = dqflush(qvp);
689 	if (error != 0)
690 		return (error);
691 
692 	/*
693 	 * Clear um_quotas before closing the quota vnode to prevent
694 	 * access to the closed vnode from dqget/dqsync
695 	 */
696 	UFS_LOCK(ump);
697 	ump->um_quotas[type] = NULLVP;
698 	ump->um_cred[type] = NOCRED;
699 	UFS_UNLOCK(ump);
700 
701 	vn_lock(qvp, LK_EXCLUSIVE | LK_RETRY);
702 	qvp->v_vflag &= ~VV_SYSTEM;
703 	VOP_UNLOCK(qvp);
704 	error = vn_close(qvp, FREAD|FWRITE, td->td_ucred, td);
705 	crfree(cr);
706 
707 	return (error);
708 }
709 
710 static int
711 quotaoff_inchange1(struct thread *td, struct mount *mp, int type)
712 {
713 	int error;
714 	bool need_resume;
715 
716 	/*
717 	 * mp is already suspended on unmount.  If not, suspend it, to
718 	 * avoid the situation where quotaoff operation eventually
719 	 * failing due to SU structures still keeping references on
720 	 * dquots, but vnode's references are already clean.  This
721 	 * would cause quota accounting leak and asserts otherwise.
722 	 * Note that the thread has already called vn_start_write().
723 	 */
724 	if (mp->mnt_susp_owner == td) {
725 		need_resume = false;
726 	} else {
727 		error = vfs_write_suspend_umnt(mp);
728 		if (error != 0)
729 			return (error);
730 		need_resume = true;
731 	}
732 	error = quotaoff1(td, mp, type);
733 	if (need_resume)
734 		vfs_write_resume(mp, VR_START_WRITE);
735 	return (error);
736 }
737 
738 /*
739  * Turns off quotas, assumes that ump->um_qflags are already checked
740  * and QTF_CLOSING is set to indicate operation in progress. Fixes
741  * ump->um_qflags and mp->mnt_flag after.
742  */
743 int
744 quotaoff_inchange(struct thread *td, struct mount *mp, int type)
745 {
746 	struct ufsmount *ump;
747 	int error, i;
748 
749 	error = quotaoff_inchange1(td, mp, type);
750 
751 	ump = VFSTOUFS(mp);
752 	UFS_LOCK(ump);
753 	ump->um_qflags[type] &= ~QTF_CLOSING;
754 	for (i = 0; i < MAXQUOTAS; i++)
755 		if (ump->um_quotas[i] != NULLVP)
756 			break;
757 	if (i == MAXQUOTAS) {
758 		MNT_ILOCK(mp);
759 		mp->mnt_flag &= ~MNT_QUOTA;
760 		mp->mnt_stat.f_flags &= ~MNT_QUOTA;
761 		MNT_IUNLOCK(mp);
762 	}
763 	UFS_UNLOCK(ump);
764 	return (error);
765 }
766 
767 /*
768  * Q_QUOTAOFF - turn off disk quotas for a filesystem.
769  */
770 int
771 quotaoff(struct thread *td, struct mount *mp, int type)
772 {
773 	struct ufsmount *ump;
774 	int error;
775 
776 	error = priv_check(td, PRIV_UFS_QUOTAOFF);
777 	if (error)
778 		return (error);
779 
780 	ump = VFSTOUFS(mp);
781 	UFS_LOCK(ump);
782 	if ((ump->um_qflags[type] & (QTF_OPENING|QTF_CLOSING)) != 0) {
783 		UFS_UNLOCK(ump);
784 		return (EALREADY);
785 	}
786 	ump->um_qflags[type] |= QTF_CLOSING;
787 	UFS_UNLOCK(ump);
788 
789 	return (quotaoff_inchange(td, mp, type));
790 }
791 
792 /*
793  * Q_GETQUOTA - return current values in a dqblk structure.
794  */
795 static int
796 _getquota(struct thread *td, struct mount *mp, u_long id, int type,
797     struct dqblk64 *dqb)
798 {
799 	struct dquot *dq;
800 	int error;
801 
802 	switch (type) {
803 	case USRQUOTA:
804 		if ((td->td_ucred->cr_uid != id) && !unprivileged_get_quota) {
805 			error = priv_check(td, PRIV_VFS_GETQUOTA);
806 			if (error)
807 				return (error);
808 		}
809 		break;
810 
811 	case GRPQUOTA:
812 		if (!groupmember(id, td->td_ucred) &&
813 		    !unprivileged_get_quota) {
814 			error = priv_check(td, PRIV_VFS_GETQUOTA);
815 			if (error)
816 				return (error);
817 		}
818 		break;
819 
820 	default:
821 		return (EINVAL);
822 	}
823 
824 	dq = NODQUOT;
825 	error = dqget(NULLVP, id, VFSTOUFS(mp), type, &dq);
826 	if (error)
827 		return (error);
828 	*dqb = dq->dq_dqb;
829 	dqrele(NULLVP, dq);
830 	return (error);
831 }
832 
833 /*
834  * Q_SETQUOTA - assign an entire dqblk structure.
835  */
836 static int
837 _setquota(struct thread *td, struct mount *mp, u_long id, int type,
838     struct dqblk64 *dqb)
839 {
840 	struct dquot *dq;
841 	struct dquot *ndq;
842 	struct ufsmount *ump;
843 	struct dqblk64 newlim;
844 	int error;
845 
846 	error = priv_check(td, PRIV_VFS_SETQUOTA);
847 	if (error)
848 		return (error);
849 
850 	newlim = *dqb;
851 
852 	ndq = NODQUOT;
853 	ump = VFSTOUFS(mp);
854 
855 	error = dqget(NULLVP, id, ump, type, &ndq);
856 	if (error)
857 		return (error);
858 	dq = ndq;
859 	DQI_LOCK(dq);
860 	DQI_WAIT(dq, PINOD+1, "setqta");
861 	/*
862 	 * Copy all but the current values.
