Home | History | Annotate | Line # | Download | only in ffs
ffs_alloc.c revision 1.157
      1 /*	$NetBSD: ffs_alloc.c,v 1.157 2017/07/12 09:30:16 hannken Exp $	*/
      2 
      3 /*-
      4  * Copyright (c) 2008, 2009 The NetBSD Foundation, Inc.
      5  * All rights reserved.
      6  *
      7  * This code is derived from software contributed to The NetBSD Foundation
      8  * by Wasabi Systems, Inc.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  * POSSIBILITY OF SUCH DAMAGE.
     30  */
     31 
     32 /*
     33  * Copyright (c) 2002 Networks Associates Technology, Inc.
     34  * All rights reserved.
     35  *
     36  * This software was developed for the FreeBSD Project by Marshall
     37  * Kirk McKusick and Network Associates Laboratories, the Security
     38  * Research Division of Network Associates, Inc. under DARPA/SPAWAR
     39  * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS
     40  * research program
     41  *
     42  * Copyright (c) 1982, 1986, 1989, 1993
     43  *	The Regents of the University of California.  All rights reserved.
     44  *
     45  * Redistribution and use in source and binary forms, with or without
     46  * modification, are permitted provided that the following conditions
     47  * are met:
     48  * 1. Redistributions of source code must retain the above copyright
     49  *    notice, this list of conditions and the following disclaimer.
     50  * 2. Redistributions in binary form must reproduce the above copyright
     51  *    notice, this list of conditions and the following disclaimer in the
     52  *    documentation and/or other materials provided with the distribution.
     53  * 3. Neither the name of the University nor the names of its contributors
     54  *    may be used to endorse or promote products derived from this software
     55  *    without specific prior written permission.
     56  *
     57  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     58  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     59  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     60  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     61  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     62  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     63  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     64  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     65  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     66  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     67  * SUCH DAMAGE.
     68  *
     69  *	@(#)ffs_alloc.c	8.19 (Berkeley) 7/13/95
     70  */
     71 
     72 #include <sys/cdefs.h>
     73 __KERNEL_RCSID(0, "$NetBSD: ffs_alloc.c,v 1.157 2017/07/12 09:30:16 hannken Exp $");
     74 
     75 #if defined(_KERNEL_OPT)
     76 #include "opt_ffs.h"
     77 #include "opt_quota.h"
     78 #include "opt_uvm_page_trkown.h"
     79 #endif
     80 
     81 #include <sys/param.h>
     82 #include <sys/systm.h>
     83 #include <sys/buf.h>
     84 #include <sys/cprng.h>
     85 #include <sys/kauth.h>
     86 #include <sys/kernel.h>
     87 #include <sys/mount.h>
     88 #include <sys/proc.h>
     89 #include <sys/syslog.h>
     90 #include <sys/vnode.h>
     91 #include <sys/wapbl.h>
     92 #include <sys/cprng.h>
     93 
     94 #include <miscfs/specfs/specdev.h>
     95 #include <ufs/ufs/quota.h>
     96 #include <ufs/ufs/ufsmount.h>
     97 #include <ufs/ufs/inode.h>
     98 #include <ufs/ufs/ufs_extern.h>
     99 #include <ufs/ufs/ufs_bswap.h>
    100 #include <ufs/ufs/ufs_wapbl.h>
    101 
    102 #include <ufs/ffs/fs.h>
    103 #include <ufs/ffs/ffs_extern.h>
    104 
    105 #ifdef UVM_PAGE_TRKOWN
    106 #include <uvm/uvm.h>
    107 #endif
    108 
    109 static daddr_t ffs_alloccg(struct inode *, int, daddr_t, int, int, int);
    110 static daddr_t ffs_alloccgblk(struct inode *, struct buf *, daddr_t, int, int);
    111 static ino_t ffs_dirpref(struct inode *);
    112 static daddr_t ffs_fragextend(struct inode *, int, daddr_t, int, int);
    113 static void ffs_fserr(struct fs *, kauth_cred_t, const char *);
    114 static daddr_t ffs_hashalloc(struct inode *, int, daddr_t, int, int, int,
    115     daddr_t (*)(struct inode *, int, daddr_t, int, int, int));
    116 static daddr_t ffs_nodealloccg(struct inode *, int, daddr_t, int, int, int);
    117 static int32_t ffs_mapsearch(struct fs *, struct cg *,
    118 				      daddr_t, int);
    119 static void ffs_blkfree_common(struct ufsmount *, struct fs *, dev_t, struct buf *,
    120     daddr_t, long, bool);
    121 static void ffs_freefile_common(struct ufsmount *, struct fs *, dev_t, struct buf *, ino_t,
    122     int, bool);
    123 
    124 /* if 1, changes in optimalization strategy are logged */
    125 int ffs_log_changeopt = 0;
    126 
    127 /* in ffs_tables.c */
    128 extern const int inside[], around[];
    129 extern const u_char * const fragtbl[];
    130 
    131 /* Basic consistency check for block allocations */
    132 static int
    133 ffs_check_bad_allocation(const char *func, struct fs *fs, daddr_t bno,
    134     long size, dev_t dev, ino_t inum)
    135 {
    136 	if ((u_int)size > fs->fs_bsize || ffs_fragoff(fs, size) != 0 ||
    137 	    ffs_fragnum(fs, bno) + ffs_numfrags(fs, size) > fs->fs_frag) {
    138 		panic("%s: bad size: dev = 0x%llx, bno = %" PRId64
    139 		    " bsize = %d, size = %ld, fs = %s", func,
    140 		    (long long)dev, bno, fs->fs_bsize, size, fs->fs_fsmnt);
    141 	}
    142 
    143 	if (bno >= fs->fs_size) {
    144 		printf("%s: bad block %" PRId64 ", ino %llu\n", func, bno,
    145 		    (unsigned long long)inum);
    146 		ffs_fserr(fs, NOCRED, "bad block");
    147 		return EINVAL;
    148 	}
    149 	return 0;
    150 }
    151 
    152 /*
    153  * Allocate a block in the file system.
    154  *
    155  * The size of the requested block is given, which must be some
    156  * multiple of fs_fsize and <= fs_bsize.
    157  * A preference may be optionally specified. If a preference is given
    158  * the following hierarchy is used to allocate a block:
    159  *   1) allocate the requested block.
    160  *   2) allocate a rotationally optimal block in the same cylinder.
    161  *   3) allocate a block in the same cylinder group.
    162  *   4) quadradically rehash into other cylinder groups, until an
    163  *      available block is located.
    164  * If no block preference is given the following hierarchy is used
    165  * to allocate a block:
    166  *   1) allocate a block in the cylinder group that contains the
    167  *      inode for the file.
    168  *   2) quadradically rehash into other cylinder groups, until an
    169  *      available block is located.
    170  *
    171  * => called with um_lock held
    172  * => releases um_lock before returning
    173  */
    174 int
    175 ffs_alloc(struct inode *ip, daddr_t lbn, daddr_t bpref, int size,
    176     int flags, kauth_cred_t cred, daddr_t *bnp)
    177 {
    178 	struct ufsmount *ump;
    179 	struct fs *fs;
    180 	daddr_t bno;
    181 	int cg;
    182 #if defined(QUOTA) || defined(QUOTA2)
    183 	int error;
    184 #endif
    185 
    186 	fs = ip->i_fs;
    187 	ump = ip->i_ump;
    188 
    189 	KASSERT(mutex_owned(&ump->um_lock));
    190 
    191 #ifdef UVM_PAGE_TRKOWN
    192 
    193 	/*
    194 	 * Sanity-check that allocations within the file size
    195 	 * do not allow other threads to read the stale contents
    196 	 * of newly allocated blocks.
    197 	 * Usually pages will exist to cover the new allocation.
    198 	 * There is an optimization in ffs_write() where we skip
    199 	 * creating pages if several conditions are met:
    200 	 *  - the file must not be mapped (in any user address space).
    201 	 *  - the write must cover whole pages and whole blocks.
    202 	 * If those conditions are not met then pages must exist and
    203 	 * be locked by the current thread.
    204 	 */
    205 
    206 	if (ITOV(ip)->v_type == VREG &&
    207 	    ffs_lblktosize(fs, (voff_t)lbn) < round_page(ITOV(ip)->v_size)) {
    208 		struct vm_page *pg;
    209 		struct vnode *vp = ITOV(ip);
    210 		struct uvm_object *uobj = &vp->v_uobj;
    211 		voff_t off = trunc_page(ffs_lblktosize(fs, lbn));
    212 		voff_t endoff = round_page(ffs_lblktosize(fs, lbn) + size);
    213 
    214 		mutex_enter(uobj->vmobjlock);
    215 		while (off < endoff) {
    216 			pg = uvm_pagelookup(uobj, off);
    217 			KASSERT((pg == NULL && (vp->v_vflag & VV_MAPPED) == 0 &&
    218 				 (size & PAGE_MASK) == 0 &&
    219 				 ffs_blkoff(fs, size) == 0) ||
    220 				(pg != NULL && pg->owner == curproc->p_pid &&
    221 				 pg->lowner == curlwp->l_lid));
    222 			off += PAGE_SIZE;
    223 		}
    224 		mutex_exit(uobj->vmobjlock);
    225 	}
    226 #endif
    227 
    228 	*bnp = 0;
    229 
    230 	KASSERTMSG((cred != NOCRED), "missing credential");
    231 	KASSERTMSG(((u_int)size <= fs->fs_bsize),
    232 	    "bad size: dev = 0x%llx, bsize = %d, size = %d, fs = %s",
    233 	    (unsigned long long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
    234 	KASSERTMSG((ffs_fragoff(fs, size) == 0),
    235 	    "bad size: dev = 0x%llx, bsize = %d, size = %d, fs = %s",
    236 	    (unsigned long long)ip->i_dev, fs->fs_bsize, size, fs->fs_fsmnt);
    237 
    238 	if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0)
    239 		goto nospace;
    240 	if (freespace(fs, fs->fs_minfree) <= 0 &&
    241 	    kauth_authorize_system(cred, KAUTH_SYSTEM_FS_RESERVEDSPACE, 0, NULL,
    242 	    NULL, NULL) != 0)
    243 		goto nospace;
    244 #if defined(QUOTA) || defined(QUOTA2)
    245 	mutex_exit(&ump->um_lock);
    246 	if ((error = chkdq(ip, btodb(size), cred, 0)) != 0)
    247 		return (error);
    248 	mutex_enter(&ump->um_lock);
    249 #endif
    250 
    251 	if (bpref >= fs->fs_size)
    252 		bpref = 0;
    253 	if (bpref == 0)
    254 		cg = ino_to_cg(fs, ip->i_number);
    255 	else
    256 		cg = dtog(fs, bpref);
    257 	bno = ffs_hashalloc(ip, cg, bpref, size, 0, flags, ffs_alloccg);
    258 	if (bno > 0) {
    259 		DIP_ADD(ip, blocks, btodb(size));
    260 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    261 		*bnp = bno;
    262 		return (0);
    263 	}
    264 #if defined(QUOTA) || defined(QUOTA2)
    265 	/*
    266 	 * Restore user's disk quota because allocation failed.
    267 	 */
    268 	(void) chkdq(ip, -btodb(size), cred, FORCE);
    269 #endif
    270 	if (flags & B_CONTIG) {
    271 		/*
    272 		 * XXX ump->um_lock handling is "suspect" at best.
    273 		 * For the case where ffs_hashalloc() fails early
    274 		 * in the B_CONTIG case we reach here with um_lock
    275 		 * already unlocked, so we can't release it again
    276 		 * like in the normal error path.  See kern/39206.
    277 		 *
    278 		 *
    279 		 * Fail silently - it's up to our caller to report
    280 		 * errors.
    281 		 */
    282 		return (ENOSPC);
    283 	}
    284 nospace:
    285 	mutex_exit(&ump->um_lock);
    286 	ffs_fserr(fs, cred, "file system full");
    287 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
    288 	return (ENOSPC);
    289 }
    290 
    291 /*
    292  * Reallocate a fragment to a bigger size
    293  *
    294  * The number and size of the old block is given, and a preference
    295  * and new size is also specified. The allocator attempts to extend
    296  * the original block. Failing that, the regular block allocator is
    297  * invoked to get an appropriate block.
