xref: /freebsd/sys/ufs/ufs/ufs_quota.c (revision 8a0a413e)
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 	/*
163 	 * Disk quotas must be turned off for system files.  Currently
164 	 * snapshot and quota files.
165 	 */
166 	if ((vp->v_vflag & VV_SYSTEM) != 0)
167 		return (0);
168 	/*
169 	 * XXX: Turn off quotas for files with a negative UID or GID.
170 	 * This prevents the creation of 100GB+ quota files.
171 	 */
172 	if ((int)ip->i_uid < 0 || (int)ip->i_gid < 0)
173 		return (0);
174 #ifdef DIAGNOSTIC
175 	if ((flags & CHOWN) == 0)
176 		chkdquot(ip);
177 #endif
178 	if (change == 0)
179 		return (0);
180 	if (change < 0) {
181 		for (i = 0; i < MAXQUOTAS; i++) {
182 			if ((dq = ip->i_dquot[i]) == NODQUOT)
183 				continue;
184 			DQI_LOCK(dq);
185 			DQI_WAIT(dq, PINOD+1, "chkdq1");
186 			ncurblocks = dq->dq_curblocks + change;
187 			if (ncurblocks >= 0)
188 				dq->dq_curblocks = ncurblocks;
189 			else
190 				dq->dq_curblocks = 0;
191 			dq->dq_flags &= ~DQ_BLKS;
192 			dq->dq_flags |= DQ_MOD;
193 			DQI_UNLOCK(dq);
194 		}
195 		return (0);
196 	}
197 	if ((flags & FORCE) == 0 &&
198 	    priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA, 0))
199 		do_check = 1;
200 	else
201 		do_check = 0;
202 	for (i = 0; i < MAXQUOTAS; i++) {
203 		if ((dq = ip->i_dquot[i]) == NODQUOT)
204 			continue;
205 		warn = 0;
206 		DQI_LOCK(dq);
207 		DQI_WAIT(dq, PINOD+1, "chkdq2");
208 		if (do_check) {
209 			error = chkdqchg(ip, change, cred, i, &warn);
210 			if (error) {
211 				/*
212 				 * Roll back user quota changes when
213 				 * group quota failed.
214 				 */
215 				while (i > 0) {
216 					--i;
217 					dq = ip->i_dquot[i];
218 					if (dq == NODQUOT)
219 						continue;
220 					DQI_LOCK(dq);
221 					DQI_WAIT(dq, PINOD+1, "chkdq3");
222 					ncurblocks = dq->dq_curblocks - change;
223 					if (ncurblocks >= 0)
224 						dq->dq_curblocks = ncurblocks;
225 					else
226 						dq->dq_curblocks = 0;
227 					dq->dq_flags &= ~DQ_BLKS;
228 					dq->dq_flags |= DQ_MOD;
229 					DQI_UNLOCK(dq);
230 				}
231 				return (error);
232 			}
233 		}
234 		/* Reset timer when crossing soft limit */
235 		if (dq->dq_curblocks + change >= dq->dq_bsoftlimit &&
236 		    dq->dq_curblocks < dq->dq_bsoftlimit)
237 			dq->dq_btime = time_second + ITOUMP(ip)->um_btime[i];
238 		dq->dq_curblocks += change;
239 		dq->dq_flags |= DQ_MOD;
240 		DQI_UNLOCK(dq);
241 		if (warn)
242 			uprintf("\n%s: warning, %s disk quota exceeded\n",
243 			    ITOVFS(ip)->mnt_stat.f_mntonname,
244 			    quotatypes[i]);
245 	}
246 	return (0);
247 }
248 
249 /*
250  * Check for a valid change to a users allocation.
251  * Issue an error message if appropriate.
252  */
253 static int
254 chkdqchg(struct inode *ip, ufs2_daddr_t change, struct ucred *cred,
255     int type, int *warn)
256 {
257 	struct dquot *dq = ip->i_dquot[type];
258 	ufs2_daddr_t ncurblocks = dq->dq_curblocks + change;
259 
260 	/*
261 	 * If user would exceed their hard limit, disallow space allocation.
262 	 */
263 	if (ncurblocks >= dq->dq_bhardlimit && dq->dq_bhardlimit) {
264 		if ((dq->dq_flags & DQ_BLKS) == 0 &&
265 		    ip->i_uid == cred->cr_uid) {
266 			dq->dq_flags |= DQ_BLKS;
267 			DQI_UNLOCK(dq);
268 			uprintf("\n%s: write failed, %s disk limit reached\n",
269 			    ITOVFS(ip)->mnt_stat.f_mntonname,
270 			    quotatypes[type]);
271 			return (EDQUOT);
272 		}
273 		DQI_UNLOCK(dq);
274 		return (EDQUOT);
275 	}
276 	/*
277 	 * If user is over their soft limit for too long, disallow space
278 	 * allocation. Reset time limit as they cross their soft limit.
279 	 */
280 	if (ncurblocks >= dq->dq_bsoftlimit && dq->dq_bsoftlimit) {
281 		if (dq->dq_curblocks < dq->dq_bsoftlimit) {
282 			dq->dq_btime = time_second + ITOUMP(ip)->um_btime[type];
283 			if (ip->i_uid == cred->cr_uid)
284 				*warn = 1;
285 			return (0);
286 		}
287 		if (time_second > dq->dq_btime) {
288 			if ((dq->dq_flags & DQ_BLKS) == 0 &&
289 			    ip->i_uid == cred->cr_uid) {
290 				dq->dq_flags |= DQ_BLKS;
291 				DQI_UNLOCK(dq);
292 				uprintf("\n%s: write failed, %s "
293 				    "disk quota exceeded for too long\n",
294 				    ITOVFS(ip)->mnt_stat.f_mntonname,
295 				    quotatypes[type]);
296 				return (EDQUOT);
297 			}
298 			DQI_UNLOCK(dq);
299 			return (EDQUOT);
300 		}
301 	}
302 	return (0);
303 }
304 
305 /*
306  * Check the inode limit, applying corrective action.
307  */
308 int
309 chkiq(struct inode *ip, int change, struct ucred *cred, int flags)
310 {
311 	struct dquot *dq;
312 	int i, error, warn, do_check;
313 
314 #ifdef DIAGNOSTIC
315 	if ((flags & CHOWN) == 0)
316 		chkdquot(ip);
317 #endif
318 	if (change == 0)
319 		return (0);
320 	if (change < 0) {
321 		for (i = 0; i < MAXQUOTAS; i++) {
322 			if ((dq = ip->i_dquot[i]) == NODQUOT)
323 				continue;
324 			DQI_LOCK(dq);
325 			DQI_WAIT(dq, PINOD+1, "chkiq1");
326 			if (dq->dq_curinodes >= -change)
327 				dq->dq_curinodes += change;
328 			else
329 				dq->dq_curinodes = 0;
330 			dq->dq_flags &= ~DQ_INODS;
331 			dq->dq_flags |= DQ_MOD;
332 			DQI_UNLOCK(dq);
333 		}
334 		return (0);
335 	}
336 	if ((flags & FORCE) == 0 &&
337 	    priv_check_cred(cred, PRIV_VFS_EXCEEDQUOTA, 0))
338 		do_check = 1;
339 	else
340 		do_check = 0;
341 	for (i = 0; i < MAXQUOTAS; i++) {
342 		if ((dq = ip->i_dquot[i]) == NODQUOT)
343 			continue;
344 		warn = 0;
345 		DQI_LOCK(dq);
346 		DQI_WAIT(dq, PINOD+1, "chkiq2");
347 		if (do_check) {
348 			error = chkiqchg(ip, change, cred, i, &warn);
349 			if (error) {
350 				/*
351 				 * Roll back user quota changes when
352 				 * group quota failed.
