Home | History | Annotate | Line # | Download | only in lfs
ulfs_quota2.c revision 1.7
      1 /*	$NetBSD: ulfs_quota2.c,v 1.7 2013/07/28 00:29:18 dholland Exp $	*/
      2 /*  from NetBSD: ufs_quota2.c,v 1.35 2012/09/27 07:47:56 bouyer Exp  */
      3 
      4 /*-
      5   * Copyright (c) 2010 Manuel Bouyer
      6   * All rights reserved.
      7   *
      8   * Redistribution and use in source and binary forms, with or without
      9   * modification, are permitted provided that the following conditions
     10   * are met:
     11   * 1. Redistributions of source code must retain the above copyright
     12   *    notice, this list of conditions and the following disclaimer.
     13   * 2. Redistributions in binary form must reproduce the above copyright
     14   *    notice, this list of conditions and the following disclaimer in the
     15   *    documentation and/or other materials provided with the distribution.
     16   *
     17   * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     18   * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     19   * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     20   * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     21   * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     22   * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     23   * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     24   * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     25   * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     26   * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     27   * POSSIBILITY OF SUCH DAMAGE.
     28   */
     29 
     30 #include <sys/cdefs.h>
     31 __KERNEL_RCSID(0, "$NetBSD: ulfs_quota2.c,v 1.7 2013/07/28 00:29:18 dholland Exp $");
     32 
     33 #include <sys/buf.h>
     34 #include <sys/param.h>
     35 #include <sys/kernel.h>
     36 #include <sys/systm.h>
     37 #include <sys/namei.h>
     38 #include <sys/file.h>
     39 #include <sys/proc.h>
     40 #include <sys/vnode.h>
     41 #include <sys/mount.h>
     42 #include <sys/fstrans.h>
     43 #include <sys/kauth.h>
     44 #include <sys/wapbl.h>
     45 #include <sys/quota.h>
     46 #include <sys/quotactl.h>
     47 
     48 #include <ufs/lfs/lfs_extern.h>
     49 
     50 #include <ufs/lfs/ulfs_quota2.h>
     51 #include <ufs/lfs/ulfs_inode.h>
     52 #include <ufs/lfs/ulfsmount.h>
     53 #include <ufs/lfs/ulfs_bswap.h>
     54 #include <ufs/lfs/ulfs_extern.h>
     55 #include <ufs/lfs/ulfs_quota.h>
     56 
     57 /*
     58  * LOCKING:
     59  * Data in the entries are protected by the associated struct dquot's
     60  * dq_interlock (this means we can't read or change a quota entry without
     61  * grabing a dquot for it).
     62  * The header and lists (including pointers in the data entries, and q2e_uid)
     63  * are protected by the global dqlock.
     64  * the locking order is dq_interlock -> dqlock
     65  */
     66 
     67 static int quota2_bwrite(struct mount *, struct buf *);
     68 static int getinoquota2(struct inode *, bool, bool, struct buf **,
     69     struct quota2_entry **);
     70 static int getq2h(struct ulfsmount *, int, struct buf **,
     71     struct quota2_header **, int);
     72 static int getq2e(struct ulfsmount *, int, daddr_t, int, struct buf **,
     73     struct quota2_entry **, int);
     74 static int quota2_walk_list(struct ulfsmount *, struct buf *, int,
     75     uint64_t *, int, void *,
     76     int (*func)(struct ulfsmount *, uint64_t *, struct quota2_entry *,
     77       uint64_t, void *));
     78 
     79 static const char *limnames[] = INITQLNAMES;
     80 
     81 static void
     82 quota2_dict_update_q2e_limits(int objtype, const struct quotaval *val,
     83     struct quota2_entry *q2e)
     84 {
     85 	/* make sure we can index q2e_val[] by the fs-independent objtype */
     86 	CTASSERT(QUOTA_OBJTYPE_BLOCKS == QL_BLOCK);
     87 	CTASSERT(QUOTA_OBJTYPE_FILES == QL_FILE);
     88 
     89 	q2e->q2e_val[objtype].q2v_hardlimit = val->qv_hardlimit;
     90 	q2e->q2e_val[objtype].q2v_softlimit = val->qv_softlimit;
     91 	q2e->q2e_val[objtype].q2v_grace = val->qv_grace;
     92 }
     93 
     94 /*
     95  * Convert internal representation to FS-independent representation.
     96  * (Note that while the two types are currently identical, the
     97  * internal representation is an on-disk struct and the FS-independent
     98  * representation is not, and they might diverge in the future.)
     99  */
    100 static void
    101 q2val_to_quotaval(struct quota2_val *q2v, struct quotaval *qv)
    102 {
    103 	qv->qv_softlimit = q2v->q2v_softlimit;
    104 	qv->qv_hardlimit = q2v->q2v_hardlimit;
    105 	qv->qv_usage = q2v->q2v_cur;
    106 	qv->qv_expiretime = q2v->q2v_time;
    107 	qv->qv_grace = q2v->q2v_grace;
    108 }
    109 
    110 /*
    111  * Convert a quota2entry and default-flag to the FS-independent
    112  * representation.
    113  */
    114 static void
    115 q2e_to_quotaval(struct quota2_entry *q2e, int def,
    116 	       id_t *id, int objtype, struct quotaval *ret)
    117 {
    118 	if (def) {
    119 		*id = QUOTA_DEFAULTID;
    120 	} else {
    121 		*id = q2e->q2e_uid;
    122 	}
    123 
    124 	KASSERT(objtype >= 0 && objtype < N_QL);
    125 	q2val_to_quotaval(&q2e->q2e_val[objtype], ret);
    126 }
    127 
    128 
    129 static int
    130 quota2_bwrite(struct mount *mp, struct buf *bp)
    131 {
    132 	if (mp->mnt_flag & MNT_SYNCHRONOUS)
    133 		return bwrite(bp);
    134 	else {
    135 		bdwrite(bp);
    136 		return 0;
    137 	}
    138 }
    139 
    140 static int
    141 getq2h(struct ulfsmount *ump, int type,
    142     struct buf **bpp, struct quota2_header **q2hp, int flags)
    143 {
    144 #ifdef LFS_EI
    145 	const int needswap = ULFS_MPNEEDSWAP(ump);
    146 #endif
    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,
    153 	    ump->um_cred[type], flags, &bp);
    154 	if (error)
    155 		return error;
    156 	if (bp->b_resid != 0)
    157 		panic("dq2get: %s quota file truncated", lfs_quotatypes[type]);
    158 
    159 	q2h = (void *)bp->b_data;
    160 	if (ulfs_rw32(q2h->q2h_magic_number, needswap) != Q2_HEAD_MAGIC ||
    161 	    q2h->q2h_type != type)
    162 		panic("dq2get: corrupted %s quota header", lfs_quotatypes[type]);
    163 	*bpp = bp;
    164 	*q2hp = q2h;
    165 	return 0;
    166 }
    167 
    168 static int
    169 getq2e(struct ulfsmount *ump, int type, daddr_t lblkno, int blkoffset,
    170     struct buf **bpp, struct quota2_entry **q2ep, int flags)
    171 {
    172 	int error;
    173 	struct buf *bp;
    174 
    175 	if (blkoffset & (sizeof(uint64_t) - 1)) {
    176 		panic("dq2get: %s quota file corrupted",
    177 		    lfs_quotatypes[type]);
    178 	}
    179 	error = bread(ump->um_quotas[type], lblkno, ump->umq2_bsize,
    180 	    ump->um_cred[type], flags, &bp);
    181 	if (error)
    182 		return error;
    183 	if (bp->b_resid != 0) {
    184 		panic("dq2get: %s quota file corrupted",
    185 		    lfs_quotatypes[type]);
    186 	}