    298  *
    299  * => called with um_lock held
    300  * => return with um_lock released
    301  */
    302 int
    303 ffs_realloccg(struct inode *ip, daddr_t lbprev, daddr_t bpref, int osize,
    304     int nsize, kauth_cred_t cred, struct buf **bpp, daddr_t *blknop)
    305 {
    306 	struct ufsmount *ump;
    307 	struct fs *fs;
    308 	struct buf *bp;
    309 	int cg, request, error;
    310 	daddr_t bprev, bno;
    311 
    312 	fs = ip->i_fs;
    313 	ump = ip->i_ump;
    314 
    315 	KASSERT(mutex_owned(&ump->um_lock));
    316 
    317 #ifdef UVM_PAGE_TRKOWN
    318 
    319 	/*
    320 	 * Sanity-check that allocations within the file size
    321 	 * do not allow other threads to read the stale contents
    322 	 * of newly allocated blocks.
    323 	 * Unlike in ffs_alloc(), here pages must always exist
    324 	 * for such allocations, because only the last block of a file
    325 	 * can be a fragment and ffs_write() will reallocate the
    326 	 * fragment to the new size using ufs_balloc_range(),
    327 	 * which always creates pages to cover blocks it allocates.
    328 	 */
    329 
    330 	if (ITOV(ip)->v_type == VREG) {
    331 		struct vm_page *pg;
    332 		struct uvm_object *uobj = &ITOV(ip)->v_uobj;
    333 		voff_t off = trunc_page(ffs_lblktosize(fs, lbprev));
    334 		voff_t endoff = round_page(ffs_lblktosize(fs, lbprev) + osize);
    335 
    336 		mutex_enter(uobj->vmobjlock);
    337 		while (off < endoff) {
    338 			pg = uvm_pagelookup(uobj, off);
    339 			KASSERT(pg->owner == curproc->p_pid &&
    340 				pg->lowner == curlwp->l_lid);
    341 			off += PAGE_SIZE;
    342 		}
    343 		mutex_exit(uobj->vmobjlock);
    344 	}
    345 #endif
    346 
    347 	KASSERTMSG((cred != NOCRED), "missing credential");
    348 	KASSERTMSG(((u_int)osize <= fs->fs_bsize),
    349 	    "bad size: dev=0x%llx, bsize=%d, osize=%d, nsize=%d, fs=%s",
    350 	    (unsigned long long)ip->i_dev, fs->fs_bsize, osize, nsize,
    351 	    fs->fs_fsmnt);
    352 	KASSERTMSG((ffs_fragoff(fs, osize) == 0),
    353 	    "bad size: dev=0x%llx, bsize=%d, osize=%d, nsize=%d, fs=%s",
    354 	    (unsigned long long)ip->i_dev, fs->fs_bsize, osize, nsize,
    355 	    fs->fs_fsmnt);
    356 	KASSERTMSG(((u_int)nsize <= fs->fs_bsize),
    357 	    "bad size: dev=0x%llx, bsize=%d, osize=%d, nsize=%d, fs=%s",
    358 	    (unsigned long long)ip->i_dev, fs->fs_bsize, osize, nsize,
    359 	    fs->fs_fsmnt);
    360 	KASSERTMSG((ffs_fragoff(fs, nsize) == 0),
    361 	    "bad size: dev=0x%llx, bsize=%d, osize=%d, nsize=%d, fs=%s",
    362 	    (unsigned long long)ip->i_dev, fs->fs_bsize, osize, nsize,
    363 	    fs->fs_fsmnt);
    364 
    365 	if (freespace(fs, fs->fs_minfree) <= 0 &&
    366 	    kauth_authorize_system(cred, KAUTH_SYSTEM_FS_RESERVEDSPACE, 0, NULL,
    367 	    NULL, NULL) != 0) {
    368 		mutex_exit(&ump->um_lock);
    369 		goto nospace;
    370 	}
    371 	if (fs->fs_magic == FS_UFS2_MAGIC)
    372 		bprev = ufs_rw64(ip->i_ffs2_db[lbprev], UFS_FSNEEDSWAP(fs));
    373 	else
    374 		bprev = ufs_rw32(ip->i_ffs1_db[lbprev], UFS_FSNEEDSWAP(fs));
    375 
    376 	if (bprev == 0) {
    377 		panic("%s: bad bprev: dev = 0x%llx, bsize = %d, bprev = %"
    378 		    PRId64 ", fs = %s", __func__,
    379 		    (unsigned long long)ip->i_dev, fs->fs_bsize, bprev,
    380 		    fs->fs_fsmnt);
    381 	}
    382 	mutex_exit(&ump->um_lock);
    383 
    384 	/*
    385 	 * Allocate the extra space in the buffer.
    386 	 */
    387 	if (bpp != NULL &&
    388 	    (error = bread(ITOV(ip), lbprev, osize, 0, &bp)) != 0) {
    389 		return (error);
    390 	}
    391 #if defined(QUOTA) || defined(QUOTA2)
    392 	if ((error = chkdq(ip, btodb(nsize - osize), cred, 0)) != 0) {
    393 		if (bpp != NULL) {
    394 			brelse(bp, 0);
    395 		}
    396 		return (error);
    397 	}
    398 #endif
    399 	/*
    400 	 * Check for extension in the existing location.
    401 	 */
    402 	cg = dtog(fs, bprev);
    403 	mutex_enter(&ump->um_lock);
    404 	if ((bno = ffs_fragextend(ip, cg, bprev, osize, nsize)) != 0) {
    405 		DIP_ADD(ip, blocks, btodb(nsize - osize));
    406 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    407 
    408 		if (bpp != NULL) {
    409 			if (bp->b_blkno != FFS_FSBTODB(fs, bno)) {
    410 				panic("%s: bad blockno %#llx != %#llx",
    411 				    __func__, (unsigned long long) bp->b_blkno,
    412 				    (unsigned long long)FFS_FSBTODB(fs, bno));
    413 			}
    414 			allocbuf(bp, nsize, 1);
    415 			memset((char *)bp->b_data + osize, 0, nsize - osize);
    416 			mutex_enter(bp->b_objlock);
    417 			KASSERT(!cv_has_waiters(&bp->b_done));
    418 			bp->b_oflags |= BO_DONE;
    419 			mutex_exit(bp->b_objlock);
    420 			*bpp = bp;
    421 		}
    422 		if (blknop != NULL) {
    423 			*blknop = bno;
    424 		}
    425 		return (0);
    426 	}
    427 	/*
    428 	 * Allocate a new disk location.
    429 	 */
    430 	if (bpref >= fs->fs_size)
    431 		bpref = 0;
    432 	switch ((int)fs->fs_optim) {
    433 	case FS_OPTSPACE:
    434 		/*
    435 		 * Allocate an exact sized fragment. Although this makes
    436 		 * best use of space, we will waste time relocating it if
    437 		 * the file continues to grow. If the fragmentation is
    438 		 * less than half of the minimum free reserve, we choose
    439 		 * to begin optimizing for time.
    440 		 */
    441 		request = nsize;
    442 		if (fs->fs_minfree < 5 ||
    443 		    fs->fs_cstotal.cs_nffree >
    444 		    fs->fs_dsize * fs->fs_minfree / (2 * 100))
    445 			break;
    446 
    447 		if (ffs_log_changeopt) {
    448 			log(LOG_NOTICE,
    449 				"%s: optimization changed from SPACE to TIME\n",
    450 				fs->fs_fsmnt);
    451 		}
    452 
    453 		fs->fs_optim = FS_OPTTIME;
    454 		break;
    455 	case FS_OPTTIME:
    456 		/*
    457 		 * At this point we have discovered a file that is trying to
    458 		 * grow a small fragment to a larger fragment. To save time,
    459 		 * we allocate a full sized block, then free the unused portion.
    460 		 * If the file continues to grow, the `ffs_fragextend' call
    461 		 * above will be able to grow it in place without further
    462 		 * copying. If aberrant programs cause disk fragmentation to
    463 		 * grow within 2% of the free reserve, we choose to begin
    464 		 * optimizing for space.
    465 		 */
    466 		request = fs->fs_bsize;
    467 		if (fs->fs_cstotal.cs_nffree <
    468 		    fs->fs_dsize * (fs->fs_minfree - 2) / 100)
    469 			break;
    470 
    471 		if (ffs_log_changeopt) {
    472 			log(LOG_NOTICE,
    473 				"%s: optimization changed from TIME to SPACE\n",
    474 				fs->fs_fsmnt);
    475 		}
    476 
    477 		fs->fs_optim = FS_OPTSPACE;
    478 		break;
    479 	default:
    480 		panic("%s: bad optim: dev = 0x%llx, optim = %d, fs = %s",
    481 		    __func__, (unsigned long long)ip->i_dev, fs->fs_optim,
    482 		    fs->fs_fsmnt);
    483 		/* NOTREACHED */
    484 	}
    485 	bno = ffs_hashalloc(ip, cg, bpref, request, nsize, 0, ffs_alloccg);
    486 	if (bno > 0) {
    487 		/*
    488 		 * Use forced deallocation registration, we can't handle
    489 		 * failure here. This is safe, as this place is ever hit
    490 		 * maximum once per write operation, when fragment is extended
    491 		 * to longer fragment, or a full block.
    492 		 */
    493 		if ((ip->i_ump->um_mountp->mnt_wapbl) &&
    494 		    (ITOV(ip)->v_type != VREG)) {
    495 			/* this should never fail */
    496 			error = UFS_WAPBL_REGISTER_DEALLOCATION_FORCE(
    497 			    ip->i_ump->um_mountp, FFS_FSBTODB(fs, bprev),
    498 			    osize);
    499 			if (error)
    500 				panic("ffs_realloccg: dealloc registration failed");
    501 		} else {
    502 			ffs_blkfree(fs, ip->i_devvp, bprev, (long)osize,
    503 			    ip->i_number);
    504 		}
    505 		DIP_ADD(ip, blocks, btodb(nsize - osize));
    506 		ip->i_flag |= IN_CHANGE | IN_UPDATE;
    507 		if (bpp != NULL) {
    508 			bp->b_blkno = FFS_FSBTODB(fs, bno);
    509 			allocbuf(bp, nsize, 1);
    510 			memset((char *)bp->b_data + osize, 0, (u_int)nsize - osize);
    511 			mutex_enter(bp->b_objlock);
    512 			KASSERT(!cv_has_waiters(&bp->b_done));
    513 			bp->b_oflags |= BO_DONE;
    514 			mutex_exit(bp->b_objlock);
    515 			*bpp = bp;
    516 		}
    517 		if (blknop != NULL) {
    518 			*blknop = bno;
    519 		}
    520 		return (0);
    521 	}
    522 	mutex_exit(&ump->um_lock);
    523 
    524 #if defined(QUOTA) || defined(QUOTA2)
    525 	/*
    526 	 * Restore user's disk quota because allocation failed.
    527 	 */
    528 	(void) chkdq(ip, -btodb(nsize - osize), cred, FORCE);
    529 #endif
    530 	if (bpp != NULL) {
    531 		brelse(bp, 0);
    532 	}
    533 
    534 nospace:
    535 	/*
    536 	 * no space available
    537 	 */
    538 	ffs_fserr(fs, cred, "file system full");
    539 	uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt);
    540 	return (ENOSPC);
    541 }
    542 
    543 /*
    544  * Allocate an inode in the file system.
    545  *
    546  * If allocating a directory, use ffs_dirpref to select the inode.
    547  * If allocating in a directory, the following hierarchy is followed:
    548  *   1) allocate the preferred inode.
    549  *   2) allocate an inode in the same cylinder group.
    550  *   3) quadradically rehash into other cylinder groups, until an
    551  *      available inode is located.
    552  * If no inode preference is given the following hierarchy is used
    553  * to allocate an inode:
    554  *   1) allocate an inode in cylinder group 0.
    555  *   2) quadradically rehash into other cylinder groups, until an
    556  *      available inode is located.
