lfs_segment.c revision 1.32.2.2
1 1.32.2.2 perseant /* $NetBSD: lfs_segment.c,v 1.32.2.2 1999/11/06 20:33:06 perseant Exp $ */ 2 1.32.2.2 perseant 3 1.32.2.2 perseant /*- 4 1.32.2.2 perseant * Copyright (c) 1999 The NetBSD Foundation, Inc. 5 1.32.2.2 perseant * All rights reserved. 6 1.32.2.2 perseant * 7 1.32.2.2 perseant * This code is derived from software contributed to The NetBSD Foundation 8 1.32.2.2 perseant * by Konrad E. Schroder <perseant (at) hhhh.org>. 9 1.32.2.2 perseant * 10 1.32.2.2 perseant * Redistribution and use in source and binary forms, with or without 11 1.32.2.2 perseant * modification, are permitted provided that the following conditions 12 1.32.2.2 perseant * are met: 13 1.32.2.2 perseant * 1. Redistributions of source code must retain the above copyright 14 1.32.2.2 perseant * notice, this list of conditions and the following disclaimer. 15 1.32.2.2 perseant * 2. Redistributions in binary form must reproduce the above copyright 16 1.32.2.2 perseant * notice, this list of conditions and the following disclaimer in the 17 1.32.2.2 perseant * documentation and/or other materials provided with the distribution. 18 1.32.2.2 perseant * 3. All advertising materials mentioning features or use of this software 19 1.32.2.2 perseant * must display the following acknowledgement: 20 1.32.2.2 perseant * This product includes software developed by the NetBSD 21 1.32.2.2 perseant * Foundation, Inc. and its contributors. 22 1.32.2.2 perseant * 4. Neither the name of The NetBSD Foundation nor the names of its 23 1.32.2.2 perseant * contributors may be used to endorse or promote products derived 24 1.32.2.2 perseant * from this software without specific prior written permission. 25 1.32.2.2 perseant * 26 1.32.2.2 perseant * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 1.32.2.2 perseant * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 1.32.2.2 perseant * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 1.32.2.2 perseant * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 1.32.2.2 perseant * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 1.32.2.2 perseant * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 1.32.2.2 perseant * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 1.32.2.2 perseant * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 1.32.2.2 perseant * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 1.32.2.2 perseant * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 1.32.2.2 perseant * POSSIBILITY OF SUCH DAMAGE. 37 1.32.2.2 perseant */ 38 1.32.2.2 perseant /* 39 1.32.2.2 perseant * Copyright (c) 1991, 1993 40 1.32.2.2 perseant * The Regents of the University of California. All rights reserved. 41 1.32.2.2 perseant * 42 1.32.2.2 perseant * Redistribution and use in source and binary forms, with or without 43 1.32.2.2 perseant * modification, are permitted provided that the following conditions 44 1.32.2.2 perseant * are met: 45 1.32.2.2 perseant * 1. Redistributions of source code must retain the above copyright 46 1.32.2.2 perseant * notice, this list of conditions and the following disclaimer. 47 1.32.2.2 perseant * 2. Redistributions in binary form must reproduce the above copyright 48 1.32.2.2 perseant * notice, this list of conditions and the following disclaimer in the 49 1.32.2.2 perseant * documentation and/or other materials provided with the distribution. 50 1.32.2.2 perseant * 3. All advertising materials mentioning features or use of this software 51 1.32.2.2 perseant * must display the following acknowledgement: 52 1.32.2.2 perseant * This product includes software developed by the University of 53 1.32.2.2 perseant * California, Berkeley and its contributors. 54 1.32.2.2 perseant * 4. Neither the name of the University nor the names of its contributors 55 1.32.2.2 perseant * may be used to endorse or promote products derived from this software 56 1.32.2.2 perseant * without specific prior written permission. 57 1.32.2.2 perseant * 58 1.32.2.2 perseant * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 59 1.32.2.2 perseant * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 60 1.32.2.2 perseant * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 61 1.32.2.2 perseant * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 62 1.32.2.2 perseant * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 63 1.32.2.2 perseant * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 64 1.32.2.2 perseant * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 65 1.32.2.2 perseant * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 66 1.32.2.2 perseant * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 67 1.32.2.2 perseant * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 68 1.32.2.2 perseant * SUCH DAMAGE. 69 1.32.2.2 perseant * 70 1.32.2.2 perseant * @(#)lfs_segment.c 8.10 (Berkeley) 6/10/95 71 1.32.2.2 perseant */ 72 1.32.2.2 perseant 73 1.32.2.2 perseant #define ivndebug(vp,str) printf("ino %d: %s\n",VTOI(vp)->i_number,(str)) 74 1.32.2.2 perseant 75 1.32.2.2 perseant #include "opt_ddb.h" 76 1.32.2.2 perseant #include <sys/param.h> 77 1.32.2.2 perseant #include <sys/systm.h> 78 1.32.2.2 perseant #include <sys/namei.h> 79 1.32.2.2 perseant #include <sys/kernel.h> 80 1.32.2.2 perseant #include <sys/resourcevar.h> 81 1.32.2.2 perseant #include <sys/file.h> 82 1.32.2.2 perseant #include <sys/stat.h> 83 1.32.2.2 perseant #include <sys/buf.h> 84 1.32.2.2 perseant #include <sys/proc.h> 85 1.32.2.2 perseant #include <sys/conf.h> 86 1.32.2.2 perseant #include <sys/vnode.h> 87 1.32.2.2 perseant #include <sys/malloc.h> 88 1.32.2.2 perseant #include <sys/mount.h> 89 1.32.2.2 perseant 90 1.32.2.2 perseant #include <miscfs/specfs/specdev.h> 91 1.32.2.2 perseant #include <miscfs/fifofs/fifo.h> 92 1.32.2.2 perseant 93 1.32.2.2 perseant #include <ufs/ufs/quota.h> 94 1.32.2.2 perseant #include <ufs/ufs/inode.h> 95 1.32.2.2 perseant #include <ufs/ufs/dir.h> 96 1.32.2.2 perseant #include <ufs/ufs/ufsmount.h> 97 1.32.2.2 perseant #include <ufs/ufs/ufs_extern.h> 98 1.32.2.2 perseant 99 1.32.2.2 perseant #include <ufs/lfs/lfs.h> 100 1.32.2.2 perseant #include <ufs/lfs/lfs_extern.h> 101 1.32.2.2 perseant 102 1.32.2.2 perseant extern int count_lock_queue __P((void)); 103 1.32.2.2 perseant extern struct simplelock vnode_free_list_slock; /* XXX */ 104 1.32.2.2 perseant extern TAILQ_HEAD(freelst, vnode) vnode_free_list; /* XXX */ 105 1.32.2.2 perseant 106 1.32.2.2 perseant /* 107 1.32.2.2 perseant * Determine if it's OK to start a partial in this segment, or if we need 108 1.32.2.2 perseant * to go on to a new segment. 