Home | History | Annotate | Line # | Download | only in lfs
ulfs_quota2.c revision 1.10
      1 /*	$NetBSD: ulfs_quota2.c,v 1.10 2013/07/28 01:22:55 dholland Exp $	*/
      2 /*  from NetBSD: ufs_quota2.c,v 1.35 2012/09/27 07:47:56 bouyer Exp  */
      3 /*  from NetBSD: ffs_quota2.c,v 1.4 2011/06/12 03:36:00 rmind 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 1.10 2013/07/28 01:22:55 dholland 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/fstrans.h>
     44 #include <sys/kauth.h>
     45 #include <sys/wapbl.h>
     46 #include <sys/quota.h>
     47 #include <sys/quotactl.h>
     48 
     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  * grabing 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 #ifdef LFS_EI
    146 	struct lfs *fs = ump->um_lfs;
    147 	const int needswap = ULFS_MPNEEDSWAP(fs);
    148 #endif
    149 	int error;
    150 	struct buf *bp;
    151 	struct quota2_header *q2h;
    152 
    153 	KASSERT(mutex_owned(&lfs_dqlock));
    154 	error = bread(ump->um_quotas[type], 0, ump->umq2_bsize,
    155 	    ump->um_cred[type], flags, &bp);
    156 	if (error)
    157 		return error;
    158 	if (bp->b_resid != 0)
    159 		panic("dq2get: %s quota file truncated", lfs_quotatypes[type]);
    160 
    161 	q2h = (void *)bp->b_data;
    162 	if (ulfs_rw32(q2h->q2h_magic_number, needswap) != Q2_HEAD_MAGIC ||
    163 	    q2h->q2h_type != type)
    164 		panic("dq2get: corrupted %s quota header", lfs_quotatypes[type]);
    165 	*bpp = bp;
    166 	*q2hp = q2h;
    167 	return 0;
    168 }
    169 
    170 static int
    171 getq2e(struct ulfsmount *ump, int type, daddr_t lblkno, int blkoffset,
    172     struct buf **bpp, struct quota2_entry **q2ep, int flags)
    173 {
    174 	int error;
    175 	struct buf *bp;
    176 
    177 	if (blkoffset & (sizeof(uint64_t) - 1)) {
    178 		panic("dq2get: %s quota file corrupted",
    179 		    lfs_quotatypes[type]);
    180 	}
    181 	error = bread(ump->um_quotas[type], lblkno, ump->umq2_bsize,
    182 	    ump->um_cred[type], flags, &bp);
    183 	if (error)
    184 		return error;
    185 	if (bp->b_resid != 0) {
    186 		panic("dq2get: %s quota file corrupted",
    187 		    lfs_quotatypes[type]);
    188 	}
    189 	*q2ep = (void *)((char *)bp->b_data + blkoffset);
    190 	*bpp = bp;
    191 	return 0;
    192 }
    193 
    194 /* walk a quota entry list, calling the callback for each entry */
    195 #define Q2WL_ABORT 0x10000000
    196 
    197 static int
    198 quota2_walk_list(struct ulfsmount *ump, struct buf *hbp, int type,
    199     uint64_t *offp, int flags, void *a,
    200     int (*func)(struct ulfsmount *, uint64_t *, struct quota2_entry *, uint64_t, void *))
    201 {
    202 #ifdef LFS_EI
    203 	struct lfs *fs = ump->um_lfs;
    204 	const int needswap = ULFS_MPNEEDSWAP(fs);
    205 #endif
    206 	daddr_t off = ulfs_rw64(*offp, needswap);
    207 	struct buf *bp, *obp = hbp;
    208 	int ret = 0, ret2 = 0;
    209 	struct quota2_entry *q2e;
    210 	daddr_t lblkno, blkoff, olblkno = 0;
    211 
    212 	KASSERT(mutex_owner(&lfs_dqlock));
    213 
    214 	while (off != 0) {
    215 		lblkno = (off >> ump->um_mountp->mnt_fs_bshift);
    216 		blkoff = (off & ump->umq2_bmask);
    217 		if (lblkno == 0) {
    218 			/* in the header block */
    219 			bp = hbp;
    220 		} else if (lblkno == olblkno) {
    221 			/* still in the same buf */
    222 			bp = obp;
    223 		} else {
    224 			ret = bread(ump->um_quotas[type], lblkno,
    225 			    ump->umq2_bsize,
    226 			    ump->um_cred[type], flags, &bp);
    227 			if (ret)
    228 				return ret;
    229 			if (bp->b_resid != 0) {
    230 				panic("quota2_walk_list: %s quota file corrupted",
    231 				    lfs_quotatypes[type]);
    232 			}
    233 		}
    234 		q2e = (void *)((char *)(bp->b_data) + blkoff);
    235 		ret = (*func)(ump, offp, q2e, off, a);
    236 		if (off != ulfs_rw64(*offp, needswap)) {
    237 			/* callback changed parent's pointer, redo */
    238 			off = ulfs_rw64(*offp, needswap);
    239 			if (bp != hbp && bp != obp)
    240 				ret2 = bwrite(bp);
    241 		} else {
    242 			/* parent if now current */
    243 			if (obp != bp && obp != hbp) {
    244 				if (flags & B_MODIFY)
    245 					ret2 = bwrite(obp);
    246 				else
    247 					brelse(obp, 0);
    248 			}
    249 			obp = bp;
    250 			olblkno = lblkno;
    251 			offp = &(q2e->q2e_next);
    252 			off = ulfs_rw64(*offp, needswap);
    253 		}
    254 		if (ret)
    255 			break;
    256 		if (ret2) {
    257 			ret = ret2;
    258 			break;
    259 		}
    260 	}
    261 	if (obp != hbp) {
    262 		if (flags & B_MODIFY)
    263 			ret2 = bwrite(obp);
    264 		else
    265 			brelse(obp, 0);
    266 	}
