Home | History | Annotate | Line # | Download | only in lfs
      1  1.96  riastrad /*	$NetBSD: lfs_balloc.c,v 1.96 2020/09/05 16:30:13 riastradh Exp $	*/
      2   1.2       cgd 
      3  1.11  perseant /*-
      4  1.36  perseant  * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc.
      5  1.11  perseant  * All rights reserved.
      6  1.11  perseant  *
      7  1.11  perseant  * This code is derived from software contributed to The NetBSD Foundation
      8  1.11  perseant  * by Konrad E. Schroder <perseant (at) hhhh.org>.
      9  1.11  perseant  *
     10  1.11  perseant  * Redistribution and use in source and binary forms, with or without
     11  1.11  perseant  * modification, are permitted provided that the following conditions
     12  1.11  perseant  * are met:
     13  1.11  perseant  * 1. Redistributions of source code must retain the above copyright
     14  1.11  perseant  *    notice, this list of conditions and the following disclaimer.
     15  1.11  perseant  * 2. Redistributions in binary form must reproduce the above copyright
     16  1.11  perseant  *    notice, this list of conditions and the following disclaimer in the
     17  1.11  perseant  *    documentation and/or other materials provided with the distribution.
     18  1.11  perseant  *
     19  1.11  perseant  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
     20  1.11  perseant  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
     21  1.11  perseant  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
     22  1.11  perseant  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
     23  1.11  perseant  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     24  1.11  perseant  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     25  1.11  perseant  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     26  1.11  perseant  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     27  1.11  perseant  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     28  1.11  perseant  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
     29  1.11  perseant  * POSSIBILITY OF SUCH DAMAGE.
     30  1.11  perseant  */
     31   1.1   mycroft /*
     32   1.1   mycroft  * Copyright (c) 1989, 1991, 1993
     33   1.1   mycroft  *	The Regents of the University of California.  All rights reserved.
     34   1.1   mycroft  *
     35   1.1   mycroft  * Redistribution and use in source and binary forms, with or without
     36   1.1   mycroft  * modification, are permitted provided that the following conditions
     37   1.1   mycroft  * are met:
     38   1.1   mycroft  * 1. Redistributions of source code must retain the above copyright
     39   1.1   mycroft  *    notice, this list of conditions and the following disclaimer.
     40   1.1   mycroft  * 2. Redistributions in binary form must reproduce the above copyright
     41   1.1   mycroft  *    notice, this list of conditions and the following disclaimer in the
     42   1.1   mycroft  *    documentation and/or other materials provided with the distribution.
     43  1.43       agc  * 3. Neither the name of the University nor the names of its contributors
     44   1.1   mycroft  *    may be used to endorse or promote products derived from this software
     45   1.1   mycroft  *    without specific prior written permission.
     46   1.1   mycroft  *
     47   1.1   mycroft  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     48   1.1   mycroft  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     49   1.1   mycroft  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     50   1.1   mycroft  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     51   1.1   mycroft  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     52   1.1   mycroft  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     53   1.1   mycroft  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     54   1.1   mycroft  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     55   1.1   mycroft  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     56   1.1   mycroft  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     57   1.1   mycroft  * SUCH DAMAGE.
     58   1.1   mycroft  *
     59   1.5      fvdl  *	@(#)lfs_balloc.c	8.4 (Berkeley) 5/8/95
     60   1.1   mycroft  */
     61  1.30     lukem 
     62  1.30     lukem #include <sys/cdefs.h>
     63  1.96  riastrad __KERNEL_RCSID(0, "$NetBSD: lfs_balloc.c,v 1.96 2020/09/05 16:30:13 riastradh Exp $");
     64   1.8    scottr 
     65  1.28       mrg #if defined(_KERNEL_OPT)
     66   1.8    scottr #include "opt_quota.h"
     67   1.9    scottr #endif
     68   1.8    scottr 
     69   1.1   mycroft #include <sys/param.h>
     70   1.3  christos #include <sys/systm.h>
     71   1.1   mycroft #include <sys/buf.h>
     72   1.1   mycroft #include <sys/proc.h>
     73   1.1   mycroft #include <sys/vnode.h>
     74   1.1   mycroft #include <sys/mount.h>
     75   1.1   mycroft #include <sys/resourcevar.h>
     76  1.55  perseant #include <sys/tree.h>
     77   1.1   mycroft #include <sys/trace.h>
     78  1.61      elad #include <sys/kauth.h>
     79   1.1   mycroft 
     80   1.1   mycroft #include <miscfs/specfs/specdev.h>
     81   1.1   mycroft 
     82  1.73  dholland #include <ufs/lfs/ulfs_quotacommon.h>
     83  1.73  dholland #include <ufs/lfs/ulfs_inode.h>
     84  1.73  dholland #include <ufs/lfs/ulfsmount.h>
     85  1.73  dholland #include <ufs/lfs/ulfs_extern.h>
     86   1.1   mycroft 
     87   1.1   mycroft #include <ufs/lfs/lfs.h>
     88  1.84  dholland #include <ufs/lfs/lfs_accessors.h>
     89   1.1   mycroft #include <ufs/lfs/lfs_extern.h>
     90  1.78  dholland #include <ufs/lfs/lfs_kernel.h>
     91   1.5      fvdl 
     92  1.96  riastrad #include <uvm/uvm_extern.h>
     93  1.36  perseant 
     94  1.89  dholland static int lfs_fragextend(struct vnode *, int, int, daddr_t, struct buf **,
     95  1.89  dholland 			  kauth_cred_t);
     96   1.5      fvdl 
     97  1.49  perseant u_int64_t locked_fakequeue_count;
     98  1.49  perseant 
     99  1.16  perseant /*
    100  1.90  dholland  * Allocate a block, and do inode and filesystem block accounting for
    101  1.90  dholland  * it and for any indirect blocks that may need to be created in order
    102  1.90  dholland  * to handle this block.