    557  *
    558  * => um_lock not held upon entry or return
    559  */
    560 int
    561 ffs_valloc(struct vnode *pvp, int mode, kauth_cred_t cred, ino_t *inop)
    562 {
    563 	struct ufsmount *ump;
    564 	struct inode *pip;
    565 	struct fs *fs;
    566 	ino_t ino, ipref;
    567 	int cg, error;
    568 
    569 	UFS_WAPBL_JUNLOCK_ASSERT(pvp->v_mount);
    570 
    571 	pip = VTOI(pvp);
    572 	fs = pip->i_fs;
    573 	ump = pip->i_ump;
    574 
    575 	error = UFS_WAPBL_BEGIN(pvp->v_mount);
    576 	if (error) {
    577 		return error;
    578 	}
    579 	mutex_enter(&ump->um_lock);
    580 	if (fs->fs_cstotal.cs_nifree == 0)
    581 		goto noinodes;
    582 
    583 	if ((mode & IFMT) == IFDIR)
    584 		ipref = ffs_dirpref(pip);
    585 	else
    586 		ipref = pip->i_number;
    587 	if (ipref >= fs->fs_ncg * fs->fs_ipg)
    588 		ipref = 0;
    589 	cg = ino_to_cg(fs, ipref);
    590 	/*
    591 	 * Track number of dirs created one after another
    592 	 * in a same cg without intervening by files.
    593 	 */
    594 	if ((mode & IFMT) == IFDIR) {
    595 		if (fs->fs_contigdirs[cg] < 255)
    596 			fs->fs_contigdirs[cg]++;
    597 	} else {
    598 		if (fs->fs_contigdirs[cg] > 0)
    599 			fs->fs_contigdirs[cg]--;
    600 	}
    601 	ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0, 0, ffs_nodealloccg);
    602 	if (ino == 0)
    603 		goto noinodes;
    604 	UFS_WAPBL_END(pvp->v_mount);
    605 	*inop = ino;
    606 	return 0;
    607 
    608 noinodes:
    609 	mutex_exit(&ump->um_lock);
    610 	UFS_WAPBL_END(pvp->v_mount);
    611 	ffs_fserr(fs, cred, "out of inodes");
    612 	uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt);
    613 	return ENOSPC;
    614 }
    615 
    616 /*
    617  * Find a cylinder group in which to place a directory.
    618  *
    619  * The policy implemented by this algorithm is to allocate a
    620  * directory inode in the same cylinder group as its parent
    621  * directory, but also to reserve space for its files inodes
    622  * and data. Restrict the number of directories which may be
    623  * allocated one after another in the same cylinder group
    624  * without intervening allocation of files.
    625  *
    626  * If we allocate a first level directory then force allocation
    627  * in another cylinder group.
    628  */
    629 static ino_t
    630 ffs_dirpref(struct inode *pip)
    631 {
    632 	register struct fs *fs;
    633 	int cg, prefcg;
    634 	int64_t dirsize, cgsize, curdsz;
    635 	int avgifree, avgbfree, avgndir;
    636 	int minifree, minbfree, maxndir;
    637 	int mincg, minndir;
    638 	int maxcontigdirs;
    639 
    640 	KASSERT(mutex_owned(&pip->i_ump->um_lock));
    641 
    642 	fs = pip->i_fs;
    643 
    644 	avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg;
    645 	avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
    646 	avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg;
    647 
    648 	/*
    649 	 * Force allocation in another cg if creating a first level dir.
    650 	 */
    651 	if (ITOV(pip)->v_vflag & VV_ROOT) {
    652 		prefcg = cprng_fast32() % fs->fs_ncg;
    653 		mincg = prefcg;
    654 		minndir = fs->fs_ipg;
    655 		for (cg = prefcg; cg < fs->fs_ncg; cg++)
    656 			if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
    657 			    fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
    658 			    fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    659 				mincg = cg;
    660 				minndir = fs->fs_cs(fs, cg).cs_ndir;
    661 			}
    662 		for (cg = 0; cg < prefcg; cg++)
    663 			if (fs->fs_cs(fs, cg).cs_ndir < minndir &&
    664 			    fs->fs_cs(fs, cg).cs_nifree >= avgifree &&
    665 			    fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    666 				mincg = cg;
    667 				minndir = fs->fs_cs(fs, cg).cs_ndir;
    668 			}
    669 		return ((ino_t)(fs->fs_ipg * mincg));
    670 	}
    671 
    672 	/*
    673 	 * Count various limits which used for
    674 	 * optimal allocation of a directory inode.
    675 	 * Try cylinder groups with >75% avgifree and avgbfree.
    676 	 * Avoid cylinder groups with no free blocks or inodes as that
    677 	 * triggers an I/O-expensive cylinder group scan.
    678 	 */
    679 	maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg);
    680 	minifree = avgifree - avgifree / 4;
    681 	if (minifree < 1)
    682 		minifree = 1;
    683 	minbfree = avgbfree - avgbfree / 4;
    684 	if (minbfree < 1)
    685 		minbfree = 1;
    686 	cgsize = (int64_t)fs->fs_fsize * fs->fs_fpg;
    687 	dirsize = (int64_t)fs->fs_avgfilesize * fs->fs_avgfpdir;
    688 	if (avgndir != 0) {
    689 		curdsz = (cgsize - (int64_t)avgbfree * fs->fs_bsize) / avgndir;
    690 		if (dirsize < curdsz)
    691 			dirsize = curdsz;
    692 	}
    693 	if (cgsize < dirsize * 255)
    694 		maxcontigdirs = (avgbfree * fs->fs_bsize) / dirsize;
    695 	else
    696 		maxcontigdirs = 255;
    697 	if (fs->fs_avgfpdir > 0)
    698 		maxcontigdirs = min(maxcontigdirs,
    699 				    fs->fs_ipg / fs->fs_avgfpdir);
    700 	if (maxcontigdirs == 0)
    701 		maxcontigdirs = 1;
    702 
    703 	/*
    704 	 * Limit number of dirs in one cg and reserve space for
    705 	 * regular files, but only if we have no deficit in
    706 	 * inodes or space.
    707 	 */
    708 	prefcg = ino_to_cg(fs, pip->i_number);
    709 	for (cg = prefcg; cg < fs->fs_ncg; cg++)
    710 		if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
    711 		    fs->fs_cs(fs, cg).cs_nifree >= minifree &&
    712 	    	    fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
    713 			if (fs->fs_contigdirs[cg] < maxcontigdirs)
    714 				return ((ino_t)(fs->fs_ipg * cg));
    715 		}
    716 	for (cg = 0; cg < prefcg; cg++)
    717 		if (fs->fs_cs(fs, cg).cs_ndir < maxndir &&
    718 		    fs->fs_cs(fs, cg).cs_nifree >= minifree &&
    719 	    	    fs->fs_cs(fs, cg).cs_nbfree >= minbfree) {
    720 			if (fs->fs_contigdirs[cg] < maxcontigdirs)
    721 				return ((ino_t)(fs->fs_ipg * cg));
    722 		}
    723 	/*
    724 	 * This is a backstop when we are deficient in space.
    725 	 */
    726 	for (cg = prefcg; cg < fs->fs_ncg; cg++)
    727 		if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
    728 			return ((ino_t)(fs->fs_ipg * cg));
    729 	for (cg = 0; cg < prefcg; cg++)
    730 		if (fs->fs_cs(fs, cg).cs_nifree >= avgifree)
    731 			break;
    732 	return ((ino_t)(fs->fs_ipg * cg));
    733 }
    734 
    735 /*
    736  * Select the desired position for the next block in a file.  The file is
    737  * logically divided into sections. The first section is composed of the
    738  * direct blocks. Each additional section contains fs_maxbpg blocks.
    739  *
    740  * If no blocks have been allocated in the first section, the policy is to
    741  * request a block in the same cylinder group as the inode that describes
    742  * the file. If no blocks have been allocated in any other section, the
    743  * policy is to place the section in a cylinder group with a greater than
    744  * average number of free blocks.  An appropriate cylinder group is found
    745  * by using a rotor that sweeps the cylinder groups. When a new group of
    746  * blocks is needed, the sweep begins in the cylinder group following the
    747  * cylinder group from which the previous allocation was made. The sweep
    748  * continues until a cylinder group with greater than the average number
    749  * of free blocks is found. If the allocation is for the first block in an
    750  * indirect block, the information on the previous allocation is unavailable;
    751  * here a best guess is made based upon the logical block number being
    752  * allocated.
    753  *
    754  * If a section is already partially allocated, the policy is to
    755  * contiguously allocate fs_maxcontig blocks.  The end of one of these
    756  * contiguous blocks and the beginning of the next is laid out
    757  * contigously if possible.
    758  *
    759  * => um_lock held on entry and exit
    760  */
    761 daddr_t
    762 ffs_blkpref_ufs1(struct inode *ip, daddr_t lbn, int indx, int flags,
    763     int32_t *bap /* XXX ondisk32 */)
    764 {
    765 	struct fs *fs;
    766 	int cg;
    767 	int avgbfree, startcg;
    768 
    769 	KASSERT(mutex_owned(&ip->i_ump->um_lock));
    770 
    771 	fs = ip->i_fs;
    772 
    773 	/*
    774 	 * If allocating a contiguous file with B_CONTIG, use the hints
    775 	 * in the inode extentions to return the desired block.
    776 	 *
    777 	 * For metadata (indirect blocks) return the address of where
    778 	 * the first indirect block resides - we'll scan for the next
    779 	 * available slot if we need to allocate more than one indirect
    780 	 * block.  For data, return the address of the actual block
    781 	 * relative to the address of the first data block.
    782 	 */
    783 	if (flags & B_CONTIG) {
    784 		KASSERT(ip->i_ffs_first_data_blk != 0);
    785 		KASSERT(ip->i_ffs_first_indir_blk != 0);
    786 		if (flags & B_METAONLY)
    787 			return ip->i_ffs_first_indir_blk;
    788 		else
    789 			return ip->i_ffs_first_data_blk + ffs_blkstofrags(fs, lbn);
    790 	}
    791 
    792 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
    793 		if (lbn < UFS_NDADDR + FFS_NINDIR(fs)) {
    794 			cg = ino_to_cg(fs, ip->i_number);
    795 			return (cgbase(fs, cg) + fs->fs_frag);
    796 		}
    797 		/*
    798 		 * Find a cylinder with greater than average number of
    799 		 * unused data blocks.
    800 		 */
    801 		if (indx == 0 || bap[indx - 1] == 0)
    802 			startcg =
    803 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
    804 		else
    805 			startcg = dtog(fs,
    806 				ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
    807 		startcg %= fs->fs_ncg;
    808 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
    809 		for (cg = startcg; cg < fs->fs_ncg; cg++)
    810 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    811 				return (cgbase(fs, cg) + fs->fs_frag);
    812 			}
    813 		for (cg = 0; cg < startcg; cg++)
    814 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    815 				return (cgbase(fs, cg) + fs->fs_frag);
    816 			}
    817 		return (0);
    818 	}
    819 	/*
    820 	 * We just always try to lay things out contiguously.
    821 	 */
    822 	return ufs_rw32(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
    823 }
    824 
    825 daddr_t
    826 ffs_blkpref_ufs2(struct inode *ip, daddr_t lbn, int indx, int flags,
    827     int64_t *bap)
    828 {
    829 	struct fs *fs;
    830 	int cg;
    831 	int avgbfree, startcg;
    832 
    833 	KASSERT(mutex_owned(&ip->i_ump->um_lock));
    834 
    835 	fs = ip->i_fs;
    836 
    837 	/*
    838 	 * If allocating a contiguous file with B_CONTIG, use the hints
    839 	 * in the inode extentions to return the desired block.
    840 	 *
    841 	 * For metadata (indirect blocks) return the address of where
    842 	 * the first indirect block resides - we'll scan for the next
    843 	 * available slot if we need to allocate more than one indirect
    844 	 * block.  For data, return the address of the actual block
    845 	 * relative to the address of the first data block.