353 				 */
354 				while (i > 0) {
355 					--i;
356 					dq = ip->i_dquot[i];
357 					if (dq == NODQUOT)
358 						continue;
359 					DQI_LOCK(dq);
360 					DQI_WAIT(dq, PINOD+1, "chkiq3");
361 					if (dq->dq_curinodes >= change)
362 						dq->dq_curinodes -= change;
363 					else
364 						dq->dq_curinodes = 0;
365 					dq->dq_flags &= ~DQ_INODS;
366 					dq->dq_flags |= DQ_MOD;
367 					DQI_UNLOCK(dq);
368 				}
369 				return (error);
370 			}
371 		}
372 		/* Reset timer when crossing soft limit */
373 		if (dq->dq_curinodes + change >= dq->dq_isoftlimit &&
374 		    dq->dq_curinodes < dq->dq_isoftlimit)
375 			dq->dq_itime = time_second + ITOUMP(ip)->um_itime[i];
376 		dq->dq_curinodes += change;
377 		dq->dq_flags |= DQ_MOD;
378 		DQI_UNLOCK(dq);
379 		if (warn)
380 			uprintf("\n%s: warning, %s inode quota exceeded\n",
381 			    ITOVFS(ip)->mnt_stat.f_mntonname,
382 			    quotatypes[i]);
383 	}
384 	return (0);
385 }
386 
387 /*
388  * Check for a valid change to a users allocation.
389  * Issue an error message if appropriate.
390  */
391 static int
392 chkiqchg(struct inode *ip, int change, struct ucred *cred, int type, int *warn)
393 {
394 	struct dquot *dq = ip->i_dquot[type];
395 	ino_t ncurinodes = dq->dq_curinodes + change;
396 
397 	/*
398 	 * If user would exceed their hard limit, disallow inode allocation.
399 	 */
400 	if (ncurinodes >= dq->dq_ihardlimit && dq->dq_ihardlimit) {
401 		if ((dq->dq_flags & DQ_INODS) == 0 &&
402 		    ip->i_uid == cred->cr_uid) {
403 			dq->dq_flags |= DQ_INODS;
404 			DQI_UNLOCK(dq);
405 			uprintf("\n%s: write failed, %s inode limit reached\n",
406 			    ITOVFS(ip)->mnt_stat.f_mntonname,
407 			    quotatypes[type]);
408 			return (EDQUOT);
409 		}
410 		DQI_UNLOCK(dq);
411 		return (EDQUOT);
412 	}
413 	/*
414 	 * If user is over their soft limit for too long, disallow inode
415 	 * allocation. Reset time limit as they cross their soft limit.
416 	 */
417 	if (ncurinodes >= dq->dq_isoftlimit && dq->dq_isoftlimit) {
418 		if (dq->dq_curinodes < dq->dq_isoftlimit) {
419 			dq->dq_itime = time_second + ITOUMP(ip)->um_itime[type];
420 			if (ip->i_uid == cred->cr_uid)
421 				*warn = 1;
422 			return (0);
423 		}
424 		if (time_second > dq->dq_itime) {
425 			if ((dq->dq_flags & DQ_INODS) == 0 &&
426 			    ip->i_uid == cred->cr_uid) {
427 				dq->dq_flags |= DQ_INODS;
428 				DQI_UNLOCK(dq);
429 				uprintf("\n%s: write failed, %s "
430 				    "inode quota exceeded for too long\n",
431 				    ITOVFS(ip)->mnt_stat.f_mntonname,
432 				    quotatypes[type]);
433 				return (EDQUOT);
434 			}
435 			DQI_UNLOCK(dq);
436 			return (EDQUOT);
437 		}
438 	}
439 	return (0);
440 }
441 
442 #ifdef DIAGNOSTIC
443 /*
444  * On filesystems with quotas enabled, it is an error for a file to change
445  * size and not to have a dquot structure associated with it.
446  */
447 static void
448 chkdquot(struct inode *ip)
449 {
450 	struct ufsmount *ump;
451 	struct vnode *vp;
452 	int i;
453 
454 	ump = ITOUMP(ip);
455 	vp = ITOV(ip);
456 
457 	/*
458 	 * Disk quotas must be turned off for system files.  Currently
459 	 * these are snapshots and quota files.
460 	 */
461 	if ((vp->v_vflag & VV_SYSTEM) != 0)
462 		return;
463 	/*
464 	 * XXX: Turn off quotas for files with a negative UID or GID.
465 	 * This prevents the creation of 100GB+ quota files.
466 	 */
467 	if ((int)ip->i_uid < 0 || (int)ip->i_gid < 0)
468 		return;
469 
470 	UFS_LOCK(ump);
471 	for (i = 0; i < MAXQUOTAS; i++) {
472 		if (ump->um_quotas[i] == NULLVP ||
473 		    (ump->um_qflags[i] & (QTF_OPENING|QTF_CLOSING)))
474 			continue;
475 		if (ip->i_dquot[i] == NODQUOT) {
476 			UFS_UNLOCK(ump);
477 			vn_printf(ITOV(ip), "chkdquot: missing dquot ");
478 			panic("chkdquot: missing dquot");
479 		}
480 	}
481 	UFS_UNLOCK(ump);
482 }
483 #endif
484 
485 /*
486  * Code to process quotactl commands.
487  */
488 
489 /*
490  * Q_QUOTAON - set up a quota file for a particular filesystem.
491  */
492 int
493 quotaon(struct thread *td, struct mount *mp, int type, void *fname)
494 {
495 	struct ufsmount *ump;
496 	struct vnode *vp, **vpp;
497 	struct vnode *mvp;
498 	struct dquot *dq;
499 	int error, flags;
500 	struct nameidata nd;
501 
502 	error = priv_check(td, PRIV_UFS_QUOTAON);
503 	if (error != 0) {
504 		vfs_unbusy(mp);
505 		return (error);
506 	}
507 
508 	if ((mp->mnt_flag & MNT_RDONLY) != 0) {
509 		vfs_unbusy(mp);
510 		return (EROFS);
511 	}
512 
513 	ump = VFSTOUFS(mp);
514 	dq = NODQUOT;
515 
516 	NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, fname, td);
517 	flags = FREAD | FWRITE;
518 	vfs_ref(mp);
519 	vfs_unbusy(mp);
520 	error = vn_open(&nd, &flags, 0, NULL);
521 	if (error != 0) {
522 		vfs_rel(mp);
523 		return (error);
524 	}
525 	NDFREE(&nd, NDF_ONLY_PNBUF);
526 	vp = nd.ni_vp;
527 	error = vfs_busy(mp, MBF_NOWAIT);
528 	vfs_rel(mp);
529 	if (error == 0) {
530 		if (vp->v_type != VREG) {
531 			error = EACCES;
532 			vfs_unbusy(mp);
533 		}
534 	}
535 	if (error != 0) {
536 		VOP_UNLOCK(vp, 0);
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, 0);
545 		(void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td);
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_unbusy(mp);
558 		return (error);
559 	}
560 	VOP_UNLOCK(vp, 0);
561 	MNT_ILOCK(mp);
562 	mp->mnt_flag |= MNT_QUOTA;
563 	MNT_IUNLOCK(mp);
564 
565 	vpp = &ump->um_quotas[type];
566 	if (*vpp != vp)
567 		quotaoff1(td, mp, type);
568 
569 	/*
570 	 * When the directory vnode containing the quota file is
571 	 * inactivated, due to the shared lookup of the quota file
572 	 * vput()ing the dvp, the qsyncvp() call for the containing
573 	 * directory would try to acquire the quota lock exclusive.
574 	 * At the same time, lookup already locked the quota vnode
575 	 * shared.  Mark the quota vnode lock as allowing recursion
576 	 * and automatically converting shared locks to exclusive.