    103  1.90  dholland  *
    104  1.90  dholland  * Blocks which have never been accounted for (i.e., which "do not
    105  1.90  dholland  * exist") have disk address 0, which is translated by ulfs_bmap to
    106  1.90  dholland  * the special value UNASSIGNED == -1, as in historical FFS-related
    107  1.90  dholland  * code.
    108  1.90  dholland  *
    109  1.90  dholland  * Blocks which have been accounted for but which have not yet been
    110  1.90  dholland  * written to disk are given the new special disk address UNWRITTEN ==
    111  1.90  dholland  * -2, so that they can be differentiated from completely new blocks.
    112  1.90  dholland  *
    113  1.90  dholland  * Note: it seems that bpp is passed as NULL for blocks that are file
    114  1.90  dholland  * pages that will be handled by UVM and not the buffer cache.
    115  1.90  dholland  *
    116  1.90  dholland  * XXX: locking?
    117  1.16  perseant  */
    118  1.75  dholland /* VOP_BWRITE ULFS_NIADDR+2 times */
    119   1.1   mycroft int
    120  1.61      elad lfs_balloc(struct vnode *vp, off_t startoffset, int iosize, kauth_cred_t cred,
    121  1.57      yamt     int flags, struct buf **bpp)
    122  1.14      fvdl {
    123   1.5      fvdl 	int offset;
    124  1.46   mycroft 	daddr_t daddr, idaddr;
    125  1.57      yamt 	struct buf *ibp, *bp;
    126   1.1   mycroft 	struct inode *ip;
    127   1.1   mycroft 	struct lfs *fs;
    128  1.75  dholland 	struct indir indirs[ULFS_NIADDR+2], *idp;
    129  1.35      fvdl 	daddr_t	lbn, lastblock;
    130  1.69   mlelstv 	int bcount;
    131  1.11  perseant 	int error, frags, i, nsize, osize, num;
    132  1.14      fvdl 
    133   1.1   mycroft 	ip = VTOI(vp);
    134   1.1   mycroft 	fs = ip->i_lfs;
    135  1.90  dholland 
    136  1.90  dholland 	/* Declare to humans that we might have the seglock here */
    137  1.90  dholland 	ASSERT_MAYBE_SEGLOCK(fs);
    138  1.90  dholland 
    139  1.90  dholland 
    140  1.90  dholland 	/* offset within block */
    141  1.77  christos 	offset = lfs_blkoff(fs, startoffset);
    142  1.90  dholland 
    143  1.90  dholland 	/* This is usually but not always exactly the block size */
    144  1.82  dholland 	KASSERT(iosize <= lfs_sb_getbsize(fs));
    145  1.90  dholland 
    146  1.90  dholland 	/* block number (within file) */
    147  1.77  christos 	lbn = lfs_lblkno(fs, startoffset);
    148  1.90  dholland 
    149  1.90  dholland 	/*
    150  1.90  dholland 	 * This checks for whether pending stuff needs to be flushed
    151  1.90  dholland 	 * out and potentially waits. It's been disabled since UBC
    152  1.90  dholland 	 * support was added to LFS in 2003. -- dholland 20160806
    153  1.90  dholland 	 */
    154  1.36  perseant 	/* (void)lfs_check(vp, lbn, 0); */
    155  1.36  perseant 
    156  1.52  perseant 
    157  1.50     perry 	/*
    158   1.1   mycroft 	 * Three cases: it's a block beyond the end of file, it's a block in
    159   1.1   mycroft 	 * the file that may or may not have been assigned a disk address or
    160  1.19  perseant 	 * we're writing an entire block.
    161  1.19  perseant 	 *
    162  1.19  perseant 	 * Note, if the daddr is UNWRITTEN, the block already exists in
    163  1.37  perseant 	 * the cache (it was read or written earlier).	If so, make sure
    164  1.19  perseant 	 * we don't count it as a new block or zero out its contents. If
    165  1.19  perseant 	 * it did not, make sure we allocate any necessary indirect
    166  1.19  perseant 	 * blocks.
    167  1.19  perseant 	 *
    168   1.5      fvdl 	 * If we are writing a block beyond the end of the file, we need to
    169  1.11  perseant 	 * check if the old last block was a fragment.	If it was, we need
    170   1.5      fvdl 	 * to rewrite it.
    171   1.1   mycroft 	 */
    172  1.50     perry 
    173  1.36  perseant 	if (bpp)
    174  1.36  perseant 		*bpp = NULL;
    175  1.50     perry 
    176  1.90  dholland 	/* Last block number in file */
    177  1.77  christos 	lastblock = lfs_lblkno(fs, ip->i_size);
    178  1.90  dholland 
    179  1.75  dholland 	if (lastblock < ULFS_NDADDR && lastblock < lbn) {
    180  1.90  dholland 		/*
    181  1.90  dholland 		 * The file is small enough to have fragments, and we're
    182  1.90  dholland 		 * allocating past EOF.
    183  1.90  dholland 		 *
    184  1.90  dholland 		 * If the last block was a fragment we need to rewrite it
    185  1.90  dholland 		 * as a full block.
