Home | History | Annotate | Line # | Download | only in lfs
      1 /*	$NetBSD: ulfs_quota2.c,v 1.35 2022/05/28 22:08:46 andvar Exp $	*/
      2 /*  from NetBSD: ufs_quota2.c,v 1.40 2015/03/28 19:24:05 maxv Exp Exp  */
      3 /*  from NetBSD: ffs_quota2.c,v 1.5 2015/02/22 14:12:48 maxv Exp  */
      4 
      5 /*-
      6   * Copyright (c) 2010 Manuel Bouyer
      7   * All rights reserved.
      8   *
      9   * Redistribution and use in source and binary forms, with or without
     10   * modification, are permitted provided that the following conditions
     11   * are met:
     12   * 1. Redistributions of source code must retain the above copyright
     13   *    notice, this list of conditions and the following disclaimer.
     14   * 2. Redistributions in binary form must reproduce the above copyright
     15   *    notice, this list of conditions and the following disclaimer in the
     16   *    documentation and/or other materials provided with the distribution.
     17   *
     18   * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     19   * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     20   * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     21   * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     22   * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     23   * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     24   * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     25   * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     26   * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     27   * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     28   * POSSIBILITY OF SUCH DAMAGE.
     29   */
     30 
     31 #include <sys/cdefs.h>
     32 __KERNEL_RCSID(0, "$NetBSD: ulfs_quota2.c,v 1.35 2022/05/28 22:08:46 andvar Exp $");
     33 
     34 #include <sys/buf.h>
     35 #include <sys/param.h>
     36 #include <sys/kernel.h>
     37 #include <sys/systm.h>
     38 #include <sys/namei.h>
     39 #include <sys/file.h>
     40 #include <sys/proc.h>
     41 #include <sys/vnode.h>
     42 #include <sys/mount.h>
     43 #include <sys/kauth.h>
     44 #include <sys/quota.h>
     45 #include <sys/quotactl.h>
     46 
     47 #include <ufs/lfs/lfs.h>
     48 #include <ufs/lfs/lfs_accessors.h>
     49 #include <ufs/lfs/lfs_extern.h>
     50 
     51 #include <ufs/lfs/ulfs_quota2.h>
     52 #include <ufs/lfs/ulfs_inode.h>
     53 #include <ufs/lfs/ulfsmount.h>
     54 #include <ufs/lfs/ulfs_bswap.h>
     55 #include <ufs/lfs/ulfs_extern.h>
     56 #include <ufs/lfs/ulfs_quota.h>
     57 
     58 /*
     59  * LOCKING:
     60  * Data in the entries are protected by the associated struct dquot's
     61  * dq_interlock (this means we can't read or change a quota entry without
     62  * grabbing a dquot for it).
     63  * The header and lists (including pointers in the data entries, and q2e_uid)
     64  * are protected by the global dqlock.
     65  * the locking order is dq_interlock -> dqlock
     66  */
     67 
     68 static int quota2_bwrite(struct mount *, struct buf *);
     69 static int getinoquota2(struct inode *, bool, bool, struct buf **,
     70     struct quota2_entry **);
     71 static int getq2h(struct ulfsmount *, int, struct buf **,
     72     struct quota2_header **, int);
     73 static int getq2e(struct ulfsmount *, int, daddr_t, int, struct buf **,
     74     struct quota2_entry **, int);
     75 static int quota2_walk_list(struct ulfsmount *, struct buf *, int,
     76     uint64_t *, int, void *,
     77     int (*func)(struct ulfsmount *, uint64_t *, struct quota2_entry *,
     78       uint64_t, void *));
     79 
     80 static const char *limnames[] = INITQLNAMES;
     81 
     82 static void
     83 quota2_dict_update_q2e_limits(int objtype, const struct quotaval *val,
     84     struct quota2_entry *q2e)
     85 {
     86 	/* make sure we can index q2e_val[] by the fs-independent objtype */
     87 	CTASSERT(QUOTA_OBJTYPE_BLOCKS == QL_BLOCK);
     88 	CTASSERT(QUOTA_OBJTYPE_FILES == QL_FILE);
     89 
     90 	q2e->q2e_val[objtype].q2v_hardlimit = val->qv_hardlimit;
     91 	q2e->q2e_val[objtype].q2v_softlimit = val->qv_softlimit;
     92 	q2e->q2e_val[objtype].q2v_grace = val->qv_grace;
     93 }
     94 
     95 /*
     96  * Convert internal representation to FS-independent representation.
     97  * (Note that while the two types are currently identical, the
     98  * internal representation is an on-disk struct and the FS-independent
     99  * representation is not, and they might diverge in the future.)
    100  */
    101 static void
    102 q2val_to_quotaval(struct quota2_val *q2v, struct quotaval *qv)
    103 {
    104 	qv->qv_softlimit = q2v->q2v_softlimit;
    105 	qv->qv_hardlimit = q2v->q2v_hardlimit;
    106 	qv->qv_usage = q2v->q2v_cur;
    107 	qv->qv_expiretime = q2v->q2v_time;
    108 	qv->qv_grace = q2v->q2v_grace;
    109 }
    110 
    111 /*
    112  * Convert a quota2entry and default-flag to the FS-independent
    113  * representation.
    114  */
    115 static void
    116 q2e_to_quotaval(struct quota2_entry *q2e, int def,
    117 	       id_t *id, int objtype, struct quotaval *ret)
    118 {
    119 	if (def) {
    120 		*id = QUOTA_DEFAULTID;
    121 	} else {
    122 		*id = q2e->q2e_uid;
    123 	}
    124 
    125 	KASSERT(objtype >= 0 && objtype < N_QL);
    126 	q2val_to_quotaval(&q2e->q2e_val[objtype], ret);
    127 }
    128 
    129 
    130 static int
    131 quota2_bwrite(struct mount *mp, struct buf *bp)
    132 {
    133 	if (mp->mnt_flag & MNT_SYNCHRONOUS)
    134 		return bwrite(bp);
    135 	else {
    136 		bdwrite(bp);
    137 		return 0;
    138 	}
    139 }
    140 
    141 static int
    142 getq2h(struct ulfsmount *ump, int type,
    143     struct buf **bpp, struct quota2_header **q2hp, int flags)
    144 {
    145 	struct lfs *fs = ump->um_lfs;
    146 	const int needswap = ULFS_MPNEEDSWAP(fs);
    147 	int error;
    148 	struct buf *bp;
    149 	struct quota2_header *q2h;
    150 
    151 	KASSERT(mutex_owned(&lfs_dqlock));
    152 	error = bread(ump->um_quotas[type], 0, ump->umq2_bsize, flags, &bp);
    153 	if (error)
    154 		return error;
    155 	if (bp->b_resid != 0)
    156 		panic("dq2get: %s quota file truncated", lfs_quotatypes[type]);
    157 
    158 	q2h = (void *)bp->b_data;
    159 	if (ulfs_rw32(q2h->q2h_magic_number, needswap) != Q2_HEAD_MAGIC ||
    160 	    q2h->q2h_type != type)
    161 		panic("dq2get: corrupted %s quota header", lfs_quotatypes[type]);
    162 	*bpp = bp;
    163 	*q2hp = q2h;
    164 	return 0;
    165 }
    166 
    167 static int
    168 getq2e(struct ulfsmount *ump, int type, daddr_t lblkno, int blkoffset,
    169     struct buf **bpp, struct quota2_entry **q2ep, int flags)
    170 {
    171 	int error;
    172 	struct buf *bp;
    173 
    174 	if (blkoffset & (sizeof(uint64_t) - 1)) {
    175 		panic("dq2get: %s quota file corrupted",
    176 		    lfs_quotatypes[type]);
    177 	}
    178 	error = bread(ump->um_quotas[type], lblkno, ump->umq2_bsize, flags, &bp);
    179 	if (error)
    180 		return error;
    181 	if (bp->b_resid != 0) {
    182 		panic("dq2get: %s quota file corrupted",