577 	 *
578 	 * Also mark quota vnode as system.
579 	 */
580 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
581 	vp->v_vflag |= VV_SYSTEM;
582 	VN_LOCK_AREC(vp);
583 	VN_LOCK_DSHARE(vp);
584 	VOP_UNLOCK(vp, 0);
585 	*vpp = vp;
586 	/*
587 	 * Save the credential of the process that turned on quotas.
588 	 * Set up the time limits for this quota.
589 	 */
590 	ump->um_cred[type] = crhold(td->td_ucred);
591 	ump->um_btime[type] = MAX_DQ_TIME;
592 	ump->um_itime[type] = MAX_IQ_TIME;
593 	if (dqget(NULLVP, 0, ump, type, &dq) == 0) {
594 		if (dq->dq_btime > 0)
595 			ump->um_btime[type] = dq->dq_btime;
596 		if (dq->dq_itime > 0)
597 			ump->um_itime[type] = dq->dq_itime;
598 		dqrele(NULLVP, dq);
599 	}
600 	/*
601 	 * Allow the getdq from getinoquota below to read the quota
602 	 * from file.
603 	 */
604 	UFS_LOCK(ump);
605 	ump->um_qflags[type] &= ~QTF_CLOSING;
606 	UFS_UNLOCK(ump);
607 	/*
608 	 * Search vnodes associated with this mount point,
609 	 * adding references to quota file being opened.
610 	 * NB: only need to add dquot's for inodes being modified.
611 	 */
612 again:
613 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
614 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
615 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
616 			goto again;
617 		}
618 		if (vp->v_type == VNON || vp->v_writecount == 0) {
619 			VOP_UNLOCK(vp, 0);
620 			vrele(vp);
621 			continue;
622 		}
623 		error = getinoquota(VTOI(vp));
624 		VOP_UNLOCK(vp, 0);
625 		vrele(vp);
626 		if (error) {
627 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
628 			break;
629 		}
630 	}
631 
632         if (error)
633 		quotaoff_inchange(td, mp, type);
634 	UFS_LOCK(ump);
635 	ump->um_qflags[type] &= ~QTF_OPENING;
636 	KASSERT((ump->um_qflags[type] & QTF_CLOSING) == 0,
637 		("quotaon: leaking flags"));
638 	UFS_UNLOCK(ump);
639 
640 	vfs_unbusy(mp);
641 	return (error);
642 }
643 
644 /*
645  * Main code to turn off disk quotas for a filesystem. Does not change
646  * flags.
647  */
648 static int
649 quotaoff1(struct thread *td, struct mount *mp, int type)
650 {
651 	struct vnode *vp;
652 	struct vnode *qvp, *mvp;
653 	struct ufsmount *ump;
654 	struct dquot *dq;
655 	struct inode *ip;
656 	struct ucred *cr;
657 	int error;
658 
659 	ump = VFSTOUFS(mp);
660 
661 	UFS_LOCK(ump);
662 	KASSERT((ump->um_qflags[type] & QTF_CLOSING) != 0,
663 		("quotaoff1: flags are invalid"));
664 	if ((qvp = ump->um_quotas[type]) == NULLVP) {
665 		UFS_UNLOCK(ump);
666 		return (0);
667 	}
668 	cr = ump->um_cred[type];
669 	UFS_UNLOCK(ump);
670 
671 	/*
672 	 * Search vnodes associated with this mount point,
673 	 * deleting any references to quota file being closed.
674 	 */
675 again:
676 	MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
677 		if (vp->v_type == VNON) {
678 			VI_UNLOCK(vp);
679 			continue;
680 		}
681 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
682 			MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
683 			goto again;
684 		}
685 		ip = VTOI(vp);
686 		dq = ip->i_dquot[type];
687 		ip->i_dquot[type] = NODQUOT;
688 		dqrele(vp, dq);
689 		VOP_UNLOCK(vp, 0);
690 		vrele(vp);
691 	}
692 
693 	error = dqflush(qvp);
694 	if (error != 0)
695 		return (error);
696 
697 	/*
698 	 * Clear um_quotas before closing the quota vnode to prevent
699 	 * access to the closed vnode from dqget/dqsync
700 	 */
701 	UFS_LOCK(ump);
702 	ump->um_quotas[type] = NULLVP;
703 	ump->um_cred[type] = NOCRED;
704 	UFS_UNLOCK(ump);
705 
706 	vn_lock(qvp, LK_EXCLUSIVE | LK_RETRY);
707 	qvp->v_vflag &= ~VV_SYSTEM;
708 	VOP_UNLOCK(qvp, 0);
709 	error = vn_close(qvp, FREAD|FWRITE, td->td_ucred, td);
710 	crfree(cr);
711 
712 	return (error);
713 }
714 
715 /*
716  * Turns off quotas, assumes that ump->um_qflags are already checked
717  * and QTF_CLOSING is set to indicate operation in progress. Fixes
718  * ump->um_qflags and mp->mnt_flag after.
719  */
720 int
721 quotaoff_inchange(struct thread *td, struct mount *mp, int type)
722 {
723 	struct ufsmount *ump;
724 	int i;
725 	int error;
726 
727 	error = quotaoff1(td, mp, type);
728 
729 	ump = VFSTOUFS(mp);
730 	UFS_LOCK(ump);
731 	ump->um_qflags[type] &= ~QTF_CLOSING;
732 	for (i = 0; i < MAXQUOTAS; i++)
733 		if (ump->um_quotas[i] != NULLVP)
734 			break;
735 	if (i == MAXQUOTAS) {
736 		MNT_ILOCK(mp);
737 		mp->mnt_flag &= ~MNT_QUOTA;
738 		MNT_IUNLOCK(mp);
739 	}
740 	UFS_UNLOCK(ump);
741 	return (error);
742 }
743 
744 /*
745  * Q_QUOTAOFF - turn off disk quotas for a filesystem.
746  */
747 int
748 quotaoff(struct thread *td, struct mount *mp, int type)
749 {
750 	struct ufsmount *ump;
751 	int error;
752 
753 	error = priv_check(td, PRIV_UFS_QUOTAOFF);
754 	if (error)
755 		return (error);
756 
757 	ump = VFSTOUFS(mp);
758 	UFS_LOCK(ump);
759 	if ((ump->um_qflags[type] & (QTF_OPENING|QTF_CLOSING)) != 0) {
760 		UFS_UNLOCK(ump);
761 		return (EALREADY);
762 	}
763 	ump->um_qflags[type] |= QTF_CLOSING;
764 	UFS_UNLOCK(ump);
765 
766 	return (quotaoff_inchange(td, mp, type));
767 }
768 
769 /*
770  * Q_GETQUOTA - return current values in a dqblk structure.
771  */
772 static int
773 _getquota(struct thread *td, struct mount *mp, u_long id, int type,
774     struct dqblk64 *dqb)
775 {
776 	struct dquot *dq;
777 	int error;
778 
779 	switch (type) {
780 	case USRQUOTA:
781 		if ((td->td_ucred->cr_uid != id) && !unprivileged_get_quota) {
782 			error = priv_check(td, PRIV_VFS_GETQUOTA);
783 			if (error)
784 				return (error);
785 		}
786 		break;
787 
788 	case GRPQUOTA:
789 		if (!groupmember(id, td->td_ucred) &&
790 		    !unprivileged_get_quota) {
791 			error = priv_check(td, PRIV_VFS_GETQUOTA);
792 			if (error)
793 				return (error);
794 		}
795 		break;
796 
797 	default:
798 		return (EINVAL);
799 	}
800 
801 	dq = NODQUOT;
802 	error = dqget(NULLVP, id, VFSTOUFS(mp), type, &dq);
803 	if (error)
804 		return (error);
805 	*dqb = dq->dq_dqb;
806 	dqrele(NULLVP, dq);
807 	return (error);
808 }
809 
810 /*
811  * Q_SETQUOTA - assign an entire dqblk structure.
