ext2_alloc.c revision 16322
1/* 2 * modified for Lites 1.1 3 * 4 * Aug 1995, Godmar Back (gback@cs.utah.edu) 5 * University of Utah, Department of Computer Science 6 */ 7/* 8 * Copyright (c) 1982, 1986, 1989, 1993 9 * The Regents of the University of California. All rights reserved. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the University of 22 * California, Berkeley and its contributors. 23 * 4. Neither the name of the University nor the names of its contributors 24 * may be used to endorse or promote products derived from this software 25 * without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 30 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 37 * SUCH DAMAGE. 38 * 39 * @(#)ext2_alloc.c 8.8 (Berkeley) 2/21/94 40 */ 41 42#if !defined(__FreeBSD__) 43#include "quota.h" 44#include "diagnostic.h" 45#else 46#include "opt_quota.h" 47#endif 48 49#include <sys/param.h> 50#include <sys/systm.h> 51#include <sys/buf.h> 52#include <sys/proc.h> 53#include <sys/vnode.h> 54#include <sys/stat.h> 55#include <sys/mount.h> 56#include <sys/kernel.h> 57#include <sys/syslog.h> 58 59#include <vm/vm.h> 60 61#include <ufs/ufs/quota.h> 62#include <ufs/ufs/inode.h> 63 64#include <gnu/ext2fs/ext2_fs.h> 65#include <gnu/ext2fs/ext2_fs_sb.h> 66#include <gnu/ext2fs/fs.h> 67#include <gnu/ext2fs/ext2_extern.h> 68 69extern u_long nextgennumber; 70 71static void ext2_fserr __P((struct ext2_sb_info *, u_int, char *)); 72 73/* 74 * Linux calls this functions at the following locations: 75 * (1) the inode is freed 76 * (2) a preallocation miss occurs 77 * (3) truncate is called 78 * (4) release_file is called and f_mode & 2 79 * 80 * I call it in ext2_inactive, ext2_truncate, ext2_vfree and in (2) 81 * the call in vfree might be redundant 82 */ 83void ext2_discard_prealloc (struct inode * ip) 84{ 85#ifdef EXT2_PREALLOCATE 86 if (ip->i_prealloc_count) { 87 int i = ip->i_prealloc_count; 88 ip->i_prealloc_count = 0; 89 ext2_free_blocks (ITOV(ip)->v_mount, 90 ip->i_prealloc_block, 91 i); 92 } 93#endif 94} 95 96/* 97 * Allocate a block in the file system. 98 * 99 * this takes the framework from ffs_alloc. To implement the 100 * actual allocation, it calls ext2_new_block, the ported version 101 * of the same Linux routine. 102 * 103 * we note that this is always called in connection with ext2_blkpref 104 * 105 * preallocation is done as Linux does it 106 */ 107int 108ext2_alloc(ip, lbn, bpref, size, cred, bnp) 109 register struct inode *ip; 110 daddr_t lbn, bpref; 111 int size; 112 struct ucred *cred; 113 daddr_t *bnp; 114{ 115 register struct ext2_sb_info *fs; 116 daddr_t bno; 117#if QUOTA 118 int error; 119#endif 120 121 *bnp = 0; 122 fs = ip->i_e2fs; 123#if DIAGNOSTIC 124 if ((u_int)size > fs->s_blocksize || blkoff(fs, size) != 0) { 125 printf("dev = 0x%x, bsize = %d, size = %d, fs = %s\n", 126 ip->i_dev, fs->s_blocksize, size, fs->fs_fsmnt); 127 panic("ext2_alloc: bad size"); 128 } 129 if (cred == NOCRED) 130 panic("ext2_alloc: missing credential\n"); 131#endif /* DIAGNOSTIC */ 132 if (size == fs->s_blocksize && fs->s_es->s_free_blocks_count == 0) 133 goto nospace; 134 if (cred->cr_uid != 0 && 135 fs->s_es->s_free_blocks_count < fs->s_es->s_r_blocks_count) 136 goto nospace; 137#if QUOTA 138 if (error = chkdq(ip, (long)btodb(size), cred, 0)) 139 return (error); 140#endif 141 if (bpref >= fs->s_es->s_blocks_count) 142 bpref = 0; 143 /* call the Linux code */ 144#ifdef EXT2_PREALLOCATE 145 /* To have a preallocation hit, we must 146 * - have at least one block preallocated 147 * - and our preferred block must have that block number or one below 148 */ 149 if (ip->i_prealloc_count && 150 (bpref == ip->i_prealloc_block || 151 bpref + 1 == ip->i_prealloc_block)) 152 { 153 bno = ip->i_prealloc_block++; 154 ip->i_prealloc_count--; 155 /* ext2_debug ("preallocation hit (%lu/%lu).\n", 156 ++alloc_hits, ++alloc_attempts); */ 157 158 /* Linux gets, clears, and releases the buffer at this 159 point - we don't have to that; we leave it to the caller 160 */ 161 } else { 162 ext2_discard_prealloc (ip); 163 /* ext2_debug ("preallocation miss (%lu/%lu).\n", 164 alloc_hits, ++alloc_attempts); */ 165 if (S_ISREG(ip->i_mode)) 166 bno = ext2_new_block 167 (ITOV(ip)->v_mount, bpref, 168 &ip->i_prealloc_count, 169 &ip->i_prealloc_block); 170 else 171 bno = (daddr_t)ext2_new_block(ITOV(ip)->v_mount, 172 bpref, 0, 0); 173 } 174#else 175 bno = (daddr_t)ext2_new_block(ITOV(ip)->v_mount, bpref, 0, 0); 176#endif 177 178 if (bno > 0) { 179 /* set next_alloc fields as done in block_getblk */ 180 ip->i_next_alloc_block = lbn; 181 ip->i_next_alloc_goal = bno; 182 183 ip->i_blocks += btodb(size); 184 ip->i_flag |= IN_CHANGE | IN_UPDATE; 185 *bnp = bno; 186 return (0); 187 } 188#if QUOTA 189 /* 190 * Restore user's disk quota because allocation failed. 191 */ 192 (void) chkdq(ip, (long)-btodb(size), cred, FORCE); 193#endif 194nospace: 195 ext2_fserr(fs, cred->cr_uid, "file system full"); 196 uprintf("\n%s: write failed, file system is full\n", fs->fs_fsmnt); 197 return (ENOSPC); 198} 199 200/* 201 * Reallocate a sequence of blocks into a contiguous sequence of blocks. 202 * 203 * The vnode and an array of buffer pointers for a range of sequential 204 * logical blocks to be made contiguous is given. The allocator attempts 205 * to find a range of sequential blocks starting as close as possible to 206 * an fs_rotdelay offset from the end of the allocation for the logical 207 * block immediately preceeding the current range. If successful, the 208 * physical block numbers in the buffer pointers and in the inode are 209 * changed to reflect the new allocation. If unsuccessful, the allocation 210 * is left unchanged. The success in doing the reallocation is returned. 211 * Note that the error return is not reflected back to the user. Rather 212 * the previous block allocation will be used. 213 */ 214 215#ifdef FANCY_REALLOC 216#include <sys/sysctl.h> 217static int doasyncfree = 1; 218#ifdef OPT_DEBUG 219SYSCTL_INT(_debug, 14, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, ""); 220#endif /* OPT_DEBUG */ 221#endif 222 223int 224ext2_reallocblks(ap) 225 struct vop_reallocblks_args /* { 226 struct vnode *a_vp; 227 struct cluster_save *a_buflist; 228 } */ *ap; 229{ 230#ifndef FANCY_REALLOC 231/* printf("ext2_reallocblks not implemented\n"); */ 232return ENOSPC; 233#else 234 235 struct ext2_sb_info *fs; 236 struct inode *ip; 237 struct vnode *vp; 238 struct buf *sbp, *ebp; 239 daddr_t *bap, *sbap, *ebap; 240 struct