1 /* $NetBSD: lfs_balloc.c,v 1.32 2002/05/14 20:03:53 perseant Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant@hhhh.org>. 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 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 /* 39 * Copyright (c) 1989, 1991, 1993 40 * The Regents of the University of California. All rights reserved. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. All advertising materials mentioning features or use of this software 51 * must display the following acknowledgement: 52 * This product includes software developed by the University of 53 * California, Berkeley and its contributors. 54 * 4. Neither the name of the University nor the names of its contributors 55 * may be used to endorse or promote products derived from this software 56 * without specific prior written permission. 57 * 58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 68 * SUCH DAMAGE. 69 * 70 * @(#)lfs_balloc.c 8.4 (Berkeley) 5/8/95 71 */ 72 73 #include <sys/cdefs.h> 74 __KERNEL_RCSID(0, "$NetBSD: lfs_balloc.c,v 1.32 2002/05/14 20:03:53 perseant Exp $"); 75 76 #if defined(_KERNEL_OPT) 77 #include "opt_quota.h" 78 #endif 79 80 #include <sys/param.h> 81 #include <sys/systm.h> 82 #include <sys/buf.h> 83 #include <sys/proc.h> 84 #include <sys/vnode.h> 85 #include <sys/mount.h> 86 #include <sys/resourcevar.h> 87 #include <sys/trace.h> 88 89 #include <miscfs/specfs/specdev.h> 90 91 #include <ufs/ufs/quota.h> 92 #include <ufs/ufs/inode.h> 93 #include <ufs/ufs/ufsmount.h> 94 #include <ufs/ufs/ufs_extern.h> 95 96 #include <ufs/lfs/lfs.h> 97 #include <ufs/lfs/lfs_extern.h> 98 99 int lfs_fragextend(struct vnode *, int, int, ufs_daddr_t, struct buf **, struct ucred *); 100 101 /* 102 * Allocate a block, and to inode and filesystem block accounting for it 103 * and for any indirect blocks the may need to be created in order for 104 * this block to be created. 105 * 106 * Blocks which have never been accounted for (i.e., which "do not exist") 107 * have disk address 0, which is translated by ufs_bmap to the special value 108 * UNASSIGNED == -1, as in the historical UFS. 109 * 110 * Blocks which have been accounted for but which have not yet been written 111 * to disk are given the new special disk address UNWRITTEN == -2, so that 112 * they can be differentiated from completely new blocks. 113 */ 114 /* VOP_BWRITE NIADDR+2 times */ 115 int 116 lfs_balloc(void *v) 117 { 118 struct vop_balloc_args /* { 119 struct vnode *a_vp; 120 off_t a_startoffset; 121 int a_size; 122 struct ucred *a_cred; 123 int a_flags; 124 struct buf *a_bpp; 125 } */ *ap = v; 126 struct vnode *vp; 127 int offset; 128 u_long iosize; 129 daddr_t daddr, idaddr; 130 struct buf *ibp, *bp; 131 struct inode *ip; 132 struct lfs *fs; 133 struct indir indirs[NIADDR+2], *idp; 134 ufs_daddr_t lbn, lastblock; 135 int bb, bcount; 136 int error, frags, i, nsize, osize, num; 137 138 vp = ap->a_vp; 139 ip = VTOI(vp); 140 fs = ip->i_lfs; 141 offset = blkoff(fs, ap->a_startoffset); 142 iosize = ap->a_size; 143 lbn = lblkno(fs, ap->a_startoffset); 144 (void)lfs_check(vp, lbn, 0); 145 146 /* 147 * Three cases: it's a block beyond the end of file, it's a block in 148 * the file that may or may not have been assigned a disk address or 149 * we're writing an entire block. 150 * 151 * Note, if the daddr is UNWRITTEN, the block already exists in 152 * the cache (it was read or written earlier). If so, make sure 153 * we don't count it as a new block or zero out its contents. If 154 * it did not, make sure we allocate any necessary indirect 155 * blocks. 