    186  1.90  dholland 		 */
    187  1.77  christos 		osize = lfs_blksize(fs, ip, lastblock);
    188  1.82  dholland 		if (osize < lfs_sb_getbsize(fs) && osize > 0) {
    189  1.82  dholland 			if ((error = lfs_fragextend(vp, osize, lfs_sb_getbsize(fs),
    190  1.36  perseant 						    lastblock,
    191  1.57      yamt 						    (bpp ? &bp : NULL), cred)))
    192  1.31       chs 				return (error);
    193  1.90  dholland 			/* Update the file size with what we just did (only) */
    194  1.87  dholland 			ip->i_size = (lastblock + 1) * lfs_sb_getbsize(fs);
    195  1.87  dholland 			lfs_dino_setsize(fs, ip->i_din, ip->i_size);
    196  1.40      fvdl 			uvm_vnp_setsize(vp, ip->i_size);
    197  1.94      maya 			ip->i_state |= IN_CHANGE | IN_UPDATE;
    198  1.90  dholland 			/* if we got a buffer for this, write it out now */
    199  1.36  perseant 			if (bpp)
    200  1.70   hannken 				(void) VOP_BWRITE(bp->b_vp, bp);
    201   1.5      fvdl 		}
    202   1.5      fvdl 	}
    203   1.5      fvdl 
    204   1.5      fvdl 	/*
    205   1.5      fvdl 	 * If the block we are writing is a direct block, it's the last
    206   1.5      fvdl 	 * block in the file, and offset + iosize is less than a full
    207   1.5      fvdl 	 * block, we can write one or more fragments.  There are two cases:
    208   1.5      fvdl 	 * the block is brand new and we should allocate it the correct
    209   1.5      fvdl 	 * size or it already exists and contains some fragments and
    210   1.5      fvdl 	 * may need to extend it.
    211   1.5      fvdl 	 */
    212  1.77  christos 	if (lbn < ULFS_NDADDR && lfs_lblkno(fs, ip->i_size) <= lbn) {
    213  1.77  christos 		osize = lfs_blksize(fs, ip, lbn);
    214  1.77  christos 		nsize = lfs_fragroundup(fs, offset + iosize);
    215  1.77  christos 		if (lfs_lblktosize(fs, lbn) >= ip->i_size) {
    216   1.5      fvdl 			/* Brand new block or fragment */
    217  1.77  christos 			frags = lfs_numfrags(fs, nsize);
    218  1.69   mlelstv 			if (!ISSPACE(fs, frags, cred))
    219  1.51  perseant 				return ENOSPC;
    220  1.36  perseant 			if (bpp) {
    221  1.57      yamt 				*bpp = bp = getblk(vp, lbn, nsize, 0, 0);
    222  1.36  perseant 				bp->b_blkno = UNWRITTEN;
    223  1.57      yamt 				if (flags & B_CLRBUF)
    224  1.38  perseant 					clrbuf(bp);
    225  1.36  perseant 			}
    226  1.90  dholland 
    227  1.90  dholland 			/*
    228  1.90  dholland 			 * Update the effective block count (this count
    229  1.90  dholland 			 * includes blocks that don't have an on-disk
    230  1.90  dholland 			 * presence or location yet)
    231  1.90  dholland 			 */
    232  1.69   mlelstv 			ip->i_lfs_effnblks += frags;
    233  1.90  dholland 
    234  1.90  dholland 			/* account for the space we're taking */
    235  1.64        ad 			mutex_enter(&lfs_lock);
    236  1.85  dholland 			lfs_sb_subbfree(fs, frags);
    237  1.64        ad 			mutex_exit(&lfs_lock);
    238  1.90  dholland 
    239  1.90  dholland 			/* update the inode */
    240  1.87  dholland 			lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN);
    241  1.19  perseant 		} else {
    242  1.90  dholland 			/* extending a block that already has fragments */
    243  1.90  dholland 
    244  1.11  perseant 			if (nsize <= osize) {
    245  1.11  perseant 				/* No need to extend */
    246  1.67   hannken 				if (bpp && (error = bread(vp, lbn, osize,
    247  1.81      maxv 				    0, &bp)))
    248  1.11  perseant 					return error;
    249  1.11  perseant 			} else {
    250  1.11  perseant 				/* Extend existing block */
    251  1.11  perseant 				if ((error =
    252  1.36  perseant 				     lfs_fragextend(vp, osize, nsize, lbn,
    253  1.57      yamt 						    (bpp ? &bp : NULL), cred)))
    254  1.19  perseant 					return error;
    255  1.11  perseant 			}
    256  1.36  perseant 			if (bpp)
    257  1.36  perseant 				*bpp = bp;
    258   1.5      fvdl 		}
    259  1.19  perseant 		return 0;
    260  1.19  perseant 	}
    261  1.19  perseant 
    262  1.90  dholland 	/*
    263  1.90  dholland 	 * Look up what's already here.
    264  1.90  dholland 	 */
    265  1.90  dholland 
    266  1.75  dholland 	error = ulfs_bmaparray(vp, lbn, &daddr, &indirs[0], &num, NULL, NULL);
    267  1.19  perseant 	if (error)
    268  1.19  perseant 		return (error);
    269  1.49  perseant 
    270  1.86  dholland 	KASSERT(daddr <= LFS_MAX_DADDR(fs));
    271  1.49  perseant 
    272  1.19  perseant 	/*
    273  1.19  perseant 	 * Do byte accounting all at once, so we can gracefully fail *before*
    274  1.19  perseant 	 * we start assigning blocks.