812  */
813 static int
814 _setquota(struct thread *td, struct mount *mp, u_long id, int type,
815     struct dqblk64 *dqb)
816 {
817 	struct dquot *dq;
818 	struct dquot *ndq;
819 	struct ufsmount *ump;
820 	struct dqblk64 newlim;
821 	int error;
822 
823 	error = priv_check(td, PRIV_VFS_SETQUOTA);
824 	if (error)
825 		return (error);
826 
827 	newlim = *dqb;
828 
829 	ndq = NODQUOT;
830 	ump = VFSTOUFS(mp);
831 
832 	error = dqget(NULLVP, id, ump, type, &ndq);
833 	if (error)
834 		return (error);
835 	dq = ndq;
836 	DQI_LOCK(dq);
837 	DQI_WAIT(dq, PINOD+1, "setqta");
838 	/*
839 	 * Copy all but the current values.
840 	 * Reset time limit if previously had no soft limit or were
841 	 * under it, but now have a soft limit and are over it.
842 	 */
843 	newlim.dqb_curblocks = dq->dq_curblocks;
844 	newlim.dqb_curinodes = dq->dq_curinodes;
845 	if (dq->dq_id != 0) {
846 		newlim.dqb_btime = dq->dq_btime;
847 		newlim.dqb_itime = dq->dq_itime;
848 	}
849 	if (newlim.dqb_bsoftlimit &&
850 	    dq->dq_curblocks >= newlim.dqb_bsoftlimit &&
851 	    (dq->dq_bsoftlimit == 0 || dq->dq_curblocks < dq->dq_bsoftlimit))
852 		newlim.dqb_btime = time_second + ump->um_btime[type];
853 	if (newlim.dqb_isoftlimit &&
854 	    dq->dq_curinodes >= newlim.dqb_isoftlimit &&
855 	    (dq->dq_isoftlimit == 0 || dq->dq_curinodes < dq->dq_isoftlimit))
856 		newlim.dqb_itime = time_second + ump->um_itime[type];
857 	dq->dq_dqb = newlim;
858 	if (dq->dq_curblocks < dq->dq_bsoftlimit)
859 		dq->dq_flags &= ~DQ_BLKS;
860 	if (dq->dq_curinodes < dq->dq_isoftlimit)
861 		dq->dq_flags &= ~DQ_INODS;
862 	if (dq->dq_isoftlimit == 0 && dq->dq_bsoftlimit == 0 &&
863 	    dq->dq_ihardlimit == 0 && dq->dq_bhardlimit == 0)
864 		dq->dq_flags |= DQ_FAKE;
865 	else
866 		dq->dq_flags &= ~DQ_FAKE;
867 	dq->dq_flags |= DQ_MOD;
868 	DQI_UNLOCK(dq);
869 	dqrele(NULLVP, dq);
870 	return (0);
871 }
872 
873 /*
874  * Q_SETUSE - set current inode and block usage.
875  */
876 static int
877 _setuse(struct thread *td, struct mount *mp, u_long id, int type,
878     struct dqblk64 *dqb)
879 {
880 	struct dquot *dq;
881 	struct ufsmount *ump;
882 	struct dquot *ndq;
883 	struct dqblk64 usage;
884 	int error;
885 
886 	error = priv_check(td, PRIV_UFS_SETUSE);
887 	if (error)
888 		return (error);
889 
890 	usage = *dqb;
891 
892 	ump = VFSTOUFS(mp);
893 	ndq = NODQUOT;
894 
895 	error = dqget(NULLVP, id, ump, type, &ndq);
896 	if (error)
897 		return (error);
898 	dq = ndq;
899 	DQI_LOCK(dq);
900 	DQI_WAIT(dq, PINOD+1, "setuse");
901 	/*
902 	 * Reset time limit if have a soft limit and were
903 	 * previously under it, but are now over it.
904 	 */
905 	if (dq->dq_bsoftlimit && dq->dq_curblocks < dq->dq_bsoftlimit &&
906 	    usage.dqb_curblocks >= dq->dq_bsoftlimit)
907 		dq->dq_btime = time_second + ump->um_btime[type];
908 	if (dq->dq_isoftlimit && dq->dq_curinodes < dq->dq_isoftlimit &&
909 	    usage.dqb_curinodes >= dq->dq_isoftlimit)
910 		dq->dq_itime = time_second + ump->um_itime[type];
911 	dq->dq_curblocks = usage.dqb_curblocks;
912 	dq->dq_curinodes = usage.dqb_curinodes;
913 	if (dq->dq_curblocks < dq->dq_bsoftlimit)
914 		dq->dq_flags &= ~DQ_BLKS;
915 	if (dq->dq_curinodes < dq->dq_isoftlimit)
916 		dq->dq_flags &= ~DQ_INODS;
917 	dq->dq_flags |= DQ_MOD;
918 	DQI_UNLOCK(dq);
919 	dqrele(NULLVP, dq);
920 	return (0);
921 }
922 
923 int
924 getquota32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
925 {
926 	struct dqblk32 dqb32;
927 	struct dqblk64 dqb64;
928 	int error;
929 
930 	error = _getquota(td, mp, id, type, &dqb64);
931 	if (error)
932 		return (error);
933 	dqb64_dqb32(&dqb64, &dqb32);
934 	error = copyout(&dqb32, addr, sizeof(dqb32));
935 	return (error);
936 }
937 
938 int
939 setquota32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
940 {
941 	struct dqblk32 dqb32;
942 	struct dqblk64 dqb64;
943 	int error;
944 
945 	error = copyin(addr, &dqb32, sizeof(dqb32));
946 	if (error)
947 		return (error);
948 	dqb32_dqb64(&dqb32, &dqb64);
949 	error = _setquota(td, mp, id, type, &dqb64);
950 	return (error);
951 }
952 
953 int
954 setuse32(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
955 {
956 	struct dqblk32 dqb32;
957 	struct dqblk64 dqb64;
958 	int error;
959 
960 	error = copyin(addr, &dqb32, sizeof(dqb32));
961 	if (error)
962 		return (error);
963 	dqb32_dqb64(&dqb32, &dqb64);
964 	error = _setuse(td, mp, id, type, &dqb64);
965 	return (error);
966 }
967 
968 int
969 getquota(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
970 {
971 	struct dqblk64 dqb64;
972 	int error;
973 
974 	error = _getquota(td, mp, id, type, &dqb64);
975 	if (error)
976 		return (error);
977 	error = copyout(&dqb64, addr, sizeof(dqb64));
978 	return (error);
979 }
980 
981 int
982 setquota(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
983 {
984 	struct dqblk64 dqb64;
985 	int error;
986 
987 	error = copyin(addr, &dqb64, sizeof(dqb64));
988 	if (error)
989 		return (error);
990 	error = _setquota(td, mp, id, type, &dqb64);
991 	return (error);
992 }
993 
994 int
995 setuse(struct thread *td, struct mount *mp, u_long id, int type, void *addr)
996 {
997 	struct dqblk64 dqb64;
998 	int error;
999 
1000 	error = copyin(addr, &dqb64, sizeof(dqb64));
1001 	if (error)
1002 		return (error);
1003 	error = _setuse(td, mp, id, type, &dqb64);
1004 	return (error);
1005 }
1006 
1007 /*
1008  * Q_GETQUOTASIZE - get bit-size of quota file fields
1009  */
1010 int
1011 getquotasize(struct thread *td, struct mount *mp, u_long id, int type,
1012     void *sizep)
1013 {
1014 	struct ufsmount *ump = VFSTOUFS(mp);
1015 	int bitsize;
1016 
1017 	UFS_LOCK(ump);
1018 	if (ump->um_quotas[type] == NULLVP ||
1019 	    (ump->um_qflags[type] & QTF_CLOSING)) {
1020 		UFS_UNLOCK(ump);
1021 		return (EINVAL);
1022 	}
1023 	if ((ump->um_qflags[type] & QTF_64BIT) != 0)
1024 		bitsize = 64;
1025 	else
1026 		bitsize = 32;
1027 	UFS_UNLOCK(ump);
1028 	return (copyout(&bitsize, sizep, sizeof(int)));
1029 }
1030 
1031 /*
1032  * Q_SYNC - sync quota files to disk.