156 * 157 * If we are writing a block beyond the end of the file, we need to 158 * check if the old last block was a fragment. If it was, we need 159 * to rewrite it. 160 */ 161 162 *ap->a_bpp = NULL; 163 164 /* Check for block beyond end of file and fragment extension needed. */ 165 lastblock = lblkno(fs, ip->i_ffs_size); 166 if (lastblock < NDADDR && lastblock < lbn) { 167 osize = blksize(fs, ip, lastblock); 168 if (osize < fs->lfs_bsize && osize > 0) { 169 if ((error = lfs_fragextend(vp, osize, fs->lfs_bsize, 170 lastblock, &bp, 171 ap->a_cred))) 172 return (error); 173 ip->i_ffs_size = (lastblock + 1) * fs->lfs_bsize; 174 uvm_vnp_setsize(vp, ip->i_ffs_size); 175 ip->i_flag |= IN_CHANGE | IN_UPDATE; 176 (void) VOP_BWRITE(bp); 177 } 178 } 179 180 /* 181 * If the block we are writing is a direct block, it's the last 182 * block in the file, and offset + iosize is less than a full 183 * block, we can write one or more fragments. There are two cases: 184 * the block is brand new and we should allocate it the correct 185 * size or it already exists and contains some fragments and 186 * may need to extend it. 187 */ 188 if (lbn < NDADDR && lblkno(fs, ip->i_ffs_size) <= lbn) { 189 osize = blksize(fs, ip, lbn); 190 nsize = fragroundup(fs, offset + iosize); 191 if (lblktosize(fs, lbn) >= ip->i_ffs_size) { 192 /* Brand new block or fragment */ 193 frags = numfrags(fs, nsize); 194 bb = fragstofsb(fs, frags); 195 *ap->a_bpp = bp = getblk(vp, lbn, nsize, 0, 0); 196 ip->i_lfs_effnblks += bb; 197 ip->i_lfs->lfs_bfree -= bb; 198 ip->i_ffs_db[lbn] = bp->b_blkno = UNWRITTEN; 199 } else { 200 if (nsize <= osize) { 201 /* No need to extend */ 202 if ((error = bread(vp, lbn, osize, NOCRED, &bp))) 203 return error; 204 } else { 205 /* Extend existing block */ 206 if ((error = 207 lfs_fragextend(vp, osize, nsize, lbn, &bp, 208 ap->a_cred))) 209 return error; 210 } 211 *ap->a_bpp = bp; 212 } 213 return 0; 214 } 215 216 error = ufs_bmaparray(vp, lbn, &daddr, &indirs[0], &num, NULL ); 217 if (error) 218 return (error); 219 /* 220 * Do byte accounting all at once, so we can gracefully fail *before* 221 * we start assigning blocks. 222 */ 223 bb = VFSTOUFS(vp->v_mount)->um_seqinc; 224 bcount = 0; 225 if (daddr == UNASSIGNED) { 226 bcount = bb; 227 } 228 for (i = 1; i < num; ++i) { 229 if (!indirs[i].in_exists) { 230 bcount += bb; 231 } 232 } 233 if (ISSPACE(fs, bcount, ap->a_cred)) { 234 ip->i_lfs->lfs_bfree -= bcount; 235 ip->i_lfs_effnblks += bcount; 236 } else { 237 return ENOSPC; 238 } 239 240 if (daddr == UNASSIGNED) { 241 if (num > 0 && ip->i_ffs_ib[indirs[0].in_off] == 0) { 242 ip->i_ffs_ib[indirs[0].in_off] = UNWRITTEN; 243 } 244 245 /* 246 * Create new indirect blocks if necessary 247 */ 248 if (num > 1) 249 idaddr = ip->i_ffs_ib[indirs[0].in_off]; 250 for (i = 1; i < num; ++i) { 251 ibp = getblk(vp, indirs[i].in_lbn, fs->lfs_bsize, 0,0); 252 if (!indirs[i].in_exists) { 253 clrbuf(ibp); 254 ibp->b_blkno = UNWRITTEN; 255 } else if (!(ibp->b_flags & (B_DELWRI | B_DONE))) { 256 ibp->b_blkno = fsbtodb(fs, idaddr); 257 ibp->b_flags |= B_READ; 258 VOP_STRATEGY(ibp); 259 biowait(ibp); 260 } 261 /* 262 * This block exists, but the next one may not. 263 * If that is the case mark it UNWRITTEN to keep 264 * the accounting straight. 265 */ 266 if (((daddr_t *)ibp->b_data)[indirs[i].in_off] == 0) 267 ((daddr_t *)ibp->b_data)[indirs[i].in_off] = 268 UNWRITTEN; 269 idaddr = ((daddr_t *)ibp->b_data)[indirs[i].in_off]; 270 if ((error = VOP_BWRITE(ibp))) { 271 return error; 272 } 273 } 274 } 275 276 277 /* 278 * Get the existing block from the cache. 279 */ 280 frags = fsbtofrags(fs, bb); 281 *ap->a_bpp = bp = getblk(vp, lbn, blksize(fs, ip, lbn), 0, 0); 282 283 /* 284 * The block we are writing may be a brand new block 285 * in which case we need to do accounting. 