    846 	 */
    847 	if (flags & B_CONTIG) {
    848 		KASSERT(ip->i_ffs_first_data_blk != 0);
    849 		KASSERT(ip->i_ffs_first_indir_blk != 0);
    850 		if (flags & B_METAONLY)
    851 			return ip->i_ffs_first_indir_blk;
    852 		else
    853 			return ip->i_ffs_first_data_blk + ffs_blkstofrags(fs, lbn);
    854 	}
    855 
    856 	if (indx % fs->fs_maxbpg == 0 || bap[indx - 1] == 0) {
    857 		if (lbn < UFS_NDADDR + FFS_NINDIR(fs)) {
    858 			cg = ino_to_cg(fs, ip->i_number);
    859 			return (cgbase(fs, cg) + fs->fs_frag);
    860 		}
    861 		/*
    862 		 * Find a cylinder with greater than average number of
    863 		 * unused data blocks.
    864 		 */
    865 		if (indx == 0 || bap[indx - 1] == 0)
    866 			startcg =
    867 			    ino_to_cg(fs, ip->i_number) + lbn / fs->fs_maxbpg;
    868 		else
    869 			startcg = dtog(fs,
    870 				ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + 1);
    871 		startcg %= fs->fs_ncg;
    872 		avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg;
    873 		for (cg = startcg; cg < fs->fs_ncg; cg++)
    874 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    875 				return (cgbase(fs, cg) + fs->fs_frag);
    876 			}
    877 		for (cg = 0; cg < startcg; cg++)
    878 			if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) {
    879 				return (cgbase(fs, cg) + fs->fs_frag);
    880 			}
    881 		return (0);
    882 	}
    883 	/*
    884 	 * We just always try to lay things out contiguously.
    885 	 */
    886 	return ufs_rw64(bap[indx - 1], UFS_FSNEEDSWAP(fs)) + fs->fs_frag;
    887 }
    888 
    889 
    890 /*
    891  * Implement the cylinder overflow algorithm.
    892  *
    893  * The policy implemented by this algorithm is:
    894  *   1) allocate the block in its requested cylinder group.
    895  *   2) quadradically rehash on the cylinder group number.
    896  *   3) brute force search for a free block.
    897  *
    898  * => called with um_lock held
    899  * => returns with um_lock released on success, held on failure
    900  *    (*allocator releases lock on success, retains lock on failure)
    901  */
    902 /*VARARGS5*/
    903 static daddr_t
    904 ffs_hashalloc(struct inode *ip, int cg, daddr_t pref,
    905     int size /* size for data blocks, mode for inodes */,
    906     int realsize,
    907     int flags,
    908     daddr_t (*allocator)(struct inode *, int, daddr_t, int, int, int))
    909 {
    910 	struct fs *fs;
    911 	daddr_t result;
    912 	int i, icg = cg;
    913 
    914 	fs = ip->i_fs;
    915 	/*
    916 	 * 1: preferred cylinder group
    917 	 */
    918 	result = (*allocator)(ip, cg, pref, size, realsize, flags);
    919 	if (result)
    920 		return (result);
    921 
    922 	if (flags & B_CONTIG)
    923 		return (result);
    924 	/*
    925 	 * 2: quadratic rehash
    926 	 */
    927 	for (i = 1; i < fs->fs_ncg; i *= 2) {
    928 		cg += i;
    929 		if (cg >= fs->fs_ncg)
    930 			cg -= fs->fs_ncg;
    931 		result = (*allocator)(ip, cg, 0, size, realsize, flags);
    932 		if (result)
    933 			return (result);
    934 	}
    935 	/*
    936 	 * 3: brute force search
    937 	 * Note that we start at i == 2, since 0 was checked initially,
    938 	 * and 1 is always checked in the quadratic rehash.
    939 	 */
    940 	cg = (icg + 2) % fs->fs_ncg;
    941 	for (i = 2; i < fs->fs_ncg; i++) {
    942 		result = (*allocator)(ip, cg, 0, size, realsize, flags);
    943 		if (result)
    944 			return (result);
    945 		cg++;
    946 		if (cg == fs->fs_ncg)
    947 			cg = 0;
    948 	}
    949 	return (0);
    950 }
    951 
    952 /*
    953  * Determine whether a fragment can be extended.
    954  *
    955  * Check to see if the necessary fragments are available, and
    956  * if they are, allocate them.
    957  *
    958  * => called with um_lock held
    959  * => returns with um_lock released on success, held on failure
    960  */
    961 static daddr_t
    962 ffs_fragextend(struct inode *ip, int cg, daddr_t bprev, int osize, int nsize)
    963 {
    964 	struct ufsmount *ump;
    965 	struct fs *fs;
    966 	struct cg *cgp;
    967 	struct buf *bp;
    968 	daddr_t bno;
    969 	int frags, bbase;
    970 	int i, error;
    971 	u_int8_t *blksfree;
    972 
    973 	fs = ip->i_fs;
    974 	ump = ip->i_ump;
    975 
    976 	KASSERT(mutex_owned(&ump->um_lock));
    977 
    978 	if (fs->fs_cs(fs, cg).cs_nffree < ffs_numfrags(fs, nsize - osize))
    979 		return (0);
    980 	frags = ffs_numfrags(fs, nsize);
    981 	bbase = ffs_fragnum(fs, bprev);
    982 	if (bbase > ffs_fragnum(fs, (bprev + frags - 1))) {
    983 		/* cannot extend across a block boundary */
    984 		return (0);
    985 	}
    986 	mutex_exit(&ump->um_lock);
    987 	error = bread(ip->i_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
    988 		(int)fs->fs_cgsize, B_MODIFY, &bp);
    989 	if (error)
    990 		goto fail;
    991 	cgp = (struct cg *)bp->b_data;
    992 	if (!cg_chkmagic(cgp, UFS_FSNEEDSWAP(fs)))
    993 		goto fail;
    994 	cgp->cg_old_time = ufs_rw32(time_second, UFS_FSNEEDSWAP(fs));
    995 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
    996 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
    997 		cgp->cg_time = ufs_rw64(time_second, UFS_FSNEEDSWAP(fs));
    998 	bno = dtogd(fs, bprev);
    999 	blksfree = cg_blksfree(cgp, UFS_FSNEEDSWAP(fs));
   1000 	for (i = ffs_numfrags(fs, osize); i < frags; i++)
   1001 		if (isclr(blksfree, bno + i))
   1002 			goto fail;
   1003 	/*
   1004 	 * the current fragment can be extended
   1005 	 * deduct the count on fragment being extended into
   1006 	 * increase the count on the remaining fragment (if any)
   1007 	 * allocate the extended piece
   1008 	 */
   1009 	for (i = frags; i < fs->fs_frag - bbase; i++)
   1010 		if (isclr(blksfree, bno + i))
   1011 			break;
   1012 	ufs_add32(cgp->cg_frsum[i - ffs_numfrags(fs, osize)], -1, UFS_FSNEEDSWAP(fs));
   1013 	if (i != frags)
   1014 		ufs_add32(cgp->cg_frsum[i - frags], 1, UFS_FSNEEDSWAP(fs));
   1015 	mutex_enter(&ump->um_lock);
   1016 	for (i = ffs_numfrags(fs, osize); i < frags; i++) {
   1017 		clrbit(blksfree, bno + i);
   1018 		ufs_add32(cgp->cg_cs.cs_nffree, -1, UFS_FSNEEDSWAP(fs));
   1019 		fs->fs_cstotal.cs_nffree--;
   1020 		fs->fs_cs(fs, cg).cs_nffree--;
   1021 	}
   1022 	fs->fs_fmod = 1;
   1023 	ACTIVECG_CLR(fs, cg);
   1024 	mutex_exit(&ump->um_lock);
   1025 	bdwrite(bp);
   1026 	return (bprev);
   1027 
   1028  fail:
   1029  	if (bp != NULL)
   1030 		brelse(bp, 0);
   1031  	mutex_enter(&ump->um_lock);
   1032  	return (0);
   1033 }
   1034 
   1035 /*
   1036  * Determine whether a block can be allocated.
   1037  *
   1038  * Check to see if a block of the appropriate size is available,
   1039  * and if it is, allocate it.
   1040  */
   1041 static daddr_t
   1042 ffs_alloccg(struct inode *ip, int cg, daddr_t bpref, int size, int realsize,
   1043     int flags)
   1044 {
   1045 	struct ufsmount *ump;
   1046 	struct fs *fs = ip->i_fs;
   1047 	struct cg *cgp;
   1048 	struct buf *bp;
   1049 	int32_t bno;
   1050 	daddr_t blkno;
   1051 	int error, frags, allocsiz, i;
   1052 	u_int8_t *blksfree;
   1053 	const int needswap = UFS_FSNEEDSWAP(fs);
   1054 
   1055 	ump = ip->i_ump;
   1056 
   1057 	KASSERT(mutex_owned(&ump->um_lock));
   1058 
   1059 	if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize)
   1060 		return (0);
   1061 	mutex_exit(&ump->um_lock);
   1062 	error = bread(ip->i_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
   1063 		(int)fs->fs_cgsize, B_MODIFY, &bp);
   1064 	if (error)
   1065 		goto fail;
   1066 	cgp = (struct cg *)bp->b_data;
   1067 	if (!cg_chkmagic(cgp, needswap) ||
   1068 	    (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize))
   1069 		goto fail;
   1070 	cgp->cg_old_time = ufs_rw32(time_second, needswap);
   1071 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
   1072 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
   1073 		cgp->cg_time = ufs_rw64(time_second, needswap);
   1074 	if (size == fs->fs_bsize) {
   1075 		mutex_enter(&ump->um_lock);
   1076 		blkno = ffs_alloccgblk(ip, bp, bpref, realsize, flags);
   1077 		ACTIVECG_CLR(fs, cg);
   1078 		mutex_exit(&ump->um_lock);
   1079 
   1080 		/*
   1081 		 * If actually needed size is lower, free the extra blocks now.
   1082 		 * This is safe to call here, there is no outside reference
   1083 		 * to this block yet. It is not necessary to keep um_lock
   1084 		 * locked.
   1085 		 */
   1086 		if (realsize != 0 && realsize < size) {
   1087 			ffs_blkfree_common(ip->i_ump, ip->i_fs,
   1088 			    ip->i_devvp->v_rdev,
   1089 			    bp, blkno + ffs_numfrags(fs, realsize),
   1090 			    (long)(size - realsize), false);
   1091 		}
   1092 
   1093 		bdwrite(bp);
   1094 		return (blkno);
   1095 	}
   1096 	/*
   1097 	 * check to see if any fragments are already available
   1098 	 * allocsiz is the size which will be allocated, hacking
   1099 	 * it down to a smaller size if necessary
   1100 	 */
   1101 	blksfree = cg_blksfree(cgp, needswap);
   1102 	frags = ffs_numfrags(fs, size);
   1103 	for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++)
   1104 		if (cgp->cg_frsum[allocsiz] != 0)
   1105 			break;
   1106 	if (allocsiz == fs->fs_frag) {
   1107 		/*
   1108 		 * no fragments were available, so a block will be
   1109 		 * allocated, and hacked up
   1110 		 */
   1111 		if (cgp->cg_cs.cs_nbfree == 0)
   1112 			goto fail;
   1113 		mutex_enter(&ump->um_lock);
   1114 		blkno = ffs_alloccgblk(ip, bp, bpref, realsize, flags);
   1115 		bno = dtogd(fs, blkno);
   1116 		for (i = frags; i < fs->fs_frag; i++)
   1117 			setbit(blksfree, bno + i);
   1118 		i = fs->fs_frag - frags;
   1119 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
   1120 		fs->fs_cstotal.cs_nffree += i;
   1121 		fs->fs_cs(fs, cg).cs_nffree += i;
   1122 		fs->fs_fmod = 1;
   1123 		ufs_add32(cgp->cg_frsum[i], 1, needswap);
   1124 		ACTIVECG_CLR(fs, cg);
   1125 		mutex_exit(&ump->um_lock);
   1126 		bdwrite(bp);
   1127 		return (blkno);
   1128 	}
   1129 	bno = ffs_mapsearch(fs, cgp, bpref, allocsiz);
   1130 #if 0
   1131 	/*
   1132 	 * XXX fvdl mapsearch will panic, and never return -1
   1133 	 *          also: returning NULL as daddr_t ?