    187 	*q2ep = (void *)((char *)bp->b_data + blkoffset);
    188 	*bpp = bp;
    189 	return 0;
    190 }
    191 
    192 /* walk a quota entry list, calling the callback for each entry */
    193 #define Q2WL_ABORT 0x10000000
    194 
    195 static int
    196 quota2_walk_list(struct ulfsmount *ump, struct buf *hbp, int type,
    197     uint64_t *offp, int flags, void *a,
    198     int (*func)(struct ulfsmount *, uint64_t *, struct quota2_entry *, uint64_t, void *))
    199 {
    200 #ifdef LFS_EI
    201 	const int needswap = ULFS_MPNEEDSWAP(ump);
    202 #endif
    203 	daddr_t off = ulfs_rw64(*offp, needswap);
    204 	struct buf *bp, *obp = hbp;
    205 	int ret = 0, ret2 = 0;
    206 	struct quota2_entry *q2e;
    207 	daddr_t lblkno, blkoff, olblkno = 0;
    208 
    209 	KASSERT(mutex_owner(&lfs_dqlock));
    210 
    211 	while (off != 0) {
    212 		lblkno = (off >> ump->um_mountp->mnt_fs_bshift);
    213 		blkoff = (off & ump->umq2_bmask);
    214 		if (lblkno == 0) {
    215 			/* in the header block */
    216 			bp = hbp;
    217 		} else if (lblkno == olblkno) {
    218 			/* still in the same buf */
    219 			bp = obp;
    220 		} else {
    221 			ret = bread(ump->um_quotas[type], lblkno,
    222 			    ump->umq2_bsize,
    223 			    ump->um_cred[type], flags, &bp);
    224 			if (ret)
    225 				return ret;
    226 			if (bp->b_resid != 0) {
    227 				panic("quota2_walk_list: %s quota file corrupted",
    228 				    lfs_quotatypes[type]);
    229 			}
    230 		}
    231 		q2e = (void *)((char *)(bp->b_data) + blkoff);
    232 		ret = (*func)(ump, offp, q2e, off, a);
    233 		if (off != ulfs_rw64(*offp, needswap)) {
    234 			/* callback changed parent's pointer, redo */
    235 			off = ulfs_rw64(*offp, needswap);
    236 			if (bp != hbp && bp != obp)
    237 				ret2 = bwrite(bp);
    238 		} else {
    239 			/* parent if now current */
    240 			if (obp != bp && obp != hbp) {
    241 				if (flags & B_MODIFY)
    242 					ret2 = bwrite(obp);
    243 				else
    244 					brelse(obp, 0);
    245 			}
    246 			obp = bp;
    247 			olblkno = lblkno;
    248 			offp = &(q2e->q2e_next);
    249 			off = ulfs_rw64(*offp, needswap);
    250 		}
    251 		if (ret)
    252 			break;
    253 		if (ret2) {
    254 			ret = ret2;
    255 			break;
    256 		}
    257 	}
    258 	if (obp != hbp) {
    259 		if (flags & B_MODIFY)
    260 			ret2 = bwrite(obp);
    261 		else
    262 			brelse(obp, 0);
    263 	}
    264 	if (ret & Q2WL_ABORT)
    265 		return 0;
    266 	if (ret == 0)
    267 		return ret2;
    268 	return ret;
    269 }
    270 
    271 int
    272 lfsquota2_umount(struct mount *mp, int flags)
    273 {
    274 	int i, error;
    275 	struct ulfsmount *ump = VFSTOULFS(mp);
    276 
    277 	if ((ump->um_flags & ULFS_QUOTA2) == 0)
    278 		return 0;
    279 
    280 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    281 		if (ump->um_quotas[i] != NULLVP) {
    282 			error = vn_close(ump->um_quotas[i], FREAD|FWRITE,
    283 			    ump->um_cred[i]);
    284 			if (error) {
    285 				printf("quota2_umount failed: close(%p) %d\n",
    286 				    ump->um_quotas[i], error);
    287 				return error;
    288 			}
    289 		}
    290 		ump->um_quotas[i] = NULLVP;
    291 	}
    292 	return 0;
    293 }
    294 
    295 static int
    296 quota2_q2ealloc(struct ulfsmount *ump, int type, uid_t uid, struct dquot *dq)
    297 {
    298 	int error, error2;
    299 	struct buf *hbp, *bp;
    300 	struct quota2_header *q2h;
    301 	struct quota2_entry *q2e;
    302 	daddr_t offset;
    303 	u_long hash_mask;
    304 	const int needswap = ULFS_MPNEEDSWAP(ump);
    305 
    306 	KASSERT(mutex_owned(&dq->dq_interlock));
    307 	KASSERT(mutex_owned(&lfs_dqlock));
    308 	error = getq2h(ump, type, &hbp, &q2h, B_MODIFY);
    309 	if (error)
    310 		return error;
    311 	offset = ulfs_rw64(q2h->q2h_free, needswap);
    312 	if (offset == 0) {
    313 		struct vnode *vp = ump->um_quotas[type];
    314 		struct inode *ip = VTOI(vp);
    315 		uint64_t size = ip->i_size;
    316 		/* need to alocate a new disk block */
    317 		error = ULFS_BALLOC(vp, size, ump->umq2_bsize,
    318 		    ump->um_cred[type], B_CLRBUF | B_SYNC, &bp);
    319 		if (error) {
    320 			brelse(hbp, 0);
    321 			return error;
    322 		}
    323 		KASSERT((ip->i_size % ump->umq2_bsize) == 0);
    324 		ip->i_size += ump->umq2_bsize;
    325 		DIP_ASSIGN(ip, size, ip->i_size);
    326 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    327 		uvm_vnp_setsize(vp, ip->i_size);
    328 		lfsquota2_addfreeq2e(q2h, bp->b_data, size, ump->umq2_bsize,
    329 		    needswap);
    330 		error = bwrite(bp);
    331 		error2 = ULFS_UPDATE(vp, NULL, NULL, UPDATE_WAIT);
    332 		if (error || error2) {
    333 			brelse(hbp, 0);
    334 			if (error)
    335 				return error;
    336 			return error2;
    337 		}
    338 		offset = ulfs_rw64(q2h->q2h_free, needswap);
    339 		KASSERT(offset != 0);
    340 	}
    341 	dq->dq2_lblkno = (offset >> ump->um_mountp->mnt_fs_bshift);
    342 	dq->dq2_blkoff = (offset & ump->umq2_bmask);
    343 	if (dq->dq2_lblkno == 0) {
    344 		bp = hbp;
    345 		q2e = (void *)((char *)bp->b_data + dq->dq2_blkoff);
    346 	} else {
    347 		error = getq2e(ump, type, dq->dq2_lblkno,
    348 		    dq->dq2_blkoff, &bp, &q2e, B_MODIFY);
    349 		if (error) {
    350 			brelse(hbp, 0);
    351 			return error;
    352 		}
    353 	}
    354 	hash_mask = ((1 << q2h->q2h_hash_shift) - 1);
    355 	/* remove from free list */
    356 	q2h->q2h_free = q2e->q2e_next;
    357 
    358 	memcpy(q2e, &q2h->q2h_defentry, sizeof(*q2e));
    359 	q2e->q2e_uid = ulfs_rw32(uid, needswap);
    360 	/* insert in hash list */
    361 	q2e->q2e_next = q2h->q2h_entries[uid & hash_mask];
    362 	q2h->q2h_entries[uid & hash_mask] = ulfs_rw64(offset, needswap);
    363 	if (hbp != bp) {
    364 		bwrite(hbp);
    365 	}
    366 	bwrite(bp);
    367 	return 0;
    368 }
    369 
    370 static int
    371 getinoquota2(struct inode *ip, bool alloc, bool modify, struct buf **bpp,
    372     struct quota2_entry **q2ep)
    373 {
    374 	int error;
    375 	int i;
    376 	struct dquot *dq;
    377 	struct ulfsmount *ump = ip->i_ump;
    378 	u_int32_t ino_ids[ULFS_MAXQUOTAS];
    379 
    380 	error = lfs_getinoquota(ip);
    381 	if (error)
    382 		return error;
    383 
    384         ino_ids[ULFS_USRQUOTA] = ip->i_uid;
    385         ino_ids[ULFS_GRPQUOTA] = ip->i_gid;
    386 	/* first get the interlock for all dquot */
    387 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    388 		dq = ip->i_dquot[i];
    389 		if (dq == NODQUOT)
    390 			continue;
    391 		mutex_enter(&dq->dq_interlock);
    392 	}
    393 	/* now get the corresponding quota entry */
    394 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    395 		bpp[i] = NULL;
    396 		q2ep[i] = NULL;
    397 		dq = ip->i_dquot[i];
    398 		if (dq == NODQUOT)
    399 			continue;
    400 		if (__predict_false(ump->um_quotas[i] == NULL)) {
    401 			/*
    402 			 * quotas have been turned off. This can happen
    403 			 * at umount time.