    275  1.19  perseant 	 */
    276  1.79  dholland 	frags = fs->um_seqinc;
    277  1.90  dholland 	bcount = 0; /* number of frags we need */
    278  1.19  perseant 	if (daddr == UNASSIGNED) {
    279  1.90  dholland 		/* no block yet, going to need a whole block */
    280  1.69   mlelstv 		bcount = frags;
    281  1.19  perseant 	}
    282  1.19  perseant 	for (i = 1; i < num; ++i) {
    283  1.19  perseant 		if (!indirs[i].in_exists) {
    284  1.90  dholland 			/* need an indirect block at this level */
    285  1.69   mlelstv 			bcount += frags;
    286  1.19  perseant 		}
    287  1.19  perseant 	}
    288  1.57      yamt 	if (ISSPACE(fs, bcount, cred)) {
    289  1.90  dholland 		/* update the superblock's free block count */
    290  1.64        ad 		mutex_enter(&lfs_lock);
    291  1.85  dholland 		lfs_sb_subbfree(fs, bcount);
    292  1.64        ad 		mutex_exit(&lfs_lock);
    293  1.90  dholland 		/* update the file's effective block count */
    294  1.21  perseant 		ip->i_lfs_effnblks += bcount;
    295   1.5      fvdl 	} else {
    296  1.90  dholland 		/* whoops, no can do */
    297  1.19  perseant 		return ENOSPC;
    298  1.19  perseant 	}
    299  1.19  perseant 
    300  1.19  perseant 	if (daddr == UNASSIGNED) {
    301  1.90  dholland 		/*
    302  1.90  dholland 		 * There is nothing here yet.
    303  1.90  dholland 		 */
    304  1.90  dholland 
    305  1.90  dholland 		/*
    306  1.90  dholland 		 * If there's no indirect block in the inode, change it
    307  1.90  dholland 		 * to UNWRITTEN to indicate that it exists but doesn't
    308  1.90  dholland 		 * have an on-disk address yet.
    309  1.90  dholland 		 *
    310  1.90  dholland 		 * (Question: where's the block data initialized?)
    311  1.90  dholland 		 */
    312  1.87  dholland 		if (num > 0 && lfs_dino_getib(fs, ip->i_din, indirs[0].in_off) == 0) {
    313  1.87  dholland 			lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, UNWRITTEN);
    314  1.19  perseant 		}
    315  1.19  perseant 
    316   1.5      fvdl 		/*
    317  1.90  dholland 		 * If we need more layers of indirect blocks, create what
    318  1.90  dholland 		 * we need.
    319   1.5      fvdl 		 */
    320  1.46   mycroft 		if (num > 1) {
    321  1.90  dholland 			/*
    322  1.90  dholland 			 * The outermost indirect block address is the one
    323  1.90  dholland 			 * in the inode, so fetch that.
    324  1.90  dholland 			 */
    325  1.87  dholland 			idaddr = lfs_dino_getib(fs, ip->i_din, indirs[0].in_off);
    326  1.90  dholland 			/*
    327  1.90  dholland 			 * For each layer of indirection...
    328  1.90  dholland 			 */
    329  1.46   mycroft 			for (i = 1; i < num; ++i) {
    330  1.90  dholland 				/*
    331  1.90  dholland 				 * Get a buffer for the indirect block data.
    332  1.90  dholland 				 *
    333  1.90  dholland 				 * (XXX: the logic here seems twisted. What's
    334  1.90  dholland 				 * wrong with testing in_exists first and then
    335  1.90  dholland 				 * doing either bread or getblk to get a
    336  1.90  dholland 				 * buffer?)
    337  1.90  dholland 				 */
    338  1.46   mycroft 				ibp = getblk(vp, indirs[i].in_lbn,
    339  1.82  dholland 				    lfs_sb_getbsize(fs), 0,0);
    340  1.46   mycroft 				if (!indirs[i].in_exists) {
    341  1.90  dholland 					/*
    342  1.90  dholland 					 * There isn't actually a block here,
    343  1.90  dholland 					 * so clear the buffer data and mark
    344  1.90  dholland 					 * the address of the block as
    345  1.90  dholland 					 * UNWRITTEN.
    346  1.90  dholland 					 */
    347  1.46   mycroft 					clrbuf(ibp);
    348  1.46   mycroft 					ibp->b_blkno = UNWRITTEN;
    349  1.64        ad 				} else if (!(ibp->b_oflags & (BO_DELWRI | BO_DONE))) {
    350  1.90  dholland 					/*
    351  1.90  dholland 					 * Otherwise read it in.
    352  1.90  dholland 					 */
    353  1.77  christos 					ibp->b_blkno = LFS_FSBTODB(fs, idaddr);
    354  1.46   mycroft 					ibp->b_flags |= B_READ;
    355  1.48   hannken 					VOP_STRATEGY(vp, ibp);
    356  1.46   mycroft 					biowait(ibp);
    357  1.46   mycroft 				}
    358  1.90  dholland 
    359  1.46   mycroft 				/*
    360  1.90  dholland 				 * Now this indirect block exists, but
    361  1.90  dholland 				 * the next one down may not yet. If
    362  1.90  dholland 				 * so, set it to UNWRITTEN. This keeps
    363  1.46   mycroft 				 * the accounting straight.