   1134 	 */
   1135 	if (bno < 0)
   1136 		goto fail;
   1137 #endif
   1138 	for (i = 0; i < frags; i++)
   1139 		clrbit(blksfree, bno + i);
   1140 	mutex_enter(&ump->um_lock);
   1141 	ufs_add32(cgp->cg_cs.cs_nffree, -frags, needswap);
   1142 	fs->fs_cstotal.cs_nffree -= frags;
   1143 	fs->fs_cs(fs, cg).cs_nffree -= frags;
   1144 	fs->fs_fmod = 1;
   1145 	ufs_add32(cgp->cg_frsum[allocsiz], -1, needswap);
   1146 	if (frags != allocsiz)
   1147 		ufs_add32(cgp->cg_frsum[allocsiz - frags], 1, needswap);
   1148 	blkno = cgbase(fs, cg) + bno;
   1149 	ACTIVECG_CLR(fs, cg);
   1150 	mutex_exit(&ump->um_lock);
   1151 	bdwrite(bp);
   1152 	return blkno;
   1153 
   1154  fail:
   1155  	if (bp != NULL)
   1156 		brelse(bp, 0);
   1157  	mutex_enter(&ump->um_lock);
   1158  	return (0);
   1159 }
   1160 
   1161 /*
   1162  * Allocate a block in a cylinder group.
   1163  *
   1164  * This algorithm implements the following policy:
   1165  *   1) allocate the requested block.
   1166  *   2) allocate a rotationally optimal block in the same cylinder.
   1167  *   3) allocate the next available block on the block rotor for the
   1168  *      specified cylinder group.
   1169  * Note that this routine only allocates fs_bsize blocks; these
   1170  * blocks may be fragmented by the routine that allocates them.
   1171  */
   1172 static daddr_t
   1173 ffs_alloccgblk(struct inode *ip, struct buf *bp, daddr_t bpref, int realsize,
   1174     int flags)
   1175 {
   1176 	struct fs *fs = ip->i_fs;
   1177 	struct cg *cgp;
   1178 	int cg;
   1179 	daddr_t blkno;
   1180 	int32_t bno;
   1181 	u_int8_t *blksfree;
   1182 	const int needswap = UFS_FSNEEDSWAP(fs);
   1183 
   1184 	KASSERT(mutex_owned(&ip->i_ump->um_lock));
   1185 
   1186 	cgp = (struct cg *)bp->b_data;
   1187 	blksfree = cg_blksfree(cgp, needswap);
   1188 	if (bpref == 0 || dtog(fs, bpref) != ufs_rw32(cgp->cg_cgx, needswap)) {
   1189 		bpref = ufs_rw32(cgp->cg_rotor, needswap);
   1190 	} else {
   1191 		bpref = ffs_blknum(fs, bpref);
   1192 		bno = dtogd(fs, bpref);
   1193 		/*
   1194 		 * if the requested block is available, use it
   1195 		 */
   1196 		if (ffs_isblock(fs, blksfree, ffs_fragstoblks(fs, bno)))
   1197 			goto gotit;
   1198 		/*
   1199 		 * if the requested data block isn't available and we are
   1200 		 * trying to allocate a contiguous file, return an error.
   1201 		 */
   1202 		if ((flags & (B_CONTIG | B_METAONLY)) == B_CONTIG)
   1203 			return (0);
   1204 	}
   1205 
   1206 	/*
   1207 	 * Take the next available block in this cylinder group.
   1208 	 */
   1209 	bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag);
   1210 #if 0
   1211 	/*
   1212 	 * XXX jdolecek ffs_mapsearch() succeeds or panics
   1213 	 */
   1214 	if (bno < 0)
   1215 		return (0);
   1216 #endif
   1217 	cgp->cg_rotor = ufs_rw32(bno, needswap);
   1218 gotit:
   1219 	blkno = ffs_fragstoblks(fs, bno);
   1220 	ffs_clrblock(fs, blksfree, blkno);
   1221 	ffs_clusteracct(fs, cgp, blkno, -1);
   1222 	ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
   1223 	fs->fs_cstotal.cs_nbfree--;
   1224 	fs->fs_cs(fs, ufs_rw32(cgp->cg_cgx, needswap)).cs_nbfree--;
   1225 	if ((fs->fs_magic == FS_UFS1_MAGIC) &&
   1226 	    ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) {
   1227 		int cylno;
   1228 		cylno = old_cbtocylno(fs, bno);
   1229 		KASSERT(cylno >= 0);
   1230 		KASSERT(cylno < fs->fs_old_ncyl);
   1231 		KASSERT(old_cbtorpos(fs, bno) >= 0);
   1232 		KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, bno) < fs->fs_old_nrpos);
   1233 		ufs_add16(old_cg_blks(fs, cgp, cylno, needswap)[old_cbtorpos(fs, bno)], -1,
   1234 		    needswap);
   1235 		ufs_add32(old_cg_blktot(cgp, needswap)[cylno], -1, needswap);
   1236 	}
   1237 	fs->fs_fmod = 1;
   1238 	cg = ufs_rw32(cgp->cg_cgx, needswap);
   1239 	blkno = cgbase(fs, cg) + bno;
   1240 	return (blkno);
   1241 }
   1242 
   1243 /*
   1244  * Determine whether an inode can be allocated.
   1245  *
   1246  * Check to see if an inode is available, and if it is,
   1247  * allocate it using the following policy:
   1248  *   1) allocate the requested inode.
   1249  *   2) allocate the next available inode after the requested
   1250  *      inode in the specified cylinder group.
   1251  */
   1252 static daddr_t
   1253 ffs_nodealloccg(struct inode *ip, int cg, daddr_t ipref, int mode, int realsize,
   1254     int flags)
   1255 {
   1256 	struct ufsmount *ump = ip->i_ump;
   1257 	struct fs *fs = ip->i_fs;
   1258 	struct cg *cgp;
   1259 	struct buf *bp, *ibp;
   1260 	u_int8_t *inosused;
   1261 	int error, start, len, loc, map, i;
   1262 	int32_t initediblk;
   1263 	daddr_t nalloc;
   1264 	struct ufs2_dinode *dp2;
   1265 	const int needswap = UFS_FSNEEDSWAP(fs);
   1266 
   1267 	KASSERT(mutex_owned(&ump->um_lock));
   1268 	UFS_WAPBL_JLOCK_ASSERT(ip->i_ump->um_mountp);
   1269 
   1270 	if (fs->fs_cs(fs, cg).cs_nifree == 0)
   1271 		return (0);
   1272 	mutex_exit(&ump->um_lock);
   1273 	ibp = NULL;
   1274 	initediblk = -1;
   1275 retry:
   1276 	error = bread(ip->i_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
   1277 		(int)fs->fs_cgsize, B_MODIFY, &bp);
   1278 	if (error)
   1279 		goto fail;
   1280 	cgp = (struct cg *)bp->b_data;
   1281 	if (!cg_chkmagic(cgp, needswap) || cgp->cg_cs.cs_nifree == 0)
   1282 		goto fail;
   1283 
   1284 	if (ibp != NULL &&
   1285 	    initediblk != ufs_rw32(cgp->cg_initediblk, needswap)) {
   1286 		/* Another thread allocated more inodes so we retry the test. */
   1287 		brelse(ibp, 0);
   1288 		ibp = NULL;
   1289 	}
   1290 	/*
   1291 	 * Check to see if we need to initialize more inodes.
   1292 	 */
   1293 	if (fs->fs_magic == FS_UFS2_MAGIC && ibp == NULL) {
   1294 		initediblk = ufs_rw32(cgp->cg_initediblk, needswap);
   1295 		nalloc = fs->fs_ipg - ufs_rw32(cgp->cg_cs.cs_nifree, needswap);
   1296 		if (nalloc + FFS_INOPB(fs) > initediblk &&
   1297 		    initediblk < ufs_rw32(cgp->cg_niblk, needswap)) {
   1298 			/*
   1299 			 * We have to release the cg buffer here to prevent
   1300 			 * a deadlock when reading the inode block will
   1301 			 * run a copy-on-write that might use this cg.
   1302 			 */
   1303 			brelse(bp, 0);
   1304 			bp = NULL;
   1305 			error = ffs_getblk(ip->i_devvp, FFS_FSBTODB(fs,
   1306 			    ino_to_fsba(fs, cg * fs->fs_ipg + initediblk)),
   1307 			    FFS_NOBLK, fs->fs_bsize, false, &ibp);
   1308 			if (error)
   1309 				goto fail;
   1310 			goto retry;
   1311 		}
   1312 	}
   1313 
   1314 	cgp->cg_old_time = ufs_rw32(time_second, needswap);
   1315 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
   1316 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
   1317 		cgp->cg_time = ufs_rw64(time_second, needswap);
   1318 	inosused = cg_inosused(cgp, needswap);
   1319 	if (ipref) {
   1320 		ipref %= fs->fs_ipg;
   1321 		if (isclr(inosused, ipref))
   1322 			goto gotit;
   1323 	}
   1324 	start = ufs_rw32(cgp->cg_irotor, needswap) / NBBY;
   1325 	len = howmany(fs->fs_ipg - ufs_rw32(cgp->cg_irotor, needswap),
   1326 		NBBY);
   1327 	loc = skpc(0xff, len, &inosused[start]);
   1328 	if (loc == 0) {
   1329 		len = start + 1;
   1330 		start = 0;
   1331 		loc = skpc(0xff, len, &inosused[0]);
   1332 		if (loc == 0) {
   1333 			panic("%s: map corrupted: cg=%d, irotor=%d, fs=%s",
   1334 			    __func__, cg, ufs_rw32(cgp->cg_irotor, needswap),
   1335 			    fs->fs_fsmnt);
   1336 			/* NOTREACHED */
   1337 		}
   1338 	}
   1339 	i = start + len - loc;
   1340 	map = inosused[i] ^ 0xff;
   1341 	if (map == 0) {
   1342 		panic("%s: block not in map: fs=%s", __func__, fs->fs_fsmnt);
   1343 	}
   1344 	ipref = i * NBBY + ffs(map) - 1;
   1345 	cgp->cg_irotor = ufs_rw32(ipref, needswap);
   1346 gotit:
   1347 	UFS_WAPBL_REGISTER_INODE(ip->i_ump->um_mountp, cg * fs->fs_ipg + ipref,
   1348 	    mode);
   1349 	/*
   1350 	 * Check to see if we need to initialize more inodes.
   1351 	 */
   1352 	if (ibp != NULL) {
   1353 		KASSERT(initediblk == ufs_rw32(cgp->cg_initediblk, needswap));
   1354 		memset(ibp->b_data, 0, fs->fs_bsize);
   1355 		dp2 = (struct ufs2_dinode *)(ibp->b_data);
   1356 		for (i = 0; i < FFS_INOPB(fs); i++) {
   1357 			/*
   1358 			 * Don't bother to swap, it's supposed to be
   1359 			 * random, after all.
   1360 			 */
   1361 			dp2->di_gen = (cprng_fast32() & INT32_MAX) / 2 + 1;
   1362 			dp2++;
   1363 		}
   1364 		initediblk += FFS_INOPB(fs);
   1365 		cgp->cg_initediblk = ufs_rw32(initediblk, needswap);
   1366 	}
   1367 
   1368 	mutex_enter(&ump->um_lock);
   1369 	ACTIVECG_CLR(fs, cg);
   1370 	setbit(inosused, ipref);
   1371 	ufs_add32(cgp->cg_cs.cs_nifree, -1, needswap);
   1372 	fs->fs_cstotal.cs_nifree--;
   1373 	fs->fs_cs(fs, cg).cs_nifree--;
   1374 	fs->fs_fmod = 1;
   1375 	if ((mode & IFMT) == IFDIR) {
   1376 		ufs_add32(cgp->cg_cs.cs_ndir, 1, needswap);
   1377 		fs->fs_cstotal.cs_ndir++;
   1378 		fs->fs_cs(fs, cg).cs_ndir++;
   1379 	}
   1380 	mutex_exit(&ump->um_lock);
   1381 	if (ibp != NULL) {
   1382 		bwrite(ibp);
   1383 		bwrite(bp);
   1384 	} else
   1385 		bdwrite(bp);
   1386 	return (cg * fs->fs_ipg + ipref);
   1387  fail:
   1388 	if (bp != NULL)
   1389 		brelse(bp, 0);
   1390 	if (ibp != NULL)
   1391 		brelse(ibp, 0);
   1392 	mutex_enter(&ump->um_lock);
   1393 	return (0);
   1394 }
   1395 
   1396 /*
   1397  * Allocate a block or fragment.