    183 		    lfs_quotatypes[type]);
    184 	}
    185 	*q2ep = (void *)((char *)bp->b_data + blkoffset);
    186 	*bpp = bp;
    187 	return 0;
    188 }
    189 
    190 /* walk a quota entry list, calling the callback for each entry */
    191 #define Q2WL_ABORT 0x10000000
    192 
    193 static int
    194 quota2_walk_list(struct ulfsmount *ump, struct buf *hbp, int type,
    195     uint64_t *offp, int flags, void *a,
    196     int (*func)(struct ulfsmount *, uint64_t *, struct quota2_entry *, uint64_t, void *))
    197 {
    198 	struct lfs *fs = ump->um_lfs;
    199 	const int needswap = ULFS_MPNEEDSWAP(fs);
    200 	daddr_t off = ulfs_rw64(*offp, needswap);
    201 	struct buf *bp, *obp = hbp;
    202 	int ret = 0, ret2 = 0;
    203 	struct quota2_entry *q2e;
    204 	daddr_t lblkno, blkoff, olblkno = 0;
    205 
    206 	KASSERT(mutex_owned(&lfs_dqlock));
    207 
    208 	while (off != 0) {
    209 		lblkno = (off >> ump->um_mountp->mnt_fs_bshift);
    210 		blkoff = (off & ump->umq2_bmask);
    211 		if (lblkno == 0) {
    212 			/* in the header block */
    213 			bp = hbp;
    214 		} else if (lblkno == olblkno) {
    215 			/* still in the same buf */
    216 			bp = obp;
    217 		} else {
    218 			ret = bread(ump->um_quotas[type], lblkno,
    219 			    ump->umq2_bsize, flags, &bp);
    220 			if (ret)
    221 				return ret;
    222 			if (bp->b_resid != 0) {
    223 				panic("quota2_walk_list: %s quota file corrupted",
    224 				    lfs_quotatypes[type]);
    225 			}
    226 		}
    227 		q2e = (void *)((char *)(bp->b_data) + blkoff);
    228 		ret = (*func)(ump, offp, q2e, off, a);
    229 		if (off != ulfs_rw64(*offp, needswap)) {
    230 			/* callback changed parent's pointer, redo */
    231 			off = ulfs_rw64(*offp, needswap);
    232 			if (bp != hbp && bp != obp)
    233 				ret2 = bwrite(bp);
    234 		} else {
    235 			/* parent if now current */
    236 			if (obp != bp && obp != hbp) {
    237 				if (flags & B_MODIFY)
    238 					ret2 = bwrite(obp);
    239 				else
    240 					brelse(obp, 0);
    241 			}
    242 			obp = bp;
    243 			olblkno = lblkno;
    244 			offp = &(q2e->q2e_next);
    245 			off = ulfs_rw64(*offp, needswap);
    246 		}
    247 		if (ret)
    248 			break;
    249 		if (ret2) {
    250 			ret = ret2;
    251 			break;
    252 		}
    253 	}
    254 	if (obp != hbp) {
    255 		if (flags & B_MODIFY)
    256 			ret2 = bwrite(obp);
    257 		else
    258 			brelse(obp, 0);
    259 	}
    260 	if (ret & Q2WL_ABORT)
    261 		return 0;
    262 	if (ret == 0)
    263 		return ret2;
    264 	return ret;
    265 }
    266 
    267 int
    268 lfsquota2_umount(struct mount *mp, int flags)
    269 {
    270 	int i, error;
    271 	struct ulfsmount *ump = VFSTOULFS(mp);
    272 	struct lfs *fs = ump->um_lfs;
    273 
    274 	if ((fs->um_flags & ULFS_QUOTA2) == 0)
    275 		return 0;
    276 
    277 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    278 		if (ump->um_quotas[i] != NULLVP) {
    279 			error = vn_close(ump->um_quotas[i], FREAD|FWRITE,
    280 			    ump->um_cred[i]);
    281 			if (error) {
    282 				printf("quota2_umount failed: close(%p) %d\n",
    283 				    ump->um_quotas[i], error);
    284 				return error;
    285 			}
    286 		}
    287 		ump->um_quotas[i] = NULLVP;
    288 	}
    289 	return 0;
    290 }
    291 
    292 static int
    293 quota2_q2ealloc(struct ulfsmount *ump, int type, uid_t uid, struct dquot *dq)
    294 {
    295 	int error, error2;
    296 	struct buf *hbp, *bp;
    297 	struct quota2_header *q2h;
    298 	struct quota2_entry *q2e;
    299 	daddr_t offset;
    300 	u_long hash_mask;
    301 	struct lfs *fs = ump->um_lfs;
    302 	const int needswap = ULFS_MPNEEDSWAP(fs);
    303 
    304 	KASSERT(mutex_owned(&dq->dq_interlock));
    305 	KASSERT(mutex_owned(&lfs_dqlock));
    306 	error = getq2h(ump, type, &hbp, &q2h, B_MODIFY);
    307 	if (error)
    308 		return error;
    309 	offset = ulfs_rw64(q2h->q2h_free, needswap);
    310 	if (offset == 0) {
    311 		struct vnode *vp = ump->um_quotas[type];
    312 		struct inode *ip = VTOI(vp);
    313 		uint64_t size = ip->i_size;
    314 		/* need to allocate a new disk block */
    315 		error = lfs_balloc(vp, size, ump->umq2_bsize,
    316 		    ump->um_cred[type], B_CLRBUF | B_SYNC, &bp);
    317 		if (error) {
    318 			brelse(hbp, 0);
    319 			return error;
    320 		}
    321 		KASSERT((ip->i_size % ump->umq2_bsize) == 0);
    322 		ip->i_size += ump->umq2_bsize;
    323 		DIP_ASSIGN(ip, size, ip->i_size);
    324 		ip->i_state |= IN_CHANGE | IN_UPDATE;
    325 		uvm_vnp_setsize(vp, ip->i_size);
    326 		lfsquota2_addfreeq2e(q2h, bp->b_data, size, ump->umq2_bsize,
    327 		    needswap);
    328 		error = bwrite(bp);
    329 		error2 = lfs_update(vp, NULL, NULL, UPDATE_WAIT);
    330 		if (error || error2) {
    331 			brelse(hbp, 0);
    332 			if (error)
    333 				return error;
    334 			return error2;
    335 		}
    336 		offset = ulfs_rw64(q2h->q2h_free, needswap);
    337 		KASSERT(offset != 0);
    338 	}
    339 	dq->dq2_lblkno = (offset >> ump->um_mountp->mnt_fs_bshift);
    340 	dq->dq2_blkoff = (offset & ump->umq2_bmask);
    341 	if (dq->dq2_lblkno == 0) {
    342 		bp = hbp;
    343 		q2e = (void *)((char *)bp->b_data + dq->dq2_blkoff);
    344 	} else {
    345 		error = getq2e(ump, type, dq->dq2_lblkno,
    346 		    dq->dq2_blkoff, &bp, &q2e, B_MODIFY);
    347 		if (error) {
    348 			brelse(hbp, 0);
    349 			return error;
    350 		}
    351 	}
    352 	hash_mask = ((1 << q2h->q2h_hash_shift) - 1);
    353 	/* remove from free list */
    354 	q2h->q2h_free = q2e->q2e_next;
    355 
    356 	memcpy(q2e, &q2h->q2h_defentry, sizeof(*q2e));
    357 	q2e->q2e_uid = ulfs_rw32(uid, needswap);
    358 	/* insert in hash list */
    359 	q2e->q2e_next = q2h->q2h_entries[uid & hash_mask];
    360 	q2h->q2h_entries[uid & hash_mask] = ulfs_rw64(offset, needswap);
    361 	if (hbp != bp) {
    362 		bwrite(hbp);
    363 	}
    364 	bwrite(bp);
    365 	return 0;
    366 }
    367 
    368 static int
    369 getinoquota2(struct inode *ip, bool alloc, bool modify, struct buf **bpp,
    370     struct quota2_entry **q2ep)
    371 {
    372 	int error;
    373 	int i;
    374 	struct dquot *dq;
    375 	struct ulfsmount *ump = ip->i_ump;
    376 	u_int32_t ino_ids[ULFS_MAXQUOTAS];
    377 
    378 	error = lfs_getinoquota(ip);
    379 	if (error)
    380 		return error;
    381 
    382         ino_ids[ULFS_USRQUOTA] = ip->i_uid;
    383         ino_ids[ULFS_GRPQUOTA] = ip->i_gid;
    384 	/* first get the interlock for all dquot */
    385 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    386 		dq = ip->i_dquot[i];
    387 		if (dq == NODQUOT)
    388 			continue;
    389 		mutex_enter(&dq->dq_interlock);
    390 	}
    391 	/* now get the corresponding quota entry */
    392 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    393 		bpp[i] = NULL;
    394 		q2ep[i] = NULL;
    395 		dq = ip->i_dquot[i];
    396 		if (dq == NODQUOT)
    397 			continue;
    398 		if (__predict_false(ump->um_quotas[i] == NULL)) {
    399 			/*
    400 			 * quotas have been turned off. This can happen
    401 			 * at umount time.