cluster_save *buflist; 241 daddr_t start_lbn, end_lbn, soff, eoff, newblk, blkno; 242 struct indir start_ap[NIADDR + 1], end_ap[NIADDR + 1], *idp; 243 int i, len, start_lvl, end_lvl, pref, ssize; 244 245 vp = ap->a_vp; 246 ip = VTOI(vp); 247 fs = ip->i_e2fs; 248#ifdef UNKLAR 249 if (fs->fs_contigsumsize <= 0) 250 return (ENOSPC); 251#endif 252 buflist = ap->a_buflist; 253 len = buflist->bs_nchildren; 254 start_lbn = buflist->bs_children[0]->b_lblkno; 255 end_lbn = start_lbn + len - 1; 256#if DIAGNOSTIC 257 for (i = 1; i < len; i++) 258 if (buflist->bs_children[i]->b_lblkno != start_lbn + i) 259 panic("ext2_reallocblks: non-cluster"); 260#endif 261 /* 262 * If the latest allocation is in a new cylinder group, assume that 263 * the filesystem has decided to move and do not force it back to 264 * the previous cylinder group. 265 */ 266 if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != 267 dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) 268 return (ENOSPC); 269 if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || 270 ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) 271 return (ENOSPC); 272 /* 273 * Get the starting offset and block map for the first block. 274 */ 275 if (start_lvl == 0) { 276 sbap = &ip->i_db[0]; 277 soff = start_lbn; 278 } else { 279 idp = &start_ap[start_lvl - 1]; 280 if (bread(vp, idp->in_lbn, (int)fs->s_blocksize, NOCRED, &sbp)) { 281 brelse(sbp); 282 return (ENOSPC); 283 } 284 sbap = (daddr_t *)sbp->b_data; 285 soff = idp->in_off; 286 } 287 /* 288 * Find the preferred location for the cluster. 289 */ 290 pref = ext2_blkpref(ip, start_lbn, soff, sbap); 291 /* 292 * If the block range spans two block maps, get the second map. 293 */ 294 if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { 295 ssize = len; 296 } else { 297#if DIAGNOSTIC 298 if (start_ap[start_lvl-1].in_lbn == idp->in_lbn) 299 panic("ext2_reallocblk: start == end"); 300#endif 301 ssize = len - (idp->in_off + 1); 302 if (bread(vp, idp->in_lbn, (int)fs->s_blocksize, NOCRED, &ebp)) 303 goto fail; 304 ebap = (daddr_t *)ebp->b_data; 305 } 306 /* 307 * Search the block map looking for an allocation of the desired size. 308 */ 309 if ((newblk = (daddr_t)ext2_hashalloc(ip, dtog(fs, pref), (long)pref, 310 len, (u_long (*)())ext2_clusteralloc)) == 0) 311 goto fail; 312 /* 313 * We have found a new contiguous block. 314 * 315 * First we have to replace the old block pointers with the new 316 * block pointers in the inode and indirect blocks associated 317 * with the file. 318 */ 319 blkno = newblk; 320 for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->s_frags_per_block) { 321 if (i == ssize) 322 bap = ebap; 323#if DIAGNOSTIC 324 if (buflist->bs_children[i]->b_blkno != fsbtodb(fs, *bap)) 325 panic("ext2_reallocblks: alloc mismatch"); 326#endif 327 *bap++ = blkno; 328 } 329 /* 330 * Next we must write out the modified inode and indirect blocks. 331 * For strict correctness, the writes should be synchronous since 332 * the old block values may have been written to disk. In practise 333 * they are almost never written, but if we are concerned about 334 * strict correctness, the `doasyncfree' flag should be set to zero. 335 * 336 * The test on `doasyncfree' should be changed to test a flag 337 * that shows whether the associated buffers and inodes have 338 * been written. The flag should be set when the cluster is 339 * started and cleared whenever the buffer or inode is flushed. 