    364  1.46   mycroft 				 */
    365  1.88  dholland 				if (lfs_iblock_get(fs, ibp->b_data, indirs[i].in_off) == 0)
    366  1.88  dholland 					lfs_iblock_set(fs, ibp->b_data, indirs[i].in_off,
    367  1.88  dholland 						UNWRITTEN);
    368  1.90  dholland 
    369  1.90  dholland 				/* get the block for the next iteration */
    370  1.88  dholland 				idaddr = lfs_iblock_get(fs, ibp->b_data, indirs[i].in_off);
    371  1.92      maya 
    372  1.52  perseant 				if (vp == fs->lfs_ivnode) {
    373  1.52  perseant 					LFS_ENTER_LOG("balloc", __FILE__,
    374  1.52  perseant 						__LINE__, indirs[i].in_lbn,
    375  1.52  perseant 						ibp->b_flags, curproc->p_pid);
    376  1.52  perseant 				}
    377  1.90  dholland 				/*
    378  1.90  dholland 				 * Write out the updated indirect block. Note
    379  1.90  dholland 				 * that this writes it out even if we didn't
    380  1.90  dholland 				 * modify it - ultimately because the final
    381  1.90  dholland 				 * block didn't exist we'll need to write a
    382  1.90  dholland 				 * new version of all the blocks that lead to
    383  1.90  dholland 				 * it. Hopefully all that gets in before any
    384  1.90  dholland 				 * actual disk I/O so we don't end up writing
    385  1.90  dholland 				 * any of them twice... this is currently not
    386  1.90  dholland 				 * very clear.
    387  1.90  dholland 				 */
    388  1.70   hannken 				if ((error = VOP_BWRITE(ibp->b_vp, ibp)))
    389  1.46   mycroft 					return error;
    390  1.19  perseant 			}
    391  1.19  perseant 		}
    392  1.50     perry 	}
    393  1.19  perseant 
    394  1.19  perseant 
    395  1.19  perseant 	/*
    396  1.36  perseant 	 * Get the existing block from the cache, if requested.
    397  1.19  perseant 	 */
    398  1.36  perseant 	if (bpp)
    399  1.77  christos 		*bpp = bp = getblk(vp, lbn, lfs_blksize(fs, ip, lbn), 0, 0);
    400  1.50     perry 
    401  1.49  perseant 	/*
    402  1.49  perseant 	 * Do accounting on blocks that represent pages.
    403  1.49  perseant 	 */
    404  1.49  perseant 	if (!bpp)
    405  1.49  perseant 		lfs_register_block(vp, lbn);
    406  1.49  perseant 
    407  1.50     perry 	/*
    408   1.5      fvdl 	 * The block we are writing may be a brand new block
    409  1.19  perseant 	 * in which case we need to do accounting.
    410  1.15  perseant 	 *
    411  1.75  dholland 	 * We can tell a truly new block because ulfs_bmaparray will say
    412  1.90  dholland 	 * it is UNASSIGNED. Once we allocate it we will assign it the
    413  1.19  perseant 	 * disk address UNWRITTEN.
    414   1.5      fvdl 	 */
    415  1.16  perseant 	if (daddr == UNASSIGNED) {
    416  1.36  perseant 		if (bpp) {
    417  1.57      yamt 			if (flags & B_CLRBUF)
    418  1.36  perseant 				clrbuf(bp);
    419  1.50     perry 
    420  1.36  perseant 			/* Note the new address */
    421  1.36  perseant 			bp->b_blkno = UNWRITTEN;
    422  1.36  perseant 		}
    423  1.50     perry 
    424  1.19  perseant 		switch (num) {
    425  1.19  perseant 		    case 0:
    426  1.90  dholland 			/* direct block - update the inode */
    427  1.87  dholland 			lfs_dino_setdb(fs, ip->i_din, lbn, UNWRITTEN);
    428  1.19  perseant 			break;
    429  1.19  perseant 		    case 1:
    430  1.90  dholland 			/*
    431  1.90  dholland 			 * using a single indirect block - update the inode
    432  1.90  dholland 			 *
    433  1.90  dholland 			 * XXX: is this right? We already set this block
    434  1.90  dholland 			 * pointer above. I think we want to be writing *in*
    435  1.90  dholland 			 * the single indirect block and this case shouldn't
    436  1.90  dholland 			 * exist. (just case 0 and default)
    437  1.90  dholland 			 *    -- dholland 20160806
    438  1.90  dholland 			 */
    439  1.87  dholland 			lfs_dino_setib(fs, ip->i_din, indirs[0].in_off, UNWRITTEN);
    440  1.19  perseant 			break;
    441  1.19  perseant 		    default:
    442  1.90  dholland 			/*
    443  1.90  dholland 			 * using multiple indirect blocks - update the
    444  1.90  dholland 			 * innermost one
    445  1.90  dholland 			 */
    446  1.19  perseant 			idp = &indirs[num - 1];
    447  1.82  dholland 			if (bread(vp, idp->in_lbn, lfs_sb_getbsize(fs),
    448  1.67   hannken 				  B_MODIFY, &ibp))
    449  1.35      fvdl 				panic("lfs_balloc: bread bno %lld",
    450  1.35      fvdl 				    (long long)idp->in_lbn);
    451  1.88  dholland 			lfs_iblock_set(fs, ibp->b_data, idp->in_off, UNWRITTEN);
    452  1.92      maya 
    453  1.52  perseant 			if (vp == fs->lfs_ivnode) {
    454  1.52  perseant 				LFS_ENTER_LOG("balloc", __FILE__,
    455  1.52  perseant 					__LINE__, idp->in_lbn,
    456  1.52  perseant 					ibp->b_flags, curproc->p_pid);
    457  1.52  perseant 			}
    458  1.92      maya 
    459  1.70   hannken 			VOP_BWRITE(ibp->b_vp, ibp);
    460  1.15  perseant 		}
    461  1.64        ad 	} else if (bpp && !(bp->b_oflags & (BO_DONE|BO_DELWRI))) {
    462  1.15  perseant 		/*
    463  1.15  perseant 		 * Not a brand new block, also not in the cache;
    464  1.15  perseant 		 * read it in from disk.