    402 			 */
    403 			mutex_exit(&dq->dq_interlock);
    404 			lfs_dqrele(NULLVP, dq);
    405 			ip->i_dquot[i] = NULL;
    406 			continue;
    407 		}
    408 
    409 		if ((dq->dq2_lblkno | dq->dq2_blkoff) == 0) {
    410 			if (!alloc) {
    411 				continue;
    412 			}
    413 			/* need to alloc a new on-disk quot */
    414 			mutex_enter(&lfs_dqlock);
    415 			error = quota2_q2ealloc(ump, i, ino_ids[i], dq);
    416 			mutex_exit(&lfs_dqlock);
    417 			if (error)
    418 				return error;
    419 		}
    420 		KASSERT(dq->dq2_lblkno != 0 || dq->dq2_blkoff != 0);
    421 		error = getq2e(ump, i, dq->dq2_lblkno,
    422 		    dq->dq2_blkoff, &bpp[i], &q2ep[i],
    423 		    modify ? B_MODIFY : 0);
    424 		if (error)
    425 			return error;
    426 	}
    427 	return 0;
    428 }
    429 
    430 __inline static int __unused
    431 lfsquota2_check_limit(struct quota2_val *q2v, uint64_t change, time_t now)
    432 {
    433 	return lfsquota_check_limit(q2v->q2v_cur, change, q2v->q2v_softlimit,
    434 	    q2v->q2v_hardlimit, q2v->q2v_time, now);
    435 }
    436 
    437 static int
    438 quota2_check(struct inode *ip, int vtype, int64_t change, kauth_cred_t cred,
    439     int flags)
    440 {
    441 	int error;
    442 	struct buf *bp[ULFS_MAXQUOTAS];
    443 	struct quota2_entry *q2e[ULFS_MAXQUOTAS];
    444 	struct quota2_val *q2vp;
    445 	struct dquot *dq;
    446 	uint64_t ncurblks;
    447 	struct ulfsmount *ump = ip->i_ump;
    448 	struct lfs *fs = ip->i_lfs;
    449 	struct mount *mp = ump->um_mountp;
    450 	const int needswap = ULFS_MPNEEDSWAP(fs);
    451 	int i;
    452 
    453 	if ((error = getinoquota2(ip, change > 0, change != 0, bp, q2e)) != 0)
    454 		return error;
    455 	if (change == 0) {
    456 		for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    457 			dq = ip->i_dquot[i];
    458 			if (dq == NODQUOT)
    459 				continue;
    460 			if (bp[i])
    461 				brelse(bp[i], 0);
    462 			mutex_exit(&dq->dq_interlock);
    463 		}
    464 		return 0;
    465 	}
    466 	if (change < 0) {
    467 		for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    468 			dq = ip->i_dquot[i];
    469 			if (dq == NODQUOT)
    470 				continue;
    471 			if (q2e[i] == NULL) {
    472 				mutex_exit(&dq->dq_interlock);
    473 				continue;
    474 			}
    475 			q2vp = &q2e[i]->q2e_val[vtype];
    476 			ncurblks = ulfs_rw64(q2vp->q2v_cur, needswap);
    477 			if (ncurblks < -change)
    478 				ncurblks = 0;
    479 			else
    480 				ncurblks += change;
    481 			q2vp->q2v_cur = ulfs_rw64(ncurblks, needswap);
    482 			quota2_bwrite(mp, bp[i]);
    483 			mutex_exit(&dq->dq_interlock);
    484 		}
    485 		return 0;
    486 	}
    487 	/* see if the allocation is allowed */
    488 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    489 		struct quota2_val q2v;
    490 		int ql_stat;
    491 		dq = ip->i_dquot[i];
    492 		if (dq == NODQUOT)
    493 			continue;
    494 		KASSERT(q2e[i] != NULL);
    495 		lfsquota2_ulfs_rwq2v(&q2e[i]->q2e_val[vtype], &q2v, needswap);
    496 		ql_stat = lfsquota2_check_limit(&q2v, change, time_second);
    497 
    498 		if ((flags & FORCE) == 0 &&
    499 		    kauth_authorize_system(cred, KAUTH_SYSTEM_FS_QUOTA,
    500 		    KAUTH_REQ_SYSTEM_FS_QUOTA_NOLIMIT,
    501 		    KAUTH_ARG(i), KAUTH_ARG(vtype), NULL) != 0) {
    502 			/* enforce this limit */
    503 			switch(QL_STATUS(ql_stat)) {
    504 			case QL_S_DENY_HARD:
    505 				if ((dq->dq_flags & DQ_WARN(vtype)) == 0) {
    506 					uprintf("\n%s: write failed, %s %s "
    507 					    "limit reached\n",
    508 					    mp->mnt_stat.f_mntonname,
    509 					    lfs_quotatypes[i], limnames[vtype]);
    510 					dq->dq_flags |= DQ_WARN(vtype);
    511 				}
    512 				error = EDQUOT;
    513 				break;
    514 			case QL_S_DENY_GRACE:
    515 				if ((dq->dq_flags & DQ_WARN(vtype)) == 0) {
    516 					uprintf("\n%s: write failed, %s %s "
    517 					    "limit reached\n",
    518 					    mp->mnt_stat.f_mntonname,
    519 					    lfs_quotatypes[i], limnames[vtype]);
    520 					dq->dq_flags |= DQ_WARN(vtype);
    521 				}
    522 				error = EDQUOT;
    523 				break;
    524 			case QL_S_ALLOW_SOFT:
    525 				if ((dq->dq_flags & DQ_WARN(vtype)) == 0) {
    526 					uprintf("\n%s: warning, %s %s "
    527 					    "quota exceeded\n",
    528 					    mp->mnt_stat.f_mntonname,
    529 					    lfs_quotatypes[i], limnames[vtype]);
    530 					dq->dq_flags |= DQ_WARN(vtype);
    531 				}
    532 				break;
    533 			}
    534 		}
    535 		/*
    536 		 * always do this; we don't know if the allocation will
    537 		 * succed or not in the end. if we don't do the allocation
    538 		 * q2v_time will be ignored anyway
    539 		 */
    540 		if (ql_stat & QL_F_CROSS) {
    541 			q2v.q2v_time = time_second + q2v.q2v_grace;
    542 			lfsquota2_ulfs_rwq2v(&q2v, &q2e[i]->q2e_val[vtype],
    543 			    needswap);
    544 		}
    545 	}
    546 
    547 	/* now do the allocation if allowed */
    548 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    549 		dq = ip->i_dquot[i];
    550 		if (dq == NODQUOT)
    551 			continue;
    552 		KASSERT(q2e[i] != NULL);
    553 		if (error == 0) {
    554 			q2vp = &q2e[i]->q2e_val[vtype];
    555 			ncurblks = ulfs_rw64(q2vp->q2v_cur, needswap);
    556 			q2vp->q2v_cur = ulfs_rw64(ncurblks + change, needswap);
    557 			quota2_bwrite(mp, bp[i]);
    558 		} else
    559 			brelse(bp[i], 0);
    560 		mutex_exit(&dq->dq_interlock);
    561 	}
    562 	return error;
    563 }
    564 
    565 int
    566 lfs_chkdq2(struct inode *ip, int64_t change, kauth_cred_t cred, int flags)
    567 {
    568 	return quota2_check(ip, QL_BLOCK, change, cred, flags);
    569 }
    570 
    571 int
    572 lfs_chkiq2(struct inode *ip, int32_t change, kauth_cred_t cred, int flags)
    573 {
    574 	return quota2_check(ip, QL_FILE, change, cred, flags);
    575 }
    576 
    577 int
    578 lfsquota2_handle_cmd_put(struct ulfsmount *ump, const struct quotakey *key,
    579     const struct quotaval *val)
    580 {
    581 	int error;
    582 	struct dquot *dq;
    583 	struct quota2_header *q2h;
    584 	struct quota2_entry q2e, *q2ep;
    585 	struct buf *bp;
    586 	struct lfs *fs = ump->um_lfs;
    587 	const int needswap = ULFS_MPNEEDSWAP(fs);
    588 
    589 	/* make sure we can index by the fs-independent idtype */
    590 	CTASSERT(QUOTA_IDTYPE_USER == ULFS_USRQUOTA);
    591 	CTASSERT(QUOTA_IDTYPE_GROUP == ULFS_GRPQUOTA);
    592 
    593 	if (ump->um_quotas[key->qk_idtype] == NULLVP)
    594 		return ENODEV;
    595 
    596 	if (key->qk_id == QUOTA_DEFAULTID) {
    597 		mutex_enter(&lfs_dqlock);
    598 		error = getq2h(ump, key->qk_idtype, &bp, &q2h, B_MODIFY);
    599 		if (error) {
    600 			mutex_exit(&lfs_dqlock);
    601 			goto out_error;
    602 		}
    603 		lfsquota2_ulfs_rwq2e(&q2h->q2h_defentry, &q2e, needswap);
    604 		quota2_dict_update_q2e_limits(key->qk_objtype, val, &q2e);
    605 		lfsquota2_ulfs_rwq2e(&q2e, &q2h->q2h_defentry, needswap);
    606 		mutex_exit(&lfs_dqlock);
    607 		quota2_bwrite(ump->um_mountp, bp);
    608 		goto out_error;
    609 	}
    610 
    611 	error = lfs_dqget(NULLVP, key->qk_id, ump, key->qk_idtype, &dq);
    612 	if (error)
    613 		goto out_error;
    614 
    615 	mutex_enter(&dq->dq_interlock);
    616 	if (dq->dq2_lblkno == 0 && dq->dq2_blkoff == 0) {
    617 		/* need to alloc a new on-disk quot */
    618 		mutex_enter(&lfs_dqlock);
    619 		error = quota2_q2ealloc(ump, key->qk_idtype, key->qk_id, dq);
    620 		mutex_exit(&lfs_dqlock);
    621 		if (error)
    622 			goto out_il;
    623 	}
    624 	KASSERT(dq->dq2_lblkno != 0 || dq->dq2_blkoff != 0);
    625 	error = getq2e(ump, key->qk_idtype, dq->dq2_lblkno,
    626 	    dq->dq2_blkoff, &bp, &q2ep, B_MODIFY);
    627 	if (error)
    628 		goto out_il;
    629 
    630 	lfsquota2_ulfs_rwq2e(q2ep, &q2e, needswap);
    631 	/*
    632 	 * Reset time limit if previously had no soft limit or were
    633 	 * under it, but now have a soft limit and are over it.