863 	 * Reset time limit if previously had no soft limit or were
864 	 * under it, but now have a soft limit and are over it.
865 	 */
866 	newlim.dqb_curblocks = dq->dq_curblocks;
867 	newlim.dqb_curinodes = dq->dq_curinodes;
868 	if (dq->dq_id != 0) {
869 		newlim.dqb_btime = dq->dq_btime;
870 		newlim.dqb_itime = dq->dq_itime;
871 	}
872 	if (newlim.dqb_bsoftlimit &&
873 	    dq->dq_curblocks >= newlim.dqb_bsoftlimit &&
874 	    (dq->dq_bsoftlimit == 0 || dq->dq_curblocks < dq->dq_bsoftlimit))
875 		newlim.dqb_btime = time_second + ump->um_btime[type];
876 	if (newlim.dqb_isoftlimit &&
877 	    dq->dq_curinodes >= newlim.dqb_isoftlimit &&
878 	    (dq->dq_isoftlimit == 0 || dq->dq_curinodes < dq->dq_isoftlimit))
879 		newlim.dqb_itime = time_second + ump->um_itime[type];
880 	dq->dq_dqb = newlim;
881 	if (dq->dq_curblocks < dq->dq_bsoftlimit)
882 		dq->dq_flags &= ~DQ_BLKS;
883 	if (dq->dq_curinodes < dq->dq_isoftlimit)
884 		dq->dq_flags &= ~DQ_INODS;
885 	if (dq->dq_isoftlimit == 0 && dq->dq_bsoftlimit == 0 &&
886 	    dq->dq_ihardlimit == 0 && dq->dq_bhardlimit == 0)
887 		dq->dq_flags |= DQ_FAKE;
888 	else
889 		dq->dq_flags &= ~DQ_FAKE;
890 	dq->dq_flags |= DQ_MOD;
891 	DQI_UNLOCK(dq);
892 	dqrele(NULLVP, dq);
893 	return (0);
894 }
895 
896 /*
897  * Q_SETUSE - set current inode and block usage.
898  */
899 static int
900 _setuse(struct thread *td, struct mount *mp, u_long id, int type,
901     struct dqblk64 *dqb)
902 {
903 	struct dquot *dq;
904 	struct ufsmount *ump;
905 	struct dquot *ndq;
906 	struct dqblk64 usage;
907 	int error;
908 
909 	error = priv_check(td, PRIV_UFS_SETUSE);
910 	if (error)
911 		return (error);
912 
913 	usage = *dqb;
914 
915 	ump = VFSTOUFS(mp);
916 	ndq = NODQUOT;
917 
918 	error = dqget(NULLVP, id, ump, type, &ndq);
919 	if (error)
920 		return (error);
921 	dq = ndq;
922 	DQI_LOCK(dq);
923 	DQI_WAIT(dq, PINOD+1, "setuse");
924 	/*
925 	 * Reset time limit if have a soft limit and were
926 	 * previously under it, but are now over it.
927 	 */
928 	if (dq->dq_bsoftlimit && dq->dq_curblocks < dq->dq_bsoftlimit &&
929 	    usage.dqb_curblocks >= dq->dq_bsoftlimit)
930 		dq->dq_btime = time_second + ump->um_btime[type];
931 	if (dq->dq_isoftlimit && dq->dq_curinodes < dq->dq_isoftlimit &&
932 	    usage.dqb_curinodes >= dq->dq_isoftlimit)
933 		dq->dq_itime = time_second + ump->um_itime[type];
934 	dq->dq_curblocks = usage.dqb_curblocks;
935 	dq->dq_curinodes = usage.dqb_curinodes;
936 	if (dq->dq_curblocks < dq->dq_bsoftlimit)
937 		dq->dq_flags &= ~DQ_BLKS;
938 	if (dq->dq_curinodes < dq->dq_isoftlimit)
939 		dq->dq_flags &= ~DQ_INODS;
940 	dq->dq_flags |= DQ_MOD;
941 	DQI_UNLOCK(dq);
942 	dqrele(NULLVP, dq);
943 	return (0);
944 }
945 
946 int
947 getquota32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
948 {
949 	struct dqblk32 dqb32;
950 	struct dqblk64 dqb64;
951 	int error;
952 
953 	error = _getquota(td, mp, id, type, &dqb64);
954 	if (error)
955 		return (error);
956 	dqb64_dqb32(&dqb64, &dqb32);
957 	error = copyout(&dqb32, addr, sizeof(dqb32));
958 	return (error);
959 }
960 
961 int
962 setquota32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
963 {
964 	struct dqblk32 dqb32;
965 	struct dqblk64 dqb64;
966 	int error;
967 
968 	error = copyin(addr, &dqb32, sizeof(dqb32));
969 	if (error)
970 		return (error);
971 	dqb32_dqb64(&dqb32, &dqb64);
972 	error = _setquota(td, mp, id, type, &dqb64);
973 	return (error);
974 }
975 
976 int
977 setuse32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
978 {
979 	struct dqblk32 dqb32;
980 	struct dqblk64 dqb64;
981 	int error;
982 
983 	error = copyin(addr, &dqb32, sizeof(dqb32));
984 	if (error)
985 		return (error);
986 	dqb32_dqb64(&dqb32, &dqb64);
987 	error = _setuse(td, mp, id, type, &dqb64);
988 	return (error);
989 }
990 
991 int
992 getquota(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
993 {
994 	struct dqblk64 dqb64;
995 	int error;
996 
997 	error = _getquota(td, mp, id, type, &dqb64);
998 	if (error)
999 		return (error);
1000 	error = copyout(&dqb64, addr, sizeof(dqb64));
1001 	return (error);
1002 }
1003 
1004 int
1005 setquota(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
1006 {
1007 	struct dqblk64 dqb64;
1008 	int error;
1009 
1010 	error = copyin(addr, &dqb64, sizeof(dqb64));
1011 	if (error)
1012 		return (error);
1013 	error = _setquota(td, mp, id, type, &dqb64);
1014 	return (error);
1015 }
1016 
1017 int
1018 setuse(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
1019 {
1020 	struct dqblk64 dqb64;
1021 	int error;
1022 
1023 	error = copyin(addr, &dqb64, sizeof(dqb64));
1024 	if (error)
1025 		return (error);
1026 	error = _setuse(td, mp, id, type, &dqb64);
1027 	return (error);
1028 }
1029 
1030 /*
1031  * Q_GETQUOTASIZE - get bit-size of quota file fields
1032  */
1033 int
1034 getquotasize(struct thread *td, struct mount *mp, u_long id, int type,
1035     void *sizep)
1036 {
1037 	struct ufsmount *ump = VFSTOUFS(mp);
1038 	int bitsize;
1039 
1040 	UFS_LOCK(ump);
1041 	if (ump->um_quotas[type] == NULLVP ||
1042 	    (ump->um_qflags[type] & QTF_CLOSING)) {
1043 		UFS_UNLOCK(ump);
1044 		return (EINVAL);
1045 	}
1046 	if ((ump->um_qflags[type] & QTF_64BIT) != 0)
1047 		bitsize = 64;
1048 	else
1049 		bitsize = 32;
1050 	UFS_UNLOCK(ump);
1051 	return (copyout(&bitsize, sizep, sizeof(int)));
1052 }
1053 
1054 /*
1055  * Q_SYNC - sync quota files to disk.