    404 			 */
    405 			mutex_exit(&dq->dq_interlock);
    406 			lfs_dqrele(NULLVP, dq);
    407 			ip->i_dquot[i] = NULL;
    408 			continue;
    409 		}
    410 
    411 		if ((dq->dq2_lblkno | dq->dq2_blkoff) == 0) {
    412 			if (!alloc) {
    413 				continue;
    414 			}
    415 			/* need to alloc a new on-disk quot */
    416 			mutex_enter(&lfs_dqlock);
    417 			error = quota2_q2ealloc(ump, i, ino_ids[i], dq);
    418 			mutex_exit(&lfs_dqlock);
    419 			if (error)
    420 				return error;
    421 		}
    422 		KASSERT(dq->dq2_lblkno != 0 || dq->dq2_blkoff != 0);
    423 		error = getq2e(ump, i, dq->dq2_lblkno,
    424 		    dq->dq2_blkoff, &bpp[i], &q2ep[i],
    425 		    modify ? B_MODIFY : 0);
    426 		if (error)
    427 			return error;
    428 	}
    429 	return 0;
    430 }
    431 
    432 __inline static int __unused
    433 lfsquota2_check_limit(struct quota2_val *q2v, uint64_t change, time_t now)
    434 {
    435 	return lfsquota_check_limit(q2v->q2v_cur, change, q2v->q2v_softlimit,
    436 	    q2v->q2v_hardlimit, q2v->q2v_time, now);
    437 }
    438 
    439 static int
    440 quota2_check(struct inode *ip, int vtype, int64_t change, kauth_cred_t cred,
    441     int flags)
    442 {
    443 	int error;
    444 	struct buf *bp[ULFS_MAXQUOTAS];
    445 	struct quota2_entry *q2e[ULFS_MAXQUOTAS];
    446 	struct quota2_val *q2vp;
    447 	struct dquot *dq;
    448 	uint64_t ncurblks;
    449 	struct ulfsmount *ump = ip->i_ump;
    450 	struct mount *mp = ump->um_mountp;
    451 	const int needswap = ULFS_MPNEEDSWAP(ump);
    452 	int i;
    453 
    454 	if ((error = getinoquota2(ip, change > 0, change != 0, bp, q2e)) != 0)
    455 		return error;
    456 	if (change == 0) {
    457 		for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    458 			dq = ip->i_dquot[i];
    459 			if (dq == NODQUOT)
    460 				continue;
    461 			if (bp[i])
    462 				brelse(bp[i], 0);
    463 			mutex_exit(&dq->dq_interlock);
    464 		}
    465 		return 0;
    466 	}
    467 	if (change < 0) {
    468 		for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    469 			dq = ip->i_dquot[i];
    470 			if (dq == NODQUOT)
    471 				continue;
    472 			if (q2e[i] == NULL) {
    473 				mutex_exit(&dq->dq_interlock);
    474 				continue;
    475 			}
    476 			q2vp = &q2e[i]->q2e_val[vtype];
    477 			ncurblks = ulfs_rw64(q2vp->q2v_cur, needswap);
    478 			if (ncurblks < -change)
    479 				ncurblks = 0;
    480 			else
    481 				ncurblks += change;
    482 			q2vp->q2v_cur = ulfs_rw64(ncurblks, needswap);
    483 			quota2_bwrite(mp, bp[i]);
    484 			mutex_exit(&dq->dq_interlock);
    485 		}
    486 		return 0;
    487 	}
    488 	/* see if the allocation is allowed */
    489 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    490 		struct quota2_val q2v;
    491 		int ql_stat;
    492 		dq = ip->i_dquot[i];
    493 		if (dq == NODQUOT)
    494 			continue;
    495 		KASSERT(q2e[i] != NULL);
    496 		lfsquota2_ulfs_rwq2v(&q2e[i]->q2e_val[vtype], &q2v, needswap);
    497 		ql_stat = lfsquota2_check_limit(&q2v, change, time_second);
    498 
    499 		if ((flags & FORCE) == 0 &&
    500 		    kauth_authorize_system(cred, KAUTH_SYSTEM_FS_QUOTA,
    501 		    KAUTH_REQ_SYSTEM_FS_QUOTA_NOLIMIT,
    502 		    KAUTH_ARG(i), KAUTH_ARG(vtype), NULL) != 0) {
    503 			/* enforce this limit */
    504 			switch(QL_STATUS(ql_stat)) {
    505 			case QL_S_DENY_HARD:
    506 				if ((dq->dq_flags & DQ_WARN(vtype)) == 0) {
    507 					uprintf("\n%s: write failed, %s %s "
    508 					    "limit reached\n",
    509 					    mp->mnt_stat.f_mntonname,
    510 					    lfs_quotatypes[i], limnames[vtype]);
    511 					dq->dq_flags |= DQ_WARN(vtype);
    512 				}
    513 				error = EDQUOT;
    514 				break;
    515 			case QL_S_DENY_GRACE:
    516 				if ((dq->dq_flags & DQ_WARN(vtype)) == 0) {
    517 					uprintf("\n%s: write failed, %s %s "
    518 					    "limit reached\n",
    519 					    mp->mnt_stat.f_mntonname,
    520 					    lfs_quotatypes[i], limnames[vtype]);
    521 					dq->dq_flags |= DQ_WARN(vtype);
    522 				}
    523 				error = EDQUOT;
    524 				break;
    525 			case QL_S_ALLOW_SOFT:
    526 				if ((dq->dq_flags & DQ_WARN(vtype)) == 0) {
    527 					uprintf("\n%s: warning, %s %s "
    528 					    "quota exceeded\n",
    529 					    mp->mnt_stat.f_mntonname,
    530 					    lfs_quotatypes[i], limnames[vtype]);
    531 					dq->dq_flags |= DQ_WARN(vtype);
    532 				}
    533 				break;
    534 			}
    535 		}
    536 		/*
    537 		 * always do this; we don't know if the allocation will
    538 		 * succed or not in the end. if we don't do the allocation
    539 		 * q2v_time will be ignored anyway
    540 		 */
    541 		if (ql_stat & QL_F_CROSS) {
    542 			q2v.q2v_time = time_second + q2v.q2v_grace;
    543 			lfsquota2_ulfs_rwq2v(&q2v, &q2e[i]->q2e_val[vtype],
    544 			    needswap);
    545 		}
    546 	}
    547 
    548 	/* now do the allocation if allowed */
    549 	for (i = 0; i < ULFS_MAXQUOTAS; i++) {
    550 		dq = ip->i_dquot[i];
    551 		if (dq == NODQUOT)
    552 			continue;
    553 		KASSERT(q2e[i] != NULL);
    554 		if (error == 0) {
    555 			q2vp = &q2e[i]->q2e_val[vtype];
    556 			ncurblks = ulfs_rw64(q2vp->q2v_cur, needswap);
    557 			q2vp->q2v_cur = ulfs_rw64(ncurblks + change, needswap);
    558 			quota2_bwrite(mp, bp[i]);
    559 		} else
    560 			brelse(bp[i], 0);
    561 		mutex_exit(&dq->dq_interlock);
    562 	}
    563 	return error;
    564 }
    565 
    566 int
    567 lfs_chkdq2(struct inode *ip, int64_t change, kauth_cred_t cred, int flags)
    568 {
    569 	return quota2_check(ip, QL_BLOCK, change, cred, flags);
    570 }
    571 
    572 int
    573 lfs_chkiq2(struct inode *ip, int32_t change, kauth_cred_t cred, int flags)
    574 {
    575 	return quota2_check(ip, QL_FILE, change, cred, flags);