    634 	 */
    635 	if (val->qv_softlimit &&
    636 	    q2e.q2e_val[key->qk_objtype].q2v_cur >= val->qv_softlimit &&
    637 	    (q2e.q2e_val[key->qk_objtype].q2v_softlimit == 0 ||
    638 	     q2e.q2e_val[key->qk_objtype].q2v_cur < q2e.q2e_val[key->qk_objtype].q2v_softlimit))
    639 		q2e.q2e_val[key->qk_objtype].q2v_time = time_second + val->qv_grace;
    640 	quota2_dict_update_q2e_limits(key->qk_objtype, val, &q2e);
    641 	lfsquota2_ulfs_rwq2e(&q2e, q2ep, needswap);
    642 	quota2_bwrite(ump->um_mountp, bp);
    643 
    644 out_il:
    645 	mutex_exit(&dq->dq_interlock);
    646 	lfs_dqrele(NULLVP, dq);
    647 out_error:
    648 	return error;
    649 }
    650 
    651 struct dq2clear_callback {
    652 	uid_t id;
    653 	struct dquot *dq;
    654 	struct quota2_header *q2h;
    655 };
    656 
    657 static int
    658 dq2clear_callback(struct ulfsmount *ump, uint64_t *offp, struct quota2_entry *q2e,
    659     uint64_t off, void *v)
    660 {
    661 	struct dq2clear_callback *c = v;
    662 	struct lfs *fs = ump->um_lfs;
    663 	const int needswap = ULFS_MPNEEDSWAP(fs);
    664 	uint64_t myoff;
    665 
    666 	if (ulfs_rw32(q2e->q2e_uid, needswap) == c->id) {
    667 		KASSERT(mutex_owned(&c->dq->dq_interlock));
    668 		c->dq->dq2_lblkno = 0;
    669 		c->dq->dq2_blkoff = 0;
    670 		myoff = *offp;
    671 		/* remove from hash list */
    672 		*offp = q2e->q2e_next;
    673 		/* add to free list */
    674 		q2e->q2e_next = c->q2h->q2h_free;
    675 		c->q2h->q2h_free = myoff;
    676 		return Q2WL_ABORT;
    677 	}
    678 	return 0;
    679 }
    680 int
    681 lfsquota2_handle_cmd_del(struct ulfsmount *ump, const struct quotakey *qk)
    682 {
    683 	int idtype;
    684 	id_t id;
    685 	int objtype;
    686 	int error, i, canfree;
    687 	struct dquot *dq;
    688 	struct quota2_header *q2h;
    689 	struct quota2_entry q2e, *q2ep;
    690 	struct buf *hbp, *bp;
    691 	u_long hash_mask;
    692 	struct dq2clear_callback c;
    693 
    694 	idtype = qk->qk_idtype;
    695 	id = qk->qk_id;
    696 	objtype = qk->qk_objtype;
    697 
    698 	if (ump->um_quotas[idtype] == NULLVP)
    699 		return ENODEV;
    700 	if (id == QUOTA_DEFAULTID)
    701 		return EOPNOTSUPP;
    702 
    703 	/* get the default entry before locking the entry's buffer */
    704 	mutex_enter(&lfs_dqlock);
    705 	error = getq2h(ump, idtype, &hbp, &q2h, 0);
    706 	if (error) {
    707 		mutex_exit(&lfs_dqlock);
    708 		return error;
    709 	}
    710 	/* we'll copy to another disk entry, so no need to swap */
    711 	memcpy(&q2e, &q2h->q2h_defentry, sizeof(q2e));
    712 	mutex_exit(&lfs_dqlock);
    713 	brelse(hbp, 0);
    714 
    715 	error = lfs_dqget(NULLVP, id, ump, idtype, &dq);
    716 	if (error)
    717 		return error;
    718 
    719 	mutex_enter(&dq->dq_interlock);
    720 	if (dq->dq2_lblkno == 0 && dq->dq2_blkoff == 0) {
    721 		/* already clear, nothing to do */
    722 		error = ENOENT;
    723 		goto out_il;
    724 	}
    725 
    726 	error = getq2e(ump, idtype, dq->dq2_lblkno, dq->dq2_blkoff,
    727 	    &bp, &q2ep, B_MODIFY);
    728 	if (error)
    729 		goto out_error;
    730 
    731 	/* make sure we can index by the objtype passed in */
    732 	CTASSERT(QUOTA_OBJTYPE_BLOCKS == QL_BLOCK);
    733 	CTASSERT(QUOTA_OBJTYPE_FILES == QL_FILE);
    734 
    735 	/* clear the requested objtype by copying from the default entry */
    736 	q2ep->q2e_val[objtype].q2v_softlimit =
    737 		q2e.q2e_val[objtype].q2v_softlimit;
    738 	q2ep->q2e_val[objtype].q2v_hardlimit =
    739 		q2e.q2e_val[objtype].q2v_hardlimit;
    740 	q2ep->q2e_val[objtype].q2v_grace =
    741 		q2e.q2e_val[objtype].q2v_grace;
    742 	q2ep->q2e_val[objtype].q2v_time = 0;
    743 
    744 	/* if this entry now contains no information, we can free it */
    745 	canfree = 1;
    746 	for (i = 0; i < N_QL; i++) {
    747 		if (q2ep->q2e_val[i].q2v_cur != 0 ||
    748 		    (q2ep->q2e_val[i].q2v_softlimit !=
    749 		     q2e.q2e_val[i].q2v_softlimit) ||
    750 		    (q2ep->q2e_val[i].q2v_hardlimit !=
    751 		     q2e.q2e_val[i].q2v_hardlimit) ||
    752 		    (q2ep->q2e_val[i].q2v_grace !=
    753 		     q2e.q2e_val[i].q2v_grace)) {
    754 			canfree = 0;
    755 			break;
    756 		}
    757 		/* note: do not need to check q2v_time */
    758 	}
    759 
    760 	if (canfree == 0) {
    761 		quota2_bwrite(ump->um_mountp, bp);
    762 		goto out_error;
    763 	}
    764 	/* we can free it. release bp so we can walk the list */
    765 	brelse(bp, 0);
    766 	mutex_enter(&lfs_dqlock);
    767 	error = getq2h(ump, idtype, &hbp, &q2h, 0);
    768 	if (error)
    769 		goto out_dqlock;
    770 
    771 	hash_mask = ((1 << q2h->q2h_hash_shift) - 1);
    772 	c.dq = dq;
    773 	c.id = id;
    774 	c.q2h = q2h;
    775 	error = quota2_walk_list(ump, hbp, idtype,
    776 	    &q2h->q2h_entries[id & hash_mask], B_MODIFY, &c,
    777 	    dq2clear_callback);
    778 
    779 	bwrite(hbp);
    780 
    781 out_dqlock:
    782 	mutex_exit(&lfs_dqlock);
    783 out_error:
    784 out_il:
    785 	mutex_exit(&dq->dq_interlock);
    786 	lfs_dqrele(NULLVP, dq);
    787 	return error;
    788 }
    789 
    790 static int
    791 quota2_fetch_q2e(struct ulfsmount *ump, const struct quotakey *qk,
    792     struct quota2_entry *ret)
    793 {
    794 	struct dquot *dq;
    795 	int error;
    796 	struct quota2_entry *q2ep;
    797 	struct buf *bp;
    798 	struct lfs *fs = ump->um_lfs;
    799 	const int needswap = ULFS_MPNEEDSWAP(fs);
    800 
    801 	error = lfs_dqget(NULLVP, qk->qk_id, ump, qk->qk_idtype, &dq);
    802 	if (error)
    803 		return error;
    804 
    805 	mutex_enter(&dq->dq_interlock);
    806 	if (dq->dq2_lblkno == 0 && dq->dq2_blkoff == 0) {
    807 		mutex_exit(&dq->dq_interlock);
    808 		lfs_dqrele(NULLVP, dq);
    809 		return ENOENT;
    810 	}
    811 	error = getq2e(ump, qk->qk_idtype, dq->dq2_lblkno, dq->dq2_blkoff,
    812 	    &bp, &q2ep, 0);
    813 	if (error) {
    814 		mutex_exit(&dq->dq_interlock);
    815 		lfs_dqrele(NULLVP, dq);
    816 		return error;
    817 	}
    818 	lfsquota2_ulfs_rwq2e(q2ep, ret, needswap);
    819 	brelse(bp, 0);
    820 	mutex_exit(&dq->dq_interlock);