109 1.32.2.2 perseant */ 110 1.32.2.2 perseant #define LFS_PARTIAL_FITS(fs) \ 111 1.32.2.2 perseant ((fs)->lfs_dbpseg - ((fs)->lfs_offset - (fs)->lfs_curseg) > \ 112 1.32.2.2 perseant 1 << (fs)->lfs_fsbtodb) 113 1.32.2.2 perseant 114 1.32.2.2 perseant void lfs_callback __P((struct buf *)); 115 1.32.2.2 perseant int lfs_gather __P((struct lfs *, struct segment *, 116 1.32.2.2 perseant struct vnode *, int (*) __P((struct lfs *, struct buf *)))); 117 1.32.2.2 perseant int lfs_gatherblock __P((struct segment *, struct buf *, int *)); 118 1.32.2.2 perseant void lfs_iset __P((struct inode *, ufs_daddr_t, time_t)); 119 1.32.2.2 perseant int lfs_match_fake __P((struct lfs *, struct buf *)); 120 1.32.2.2 perseant int lfs_match_data __P((struct lfs *, struct buf *)); 121 1.32.2.2 perseant int lfs_match_dindir __P((struct lfs *, struct buf *)); 122 1.32.2.2 perseant int lfs_match_indir __P((struct lfs *, struct buf *)); 123 1.32.2.2 perseant int lfs_match_tindir __P((struct lfs *, struct buf *)); 124 1.32.2.2 perseant void lfs_newseg __P((struct lfs *)); 125 1.32.2.2 perseant void lfs_shellsort __P((struct buf **, ufs_daddr_t *, register int)); 126 1.32.2.2 perseant void lfs_supercallback __P((struct buf *)); 127 1.32.2.2 perseant void lfs_updatemeta __P((struct segment *)); 128 1.32.2.2 perseant int lfs_vref __P((struct vnode *)); 129 1.32.2.2 perseant void lfs_vunref __P((struct vnode *)); 130 1.32.2.2 perseant void lfs_writefile __P((struct lfs *, struct segment *, struct vnode *)); 131 1.32.2.2 perseant int lfs_writeinode __P((struct lfs *, struct segment *, struct inode *)); 132 1.32.2.2 perseant int lfs_writeseg __P((struct lfs *, struct segment *)); 133 1.32.2.2 perseant void lfs_writesuper __P((struct lfs *, daddr_t)); 134 1.32.2.2 perseant int lfs_writevnodes __P((struct lfs *fs, struct mount *mp, 135 1.32.2.2 perseant struct segment *sp, int dirops)); 136 1.32.2.2 perseant 137 1.32.2.2 perseant int lfs_allclean_wakeup; /* Cleaner wakeup address. */ 138 1.32.2.2 perseant int lfs_writeindir = 1; /* whether to flush indir on non-ckp */ 139 1.32.2.2 perseant int lfs_clean_vnhead = 0; /* Allow freeing to head of vn list */ 140 1.32.2.2 perseant int lfs_dirvcount = 0; /* # active dirops */ 141 1.32.2.2 perseant 142 1.32.2.2 perseant /* Statistics Counters */ 143 1.32.2.2 perseant int lfs_dostats = 1; 144 1.32.2.2 perseant struct lfs_stats lfs_stats; 145 1.32.2.2 perseant 146 1.32.2.2 perseant /* op values to lfs_writevnodes */ 147 1.32.2.2 perseant #define VN_REG 0 148 1.32.2.2 perseant #define VN_DIROP 1 149 1.32.2.2 perseant #define VN_EMPTY 2 150 1.32.2.2 perseant #define VN_CLEAN 3 151 1.32.2.2 perseant 152 1.32.2.2 perseant #define LFS_MAX_ACTIVE 10 153 1.32.2.2 perseant 154 1.32.2.2 perseant /* 155 1.32.2.2 perseant * XXX KS - Set modification time on the Ifile, so the cleaner can 156 1.32.2.2 perseant * read the fs mod time off of it. We don't set IN_UPDATE here, 157 1.32.2.2 perseant * since we don't really need this to be flushed to disk (and in any 158 1.32.2.2 perseant * case that wouldn't happen to the Ifile until we checkpoint). 159 1.32.2.2 perseant */ 160 1.32.2.2 perseant void 161 1.32.2.2 perseant lfs_imtime(fs) 162 1.32.2.2 perseant struct lfs *fs; 163 1.32.2.2 perseant { 164 1.32.2.2 perseant struct timespec ts; 165 1.32.2.2 perseant struct inode *ip; 166 1.32.2.2 perseant 167 1.32.2.2 perseant TIMEVAL_TO_TIMESPEC(&time, &ts); 168 1.32.2.2 perseant ip = VTOI(fs->lfs_ivnode); 169 1.32.2.2 perseant ip->i_ffs_mtime = ts.tv_sec; 170 1.32.2.2 perseant ip->i_ffs_mtimensec = ts.tv_nsec; 171 1.32.2.2 perseant } 172 1.32.2.2 perseant 173 1.32.2.2 perseant /* 174 1.32.2.2 perseant * Ifile and meta data blocks are not marked busy, so segment writes MUST be 175 1.32.2.2 perseant * single threaded. Currently, there are two paths into lfs_segwrite, sync() 176 1.32.2.2 perseant * and getnewbuf(). They both mark the file system busy. Lfs_vflush() 177 1.32.2.2 perseant * explicitly marks the file system busy. So lfs_segwrite is safe. I think. 178 1.32.2.2 perseant */ 179 1.32.2.2 perseant 180 1.32.2.2 perseant #define SET_FLUSHING(fs,vp) (fs)->lfs_flushvp = (vp) 181 1.32.2.2 perseant #define IS_FLUSHING(fs,vp) ((fs)->lfs_flushvp == (vp)) 182 1.32.2.2 perseant #define CLR_FLUSHING(fs,vp) (fs)->lfs_flushvp = NULL 183 1.32.2.2 perseant 184 1.32.2.2 perseant int 185 1.32.2.2 perseant lfs_vflush(vp) 186 1.32.2.2 perseant struct vnode *vp; 187 1.32.2.2 perseant { 188 1.32.2.2 perseant struct inode *ip; 189 1.32.2.2 perseant struct lfs *fs; 190 1.32.2.2 perseant struct segment *sp; 191 1.32.2.2 perseant struct buf *bp, *nbp; 192 1.32.2.2 perseant int error, s; 193 1.32.2.2 perseant 194 1.32.2.2 perseant ip = VTOI(vp); 195 1.32.2.2 perseant fs = VFSTOUFS(vp->v_mount)->um_lfs; 196 1.32.2.2 perseant 197 1.32.2.2 perseant if(ip->i_flag & IN_CLEANING) { 198 1.32.2.2 perseant #ifdef DEBUG_LFS 199 1.32.2.2 perseant ivndebug(vp,"vflush/in_cleaning"); 200 1.32.2.2 perseant #endif 201 1.32.2.2 perseant ip->i_flag &= ~IN_CLEANING; 202 1.32.2.2 