    465  1.15  perseant 		 */
    466  1.82  dholland 		if (iosize == lfs_sb_getbsize(fs))
    467   1.5      fvdl 			/* Optimization: I/O is unnecessary. */
    468   1.5      fvdl 			bp->b_blkno = daddr;
    469  1.15  perseant 		else {
    470   1.5      fvdl 			/*
    471   1.5      fvdl 			 * We need to read the block to preserve the
    472   1.5      fvdl 			 * existing bytes.
    473   1.5      fvdl 			 */
    474   1.1   mycroft 			bp->b_blkno = daddr;
    475   1.1   mycroft 			bp->b_flags |= B_READ;
    476  1.48   hannken 			VOP_STRATEGY(vp, bp);
    477  1.31       chs 			return (biowait(bp));
    478   1.1   mycroft 		}
    479   1.1   mycroft 	}
    480  1.50     perry 
    481   1.5      fvdl 	return (0);
    482   1.5      fvdl }
    483   1.5      fvdl 
    484  1.90  dholland /*
    485  1.90  dholland  * Extend a file that uses fragments with more fragments.
    486  1.90  dholland  *
    487  1.90  dholland  * XXX: locking?
    488  1.90  dholland  */
    489  1.25  perseant /* VOP_BWRITE 1 time */
    490  1.89  dholland static int
    491  1.89  dholland lfs_fragextend(struct vnode *vp, int osize, int nsize, daddr_t lbn,
    492  1.89  dholland     struct buf **bpp, kauth_cred_t cred)
    493   1.5      fvdl {
    494   1.5      fvdl 	struct inode *ip;
    495   1.5      fvdl 	struct lfs *fs;
    496  1.69   mlelstv 	long frags;
    497   1.5      fvdl 	int error;
    498  1.90  dholland 	size_t obufsize;
    499  1.90  dholland 
    500  1.90  dholland 	/* XXX move this to a header file */
    501  1.90  dholland 	/* (XXX: except it's not clear what purpose it serves) */
    502  1.11  perseant 	extern long locked_queue_bytes;
    503  1.90  dholland 
    504  1.91  dholland 	ip = VTOI(vp);
    505  1.91  dholland 	fs = ip->i_lfs;
    506  1.91  dholland 
    507  1.90  dholland 	/*
    508  1.90  dholland 	 * XXX: is there some reason we know more about the seglock
    509  1.90  dholland 	 * state here than at the top of lfs_balloc?
    510  1.90  dholland 	 */
    511  1.90  dholland 	ASSERT_NO_SEGLOCK(fs);
    512   1.5      fvdl 
    513  1.90  dholland 	/* number of frags we're adding */
    514  1.77  christos 	frags = (long)lfs_numfrags(fs, nsize - osize);
    515  1.90  dholland 
    516  1.18  perseant 	error = 0;
    517  1.18  perseant 
    518  1.18  perseant 	/*
    519  1.36  perseant 	 * Get the seglock so we don't enlarge blocks while a segment
    520  1.36  perseant 	 * is being written.  If we're called with bpp==NULL, though,
    521  1.36  perseant 	 * we are only pretending to change a buffer, so we don't have to
    522  1.36  perseant 	 * lock.
    523  1.90  dholland 	 *
    524  1.90  dholland 	 * XXX: the above comment is lying, as fs->lfs_fraglock is not
    525  1.90  dholland 	 * the segment lock.
    526  1.18  perseant 	 */
    527  1.26  perseant     top:
    528  1.36  perseant 	if (bpp) {
    529  1.62        ad 		rw_enter(&fs->lfs_fraglock, RW_READER);
    530  1.36  perseant 	}
    531  1.36  perseant 
    532  1.90  dholland 	/* check if we actually have enough frags available */
    533  1.69   mlelstv 	if (!ISSPACE(fs, frags, cred)) {
    534  1.18  perseant 		error = ENOSPC;
    535  1.18  perseant 		goto out;
    536   1.5      fvdl 	}
    537  1.36  perseant 
    538  1.36  perseant 	/*
    539  1.36  perseant 	 * If we are not asked to actually return the block, all we need
    540  1.36  perseant 	 * to do is allocate space for it.  UBC will handle dirtying the
    541  1.36  perseant 	 * appropriate things and making sure it all goes to disk.
    542  1.36  perseant 	 * Don't bother to read in that case.