1056  */
1057 int
1058 qsync(struct mount *mp)
1059 {
1060 	struct ufsmount *ump = VFSTOUFS(mp);
1061 	struct vnode *vp, *mvp;
1062 	struct dquot *dq;
1063 	int i, error;
1064 
1065 	/*
1066 	 * Check if the mount point has any quotas.
1067 	 * If not, simply return.
1068 	 */
1069 	for (i = 0; i < MAXQUOTAS; i++)
1070 		if (ump->um_quotas[i] != NULLVP)
1071 			break;
1072 	if (i == MAXQUOTAS)
1073 		return (0);
1074 	/*
1075 	 * Search vnodes associated with this mount point,
1076 	 * synchronizing any modified dquot structures.
1077 	 */
1078 again:
1079 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
1080 		if (vp->v_type == VNON) {
1081 			VI_UNLOCK(vp);
1082 			continue;
1083 		}
1084 		error = vget(vp, LK_EXCLUSIVE | LK_INTERLOCK);
1085 		if (error) {
1086 			if (error == ENOENT) {
1087 				MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
1088 				goto again;
1089 			}
1090 			continue;
1091 		}
1092 		for (i = 0; i < MAXQUOTAS; i++) {
1093 			dq = VTOI(vp)->i_dquot[i];
1094 			if (dq != NODQUOT)
1095 				dqsync(vp, dq);
1096 		}
1097 		vput(vp);
1098 	}
1099 	return (0);
1100 }
1101 
1102 /*
1103  * Sync quota file for given vnode to disk.
1104  */
1105 int
1106 qsyncvp(struct vnode *vp)
1107 {
1108 	struct ufsmount *ump = VFSTOUFS(vp->v_mount);
1109 	struct dquot *dq;
1110 	int i;
1111 
1112 	/*
1113 	 * Check if the mount point has any quotas.
1114 	 * If not, simply return.
1115 	 */
1116 	for (i = 0; i < MAXQUOTAS; i++)
1117 		if (ump->um_quotas[i] != NULLVP)
1118 			break;
1119 	if (i == MAXQUOTAS)
1120 		return (0);
1121 	/*
1122 	 * Search quotas associated with this vnode
1123 	 * synchronizing any modified dquot structures.
1124 	 */
1125 	for (i = 0; i < MAXQUOTAS; i++) {
1126 		dq = VTOI(vp)->i_dquot[i];
1127 		if (dq != NODQUOT)
1128 			dqsync(vp, dq);
1129 	}
1130 	return (0);
1131 }
1132 
1133 /*
1134  * Code pertaining to management of the in-core dquot data structures.
1135  */
1136 #define DQHASH(dqvp, id) \
1137 	(&dqhashtbl[((((intptr_t)(dqvp)) >> 8) + id) & dqhash])
1138 static LIST_HEAD(dqhash, dquot) *dqhashtbl;
1139 static u_long dqhash;
1140 
1141 /*
1142  * Dquot free list.
1143  */
1144 #define	DQUOTINC	5	/* minimum free dquots desired */
1145 static TAILQ_HEAD(dqfreelist, dquot) dqfreelist;
1146 static long numdquot, desireddquot = DQUOTINC;
1147 
1148 /*
1149  * Lock to protect quota hash, dq free list and dq_cnt ref counters of
1150  * _all_ dqs.
1151  */
1152 struct mtx dqhlock;
1153 
1154 #define	DQH_LOCK()	mtx_lock(&dqhlock)
1155 #define	DQH_UNLOCK()	mtx_unlock(&dqhlock)
1156 
1157 static struct dquot *dqhashfind(struct dqhash *dqh, u_long id,
1158 	struct vnode *dqvp);
1159 
1160 /*
1161  * Initialize the quota system.
1162  */
1163 void
1164 dqinit(void)
1165 {
1166 
1167 	mtx_init(&dqhlock, "dqhlock", NULL, MTX_DEF);
1168 	dqhashtbl = hashinit(desiredvnodes, M_DQUOT, &dqhash);
1169 	TAILQ_INIT(&dqfreelist);
1170 }
1171 
1172 /*
1173  * Shut down the quota system.