    576 }
    577 
    578 int
    579 lfsquota2_handle_cmd_put(struct ulfsmount *ump, const struct quotakey *key,
    580     const struct quotaval *val)
    581 {
    582 	int error;
    583 	struct dquot *dq;
    584 	struct quota2_header *q2h;
    585 	struct quota2_entry q2e, *q2ep;
    586 	struct buf *bp;
    587 	const int needswap = ULFS_MPNEEDSWAP(ump);
    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_wapbl;
    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_wapbl;
    609 	}
    610 
    611 	error = lfs_dqget(NULLVP, key->qk_id, ump, key->qk_idtype, &dq);
    612 	if (error)
    613 		goto out_wapbl;
    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 	quota2_dict_update_q2e_limits(key->qk_objtype, val, &q2e);
    632 	lfsquota2_ulfs_rwq2e(&q2e, q2ep, needswap);
    633 	quota2_bwrite(ump->um_mountp, bp);
    634 
    635 out_il:
    636 	mutex_exit(&dq->dq_interlock);
    637 	lfs_dqrele(NULLVP, dq);
    638 out_wapbl:
    639 	return error;
    640 }
    641 
    642 struct dq2clear_callback {
    643 	uid_t id;
    644 	struct dquot *dq;
    645 	struct quota2_header *q2h;
    646 };
    647 
    648 static int
    649 dq2clear_callback(struct ulfsmount *ump, uint64_t *offp, struct quota2_entry *q2e,
    650     uint64_t off, void *v)
    651 {
    652 	struct dq2clear_callback *c = v;
    653 #ifdef LFS_EI
    654 	const int needswap = ULFS_MPNEEDSWAP(ump);
    655 #endif
    656 	uint64_t myoff;
    657 
    658 	if (ulfs_rw32(q2e->q2e_uid, needswap) == c->id) {
    659 		KASSERT(mutex_owned(&c->dq->dq_interlock));
    660 		c->dq->dq2_lblkno = 0;
    661 		c->dq->dq2_blkoff = 0;
    662 		myoff = *offp;
    663 		/* remove from hash list */
    664 		*offp = q2e->q2e_next;
    665 		/* add to free list */
    666 		q2e->q2e_next = c->q2h->q2h_free;
    667 		c->q2h->q2h_free = myoff;
    668 		return Q2WL_ABORT;
    669 	}
    670 	return 0;
    671 }
    672 int
    673 lfsquota2_handle_cmd_delete(struct ulfsmount *ump, const struct quotakey *qk)
    674 {
    675 	int idtype;
    676 	id_t id;
    677 	int objtype;
    678 	int error, i, canfree;
    679 	struct dquot *dq;
    680 	struct quota2_header *q2h;
    681 	struct quota2_entry q2e, *q2ep;
    682 	struct buf *hbp, *bp;
    683 	u_long hash_mask;
    684 	struct dq2clear_callback c;
    685 
    686 	idtype = qk->qk_idtype;
    687 	id = qk->qk_id;
    688 	objtype = qk->qk_objtype;
    689 
    690 	if (ump->um_quotas[idtype] == NULLVP)
    691 		return ENODEV;
    692 	if (id == QUOTA_DEFAULTID)
    693 		return EOPNOTSUPP;
    694 
    695 	/* get the default entry before locking the entry's buffer */
    696 	mutex_enter(&lfs_dqlock);
    697 	error = getq2h(ump, idtype, &hbp, &q2h, 0);
    698 	if (error) {
    699 		mutex_exit(&lfs_dqlock);
    700 		return error;
    701 	}
    702 	/* we'll copy to another disk entry, so no need to swap */
    703 	memcpy(&q2e, &q2h->q2h_defentry, sizeof(q2e));
    704 	mutex_exit(&lfs_dqlock);
    705 	brelse(hbp, 0);
    706 
    707 	error = lfs_dqget(NULLVP, id, ump, idtype, &dq);
    708 	if (error)
    709 		return error;
    710 
    711 	mutex_enter(&dq->dq_interlock);
    712 	if (dq->dq2_lblkno == 0 && dq->dq2_blkoff == 0) {
    713 		/* already clear, nothing to do */
    714 		error = ENOENT;
    715 		goto out_il;
    716 	}
    717 
    718 	error = getq2e(ump, idtype, dq->dq2_lblkno, dq->dq2_blkoff,
    719 	    &bp, &q2ep, B_MODIFY);
    720 	if (error)
    721 		goto out_wapbl;
    722 
    723 	/* make sure we can index by the objtype passed in */
    724 	CTASSERT(QUOTA_OBJTYPE_BLOCKS == QL_BLOCK);
    725 	CTASSERT(QUOTA_OBJTYPE_FILES == QL_FILE);
    726 
    727 	/* clear the requested objtype by copying from the default entry */
    728 	q2ep->q2e_val[objtype].q2v_softlimit =
    729 		q2e.q2e_val[objtype].q2v_softlimit;
    730 	q2ep->q2e_val[objtype].q2v_hardlimit =
    731 		q2e.q2e_val[objtype].q2v_hardlimit;
    732 	q2ep->q2e_val[objtype].q2v_grace =
    733 		q2e.q2e_val[objtype].q2v_grace;
    734 	q2ep->q2e_val[objtype].q2v_time = 0;
    735 
    736 	/* if this entry now contains no information, we can free it */
    737 	canfree = 1;
    738 	for (i = 0; i < N_QL; i++) {
    739 		if (q2ep->q2e_val[i].q2v_cur != 0 ||
    740 		    (q2ep->q2e_val[i].q2v_softlimit !=
    741 		     q2e.q2e_val[i].q2v_softlimit) ||
    742 		    (q2ep->q2e_val[i].q2v_hardlimit !=
    743 		     q2e.q2e_val[i].q2v_hardlimit) ||
    744 		    (q2ep->q2e_val[i].q2v_grace !=
    745 		     q2e.q2e_val[i].q2v_grace)) {
    746 			canfree = 0;
    747 			break;
    748 		}
    749 		/* note: do not need to check q2v_time */
    750 	}
    751 
    752 	if (canfree == 0) {
    753 		quota2_bwrite(ump->um_mountp, bp);
    754 		goto out_wapbl;
    755 	}
    756 	/* we can free it. release bp so we can walk the list */
    757 	brelse(bp, 0);
    758 	mutex_enter(&lfs_dqlock);
    759 	error = getq2h(ump, idtype, &hbp, &q2h, 0);
    760 	if (error)
    761 		goto out_dqlock;
    762 
    763 	hash_mask = ((1 << q2h->q2h_hash_shift) - 1);
    764 	c.dq = dq;
    765 	c.id = id;
    766 	c.q2h = q2h;
    767 	error = quota2_walk_list(ump, hbp, idtype,
    768 	    &q2h->q2h_entries[id & hash_mask], B_MODIFY, &c,
    769 	    dq2clear_callback);
    770 
    771 	bwrite(hbp);
    772 
    773 out_dqlock:
    774 	mutex_exit(&lfs_dqlock);
    775 out_wapbl:
    776 out_il:
    777 	mutex_exit(&dq->dq_interlock);
    778 	lfs_dqrele(NULLVP, dq);
    779 	return error;
    780 }
    781 
    782 static int
    783 quota2_fetch_q2e(struct ulfsmount *ump, const struct quotakey *qk,
    784     struct quota2_entry *ret)
    785 {
    786 	struct dquot *dq;
    787 	int error;
    788 	struct quota2_entry *q2ep;
    789 	struct buf *bp;
    790 	const int needswap = ULFS_MPNEEDSWAP(ump);
    791 
    792 	error = lfs_dqget(NULLVP, qk->qk_id, ump, qk->qk_idtype, &dq);