1033  */
1034 int
1035 qsync(struct mount *mp)
1036 {
1037 	struct ufsmount *ump = VFSTOUFS(mp);
1038 	struct thread *td = curthread;		/* XXX */
1039 	struct vnode *vp, *mvp;
1040 	struct dquot *dq;
1041 	int i, error;
1042 
1043 	/*
1044 	 * Check if the mount point has any quotas.
1045 	 * If not, simply return.
1046 	 */
1047 	for (i = 0; i < MAXQUOTAS; i++)
1048 		if (ump->um_quotas[i] != NULLVP)
1049 			break;
1050 	if (i == MAXQUOTAS)
1051 		return (0);
1052 	/*
1053 	 * Search vnodes associated with this mount point,
1054 	 * synchronizing any modified dquot structures.
1055 	 */
1056 again:
1057 	MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
1058 		if (vp->v_type == VNON) {
1059 			VI_UNLOCK(vp);
1060 			continue;
1061 		}
1062 		error = vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td);
1063 		if (error) {
1064 			if (error == ENOENT) {
1065 				MNT_VNODE_FOREACH_ACTIVE_ABORT(mp, mvp);
1066 				goto again;
1067 			}
1068 			continue;
1069 		}
1070 		for (i = 0; i < MAXQUOTAS; i++) {
1071 			dq = VTOI(vp)->i_dquot[i];
1072 			if (dq != NODQUOT)
1073 				dqsync(vp, dq);
1074 		}
1075 		vput(vp);
1076 	}
1077 	return (0);
1078 }
1079 
1080 /*
1081  * Sync quota file for given vnode to disk.
1082  */
1083 int
1084 qsyncvp(struct vnode *vp)
1085 {
1086 	struct ufsmount *ump = VFSTOUFS(vp->v_mount);
1087 	struct dquot *dq;
1088 	int i;
1089 
1090 	/*
1091 	 * Check if the mount point has any quotas.
1092 	 * If not, simply return.
1093 	 */
1094 	for (i = 0; i < MAXQUOTAS; i++)
1095 		if (ump->um_quotas[i] != NULLVP)
1096 			break;
1097 	if (i == MAXQUOTAS)
1098 		return (0);
1099 	/*
1100 	 * Search quotas associated with this vnode
1101 	 * synchronizing any modified dquot structures.
1102 	 */
1103 	for (i = 0; i < MAXQUOTAS; i++) {
1104 		dq = VTOI(vp)->i_dquot[i];
1105 		if (dq != NODQUOT)
1106 			dqsync(vp, dq);
1107 	}
1108 	return (0);
1109 }
1110 
1111 /*
1112  * Code pertaining to management of the in-core dquot data structures.
1113  */
1114 #define DQHASH(dqvp, id) \
1115 	(&dqhashtbl[((((intptr_t)(dqvp)) >> 8) + id) & dqhash])
1116 static LIST_HEAD(dqhash, dquot) *dqhashtbl;
1117 static u_long dqhash;
1118 
1119 /*
1120  * Dquot free list.
1121  */
1122 #define	DQUOTINC	5	/* minimum free dquots desired */
1123 static TAILQ_HEAD(dqfreelist, dquot) dqfreelist;
1124 static long numdquot, desireddquot = DQUOTINC;
1125 
1126 /*
1127  * Lock to protect quota hash, dq free list and dq_cnt ref counters of
1128  * _all_ dqs.
1129  */
1130 struct mtx dqhlock;
1131 
1132 #define	DQH_LOCK()	mtx_lock(&dqhlock)
1133 #define	DQH_UNLOCK()	mtx_unlock(&dqhlock)
1134 
1135 static struct dquot *dqhashfind(struct dqhash *dqh, u_long id,
1136 	struct vnode *dqvp);
1137 
1138 /*
1139  * Initialize the quota system.
1140  */
1141 void
1142 dqinit(void)
1143 {
1144 
1145 	mtx_init(&dqhlock, "dqhlock", NULL, MTX_DEF);
1146 	dqhashtbl = hashinit(desiredvnodes, M_DQUOT, &dqhash);
1147 	TAILQ_INIT(&dqfreelist);
1148 }
1149 
1150 /*
1151  * Shut down the quota system.
1152  */
1153 void
1154 dquninit(void)
1155 {
1156 	struct dquot *dq;
1157 
1158 	hashdestroy(dqhashtbl, M_DQUOT, dqhash);
1159 	while ((dq = TAILQ_FIRST(&dqfreelist)) != NULL) {
1160 		TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
1161 		mtx_destroy(&dq->dq_lock);
1162 		free(dq, M_DQUOT);
1163 	}
1164 	mtx_destroy(&dqhlock);
1165 }
1166 
1167 static struct dquot *
1168 dqhashfind(struct dqhash *dqh, u_long id, struct vnode *dqvp)
1169 {
1170 	struct dquot *dq;
1171 
1172 	mtx_assert(&dqhlock, MA_OWNED);
1173 	LIST_FOREACH(dq, dqh, dq_hash) {
1174 		if (dq->dq_id != id ||
1175 		    dq->dq_ump->um_quotas[dq->dq_type] != dqvp)
1176 			continue;
1177 		/*
1178 		 * Cache hit with no references.  Take
1179 		 * the structure off the free list.
1180 		 */
1181 		if (dq->dq_cnt == 0)
1182 			TAILQ_REMOVE(&dqfreelist, dq, dq_freelist);
1183 		DQREF(dq);
1184 		return (dq);
1185 	}
1186 	return (NODQUOT);
1187 }
1188 
1189 /*
1190  * Determine the quota file type.
1191  *
1192  * A 32-bit quota file is simply an array of struct dqblk32.
1193  *
1194  * A 64-bit quota file is a struct dqhdr64 followed by an array of struct
1195  * dqblk64.  The header contains various magic bits which allow us to be
1196  * reasonably confident that it is indeeda 64-bit quota file and not just
1197  * a 32-bit quota file that just happens to "look right".
1198  *
1199  */
1200 static int
1201 dqopen(struct vnode *vp, struct ufsmount *ump, int type)
1202 {
1203 	struct dqhdr64 dqh;
1204 	struct iovec aiov;
1205 	struct uio auio;
1206 	int error;
1207 
1208 	ASSERT_VOP_LOCKED(vp, "dqopen");
1209 	auio.uio_iov = &aiov;
1210 	auio.uio_iovcnt = 1;
1211 	aiov.iov_base = &dqh;
1212 	aiov.iov_len = sizeof(dqh);
1213 	auio.uio_resid = sizeof(dqh);
1214 	auio.uio_offset = 0;
1215 	auio.uio_segflg = UIO_SYSSPACE;
1216 	auio.uio_rw = UIO_READ;
1217 	auio.uio_td = (struct thread *)0;
1218 	error = VOP_READ(vp, &auio, 0, ump->um_cred[type]);
1219 
1220 	if (error != 0)
1221 		return (error);
1222 	if (auio.uio_resid > 0) {
1223 		/* assume 32 bits */
1224 		return (0);
1225 	}
1226 
1227 	UFS_LOCK(ump);
1228 	if (strcmp(dqh.dqh_magic, Q_DQHDR64_MAGIC) == 0 &&
1229 	    be32toh(dqh.dqh_version) == Q_DQHDR64_VERSION &&
1230 	    be32toh(dqh.dqh_hdrlen) == (uint32_t)sizeof(struct dqhdr64) &&
1231 	    be32toh(dqh.dqh_reclen) == (uint32_t)sizeof(struct dqblk64)) {
1232 		/* XXX: what if the magic matches, but the sizes are wrong? */
1233 		ump->um_qflags[type] |= QTF_64BIT;
1234 	} else {
1235 		ump->um_qflags[type] &= ~QTF_64BIT;
1236 	}
1237 	UFS_UNLOCK(ump);
1238 
1239 	return (0);
1240 }
1241 
1242 /*
1243  * Obtain a dquot structure for the specified identifier and quota file
1244  * reading the information from the file if necessary.