   1398  *
   1399  * The specified block or fragment is removed from the
   1400  * free map, possibly fragmenting a block in the process.
   1401  *
   1402  * This implementation should mirror fs_blkfree
   1403  *
   1404  * => um_lock not held on entry or exit
   1405  */
   1406 int
   1407 ffs_blkalloc(struct inode *ip, daddr_t bno, long size)
   1408 {
   1409 	int error;
   1410 
   1411 	error = ffs_check_bad_allocation(__func__, ip->i_fs, bno, size,
   1412 	    ip->i_dev, ip->i_uid);
   1413 	if (error)
   1414 		return error;
   1415 
   1416 	return ffs_blkalloc_ump(ip->i_ump, bno, size);
   1417 }
   1418 
   1419 int
   1420 ffs_blkalloc_ump(struct ufsmount *ump, daddr_t bno, long size)
   1421 {
   1422 	struct fs *fs = ump->um_fs;
   1423 	struct cg *cgp;
   1424 	struct buf *bp;
   1425 	int32_t fragno, cgbno;
   1426 	int i, error, cg, blk, frags, bbase;
   1427 	u_int8_t *blksfree;
   1428 	const int needswap = UFS_FSNEEDSWAP(fs);
   1429 
   1430 	KASSERT((u_int)size <= fs->fs_bsize && ffs_fragoff(fs, size) == 0 &&
   1431 	    ffs_fragnum(fs, bno) + ffs_numfrags(fs, size) <= fs->fs_frag);
   1432 	KASSERT(bno < fs->fs_size);
   1433 
   1434 	cg = dtog(fs, bno);
   1435 	error = bread(ump->um_devvp, FFS_FSBTODB(fs, cgtod(fs, cg)),
   1436 		(int)fs->fs_cgsize, B_MODIFY, &bp);
   1437 	if (error) {
   1438 		return error;
   1439 	}
   1440 	cgp = (struct cg *)bp->b_data;
   1441 	if (!cg_chkmagic(cgp, needswap)) {
   1442 		brelse(bp, 0);
   1443 		return EIO;
   1444 	}
   1445 	cgp->cg_old_time = ufs_rw32(time_second, needswap);
   1446 	cgp->cg_time = ufs_rw64(time_second, needswap);
   1447 	cgbno = dtogd(fs, bno);
   1448 	blksfree = cg_blksfree(cgp, needswap);
   1449 
   1450 	mutex_enter(&ump->um_lock);
   1451 	if (size == fs->fs_bsize) {
   1452 		fragno = ffs_fragstoblks(fs, cgbno);
   1453 		if (!ffs_isblock(fs, blksfree, fragno)) {
   1454 			mutex_exit(&ump->um_lock);
   1455 			brelse(bp, 0);
   1456 			return EBUSY;
   1457 		}
   1458 		ffs_clrblock(fs, blksfree, fragno);
   1459 		ffs_clusteracct(fs, cgp, fragno, -1);
   1460 		ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
   1461 		fs->fs_cstotal.cs_nbfree--;
   1462 		fs->fs_cs(fs, cg).cs_nbfree--;
   1463 	} else {
   1464 		bbase = cgbno - ffs_fragnum(fs, cgbno);
   1465 
   1466 		frags = ffs_numfrags(fs, size);
   1467 		for (i = 0; i < frags; i++) {
   1468 			if (isclr(blksfree, cgbno + i)) {
   1469 				mutex_exit(&ump->um_lock);
   1470 				brelse(bp, 0);
   1471 				return EBUSY;
   1472 			}
   1473 		}
   1474 		/*
   1475 		 * if a complete block is being split, account for it
   1476 		 */
   1477 		fragno = ffs_fragstoblks(fs, bbase);
   1478 		if (ffs_isblock(fs, blksfree, fragno)) {
   1479 			ufs_add32(cgp->cg_cs.cs_nffree, fs->fs_frag, needswap);
   1480 			fs->fs_cstotal.cs_nffree += fs->fs_frag;
   1481 			fs->fs_cs(fs, cg).cs_nffree += fs->fs_frag;
   1482 			ffs_clusteracct(fs, cgp, fragno, -1);
   1483 			ufs_add32(cgp->cg_cs.cs_nbfree, -1, needswap);
   1484 			fs->fs_cstotal.cs_nbfree--;
   1485 			fs->fs_cs(fs, cg).cs_nbfree--;
   1486 		}
   1487 		/*
   1488 		 * decrement the counts associated with the old frags
   1489 		 */
   1490 		blk = blkmap(fs, blksfree, bbase);
   1491 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
   1492 		/*
   1493 		 * allocate the fragment
   1494 		 */
   1495 		for (i = 0; i < frags; i++) {
   1496 			clrbit(blksfree, cgbno + i);
   1497 		}
   1498 		ufs_add32(cgp->cg_cs.cs_nffree, -i, needswap);
   1499 		fs->fs_cstotal.cs_nffree -= i;
   1500 		fs->fs_cs(fs, cg).cs_nffree -= i;
   1501 		/*
   1502 		 * add back in counts associated with the new frags
   1503 		 */
   1504 		blk = blkmap(fs, blksfree, bbase);
   1505 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
   1506 	}
   1507 	fs->fs_fmod = 1;
   1508 	ACTIVECG_CLR(fs, cg);
   1509 	mutex_exit(&ump->um_lock);
   1510 	bdwrite(bp);
   1511 	return 0;
   1512 }
   1513 
   1514 /*
   1515  * Free a block or fragment.
   1516  *
   1517  * The specified block or fragment is placed back in the
   1518  * free map. If a fragment is deallocated, a possible
   1519  * block reassembly is checked.
   1520  *
   1521  * => um_lock not held on entry or exit
   1522  */
   1523 static void
   1524 ffs_blkfree_cg(struct fs *fs, struct vnode *devvp, daddr_t bno, long size)
   1525 {
   1526 	struct cg *cgp;
   1527 	struct buf *bp;
   1528 	struct ufsmount *ump;
   1529 	daddr_t cgblkno;
   1530 	int error, cg;
   1531 	dev_t dev;
   1532 	const bool devvp_is_snapshot = (devvp->v_type != VBLK);
   1533 	const int needswap = UFS_FSNEEDSWAP(fs);
   1534 
   1535 	KASSERT(!devvp_is_snapshot);
   1536 
   1537 	cg = dtog(fs, bno);
   1538 	dev = devvp->v_rdev;
   1539 	ump = VFSTOUFS(spec_node_getmountedfs(devvp));
   1540 	KASSERT(fs == ump->um_fs);
   1541 	cgblkno = FFS_FSBTODB(fs, cgtod(fs, cg));
   1542 
   1543 	error = bread(devvp, cgblkno, (int)fs->fs_cgsize,
   1544 	    B_MODIFY, &bp);
   1545 	if (error) {
   1546 		return;
   1547 	}
   1548 	cgp = (struct cg *)bp->b_data;
   1549 	if (!cg_chkmagic(cgp, needswap)) {
   1550 		brelse(bp, 0);
   1551 		return;
   1552 	}
   1553 
   1554 	ffs_blkfree_common(ump, fs, dev, bp, bno, size, devvp_is_snapshot);
   1555 
   1556 	bdwrite(bp);
   1557 }
   1558 
   1559 struct discardopdata {
   1560 	struct work wk; /* must be first */
   1561 	struct vnode *devvp;
   1562 	daddr_t bno;
   1563 	long size;
   1564 };
   1565 
   1566 struct discarddata {
   1567 	struct fs *fs;
   1568 	struct discardopdata *entry;
   1569 	long maxsize;
   1570 	kmutex_t entrylk;
   1571 	struct workqueue *wq;
   1572 	int wqcnt, wqdraining;
   1573 	kmutex_t wqlk;
   1574 	kcondvar_t wqcv;
   1575 	/* timer for flush? */
   1576 };
   1577 
   1578 static void
   1579 ffs_blkfree_td(struct fs *fs, struct discardopdata *td)
   1580 {
   1581 	struct mount *mp = spec_node_getmountedfs(td->devvp);
   1582 	long todo;
   1583 	int error;
   1584 
   1585 	while (td->size) {
   1586 		todo = min(td->size,
   1587 		  ffs_lfragtosize(fs, (fs->fs_frag - ffs_fragnum(fs, td->bno))));
   1588 		error = UFS_WAPBL_BEGIN(mp);
   1589 		if (error) {
   1590 			printf("ffs: failed to begin wapbl transaction"
   1591 			    " for discard: %d\n", error);
   1592 			break;
   1593 		}
   1594 		ffs_blkfree_cg(fs, td->devvp, td->bno, todo);
   1595 		UFS_WAPBL_END(mp);
   1596 		td->bno += ffs_numfrags(fs, todo);
   1597 		td->size -= todo;
   1598 	}
   1599 }
   1600 
   1601 static void
   1602 ffs_discardcb(struct work *wk, void *arg)
   1603 {
   1604 	struct discardopdata *td = (void *)wk;
   1605 	struct discarddata *ts = arg;
   1606 	struct fs *fs = ts->fs;
   1607 	off_t start, len;
   1608 #ifdef TRIMDEBUG
   1609 	int error;
   1610 #endif
   1611 
   1612 /* like FSBTODB but emits bytes; XXX move to fs.h */
   1613 #ifndef FFS_FSBTOBYTES
   1614 #define FFS_FSBTOBYTES(fs, b) ((b) << (fs)->fs_fshift)
   1615 #endif
   1616 
   1617 	start = FFS_FSBTOBYTES(fs, td->bno);
   1618 	len = td->size;
   1619 #ifdef TRIMDEBUG
   1620 	error =
   1621 #endif
   1622 		VOP_FDISCARD(td->devvp, start, len);
   1623 #ifdef TRIMDEBUG
   1624 	printf("trim(%" PRId64 ",%ld):%d\n", td->bno, td->size, error);
   1625 #endif
   1626 
   1627 	ffs_blkfree_td(fs, td);
   1628 	kmem_free(td, sizeof(*td));
   1629 	mutex_enter(&ts->wqlk);
   1630 	ts->wqcnt--;
   1631 	if (ts->wqdraining && !ts->wqcnt)
   1632 		cv_signal(&ts->wqcv);
   1633 	mutex_exit(&ts->wqlk);
   1634 }
   1635 
   1636 void *
   1637 ffs_discard_init(struct vnode *devvp, struct fs *fs)
   1638 {
   1639 	struct discarddata *ts;
   1640 	int error;
   1641 
   1642 	ts = kmem_zalloc(sizeof (*ts), KM_SLEEP);
   1643 	error = workqueue_create(&ts->wq, "trimwq", ffs_discardcb, ts,
   1644 				 0, 0, 0);
   1645 	if (error) {
   1646 		kmem_free(ts, sizeof (*ts));
   1647 		return NULL;
   1648 	}
   1649 	mutex_init(&ts->entrylk, MUTEX_DEFAULT, IPL_NONE);
   1650 	mutex_init(&ts->wqlk, MUTEX_DEFAULT, IPL_NONE);
   1651 	cv_init(&ts->wqcv, "trimwqcv");
   1652 	ts->maxsize = 100*1024; /* XXX */
   1653 	ts->fs = fs;
   1654 	return ts;
   1655 }
   1656 
   1657 void
   1658 ffs_discard_finish(void *vts, int flags)
   1659 {
   1660 	struct discarddata *ts = vts;
   1661 	struct discardopdata *td = NULL;
   1662 	int res = 0;
   1663 
   1664 	/* wait for workqueue to drain */
   1665 	mutex_enter(&ts->wqlk);
   1666 	if (ts->wqcnt) {
   1667 		ts->wqdraining = 1;
   1668 		res = cv_timedwait(&ts->wqcv, &ts->wqlk, mstohz(5000));
   1669 	}
   1670 	mutex_exit(&ts->wqlk);
   1671 	if (res)
   1672 		printf("ffs_discarddata drain timeout\n");
   1673 
   1674 	mutex_enter(&ts->entrylk);
   1675 	if (ts->entry) {
   1676 		td = ts->entry;
   1677 		ts->entry = NULL;
   1678 	}
   1679 	mutex_exit(&ts->entrylk);
   1680 	if (td) {
   1681 		/* XXX don't tell disk, its optional */