1174  */
1175 void
1176 dquninit(void)
1177 {
1178 	struct dquot *dq;
1179 
1180 	hashdestroy(dqhashtbl, M_DQUOT, dqhash);
1181 	while ((dq = TAILQ_FIRST(&dqfreelist)) != NULL) {
1182 		TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
1183 		mtx_destroy(&dq->dq_lock);
1184 		free(dq, M_DQUOT);
1185 	}
1186 	mtx_destroy(&dqhlock);
1187 }
1188 
1189 static struct dquot *
1190 dqhashfind(struct dqhash *dqh, u_long id, struct vnode *dqvp)
1191 {
1192 	struct dquot *dq;
1193 
1194 	mtx_assert(&dqhlock, MA_OWNED);
1195 	LIST_FOREACH(dq, dqh, dq_hash) {
1196 		if (dq->dq_id != id ||
1197 		    dq->dq_ump->um_quotas[dq->dq_type] != dqvp)
1198 			continue;
1199 		/*
1200 		 * Cache hit with no references.  Take
1201 		 * the structure off the free list.
1202 		 */
1203 		if (dq->dq_cnt == 0)
1204 			TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
1205 		DQREF(dq);
1206 		return (dq);
1207 	}
1208 	return (NODQUOT);
1209 }
1210 
1211 /*
1212  * Determine the quota file type.
1213  *
1214  * A 32-bit quota file is simply an array of struct dqblk32.
1215  *
1216  * A 64-bit quota file is a struct dqhdr64 followed by an array of struct
1217  * dqblk64.  The header contains various magic bits which allow us to be
1218  * reasonably confident that it is indeeda 64-bit quota file and not just
1219  * a 32-bit quota file that just happens to "look right".
1220  *
1221  */
1222 static int
1223 dqopen(struct vnode *vp, struct ufsmount *ump, int type)
1224 {
1225 	struct dqhdr64 dqh;
1226 	struct iovec aiov;
1227 	struct uio auio;
1228 	int error;
1229 
1230 	ASSERT_VOP_LOCKED(vp, "dqopen");
1231 	auio.uio_iov = &aiov;
1232 	auio.uio_iovcnt = 1;
1233 	aiov.iov_base = &dqh;
1234 	aiov.iov_len = sizeof(dqh);
1235 	auio.uio_resid = sizeof(dqh);
1236 	auio.uio_offset = 0;
1237 	auio.uio_segflg = UIO_SYSSPACE;
1238 	auio.uio_rw = UIO_READ;
1239 	auio.uio_td = (struct thread *)0;
1240 	error = VOP_READ(vp, &auio, 0, ump->um_cred[type]);
1241 
1242 	if (error != 0)
1243 		return (error);
1244 	if (auio.uio_resid > 0) {
1245 		/* assume 32 bits */
1246 		return (0);
1247 	}
1248 
1249 	UFS_LOCK(ump);
1250 	if (strcmp(dqh.dqh_magic, Q_DQHDR64_MAGIC) == 0 &&
1251 	    be32toh(dqh.dqh_version) == Q_DQHDR64_VERSION &&
1252 	    be32toh(dqh.dqh_hdrlen) == (uint32_t)sizeof(struct dqhdr64) &&
1253 	    be32toh(dqh.dqh_reclen) == (uint32_t)sizeof(struct dqblk64)) {
1254 		/* XXX: what if the magic matches, but the sizes are wrong? */
1255 		ump->um_qflags[type] |= QTF_64BIT;
1256 	} else {
1257 		ump->um_qflags[type] &= ~QTF_64BIT;
1258 	}
1259 	UFS_UNLOCK(ump);
1260 
1261 	return (0);
1262 }
1263 
1264 /*
1265  * Obtain a dquot structure for the specified identifier and quota file
1266  * reading the information from the file if necessary.
1267  */
1268 static int
1269 dqget(struct vnode *vp, u_long id, struct ufsmount *ump, int type,
1270     struct dquot **dqp)
1271 {
1272 	uint8_t buf[sizeof(struct dqblk64)];
1273 	off_t base, recsize;
1274 	struct dquot *dq, *dq1;
1275 	struct dqhash *dqh;
1276 	struct vnode *dqvp;
1277 	struct iovec aiov;
1278 	struct uio auio;
1279 	int dqvplocked, error;
1280 
1281 #ifdef DEBUG_VFS_LOCKS
1282 	if (vp != NULLVP)
1283 		ASSERT_VOP_ELOCKED(vp, "dqget");
1284 #endif
1285 
1286 	if (vp != NULLVP && *dqp != NODQUOT) {
1287 		return (0);
1288 	}
1289 
1290 	/* XXX: Disallow negative id values to prevent the
1291 	* creation of 100GB+ quota data files.
1292 	*/
1293 	if ((int)id < 0)
1294 		return (EINVAL);
1295 
1296 	UFS_LOCK(ump);
1297 	dqvp = ump->um_quotas[type];
1298 	if (dqvp == NULLVP || (ump->um_qflags[type] & QTF_CLOSING)) {
1299 		*dqp = NODQUOT;
1300 		UFS_UNLOCK(ump);
1301 		return (EINVAL);
1302 	}
1303 	vref(dqvp);
1304 	UFS_UNLOCK(ump);
1305 	error = 0;
1306 	dqvplocked = 0;
1307 
1308 	/*
1309 	 * Check the cache first.
1310 	 */
1311 	dqh = DQHASH(dqvp, id);
1312 	DQH_LOCK();
1313 	dq = dqhashfind(dqh, id, dqvp);
1314 	if (dq != NULL) {
1315 		DQH_UNLOCK();
1316 hfound:		DQI_LOCK(dq);
1317 		DQI_WAIT(dq, PINOD+1, "dqget");
1318 		DQI_UNLOCK(dq);
1319 		if (dq->dq_ump == NULL) {
1320 			dqrele(vp, dq);
1321 			dq = NODQUOT;
1322 			error = EIO;
1323 		}
1324 		*dqp = dq;
1325 		if (dqvplocked)
1326 			vput(dqvp);
1327 		else
1328 			vrele(dqvp);
1329 		return (error);
1330 	}
1331 
1332 	/*
1333 	 * Quota vnode lock is before DQ_LOCK. Acquire dqvp lock there
1334 	 * since new dq will appear on the hash chain DQ_LOCKed.