    267 	if (ret & Q2WL_ABORT)
    268 		return 0;
    269 	if (ret == 0)
    270 		return ret2;
    271 	return ret;
    272 }
    273 
    274 int
    275 lfsquota2_umount(struct mount *mp, int flags)
    276 {
    277 	int i, error;
    278 	struct ulfsmount *ump = VFSTOULFS(mp);
    279 	struct lfs *fs = ump->um_lfs;
    280 
    281 	if ((fs->um_flags & ULFS_QUOTA2) == 0)
    282 		return 0;
    283 
    284 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    285 		if (ump->um_quotas[i] != NULLVP) {
    286 			error = vn_close(ump->um_quotas[i], FREAD|FWRITE,
    287 			    ump->um_cred[i]);
    288 			if (error) {
    289 				printf("quota2_umount failed: close(%p) %d\n",
    290 				    ump->um_quotas[i], error);
    291 				return error;
    292 			}
    293 		}
    294 		ump->um_quotas[i] = NULLVP;
    295 	}
    296 	return 0;
    297 }
    298 
    299 static int
    300 quota2_q2ealloc(struct ulfsmount *ump, int type, uid_t uid, struct dquot *dq)
    301 {
    302 	int error, error2;
    303 	struct buf *hbp, *bp;
    304 	struct quota2_header *q2h;
    305 	struct quota2_entry *q2e;
    306 	daddr_t offset;
    307 	u_long hash_mask;
    308 	struct lfs *fs = ump->um_lfs;
    309 	const int needswap = ULFS_MPNEEDSWAP(fs);
    310 
    311 	KASSERT(mutex_owned(&dq->dq_interlock));
    312 	KASSERT(mutex_owned(&lfs_dqlock));
    313 	error = getq2h(ump, type, &hbp, &q2h, B_MODIFY);
    314 	if (error)
    315 		return error;
    316 	offset = ulfs_rw64(q2h->q2h_free, needswap);
    317 	if (offset == 0) {
    318 		struct vnode *vp = ump->um_quotas[type];
    319 		struct inode *ip = VTOI(vp);
    320 		uint64_t size = ip->i_size;
    321 		/* need to alocate a new disk block */
    322 		error = lfs_balloc(vp, size, ump->umq2_bsize,
    323 		    ump->um_cred[type], B_CLRBUF | B_SYNC, &bp);
    324 		if (error) {
    325 			brelse(hbp, 0);
    326 			return error;
    327 		}
    328 		KASSERT((ip->i_size % ump->umq2_bsize) == 0);
    329 		ip->i_size += ump->umq2_bsize;
    330 		DIP_ASSIGN(ip, size, ip->i_size);
    331 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    332 		uvm_vnp_setsize(vp, ip->i_size);
    333 		lfsquota2_addfreeq2e(q2h, bp->b_data, size, ump->umq2_bsize,
    334 		    needswap);
    335 		error = bwrite(bp);
    336 		error2 = lfs_update(vp, NULL, NULL, UPDATE_WAIT);
    337 		if (error || error2) {
    338 			brelse(hbp, 0);
    339 			if (error)
    340 				return error;
    341 			return error2;
    342 		}
    343 		offset = ulfs_rw64(q2h->q2h_free, needswap);
    344 		KASSERT(offset != 0);
    345 	}
    346 	dq->dq2_lblkno = (offset >> ump->um_mountp->mnt_fs_bshift);
    347 	dq->dq2_blkoff = (offset & ump->umq2_bmask);
    348 	if (dq->dq2_lblkno == 0) {
    349 		bp = hbp;
    350 		q2e = (void *)((char *)bp->b_data + dq->dq2_blkoff);
    351 	} else {
    352 		error = getq2e(ump, type, dq->dq2_lblkno,
    353 		    dq->dq2_blkoff, &bp, &q2e, B_MODIFY);
    354 		if (error) {
    355 			brelse(hbp, 0);
    356 			return error;
    357 		}
    358 	}
    359 	hash_mask = ((1 << q2h->q2h_hash_shift) - 1);
    360 	/* remove from free list */
    361 	q2h->q2h_free = q2e->q2e_next;
    362 
    363 	memcpy(q2e, &q2h->q2h_defentry, sizeof(*q2e));
    364 	q2e->q2e_uid = ulfs_rw32(uid, needswap);
    365 	/* insert in hash list */
    366 	q2e->q2e_next = q2h->q2h_entries[uid & hash_mask];
    367 	q2h->q2h_entries[uid & hash_mask] = ulfs_rw64(offset, needswap);
    368 	if (hbp != bp) {
    369 		bwrite(hbp);
    370 	}
    371 	bwrite(bp);
    372 	return 0;
    373 }
    374 
    375 static int
    376 getinoquota2(struct inode *ip, bool alloc, bool modify, struct buf **bpp,
    377     struct quota2_entry **q2ep)
    378 {
    379 	int error;
    380 	int i;
    381 	struct dquot *dq;
    382 	struct ulfsmount *ump = ip->i_ump;
    383 	u_int32_t ino_ids[ULFS_MAXQUOTAS];
    384 
    385 	error = lfs_getinoquota(ip);
    386 	if (error)
    387 		return error;
    388 
    389         ino_ids[ULFS_USRQUOTA] = ip->i_uid;
    390         ino_ids[ULFS_GRPQUOTA] = ip->i_gid;
    391 	/* first get the interlock for all dquot */
    392 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    393 		dq = ip->i_dquot[i];
    394 		if (dq == NODQUOT)
    395 			continue;
    396 		mutex_enter(&dq->dq_interlock);
    397 	}
    398 	/* now get the corresponding quota entry */
    399 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    400 		bpp[i] = NULL;
    401 		q2ep[i] = NULL;
    402 		dq = ip->i_dquot[i];
    403 		if (dq == NODQUOT)
    404 			continue;
    405 		if (__predict_false(ump->um_quotas[i] == NULL)) {
    406 			/*
    407 			 * quotas have been turned off. This can happen
    408 			 * at umount time.