1245  */
1246 static int
1247 dqget(struct vnode *vp, u_long id, struct ufsmount *ump, int type,
1248     struct dquot **dqp)
1249 {
1250 	uint8_t buf[sizeof(struct dqblk64)];
1251 	off_t base, recsize;
1252 	struct dquot *dq, *dq1;
1253 	struct dqhash *dqh;
1254 	struct vnode *dqvp;
1255 	struct iovec aiov;
1256 	struct uio auio;
1257 	int dqvplocked, error;
1258 
1259 #ifdef DEBUG_VFS_LOCKS
1260 	if (vp != NULLVP)
1261 		ASSERT_VOP_ELOCKED(vp, "dqget");
1262 #endif
1263 
1264 	if (vp != NULLVP && *dqp != NODQUOT) {
1265 		return (0);
1266 	}
1267 
1268 	/* XXX: Disallow negative id values to prevent the
1269 	* creation of 100GB+ quota data files.
1270 	*/
1271 	if ((int)id < 0)
1272 		return (EINVAL);
1273 
1274 	UFS_LOCK(ump);
1275 	dqvp = ump->um_quotas[type];
1276 	if (dqvp == NULLVP || (ump->um_qflags[type] & QTF_CLOSING)) {
1277 		*dqp = NODQUOT;
1278 		UFS_UNLOCK(ump);
1279 		return (EINVAL);
1280 	}
1281 	vref(dqvp);
1282 	UFS_UNLOCK(ump);
1283 	error = 0;
1284 	dqvplocked = 0;
1285 
1286 	/*
1287 	 * Check the cache first.
1288 	 */
1289 	dqh = DQHASH(dqvp, id);
1290 	DQH_LOCK();
1291 	dq = dqhashfind(dqh, id, dqvp);
1292 	if (dq != NULL) {
1293 		DQH_UNLOCK();
1294 hfound:		DQI_LOCK(dq);
1295 		DQI_WAIT(dq, PINOD+1, "dqget");
1296 		DQI_UNLOCK(dq);
1297 		if (dq->dq_ump == NULL) {
1298 			dqrele(vp, dq);
1299 			dq = NODQUOT;
1300 			error = EIO;
1301 		}
1302 		*dqp = dq;
1303 		if (dqvplocked)
1304 			vput(dqvp);
1305 		else
1306 			vrele(dqvp);
1307 		return (error);
1308 	}
1309 
1310 	/*
1311 	 * Quota vnode lock is before DQ_LOCK. Acquire dqvp lock there
1312 	 * since new dq will appear on the hash chain DQ_LOCKed.
1313 	 */
1314 	if (vp != dqvp) {
1315 		DQH_UNLOCK();
1316 		vn_lock(dqvp, LK_SHARED | LK_RETRY);
1317 		dqvplocked = 1;
1318 		DQH_LOCK();
1319 		/*
1320 		 * Recheck the cache after sleep for quota vnode lock.
1321 		 */
1322 		dq = dqhashfind(dqh, id, dqvp);
1323 		if (dq != NULL) {
1324 			DQH_UNLOCK();
1325 			goto hfound;
1326 		}
1327 	}
1328 
1329 	/*
1330 	 * Not in cache, allocate a new one or take it from the
1331 	 * free list.
1332 	 */
1333 	if (TAILQ_FIRST(&dqfreelist) == NODQUOT &&
1334 	    numdquot < MAXQUOTAS * desiredvnodes)
1335 		desireddquot += DQUOTINC;
1336 	if (numdquot < desireddquot) {
1337 		numdquot++;
1338 		DQH_UNLOCK();
1339 		dq1 = malloc(sizeof *dq1, M_DQUOT, M_WAITOK | M_ZERO);
1340 		mtx_init(&dq1->dq_lock, "dqlock", NULL, MTX_DEF);
1341 		DQH_LOCK();
1342 		/*
1343 		 * Recheck the cache after sleep for memory.
1344 		 */
1345 		dq = dqhashfind(dqh, id, dqvp);
1346 		if (dq != NULL) {
1347 			numdquot--;
1348 			DQH_UNLOCK();
1349 			mtx_destroy(&dq1->dq_lock);
1350 			free(dq1, M_DQUOT);
1351 			goto hfound;
1352 		}
1353 		dq = dq1;
1354 	} else {
1355 		if ((dq = TAILQ_FIRST(&dqfreelist)) == NULL) {
1356 			DQH_UNLOCK();
1357 			tablefull("dquot");
1358 			*dqp = NODQUOT;
1359 			if (dqvplocked)
1360 				vput(dqvp);
1361 			else
1362 				vrele(dqvp);
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 	error = VOP_READ(dqvp, &auio, 0, ump->um_cred[type]);
1406 	if (auio.uio_resid == recsize && error == 0) {
1407 		bzero(&dq->dq_dqb, sizeof(dq->dq_dqb));
1408 	} else {
1409 		if (ump->um_qflags[type] & QTF_64BIT)
1410 			dqb64_dq((struct dqblk64 *)buf, dq);
1411 		else
1412 			dqb32_dq((struct dqblk32 *)buf, dq);
1413 	}
1414 	if (dqvplocked)
1415 		vput(dqvp);
1416 	else
1417 		vrele(dqvp);
1418 	/*
1419 	 * I/O error in reading quota file, release
1420 	 * quota structure and reflect problem to caller.
1421 	 */
1422 	if (error) {
1423 		DQH_LOCK();
1424 		dq->dq_ump = NULL;
1425 		LIST_REMOVE(dq, dq_hash);
1426 		DQH_UNLOCK();
1427 		DQI_LOCK(dq);
1428 		if (dq->dq_flags & DQ_WANT)
1429 			wakeup(dq);
1430 		dq->dq_flags = 0;
1431 		DQI_UNLOCK(dq);
1432 		dqrele(vp, dq);
1433 		*dqp = NODQUOT;
1434 		return (error);
1435 	}
1436 	DQI_LOCK(dq);
1437 	/*
1438 	 * Check for no limit to enforce.
1439 	 * Initialize time values if necessary.
1440 	 */
1441 	if (dq->dq_isoftlimit == 0 && dq->dq_bsoftlimit == 0 &&
1442 	    dq->dq_ihardlimit == 0 && dq->dq_bhardlimit == 0)
1443 		dq->dq_flags |= DQ_FAKE;
1444 	if (dq->dq_id != 0) {
1445 		if (dq->dq_btime == 0) {
1446 			dq->dq_btime = time_second + ump->um_btime[type];
1447 			if (dq->dq_bsoftlimit &&
1448 			    dq->dq_curblocks >= dq->dq_bsoftlimit)
1449 				dq->dq_flags |= DQ_MOD;
1450 		}
1451 		if (dq->dq_itime == 0) {
1452 			dq->dq_itime = time_second + ump->um_itime[type];
1453 			if (dq->dq_isoftlimit &&
1454 			    dq->dq_curinodes >= dq->dq_isoftlimit)
1455 				dq->dq_flags |= DQ_MOD;
1456 		}
1457 	}
1458 	DQI_WAKEUP(dq);
1459 	DQI_UNLOCK(dq);
1460 	*dqp = dq;
1461 	return (0);
1462 }
1463 
1464 #ifdef DIAGNOSTIC
1465 /*
1466  * Obtain a reference to a dquot.