1335 	 */
1336 	if (vp != dqvp) {
1337 		DQH_UNLOCK();
1338 		vn_lock(dqvp, LK_SHARED | LK_RETRY);
1339 		dqvplocked = 1;
1340 		DQH_LOCK();
1341 		/*
1342 		 * Recheck the cache after sleep for quota vnode lock.
1343 		 */
1344 		dq = dqhashfind(dqh, id, dqvp);
1345 		if (dq != NULL) {
1346 			DQH_UNLOCK();
1347 			goto hfound;
1348 		}
1349 	}
1350 
1351 	/*
1352 	 * Not in cache, allocate a new one or take it from the
1353 	 * free list.
1354 	 */
1355 	if (TAILQ_FIRST(&dqfreelist) == NODQUOT &&
1356 	    numdquot < MAXQUOTAS * desiredvnodes)
1357 		desireddquot += DQUOTINC;
1358 	if (numdquot < desireddquot) {
1359 		numdquot++;
1360 		DQH_UNLOCK();
1361 		dq1 = malloc(sizeof *dq1, M_DQUOT, M_WAITOK | M_ZERO);
1362 		mtx_init(&dq1->dq_lock, "dqlock", NULL, MTX_DEF);
1363 		DQH_LOCK();
1364 		/*
1365 		 * Recheck the cache after sleep for memory.
1366 		 */
1367 		dq = dqhashfind(dqh, id, dqvp);
1368 		if (dq != NULL) {
1369 			numdquot--;
1370 			DQH_UNLOCK();
1371 			mtx_destroy(&dq1->dq_lock);
1372 			free(dq1, M_DQUOT);
1373 			goto hfound;
1374 		}
1375 		dq = dq1;
1376 	} else {
1377 		if ((dq = TAILQ_FIRST(&dqfreelist)) == NULL) {
1378 			DQH_UNLOCK();
1379 			tablefull("dquot");
1380 			*dqp = NODQUOT;
1381 			if (dqvplocked)
1382 				vput(dqvp);
1383 			else
1384 				vrele(dqvp);
1385 			return (EUSERS);
1386 		}
1387 		if (dq->dq_cnt || (dq->dq_flags & DQ_MOD))
1388 			panic("dqget: free dquot isn't %p", dq);
1389 		TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
1390 		if (dq->dq_ump != NULL)
1391 			LIST_REMOVE(dq, dq_hash);
1392 	}
1393 
1394 	/*
1395 	 * Dq is put into hash already locked to prevent parallel
1396 	 * usage while it is being read from file.
1397 	 */
1398 	dq->dq_flags = DQ_LOCK;
1399 	dq->dq_id = id;
1400 	dq->dq_type = type;
1401 	dq->dq_ump = ump;
1402 	LIST_INSERT_HEAD(dqh, dq, dq_hash);
1403 	DQREF(dq);
1404 	DQH_UNLOCK();
1405 
1406 	/*
1407 	 * Read the requested quota record from the quota file, performing
1408 	 * any necessary conversions.
1409 	 */
1410 	if (ump->um_qflags[type] & QTF_64BIT) {
1411 		recsize = sizeof(struct dqblk64);
1412 		base = sizeof(struct dqhdr64);
1413 	} else {
1414 		recsize = sizeof(struct dqblk32);
1415 		base = 0;
1416 	}
1417 	auio.uio_iov = &aiov;
1418 	auio.uio_iovcnt = 1;
1419 	aiov.iov_base = buf;
1420 	aiov.iov_len = recsize;
1421 	auio.uio_resid = recsize;
1422 	auio.uio_offset = base + id * recsize;
1423 	auio.uio_segflg = UIO_SYSSPACE;
1424 	auio.uio_rw = UIO_READ;
1425 	auio.uio_td = (struct thread *)0;
1426 
1427 	error = VOP_READ(dqvp, &auio, 0, ump->um_cred[type]);
1428 	if (auio.uio_resid == recsize && error == 0) {
1429 		bzero(&dq->dq_dqb, sizeof(dq->dq_dqb));
1430 	} else {
1431 		if (ump->um_qflags[type] & QTF_64BIT)
1432 			dqb64_dq((struct dqblk64 *)buf, dq);
1433 		else
1434 			dqb32_dq((struct dqblk32 *)buf, dq);
1435 	}
1436 	if (dqvplocked)
1437 		vput(dqvp);
1438 	else
1439 		vrele(dqvp);
1440 	/*
1441 	 * I/O error in reading quota file, release
1442 	 * quota structure and reflect problem to caller.
1443 	 */
1444 	if (error) {
1445 		DQH_LOCK();
1446 		dq->dq_ump = NULL;
1447 		LIST_REMOVE(dq, dq_hash);
1448 		DQH_UNLOCK();
1449 		DQI_LOCK(dq);
1450 		if (dq->dq_flags & DQ_WANT)
1451 			wakeup(dq);
1452 		dq->dq_flags = 0;
1453 		DQI_UNLOCK(dq);
1454 		dqrele(vp, dq);
1455 		*dqp = NODQUOT;
1456 		return (error);
1457 	}
1458 	DQI_LOCK(dq);
1459 	/*
1460 	 * Check for no limit to enforce.
1461 	 * Initialize time values if necessary.