    821 	lfs_dqrele(NULLVP, dq);
    822 
    823 	return 0;
    824 }
    825 
    826 static int
    827 quota2_fetch_quotaval(struct ulfsmount *ump, const struct quotakey *qk,
    828     struct quotaval *ret)
    829 {
    830 	struct dquot *dq;
    831 	int error;
    832 	struct quota2_entry *q2ep, q2e;
    833 	struct buf  *bp;
    834 	struct lfs *fs = ump->um_lfs;
    835 	const int needswap = ULFS_MPNEEDSWAP(fs);
    836 	id_t id2;
    837 
    838 	error = lfs_dqget(NULLVP, qk->qk_id, ump, qk->qk_idtype, &dq);
    839 	if (error)
    840 		return error;
    841 
    842 	mutex_enter(&dq->dq_interlock);
    843 	if (dq->dq2_lblkno == 0 && dq->dq2_blkoff == 0) {
    844 		mutex_exit(&dq->dq_interlock);
    845 		lfs_dqrele(NULLVP, dq);
    846 		return ENOENT;
    847 	}
    848 	error = getq2e(ump, qk->qk_idtype, dq->dq2_lblkno, dq->dq2_blkoff,
    849 	    &bp, &q2ep, 0);
    850 	if (error) {
    851 		mutex_exit(&dq->dq_interlock);
    852 		lfs_dqrele(NULLVP, dq);
    853 		return error;
    854 	}
    855 	lfsquota2_ulfs_rwq2e(q2ep, &q2e, needswap);
    856 	brelse(bp, 0);
    857 	mutex_exit(&dq->dq_interlock);
    858 	lfs_dqrele(NULLVP, dq);
    859 
    860 	q2e_to_quotaval(&q2e, 0, &id2, qk->qk_objtype, ret);
    861 	KASSERT(id2 == qk->qk_id);
    862 	return 0;
    863 }
    864 
    865 int
    866 lfsquota2_handle_cmd_get(struct ulfsmount *ump, const struct quotakey *qk,
    867     struct quotaval *qv)
    868 {
    869 	int error;
    870 	struct quota2_header *q2h;
    871 	struct quota2_entry q2e;
    872 	struct buf *bp;
    873 	struct lfs *fs = ump->um_lfs;
    874 	const int needswap = ULFS_MPNEEDSWAP(fs);
    875 	id_t id2;
    876 
    877 	/*
    878 	 * Make sure the FS-independent codes match the internal ones,
    879 	 * so we can use the passed-in objtype without having to
    880 	 * convert it explicitly to QL_BLOCK/QL_FILE.
    881 	 */
    882 	CTASSERT(QL_BLOCK == QUOTA_OBJTYPE_BLOCKS);
    883 	CTASSERT(QL_FILE == QUOTA_OBJTYPE_FILES);
    884 	CTASSERT(N_QL == 2);
    885 
    886 	if (qk->qk_objtype < 0 || qk->qk_objtype >= N_QL) {
    887 		return EINVAL;
    888 	}
    889 
    890 	if (ump->um_quotas[qk->qk_idtype] == NULLVP)
    891 		return ENODEV;
    892 	if (qk->qk_id == QUOTA_DEFAULTID) {
    893 		mutex_enter(&lfs_dqlock);
    894 		error = getq2h(ump, qk->qk_idtype, &bp, &q2h, 0);
    895 		if (error) {
    896 			mutex_exit(&lfs_dqlock);
    897 			return error;
    898 		}
    899 		lfsquota2_ulfs_rwq2e(&q2h->q2h_defentry, &q2e, needswap);
    900 		mutex_exit(&lfs_dqlock);
    901 		brelse(bp, 0);
    902 		q2e_to_quotaval(&q2e, qk->qk_id == QUOTA_DEFAULTID, &id2,
    903 				qk->qk_objtype, qv);
    904 		(void)id2;
    905 	} else
    906 		error = quota2_fetch_quotaval(ump, qk, qv);
    907 
    908 	return error;
    909 }
    910 
    911 /*
    912  * Cursor structure we used.
    913  *
    914  * This will get stored in userland between calls so we must not assume
    915  * it isn't arbitrarily corrupted.
    916  */
    917 struct ulfsq2_cursor {
    918 	uint32_t q2c_magic;	/* magic number */
    919 	int q2c_hashsize;	/* size of hash table at last go */
    920 
    921 	int q2c_users_done;	/* true if we've returned all user data */
    922 	int q2c_groups_done;	/* true if we've returned all group data */
    923 	int q2c_defaults_done;	/* true if we've returned the default values */
    924 	int q2c_hashpos;	/* slot to start at in hash table */
    925 	int q2c_uidpos;		/* number of ids we've handled */
    926 	int q2c_blocks_done;	/* true if we've returned the blocks value */
    927 };
    928 
    929 /*
    930  * State of a single cursorget call, or at least the part of it that
    931  * needs to be passed around.
    932  */
    933 struct q2cursor_state {
    934 	/* data return pointers */
    935 	struct quotakey *keys;
    936 	struct quotaval *vals;
    937 
    938 	/* key/value counters */
    939 	unsigned maxkeyvals;
    940 	unsigned numkeys;	/* number of keys assigned */
    941 
    942 	/* ID to key/value conversion state */
    943 	int skipfirst;		/* if true skip first key/value */
    944 	int skiplast;		/* if true skip last key/value */
    945 
    946 	/* ID counters */
    947 	unsigned maxids;	/* maximum number of IDs to handle */
    948 	unsigned numids;	/* number of IDs handled */
    949 };
    950 
    951 /*
    952  * Additional structure for getids callback.
    953  */
    954 struct q2cursor_getids {
    955 	struct q2cursor_state *state;
    956 	int idtype;
    957 	unsigned skip;		/* number of ids to skip over */
    958 	unsigned new_skip;	/* number of ids to skip over next time */
    959 	unsigned skipped;	/* number skipped so far */
    960 	int stopped;		/* true if we stopped quota_walk_list early */
    961 };
    962 
    963 /*
    964  * Cursor-related functions
    965  */
    966 
    967 /* magic number */
    968 #define Q2C_MAGIC (0xbeebe111)
    969 
    970 /* extract cursor from caller form */
    971 #define Q2CURSOR(qkc) ((struct ulfsq2_cursor *)&qkc->u.qkc_space[0])
    972 
    973 /*
    974  * Check that a cursor we're handed is something like valid. If
    975  * someone munges it and it still passes these checks, they'll get
    976  * partial or odd results back but won't break anything.
    977  */
    978 static int
    979 q2cursor_check(struct ulfsq2_cursor *cursor)
    980 {
    981 	if (cursor->q2c_magic != Q2C_MAGIC) {
    982 		return EINVAL;
    983 	}
    984 	if (cursor->q2c_hashsize < 0) {
    985 		return EINVAL;
    986 	}
    987 
    988 	if (cursor->q2c_users_done != 0 && cursor->q2c_users_done != 1) {
    989 		return EINVAL;
    990 	}
    991 	if (cursor->q2c_groups_done != 0 && cursor->q2c_groups_done != 1) {
    992 		return EINVAL;
    993 	}
    994 	if (cursor->q2c_defaults_done != 0 && cursor->q2c_defaults_done != 1) {
    995 		return EINVAL;
    996 	}
    997 	if (cursor->q2c_hashpos < 0 || cursor->q2c_uidpos < 0) {
    998 		return EINVAL;
    999 	}
   1000 	if (cursor->q2c_blocks_done != 0 && cursor->q2c_blocks_done != 1) {
   1001 		return EINVAL;
   1002 	}
   1003 	return 0;
   1004 }
   1005 
   1006 /*
   1007  * Set up the q2cursor state.