340 * We can then check below to see if it is set, and do the 341 * synchronous write only when it has been cleared. 342 */ 343 if (sbap != &ip->i_db[0]) { 344 if (doasyncfree) 345 bdwrite(sbp); 346 else 347 bwrite(sbp); 348 } else { 349#if !defined(__FreeBSD__) 350 struct timeval time; 351 get_time(&time); 352#endif 353 ip->i_flag |= IN_CHANGE | IN_UPDATE; 354 if (!doasyncfree) 355 VOP_UPDATE(vp, &time, &time, MNT_WAIT); 356 } 357 if (ssize < len) 358 if (doasyncfree) 359 bdwrite(ebp); 360 else 361 bwrite(ebp); 362 /* 363 * Last, free the old blocks and assign the new blocks to the buffers. 364 */ 365 for (blkno = newblk, i = 0; i < len; i++, blkno += fs->s_frags_per_block) { 366 ext2_blkfree(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), 367 fs->s_blocksize); 368 buflist->bs_children[i]->b_blkno = fsbtodb(fs, blkno); 369 } 370 return (0); 371 372fail: 373 if (ssize < len) 374 brelse(ebp); 375 if (sbap != &ip->i_db[0]) 376 brelse(sbp); 377 return (ENOSPC); 378 379#endif /* FANCY_REALLOC */ 380} 381 382/* 383 * Allocate an inode in the file system. 384 * 385 * we leave the actual allocation strategy to the (modified) 386 * ext2_new_inode(), to make sure we get the policies right 387 */ 388int 389ext2_valloc(ap) 390 struct vop_valloc_args /* { 391 struct vnode *a_pvp; 392 int a_mode; 393 struct ucred *a_cred; 394 struct vnode **a_vpp; 395 } */ *ap; 396{ 397 register struct vnode *pvp = ap->a_pvp; 398 register struct inode *pip; 399 register struct ext2_sb_info *fs; 400 register struct inode *ip; 401 mode_t mode = ap->a_mode; 402 ino_t ino; 403 int i, error; 404#if !defined(__FreeBSD__) 405 struct timeval time; 406#endif 407 408 *ap->a_vpp = NULL; 409 pip = VTOI(pvp); 410 fs = pip->i_e2fs; 411 if (fs->s_es->s_free_inodes_count == 0) 412 goto noinodes; 413 414 /* call the Linux routine - it returns the inode number only */ 415 ino = ext2_new_inode(pip, mode); 416 417 if (ino == 0) 418 goto noinodes; 419 error = VFS_VGET(pvp->v_mount, ino, ap->a_vpp); 420 if (error) { 421 VOP_VFREE(pvp, ino, mode); 422 return (error); 423 } 424 ip = VTOI(*ap->a_vpp); 425 426 /* 427 the question is whether using VGET was such good idea at all - 428 Linux doesn't read the old inode in when it's allocating a 429 new one. I will set at least i_size & i_blocks the zero. 430 */ 431 ip->i_mode = 0; 432 ip->i_size = 0; 433 ip->i_blocks = 0; 434 ip->i_flags = 0; 435 /* now we want to make sure that the block pointers are zeroed out */ 436 for(i = 0; i < EXT2_NDIR_BLOCKS; i++) 437 ip->i_db[i] = 0; 438 439 /* 440 * Set up a new generation number for this inode. 