1462 	 */
1463 	if (dq->dq_isoftlimit == 0 && dq->dq_bsoftlimit == 0 &&
1464 	    dq->dq_ihardlimit == 0 && dq->dq_bhardlimit == 0)
1465 		dq->dq_flags |= DQ_FAKE;
1466 	if (dq->dq_id != 0) {
1467 		if (dq->dq_btime == 0) {
1468 			dq->dq_btime = time_second + ump->um_btime[type];
1469 			if (dq->dq_bsoftlimit &&
1470 			    dq->dq_curblocks >= dq->dq_bsoftlimit)
1471 				dq->dq_flags |= DQ_MOD;
1472 		}
1473 		if (dq->dq_itime == 0) {
1474 			dq->dq_itime = time_second + ump->um_itime[type];
1475 			if (dq->dq_isoftlimit &&
1476 			    dq->dq_curinodes >= dq->dq_isoftlimit)
1477 				dq->dq_flags |= DQ_MOD;
1478 		}
1479 	}
1480 	DQI_WAKEUP(dq);
1481 	DQI_UNLOCK(dq);
1482 	*dqp = dq;
1483 	return (0);
1484 }
1485 
1486 #ifdef DIAGNOSTIC
1487 /*
1488  * Obtain a reference to a dquot.
1489  */
1490 static void
1491 dqref(struct dquot *dq)
1492 {
1493 
1494 	dq->dq_cnt++;
1495 }
1496 #endif
1497 
1498 /*
1499  * Release a reference to a dquot.
1500  */
1501 void
1502 dqrele(struct vnode *vp, struct dquot *dq)
1503 {
1504 
1505 	if (dq == NODQUOT)
1506 		return;
1507 	DQH_LOCK();
1508 	KASSERT(dq->dq_cnt > 0, ("Lost dq %p reference 1", dq));
1509 	if (dq->dq_cnt > 1) {
1510 		dq->dq_cnt--;
1511 		DQH_UNLOCK();
1512 		return;
1513 	}
1514 	DQH_UNLOCK();
1515 sync:
1516 	(void) dqsync(vp, dq);
1517 
1518 	DQH_LOCK();
1519 	KASSERT(dq->dq_cnt > 0, ("Lost dq %p reference 2", dq));
1520 	if (--dq->dq_cnt > 0)
1521 	{
1522 		DQH_UNLOCK();
1523 		return;
1524 	}
1525 
1526 	/*
1527 	 * The dq may become dirty after it is synced but before it is
1528 	 * put to the free list. Checking the DQ_MOD there without
1529 	 * locking dq should be safe since no other references to the
1530 	 * dq exist.
1531 	 */
1532 	if ((dq->dq_flags & DQ_MOD) != 0) {
1533 		dq->dq_cnt++;
1534 		DQH_UNLOCK();
1535 		goto sync;
1536 	}
1537 	TAILQ_INSERT_TAIL(&dqfreelist, dq, dq_freelist);
1538 	DQH_UNLOCK();
1539 }
1540 
1541 /*
1542  * Update the disk quota in the quota file.
1543  */
1544 static int
1545 dqsync(struct vnode *vp, struct dquot *dq)
1546 {
1547 	uint8_t buf[sizeof(struct dqblk64)];
1548 	off_t base, recsize;
1549 	struct vnode *dqvp;
1550 	struct iovec aiov;
1551 	struct uio auio;
1552 	int error;
1553 	struct mount *mp;
1554 	struct ufsmount *ump;
1555 
1556 #ifdef DEBUG_VFS_LOCKS
1557 	if (vp != NULL)
1558 		ASSERT_VOP_ELOCKED(vp, "dqsync");
1559 #endif
1560 
1561 	mp = NULL;
1562 	error = 0;
1563 	if (dq == NODQUOT)
1564 		panic("dqsync: dquot");
1565 	if ((ump = dq->dq_ump) == NULL)
1566 		return (0);
1567 	UFS_LOCK(ump);
1568 	if ((dqvp = ump->um_quotas[dq->dq_type]) == NULLVP) {
1569 		if (vp == NULL) {
1570 			UFS_UNLOCK(ump);
1571 			return (0);
1572 		} else
1573 			panic("dqsync: file");
1574 	}
1575 	vref(dqvp);
1576 	UFS_UNLOCK(ump);
1577 
1578 	DQI_LOCK(dq);
1579 	if ((dq->dq_flags & DQ_MOD) == 0) {
1580 		DQI_UNLOCK(dq);
1581 		vrele(dqvp);
1582 		return (0);
1583 	}
1584 	DQI_UNLOCK(dq);
1585 
1586 	(void) vn_start_secondary_write(dqvp, &mp, V_WAIT);
1587 	if (vp != dqvp)
1588 		vn_lock(dqvp, LK_EXCLUSIVE | LK_RETRY);
1589 
1590 	DQI_LOCK(dq);
1591 	DQI_WAIT(dq, PINOD+2, "dqsync");
1592 	if ((dq->dq_flags & DQ_MOD) == 0)
1593 		goto out;
1594 	dq->dq_flags |= DQ_LOCK;
1595 	DQI_UNLOCK(dq);
1596 
1597 	/*
1598 	 * Write the quota record to the quota file, performing any
1599 	 * necessary conversions.  See dqget() for additional details.