286 * 287 * We can tell a truly new block because ufs_bmaparray will say 288 * it is UNASSIGNED. Once we allocate it we will assign it the 289 * disk address UNWRITTEN. 290 */ 291 if (daddr == UNASSIGNED) { 292 if (iosize != fs->lfs_bsize) 293 clrbuf(bp); 294 295 /* Note the new address */ 296 bp->b_blkno = UNWRITTEN; 297 298 switch (num) { 299 case 0: 300 ip->i_ffs_db[lbn] = UNWRITTEN; 301 break; 302 case 1: 303 ip->i_ffs_ib[indirs[0].in_off] = UNWRITTEN; 304 break; 305 default: 306 idp = &indirs[num - 1]; 307 if (bread(vp, idp->in_lbn, fs->lfs_bsize, NOCRED, 308 &ibp)) 309 panic("lfs_balloc: bread bno %d", idp->in_lbn); 310 ((ufs_daddr_t *)ibp->b_data)[idp->in_off] = UNWRITTEN; 311 VOP_BWRITE(ibp); 312 } 313 } else if (!(bp->b_flags & (B_DONE|B_DELWRI))) { 314 /* 315 * Not a brand new block, also not in the cache; 316 * read it in from disk. 317 */ 318 if (iosize == fs->lfs_bsize) 319 /* Optimization: I/O is unnecessary. */ 320 bp->b_blkno = daddr; 321 else { 322 /* 323 * We need to read the block to preserve the 324 * existing bytes. 325 */ 326 bp->b_blkno = daddr; 327 bp->b_flags |= B_READ; 328 VOP_STRATEGY(bp); 329 return (biowait(bp)); 330 } 331 } 332 333 return (0); 334 } 335 336 /* VOP_BWRITE 1 time */ 337 int 338 lfs_fragextend(struct vnode *vp, int osize, int nsize, ufs_daddr_t lbn, struct buf **bpp, struct ucred *cred) 339 { 340 struct inode *ip; 341 struct lfs *fs; 342 long bb; 343 int error; 344 extern long locked_queue_bytes; 345 struct buf *ibp; 346 size_t obufsize; 347 SEGUSE *sup; 348 349 ip = VTOI(vp); 350 fs = ip->i_lfs; 351 bb = (long)fragstofsb(fs, numfrags(fs, nsize - osize)); 352 error = 0; 353 354 /* 355 * Get the seglock so we don't enlarge blocks or change the segment 356 * accounting information while a segment is being written. 357 */ 358 top: 359 lfs_seglock(fs, SEGM_PROT); 360 361 if (!ISSPACE(fs, bb, cred)) { 362 error = ENOSPC; 363 goto out; 364 } 365 if ((error = bread(vp, lbn, osize, NOCRED, bpp))) { 366 brelse(*bpp); 367 goto out; 368 } 369 #ifdef QUOTA 370 if ((error = chkdq(ip, bb, cred, 0))) { 371 brelse(*bpp); 372 goto out; 373 } 374 #endif 375 /* 376 * Adjust accounting for lfs_avail. If there's not enough room, 377 * we will have to wait for the cleaner, which we can't do while 378 * holding a block busy or while holding the seglock. In that case, 379 * release both and start over after waiting. 380 */ 381 if ((*bpp)->b_flags & B_DELWRI) { 382 if (!lfs_fits(fs, bb)) { 383 brelse(*bpp); 384 #ifdef QUOTA 385 chkdq(ip, -bb, cred, 0); 386 #endif 387 lfs_segunlock(fs); 388 lfs_availwait(fs, bb); 389 goto top; 390 } 391 fs->lfs_avail -= bb; 392 } 393 394 /* 395 * Fix the allocation for this fragment so that it looks like the 396 * source segment contained a block of the new size. This overcounts; 397 * but the overcount only lasts until the block in question 398 * is written, so the on-disk live bytes count is always correct. 399 */ 400 if ((*bpp)->b_blkno > 0) { 401 LFS_SEGENTRY(sup, fs, dtosn(fs, dbtofsb(fs, (*bpp)->b_blkno)), ibp); 402 sup->su_nbytes += (nsize - osize); 403 LFS_BWRITE_LOG(ibp); 404 ip->i_ffs_blocks += bb; 405 } 406 fs->lfs_bfree -= bb; 407 ip->i_lfs_effnblks += bb; 408 ip->i_flag |= IN_CHANGE | IN_UPDATE; 409 410 LFS_DEBUG_COUNTLOCKED("frag1"); 411 412 obufsize = (*bpp)->b_bufsize; 413 allocbuf(*bpp, nsize); 414 415 /* Adjust locked-list accounting */ 416 if (((*bpp)->b_flags & (B_LOCKED | B_CALL)) == B_LOCKED) 417 locked_queue_bytes += (*bpp)->b_bufsize - obufsize; 418 419 LFS_DEBUG_COUNTLOCKED("frag2"); 420 421 bzero((char *)((*bpp)->b_data) + osize, (u_int)(nsize - osize)); 422 423 out: 424 lfs_segunlock(fs); 425 return (error); 426 } 427