perseant if(ip->i_flag & IN_MODIFIED) { 203 1.32.2.2 perseant fs->lfs_uinodes--; 204 1.32.2.2 perseant } else 205 1.32.2.2 perseant ip->i_flag |= IN_MODIFIED; 206 1.32.2.2 perseant } 207 1.32.2.2 perseant 208 1.32.2.2 perseant /* If the node is being written, wait until that is done */ 209 1.32.2.2 perseant if(WRITEINPROG(vp)) { 210 1.32.2.2 perseant #ifdef DEBUG_LFS 211 1.32.2.2 perseant ivndebug(vp,"vflush/writeinprog"); 212 1.32.2.2 perseant #endif 213 1.32.2.2 perseant tsleep(vp, PRIBIO+1, "lfs_vw", 0); 214 1.32.2.2 perseant } 215 1.32.2.2 perseant 216 1.32.2.2 perseant /* Protect against VXLOCK deadlock in vinvalbuf() */ 217 1.32.2.2 perseant lfs_seglock(fs, SEGM_SYNC); 218 1.32.2.2 perseant 219 1.32.2.2 perseant /* If we're supposed to flush a freed inode, just toss it */ 220 1.32.2.2 perseant /* XXX - seglock, so these buffers can't be gathered, right? */ 221 1.32.2.2 perseant if(ip->i_ffs_mode == 0) { 222 1.32.2.2 perseant printf("lfs_vflush: ino %d is freed, not flushing\n", 223 1.32.2.2 perseant ip->i_number); 224 1.32.2.2 perseant s = splbio(); 225 1.32.2.2 perseant for(bp=vp->v_dirtyblkhd.lh_first; bp; bp=nbp) { 226 1.32.2.2 perseant nbp = bp->b_vnbufs.le_next; 227 1.32.2.2 perseant /* Copied from lfs_writeseg */ 228 1.32.2.2 perseant if (bp->b_flags & B_CALL) { 229 1.32.2.2 perseant /* if B_CALL, it was created with newbuf */ 230 1.32.2.2 perseant lfs_freebuf(bp); 231 1.32.2.2 perseant } else { 232 1.32.2.2 perseant bremfree(bp); 233 1.32.2.2 perseant bp->b_flags &= ~(B_ERROR | B_READ | B_DELWRI | 234 1.32.2.2 perseant B_LOCKED | B_GATHERED); 235 1.32.2.2 perseant bp->b_flags |= B_DONE; 236 1.32.2.2 perseant reassignbuf(bp, vp); 237 1.32.2.2 perseant brelse(bp); 238 1.32.2.2 perseant } 239 1.32.2.2 perseant } 240 1.32.2.2 perseant splx(s); 241 1.32.2.2 perseant if(ip->i_flag & IN_CLEANING) 242 1.32.2.2 perseant fs->lfs_uinodes--; 243 1.32.2.2 perseant if(ip->i_flag & IN_MODIFIED) 244 1.32.2.2 perseant fs->lfs_uinodes--; 245 1.32.2.2 perseant ip->i_flag &= ~(IN_MODIFIED|IN_UPDATE|IN_ACCESS|IN_CHANGE|IN_CLEANING); 246 1.32.2.2 perseant printf("lfs_vflush: done not flushing ino %d\n", 247 1.32.2.2 perseant ip->i_number); 248 1.32.2.2 perseant lfs_segunlock(fs); 249 1.32.2.2 perseant return 0; 250 1.32.2.2 perseant } 251 1.32.2.2 perseant 252 1.32.2.2 perseant SET_FLUSHING(fs,vp); 253 1.32.2.2 perseant if (fs->lfs_nactive > LFS_MAX_ACTIVE) { 254 1.32.2.2 perseant error = lfs_segwrite(vp->v_mount, SEGM_SYNC|SEGM_CKP); 255 1.32.2.2 perseant CLR_FLUSHING(fs,vp); 256 1.32.2.2 perseant lfs_segunlock(fs); 257 1.32.2.2 perseant return error; 258 1.32.2.2 perseant } 259 1.32.2.2 perseant sp = fs->lfs_sp; 260 1.32.2.2 perseant 261 1.32.2.2 perseant if (vp->v_dirtyblkhd.lh_first == NULL) { 262 1.32.2.2 perseant lfs_writevnodes(fs, vp->v_mount, sp, VN_EMPTY); 263 1.32.2.2 perseant } else if((ip->i_flag & IN_CLEANING) && (fs->lfs_sp->seg_flags & SEGM_CLEAN)) { 264 1.32.2.2 perseant #ifdef DEBUG_LFS 265 1.32.2.2 perseant ivndebug(vp,"vflush/clean"); 266 1.32.2.2 perseant #endif 267 1.32.2.2 perseant lfs_writevnodes(fs, vp->v_mount, sp, VN_CLEAN); 268 1.32.2.2 perseant } 269 1.32.2.2 perseant else if(lfs_dostats) { 270 1.32.2.2 perseant if(vp->v_dirtyblkhd.lh_first || (VTOI(vp)->i_flag & (IN_MODIFIED|IN_UPDATE|IN_ACCESS|IN_CHANGE|IN_CLEANING))) 271 1.32.2.2 perseant ++lfs_stats.vflush_invoked; 272 1.32.2.2 perseant #ifdef DEBUG_LFS 273 1.32.2.2 perseant ivndebug(vp,"vflush"); 274 1.32.2.2 perseant #endif 275 1.32.2.2 perseant } 276 1.32.2.2 perseant 277 1.32.2.2 perseant #ifdef DIAGNOSTIC 278 1.32.2.2 perseant /* XXX KS This actually can happen right now, though it shouldn't(?) */ 279 1.32.2.2 perseant if(vp->v_flag & VDIROP) { 280 1.32.2.2 perseant printf("lfs_vflush: flushing VDIROP, this shouldn\'t be\n"); 281 1.32.2.2 perseant /* panic("VDIROP being flushed...this can\'t happen"); */ 282 1.32.2.2 perseant } 283 1.32.2.2 perseant if(vp->v_usecount<0) { 284 1.32.2.2 perseant printf("usecount=%ld\n",vp->v_usecount); 285 1.32.2.2 perseant panic("lfs_vflush: usecount<0"); 286 1.32.2.2 perseant } 287 1.32.2.2 perseant #endif 288 1.32.2.2 perseant 289 1.32.2.2 perseant do { 290 1.32.2.2 perseant do { 291 1.32.2.2 perseant if (vp->v_dirtyblkhd.lh_first != NULL) 292 1.32.2.2 perseant lfs_writefile(fs, sp, vp); 293 1.32.2.2 perseant } while (lfs_writeinode(fs, sp, ip)); 294 1.32.2.2 perseant } while (lfs_writeseg(fs, sp) && ip->i_number == LFS_IFILE_INUM); 295 1.32.2.2 perseant 296 1.32.2.2 perseant if(lfs_dostats) { 297 1.32.2.2 perseant ++lfs_stats.nwrites; 298 1.32.2.2 perseant if (sp->seg_flags & SEGM_SYNC) 299 1.32.2.2 perseant ++lfs_stats.nsync_writes; 300 1.32.2.2 perseant if (sp->seg_flags & SEGM_CKP) 301 1.32.2.2 perseant ++lfs_stats.ncheckpoints; 302 1.32.2.2 perseant } 303 1.32.2.2 perseant lfs_segunlock(fs); 304 1.32.2.2 perseant 305 1.32.2.2 perseant CLR_FLUSHING(fs,vp); 306 1.32.2.2 perseant return (0); 307 1.32.2.2 perseant } 308 1.32.2.2 perseant 309 1.32.2.2 perseant #ifdef DEBUG_LFS_VERBOSE 310 1.32.2.2 perseant # define vndebug(vp,str) if(VTOI(vp)->i_flag & IN_CLEANING) printf("not writing ino %d because %s (op %d)\n",VTOI(vp)->i_number,(str),op) 311 1.32.2.2 perseant #else 312 1.32.2.2 perseant # define vndebug(vp,str) 313 1.32.2.2 perseant #endif 314 1.32.2.2 perseant 315 1.32.2.2 perseant int 316 1.32.2.2 perseant lfs_writevnodes(fs, mp, sp, op) 317 1.32.2.2 perseant struct lfs *fs; 318 1.32.2.2 perseant struct mount *mp; 319 1.32.2.2 perseant struct segment *sp; 320 1.32.2.2 perseant int op; 321 1.32.2.2 perseant { 322 1.32.2.2 perseant struct inode *ip; 323 1.32.2.2 perseant struct vnode *vp; 324 1.32.2.2 perseant int inodes_written=0, only_cleaning; 325 1.32.2.2 perseant 326 1.32.2.2 perseant #ifndef LFS_NO_BACKVP_HACK 327 1.32.2.2 perseant /* BEGIN HACK */ 328 1.32.2.2 perseant #define VN_OFFSET (((caddr_t)&vp->v_mntvnodes.le_next) - (caddr_t)vp) 329 1.32.2.2 perseant #define BACK_VP(VP) ((struct vnode *)(((caddr_t)VP->v_mntvnodes.le_prev) - VN_OFFSET)) 330 1.32.2.2 perseant #define BEG_OF_VLIST ((struct vnode *)(((caddr_t)&mp->mnt_vnodelist.lh_first) - VN_OFFSET)) 331 1.32.2.2 perseant 332 1.32.2.2 perseant /* Find last vnode. */ 333 1.32.2.2 perseant loop: for (vp = mp->mnt_vnodelist.lh_first; 334 1.32.2.2 perseant vp && vp->v_mntvnodes.le_next != NULL; 335 1.32.2.2 perseant vp = vp->v_mntvnodes.le_next); 336 1.32.2.2 perseant for (; vp && vp != BEG_OF_VLIST; vp = BACK_VP(vp)) { 337 1.32.2.2 perseant #else 338 1.32.2.2 perseant loop: 339 1.32.2.2 perseant for (vp = mp->mnt_vnodelist.lh_first; 340 1.32.2.2 perseant vp != NULL; 341 1.32.2.2 perseant vp = vp->v_mntvnodes.le_next) { 342 1.32.2.2 perseant #endif 343 1.32.2.2 perseant /* 344 1.32.2.2 perseant * If the vnode that we are about to sync is no longer 345 1.32.2.2 perseant * associated with this mount point, start over. 346 1.32.2.2 perseant */ 347 1.32.2.2 perseant if (vp->v_mount != mp) 348 1.32.2.2 perseant goto loop; 349 1.32.2.2 perseant 350 1.32.2.2 perseant ip = VTOI(vp); 351 1.32.2.2 perseant if ((op == VN_DIROP && !(vp->v_flag & VDIROP)) || 352 1.32.2.2 perseant (op != VN_DIROP && op != VN_CLEAN && (vp->v_flag & VDIROP))) { 353 1.32.2.2 perseant vndebug(vp,"dirop"); 354 1.32.2.2 perseant continue; 355 1.32.2.2 perseant } 356 1.32.2.2 perseant 357 1.32.2.2 perseant if (op == VN_EMPTY && vp->v_dirtyblkhd.lh_first) { 358 1.32.2.2 perseant vndebug(vp,"empty"); 359 1.32.2.2 perseant continue; 360 1.32.2.2 perseant } 361 1.32.2.2 perseant 362 1.32.2.2 perseant if (vp->v_type == VNON) { 363 1.32.2.2 perseant continue; 364 1.32.2.2 perseant } 365 1.32.2.2 perseant 366 1.32.2.2 perseant if(op == VN_CLEAN && ip->i_number != LFS_IFILE_INUM 367 1.32.2.2 perseant && !(ip->i_flag & IN_CLEANING)) { 368 1.32.2.2 perseant vndebug(vp,"cleaning"); 369 1.32.2.2 perseant continue; 370 1.32.2.2 perseant } 371 1.32.2.2 perseant 372 1.32.2.2 perseant if (lfs_vref(vp)) { 373 1.32.2.2 perseant vndebug(vp,"vref"); 374 1.32.2.2 perseant continue; 375 1.32.2.2 perseant } 376 1.32.2.2 perseant 377 1.32.2.2 perseant #if 0 /* XXX KS - if we skip the ifile, things could go badly for us. */ 378 1.32.2.2 perseant if(WRITEINPROG(vp)) { 379 1.32.2.2 perseant lfs_vunref(vp); 380 1.32.2.2 perseant #ifdef DEBUG_LFS 381 1.32.2.2 perseant ivndebug(vp,"writevnodes/writeinprog"); 382 1.32.2.2 perseant #endif 383 1.32.2.2 perseant continue; 384 1.32.2.2 perseant } 385 1.32.2.2 perseant #endif 386 1.32.2.2 perseant only_cleaning = 0; 387 1.32.2.2 perseant /* 388 1.32.2.2 perseant * Write the inode/file if dirty and it's not the 389 1.32.2.2 perseant * the IFILE. 390 1.32.2.2 perseant */ 391 1.32.2.2 perseant if ((ip->i_flag & 392 1.32.2.2 perseant (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE | IN_CLEANING) || 393 1.32.2.2 perseant vp->v_dirtyblkhd.lh_first != NULL)) 394 1.32.2.2 perseant { 395 1.32.2.2 perseant only_cleaning = ((ip->i_flag & (IN_ACCESS|IN_CHANGE|IN_MODIFIED|IN_UPDATE|IN_CLEANING))==IN_CLEANING); 396 1.32.2.2 perseant 397 1.32.2.2 perseant if(ip->i_number != LFS_IFILE_INUM 398 1.32.2.2 perseant && vp->v_dirtyblkhd.lh_first != NULL) 399 1.32.2.2 perseant { 400 1.32.2.2 perseant lfs_writefile(fs, sp, vp); 401 1.32.2.2 perseant } 402 1.32.2.2 perseant if(vp->v_dirtyblkhd.lh_first != NULL) { 403 1.32.2.2 perseant if(WRITEINPROG(vp)) { 404 1.32.2.2 perseant #ifdef DEBUG_LFS 405 1.32.2.2 perseant ivndebug(vp,"writevnodes/write2"); 406 1.32.2.2 perseant #endif 407 1.32.2.2 perseant } else if(!(ip->i_flag & (IN_ACCESS|IN_CHANGE|IN_MODIFIED|IN_UPDATE|IN_CLEANING))) { 408 1.32.2.2 perseant #ifdef DEBUG_LFS 409 1.32.2.2 perseant printf("<%d>",ip->i_number); 410 1.32.2.2 perseant #endif 411 1.32.2.2 perseant ip->i_flag |= IN_MODIFIED; 412 1.32.2.2 perseant ++fs->lfs_uinodes; 413 1.32.2.2 perseant } 414 1.32.2.2 perseant } 415 1.32.2.2 perseant (void) lfs_writeinode(fs, sp, ip); 416 1.32.2.2 perseant inodes_written++; 417 1.32.2.2 perseant } 418 1.32.2.2 perseant 419 1.32.2.2 perseant if(vp->v_flag & VDIROP) { 420 1.32.2.2 perseant --lfs_dirvcount; 421 1.32.2.2 perseant vp->v_flag &= ~VDIROP; 422 1.32.2.2 perseant wakeup(&lfs_dirvcount); 423 1.32.2.2 perseant lfs_vunref(vp); 424 1.32.2.2 perseant } 425 1.32.2.2 perseant 426 1.32.2.2 perseant if(lfs_clean_vnhead && only_cleaning) 427 1.32.2.2 perseant lfs_vunref_head(vp); 428 1.32.2.2 perseant else 429 1.32.2.2 perseant lfs_vunref(vp); 430 1.32.2.2 perseant } 431 1.32.2.2 perseant return inodes_written; 432 1.32.2.2 perseant } 433 1.32.2.2 perseant 434 1.32.2.2 perseant int 435 1.32.2.2 perseant lfs_segwrite(mp, flags) 436 1.32.2.2 perseant struct mount *mp; 437 1.32.2.2 perseant int flags; /* Do a checkpoint. */ 438 1.32.2.2 perseant { 439 1.32.2.2 perseant struct buf *bp; 440 1.32.2.2 perseant struct inode *ip; 441 1.32.2.2 perseant struct lfs *fs; 442 1.32.2.2 perseant struct segment *sp; 443 1.32.2.2 perseant struct vnode *vp; 444 1.32.2.2 perseant SEGUSE *segusep; 445 1.32.2.2 perseant ufs_daddr_t ibno; 446 1.32.2.2 perseant int do_ckp, error, i; 447 1.32.2.2 perseant int writer_set = 0; 448 1.32.2.2 perseant int need_unlock = 0; 449 1.32.2.2 perseant 450 1.32.2.2 perseant fs = VFSTOUFS(mp)->um_lfs; 451 1.32.2.2 perseant 452 1.32.2.2 perseant lfs_imtime(fs); 453 1.32.2.2 perseant 454 1.32.2.2 perseant /* 455 1.32.2.2 perseant * If we are not the cleaner, and we have fewer than MIN_FREE_SEGS 456 1.32.2.2 perseant * clean segments, wait until cleaner writes. 457 1.32.2.2 perseant */ 458 1.32.2.2 perseant if(!(flags & SEGM_CLEAN) 459 1.32.2.2 perseant && (!fs->lfs_seglock || !