   1008  */
   1009 static void
   1010 q2cursor_initstate(struct q2cursor_state *state, struct quotakey *keys,
   1011     struct quotaval *vals, unsigned maxkeyvals, int blocks_done)
   1012 {
   1013 	state->keys = keys;
   1014 	state->vals = vals;
   1015 
   1016 	state->maxkeyvals = maxkeyvals;
   1017 	state->numkeys = 0;
   1018 
   1019 	/*
   1020 	 * For each ID there are two quotavals to return. If the
   1021 	 * maximum number of entries to return is odd, we might want
   1022 	 * to skip the first quotaval of the first ID, or the last
   1023 	 * quotaval of the last ID, but not both. So the number of IDs
   1024 	 * we want is (up to) half the number of return slots we have,
   1025 	 * rounded up.
   1026 	 */
   1027 
   1028 	state->maxids = (state->maxkeyvals + 1) / 2;
   1029 	state->numids = 0;
   1030 	if (state->maxkeyvals % 2) {
   1031 		if (blocks_done) {
   1032 			state->skipfirst = 1;
   1033 			state->skiplast = 0;
   1034 		} else {
   1035 			state->skipfirst = 0;
   1036 			state->skiplast = 1;
   1037 		}
   1038 	} else {
   1039 		state->skipfirst = 0;
   1040 		state->skiplast = 0;
   1041 	}
   1042 }
   1043 
   1044 /*
   1045  * Choose which idtype we're going to work on. If doing a full
   1046  * iteration, we do users first, then groups, but either might be
   1047  * disabled or marked to skip via cursorsetidtype(), so don't make
   1048  * silly assumptions.
   1049  */
   1050 static int
   1051 q2cursor_pickidtype(struct ulfsq2_cursor *cursor, int *idtype_ret)
   1052 {
   1053 	if (cursor->q2c_users_done == 0) {
   1054 		*idtype_ret = QUOTA_IDTYPE_USER;
   1055 	} else if (cursor->q2c_groups_done == 0) {
   1056 		*idtype_ret = QUOTA_IDTYPE_GROUP;
   1057 	} else {
   1058 		return EAGAIN;
   1059 	}
   1060 	return 0;
   1061 }
   1062 
   1063 /*
   1064  * Add an ID to the current state. Sets up either one or two keys to
   1065  * refer to it, depending on whether it's first/last and the setting
   1066  * of skipfirst. (skiplast does not need to be explicitly tested)
   1067  */
   1068 static void
   1069 q2cursor_addid(struct q2cursor_state *state, int idtype, id_t id)
   1070 {
   1071 	KASSERT(state->numids < state->maxids);
   1072 	KASSERT(state->numkeys < state->maxkeyvals);
   1073 
   1074 	if (!state->skipfirst || state->numkeys > 0) {
   1075 		state->keys[state->numkeys].qk_idtype = idtype;
   1076 		state->keys[state->numkeys].qk_id = id;
   1077 		state->keys[state->numkeys].qk_objtype = QUOTA_OBJTYPE_BLOCKS;
   1078 		state->numkeys++;
   1079 	}
   1080 	if (state->numkeys < state->maxkeyvals) {
   1081 		state->keys[state->numkeys].qk_idtype = idtype;
   1082 		state->keys[state->numkeys].qk_id = id;
   1083 		state->keys[state->numkeys].qk_objtype = QUOTA_OBJTYPE_FILES;
   1084 		state->numkeys++;
   1085 	} else {
   1086 		KASSERT(state->skiplast);
   1087 	}
   1088 	state->numids++;
   1089 }
   1090 
   1091 /*
   1092  * Callback function for getting IDs. Update counting and call addid.
   1093  */
   1094 static int
   1095 q2cursor_getids_callback(struct ulfsmount *ump, uint64_t *offp,
   1096     struct quota2_entry *q2ep, uint64_t off, void *v)
   1097 {
   1098 	struct q2cursor_getids *gi = v;
   1099 	id_t id;
   1100 	struct lfs *fs = ump->um_lfs;
   1101 	const int needswap = ULFS_MPNEEDSWAP(fs);
   1102 
   1103 	if (gi->skipped < gi->skip) {
   1104 		gi->skipped++;
   1105 		return 0;
   1106 	}
   1107 	id = ulfs_rw32(q2ep->q2e_uid, needswap);
   1108 	q2cursor_addid(gi->state, gi->idtype, id);
   1109 	gi->new_skip++;
   1110 	if (gi->state->numids >= gi->state->maxids) {
   1111 		/* got enough ids, stop now */
   1112 		gi->stopped = 1;
   1113 		return Q2WL_ABORT;
   1114 	}
   1115 	return 0;
   1116 }
   1117 
   1118 /*
   1119  * Fill in a batch of quotakeys by scanning one or more hash chains.
   1120  */
   1121 static int
   1122 q2cursor_getkeys(struct ulfsmount *ump, int idtype, struct ulfsq2_cursor *cursor,
   1123     struct q2cursor_state *state,
   1124     int *hashsize_ret, struct quota2_entry *default_q2e_ret)
   1125 {
   1126 	struct lfs *fs = ump->um_lfs;
   1127 	const int needswap = ULFS_MPNEEDSWAP(fs);
   1128 	struct buf *hbp;
   1129 	struct quota2_header *q2h;
   1130 	int quota2_hash_size;
   1131 	struct q2cursor_getids gi;
   1132 	uint64_t offset;
   1133 	int error;
   1134 
   1135 	/*
   1136 	 * Read the header block.
   1137 	 */
   1138 
   1139 	mutex_enter(&lfs_dqlock);
   1140 	error = getq2h(ump, idtype, &hbp, &q2h, 0);
   1141 	if (error) {
   1142 		mutex_exit(&lfs_dqlock);
   1143 		return error;
   1144 	}
   1145 
   1146 	/* if the table size has changed, make the caller start over */
   1147 	quota2_hash_size = ulfs_rw16(q2h->q2h_hash_size, needswap);
   1148 	if (cursor->q2c_hashsize == 0) {
   1149 		cursor->q2c_hashsize = quota2_hash_size;
   1150 	} else if (cursor->q2c_hashsize != quota2_hash_size) {
   1151 		error = EDEADLK;
   1152 		goto scanfail;
   1153 	}
   1154 
   1155 	/* grab the entry with the default values out of the header */
   1156 	lfsquota2_ulfs_rwq2e(&q2h->q2h_defentry, default_q2e_ret, needswap);
   1157 
   1158 	/* If we haven't done the defaults yet, that goes first. */
   1159 	if (cursor->q2c_defaults_done == 0) {
   1160 		q2cursor_addid(state, idtype, QUOTA_DEFAULTID);
   1161 		/* if we read both halves, mark it done */
   1162 		if (state->numids < state->maxids || !state->skiplast) {
   1163 			cursor->q2c_defaults_done = 1;
   1164 		}
   1165 	}
   1166 
   1167 	gi.state = state;
   1168 	gi.idtype = idtype;
   1169 
   1170 	while (state->numids < state->maxids) {
   1171 		if (cursor->q2c_hashpos >= quota2_hash_size) {
   1172 			/* nothing more left */
   1173 			break;
   1174 		}
   1175 
   1176 		/* scan this hash chain */
   1177 		gi.skip = cursor->q2c_uidpos;
   1178 		gi.new_skip = gi.skip;
   1179 		gi.skipped = 0;
   1180 		gi.stopped = 0;
   1181 		offset = q2h->q2h_entries[cursor->q2c_hashpos];
   1182 
   1183 		error = quota2_walk_list(ump, hbp, idtype, &offset, 0, &gi,
   1184 		    q2cursor_getids_callback);
   1185 		KASSERT(error != Q2WL_ABORT);
   1186 		if (error) {
   1187 			break;
   1188 		}
   1189 		if (gi.stopped) {
   1190 			/* callback stopped before reading whole chain */
   1191 			cursor->q2c_uidpos = gi.new_skip;
   1192 			/* if we didn't get both halves, back up */
   1193 			if (state->numids == state->maxids && state->skiplast){
   1194 				KASSERT(cursor->q2c_uidpos > 0);
   1195 				cursor->q2c_uidpos--;
   1196 			}
   1197 		} else {
   1198 			/* read whole chain */
   1199 			/* if we got both halves of the last id, advance */
   1200 			if (state->numids < state->maxids || !state->skiplast){
   1201 				cursor->q2c_uidpos = 0;
   1202 				cursor->q2c_hashpos++;
   1203 			}
   1204 		}
   1205 	}
   1206 
   1207 scanfail:
   1208 	mutex_exit(&lfs_dqlock);
   1209 	brelse(hbp, 0);
   1210 	if (error)
   1211 		return error;
   1212 
   1213 	*hashsize_ret = quota2_hash_size;
   1214 	return 0;
   1215 }
   1216 
   1217 /*
   1218  * Fetch the quotavals for the quotakeys.