1467  */
1468 static void
1469 dqref(struct dquot *dq)
1470 {
1471 
1472 	dq->dq_cnt++;
1473 }
1474 #endif
1475 
1476 /*
1477  * Release a reference to a dquot.
1478  */
1479 void
1480 dqrele(struct vnode *vp, struct dquot *dq)
1481 {
1482 
1483 	if (dq == NODQUOT)
1484 		return;
1485 	DQH_LOCK();
1486 	KASSERT(dq->dq_cnt > 0, ("Lost dq %p reference 1", dq));
1487 	if (dq->dq_cnt > 1) {
1488 		dq->dq_cnt--;
1489 		DQH_UNLOCK();
1490 		return;
1491 	}
1492 	DQH_UNLOCK();
1493 sync:
1494 	(void) dqsync(vp, dq);
1495 
1496 	DQH_LOCK();
1497 	KASSERT(dq->dq_cnt > 0, ("Lost dq %p reference 2", dq));
1498 	if (--dq->dq_cnt > 0)
1499 	{
1500 		DQH_UNLOCK();
1501 		return;
1502 	}
1503 
1504 	/*
1505 	 * The dq may become dirty after it is synced but before it is
1506 	 * put to the free list. Checking the DQ_MOD there without
1507 	 * locking dq should be safe since no other references to the
1508 	 * dq exist.
1509 	 */
1510 	if ((dq->dq_flags & DQ_MOD) != 0) {
1511 		dq->dq_cnt++;
1512 		DQH_UNLOCK();
1513 		goto sync;
1514 	}
1515 	TAILQ_INSERT_TAIL(&dqfreelist, dq, dq_freelist);
1516 	DQH_UNLOCK();
1517 }
1518 
1519 /*
1520  * Update the disk quota in the quota file.
1521  */
1522 static int
1523 dqsync(struct vnode *vp, struct dquot *dq)
1524 {
1525 	uint8_t buf[sizeof(struct dqblk64)];
1526 	off_t base, recsize;
1527 	struct vnode *dqvp;
1528 	struct iovec aiov;
1529 	struct uio auio;
1530 	int error;
1531 	struct mount *mp;
1532 	struct ufsmount *ump;
1533 
1534 #ifdef DEBUG_VFS_LOCKS
1535 	if (vp != NULL)
1536 		ASSERT_VOP_ELOCKED(vp, "dqsync");
1537 #endif
1538 
1539 	mp = NULL;
1540 	error = 0;
1541 	if (dq == NODQUOT)
1542 		panic("dqsync: dquot");
1543 	if ((ump = dq->dq_ump) == NULL)
1544 		return (0);
1545 	UFS_LOCK(ump);
1546 	if ((dqvp = ump->um_quotas[dq->dq_type]) == NULLVP) {
1547 		if (vp == NULL) {
1548 			UFS_UNLOCK(ump);
1549 			return (0);
1550 		} else
1551 			panic("dqsync: file");
1552 	}
1553 	vref(dqvp);
1554 	UFS_UNLOCK(ump);
1555 
1556 	DQI_LOCK(dq);
1557 	if ((dq->dq_flags & DQ_MOD) == 0) {
1558 		DQI_UNLOCK(dq);
1559 		vrele(dqvp);
1560 		return (0);
1561 	}
1562 	DQI_UNLOCK(dq);
1563 
1564 	(void) vn_start_secondary_write(dqvp, &mp, V_WAIT);
1565 	if (vp != dqvp)
1566 		vn_lock(dqvp, LK_EXCLUSIVE | LK_RETRY);
1567 
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 	error = VOP_WRITE(dqvp, &auio, 0, dq->dq_ump->um_cred[dq->dq_type]);
1599 	if (auio.uio_resid && error == 0)
1600 		error = EIO;
1601 
1602 	DQI_LOCK(dq);
1603 	DQI_WAKEUP(dq);
1604 	dq->dq_flags &= ~DQ_MOD;
1605 out:
1606 	DQI_UNLOCK(dq);
1607 	if (vp != dqvp)
1608 		vput(dqvp);
1609 	else
1610 		vrele(dqvp);
1611 	vn_finished_secondary_write(mp);
1612 	return (error);
1613 }
1614 
1615 /*
1616  * Flush all entries from the cache for a particular vnode.
1617  */
1618 static int
1619 dqflush(struct vnode *vp)
1620 {
1621 	struct dquot *dq, *nextdq;
1622 	struct dqhash *dqh;
1623 	int error;
1624 
1625 	/*
1626 	 * Move all dquot's that used to refer to this quota
1627 	 * file off their hash chains (they will eventually
1628 	 * fall off the head of the free list and be re-used).
1629 	 */
1630 	error = 0;
1631 	DQH_LOCK();
1632 	for (dqh = &dqhashtbl[dqhash]; dqh >= dqhashtbl; dqh--) {
1633 		for (dq = LIST_FIRST(dqh); dq; dq = nextdq) {
1634 			nextdq = LIST_NEXT(dq, dq_hash);
1635 			if (dq->dq_ump->um_quotas[dq->dq_type] != vp)
1636 				continue;
1637 			if (dq->dq_cnt)
1638 				error = EBUSY;
1639 			else {
1640 				LIST_REMOVE(dq, dq_hash);
1641 				dq->dq_ump = NULL;
1642 			}
1643 		}
1644 	}
1645 	DQH_UNLOCK();
1646 	return (error);
1647 }
1648 
1649 /*
1650  * The following three functions are provided for the adjustment of
1651  * quotas by the soft updates code.
1652  */
1653 #ifdef SOFTUPDATES
1654 /*
1655  * Acquire a reference to the quota structures associated with a vnode.
1656  * Return count of number of quota structures found.
1657  */
1658 int
1659 quotaref(vp, qrp)
1660 	struct vnode *vp;
1661 	struct dquot **qrp;
1662 {
1663 	struct inode *ip;
1664 	struct dquot *dq;
1665 	int i, found;
1666 
1667 	for (i = 0; i < MAXQUOTAS; i++)
1668 		qrp[i] = NODQUOT;
1669 	/*
1670 	 * Disk quotas must be turned off for system files.  Currently
1671 	 * snapshot and quota files.
1672 	 */
1673 	if ((vp->v_vflag & VV_SYSTEM) != 0)
1674 		return (0);
1675 	/*
1676 	 * Iterate through and copy active quotas.
1677 	 */
1678 	found = 0;
1679 	ip = VTOI(vp);
1680 	mtx_lock(&dqhlock);
1681 	for (i = 0; i < MAXQUOTAS; i++) {
1682 		if ((dq = ip->i_dquot[i]) == NODQUOT)
1683 			continue;
1684 		DQREF(dq);
1685 		qrp[i] = dq;
1686 		found++;
1687 	}
1688 	mtx_unlock(&dqhlock);
1689 	return (found);
1690 }
1691 
1692 /*
1693  * Release a set of quota structures obtained from a vnode.
1694  */
1695 void
1696 quotarele(qrp)
1697 	struct dquot **qrp;
1698 {
1699 	struct dquot *dq;
1700 	int i;
1701 
1702 	for (i = 0; i < MAXQUOTAS; i++) {
1703 		if ((dq = qrp[i]) == NODQUOT)
1704 			continue;
1705 		dqrele(NULL, dq);
1706 	}
1707 }
1708 
1709 /*
1710  * Adjust the number of blocks associated with a quota.