    543  1.36  perseant 	 */
    544  1.81      maxv 	if (bpp && (error = bread(vp, lbn, osize, 0, bpp))) {
    545  1.18  perseant 		goto out;
    546   1.5      fvdl 	}
    547  1.80  dholland #if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
    548  1.76  dholland 	if ((error = lfs_chkdq(ip, frags, cred, 0))) {
    549  1.36  perseant 		if (bpp)
    550  1.63        ad 			brelse(*bpp, 0);
    551  1.19  perseant 		goto out;
    552  1.19  perseant 	}
    553  1.19  perseant #endif
    554  1.13  perseant 	/*
    555  1.26  perseant 	 * Adjust accounting for lfs_avail.  If there's not enough room,
    556  1.26  perseant 	 * we will have to wait for the cleaner, which we can't do while
    557  1.26  perseant 	 * holding a block busy or while holding the seglock.  In that case,
    558  1.26  perseant 	 * release both and start over after waiting.
    559  1.26  perseant 	 */
    560  1.33  perseant 
    561  1.64        ad 	if (bpp && ((*bpp)->b_oflags & BO_DELWRI)) {
    562  1.69   mlelstv 		if (!lfs_fits(fs, frags)) {
    563  1.36  perseant 			if (bpp)
    564  1.63        ad 				brelse(*bpp, 0);
    565  1.80  dholland #if defined(LFS_QUOTA) || defined(LFS_QUOTA2)
    566  1.76  dholland 			lfs_chkdq(ip, -frags, cred, 0);
    567  1.26  perseant #endif
    568  1.62        ad 			rw_exit(&fs->lfs_fraglock);
    569  1.69   mlelstv 			lfs_availwait(fs, frags);
    570  1.26  perseant 			goto top;
    571  1.26  perseant 		}
    572  1.82  dholland 		lfs_sb_subavail(fs, frags);
    573  1.26  perseant 	}
    574  1.26  perseant 
    575  1.90  dholland 	/* decrease the free block count in the superblock */
    576  1.64        ad 	mutex_enter(&lfs_lock);
    577  1.85  dholland 	lfs_sb_subbfree(fs, frags);
    578  1.64        ad 	mutex_exit(&lfs_lock);
    579  1.90  dholland 	/* increase the file's effective block count */
    580  1.69   mlelstv 	ip->i_lfs_effnblks += frags;
    581  1.90  dholland 	/* mark the inode dirty */
    582  1.94      maya 	ip->i_state |= IN_CHANGE | IN_UPDATE;
    583  1.26  perseant 
    584  1.36  perseant 	if (bpp) {
    585  1.36  perseant 		obufsize = (*bpp)->b_bufsize;
    586  1.47        pk 		allocbuf(*bpp, nsize, 1);
    587  1.26  perseant 
    588  1.36  perseant 		/* Adjust locked-list accounting */
    589  1.65        ad 		if (((*bpp)->b_flags & B_LOCKED) != 0 &&
    590  1.64        ad 		    (*bpp)->b_iodone == NULL) {
    591  1.64        ad 			mutex_enter(&lfs_lock);
    592  1.36  perseant 			locked_queue_bytes += (*bpp)->b_bufsize - obufsize;
    593  1.64        ad 			mutex_exit(&lfs_lock);
    594  1.52  perseant 		}
    595  1.26  perseant 
    596  1.90  dholland 		/* zero the new space */
    597  1.68    cegger 		memset((char *)((*bpp)->b_data) + osize, 0, (u_int)(nsize - osize));
    598  1.36  perseant 	}
    599  1.18  perseant 
    600  1.18  perseant     out:
    601  1.36  perseant 	if (bpp) {
    602  1.62        ad 		rw_exit(&fs->lfs_fraglock);
    603  1.36  perseant 	}
    604  1.18  perseant 	return (error);
    605   1.1   mycroft }
    606  1.49  perseant 
    607  1.59     perry static inline int
    608  1.55  perseant lge(struct lbnentry *a, struct lbnentry *b)
    609  1.53  perseant {
    610  1.55  perseant 	return a->lbn - b->lbn;
    611  1.53  perseant }
    612  1.53  perseant 
    613  1.55  perseant SPLAY_PROTOTYPE(lfs_splay, lbnentry, entry, lge);
    614  1.55  perseant 
    615  1.55  perseant SPLAY_GENERATE(lfs_splay, lbnentry, entry, lge);
    616  1.53  perseant 
    617  1.49  perseant /*
    618  1.49  perseant  * Record this lbn as being "write pending".  We used to have this information
    619  1.49  perseant  * on the buffer headers, but since pages don't have buffer headers we
    620  1.49  perseant  * record it here instead.