1600 	 */
1601 	if (ump->um_qflags[dq->dq_type] & QTF_64BIT) {
1602 		dq_dqb64(dq, (struct dqblk64 *)buf);
1603 		recsize = sizeof(struct dqblk64);
1604 		base = sizeof(struct dqhdr64);
1605 	} else {
1606 		dq_dqb32(dq, (struct dqblk32 *)buf);
1607 		recsize = sizeof(struct dqblk32);
1608 		base = 0;
1609 	}
1610 
1611 	auio.uio_iov = &aiov;
1612 	auio.uio_iovcnt = 1;
1613 	aiov.iov_base = buf;
1614 	aiov.iov_len = recsize;
1615 	auio.uio_resid = recsize;
1616 	auio.uio_offset = base + dq->dq_id * recsize;
1617 	auio.uio_segflg = UIO_SYSSPACE;
1618 	auio.uio_rw = UIO_WRITE;
1619 	auio.uio_td = (struct thread *)0;
1620 	error = VOP_WRITE(dqvp, &auio, 0, dq->dq_ump->um_cred[dq->dq_type]);
1621 	if (auio.uio_resid && error == 0)
1622 		error = EIO;
1623 
1624 	DQI_LOCK(dq);
1625 	DQI_WAKEUP(dq);
1626 	dq->dq_flags &= ~DQ_MOD;
1627 out:
1628 	DQI_UNLOCK(dq);
1629 	if (vp != dqvp)
1630 		vput(dqvp);
1631 	else
1632 		vrele(dqvp);
1633 	vn_finished_secondary_write(mp);
1634 	return (error);
1635 }
1636 
1637 /*
1638  * Flush all entries from the cache for a particular vnode.
1639  */
1640 static int
1641 dqflush(struct vnode *vp)
1642 {
1643 	struct dquot *dq, *nextdq;
1644 	struct dqhash *dqh;
1645 	int error;
1646 
1647 	/*
1648 	 * Move all dquot's that used to refer to this quota
1649 	 * file off their hash chains (they will eventually
1650 	 * fall off the head of the free list and be re-used).
1651 	 */
1652 	error = 0;
1653 	DQH_LOCK();
1654 	for (dqh = &dqhashtbl[dqhash]; dqh >= dqhashtbl; dqh--) {
1655 		for (dq = LIST_FIRST(dqh); dq; dq = nextdq) {
1656 			nextdq = LIST_NEXT(dq, dq_hash);
1657 			if (dq->dq_ump->um_quotas[dq->dq_type] != vp)
1658 				continue;
1659 			if (dq->dq_cnt)
1660 				error = EBUSY;
1661 			else {
1662 				LIST_REMOVE(dq, dq_hash);
1663 				dq->dq_ump = NULL;
1664 			}
1665 		}
1666 	}
1667 	DQH_UNLOCK();
1668 	return (error);
1669 }
1670 
1671 /*
1672  * The following three functions are provided for the adjustment of
1673  * quotas by the soft updates code.
1674  */
1675 #ifdef SOFTUPDATES
1676 /*
1677  * Acquire a reference to the quota structures associated with a vnode.
1678  * Return count of number of quota structures found.
1679  */
1680 int
1681 quotaref(struct vnode *vp, struct dquot **qrp)
1682 {
1683 	struct inode *ip;
1684 	struct dquot *dq;
1685 	int i, found;
1686 
1687 	for (i = 0; i < MAXQUOTAS; i++)
1688 		qrp[i] = NODQUOT;
1689 	/*
1690 	 * Disk quotas must be turned off for system files.  Currently
1691 	 * snapshot and quota files.
1692 	 */
1693 	if ((vp->v_vflag & VV_SYSTEM) != 0)
1694 		return (0);
1695 	/*
1696 	 * Iterate through and copy active quotas.
1697 	 */
1698 	found = 0;
1699 	ip = VTOI(vp);
1700 	mtx_lock(&dqhlock);
1701 	for (i = 0; i < MAXQUOTAS; i++) {
1702 		if ((dq = ip->i_dquot[i]) == NODQUOT)
1703 			continue;
1704 		DQREF(dq);
1705 		qrp[i] = dq;
1706 		found++;
1707 	}
1708 	mtx_unlock(&dqhlock);
1709 	return (found);
1710 }
1711 
1712 /*
1713  * Release a set of quota structures obtained from a vnode.
1714  */
1715 void
1716 quotarele(struct dquot **qrp)
1717 {
1718 	struct dquot *dq;
1719 	int i;
1720 
1721 	for (i = 0; i < MAXQUOTAS; i++) {
1722 		if ((dq = qrp[i]) == NODQUOT)
1723 			continue;
1724 		dqrele(NULL, dq);
1725 	}
1726 }
1727 
1728 /*
1729  * Adjust the number of blocks associated with a quota.
1730  * Positive numbers when adding blocks; negative numbers when freeing blocks.
1731  */
1732 void
1733 quotaadj(struct dquot **qrp, struct ufsmount *ump, int64_t blkcount)
1734 {
1735 	struct dquot *dq;
1736 	ufs2_daddr_t ncurblocks;
1737 	int i;
1738 
1739 	if (blkcount == 0)
1740 		return;
1741 	for (i = 0; i < MAXQUOTAS; i++) {
1742 		if ((dq = qrp[i]) == NODQUOT)
1743 			continue;
1744 		DQI_LOCK(dq);
1745 		DQI_WAIT(dq, PINOD+1, "adjqta");
1746 		ncurblocks = dq->dq_curblocks + blkcount;
1747 		if (ncurblocks >= 0)
1748 			dq->dq_curblocks = ncurblocks;
1749 		else
1750 			dq->dq_curblocks = 0;
1751 		if (blkcount < 0)
1752 			dq->dq_flags &= ~DQ_BLKS;
1753 		else if (dq->dq_curblocks + blkcount >= dq->dq_bsoftlimit &&
1754 			 dq->dq_curblocks < dq->dq_bsoftlimit)
1755 			dq->dq_btime = time_second + ump->um_btime[i];
1756 		dq->dq_flags |= DQ_MOD;
1757 		DQI_UNLOCK(dq);
1758 	}
1759 }
1760 #endif /* SOFTUPDATES */
1761 
1762 /*
1763  * 32-bit / 64-bit conversion functions.
1764  *
1765  * 32-bit quota records are stored in native byte order.  Attention must
1766  * be paid to overflow issues.
1767  *
1768  * 64-bit quota records are stored in network byte order.