    409 			 */
    410 			mutex_exit(&dq->dq_interlock);
    411 			lfs_dqrele(NULLVP, dq);
    412 			ip->i_dquot[i] = NULL;
    413 			continue;
    414 		}
    415 
    416 		if ((dq->dq2_lblkno | dq->dq2_blkoff) == 0) {
    417 			if (!alloc) {
    418 				continue;
    419 			}
    420 			/* need to alloc a new on-disk quot */
    421 			mutex_enter(&lfs_dqlock);
    422 			error = quota2_q2ealloc(ump, i, ino_ids[i], dq);
    423 			mutex_exit(&lfs_dqlock);
    424 			if (error)
    425 				return error;
    426 		}
    427 		KASSERT(dq->dq2_lblkno != 0 || dq->dq2_blkoff != 0);
    428 		error = getq2e(ump, i, dq->dq2_lblkno,
    429 		    dq->dq2_blkoff, &bpp[i], &q2ep[i],
    430 		    modify ? B_MODIFY : 0);
    431 		if (error)
    432 			return error;
    433 	}
    434 	return 0;
    435 }
    436 
    437 __inline static int __unused
    438 lfsquota2_check_limit(struct quota2_val *q2v, uint64_t change, time_t now)
    439 {
    440 	return lfsquota_check_limit(q2v->q2v_cur, change, q2v->q2v_softlimit,
    441 	    q2v->q2v_hardlimit, q2v->q2v_time, now);
    442 }
    443 
    444 static int
    445 quota2_check(struct inode *ip, int vtype, int64_t change, kauth_cred_t cred,
    446     int flags)
    447 {
    448 	int error;
    449 	struct buf *bp[ULFS_MAXQUOTAS];
    450 	struct quota2_entry *q2e[ULFS_MAXQUOTAS];
    451 	struct quota2_val *q2vp;
    452 	struct dquot *dq;
    453 	uint64_t ncurblks;
    454 	struct ulfsmount *ump = ip->i_ump;
    455 	struct lfs *fs = ip->i_lfs;
    456 	struct mount *mp = ump->um_mountp;
    457 	const int needswap = ULFS_MPNEEDSWAP(fs);
    458 	int i;
    459 
    460 	if ((error = getinoquota2(ip, change > 0, change != 0, bp, q2e)) != 0)
    461 		return error;
    462 	if (change == 0) {
    463 		for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    464 			dq = ip->i_dquot[i];
    465 			if (dq == NODQUOT)
    466 				continue;
    467 			if (bp[i])
    468 				brelse(bp[i], 0);
    469 			mutex_exit(&dq->dq_interlock);
    470 		}
    471 		return 0;
    472 	}
    473 	if (change < 0) {
    474 		for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    475 			dq = ip->i_dquot[i];
    476 			if (dq == NODQUOT)
    477 				continue;
    478 			if (q2e[i] == NULL) {
    479 				mutex_exit(&dq->dq_interlock);
    480 				continue;
    481 			}
    482 			q2vp = &q2e[i]->q2e_val[vtype];
    483 			ncurblks = ulfs_rw64(q2vp->q2v_cur, needswap);
    484 			if (ncurblks < -change)
    485 				ncurblks = 0;
    486 			else
    487 				ncurblks += change;
    488 			q2vp->q2v_cur = ulfs_rw64(ncurblks, needswap);
    489 			quota2_bwrite(mp, bp[i]);
    490 			mutex_exit(&dq->dq_interlock);
    491 		}
    492 		return 0;
    493 	}
    494 	/* see if the allocation is allowed */
    495 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    496 		struct quota2_val q2v;
    497 		int ql_stat;
    498 		dq = ip->i_dquot[i];
    499 		if (dq == NODQUOT)
    500 			continue;
    501 		KASSERT(q2e[i] != NULL);
    502 		lfsquota2_ulfs_rwq2v(&q2e[i]->q2e_val[vtype], &q2v, needswap);
    503 		ql_stat = lfsquota2_check_limit(&q2v, change, time_second);
    504 
    505 		if ((flags & FORCE) == 0 &&
    506 		    kauth_authorize_system(cred, KAUTH_SYSTEM_FS_QUOTA,
    507 		    KAUTH_REQ_SYSTEM_FS_QUOTA_NOLIMIT,
    508 		    KAUTH_ARG(i), KAUTH_ARG(vtype), NULL) != 0) {
    509 			/* enforce this limit */
    510 			switch(QL_STATUS(ql_stat)) {
    511 			case QL_S_DENY_HARD:
    512 				if ((dq->dq_flags & DQ_WARN(vtype)) == 0) {
    513 					uprintf("\n%s: write failed, %s %s "
    514 					    "limit reached\n",
    515 					    mp->mnt_stat.f_mntonname,
    516 					    lfs_quotatypes[i], limnames[vtype]);
    517 					dq->dq_flags |= DQ_WARN(vtype);
    518 				}
    519 				error = EDQUOT;
    520 				break;
    521 			case QL_S_DENY_GRACE:
    522 				if ((dq->dq_flags & DQ_WARN(vtype)) == 0) {
    523 					uprintf("\n%s: write failed, %s %s "
    524 					    "limit reached\n",
    525 					    mp->mnt_stat.f_mntonname,
    526 					    lfs_quotatypes[i], limnames[vtype]);
    527 					dq->dq_flags |= DQ_WARN(vtype);
    528 				}
    529 				error = EDQUOT;
    530 				break;
    531 			case QL_S_ALLOW_SOFT:
    532 				if ((dq->dq_flags & DQ_WARN(vtype)) == 0) {
    533 					uprintf("\n%s: warning, %s %s "