1711  * Positive numbers when adding blocks; negative numbers when freeing blocks.
1712  */
1713 void
1714 quotaadj(qrp, ump, blkcount)
1715 	struct dquot **qrp;
1716 	struct ufsmount *ump;
1717 	int64_t blkcount;
1718 {
1719 	struct dquot *dq;
1720 	ufs2_daddr_t ncurblocks;
1721 	int i;
1722 
1723 	if (blkcount == 0)
1724 		return;
1725 	for (i = 0; i < MAXQUOTAS; i++) {
1726 		if ((dq = qrp[i]) == NODQUOT)
1727 			continue;
1728 		DQI_LOCK(dq);
1729 		DQI_WAIT(dq, PINOD+1, "adjqta");
1730 		ncurblocks = dq->dq_curblocks + blkcount;
1731 		if (ncurblocks >= 0)
1732 			dq->dq_curblocks = ncurblocks;
1733 		else
1734 			dq->dq_curblocks = 0;
1735 		if (blkcount < 0)
1736 			dq->dq_flags &= ~DQ_BLKS;
1737 		else if (dq->dq_curblocks + blkcount >= dq->dq_bsoftlimit &&
1738 			 dq->dq_curblocks < dq->dq_bsoftlimit)
1739 			dq->dq_btime = time_second + ump->um_btime[i];
1740 		dq->dq_flags |= DQ_MOD;
1741 		DQI_UNLOCK(dq);
1742 	}
1743 }
1744 #endif /* SOFTUPDATES */
1745 
1746 /*
1747  * 32-bit / 64-bit conversion functions.
1748  *
1749  * 32-bit quota records are stored in native byte order.  Attention must
1750  * be paid to overflow issues.
1751  *
1752  * 64-bit quota records are stored in network byte order.
1753  */
1754 
1755 #define CLIP32(u64) (u64 > UINT32_MAX ? UINT32_MAX : (uint32_t)u64)
1756 
1757 /*
1758  * Convert 32-bit host-order structure to dquot.
1759  */
1760 static void
1761 dqb32_dq(const struct dqblk32 *dqb32, struct dquot *dq)
1762 {
1763 
1764 	dq->dq_bhardlimit = dqb32->dqb_bhardlimit;
1765 	dq->dq_bsoftlimit = dqb32->dqb_bsoftlimit;
1766 	dq->dq_curblocks = dqb32->dqb_curblocks;
1767 	dq->dq_ihardlimit = dqb32->dqb_ihardlimit;
1768 	dq->dq_isoftlimit = dqb32->dqb_isoftlimit;
1769 	dq->dq_curinodes = dqb32->dqb_curinodes;
1770 	dq->dq_btime = dqb32->dqb_btime;
1771 	dq->dq_itime = dqb32->dqb_itime;
1772 }
1773 
1774 /*
1775  * Convert 64-bit network-order structure to dquot.
1776  */
1777 static void
1778 dqb64_dq(const struct dqblk64 *dqb64, struct dquot *dq)
1779 {
1780 
1781 	dq->dq_bhardlimit = be64toh(dqb64->dqb_bhardlimit);
1782 	dq->dq_bsoftlimit = be64toh(dqb64->dqb_bsoftlimit);
1783 	dq->dq_curblocks = be64toh(dqb64->dqb_curblocks);
1784 	dq->dq_ihardlimit = be64toh(dqb64->dqb_ihardlimit);
1785 	dq->dq_isoftlimit = be64toh(dqb64->dqb_isoftlimit);
1786 	dq->dq_curinodes = be64toh(dqb64->dqb_curinodes);
1787 	dq->dq_btime = be64toh(dqb64->dqb_btime);
1788 	dq->dq_itime = be64toh(dqb64->dqb_itime);
1789 }
1790 
1791 /*
1792  * Convert dquot to 32-bit host-order structure.
1793  */
1794 static void
1795 dq_dqb32(const struct dquot *dq, struct dqblk32 *dqb32)
1796 {
1797 
1798 	dqb32->dqb_bhardlimit = CLIP32(dq->dq_bhardlimit);
1799 	dqb32->dqb_bsoftlimit = CLIP32(dq->dq_bsoftlimit);
1800 	dqb32->dqb_curblocks = CLIP32(dq->dq_curblocks);
1801 	dqb32->dqb_ihardlimit = CLIP32(dq->dq_ihardlimit);
1802 	dqb32->dqb_isoftlimit = CLIP32(dq->dq_isoftlimit);
1803 	dqb32->dqb_curinodes = CLIP32(dq->dq_curinodes);
1804 	dqb32->dqb_btime = CLIP32(dq->dq_btime);
1805 	dqb32->dqb_itime = CLIP32(dq->dq_itime);
1806 }
1807 
1808 /*
1809  * Convert dquot to 64-bit network-order structure.
1810  */
1811 static void
1812 dq_dqb64(const struct dquot *dq, struct dqblk64 *dqb64)
1813 {
1814 
1815 	dqb64->dqb_bhardlimit = htobe64(dq->dq_bhardlimit);
1816 	dqb64->dqb_bsoftlimit = htobe64(dq->dq_bsoftlimit);
1817 	dqb64->dqb_curblocks = htobe64(dq->dq_curblocks);
1818 	dqb64->dqb_ihardlimit = htobe64(dq->dq_ihardlimit);
1819 	dqb64->dqb_isoftlimit = htobe64(dq->dq_isoftlimit);
1820 	dqb64->dqb_curinodes = htobe64(dq->dq_curinodes);
1821 	dqb64->dqb_btime = htobe64(dq->dq_btime);
1822 	dqb64->dqb_itime = htobe64(dq->dq_itime);
1823 }
1824 
1825 /*
1826  * Convert 64-bit host-order structure to 32-bit host-order structure.
1827  */
1828 static void
1829 dqb64_dqb32(const struct dqblk64 *dqb64, struct dqblk32 *dqb32)
1830 {
1831 
1832 	dqb32->dqb_bhardlimit = CLIP32(dqb64->dqb_bhardlimit);
1833 	dqb32->dqb_bsoftlimit = CLIP32(dqb64->dqb_bsoftlimit);
1834 	dqb32->dqb_curblocks = CLIP32(dqb64->dqb_curblocks);
1835 	dqb32->dqb_ihardlimit = CLIP32(dqb64->dqb_ihardlimit);
1836 	dqb32->dqb_isoftlimit = CLIP32(dqb64->dqb_isoftlimit);
1837 	dqb32->dqb_curinodes = CLIP32(dqb64->dqb_curinodes);
1838 	dqb32->dqb_btime = CLIP32(dqb64->dqb_btime);
1839 	dqb32->dqb_itime = CLIP32(dqb64->dqb_itime);
1840 }
1841 
1842 /*
1843  * Convert 32-bit host-order structure to 64-bit host-order structure.
1844  */
1845 static void
1846 dqb32_dqb64(const struct dqblk32 *dqb32, struct dqblk64 *dqb64)
1847 {
1848 
1849 	dqb64->dqb_bhardlimit = dqb32->dqb_bhardlimit;
1850 	dqb64->dqb_bsoftlimit = dqb32->dqb_bsoftlimit;
1851 	dqb64->dqb_curblocks = dqb32->dqb_curblocks;
1852 	dqb64->dqb_ihardlimit = dqb32->dqb_ihardlimit;
1853 	dqb64->dqb_isoftlimit = dqb32->dqb_isoftlimit;
1854 	dqb64->dqb_curinodes = dqb32->dqb_curinodes;
1855 	dqb64->dqb_btime = dqb32->dqb_btime;
1856 	dqb64->dqb_itime = dqb32->dqb_itime;
1857 }
1858