   1682 		ffs_blkfree_td(ts->fs, td);
   1683 #ifdef TRIMDEBUG
   1684 		printf("finish(%" PRId64 ",%ld)\n", td->bno, td->size);
   1685 #endif
   1686 		kmem_free(td, sizeof(*td));
   1687 	}
   1688 
   1689 	cv_destroy(&ts->wqcv);
   1690 	mutex_destroy(&ts->entrylk);
   1691 	mutex_destroy(&ts->wqlk);
   1692 	workqueue_destroy(ts->wq);
   1693 	kmem_free(ts, sizeof(*ts));
   1694 }
   1695 
   1696 void
   1697 ffs_blkfree(struct fs *fs, struct vnode *devvp, daddr_t bno, long size,
   1698     ino_t inum)
   1699 {
   1700 	struct ufsmount *ump;
   1701 	int error;
   1702 	dev_t dev;
   1703 	struct discarddata *ts;
   1704 	struct discardopdata *td;
   1705 
   1706 	dev = devvp->v_rdev;
   1707 	ump = VFSTOUFS(spec_node_getmountedfs(devvp));
   1708 	if (ffs_snapblkfree(fs, devvp, bno, size, inum))
   1709 		return;
   1710 
   1711 	error = ffs_check_bad_allocation(__func__, fs, bno, size, dev, inum);
   1712 	if (error)
   1713 		return;
   1714 
   1715 	if (!ump->um_discarddata) {
   1716 		ffs_blkfree_cg(fs, devvp, bno, size);
   1717 		return;
   1718 	}
   1719 
   1720 #ifdef TRIMDEBUG
   1721 	printf("blkfree(%" PRId64 ",%ld)\n", bno, size);
   1722 #endif
   1723 	ts = ump->um_discarddata;
   1724 	td = NULL;
   1725 
   1726 	mutex_enter(&ts->entrylk);
   1727 	if (ts->entry) {
   1728 		td = ts->entry;
   1729 		/* ffs deallocs backwards, check for prepend only */
   1730 		if (td->bno == bno + ffs_numfrags(fs, size)
   1731 		    && td->size + size <= ts->maxsize) {
   1732 			td->bno = bno;
   1733 			td->size += size;
   1734 			if (td->size < ts->maxsize) {
   1735 #ifdef TRIMDEBUG
   1736 				printf("defer(%" PRId64 ",%ld)\n", td->bno, td->size);
   1737 #endif
   1738 				mutex_exit(&ts->entrylk);
   1739 				return;
   1740 			}
   1741 			size = 0; /* mark done */
   1742 		}
   1743 		ts->entry = NULL;
   1744 	}
   1745 	mutex_exit(&ts->entrylk);
   1746 
   1747 	if (td) {
   1748 #ifdef TRIMDEBUG
   1749 		printf("enq old(%" PRId64 ",%ld)\n", td->bno, td->size);
   1750 #endif
   1751 		mutex_enter(&ts->wqlk);
   1752 		ts->wqcnt++;
   1753 		mutex_exit(&ts->wqlk);
   1754 		workqueue_enqueue(ts->wq, &td->wk, NULL);
   1755 	}
   1756 	if (!size)
   1757 		return;
   1758 
   1759 	td = kmem_alloc(sizeof(*td), KM_SLEEP);
   1760 	td->devvp = devvp;
   1761 	td->bno = bno;
   1762 	td->size = size;
   1763 
   1764 	if (td->size < ts->maxsize) { /* XXX always the case */
   1765 		mutex_enter(&ts->entrylk);
   1766 		if (!ts->entry) { /* possible race? */
   1767 #ifdef TRIMDEBUG
   1768 			printf("defer(%" PRId64 ",%ld)\n", td->bno, td->size);
   1769 #endif
   1770 			ts->entry = td;
   1771 			td = NULL;
   1772 		}
   1773 		mutex_exit(&ts->entrylk);
   1774 	}
   1775 	if (td) {
   1776 #ifdef TRIMDEBUG
   1777 		printf("enq new(%" PRId64 ",%ld)\n", td->bno, td->size);
   1778 #endif
   1779 		mutex_enter(&ts->wqlk);
   1780 		ts->wqcnt++;
   1781 		mutex_exit(&ts->wqlk);
   1782 		workqueue_enqueue(ts->wq, &td->wk, NULL);
   1783 	}
   1784 }
   1785 
   1786 /*
   1787  * Free a block or fragment from a snapshot cg copy.
   1788  *
   1789  * The specified block or fragment is placed back in the
   1790  * free map. If a fragment is deallocated, a possible
   1791  * block reassembly is checked.
   1792  *
   1793  * => um_lock not held on entry or exit
   1794  */
   1795 void
   1796 ffs_blkfree_snap(struct fs *fs, struct vnode *devvp, daddr_t bno, long size,
   1797     ino_t inum)
   1798 {
   1799 	struct cg *cgp;
   1800 	struct buf *bp;
   1801 	struct ufsmount *ump;
   1802 	daddr_t cgblkno;
   1803 	int error, cg;
   1804 	dev_t dev;
   1805 	const bool devvp_is_snapshot = (devvp->v_type != VBLK);
   1806 	const int needswap = UFS_FSNEEDSWAP(fs);
   1807 
   1808 	KASSERT(devvp_is_snapshot);
   1809 
   1810 	cg = dtog(fs, bno);
   1811 	dev = VTOI(devvp)->i_devvp->v_rdev;
   1812 	ump = VFSTOUFS(devvp->v_mount);
   1813 	cgblkno = ffs_fragstoblks(fs, cgtod(fs, cg));
   1814 
   1815 	error = ffs_check_bad_allocation(__func__, fs, bno, size, dev, inum);
   1816 	if (error)
   1817 		return;
   1818 
   1819 	error = bread(devvp, cgblkno, (int)fs->fs_cgsize,
   1820 	    B_MODIFY, &bp);
   1821 	if (error) {
   1822 		return;
   1823 	}
   1824 	cgp = (struct cg *)bp->b_data;
   1825 	if (!cg_chkmagic(cgp, needswap)) {
   1826 		brelse(bp, 0);
   1827 		return;
   1828 	}
   1829 
   1830 	ffs_blkfree_common(ump, fs, dev, bp, bno, size, devvp_is_snapshot);
   1831 
   1832 	bdwrite(bp);
   1833 }
   1834 
   1835 static void
   1836 ffs_blkfree_common(struct ufsmount *ump, struct fs *fs, dev_t dev,
   1837     struct buf *bp, daddr_t bno, long size, bool devvp_is_snapshot)
   1838 {
   1839 	struct cg *cgp;
   1840 	int32_t fragno, cgbno;
   1841 	int i, cg, blk, frags, bbase;
   1842 	u_int8_t *blksfree;
   1843 	const int needswap = UFS_FSNEEDSWAP(fs);
   1844 
   1845 	cg = dtog(fs, bno);
   1846 	cgp = (struct cg *)bp->b_data;
   1847 	cgp->cg_old_time = ufs_rw32(time_second, needswap);
   1848 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
   1849 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
   1850 		cgp->cg_time = ufs_rw64(time_second, needswap);
   1851 	cgbno = dtogd(fs, bno);
   1852 	blksfree = cg_blksfree(cgp, needswap);
   1853 	mutex_enter(&ump->um_lock);
   1854 	if (size == fs->fs_bsize) {
   1855 		fragno = ffs_fragstoblks(fs, cgbno);
   1856 		if (!ffs_isfreeblock(fs, blksfree, fragno)) {
   1857 			if (devvp_is_snapshot) {
   1858 				mutex_exit(&ump->um_lock);
   1859 				return;
   1860 			}
   1861 			panic("%s: freeing free block: dev = 0x%llx, block = %"
   1862 			    PRId64 ", fs = %s", __func__,
   1863 			    (unsigned long long)dev, bno, fs->fs_fsmnt);
   1864 		}
   1865 		ffs_setblock(fs, blksfree, fragno);
   1866 		ffs_clusteracct(fs, cgp, fragno, 1);
   1867 		ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
   1868 		fs->fs_cstotal.cs_nbfree++;
   1869 		fs->fs_cs(fs, cg).cs_nbfree++;
   1870 		if ((fs->fs_magic == FS_UFS1_MAGIC) &&
   1871 		    ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) {
   1872 			i = old_cbtocylno(fs, cgbno);
   1873 			KASSERT(i >= 0);
   1874 			KASSERT(i < fs->fs_old_ncyl);
   1875 			KASSERT(old_cbtorpos(fs, cgbno) >= 0);
   1876 			KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, cgbno) < fs->fs_old_nrpos);
   1877 			ufs_add16(old_cg_blks(fs, cgp, i, needswap)[old_cbtorpos(fs, cgbno)], 1,
   1878 			    needswap);
   1879 			ufs_add32(old_cg_blktot(cgp, needswap)[i], 1, needswap);
   1880 		}
   1881 	} else {
   1882 		bbase = cgbno - ffs_fragnum(fs, cgbno);
   1883 		/*
   1884 		 * decrement the counts associated with the old frags
   1885 		 */
   1886 		blk = blkmap(fs, blksfree, bbase);
   1887 		ffs_fragacct(fs, blk, cgp->cg_frsum, -1, needswap);
   1888 		/*
   1889 		 * deallocate the fragment
   1890 		 */
   1891 		frags = ffs_numfrags(fs, size);
   1892 		for (i = 0; i < frags; i++) {
   1893 			if (isset(blksfree, cgbno + i)) {
   1894 				panic("%s: freeing free frag: "
   1895 				    "dev = 0x%llx, block = %" PRId64
   1896 				    ", fs = %s", __func__,
   1897 				    (unsigned long long)dev, bno + i,
   1898 				    fs->fs_fsmnt);
   1899 			}
   1900 			setbit(blksfree, cgbno + i);
   1901 		}
   1902 		ufs_add32(cgp->cg_cs.cs_nffree, i, needswap);
   1903 		fs->fs_cstotal.cs_nffree += i;
   1904 		fs->fs_cs(fs, cg).cs_nffree += i;
   1905 		/*
   1906 		 * add back in counts associated with the new frags
   1907 		 */
   1908 		blk = blkmap(fs, blksfree, bbase);
   1909 		ffs_fragacct(fs, blk, cgp->cg_frsum, 1, needswap);
   1910 		/*
   1911 		 * if a complete block has been reassembled, account for it
   1912 		 */
   1913 		fragno = ffs_fragstoblks(fs, bbase);
   1914 		if (ffs_isblock(fs, blksfree, fragno)) {
   1915 			ufs_add32(cgp->cg_cs.cs_nffree, -fs->fs_frag, needswap);
   1916 			fs->fs_cstotal.cs_nffree -= fs->fs_frag;
   1917 			fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag;
   1918 			ffs_clusteracct(fs, cgp, fragno, 1);
   1919 			ufs_add32(cgp->cg_cs.cs_nbfree, 1, needswap);
   1920 			fs->fs_cstotal.cs_nbfree++;
   1921 			fs->fs_cs(fs, cg).cs_nbfree++;
   1922 			if ((fs->fs_magic == FS_UFS1_MAGIC) &&
   1923 			    ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0)) {
   1924 				i = old_cbtocylno(fs, bbase);
   1925 				KASSERT(i >= 0);
   1926 				KASSERT(i < fs->fs_old_ncyl);
   1927 				KASSERT(old_cbtorpos(fs, bbase) >= 0);
   1928 				KASSERT(fs->fs_old_nrpos == 0 || old_cbtorpos(fs, bbase) < fs->fs_old_nrpos);
   1929 				ufs_add16(old_cg_blks(fs, cgp, i, needswap)[old_cbtorpos(fs,
   1930 				    bbase)], 1, needswap);
   1931 				ufs_add32(old_cg_blktot(cgp, needswap)[i], 1, needswap);
   1932 			}
   1933 		}
   1934 	}
   1935 	fs->fs_fmod = 1;
   1936 	ACTIVECG_CLR(fs, cg);
   1937 	mutex_exit(&ump->um_lock);
   1938 }
   1939 
   1940 /*
   1941  * Free an inode.