    534 					    "quota exceeded\n",
    535 					    mp->mnt_stat.f_mntonname,
    536 					    lfs_quotatypes[i], limnames[vtype]);
    537 					dq->dq_flags |= DQ_WARN(vtype);
    538 				}
    539 				break;
    540 			}
    541 		}
    542 		/*
    543 		 * always do this; we don't know if the allocation will
    544 		 * succed or not in the end. if we don't do the allocation
    545 		 * q2v_time will be ignored anyway
    546 		 */
    547 		if (ql_stat & QL_F_CROSS) {
    548 			q2v.q2v_time = time_second + q2v.q2v_grace;
    549 			lfsquota2_ulfs_rwq2v(&q2v, &q2e[i]->q2e_val[vtype],
    550 			    needswap);
    551 		}
    552 	}
    553 
    554 	/* now do the allocation if allowed */
    555 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    556 		dq = ip->i_dquot[i];
    557 		if (dq == NODQUOT)
    558 			continue;
    559 		KASSERT(q2e[i] != NULL);
    560 		if (error == 0) {
    561 			q2vp = &q2e[i]->q2e_val[vtype];
    562 			ncurblks = ulfs_rw64(q2vp->q2v_cur, needswap);
    563 			q2vp->q2v_cur = ulfs_rw64(ncurblks + change, needswap);
    564 			quota2_bwrite(mp, bp[i]);
    565 		} else
    566 			brelse(bp[i], 0);
    567 		mutex_exit(&dq->dq_interlock);
    568 	}
    569 	return error;
    570 }
    571 
    572 int
    573 lfs_chkdq2(struct inode *ip, int64_t change, kauth_cred_t cred, int flags)
    574 {
    575 	return quota2_check(ip, QL_BLOCK, change, cred, flags);
    576 }
    577 
    578 int
    579 lfs_chkiq2(struct inode *ip, int32_t change, kauth_cred_t cred, int flags)
    580 {
    581 	return quota2_check(ip, QL_FILE, change, cred, flags);
    582 }
    583 
    584 int
    585 lfsquota2_handle_cmd_put(struct ulfsmount *ump, const struct quotakey *key,
    586     const struct quotaval *val)
    587 {
    588 	int error;
    589 	struct dquot *dq;
    590 	struct quota2_header *q2h;
    591 	struct quota2_entry q2e, *q2ep;
    592 	struct buf *bp;
    593 	struct lfs *fs = ump->um_lfs;
    594 	const int needswap = ULFS_MPNEEDSWAP(fs);
    595 
    596 	/* make sure we can index by the fs-independent idtype */
    597 	CTASSERT(QUOTA_IDTYPE_USER == ULFS_USRQUOTA);
    598 	CTASSERT(QUOTA_IDTYPE_GROUP == ULFS_GRPQUOTA);
    599 
    600 	if (ump->um_quotas[key->qk_idtype] == NULLVP)
    601 		return ENODEV;
    602 
    603 	if (key->qk_id == QUOTA_DEFAULTID) {
    604 		mutex_enter(&lfs_dqlock);
    605 		error = getq2h(ump, key->qk_idtype, &bp, &q2h, B_MODIFY);
    606 		if (error) {
    607 			mutex_exit(&lfs_dqlock);
    608 			goto out_wapbl;
    609 		}
    610 		lfsquota2_ulfs_rwq2e(&q2h->q2h_defentry, &q2e, needswap);
    611 		quota2_dict_update_q2e_limits(key->qk_objtype, val, &q2e);
    612 		lfsquota2_ulfs_rwq2e(&q2e, &q2h->q2h_defentry, needswap);
    613 		mutex_exit(&lfs_dqlock);
    614 		quota2_bwrite(ump->um_mountp, bp);
    615 		goto out_wapbl;
    616 	}
    617 
    618 	error = lfs_dqget(NULLVP, key->qk_id, ump, key->qk_idtype, &dq);
    619 	if (error)
    620 		goto out_wapbl;
    621 
    622 	mutex_enter(&dq->dq_interlock);
    623 	if (dq->dq2_lblkno == 0 && dq->dq2_blkoff == 0) {
    624 		/* need to alloc a new on-disk quot */
    625 		mutex_enter(&lfs_dqlock);
    626 		error = quota2_q2ealloc(ump, key->qk_idtype, key->qk_id, dq);
    627 		mutex_exit(&lfs_dqlock);
    628 		if (error)
    629 			goto out_il;
    630 	}
    631 	KASSERT(dq->dq2_lblkno != 0 || dq->dq2_blkoff != 0);
    632 	error = getq2e(ump, key->qk_idtype, dq->dq2_lblkno,
    633 	    dq->dq2_blkoff, &bp, &q2ep, B_MODIFY);
    634 	if (error)
    635 		goto out_il;
    636 
    637 	lfsquota2_ulfs_rwq2e(q2ep, &q2e, needswap);
    638 	quota2_dict_update_q2e_limits(key->qk_objtype, val, &q2e);
    639 	lfsquota2_ulfs_rwq2e(&q2e, q2ep, needswap);
    640 	quota2_bwrite(ump->um_mountp, bp);
    641 
    642 out_il:
    643 	mutex_exit(&dq->dq_interlock);
    644 	lfs_dqrele(NULLVP, dq);
    645 out_wapbl:
    646 	return error;
    647 }
    648 
    649 struct dq2clear_callback {
    650 	uid_t id;
    651 	struct dquot *dq;
    652 	struct quota2_header *q2h;
    653 };
    654 
    655 static int
    656 dq2clear_callback(struct ulfsmount *ump, uint64_t *offp, struct quota2_entry *q2e,
    657     uint64_t off, void *v)
    658 {
    659 	struct dq2clear_callback *c = v;
    660 #ifdef LFS_EI