    793 	if (error)
    794 		return error;
    795 
    796 	mutex_enter(&dq->dq_interlock);
    797 	if (dq->dq2_lblkno == 0 && dq->dq2_blkoff == 0) {
    798 		mutex_exit(&dq->dq_interlock);
    799 		lfs_dqrele(NULLVP, dq);
    800 		return ENOENT;
    801 	}
    802 	error = getq2e(ump, qk->qk_idtype, dq->dq2_lblkno, dq->dq2_blkoff,
    803 	    &bp, &q2ep, 0);
    804 	if (error) {
    805 		mutex_exit(&dq->dq_interlock);
    806 		lfs_dqrele(NULLVP, dq);
    807 		return error;
    808 	}
    809 	lfsquota2_ulfs_rwq2e(q2ep, ret, needswap);
    810 	brelse(bp, 0);
    811 	mutex_exit(&dq->dq_interlock);
    812 	lfs_dqrele(NULLVP, dq);
    813 
    814 	return 0;
    815 }
    816 
    817 static int
    818 quota2_fetch_quotaval(struct ulfsmount *ump, const struct quotakey *qk,
    819     struct quotaval *ret)
    820 {
    821 	struct dquot *dq;
    822 	int error;
    823 	struct quota2_entry *q2ep, q2e;
    824 	struct buf  *bp;
    825 	const int needswap = ULFS_MPNEEDSWAP(ump);
    826 	id_t id2;
    827 
    828 	error = lfs_dqget(NULLVP, qk->qk_id, ump, qk->qk_idtype, &dq);
    829 	if (error)
    830 		return error;
    831 
    832 	mutex_enter(&dq->dq_interlock);
    833 	if (dq->dq2_lblkno == 0 && dq->dq2_blkoff == 0) {
    834 		mutex_exit(&dq->dq_interlock);
    835 		lfs_dqrele(NULLVP, dq);
    836 		return ENOENT;
    837 	}
    838 	error = getq2e(ump, qk->qk_idtype, dq->dq2_lblkno, dq->dq2_blkoff,
    839 	    &bp, &q2ep, 0);
    840 	if (error) {
    841 		mutex_exit(&dq->dq_interlock);
    842 		lfs_dqrele(NULLVP, dq);
    843 		return error;
    844 	}
    845 	lfsquota2_ulfs_rwq2e(q2ep, &q2e, needswap);
    846 	brelse(bp, 0);
    847 	mutex_exit(&dq->dq_interlock);
    848 	lfs_dqrele(NULLVP, dq);
    849 
    850 	q2e_to_quotaval(&q2e, 0, &id2, qk->qk_objtype, ret);
    851 	KASSERT(id2 == qk->qk_id);
    852 	return 0;
    853 }
    854 
    855 int
    856 lfsquota2_handle_cmd_get(struct ulfsmount *ump, const struct quotakey *qk,
    857     struct quotaval *qv)
    858 {
    859 	int error;
    860 	struct quota2_header *q2h;
    861 	struct quota2_entry q2e;
    862 	struct buf *bp;
    863 	const int needswap = ULFS_MPNEEDSWAP(ump);
    864 	id_t id2;
    865 
    866 	/*
    867 	 * Make sure the FS-independent codes match the internal ones,
    868 	 * so we can use the passed-in objtype without having to
    869 	 * convert it explicitly to QL_BLOCK/QL_FILE.
    870 	 */
    871 	CTASSERT(QL_BLOCK == QUOTA_OBJTYPE_BLOCKS);
    872 	CTASSERT(QL_FILE == QUOTA_OBJTYPE_FILES);
    873 	CTASSERT(N_QL == 2);
    874 
    875 	if (qk->qk_objtype < 0 || qk->qk_objtype >= N_QL) {
    876 		return EINVAL;
    877 	}
    878 
    879 	if (ump->um_quotas[qk->qk_idtype] == NULLVP)
    880 		return ENODEV;
    881 	if (qk->qk_id == QUOTA_DEFAULTID) {
    882 		mutex_enter(&lfs_dqlock);
    883 		error = getq2h(ump, qk->qk_idtype, &bp, &q2h, 0);
    884 		if (error) {
    885 			mutex_exit(&lfs_dqlock);
    886 			return error;
    887 		}
    888 		lfsquota2_ulfs_rwq2e(&q2h->q2h_defentry, &q2e, needswap);
    889 		mutex_exit(&lfs_dqlock);
    890 		brelse(bp, 0);
    891 		q2e_to_quotaval(&q2e, qk->qk_id == QUOTA_DEFAULTID, &id2,
    892 				qk->qk_objtype, qv);
    893 		(void)id2;
    894 	} else
    895 		error = quota2_fetch_quotaval(ump, qk, qv);
    896 
    897 	return error;
    898 }
    899 
    900 /*
    901  * Cursor structure we used.
    902  *
    903  * This will get stored in userland between calls so we must not assume
    904  * it isn't arbitrarily corrupted.
    905  */
    906 struct ulfsq2_cursor {
    907 	uint32_t q2c_magic;	/* magic number */
    908 	int q2c_hashsize;	/* size of hash table at last go */
    909 
    910 	int q2c_users_done;	/* true if we've returned all user data */
    911 	int q2c_groups_done;	/* true if we've returned all group data */
    912 	int q2c_defaults_done;	/* true if we've returned the default values */
    913 	int q2c_hashpos;	/* slot to start at in hash table */
    914 	int q2c_uidpos;		/* number of ids we've handled */
    915 	int q2c_blocks_done;	/* true if we've returned the blocks value */
    916 };
    917 
    918 /*
    919  * State of a single cursorget call, or at least the part of it that
    920  * needs to be passed around.
    921  */
    922 struct q2cursor_state {
    923 	/* data return pointers */
    924 	struct quotakey *keys;
    925 	struct quotaval *vals;
    926 
    927 	/* key/value counters */
    928 	unsigned maxkeyvals;
    929 	unsigned numkeys;	/* number of keys assigned */
    930 
    931 	/* ID to key/value conversion state */
    932 	int skipfirst;		/* if true skip first key/value */
    933 	int skiplast;		/* if true skip last key/value */
    934 
    935 	/* ID counters */
    936 	unsigned maxids;	/* maximum number of IDs to handle */
    937 	unsigned numids;	/* number of IDs handled */
    938 };
    939 
    940 /*
    941  * Additional structure for getids callback.
    942  */
    943 struct q2cursor_getids {
    944 	struct q2cursor_state *state;
    945 	int idtype;
    946 	unsigned skip;		/* number of ids to skip over */
    947 	unsigned new_skip;	/* number of ids to skip over next time */
    948 	unsigned skipped;	/* number skipped so far */
    949 	int stopped;		/* true if we stopped quota_walk_list early */
    950 };
    951 
    952 /*
    953  * Cursor-related functions
    954  */
    955 
    956 /* magic number */
    957 #define Q2C_MAGIC (0xbeebe111)
    958 
    959 /* extract cursor from caller form */
    960 #define Q2CURSOR(qkc) ((struct ulfsq2_cursor *)&qkc->u.qkc_space[0])
    961 
    962 /*
    963  * Check that a cursor we're handed is something like valid. If
    964  * someone munges it and it still passes these checks, they'll get
    965  * partial or odd results back but won't break anything.