    621  1.49  perseant  */
    622  1.49  perseant void
    623  1.49  perseant lfs_register_block(struct vnode *vp, daddr_t lbn)
    624  1.49  perseant {
    625  1.49  perseant 	struct lfs *fs;
    626  1.49  perseant 	struct inode *ip;
    627  1.49  perseant 	struct lbnentry *lbp;
    628  1.53  perseant 
    629  1.53  perseant 	ip = VTOI(vp);
    630  1.49  perseant 
    631  1.49  perseant 	/* Don't count metadata */
    632  1.53  perseant 	if (lbn < 0 || vp->v_type != VREG || ip->i_number == LFS_IFILE_INUM)
    633  1.49  perseant 		return;
    634  1.49  perseant 
    635  1.49  perseant 	fs = ip->i_lfs;
    636  1.49  perseant 
    637  1.52  perseant 	ASSERT_NO_SEGLOCK(fs);
    638  1.52  perseant 
    639  1.49  perseant 	/* If no space, wait for the cleaner */
    640  1.83  dholland 	lfs_availwait(fs, lfs_btofsb(fs, 1 << lfs_sb_getbshift(fs)));
    641  1.49  perseant 
    642  1.49  perseant 	lbp = (struct lbnentry *)pool_get(&lfs_lbnentry_pool, PR_WAITOK);
    643  1.49  perseant 	lbp->lbn = lbn;
    644  1.64        ad 	mutex_enter(&lfs_lock);
    645  1.55  perseant 	if (SPLAY_INSERT(lfs_splay, &ip->i_lfs_lbtree, lbp) != NULL) {
    646  1.64        ad 		mutex_exit(&lfs_lock);
    647  1.55  perseant 		/* Already there */
    648  1.55  perseant 		pool_put(&lfs_lbnentry_pool, lbp);
    649  1.55  perseant 		return;
    650  1.55  perseant 	}
    651  1.52  perseant 
    652  1.56  perseant 	++ip->i_lfs_nbtree;
    653  1.83  dholland 	fs->lfs_favail += lfs_btofsb(fs, (1 << lfs_sb_getbshift(fs)));
    654  1.82  dholland 	fs->lfs_pages += lfs_sb_getbsize(fs) >> PAGE_SHIFT;
    655  1.49  perseant 	++locked_fakequeue_count;
    656  1.82  dholland 	lfs_subsys_pages += lfs_sb_getbsize(fs) >> PAGE_SHIFT;
    657  1.64        ad 	mutex_exit(&lfs_lock);
    658  1.49  perseant }
    659  1.49  perseant 
    660  1.49  perseant static void
    661  1.53  perseant lfs_do_deregister(struct lfs *fs, struct inode *ip, struct lbnentry *lbp)
    662  1.49  perseant {
    663  1.95  riastrad 
    664  1.95  riastrad 	KASSERT(mutex_owned(&lfs_lock));
    665  1.52  perseant 	ASSERT_MAYBE_SEGLOCK(fs);
    666  1.52  perseant 
    667  1.56  perseant 	--ip->i_lfs_nbtree;
    668  1.55  perseant 	SPLAY_REMOVE(lfs_splay, &ip->i_lfs_lbtree, lbp);
    669  1.83  dholland 	if (fs->lfs_favail > lfs_btofsb(fs, (1 << lfs_sb_getbshift(fs))))
    670  1.83  dholland 		fs->lfs_favail -= lfs_btofsb(fs, (1 << lfs_sb_getbshift(fs)));
    671  1.82  dholland 	fs->lfs_pages -= lfs_sb_getbsize(fs) >> PAGE_SHIFT;
    672  1.49  perseant 	if (locked_fakequeue_count > 0)
    673  1.49  perseant 		--locked_fakequeue_count;
    674  1.82  dholland 	lfs_subsys_pages -= lfs_sb_getbsize(fs) >> PAGE_SHIFT;
    675  1.95  riastrad 
    676  1.64        ad 	mutex_exit(&lfs_lock);
    677  1.95  riastrad 	pool_put(&lfs_lbnentry_pool, lbp);
    678  1.95  riastrad 	mutex_enter(&lfs_lock);
    679  1.64        ad 
    680  1.95  riastrad 	KASSERT(mutex_owned(&lfs_lock));
    681  1.49  perseant }
    682  1.49  perseant 
    683  1.49  perseant void
    684  1.49  perseant lfs_deregister_block(struct vnode *vp, daddr_t lbn)
    685  1.49  perseant {
    686  1.49  perseant 	struct lfs *fs;
    687  1.49  perseant 	struct inode *ip;
    688  1.49  perseant 	struct lbnentry *lbp;
    689  1.55  perseant 	struct lbnentry tmp;
    690  1.53  perseant 
    691  1.53  perseant 	ip = VTOI(vp);
    692  1.49  perseant 
    693  1.49  perseant 	/* Don't count metadata */
    694  1.53  perseant 	if (lbn < 0 || vp->v_type != VREG || ip->i_number == LFS_IFILE_INUM)
    695  1.49  perseant 		return;
    696  1.49  perseant 
    697  1.95  riastrad 	mutex_enter(&lfs_lock);
    698  1.49  perseant 	fs = ip->i_lfs;
    699  1.55  perseant 	tmp.lbn = lbn;
    700  1.95  riastrad 	if ((lbp = SPLAY_FIND(lfs_splay, &ip->i_lfs_lbtree, &tmp)) != NULL)
    701  1.95  riastrad 		lfs_do_deregister(fs, ip, lbp);
    702  1.95  riastrad 	mutex_exit(&lfs_lock);
    703  1.49  perseant }
    704  1.55  perseant 
    705  1.55  perseant void
    706  1.55  perseant lfs_deregister_all(struct vnode *vp)
    707  1.55  perseant {
    708  1.95  riastrad 	struct lbnentry *lbp;
    709  1.55  perseant 	struct lfs_splay *hd;
    710  1.55  perseant 	struct lfs *fs;
    711  1.55  perseant 	struct inode *ip;
    712  1.55  perseant 
    713  1.55  perseant 	ip = VTOI(vp);
    714  1.55  perseant 	fs = ip->i_lfs;
    715  1.55  perseant 	hd = &ip->i_lfs_lbtree;
    716  1.55  perseant 
    717  1.95  riastrad 	mutex_enter(&lfs_lock);
    718  1.95  riastrad 	while ((lbp = SPLAY_MIN(lfs_splay, hd)) != NULL)
    719  1.55  perseant 		lfs_do_deregister(fs, ip, lbp);
    720  1.95  riastrad 	mutex_exit(&lfs_lock);
    721  1.55  perseant }
    722