    661 	struct lfs *fs = ump->um_lfs;
    662 	const int needswap = ULFS_MPNEEDSWAP(fs);
    663 #endif
    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_delete(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_wapbl;
    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_wapbl;
    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_wapbl:
    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 #ifdef LFS_EI
   1101 	struct lfs *fs = ump->um_lfs;
   1102 	const int needswap = ULFS_MPNEEDSWAP(fs);
   1103 #endif
   1104 
   1105 	if (gi->skipped < gi->skip) {
   1106 		gi->skipped++;
   1107 		return 0;
   1108 	}
   1109 	id = ulfs_rw32(q2ep->q2e_uid, needswap);
   1110 	q2cursor_addid(gi->state, gi->idtype, id);
   1111 	gi->new_skip++;
   1112 	if (gi->state->numids >= gi->state->maxids) {
   1113 		/* got enough ids, stop now */
   1114 		gi->stopped = 1;
   1115 		return Q2WL_ABORT;
   1116 	}
   1117 	return 0;
   1118 }
   1119 
   1120 /*
   1121  * Fill in a batch of quotakeys by scanning one or more hash chains.
   1122  */
   1123 static int
   1124 q2cursor_getkeys(struct ulfsmount *ump, int idtype, struct ulfsq2_cursor *cursor,
   1125     struct q2cursor_state *state,
   1126     int *hashsize_ret, struct quota2_entry *default_q2e_ret)
   1127 {
   1128 	struct lfs *fs = ump->um_lfs;
   1129 	const int needswap = ULFS_MPNEEDSWAP(fs);
   1130 	struct buf *hbp;
   1131 	struct quota2_header *q2h;
   1132 	int quota2_hash_size;
   1133 	struct q2cursor_getids gi;
   1134 	uint64_t offset;
   1135 	int error;
   1136 
   1137 	/*
   1138 	 * Read the header block.
   1139 	 */
   1140 
   1141 	mutex_enter(&lfs_dqlock);
   1142 	error = getq2h(ump, idtype, &hbp, &q2h, 0);
   1143 	if (error) {
   1144 		mutex_exit(&lfs_dqlock);
   1145 		return error;
   1146 	}
   1147 
   1148 	/* if the table size has changed, make the caller start over */
   1149 	quota2_hash_size = ulfs_rw16(q2h->q2h_hash_size, needswap);
   1150 	if (cursor->q2c_hashsize == 0) {
   1151 		cursor->q2c_hashsize = quota2_hash_size;
   1152 	} else if (cursor->q2c_hashsize != quota2_hash_size) {
   1153 		error = EDEADLK;
   1154 		goto scanfail;
   1155 	}
   1156 
   1157 	/* grab the entry with the default values out of the header */
   1158 	lfsquota2_ulfs_rwq2e(&q2h->q2h_defentry, default_q2e_ret, needswap);
   1159 
   1160 	/* If we haven't done the defaults yet, that goes first. */
   1161 	if (cursor->q2c_defaults_done == 0) {
   1162 		q2cursor_addid(state, idtype, QUOTA_DEFAULTID);
   1163 		/* if we read both halves, mark it done */
   1164 		if (state->numids < state->maxids || !state->skiplast) {
   1165 			cursor->q2c_defaults_done = 1;
   1166 		}
   1167 	}
   1168 
   1169 	gi.state = state;
   1170 	gi.idtype = idtype;
   1171 
   1172 	while (state->numids < state->maxids) {
   1173 		if (cursor->q2c_hashpos >= quota2_hash_size) {
   1174 			/* nothing more left */
   1175 			break;
   1176 		}
   1177 
   1178 		/* scan this hash chain */
   1179 		gi.skip = cursor->q2c_uidpos;
   1180 		gi.new_skip = gi.skip;
   1181 		gi.skipped = 0;
   1182 		gi.stopped = 0;
   1183 		offset = q2h->q2h_entries[cursor->q2c_hashpos];
   1184 
   1185 		error = quota2_walk_list(ump, hbp, idtype, &offset, 0, &gi,
   1186 		    q2cursor_getids_callback);
   1187 		KASSERT(error != Q2WL_ABORT);
   1188 		if (error) {
   1189 			break;
   1190 		}
   1191 		if (gi.stopped) {
   1192 			/* callback stopped before reading whole chain */
   1193 			cursor->q2c_uidpos = gi.new_skip;
   1194 			/* if we didn't get both halves, back up */
   1195 			if (state->numids == state->maxids && state->skiplast){
   1196 				KASSERT(cursor->q2c_uidpos > 0);
   1197 				cursor->q2c_uidpos--;
   1198 			}
   1199 		} else {
   1200 			/* read whole chain */
   1201 			/* if we got both halves of the last id, advance */
   1202 			if (state->numids < state->maxids || !state->skiplast){
   1203 				cursor->q2c_uidpos = 0;
   1204 				cursor->q2c_hashpos++;
   1205 			}
   1206 		}
   1207 	}
   1208 
   1209 scanfail:
   1210 	mutex_exit(&lfs_dqlock);
   1211 	brelse(hbp, 0);
   1212 	if (error)
   1213 		return error;
   1214 
   1215 	*hashsize_ret = quota2_hash_size;
   1216 	return 0;
   1217 }
   1218 
   1219 /*
   1220  * Fetch the quotavals for the quotakeys.