    966  */
    967 static int
    968 q2cursor_check(struct ulfsq2_cursor *cursor)
    969 {
    970 	if (cursor->q2c_magic != Q2C_MAGIC) {
    971 		return EINVAL;
    972 	}
    973 	if (cursor->q2c_hashsize < 0) {
    974 		return EINVAL;
    975 	}
    976 
    977 	if (cursor->q2c_users_done != 0 && cursor->q2c_users_done != 1) {
    978 		return EINVAL;
    979 	}
    980 	if (cursor->q2c_groups_done != 0 && cursor->q2c_groups_done != 1) {
    981 		return EINVAL;
    982 	}
    983 	if (cursor->q2c_defaults_done != 0 && cursor->q2c_defaults_done != 1) {
    984 		return EINVAL;
    985 	}
    986 	if (cursor->q2c_hashpos < 0 || cursor->q2c_uidpos < 0) {
    987 		return EINVAL;
    988 	}
    989 	if (cursor->q2c_blocks_done != 0 && cursor->q2c_blocks_done != 1) {
    990 		return EINVAL;
    991 	}
    992 	return 0;
    993 }
    994 
    995 /*
    996  * Set up the q2cursor state.
    997  */
    998 static void
    999 q2cursor_initstate(struct q2cursor_state *state, struct quotakey *keys,
   1000     struct quotaval *vals, unsigned maxkeyvals, int blocks_done)
   1001 {
   1002 	state->keys = keys;
   1003 	state->vals = vals;
   1004 
   1005 	state->maxkeyvals = maxkeyvals;
   1006 	state->numkeys = 0;
   1007 
   1008 	/*
   1009 	 * For each ID there are two quotavals to return. If the
   1010 	 * maximum number of entries to return is odd, we might want
   1011 	 * to skip the first quotaval of the first ID, or the last
   1012 	 * quotaval of the last ID, but not both. So the number of IDs
   1013 	 * we want is (up to) half the number of return slots we have,
   1014 	 * rounded up.
   1015 	 */
   1016 
   1017 	state->maxids = (state->maxkeyvals + 1) / 2;
   1018 	state->numids = 0;
   1019 	if (state->maxkeyvals % 2) {
   1020 		if (blocks_done) {
   1021 			state->skipfirst = 1;
   1022 			state->skiplast = 0;
   1023 		} else {
   1024 			state->skipfirst = 0;
   1025 			state->skiplast = 1;
   1026 		}
   1027 	} else {
   1028 		state->skipfirst = 0;
   1029 		state->skiplast = 0;
   1030 	}
   1031 }
   1032 
   1033 /*
   1034  * Choose which idtype we're going to work on. If doing a full
   1035  * iteration, we do users first, then groups, but either might be
   1036  * disabled or marked to skip via cursorsetidtype(), so don't make
   1037  * silly assumptions.
   1038  */
   1039 static int
   1040 q2cursor_pickidtype(struct ulfsq2_cursor *cursor, int *idtype_ret)
   1041 {
   1042 	if (cursor->q2c_users_done == 0) {
   1043 		*idtype_ret = QUOTA_IDTYPE_USER;
   1044 	} else if (cursor->q2c_groups_done == 0) {
   1045 		*idtype_ret = QUOTA_IDTYPE_GROUP;
   1046 	} else {
   1047 		return EAGAIN;
   1048 	}
   1049 	return 0;
   1050 }
   1051 
   1052 /*
   1053  * Add an ID to the current state. Sets up either one or two keys to
   1054  * refer to it, depending on whether it's first/last and the setting
   1055  * of skipfirst. (skiplast does not need to be explicitly tested)
   1056  */
   1057 static void
   1058 q2cursor_addid(struct q2cursor_state *state, int idtype, id_t id)
   1059 {
   1060 	KASSERT(state->numids < state->maxids);
   1061 	KASSERT(state->numkeys < state->maxkeyvals);
   1062 
   1063 	if (!state->skipfirst || state->numkeys > 0) {
   1064 		state->keys[state->numkeys].qk_idtype = idtype;
   1065 		state->keys[state->numkeys].qk_id = id;
   1066 		state->keys[state->numkeys].qk_objtype = QUOTA_OBJTYPE_BLOCKS;
   1067 		state->numkeys++;
   1068 	}
   1069 	if (state->numkeys < state->maxkeyvals) {
   1070 		state->keys[state->numkeys].qk_idtype = idtype;
   1071 		state->keys[state->numkeys].qk_id = id;
   1072 		state->keys[state->numkeys].qk_objtype = QUOTA_OBJTYPE_FILES;
   1073 		state->numkeys++;
   1074 	} else {
   1075 		KASSERT(state->skiplast);
   1076 	}
   1077 	state->numids++;
   1078 }
   1079 
   1080 /*
   1081  * Callback function for getting IDs. Update counting and call addid.
   1082  */
   1083 static int
   1084 q2cursor_getids_callback(struct ulfsmount *ump, uint64_t *offp,
   1085     struct quota2_entry *q2ep, uint64_t off, void *v)
   1086 {
   1087 	struct q2cursor_getids *gi = v;
   1088 	id_t id;
   1089 #ifdef LFS_EI
   1090 	const int needswap = ULFS_MPNEEDSWAP(ump);
   1091 #endif
   1092 
   1093 	if (gi->skipped < gi->skip) {
   1094 		gi->skipped++;
   1095 		return 0;
   1096 	}
   1097 	id = ulfs_rw32(q2ep->q2e_uid, needswap);
   1098 	q2cursor_addid(gi->state, gi->idtype, id);
   1099 	gi->new_skip++;
   1100 	if (gi->state->numids >= gi->state->maxids) {
   1101 		/* got enough ids, stop now */
   1102 		gi->stopped = 1;
   1103 		return Q2WL_ABORT;
   1104 	}
   1105 	return 0;
   1106 }
   1107 
   1108 /*
   1109  * Fill in a batch of quotakeys by scanning one or more hash chains.
   1110  */
   1111 static int
   1112 q2cursor_getkeys(struct ulfsmount *ump, int idtype, struct ulfsq2_cursor *cursor,
   1113     struct q2cursor_state *state,
   1114     int *hashsize_ret, struct quota2_entry *default_q2e_ret)
   1115 {
   1116 	const int needswap = ULFS_MPNEEDSWAP(ump);
   1117 	struct buf *hbp;
   1118 	struct quota2_header *q2h;
   1119 	int quota2_hash_size;
   1120 	struct q2cursor_getids gi;
   1121 	uint64_t offset;
   1122 	int error;
   1123 
   1124 	/*
   1125 	 * Read the header block.