(fs->lfs_sp->seg_flags & SEGM_CLEAN))) 460 1.32.2.2 perseant { 461 1.32.2.2 perseant do { 462 1.32.2.2 perseant if (fs->lfs_nclean <= MIN_FREE_SEGS 463 1.32.2.2 perseant || fs->lfs_avail <= 0) 464 1.32.2.2 perseant { 465 1.32.2.2 perseant wakeup(&lfs_allclean_wakeup); 466 1.32.2.2 perseant wakeup(&fs->lfs_nextseg); 467 1.32.2.2 perseant error = tsleep(&fs->lfs_avail, PRIBIO + 1, 468 1.32.2.2 perseant "lfs_avail", 0); 469 1.32.2.2 perseant if (error) { 470 1.32.2.2 perseant return (error); 471 1.32.2.2 perseant } 472 1.32.2.2 perseant } 473 1.32.2.2 perseant } while (fs->lfs_nclean <= MIN_FREE_SEGS || fs->lfs_avail <= 0); 474 1.32.2.2 perseant } 475 1.32.2.2 perseant 476 1.32.2.2 perseant /* 477 1.32.2.2 perseant * Allocate a segment structure and enough space to hold pointers to 478 1.32.2.2 perseant * the maximum possible number of buffers which can be described in a 479 1.32.2.2 perseant * single summary block. 480 1.32.2.2 perseant */ 481 1.32.2.2 perseant do_ckp = (flags & SEGM_CKP) || fs->lfs_nactive > LFS_MAX_ACTIVE; 482 1.32.2.2 perseant lfs_seglock(fs, flags | (do_ckp ? SEGM_CKP : 0)); 483 1.32.2.2 perseant sp = fs->lfs_sp; 484 1.32.2.2 perseant 485 1.32.2.2 perseant /* 486 1.32.2.2 perseant * If lfs_flushvp is non-NULL, we are called from lfs_vflush, 487 1.32.2.2 perseant * in which case we have to flush *all* buffers off of this vnode. 488 1.32.2.2 perseant */ 489 1.32.2.2 perseant if((sp->seg_flags & SEGM_CLEAN) && !(fs->lfs_flushvp)) 490 1.32.2.2 perseant lfs_writevnodes(fs, mp, sp, VN_CLEAN); 491 1.32.2.2 perseant else { 492 1.32.2.2 perseant lfs_writevnodes(fs, mp, sp, VN_REG); 493 1.32.2.2 perseant /* 494 1.32.2.2 perseant * XXX KS - If we're cleaning, we can't wait for dirops, 495 1.32.2.2 perseant * because they might be waiting on us. The downside of this 496 1.32.2.2 perseant * is that, if we write anything besides cleaning blocks 497 1.32.2.2 perseant * while cleaning, the checkpoint is not completely 498 1.32.2.2 perseant * consistent. 499 1.32.2.2 perseant */ 500 1.32.2.2 perseant if(!(sp->seg_flags & SEGM_CLEAN)) { 501 1.32.2.2 perseant while(fs->lfs_dirops) 502 1.32.2.2 perseant if((error = tsleep(&fs->lfs_writer, PRIBIO + 1, 503 1.32.2.2 perseant "lfs writer", 0))) 504 1.32.2.2 perseant { 505 1.32.2.2 perseant free(sp->bpp, M_SEGMENT); 506 1.32.2.2 perseant free(sp, M_SEGMENT); 507 1.32.2.2 perseant return (error); 508 1.32.2.2 perseant } 509 1.32.2.2 perseant fs->lfs_writer++; 510 1.32.2.2 perseant writer_set=1; 511 1.32.2.2 perseant lfs_writevnodes(fs, mp, sp, VN_DIROP); 512 1.32.2.2 perseant ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 513 1.32.2.2 perseant } 514 1.32.2.2 perseant } 515 1.32.2.2 perseant 516 1.32.2.2 perseant /* 517 1.32.2.2 perseant * If we are doing a checkpoint, mark everything since the 518 1.32.2.2 perseant * last checkpoint as no longer ACTIVE. 519 1.32.2.2 perseant */ 520 1.32.2.2 perseant if (do_ckp) { 521 1.32.2.2 perseant for (ibno = fs->lfs_cleansz + fs->lfs_segtabsz; 522 1.32.2.2 perseant --ibno >= fs->lfs_cleansz; ) { 523 1.32.2.2 perseant if (bread(fs->lfs_ivnode, ibno, fs->lfs_bsize, NOCRED, &bp)) 524 1.32.2.2 perseant 525 1.32.2.2 perseant panic("lfs_segwrite: ifile read"); 526 1.32.2.2 perseant segusep = (SEGUSE *)bp->b_data; 527 1.32.2.2 perseant for (i = fs->lfs_sepb; i--; segusep++) 528 1.32.2.2 perseant segusep->su_flags &= ~SEGUSE_ACTIVE; 529 1.32.2.2 perseant 530 1.32.2.2 perseant /* But the current segment is still ACTIVE */ 531 1.32.2.2 perseant if (fs->lfs_curseg/fs->lfs_sepb==(ibno-fs->lfs_cleansz)) 532 1.32.2.2 perseant ((SEGUSE *)(bp->b_data))[fs->lfs_curseg%fs->lfs_sepb].su_flags |= SEGUSE_ACTIVE; 533 1.32.2.2 perseant error = VOP_BWRITE(bp); 534 1.32.2.2 perseant } 535 1.32.2.2 perseant } 536 1.32.2.2 perseant 537 1.32.2.2 perseant if (do_ckp || fs->lfs_doifile) { 538 1.32.2.2 perseant redo: 539 1.32.2.2 perseant vp = fs->lfs_ivnode; 540 1.32.2.2 perseant /* 541 1.32.2.2 perseant * Depending on the circumstances of our calling, the ifile 542 1.32.2.2 perseant * inode might be locked. If it is, and if it is locked by 543 1.32.2.2 perseant * us, we should VREF instead of vget here. 544 1.32.2.2 perseant */ 545 1.32.2.2 perseant need_unlock = 0; 546 1.32.2.2 perseant if(VOP_ISLOCKED(vp) 547 1.32.2.2 perseant && vp->v_lock.lk_lockholder == curproc->p_pid) { 548 1.32.2.2 perseant VREF(vp); 549 1.32.2.2 perseant } else { 550 1.32.2.2 perseant while (vget(vp, LK_EXCLUSIVE)) 551 1.32.2.2 perseant continue; 552 1.32.2.2 perseant need_unlock = 1; 553 1.32.2.2 perseant } 554 1.32.2.2 perseant ip = VTOI(vp); 555 1.32.2.2 perseant if (vp->v_dirtyblkhd.lh_first != NULL) 556 1.32.2.2 perseant lfs_writefile(fs, sp, vp); 557 1.32.2.2 perseant (void)lfs_writeinode(fs, sp, ip); 558 1.32.2.2 perseant 559 1.32.2.2 perseant /* Only vput if we used vget() above. */ 560 1.32.2.2 perseant if(need_unlock) 561 1.32.2.2 perseant vput(vp); 562 1.32.2.2 perseant else 563 1.32.2.2 perseant vrele(vp); 564 1.32.2.2 perseant 565 1.32.2.2 perseant if (lfs_writeseg(fs, sp) && do_ckp) 566 1.32.2.2 perseant goto redo; 567 1.32.2.2 perseant } else { 568 1.32.2.2 perseant (void) lfs_writeseg(fs, sp); 569 1.32.2.2 perseant } 570 1.32.2.2 perseant 571 1.32.2.2 perseant /* 572 1.32.2.2 perseant * If the I/O count is non-zero, sleep until it reaches zero. 573 1.32.2.2 perseant * At the moment, the user's process hangs around so we can 574 1.32.2.2 perseant * sleep. 575 1.32.2.2 perseant */ 576 1.32.2.2 perseant fs->lfs_doifile = 0; 577 1.32.2.2 perseant if(writer_set && --fs->lfs_writer==0) 578 1.32.2.2 perseant wakeup(&fs->lfs_dirops); 579 1.32.2.2 perseant 580 1.32.2.2 perseant if(lfs_dostats) { 581 1.32.2.2 perseant ++lfs_stats.nwrites; 582 1.32.2.2 perseant if (sp->seg_flags & SEGM_SYNC) 583 1.32.2.2 perseant ++lfs_stats.nsync_writes; 584 1.32.2.2 perseant if (sp->seg_flags & SEGM_CKP) 585 1.32.2.2 perseant ++lfs_stats.ncheckpoints; 586 1.32.2.2 perseant } 587 1.32.2.2 perseant lfs_segunlock(fs); 588 1.32.2.2 perseant return (0); 589 1.32.2.2 perseant } 590 1.32.2.2 perseant 591 1.32.2.2 perseant /* 592 1.32.2.2 perseant * Write the dirty blocks associated with a vnode. 