   1219  */
   1220 static int
   1221 q2cursor_getvals(struct ulfsmount *ump, struct q2cursor_state *state,
   1222     const struct quota2_entry *default_q2e)
   1223 {
   1224 	int hasid;
   1225 	id_t loadedid, id;
   1226 	unsigned pos;
   1227 	struct quota2_entry q2e;
   1228 	int objtype;
   1229 	int error;
   1230 
   1231 	hasid = 0;
   1232 	loadedid = 0;
   1233 	for (pos = 0; pos < state->numkeys; pos++) {
   1234 		id = state->keys[pos].qk_id;
   1235 		if (!hasid || id != loadedid) {
   1236 			hasid = 1;
   1237 			loadedid = id;
   1238 			if (id == QUOTA_DEFAULTID) {
   1239 				q2e = *default_q2e;
   1240 			} else {
   1241 				error = quota2_fetch_q2e(ump,
   1242 							 &state->keys[pos],
   1243 							 &q2e);
   1244 				if (error == ENOENT) {
   1245 					/* something changed - start over */
   1246 					error = EDEADLK;
   1247 				}
   1248 				if (error) {
   1249 					return error;
   1250 				}
   1251  			}
   1252 		}
   1253 
   1254 
   1255 		objtype = state->keys[pos].qk_objtype;
   1256 		KASSERT(objtype >= 0 && objtype < N_QL);
   1257 		q2val_to_quotaval(&q2e.q2e_val[objtype], &state->vals[pos]);
   1258 	}
   1259 
   1260 	return 0;
   1261 }
   1262 
   1263 /*
   1264  * Handle cursorget.
   1265  *
   1266  * We can't just read keys and values directly, because we can't walk
   1267  * the list with qdlock and grab dq_interlock to read the entries at
   1268  * the same time. So we're going to do two passes: one to figure out
   1269  * which IDs we want and fill in the keys, and then a second to use
   1270  * the keys to fetch the values.
   1271  */
   1272 int
   1273 lfsquota2_handle_cmd_cursorget(struct ulfsmount *ump, struct quotakcursor *qkc,
   1274     struct quotakey *keys, struct quotaval *vals, unsigned maxreturn,
   1275     unsigned *ret)
   1276 {
   1277 	int error;
   1278 	struct ulfsq2_cursor *cursor;
   1279 	struct ulfsq2_cursor newcursor;
   1280 	struct q2cursor_state state;
   1281 	struct quota2_entry default_q2e;
   1282 	int idtype;
   1283 	int quota2_hash_size = 0; /* XXXuninit */
   1284 
   1285 	/*
   1286 	 * Convert and validate the cursor.
   1287 	 */
   1288 	cursor = Q2CURSOR(qkc);
   1289 	error = q2cursor_check(cursor);
   1290 	if (error) {
   1291 		return error;
   1292 	}
   1293 
   1294 	/*
   1295 	 * Make sure our on-disk codes match the values of the
   1296 	 * FS-independent ones. This avoids the need for explicit
   1297 	 * conversion (which would be a NOP anyway and thus easily
   1298 	 * left out or called in the wrong places...)
   1299 	 */
   1300 	CTASSERT(QUOTA_IDTYPE_USER == ULFS_USRQUOTA);
   1301 	CTASSERT(QUOTA_IDTYPE_GROUP == ULFS_GRPQUOTA);
   1302 	CTASSERT(QUOTA_OBJTYPE_BLOCKS == QL_BLOCK);
   1303 	CTASSERT(QUOTA_OBJTYPE_FILES == QL_FILE);
   1304 
   1305 	/*
   1306 	 * If some of the idtypes aren't configured/enabled, arrange
   1307 	 * to skip over them.
   1308 	 */
   1309 	if (cursor->q2c_users_done == 0 &&
   1310 	    ump->um_quotas[ULFS_USRQUOTA] == NULLVP) {
   1311 		cursor->q2c_users_done = 1;
   1312 	}
   1313 	if (cursor->q2c_groups_done == 0 &&
   1314 	    ump->um_quotas[ULFS_GRPQUOTA] == NULLVP) {
   1315 		cursor->q2c_groups_done = 1;
   1316 	}
   1317 
   1318 	/* Loop over, potentially, both idtypes */
   1319 	while (1) {
   1320 
   1321 		/* Choose id type */
   1322 		error = q2cursor_pickidtype(cursor, &idtype);
   1323 		if (error == EAGAIN) {
   1324 			/* nothing more to do, return 0 */
   1325 			*ret = 0;
   1326 			return 0;
   1327 		}
   1328 		KASSERT(ump->um_quotas[idtype] != NULLVP);
   1329 
   1330 		/*
   1331 		 * Initialize the per-call iteration state. Copy the
   1332 		 * cursor state so we can update it in place but back
   1333 		 * out on error.
   1334 		 */
   1335 		q2cursor_initstate(&state, keys, vals, maxreturn,
   1336 				   cursor->q2c_blocks_done);
   1337 		newcursor = *cursor;
   1338 
   1339 		/* Assign keys */
   1340 		error = q2cursor_getkeys(ump, idtype, &newcursor, &state,
   1341 					 &quota2_hash_size, &default_q2e);
   1342 		if (error) {
   1343 			return error;
   1344 		}
   1345 
   1346 		/* Now fill in the values. */
   1347 		error = q2cursor_getvals(ump, &state, &default_q2e);
   1348 		if (error) {
   1349 			return error;
   1350 		}
   1351 
   1352 		/*
   1353 		 * Now that we aren't going to fail and lose what we
   1354 		 * did so far, we can update the cursor state.
   1355 		 */
   1356 
   1357 		if (newcursor.q2c_hashpos >= quota2_hash_size) {
   1358 			if (idtype == QUOTA_IDTYPE_USER)
   1359 				cursor->q2c_users_done = 1;
   1360 			else
   1361 				cursor->q2c_groups_done = 1;
   1362 
   1363 			/* start over on another id type */
   1364 			cursor->q2c_hashsize = 0;
   1365 			cursor->q2c_defaults_done = 0;
   1366 			cursor->q2c_hashpos = 0;
   1367 			cursor->q2c_uidpos = 0;
   1368 			cursor->q2c_blocks_done = 0;
   1369 		} else {
   1370 			*cursor = newcursor;
   1371 			cursor->q2c_blocks_done = state.skiplast;
   1372 		}
   1373 
   1374 		/*
   1375 		 * If we have something to return, return it.
   1376 		 * Otherwise, continue to the other idtype, if any,
   1377 		 * and only return zero at end of iteration.