441 * XXX check if this makes sense in ext2 442 */ 443#if !defined(__FreeBSD__) 444 get_time(&time); 445#endif 446 if (++nextgennumber < (u_long)time.tv_sec) 447 nextgennumber = time.tv_sec; 448 ip->i_gen = nextgennumber; 449/* 450printf("ext2_valloc: allocated inode %d\n", ino); 451*/ 452 return (0); 453noinodes: 454 ext2_fserr(fs, ap->a_cred->cr_uid, "out of inodes"); 455 uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt); 456 return (ENOSPC); 457} 458 459/* 460 * Select the desired position for the next block in a file. 461 * 462 * we try to mimic what Remy does in inode_getblk/block_getblk 463 * 464 * we note: blocknr == 0 means that we're about to allocate either 465 * a direct block or a pointer block at the first level of indirection 466 * (In other words, stuff that will go in i_db[] or i_ib[]) 467 * 468 * blocknr != 0 means that we're allocating a block that is none 469 * of the above. Then, blocknr tells us the number of the block 470 * that will hold the pointer 471 */ 472daddr_t 473ext2_blkpref(ip, lbn, indx, bap, blocknr) 474 struct inode *ip; 475 daddr_t lbn; 476 int indx; 477 daddr_t *bap; 478 daddr_t blocknr; 479{ 480 int tmp; 481 482 /* if the next block is actually what we thought it is, 483 then set the goal to what we thought it should be 484 */ 485 if(ip->i_next_alloc_block == lbn) 486 return ip->i_next_alloc_goal; 487 488 /* now check whether we were provided with an array that basically 489 tells us previous blocks to which we want to stay closeby 490 */ 491 if(bap) 492 for (tmp = indx - 1; tmp >= 0; tmp--) 493 if (bap[tmp]) 494 return bap[tmp]; 495 496 /* else let's fall back to the blocknr, or, if there is none, 497 follow the rule that a block should be allocated near it's inode 498 */ 499 return blocknr ? blocknr : 500 (daddr_t)(ip->i_block_group * 501 EXT2_BLOCKS_PER_GROUP(ip->i_e2fs)) + 502 ip->i_e2fs->s_es->s_first_data_block; 503} 504 505/* 506 * Free a block or fragment. 507 * 508 * pass on to the Linux code 509 */ 510void 511ext2_blkfree(ip, bno, size) 512 register struct inode *ip; 513 daddr_t bno; 514 long size; 515{ 516 register struct ext2_sb_info *fs; 517 518 fs = ip->i_e2fs; 519 /* 520 * call Linux code with mount *, block number, count 521 */ 522 ext2_free_blocks(ITOV(ip)->v_mount, bno, size / fs->s_frag_size); 523} 524 525/* 526 * Free an inode. 527 * 528 * the maintenance of the actual bitmaps is again up to the linux code 529 */ 530int 531ext2_vfree(ap) 532 struct vop_vfree_args /* { 533 struct vnode *a_pvp; 534 ino_t a_ino; 535 int a_mode; 536 } */ *ap; 537{ 538 register struct ext2_sb_info *fs; 539 register struct inode *pip; 540 ino_t ino = ap->a_ino; 541 int mode; 542 543 pip = VTOI(ap->a_pvp); 544 fs = pip->i_e2fs; 545 if ((u_int)ino >= fs->s_inodes_per_group * fs->s_groups_count) 546 panic("ifree: range: dev = 0x%x, ino = %d, fs = %s\n", 547 pip->i_dev, ino, fs->fs_fsmnt); 548 549/* ext2_debug("ext2_vfree (%d, %d) called\n", pip->i_number, ap->a_mode); 550 */ 551 ext2_discard_prealloc(pip); 552 553 /* we need to make sure that ext2_free_inode can adjust the 554 used_dir_counts in the group summary information - I'd 555 really like to know what the rationale behind this 556 'set i_mode to zero to denote an unused inode' is 557 */ 558 mode = pip->i_mode; 559 pip->i_mode = ap->a_mode; 560 ext2_free_inode(pip); 561 pip->i_mode = mode; 562 return (0); 563} 564 565/* 566 * Fserr prints the name of a file system with an error diagnostic. 567 * 568 * The form of the error message is: 569 * fs: error message 570 */ 571static void 572ext2_fserr(fs, uid, cp) 573 struct ext2_sb_info *fs; 574 u_int uid; 575 char *cp; 576{ 577 578 log(LOG_ERR, "uid %d on %s: %s\n", uid, fs->fs_fsmnt, cp); 579} 580