593 1.32.2.2 perseant */ 594 1.32.2.2 perseant void 595 1.32.2.2 perseant lfs_writefile(fs, sp, vp) 596 1.32.2.2 perseant struct lfs *fs; 597 1.32.2.2 perseant struct segment *sp; 598 1.32.2.2 perseant struct vnode *vp; 599 1.32.2.2 perseant { 600 1.32.2.2 perseant struct buf *bp; 601 1.32.2.2 perseant struct finfo *fip; 602 1.32.2.2 perseant IFILE *ifp; 603 1.32.2.2 perseant 604 1.32.2.2 perseant 605 1.32.2.2 perseant if (sp->seg_bytes_left < fs->lfs_bsize || 606 1.32.2.2 perseant sp->sum_bytes_left < sizeof(struct finfo)) 607 1.32.2.2 perseant (void) lfs_writeseg(fs, sp); 608 1.32.2.2 perseant 609 1.32.2.2 perseant sp->sum_bytes_left -= sizeof(struct finfo) - sizeof(ufs_daddr_t); 610 1.32.2.2 perseant ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 611 1.32.2.2 perseant 612 1.32.2.2 perseant if(vp->v_flag & VDIROP) 613 1.32.2.2 perseant ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 614 1.32.2.2 perseant 615 1.32.2.2 perseant fip = sp->fip; 616 1.32.2.2 perseant fip->fi_nblocks = 0; 617 1.32.2.2 perseant fip->fi_ino = VTOI(vp)->i_number; 618 1.32.2.2 perseant LFS_IENTRY(ifp, fs, fip->fi_ino, bp); 619 1.32.2.2 perseant fip->fi_version = ifp->if_version; 620 1.32.2.2 perseant brelse(bp); 621 1.32.2.2 perseant 622 1.32.2.2 perseant /* 623 1.32.2.2 perseant * It may not be necessary to write the meta-data blocks at this point, 624 1.32.2.2 perseant * as the roll-forward recovery code should be able to reconstruct the 625 1.32.2.2 perseant * list. 626 1.32.2.2 perseant * 627 1.32.2.2 perseant * We have to write them anyway, though, under two conditions: (1) the 628 1.32.2.2 perseant * vnode is being flushed (for reuse by vinvalbuf); or (2) we are 629 1.32.2.2 perseant * checkpointing. 630 1.32.2.2 perseant */ 631 1.32.2.2 perseant if((sp->seg_flags & SEGM_CLEAN) 632 1.32.2.2 perseant && VTOI(vp)->i_number != LFS_IFILE_INUM 633 1.32.2.2 perseant && !IS_FLUSHING(fs,vp)) 634 1.32.2.2 perseant { 635 1.32.2.2 perseant lfs_gather(fs, sp, vp, lfs_match_fake); 636 1.32.2.2 perseant } else 637 1.32.2.2 perseant lfs_gather(fs, sp, vp, lfs_match_data); 638 1.32.2.2 perseant 639 1.32.2.2 perseant if(lfs_writeindir 640 1.32.2.2 perseant || IS_FLUSHING(fs,vp) 641 1.32.2.2 perseant || (sp->seg_flags & SEGM_CKP)) 642 1.32.2.2 perseant { 643 1.32.2.2 perseant lfs_gather(fs, sp, vp, lfs_match_indir); 644 1.32.2.2 perseant lfs_gather(fs, sp, vp, lfs_match_dindir); 645 1.32.2.2 perseant /* XXX KS - when is TRIPLE not true? */ /* #ifdef TRIPLE */ 646 1.32.2.2 perseant lfs_gather(fs, sp, vp, lfs_match_tindir); 647 1.32.2.2 perseant /* #endif */ 648 1.32.2.2 perseant } 649 1.32.2.2 perseant fip = sp->fip; 650 1.32.2.2 perseant if (fip->fi_nblocks != 0) { 651 1.32.2.2 perseant sp->fip = (FINFO*)((caddr_t)fip + sizeof(struct finfo) + 652 1.32.2.2 perseant sizeof(ufs_daddr_t) * (fip->fi_nblocks-1)); 653 1.32.2.2 perseant sp->start_lbp = &sp->fip->fi_blocks[0]; 654 1.32.2.2 perseant } else { 655 1.32.2.2 perseant sp->sum_bytes_left += sizeof(FINFO) - sizeof(ufs_daddr_t); 656 1.32.2.2 perseant --((SEGSUM *)(sp->segsum))->ss_nfinfo; 657 1.32.2.2 perseant } 658 1.32.2.2 perseant } 659 1.32.2.2 perseant 660 1.32.2.2 perseant int 661 1.32.2.2 perseant lfs_writeinode(fs, sp, ip) 662 1.32.2.2 perseant struct lfs *fs; 663 1.32.2.2 perseant struct segment *sp; 664 1.32.2.2 perseant struct inode *ip; 665 1.32.2.2 perseant { 666 1.32.2.2 perseant struct buf *bp, *ibp; 667 1.32.2.2 perseant IFILE *ifp; 668 1.32.2.2 perseant SEGUSE *sup; 669 1.32.2.2 perseant ufs_daddr_t daddr; 670 1.32.2.2 perseant ino_t ino; 671 1.32.2.2 perseant int error, i, ndx; 672 1.32.2.2 perseant int redo_ifile = 0; 673 1.32.2.2 perseant struct timespec ts; 674 1.32.2.2 perseant int gotblk=0; 675 1.32.2.2 perseant 676 1.32.2.2 perseant if (!(ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE | IN_CLEANING))) 677 1.32.2.2 perseant return(0); 678 1.32.2.2 perseant 679 1.32.2.2 perseant /* Allocate a new inode block if necessary. */ 680 1.32.2.2 perseant if ((ip->i_number != LFS_IFILE_INUM || sp->idp==NULL) && sp->ibp == NULL) { 681 1.32.2.2 perseant /* Allocate a new segment if necessary. */ 682 1.32.2.2 perseant if (sp->seg_bytes_left < fs->lfs_bsize || 683 1.32.2.2 perseant sp->sum_bytes_left < sizeof(ufs_daddr_t)) 684 1.32.2.2 perseant (void) lfs_writeseg(fs, sp); 685 1.32.2.2 perseant 686 1.32.2.2 perseant /* Get next inode block. */ 687 1.32.2.2 perseant daddr = fs->lfs_offset; 688 1.32.2.2 perseant fs->lfs_offset += fsbtodb(fs, 1); 689 1.32.2.2 perseant sp->ibp = *sp->cbpp++ = 690 1.32.2.2 perseant getblk(VTOI(fs->lfs_ivnode)->i_devvp, daddr, fs->lfs_bsize, 0, 0); 691 1.32.2.2 perseant gotblk++; 692 1.32.2.2 perseant 693 1.32.2.2 perseant /* Zero out inode numbers */ 694 1.32.2.2 perseant for (i = 0; i < INOPB(fs); ++i) 695 1.32.2.2 perseant ((struct dinode *)sp->ibp->b_data)[i].di_inumber = 0; 696 1.32.2.2 perseant 697 1.32.2.2 perseant ++sp->start_bpp; 698 1.32.2.2 perseant fs->lfs_avail -= fsbtodb(fs, 1); 699 1.32.2.2 perseant /* Set remaining space counters. */ 700 1.32.2.2 perseant sp->seg_bytes_left -= fs->lfs_bsize; 701 1.32.2.2 perseant sp->sum_bytes_left -= sizeof(ufs_daddr_t); 702 1.32.2.2 perseant ndx = LFS_SUMMARY_SIZE / sizeof(ufs_daddr_t) - 703 1.32.2.2 perseant sp->ninodes / INOPB(fs) - 1; 704 1.32.2.2 perseant ((ufs_daddr_t *)(sp->segsum))[ndx] = daddr; 705 1.32.2.2 perseant } 706 1.32.2.2 perseant 707 1.32.2.2 perseant /* Update the inode times and copy the inode onto the inode page. */ 708 1.32.2.2 perseant if (ip->i_flag & (IN_CLEANING|IN_MODIFIED)) 709 1.32.2.2 perseant --fs->lfs_uinodes; 710 1.32.2.2 perseant TIMEVAL_TO_TIMESPEC(&time, &ts); 711 1.32.2.2 perseant LFS_ITIMES(ip, &ts, &ts, &ts); 712 1.32.2.2 perseant 713 1.32.2.2 perseant if(ip->i_flag & IN_CLEANING) 714 1.32.2.2 perseant ip->i_flag &= ~IN_CLEANING; 715 1.32.2.2 perseant else 716 1.32.2.2 perseant ip->i_flag &= ~(IN_ACCESS|IN_CHANGE|IN_MODIFIED|IN_UPDATE); 717 1.32.2.2 perseant 718 1.32.2.2 perseant /* 719 1.32.2.2 perseant * If this is the Ifile, and we've already written the Ifile in this 720 1.32.2.2 perseant * partial segment, just overwrite it (it's not on disk yet) and 721 1.32.2.2 perseant * continue. 