   1221  */
   1222 static int
   1223 q2cursor_getvals(struct ulfsmount *ump, struct q2cursor_state *state,
   1224     const struct quota2_entry *default_q2e)
   1225 {
   1226 	int hasid;
   1227 	id_t loadedid, id;
   1228 	unsigned pos;
   1229 	struct quota2_entry q2e;
   1230 	int objtype;
   1231 	int error;
   1232 
   1233 	hasid = 0;
   1234 	loadedid = 0;
   1235 	for (pos = 0; pos < state->numkeys; pos++) {
   1236 		id = state->keys[pos].qk_id;
   1237 		if (!hasid || id != loadedid) {
   1238 			hasid = 1;
   1239 			loadedid = id;
   1240 			if (id == QUOTA_DEFAULTID) {
   1241 				q2e = *default_q2e;
   1242 			} else {
   1243 				error = quota2_fetch_q2e(ump,
   1244 							 &state->keys[pos],
   1245 							 &q2e);
   1246 				if (error == ENOENT) {
   1247 					/* something changed - start over */
   1248 					error = EDEADLK;
   1249 				}
   1250 				if (error) {
   1251 					return error;
   1252 				}
   1253  			}
   1254 		}
   1255 
   1256 
   1257 		objtype = state->keys[pos].qk_objtype;
   1258 		KASSERT(objtype >= 0 && objtype < N_QL);
   1259 		q2val_to_quotaval(&q2e.q2e_val[objtype], &state->vals[pos]);
   1260 	}
   1261 
   1262 	return 0;
   1263 }
   1264 
   1265 /*
   1266  * Handle cursorget.
   1267  *
   1268  * We can't just read keys and values directly, because we can't walk
   1269  * the list with qdlock and grab dq_interlock to read the entries at
   1270  * the same time. So we're going to do two passes: one to figure out
   1271  * which IDs we want and fill in the keys, and then a second to use
   1272  * the keys to fetch the values.
   1273  */
   1274 int
   1275 lfsquota2_handle_cmd_cursorget(struct ulfsmount *ump, struct quotakcursor *qkc,
   1276     struct quotakey *keys, struct quotaval *vals, unsigned maxreturn,
   1277     unsigned *ret)
   1278 {
   1279 	int error;
   1280 	struct ulfsq2_cursor *cursor;
   1281 	struct ulfsq2_cursor newcursor;
   1282 	struct q2cursor_state state;
   1283 	struct quota2_entry default_q2e;
   1284 	int idtype;
   1285 	int quota2_hash_size;
   1286 
   1287 	/*
   1288 	 * Convert and validate the cursor.
   1289 	 */
   1290 	cursor = Q2CURSOR(qkc);
   1291 	error = q2cursor_check(cursor);
   1292 	if (error) {
   1293 		return error;
   1294 	}
   1295 
   1296 	/*
   1297 	 * Make sure our on-disk codes match the values of the
   1298 	 * FS-independent ones. This avoids the need for explicit
   1299 	 * conversion (which would be a NOP anyway and thus easily
   1300 	 * left out or called in the wrong places...)
   1301 	 */
   1302 	CTASSERT(QUOTA_IDTYPE_USER == ULFS_USRQUOTA);
   1303 	CTASSERT(QUOTA_IDTYPE_GROUP == ULFS_GRPQUOTA);
   1304 	CTASSERT(QUOTA_OBJTYPE_BLOCKS == QL_BLOCK);
   1305 	CTASSERT(QUOTA_OBJTYPE_FILES == QL_FILE);
   1306 
   1307 	/*
   1308 	 * If some of the idtypes aren't configured/enabled, arrange
   1309 	 * to skip over them.
   1310 	 */
   1311 	if (cursor->q2c_users_done == 0 &&
   1312 	    ump->um_quotas[ULFS_USRQUOTA] == NULLVP) {
   1313 		cursor->q2c_users_done = 1;
   1314 	}
   1315 	if (cursor->q2c_groups_done == 0 &&
   1316 	    ump->um_quotas[ULFS_GRPQUOTA] == NULLVP) {
   1317 		cursor->q2c_groups_done = 1;
   1318 	}
   1319 
   1320 	/* Loop over, potentially, both idtypes */
   1321 	while (1) {
   1322 
   1323 		/* Choose id type */
   1324 		error = q2cursor_pickidtype(cursor, &idtype);
   1325 		if (error == EAGAIN) {
   1326 			/* nothing more to do, return 0 */
   1327 			*ret = 0;
   1328 			return 0;
   1329 		}
   1330 		KASSERT(ump->um_quotas[idtype] != NULLVP);
   1331 
   1332 		/*
   1333 		 * Initialize the per-call iteration state. Copy the
   1334 		 * cursor state so we can update it in place but back
   1335 		 * out on error.
   1336 		 */
   1337 		q2cursor_initstate(&state, keys, vals, maxreturn,
   1338 				   cursor->q2c_blocks_done);
   1339 		newcursor = *cursor;
   1340 
   1341 		/* Assign keys */
   1342 		error = q2cursor_getkeys(ump, idtype, &newcursor, &state,
   1343 					 &quota2_hash_size, &default_q2e);
   1344 		if (error) {
   1345 			return error;
   1346 		}
   1347 
   1348 		/* Now fill in the values. */
   1349 		error = q2cursor_getvals(ump, &state, &default_q2e);
   1350 		if (error) {
   1351 			return error;
   1352 		}
   1353 
   1354 		/*
   1355 		 * Now that we aren't going to fail and lose what we
   1356 		 * did so far, we can update the cursor state.