   1126 	 */
   1127 
   1128 	mutex_enter(&lfs_dqlock);
   1129 	error = getq2h(ump, idtype, &hbp, &q2h, 0);
   1130 	if (error) {
   1131 		mutex_exit(&lfs_dqlock);
   1132 		return error;
   1133 	}
   1134 
   1135 	/* if the table size has changed, make the caller start over */
   1136 	quota2_hash_size = ulfs_rw16(q2h->q2h_hash_size, needswap);
   1137 	if (cursor->q2c_hashsize == 0) {
   1138 		cursor->q2c_hashsize = quota2_hash_size;
   1139 	} else if (cursor->q2c_hashsize != quota2_hash_size) {
   1140 		error = EDEADLK;
   1141 		goto scanfail;
   1142 	}
   1143 
   1144 	/* grab the entry with the default values out of the header */
   1145 	lfsquota2_ulfs_rwq2e(&q2h->q2h_defentry, default_q2e_ret, needswap);
   1146 
   1147 	/* If we haven't done the defaults yet, that goes first. */
   1148 	if (cursor->q2c_defaults_done == 0) {
   1149 		q2cursor_addid(state, idtype, QUOTA_DEFAULTID);
   1150 		/* if we read both halves, mark it done */
   1151 		if (state->numids < state->maxids || !state->skiplast) {
   1152 			cursor->q2c_defaults_done = 1;
   1153 		}
   1154 	}
   1155 
   1156 	gi.state = state;
   1157 	gi.idtype = idtype;
   1158 
   1159 	while (state->numids < state->maxids) {
   1160 		if (cursor->q2c_hashpos >= quota2_hash_size) {
   1161 			/* nothing more left */
   1162 			break;
   1163 		}
   1164 
   1165 		/* scan this hash chain */
   1166 		gi.skip = cursor->q2c_uidpos;
   1167 		gi.new_skip = gi.skip;
   1168 		gi.skipped = 0;
   1169 		gi.stopped = 0;
   1170 		offset = q2h->q2h_entries[cursor->q2c_hashpos];
   1171 
   1172 		error = quota2_walk_list(ump, hbp, idtype, &offset, 0, &gi,
   1173 		    q2cursor_getids_callback);
   1174 		KASSERT(error != Q2WL_ABORT);
   1175 		if (error) {
   1176 			break;
   1177 		}
   1178 		if (gi.stopped) {
   1179 			/* callback stopped before reading whole chain */
   1180 			cursor->q2c_uidpos = gi.new_skip;
   1181 			/* if we didn't get both halves, back up */
   1182 			if (state->numids == state->maxids && state->skiplast){
   1183 				KASSERT(cursor->q2c_uidpos > 0);
   1184 				cursor->q2c_uidpos--;
   1185 			}
   1186 		} else {
   1187 			/* read whole chain */
   1188 			/* if we got both halves of the last id, advance */
   1189 			if (state->numids < state->maxids || !state->skiplast){
   1190 				cursor->q2c_uidpos = 0;
   1191 				cursor->q2c_hashpos++;
   1192 			}
   1193 		}
   1194 	}
   1195 
   1196 scanfail:
   1197 	mutex_exit(&lfs_dqlock);
   1198 	brelse(hbp, 0);
   1199 	if (error)
   1200 		return error;
   1201 
   1202 	*hashsize_ret = quota2_hash_size;
   1203 	return 0;
   1204 }
   1205 
   1206 /*
   1207  * Fetch the quotavals for the quotakeys.
   1208  */
   1209 static int
   1210 q2cursor_getvals(struct ulfsmount *ump, struct q2cursor_state *state,
   1211     const struct quota2_entry *default_q2e)
   1212 {
   1213 	int hasid;
   1214 	id_t loadedid, id;
   1215 	unsigned pos;
   1216 	struct quota2_entry q2e;
   1217 	int objtype;
   1218 	int error;
   1219 
   1220 	hasid = 0;
   1221 	loadedid = 0;
   1222 	for (pos = 0; pos < state->numkeys; pos++) {
   1223 		id = state->keys[pos].qk_id;
   1224 		if (!hasid || id != loadedid) {
   1225 			hasid = 1;
   1226 			loadedid = id;
   1227 			if (id == QUOTA_DEFAULTID) {
   1228 				q2e = *default_q2e;
   1229 			} else {
   1230 				error = quota2_fetch_q2e(ump,
   1231 							 &state->keys[pos],
   1232 							 &q2e);
   1233 				if (error == ENOENT) {
   1234 					/* something changed - start over */
   1235 					error = EDEADLK;
   1236 				}
   1237 				if (error) {
   1238 					return error;
   1239 				}
   1240  			}
   1241 		}
   1242 
   1243 
   1244 		objtype = state->keys[pos].qk_objtype;
   1245 		KASSERT(objtype >= 0 && objtype < N_QL);
   1246 		q2val_to_quotaval(&q2e.q2e_val[objtype], &state->vals[pos]);
   1247 	}
   1248 
   1249 	return 0;
   1250 }
   1251 
   1252 /*
   1253  * Handle cursorget.
   1254  *
   1255  * We can't just read keys and values directly, because we can't walk
   1256  * the list with qdlock and grab dq_interlock to read the entries at
   1257  * the same time. So we're going to do two passes: one to figure out
   1258  * which IDs we want and fill in the keys, and then a second to use
   1259  * the keys to fetch the values.
   1260  */
   1261 int
   1262 lfsquota2_handle_cmd_cursorget(struct ulfsmount *ump, struct quotakcursor *qkc,
   1263     struct quotakey *keys, struct quotaval *vals, unsigned maxreturn,
   1264     unsigned *ret)
   1265 {
   1266 	int error;
   1267 	struct ulfsq2_cursor *cursor;
   1268 	struct ulfsq2_cursor newcursor;
   1269 	struct q2cursor_state state;
   1270 	struct quota2_entry default_q2e;
   1271 	int idtype;
   1272 	int quota2_hash_size;
   1273 
   1274 	/*
   1275 	 * Convert and validate the cursor.
   1276 	 */
   1277 	cursor = Q2CURSOR(qkc);
   1278 	error = q2cursor_check(cursor);
   1279 	if (error) {
   1280 		return error;
   1281 	}
   1282 
   1283 	/*
   1284 	 * Make sure our on-disk codes match the values of the
   1285 	 * FS-independent ones. This avoids the need for explicit
   1286 	 * conversion (which would be a NOP anyway and thus easily
   1287 	 * left out or called in the wrong places...)
   1288 	 */
   1289 	CTASSERT(QUOTA_IDTYPE_USER == ULFS_USRQUOTA);
   1290 	CTASSERT(QUOTA_IDTYPE_GROUP == ULFS_GRPQUOTA);
   1291 	CTASSERT(QUOTA_OBJTYPE_BLOCKS == QL_BLOCK);
   1292 	CTASSERT(QUOTA_OBJTYPE_FILES == QL_FILE);
   1293 
   1294 	/*
   1295 	 * If some of the idtypes aren't configured/enabled, arrange
   1296 	 * to skip over them.
   1297 	 */
   1298 	if (cursor->q2c_users_done == 0 &&
   1299 	    ump->um_quotas[ULFS_USRQUOTA] == NULLVP) {
   1300 		cursor->q2c_users_done = 1;
   1301 	}
   1302 	if (cursor->q2c_groups_done == 0 &&
   1303 	    ump->um_quotas[ULFS_GRPQUOTA] == NULLVP) {
   1304 		cursor->q2c_groups_done = 1;
   1305 	}
   1306 
   1307 	/* Loop over, potentially, both idtypes */
   1308 	while (1) {
   1309 
   1310 		/* Choose id type */
   1311 		error = q2cursor_pickidtype(cursor, &idtype);
   1312 		if (error == EAGAIN) {
   1313 			/* nothing more to do, return 0 */
   1314 			*ret = 0;
   1315 			return 0;
   1316 		}
   1317 		KASSERT(ump->um_quotas[idtype] != NULLVP);
   1318 
   1319 		/*
   1320 		 * Initialize the per-call iteration state. Copy the
   1321 		 * cursor state so we can update it in place but back
   1322 		 * out on error.
   1323 		 */
   1324 		q2cursor_initstate(&state, keys, vals, maxreturn,
   1325 				   cursor->q2c_blocks_done);
   1326 		newcursor = *cursor;
   1327 
   1328 		/* Assign keys */
   1329 		error = q2cursor_getkeys(ump, idtype, &newcursor, &state,
   1330 					 &quota2_hash_size, &default_q2e);
   1331 		if (error) {
   1332 			return error;
   1333 		}
   1334 
   1335 		/* Now fill in the values. */
   1336 		error = q2cursor_getvals(ump, &state, &default_q2e);
   1337 		if (error) {
   1338 			return error;
   1339 		}
   1340 
   1341 		/*
   1342 		 * Now that we aren't going to fail and lose what we
   1343 		 * did so far, we can update the cursor state.