1769  */
1770 
1771 #define CLIP32(u64) (u64 > UINT32_MAX ? UINT32_MAX : (uint32_t)u64)
1772 
1773 /*
1774  * Convert 32-bit host-order structure to dquot.
1775  */
1776 static void
1777 dqb32_dq(const struct dqblk32 *dqb32, struct dquot *dq)
1778 {
1779 
1780 	dq->dq_bhardlimit = dqb32->dqb_bhardlimit;
1781 	dq->dq_bsoftlimit = dqb32->dqb_bsoftlimit;
1782 	dq->dq_curblocks = dqb32->dqb_curblocks;
1783 	dq->dq_ihardlimit = dqb32->dqb_ihardlimit;
1784 	dq->dq_isoftlimit = dqb32->dqb_isoftlimit;
1785 	dq->dq_curinodes = dqb32->dqb_curinodes;
1786 	dq->dq_btime = dqb32->dqb_btime;
1787 	dq->dq_itime = dqb32->dqb_itime;
1788 }
1789 
1790 /*
1791  * Convert 64-bit network-order structure to dquot.
1792  */
1793 static void
1794 dqb64_dq(const struct dqblk64 *dqb64, struct dquot *dq)
1795 {
1796 
1797 	dq->dq_bhardlimit = be64toh(dqb64->dqb_bhardlimit);
1798 	dq->dq_bsoftlimit = be64toh(dqb64->dqb_bsoftlimit);
1799 	dq->dq_curblocks = be64toh(dqb64->dqb_curblocks);
1800 	dq->dq_ihardlimit = be64toh(dqb64->dqb_ihardlimit);
1801 	dq->dq_isoftlimit = be64toh(dqb64->dqb_isoftlimit);
1802 	dq->dq_curinodes = be64toh(dqb64->dqb_curinodes);
1803 	dq->dq_btime = be64toh(dqb64->dqb_btime);
1804 	dq->dq_itime = be64toh(dqb64->dqb_itime);
1805 }
1806 
1807 /*
1808  * Convert dquot to 32-bit host-order structure.
1809  */
1810 static void
1811 dq_dqb32(const struct dquot *dq, struct dqblk32 *dqb32)
1812 {
1813 
1814 	dqb32->dqb_bhardlimit = CLIP32(dq->dq_bhardlimit);
1815 	dqb32->dqb_bsoftlimit = CLIP32(dq->dq_bsoftlimit);
1816 	dqb32->dqb_curblocks = CLIP32(dq->dq_curblocks);
1817 	dqb32->dqb_ihardlimit = CLIP32(dq->dq_ihardlimit);
1818 	dqb32->dqb_isoftlimit = CLIP32(dq->dq_isoftlimit);
1819 	dqb32->dqb_curinodes = CLIP32(dq->dq_curinodes);
1820 	dqb32->dqb_btime = CLIP32(dq->dq_btime);
1821 	dqb32->dqb_itime = CLIP32(dq->dq_itime);
1822 }
1823 
1824 /*
1825  * Convert dquot to 64-bit network-order structure.
1826  */
1827 static void
1828 dq_dqb64(const struct dquot *dq, struct dqblk64 *dqb64)
1829 {
1830 
1831 	dqb64->dqb_bhardlimit = htobe64(dq->dq_bhardlimit);
1832 	dqb64->dqb_bsoftlimit = htobe64(dq->dq_bsoftlimit);
1833 	dqb64->dqb_curblocks = htobe64(dq->dq_curblocks);
1834 	dqb64->dqb_ihardlimit = htobe64(dq->dq_ihardlimit);
1835 	dqb64->dqb_isoftlimit = htobe64(dq->dq_isoftlimit);
1836 	dqb64->dqb_curinodes = htobe64(dq->dq_curinodes);
1837 	dqb64->dqb_btime = htobe64(dq->dq_btime);
1838 	dqb64->dqb_itime = htobe64(dq->dq_itime);
1839 }
1840 
1841 /*
1842  * Convert 64-bit host-order structure to 32-bit host-order structure.
1843  */
1844 static void
1845 dqb64_dqb32(const struct dqblk64 *dqb64, struct dqblk32 *dqb32)
1846 {
1847 
1848 	dqb32->dqb_bhardlimit = CLIP32(dqb64->dqb_bhardlimit);
1849 	dqb32->dqb_bsoftlimit = CLIP32(dqb64->dqb_bsoftlimit);
1850 	dqb32->dqb_curblocks = CLIP32(dqb64->dqb_curblocks);
1851 	dqb32->dqb_ihardlimit = CLIP32(dqb64->dqb_ihardlimit);
1852 	dqb32->dqb_isoftlimit = CLIP32(dqb64->dqb_isoftlimit);
1853 	dqb32->dqb_curinodes = CLIP32(dqb64->dqb_curinodes);
1854 	dqb32->dqb_btime = CLIP32(dqb64->dqb_btime);
1855 	dqb32->dqb_itime = CLIP32(dqb64->dqb_itime);
1856 }
1857 
1858 /*
1859  * Convert 32-bit host-order structure to 64-bit host-order structure.
1860  */
1861 static void
1862 dqb32_dqb64(const struct dqblk32 *dqb32, struct dqblk64 *dqb64)
1863 {
1864 
1865 	dqb64->dqb_bhardlimit = dqb32->dqb_bhardlimit;
1866 	dqb64->dqb_bsoftlimit = dqb32->dqb_bsoftlimit;
1867 	dqb64->dqb_curblocks = dqb32->dqb_curblocks;
1868 	dqb64->dqb_ihardlimit = dqb32->dqb_ihardlimit;
1869 	dqb64->dqb_isoftlimit = dqb32->dqb_isoftlimit;
1870 	dqb64->dqb_curinodes = dqb32->dqb_curinodes;
1871 	dqb64->dqb_btime = dqb32->dqb_btime;
1872 	dqb64->dqb_itime = dqb32->dqb_itime;
1873 }
1874