   1357 		 */
   1358 
   1359 		if (newcursor.q2c_hashpos >= quota2_hash_size) {
   1360 			if (idtype == QUOTA_IDTYPE_USER)
   1361 				cursor->q2c_users_done = 1;
   1362 			else
   1363 				cursor->q2c_groups_done = 1;
   1364 
   1365 			/* start over on another id type */
   1366 			cursor->q2c_hashsize = 0;
   1367 			cursor->q2c_defaults_done = 0;
   1368 			cursor->q2c_hashpos = 0;
   1369 			cursor->q2c_uidpos = 0;
   1370 			cursor->q2c_blocks_done = 0;
   1371 		} else {
   1372 			*cursor = newcursor;
   1373 			cursor->q2c_blocks_done = state.skiplast;
   1374 		}
   1375 
   1376 		/*
   1377 		 * If we have something to return, return it.
   1378 		 * Otherwise, continue to the other idtype, if any,
   1379 		 * and only return zero at end of iteration.
   1380 		 */
   1381 		if (state.numkeys > 0) {
   1382 			break;
   1383 		}
   1384 	}
   1385 
   1386 	*ret = state.numkeys;
   1387 	return 0;
   1388 }
   1389 
   1390 int
   1391 lfsquota2_handle_cmd_cursoropen(struct ulfsmount *ump, struct quotakcursor *qkc)
   1392 {
   1393 	struct ulfsq2_cursor *cursor;
   1394 
   1395 	CTASSERT(sizeof(*cursor) <= sizeof(qkc->u.qkc_space));
   1396 	cursor = Q2CURSOR(qkc);
   1397 
   1398 	cursor->q2c_magic = Q2C_MAGIC;
   1399 	cursor->q2c_hashsize = 0;
   1400 
   1401 	cursor->q2c_users_done = 0;
   1402 	cursor->q2c_groups_done = 0;
   1403 	cursor->q2c_defaults_done = 0;
   1404 	cursor->q2c_hashpos = 0;
   1405 	cursor->q2c_uidpos = 0;
   1406 	cursor->q2c_blocks_done = 0;
   1407 	return 0;
   1408 }
   1409 
   1410 int
   1411 lfsquota2_handle_cmd_cursorclose(struct ulfsmount *ump, struct quotakcursor *qkc)
   1412 {
   1413 	struct ulfsq2_cursor *cursor;
   1414 	int error;
   1415 
   1416 	cursor = Q2CURSOR(qkc);
   1417 	error = q2cursor_check(cursor);
   1418 	if (error) {
   1419 		return error;
   1420 	}
   1421 
   1422 	/* nothing to do */
   1423 
   1424 	return 0;
   1425 }
   1426 
   1427 int
   1428 lfsquota2_handle_cmd_cursorskipidtype(struct ulfsmount *ump,
   1429     struct quotakcursor *qkc, int idtype)
   1430 {
   1431 	struct ulfsq2_cursor *cursor;
   1432 	int error;
   1433 
   1434 	cursor = Q2CURSOR(qkc);
   1435 	error = q2cursor_check(cursor);
   1436 	if (error) {
   1437 		return error;
   1438 	}
   1439 
   1440 	switch (idtype) {
   1441 	    case QUOTA_IDTYPE_USER:
   1442 		cursor->q2c_users_done = 1;
   1443 		break;
   1444 	    case QUOTA_IDTYPE_GROUP:
   1445 		cursor->q2c_groups_done = 1;
   1446 		break;
   1447 	    default:
   1448 		return EINVAL;
   1449 	}
   1450 
   1451 	return 0;
   1452 }
   1453 
   1454 int
   1455 lfsquota2_handle_cmd_cursoratend(struct ulfsmount *ump, struct quotakcursor *qkc,
   1456     int *ret)
   1457 {
   1458 	struct ulfsq2_cursor *cursor;
   1459 	int error;
   1460 
   1461 	cursor = Q2CURSOR(qkc);
   1462 	error = q2cursor_check(cursor);
   1463 	if (error) {
   1464 		return error;
   1465 	}
   1466 
   1467 	*ret = (cursor->q2c_users_done && cursor->q2c_groups_done);
   1468 	return 0;
   1469 }
   1470 
   1471 int
   1472 lfsquota2_handle_cmd_cursorrewind(struct ulfsmount *ump, struct quotakcursor *qkc)
   1473 {
   1474 	struct ulfsq2_cursor *cursor;
   1475 	int error;
   1476 
   1477 	cursor = Q2CURSOR(qkc);
   1478 	error = q2cursor_check(cursor);
   1479 	if (error) {
   1480 		return error;
   1481 	}
   1482 
   1483 	cursor->q2c_hashsize = 0;
   1484 
   1485 	cursor->q2c_users_done = 0;
   1486 	cursor->q2c_groups_done = 0;
   1487 	cursor->q2c_defaults_done = 0;
   1488 	cursor->q2c_hashpos = 0;
   1489 	cursor->q2c_uidpos = 0;
   1490 	cursor->q2c_blocks_done = 0;
   1491 
   1492 	return 0;
   1493 }
   1494 
   1495 int
   1496 lfs_q2sync(struct mount *mp)
   1497 {
   1498 	return 0;
   1499 }
   1500 
   1501 struct dq2get_callback {
   1502 	uid_t id;
   1503 	struct dquot *dq;
   1504 };
   1505 
   1506 static int