   1344 		 */
   1345 
   1346 		if (newcursor.q2c_hashpos >= quota2_hash_size) {
   1347 			if (idtype == QUOTA_IDTYPE_USER)
   1348 				cursor->q2c_users_done = 1;
   1349 			else
   1350 				cursor->q2c_groups_done = 1;
   1351 
   1352 			/* start over on another id type */
   1353 			cursor->q2c_hashsize = 0;
   1354 			cursor->q2c_defaults_done = 0;
   1355 			cursor->q2c_hashpos = 0;
   1356 			cursor->q2c_uidpos = 0;
   1357 			cursor->q2c_blocks_done = 0;
   1358 		} else {
   1359 			*cursor = newcursor;
   1360 			cursor->q2c_blocks_done = state.skiplast;
   1361 		}
   1362 
   1363 		/*
   1364 		 * If we have something to return, return it.
   1365 		 * Otherwise, continue to the other idtype, if any,
   1366 		 * and only return zero at end of iteration.
   1367 		 */
   1368 		if (state.numkeys > 0) {
   1369 			break;
   1370 		}
   1371 	}
   1372 
   1373 	*ret = state.numkeys;
   1374 	return 0;
   1375 }
   1376 
   1377 int
   1378 lfsquota2_handle_cmd_cursoropen(struct ulfsmount *ump, struct quotakcursor *qkc)
   1379 {
   1380 	struct ulfsq2_cursor *cursor;
   1381 
   1382 	CTASSERT(sizeof(*cursor) <= sizeof(qkc->u.qkc_space));
   1383 	cursor = Q2CURSOR(qkc);
   1384 
   1385 	cursor->q2c_magic = Q2C_MAGIC;
   1386 	cursor->q2c_hashsize = 0;
   1387 
   1388 	cursor->q2c_users_done = 0;
   1389 	cursor->q2c_groups_done = 0;
   1390 	cursor->q2c_defaults_done = 0;
   1391 	cursor->q2c_hashpos = 0;
   1392 	cursor->q2c_uidpos = 0;
   1393 	cursor->q2c_blocks_done = 0;
   1394 	return 0;
   1395 }
   1396 
   1397 int
   1398 lfsquota2_handle_cmd_cursorclose(struct ulfsmount *ump, struct quotakcursor *qkc)
   1399 {
   1400 	struct ulfsq2_cursor *cursor;
   1401 	int error;
   1402 
   1403 	cursor = Q2CURSOR(qkc);
   1404 	error = q2cursor_check(cursor);
   1405 	if (error) {
   1406 		return error;
   1407 	}
   1408 
   1409 	/* nothing to do */
   1410 
   1411 	return 0;
   1412 }
   1413 
   1414 int
   1415 lfsquota2_handle_cmd_cursorskipidtype(struct ulfsmount *ump,
   1416     struct quotakcursor *qkc, int idtype)
   1417 {
   1418 	struct ulfsq2_cursor *cursor;
   1419 	int error;
   1420 
   1421 	cursor = Q2CURSOR(qkc);
   1422 	error = q2cursor_check(cursor);
   1423 	if (error) {
   1424 		return error;
   1425 	}
   1426 
   1427 	switch (idtype) {
   1428 	    case QUOTA_IDTYPE_USER:
   1429 		cursor->q2c_users_done = 1;
   1430 		break;
   1431 	    case QUOTA_IDTYPE_GROUP:
   1432 		cursor->q2c_groups_done = 1;
   1433 		break;
   1434 	    default:
   1435 		return EINVAL;
   1436 	}
   1437 
   1438 	return 0;
   1439 }
   1440 
   1441 int
   1442 lfsquota2_handle_cmd_cursoratend(struct ulfsmount *ump, struct quotakcursor *qkc,
   1443     int *ret)
   1444 {
   1445 	struct ulfsq2_cursor *cursor;
   1446 	int error;
   1447 
   1448 	cursor = Q2CURSOR(qkc);
   1449 	error = q2cursor_check(cursor);
   1450 	if (error) {
   1451 		return error;
   1452 	}
   1453 
   1454 	*ret = (cursor->q2c_users_done && cursor->q2c_groups_done);
   1455 	return 0;
   1456 }
   1457 
   1458 int
   1459 lfsquota2_handle_cmd_cursorrewind(struct ulfsmount *ump, struct quotakcursor *qkc)
   1460 {
   1461 	struct ulfsq2_cursor *cursor;
   1462 	int error;
   1463 
   1464 	cursor = Q2CURSOR(qkc);
   1465 	error = q2cursor_check(cursor);
   1466 	if (error) {
   1467 		return error;
   1468 	}
   1469 
   1470 	cursor->q2c_hashsize = 0;
   1471 
   1472 	cursor->q2c_users_done = 0;
   1473 	cursor->q2c_groups_done = 0;
   1474 	cursor->q2c_defaults_done = 0;
   1475 	cursor->q2c_hashpos = 0;
   1476 	cursor->q2c_uidpos = 0;
   1477 	cursor->q2c_blocks_done = 0;
   1478 
   1479 	return 0;
   1480 }
   1481 
   1482 int
   1483 lfs_q2sync(struct mount *mp)
   1484 {
   1485 	return 0;
   1486 }
   1487 
   1488 struct dq2get_callback {
   1489 	uid_t id;
   1490 	struct dquot *dq;
   1491 };
   1492 
   1493 static int
   1494 dq2get_callback(struct ulfsmount *ump, uint64_t *offp, struct quota2_entry *q2e,
   1495     uint64_t off, void *v)
   1496 {
   1497 	struct dq2get_callback *c = v;
   1498 	daddr_t lblkno;
   1499 	int blkoff;
   1500 #ifdef LFS_EI
   1501 	const int needswap = ULFS_MPNEEDSWAP(ump);
   1502 #endif
   1503 
   1504 	if (ulfs_rw32(q2e->q2e_uid, needswap) == c->id) {
   1505 		KASSERT(mutex_owned(&c->dq->dq_interlock));
   1506 		lblkno = (off >> ump->um_mountp->mnt_fs_bshift);
   1507 		blkoff = (off & ump->umq2_bmask);
   1508 		c->dq->dq2_lblkno = lblkno;
   1509 		c->dq->dq2_blkoff = blkoff;
   1510 		return Q2WL_ABORT;
   1511 	}
   1512 	return 0;
   1513 }
   1514 
   1515 int
   1516 lfs_dq2get(struct vnode *dqvp, u_long id, struct ulfsmount *ump, int type,
   1517     struct dquot *dq)
   1518 {
   1519 	struct buf *bp;
   1520 	struct quota2_header *q2h;
   1521 	int error;
   1522 	daddr_t offset;
   1523 	u_long hash_mask;
   1524 	struct dq2get_callback c = {
   1525 		.id = id,
   1526 		.dq = dq
   1527 	};
   1528 
   1529 	KASSERT(mutex_owned(&dq->dq_interlock));
   1530 	mutex_enter(&lfs_dqlock);
   1531 	error = getq2h(ump, type, &bp, &q2h, 0);
   1532 	if (error)
   1533 		goto out_mutex;
   1534 	/* look for our entry */
   1535 	hash_mask = ((1 << q2h->q2h_hash_shift) - 1);
   1536 	offset = q2h->q2h_entries[id & hash_mask];
   1537 	error = quota2_walk_list(ump, bp, type, &offset, 0, (void *)&c,
   1538 	    dq2get_callback);
   1539 	brelse(bp, 0);
   1540 out_mutex:
   1541 	mutex_exit(&lfs_dqlock);
   1542 	return error;
   1543 }
   1544 
   1545 int
   1546 lfs_dq2sync(struct vnode *vp, struct dquot *dq)
   1547 {
   1548 	return 0;
   1549 }
   1550