   1378 		 */
   1379 		if (state.numkeys > 0) {
   1380 			break;
   1381 		}
   1382 	}
   1383 
   1384 	*ret = state.numkeys;
   1385 	return 0;
   1386 }
   1387 
   1388 int
   1389 lfsquota2_handle_cmd_cursoropen(struct ulfsmount *ump, struct quotakcursor *qkc)
   1390 {
   1391 	struct ulfsq2_cursor *cursor;
   1392 
   1393 	CTASSERT(sizeof(*cursor) <= sizeof(qkc->u.qkc_space));
   1394 	cursor = Q2CURSOR(qkc);
   1395 
   1396 	cursor->q2c_magic = Q2C_MAGIC;
   1397 	cursor->q2c_hashsize = 0;
   1398 
   1399 	cursor->q2c_users_done = 0;
   1400 	cursor->q2c_groups_done = 0;
   1401 	cursor->q2c_defaults_done = 0;
   1402 	cursor->q2c_hashpos = 0;
   1403 	cursor->q2c_uidpos = 0;
   1404 	cursor->q2c_blocks_done = 0;
   1405 	return 0;
   1406 }
   1407 
   1408 int
   1409 lfsquota2_handle_cmd_cursorclose(struct ulfsmount *ump, struct quotakcursor *qkc)
   1410 {
   1411 	struct ulfsq2_cursor *cursor;
   1412 	int error;
   1413 
   1414 	cursor = Q2CURSOR(qkc);
   1415 	error = q2cursor_check(cursor);
   1416 	if (error) {
   1417 		return error;
   1418 	}
   1419 
   1420 	/* nothing to do */
   1421 
   1422 	return 0;
   1423 }
   1424 
   1425 int
   1426 lfsquota2_handle_cmd_cursorskipidtype(struct ulfsmount *ump,
   1427     struct quotakcursor *qkc, int idtype)
   1428 {
   1429 	struct ulfsq2_cursor *cursor;
   1430 	int error;
   1431 
   1432 	cursor = Q2CURSOR(qkc);
   1433 	error = q2cursor_check(cursor);
   1434 	if (error) {
   1435 		return error;
   1436 	}
   1437 
   1438 	switch (idtype) {
   1439 	    case QUOTA_IDTYPE_USER:
   1440 		cursor->q2c_users_done = 1;
   1441 		break;
   1442 	    case QUOTA_IDTYPE_GROUP:
   1443 		cursor->q2c_groups_done = 1;
   1444 		break;
   1445 	    default:
   1446 		return EINVAL;
   1447 	}
   1448 
   1449 	return 0;
   1450 }
   1451 
   1452 int
   1453 lfsquota2_handle_cmd_cursoratend(struct ulfsmount *ump, struct quotakcursor *qkc,
   1454     int *ret)
   1455 {
   1456 	struct ulfsq2_cursor *cursor;
   1457 	int error;
   1458 
   1459 	cursor = Q2CURSOR(qkc);
   1460 	error = q2cursor_check(cursor);
   1461 	if (error) {
   1462 		return error;
   1463 	}
   1464 
   1465 	*ret = (cursor->q2c_users_done && cursor->q2c_groups_done);
   1466 	return 0;
   1467 }
   1468 
   1469 int
   1470 lfsquota2_handle_cmd_cursorrewind(struct ulfsmount *ump, struct quotakcursor *qkc)
   1471 {
   1472 	struct ulfsq2_cursor *cursor;
   1473 	int error;
   1474 
   1475 	cursor = Q2CURSOR(qkc);
   1476 	error = q2cursor_check(cursor);
   1477 	if (error) {
   1478 		return error;
   1479 	}
   1480 
   1481 	cursor->q2c_hashsize = 0;
   1482 
   1483 	cursor->q2c_users_done = 0;
   1484 	cursor->q2c_groups_done = 0;
   1485 	cursor->q2c_defaults_done = 0;
   1486 	cursor->q2c_hashpos = 0;
   1487 	cursor->q2c_uidpos = 0;
   1488 	cursor->q2c_blocks_done = 0;
   1489 
   1490 	return 0;
   1491 }
   1492 
   1493 int
   1494 lfs_q2sync(struct mount *mp)
   1495 {
   1496 	return 0;
   1497 }
   1498 
   1499 struct dq2get_callback {
   1500 	uid_t id;
   1501 	struct dquot *dq;
   1502 };
   1503 
   1504 static int
   1505 dq2get_callback(struct ulfsmount *ump, uint64_t *offp, struct quota2_entry *q2e,
   1506     uint64_t off, void *v)
   1507 {
   1508 	struct dq2get_callback *c = v;
   1509 	daddr_t lblkno;
   1510 	int blkoff;
   1511 	struct lfs *fs = ump->um_lfs;
   1512 	const int needswap = ULFS_MPNEEDSWAP(fs);
   1513 
   1514 	if (ulfs_rw32(q2e->q2e_uid, needswap) == c->id) {
   1515 		KASSERT(mutex_owned(&c->dq->dq_interlock));
   1516 		lblkno = (off >> ump->um_mountp->mnt_fs_bshift);
   1517 		blkoff = (off & ump->umq2_bmask);
   1518 		c->dq->dq2_lblkno = lblkno;
   1519 		c->dq->dq2_blkoff = blkoff;
   1520 		return Q2WL_ABORT;
   1521 	}
   1522 	return 0;
   1523 }
   1524 
   1525 int
   1526 lfs_dq2get(struct vnode *dqvp, u_long id, struct ulfsmount *ump, int type,
   1527     struct dquot *dq)
   1528 {
   1529 	struct buf *bp;
   1530 	struct quota2_header *q2h;
   1531 	int error;
   1532 	daddr_t offset;
   1533 	u_long hash_mask;
   1534 	struct dq2get_callback c = {
   1535 		.id = id,
   1536 		.dq = dq
   1537 	};
   1538 
   1539 	KASSERT(mutex_owned(&dq->dq_interlock));
   1540 	mutex_enter(&lfs_dqlock);
   1541 	error = getq2h(ump, type, &bp, &q2h, 0);
   1542 	if (error)
   1543 		goto out_mutex;
   1544 	/* look for our entry */
   1545 	hash_mask = ((1 << q2h->q2h_hash_shift) - 1);
   1546 	offset = q2h->q2h_entries[id & hash_mask];
   1547 	error = quota2_walk_list(ump, bp, type, &offset, 0, (void *)&c,
   1548 	    dq2get_callback);
   1549 	brelse(bp, 0);
   1550 out_mutex:
   1551 	mutex_exit(&lfs_dqlock);
   1552 	return error;
   1553 }
   1554 
   1555 int
   1556 lfs_dq2sync(struct vnode *vp, struct dquot *dq)
   1557 {
   1558 	return 0;
   1559 }
   1560 
   1561 int
   1562 lfs_quota2_mount(struct mount *mp)
   1563 {
   1564 	struct ulfsmount *ump = VFSTOULFS(mp);
   1565 	struct lfs *fs = ump->um_lfs;
   1566 	int error;
   1567 	struct vnode *vp;
   1568 	struct lwp *l = curlwp;
   1569 
   1570 	if ((fs->lfs_use_quota2) == 0)
   1571 		return 0;
   1572 
   1573 	fs->um_flags |= ULFS_QUOTA2;
   1574 	ump->umq2_bsize = lfs_sb_getbsize(fs);
   1575 	ump->umq2_bmask = lfs_sb_getbmask(fs);
   1576 	if (fs->lfs_quota_magic != Q2_HEAD_MAGIC) {
   1577 		printf("%s: Invalid quota magic number\n",
   1578 		    mp->mnt_stat.f_mntonname);
   1579 		return EINVAL;
   1580 	}
   1581 
   1582 	error = 0;
   1583         if ((fs->lfs_quota_flags & FS_Q2_DO_TYPE(ULFS_USRQUOTA)) &&
   1584             fs->lfs_quotaino[ULFS_USRQUOTA] == 0) {
   1585                 printf("%s: No user quota inode\n",
   1586 		    mp->mnt_stat.f_mntonname);
   1587                 error = EINVAL;
   1588         }
   1589         if ((fs->lfs_quota_flags & FS_Q2_DO_TYPE(ULFS_GRPQUOTA)) &&
   1590             fs->lfs_quotaino[ULFS_GRPQUOTA] == 0) {
   1591                 printf("%s: No group quota inode\n",
   1592 		    mp->mnt_stat.f_mntonname);
   1593                 error = EINVAL;
   1594         }
   1595 	if (error)
   1596 		return error;
   1597 
   1598         if (fs->lfs_quota_flags & FS_Q2_DO_TYPE(ULFS_USRQUOTA) &&
   1599 	    ump->um_quotas[ULFS_USRQUOTA] == NULLVP) {
   1600 		error = VFS_VGET(mp, fs->lfs_quotaino[ULFS_USRQUOTA],
   1601 		    LK_EXCLUSIVE, &vp);
   1602 		if (error) {
   1603 			printf("%s: can't vget() user quota inode: %d\n",
   1604 			    mp->mnt_stat.f_mntonname, error);
   1605 			return error;
   1606 		}
   1607 		ump->um_quotas[ULFS_USRQUOTA] = vp;
   1608 		ump->um_cred[ULFS_USRQUOTA] = l->l_cred;
   1609 		mutex_enter(vp->v_interlock);
   1610 		vp->v_writecount++;
   1611 		mutex_exit(vp->v_interlock);
   1612 		VOP_UNLOCK(vp);
   1613 	}
   1614         if (fs->lfs_quota_flags & FS_Q2_DO_TYPE(ULFS_GRPQUOTA) &&
   1615 	    ump->um_quotas[ULFS_GRPQUOTA] == NULLVP) {
   1616 		error = VFS_VGET(mp, fs->lfs_quotaino[ULFS_GRPQUOTA],
   1617 		    LK_EXCLUSIVE, &vp);
   1618 		if (error) {
   1619 			vn_close(ump->um_quotas[ULFS_USRQUOTA],
   1620 			    FREAD|FWRITE, l->l_cred);
   1621 			printf("%s: can't vget() group quota inode: %d\n",
   1622 			    mp->mnt_stat.f_mntonname, error);
   1623 			return error;
   1624 		}
   1625 		ump->um_quotas[ULFS_GRPQUOTA] = vp;
   1626 		ump->um_cred[ULFS_GRPQUOTA] = l->l_cred;
   1627 		mutex_enter(vp->v_interlock);
   1628 		vp->v_vflag |= VV_SYSTEM;
   1629 		vp->v_writecount++;
   1630 		mutex_exit(vp->v_interlock);
   1631 		VOP_UNLOCK(vp);
   1632 	}
   1633 
   1634 	mp->mnt_flag |= MNT_QUOTA;
   1635 	return 0;
   1636 }
   1637