   1507 dq2get_callback(struct ulfsmount *ump, uint64_t *offp, struct quota2_entry *q2e,
   1508     uint64_t off, void *v)
   1509 {
   1510 	struct dq2get_callback *c = v;
   1511 	daddr_t lblkno;
   1512 	int blkoff;
   1513 #ifdef LFS_EI
   1514 	struct lfs *fs = ump->um_lfs;
   1515 	const int needswap = ULFS_MPNEEDSWAP(fs);
   1516 #endif
   1517 
   1518 	if (ulfs_rw32(q2e->q2e_uid, needswap) == c->id) {
   1519 		KASSERT(mutex_owned(&c->dq->dq_interlock));
   1520 		lblkno = (off >> ump->um_mountp->mnt_fs_bshift);
   1521 		blkoff = (off & ump->umq2_bmask);
   1522 		c->dq->dq2_lblkno = lblkno;
   1523 		c->dq->dq2_blkoff = blkoff;
   1524 		return Q2WL_ABORT;
   1525 	}
   1526 	return 0;
   1527 }
   1528 
   1529 int
   1530 lfs_dq2get(struct vnode *dqvp, u_long id, struct ulfsmount *ump, int type,
   1531     struct dquot *dq)
   1532 {
   1533 	struct buf *bp;
   1534 	struct quota2_header *q2h;
   1535 	int error;
   1536 	daddr_t offset;
   1537 	u_long hash_mask;
   1538 	struct dq2get_callback c = {
   1539 		.id = id,
   1540 		.dq = dq
   1541 	};
   1542 
   1543 	KASSERT(mutex_owned(&dq->dq_interlock));
   1544 	mutex_enter(&lfs_dqlock);
   1545 	error = getq2h(ump, type, &bp, &q2h, 0);
   1546 	if (error)
   1547 		goto out_mutex;
   1548 	/* look for our entry */
   1549 	hash_mask = ((1 << q2h->q2h_hash_shift) - 1);
   1550 	offset = q2h->q2h_entries[id & hash_mask];
   1551 	error = quota2_walk_list(ump, bp, type, &offset, 0, (void *)&c,
   1552 	    dq2get_callback);
   1553 	brelse(bp, 0);
   1554 out_mutex:
   1555 	mutex_exit(&lfs_dqlock);
   1556 	return error;
   1557 }
   1558 
   1559 int
   1560 lfs_dq2sync(struct vnode *vp, struct dquot *dq)
   1561 {
   1562 	return 0;
   1563 }
   1564 
   1565 int
   1566 lfs_quota2_mount(struct mount *mp)
   1567 {
   1568 	struct ulfsmount *ump = VFSTOULFS(mp);
   1569 	struct lfs *fs = ump->um_lfs;
   1570 	int error = 0;
   1571 	struct vnode *vp;
   1572 	struct lwp *l = curlwp;
   1573 
   1574 	if ((fs->lfs_use_quota2) == 0)
   1575 		return 0;
   1576 
   1577 	fs->um_flags |= ULFS_QUOTA2;
   1578 	ump->umq2_bsize = fs->lfs_bsize;
   1579 	ump->umq2_bmask = fs->lfs_bmask;
   1580 	if (fs->lfs_quota_magic != Q2_HEAD_MAGIC) {
   1581 		printf("%s: Invalid quota magic number\n",
   1582 		    mp->mnt_stat.f_mntonname);
   1583 		return EINVAL;
   1584 	}
   1585         if ((fs->lfs_quota_flags & FS_Q2_DO_TYPE(ULFS_USRQUOTA)) &&
   1586             fs->lfs_quotaino[ULFS_USRQUOTA] == 0) {
   1587                 printf("%s: no user quota inode\n",
   1588 		    mp->mnt_stat.f_mntonname);
   1589                 error = EINVAL;
   1590         }
   1591         if ((fs->lfs_quota_flags & FS_Q2_DO_TYPE(ULFS_GRPQUOTA)) &&
   1592             fs->lfs_quotaino[ULFS_GRPQUOTA] == 0) {
   1593                 printf("%s: no group quota inode\n",
   1594 		    mp->mnt_stat.f_mntonname);
   1595                 error = EINVAL;
   1596         }
   1597 	if (error)
   1598 		return error;
   1599 
   1600         if (fs->lfs_quota_flags & FS_Q2_DO_TYPE(ULFS_USRQUOTA) &&
   1601 	    ump->um_quotas[ULFS_USRQUOTA] == NULLVP) {
   1602 		error = VFS_VGET(mp, fs->lfs_quotaino[ULFS_USRQUOTA], &vp);
   1603 		if (error) {
   1604 			printf("%s: can't vget() user quota inode: %d\n",
   1605 			    mp->mnt_stat.f_mntonname, error);
   1606 			return error;
   1607 		}
   1608 		ump->um_quotas[ULFS_USRQUOTA] = vp;
   1609 		ump->um_cred[ULFS_USRQUOTA] = l->l_cred;
   1610 		mutex_enter(vp->v_interlock);
   1611 		vp->v_writecount++;
   1612 		mutex_exit(vp->v_interlock);
   1613 		VOP_UNLOCK(vp);
   1614 	}
   1615         if (fs->lfs_quota_flags & FS_Q2_DO_TYPE(ULFS_GRPQUOTA) &&
   1616 	    ump->um_quotas[ULFS_GRPQUOTA] == NULLVP) {
   1617 		error = VFS_VGET(mp, fs->lfs_quotaino[ULFS_GRPQUOTA], &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 	mp->mnt_flag |= MNT_QUOTA;
   1634 	return 0;
   1635 }
   1636