   1942  */
   1943 int
   1944 ffs_vfree(struct vnode *vp, ino_t ino, int mode)
   1945 {
   1946 
   1947 	return ffs_freefile(vp->v_mount, ino, mode);
   1948 }
   1949 
   1950 /*
   1951  * Do the actual free operation.
   1952  * The specified inode is placed back in the free map.
   1953  *
   1954  * => um_lock not held on entry or exit
   1955  */
   1956 int
   1957 ffs_freefile(struct mount *mp, ino_t ino, int mode)
   1958 {
   1959 	struct ufsmount *ump = VFSTOUFS(mp);
   1960 	struct fs *fs = ump->um_fs;
   1961 	struct vnode *devvp;
   1962 	struct cg *cgp;
   1963 	struct buf *bp;
   1964 	int error, cg;
   1965 	daddr_t cgbno;
   1966 	dev_t dev;
   1967 	const int needswap = UFS_FSNEEDSWAP(fs);
   1968 
   1969 	cg = ino_to_cg(fs, ino);
   1970 	devvp = ump->um_devvp;
   1971 	dev = devvp->v_rdev;
   1972 	cgbno = FFS_FSBTODB(fs, cgtod(fs, cg));
   1973 
   1974 	if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
   1975 		panic("%s: range: dev = 0x%llx, ino = %llu, fs = %s", __func__,
   1976 		    (long long)dev, (unsigned long long)ino, fs->fs_fsmnt);
   1977 	error = bread(devvp, cgbno, (int)fs->fs_cgsize,
   1978 	    B_MODIFY, &bp);
   1979 	if (error) {
   1980 		return (error);
   1981 	}
   1982 	cgp = (struct cg *)bp->b_data;
   1983 	if (!cg_chkmagic(cgp, needswap)) {
   1984 		brelse(bp, 0);
   1985 		return (0);
   1986 	}
   1987 
   1988 	ffs_freefile_common(ump, fs, dev, bp, ino, mode, false);
   1989 
   1990 	bdwrite(bp);
   1991 
   1992 	return 0;
   1993 }
   1994 
   1995 int
   1996 ffs_freefile_snap(struct fs *fs, struct vnode *devvp, ino_t ino, int mode)
   1997 {
   1998 	struct ufsmount *ump;
   1999 	struct cg *cgp;
   2000 	struct buf *bp;
   2001 	int error, cg;
   2002 	daddr_t cgbno;
   2003 	dev_t dev;
   2004 	const int needswap = UFS_FSNEEDSWAP(fs);
   2005 
   2006 	KASSERT(devvp->v_type != VBLK);
   2007 
   2008 	cg = ino_to_cg(fs, ino);
   2009 	dev = VTOI(devvp)->i_devvp->v_rdev;
   2010 	ump = VFSTOUFS(devvp->v_mount);
   2011 	cgbno = ffs_fragstoblks(fs, cgtod(fs, cg));
   2012 	if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
   2013 		panic("%s: range: dev = 0x%llx, ino = %llu, fs = %s", __func__,
   2014 		    (unsigned long long)dev, (unsigned long long)ino,
   2015 		    fs->fs_fsmnt);
   2016 	error = bread(devvp, cgbno, (int)fs->fs_cgsize,
   2017 	    B_MODIFY, &bp);
   2018 	if (error) {
   2019 		return (error);
   2020 	}
   2021 	cgp = (struct cg *)bp->b_data;
   2022 	if (!cg_chkmagic(cgp, needswap)) {
   2023 		brelse(bp, 0);
   2024 		return (0);
   2025 	}
   2026 	ffs_freefile_common(ump, fs, dev, bp, ino, mode, true);
   2027 
   2028 	bdwrite(bp);
   2029 
   2030 	return 0;
   2031 }
   2032 
   2033 static void
   2034 ffs_freefile_common(struct ufsmount *ump, struct fs *fs, dev_t dev,
   2035     struct buf *bp, ino_t ino, int mode, bool devvp_is_snapshot)
   2036 {
   2037 	int cg;
   2038 	struct cg *cgp;
   2039 	u_int8_t *inosused;
   2040 	const int needswap = UFS_FSNEEDSWAP(fs);
   2041 
   2042 	cg = ino_to_cg(fs, ino);
   2043 	cgp = (struct cg *)bp->b_data;
   2044 	cgp->cg_old_time = ufs_rw32(time_second, needswap);
   2045 	if ((fs->fs_magic != FS_UFS1_MAGIC) ||
   2046 	    (fs->fs_old_flags & FS_FLAGS_UPDATED))
   2047 		cgp->cg_time = ufs_rw64(time_second, needswap);
   2048 	inosused = cg_inosused(cgp, needswap);
   2049 	ino %= fs->fs_ipg;
   2050 	if (isclr(inosused, ino)) {
   2051 		printf("ifree: dev = 0x%llx, ino = %llu, fs = %s\n",
   2052 		    (unsigned long long)dev, (unsigned long long)ino +
   2053 		    cg * fs->fs_ipg, fs->fs_fsmnt);
   2054 		if (fs->fs_ronly == 0)
   2055 			panic("%s: freeing free inode", __func__);
   2056 	}
   2057 	clrbit(inosused, ino);
   2058 	if (!devvp_is_snapshot)
   2059 		UFS_WAPBL_UNREGISTER_INODE(ump->um_mountp,
   2060 		    ino + cg * fs->fs_ipg, mode);
   2061 	if (ino < ufs_rw32(cgp->cg_irotor, needswap))
   2062 		cgp->cg_irotor = ufs_rw32(ino, needswap);
   2063 	ufs_add32(cgp->cg_cs.cs_nifree, 1, needswap);
   2064 	mutex_enter(&ump->um_lock);
   2065 	fs->fs_cstotal.cs_nifree++;
   2066 	fs->fs_cs(fs, cg).cs_nifree++;
   2067 	if ((mode & IFMT) == IFDIR) {
   2068 		ufs_add32(cgp->cg_cs.cs_ndir, -1, needswap);
   2069 		fs->fs_cstotal.cs_ndir--;
   2070 		fs->fs_cs(fs, cg).cs_ndir--;
   2071 	}
   2072 	fs->fs_fmod = 1;
   2073 	ACTIVECG_CLR(fs, cg);
   2074 	mutex_exit(&ump->um_lock);
   2075 }
   2076 
   2077 /*
   2078  * Check to see if a file is free.
   2079  */
   2080 int
   2081 ffs_checkfreefile(struct fs *fs, struct vnode *devvp, ino_t ino)
   2082 {
   2083 	struct cg *cgp;
   2084 	struct buf *bp;
   2085 	daddr_t cgbno;
   2086 	int ret, cg;
   2087 	u_int8_t *inosused;
   2088 	const bool devvp_is_snapshot = (devvp->v_type != VBLK);
   2089 
   2090 	KASSERT(devvp_is_snapshot);
   2091 
   2092 	cg = ino_to_cg(fs, ino);
   2093 	if (devvp_is_snapshot)
   2094 		cgbno = ffs_fragstoblks(fs, cgtod(fs, cg));
   2095 	else
   2096 		cgbno = FFS_FSBTODB(fs, cgtod(fs, cg));
   2097 	if ((u_int)ino >= fs->fs_ipg * fs->fs_ncg)
   2098 		return 1;
   2099 	if (bread(devvp, cgbno, (int)fs->fs_cgsize, 0, &bp)) {
   2100 		return 1;
   2101 	}
   2102 	cgp = (struct cg *)bp->b_data;
   2103 	if (!cg_chkmagic(cgp, UFS_FSNEEDSWAP(fs))) {
   2104 		brelse(bp, 0);
   2105 		return 1;
   2106 	}
   2107 	inosused = cg_inosused(cgp, UFS_FSNEEDSWAP(fs));
   2108 	ino %= fs->fs_ipg;
   2109 	ret = isclr(inosused, ino);
   2110 	brelse(bp, 0);
   2111 	return ret;
   2112 }
   2113 
   2114 /*
   2115  * Find a block of the specified size in the specified cylinder group.
   2116  *
   2117  * It is a panic if a request is made to find a block if none are
   2118  * available.
   2119  */
   2120 static int32_t
   2121 ffs_mapsearch(struct fs *fs, struct cg *cgp, daddr_t bpref, int allocsiz)
   2122 {
   2123 	int32_t bno;
   2124 	int start, len, loc, i;
   2125 	int blk, field, subfield, pos;
   2126 	int ostart, olen;
   2127 	u_int8_t *blksfree;
   2128 	const int needswap = UFS_FSNEEDSWAP(fs);
   2129 
   2130 	/* KASSERT(mutex_owned(&ump->um_lock)); */
   2131 
   2132 	/*
   2133 	 * find the fragment by searching through the free block
   2134 	 * map for an appropriate bit pattern
   2135 	 */
   2136 	if (bpref)
   2137 		start = dtogd(fs, bpref) / NBBY;
   2138 	else
   2139 		start = ufs_rw32(cgp->cg_frotor, needswap) / NBBY;
   2140 	blksfree = cg_blksfree(cgp, needswap);
   2141 	len = howmany(fs->fs_fpg, NBBY) - start;
   2142 	ostart = start;
   2143 	olen = len;
   2144 	loc = scanc((u_int)len,
   2145 		(const u_char *)&blksfree[start],
   2146 		(const u_char *)fragtbl[fs->fs_frag],
   2147 		(1 << (allocsiz - 1 + (fs->fs_frag & (NBBY - 1)))));
   2148 	if (loc == 0) {
   2149 		len = start + 1;
   2150 		start = 0;
   2151 		loc = scanc((u_int)len,
   2152 			(const u_char *)&blksfree[0],
   2153 			(const u_char *)fragtbl[fs->fs_frag],
   2154 			(1 << (allocsiz - 1 + (fs->fs_frag & (NBBY - 1)))));
   2155 		if (loc == 0) {
   2156 			panic("%s: map corrupted: start=%d, len=%d, "
   2157 			    "fs = %s, offset=%d/%ld, cg %d", __func__,
   2158 			    ostart, olen, fs->fs_fsmnt,
   2159 			    ufs_rw32(cgp->cg_freeoff, needswap),
   2160 			    (long)blksfree - (long)cgp, cgp->cg_cgx);
   2161 			/* NOTREACHED */
   2162 		}
   2163 	}
   2164 	bno = (start + len - loc) * NBBY;
   2165 	cgp->cg_frotor = ufs_rw32(bno, needswap);
   2166 	/*
   2167 	 * found the byte in the map
   2168 	 * sift through the bits to find the selected frag
   2169 	 */
   2170 	for (i = bno + NBBY; bno < i; bno += fs->fs_frag) {
   2171 		blk = blkmap(fs, blksfree, bno);
   2172 		blk <<= 1;
   2173 		field = around[allocsiz];
   2174 		subfield = inside[allocsiz];
   2175 		for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) {
   2176 			if ((blk & field) == subfield)
   2177 				return (bno + pos);
   2178 			field <<= 1;
   2179 			subfield <<= 1;
   2180 		}
   2181 	}
   2182 	panic("%s: block not in map: bno=%d, fs=%s", __func__,
   2183 	    bno, fs->fs_fsmnt);
   2184 	/* return (-1); */
   2185 }
   2186 
   2187 /*
   2188  * Fserr prints the name of a file system with an error diagnostic.
   2189  *
   2190  * The form of the error message is:
   2191  *	fs: error message
   2192  */
   2193 static void
   2194 ffs_fserr(struct fs *fs, kauth_cred_t cred, const char *cp)
   2195 {
   2196 	KASSERT(cred != NULL);
   2197 
   2198 	if (cred == NOCRED || cred == FSCRED) {
   2199 		log(LOG_ERR, "pid %d, command %s, on %s: %s\n",
   2200 		    curproc->p_pid, curproc->p_comm,
   2201 		    fs->fs_fsmnt, cp);
   2202 	} else {
   2203 		log(LOG_ERR, "uid %d, pid %d, command %s, on %s: %s\n",
   2204 		    kauth_cred_getuid(cred), curproc->p_pid, curproc->p_comm,
   2205 		    fs->fs_fsmnt, cp);
   2206 	}
   2207 }
   2208