722 1.32.2.2 perseant * 723 1.32.2.2 perseant * XXX we know that the bp that we get the second time around has 724 1.32.2.2 perseant * already been gathered. 725 1.32.2.2 perseant */ 726 1.32.2.2 perseant if(ip->i_number == LFS_IFILE_INUM && sp->idp) { 727 1.32.2.2 perseant *(sp->idp) = ip->i_din.ffs_din; 728 1.32.2.2 perseant return 0; 729 1.32.2.2 perseant } 730 1.32.2.2 perseant 731 1.32.2.2 perseant bp = sp->ibp; 732 1.32.2.2 perseant ((struct dinode *)bp->b_data)[sp->ninodes % INOPB(fs)] = 733 1.32.2.2 perseant ip->i_din.ffs_din; 734 1.32.2.2 perseant 735 1.32.2.2 perseant if(ip->i_number == LFS_IFILE_INUM) /* We know sp->idp == NULL */ 736 1.32.2.2 perseant sp->idp = ((struct dinode *)bp->b_data)+(sp->ninodes % INOPB(fs)); 737 1.32.2.2 perseant if(gotblk) { 738 1.32.2.2 perseant bp->b_flags |= B_LOCKED; 739 1.32.2.2 perseant brelse(bp); 740 1.32.2.2 perseant } 741 1.32.2.2 perseant 742 1.32.2.2 perseant /* Increment inode count in segment summary block. */ 743 1.32.2.2 perseant ++((SEGSUM *)(sp->segsum))->ss_ninos; 744 1.32.2.2 perseant 745 1.32.2.2 perseant /* If this page is full, set flag to allocate a new page. */ 746 1.32.2.2 perseant if (++sp->ninodes % INOPB(fs) == 0) 747 1.32.2.2 perseant sp->ibp = NULL; 748 1.32.2.2 perseant 749 1.32.2.2 perseant /* 750 1.32.2.2 perseant * If updating the ifile, update the super-block. Update the disk 751 1.32.2.2 perseant * address and access times for this inode in the ifile. 752 1.32.2.2 perseant */ 753 1.32.2.2 perseant ino = ip->i_number; 754 1.32.2.2 perseant if (ino == LFS_IFILE_INUM) { 755 1.32.2.2 perseant daddr = fs->lfs_idaddr; 756 1.32.2.2 perseant fs->lfs_idaddr = bp->b_blkno; 757 1.32.2.2 perseant } else { 758 1.32.2.2 perseant LFS_IENTRY(ifp, fs, ino, ibp); 759 1.32.2.2 perseant daddr = ifp->if_daddr; 760 1.32.2.2 perseant ifp->if_daddr = bp->b_blkno; 761 1.32.2.2 perseant #ifdef LFS_DEBUG_NEXTFREE 762 1.32.2.2 perseant if(ino > 3 && ifp->if_nextfree) { 763 1.32.2.2 perseant vprint("lfs_writeinode",ITOV(ip)); 764 1.32.2.2 perseant printf("lfs_writeinode: updating free ino %d\n", 765 1.32.2.2 perseant ip->i_number); 766 1.32.2.2 perseant } 767 1.32.2.2 perseant #endif 768 1.32.2.2 perseant error = VOP_BWRITE(ibp); 769 1.32.2.2 perseant } 770 1.32.2.2 perseant 771 1.32.2.2 perseant /* 772 1.32.2.2 perseant * No need to update segment usage if there was no former inode address 773 1.32.2.2 perseant * or if the last inode address is in the current partial segment. 774 1.32.2.2 perseant */ 775 1.32.2.2 perseant if (daddr != LFS_UNUSED_DADDR && 776 1.32.2.2 perseant !(daddr >= fs->lfs_lastpseg && daddr <= bp->b_blkno)) { 777 1.32.2.2 perseant LFS_SEGENTRY(sup, fs, datosn(fs, daddr), bp); 778 1.32.2.2 perseant #ifdef DIAGNOSTIC 779 1.32.2.2 perseant if (sup->su_nbytes < DINODE_SIZE) { 780 1.32.2.2 perseant /* XXX -- Change to a panic. */ 781 1.32.2.2 perseant printf("lfs_writeinode: negative bytes (segment %d short by %d)\n", 782 1.32.2.2 perseant datosn(fs, daddr), (int)DINODE_SIZE - sup->su_nbytes); 783 1.32.2.2 perseant panic("lfs_writeinode: negative bytes"); 784 1.32.2.2 perseant sup->su_nbytes = DINODE_SIZE; 785 1.32.2.2 perseant } 786 1.32.2.2 perseant #endif 787 1.32.2.2 perseant sup->su_nbytes -= DINODE_SIZE; 788 1.32.2.2 perseant redo_ifile = 789 1.32.2.2 perseant (ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED)); 790 1.32.2.2 perseant error = VOP_BWRITE(bp); 791 1.32.2.2 perseant } 792 1.32.2.2 perseant return (redo_ifile); 793 1.32.2.2 perseant } 794 1.32.2.2 perseant 795 1.32.2.2 perseant int 796 1.32.2.2 perseant lfs_gatherblock(sp, bp, sptr) 797 1.32.2.2 perseant struct segment *sp; 798 1.32.2.2 perseant struct buf *bp; 799 1.32.2.2 perseant int *sptr; 800 1.32.2.2 perseant { 801 1.32.2.2 perseant struct lfs *fs; 802 1.32.2.2 perseant int version; 803 1.32.2.2 perseant 804 1.32.2.2 perseant /* 805 1.32.2.2 perseant * If full, finish this segment. We may be doing I/O, so 806 1.32.2.2 perseant * release and reacquire the splbio(). 807 1.32.2.2 perseant */ 808 1.32.2.2 perseant #ifdef DIAGNOSTIC 809 1.32.2.2 perseant if (sp->vp == NULL) 810 1.32.2.2 perseant panic ("lfs_gatherblock: Null vp in segment"); 811 1.32.2.2 perseant #endif 812 1.32.2.2 perseant fs = sp->fs; 813 1.32.2.2 perseant if (sp->sum_bytes_left < sizeof(ufs_daddr_t) || 814 1.32.2.2 perseant sp->seg_bytes_left < bp->b_bcount) { 815 1.32.2.2 perseant if (sptr) 816 1.32.2.2 perseant splx(*sptr); 817 1.32.2.2 perseant lfs_updatemeta(sp); 818 1.32.2.2 perseant 819 1.32.2.2 perseant version = sp->fip->fi_version; 820 1.32.2.2 perseant (void) lfs_writeseg(fs, sp); 821 1.32.2.2 perseant 822 1.32.2.2 perseant sp->fip->fi_version = version; 823 1.32.2.2 perseant sp->fip->fi_ino = VTOI(sp->vp)->i_number; 824 1.32.2.2 perseant /* Add the current file to the segment summary. */ 825 1.32.2.2 perseant ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 826 1.32.2.2 perseant sp->sum_bytes_left -= 827 1.32.2.2 perseant sizeof(struct finfo) - sizeof(ufs_daddr_t); 828 1.32.2.2 perseant 829 1.32.2.2 perseant if (sptr) 830 1.32.2.2 perseant *sptr = splbio(); 831 1.32.2.2 perseant return(1); 832 1.32.2.2 perseant } 833 1.32.2.2 perseant 834 1.32.2.2 perseant #ifdef DEBUG 835 1.32.2.2 perseant if(bp->b_flags & B_GATHERED) { 836 1.32.2.2 perseant printf("lfs_gatherblock: already gathered! Ino %d, lbn %d\n", 837 1.32.2.2 perseant sp->fip->fi_ino, bp->b_lblkno); 838 1.32.2.2 perseant return(0); 839 1.32.2.2 perseant } 840 1.32.2.2 perseant #endif 841 1.32.2.2 perseant /* Insert into the buffer list, update the FINFO block. */ 842 1.32.2.2 perseant bp->b_flags |= B_GATHERED; 843 1.32.2.2 perseant *sp->cbpp++ = bp; 844 1.32.2.2 perseant sp->fip->fi_blocks[sp->fip->fi_nblocks++] = bp->b_lblkno; 845 1.32.2.2 perseant 846 1.32.2.2 perseant sp->sum_bytes_left -= sizeof(ufs_daddr_t); 847 1.32.2.2 perseant sp->seg_bytes_left -= bp->b_bcount; 848 1.32.2.2 perseant return(0); 849 1.32.2.2 perseant } 850 1.32.2.2 perseant 851 1.32.2.2 perseant int 852 1.32.2.2 perseant lfs_gather(fs, sp, vp, match) 853 1.32.2.2 perseant struct lfs *fs; 854 1.32.2.2 perseant struct segment *sp; 855 1.32.2.2 perseant struct vnode *vp; 856 1.32.2.2 perseant int (*match) __P((struct lfs *, struct buf *)); 857 1.32.2.2 perseant { 858 1.32.2.2 perseant struct buf *bp; 859 1.32.2.2 perseant int s, count=0; 860 1.32.2.2 perseant 861 1.32.2.2 perseant sp->vp = vp; 862 1.32.2.2 perseant s = splbio(); 863 1.32.2.2 perseant 864 1.32.2.2 perseant #ifndef LFS_NO_BACKBUF_HACK 865 1.32.2.2 perseant loop: for (bp = vp->v_dirtyblkhd.lh_first; bp; bp = bp->b_vnbufs.le_next) { 866 1.32.2.2 perseant #else /* LFS_NO_BACKBUF_HACK */ 867 1.32.2.2 perseant /* This is a hack to see if ordering the blocks in LFS makes a difference. */ 868 1.32.2.2 perseant # define BUF_OFFSET (((void *)&bp->b_vnbufs.le_next) - (void *)bp) 869 1.32.2.2 perseant # define BACK_BUF(BP) ((struct buf *)(((void *)BP->b_vnbufs.le_prev) - BUF_OFFSET)) 870 1.32.2.2 perseant # define BEG_OF_LIST ((struct buf *)(((void *)&vp->v_dirtyblkhd.lh_first) - BUF_OFFSET)) 871 1.32.2.2 perseant /* Find last buffer. */ 872 1.32.2.2 perseant loop: for (bp = vp->v_dirtyblkhd.lh_first; bp && bp->b_vnbufs.le_next != NULL; 873 1.32.2.2 perseant bp = bp->b_vnbufs.le_next); 874 1.32.2.2 perseant for (; bp && bp != BEG_OF_LIST; bp = BACK_BUF(bp)) { 875 1.32.2.2 perseant #endif /* LFS_NO_BACKBUF_HACK */ 876 1.32.2.2 perseant if ((bp->b_flags & (B_BUSY|B_GATHERED)) || !match(fs, bp)) 877 1.32.2.2 perseant continue; 878 1.32.2.2 perseant if(vp->v_type == VBLK) { 879 1.32.2.2 perseant /* For block devices, just write the blocks. */ 880 1.32.2.2 perseant /* XXX Do we really need to even do this? */ 881 1.32.2.2 perseant #ifdef DEBUG_LFS 882 1.32.2.2 perseant if(count==0) 883 1.32.2.2 perseant printf("BLK("); 884 1.32.2.2 perseant printf("."); 885 1.32.2.2 perseant #endif 886 1.32.2.2 perseant /* Get the block before bwrite, so we don't corrupt the free list */ 887 1.32.2.2 perseant bp->b_flags |= B_BUSY; 888 1.32.2.2 perseant bremfree(bp); 889 1.32.2.2 perseant bwrite(bp); 890 1.32.2.2 perseant } else { 891 1.32.2.2 perseant #ifdef DIAGNOSTIC 892 1.32.2.2 perseant if (!(bp->b_flags & B_DELWRI)) 893 1.32.2.2 perseant panic("lfs_gather: bp not B_DELWRI"); 894 1.32.2.2 perseant if (!(bp->b_flags & B_LOCKED)) { 895 1.32.2.2 perseant printf("lfs_gather: lbn %d blk %d not B_LOCKED\n", bp->b_lblkno, bp->b_blkno); 896 1.32.2.2 perseant VOP_PRINT(bp->b_vp); 897 1.32.2.2 perseant panic("lfs_gather: bp not B_LOCKED"); 898 1.32.2.2 perseant } 899 1.32.2.2 perseant #endif 900 1.32.2.2 perseant if (lfs_gatherblock(sp, bp, &s)) { 901 1.32.2.2 perseant goto loop; 902 1.32.2.2 perseant } 903 1.32.2.2 perseant } 904 1.32.2.2 perseant count++; 905 1.32.2.2 perseant } 906 1.32.2.2 perseant splx(s); 907 1.32.2.2 perseant #ifdef DEBUG_LFS 908 1.32.2.2 perseant if(vp->v_type == VBLK && count) 909 1.32.2.2 perseant printf(")\n"); 910 1.32.2.2 perseant #endif 911 1.32.2.2 perseant lfs_updatemeta(sp); 912 1.32.2.2 perseant sp->vp = NULL; 913 1.32.2.2 perseant return count; 914 1.32.2.2 perseant } 915 1.32.2.2 perseant 916 1.32.2.2 perseant /* 917 1.32.2.2 perseant * Update the metadata that points to the blocks listed in the FINFO 918 1.32.2.2 perseant * array. 919 1.32.2.2 perseant */ 920 1.32.2.2 perseant void 921 1.32.2.2 perseant lfs_updatemeta(sp) 922 1.32.2.2 perseant struct segment *sp; 923 1.32.2.2 perseant { 924 1.32.2.2 perseant SEGUSE *sup; 925 1.32.2.2 perseant struct buf *bp; 926 1.32.2.2 perseant struct lfs *fs; 927 1.32.2.2 perseant struct vnode *vp; 928 1.32.2.2 perseant struct indir a[NIADDR + 2], *ap; 929 1.32.2.2 perseant struct inode *ip; 930 1.32.2.2 perseant ufs_daddr_t daddr, lbn, off; 931 1.32.2.2 perseant int error, i, nblocks, num; 932 1.32.2.2 perseant 933 1.32.2.2 perseant vp = sp->vp; 934 1.32.2.2 perseant nblocks = &sp->fip->fi_blocks[sp->fip->fi_nblocks] - sp->start_lbp; 935 1.32.2.2 perseant if (nblocks < 0) 936 1.32.2.2 perseant panic("This is a bad thing\n"); 937 1.32.2.2 perseant if (vp == NULL || nblocks == 0) 938 1.32.2.2 perseant return; 939 1.32.2.2 perseant 940 1.32.2.2 perseant /* Sort the blocks. */ 941 1.32.2.2 perseant /* 942 1.32.2.2 perseant * XXX KS - We have to sort even if the blocks come from the 943 1.32.2.2 perseant * cleaner, because there might be other pending blocks on the 944 1.32.2.2 perseant * same inode...and if we don't sort, and there are fragments 945 1.32.2.2 perseant * present, blocks may be written in the wrong place. 946 1.32.2.2 perseant */ 947 1.32.2.2 perseant /* if (!(sp->seg_flags & SEGM_CLEAN)) */ 948 1.32.2.2 perseant lfs_shellsort(sp->start_bpp, sp->start_lbp, nblocks); 949 1.32.2.2 perseant 950 1.32.2.2 perseant /* 951 1.32.2.2 perseant * Record the length of the last block in case it's a fragment. 952 1.32.2.2 perseant * If there are indirect blocks present, they sort last. An 953 1.32.2.2 perseant * indirect block will be lfs_bsize and its presence indicates 954 1.32.2.2 perseant * that you cannot have fragments. 955 1.32.2.2 perseant */ 956 1.32.2.2 perseant sp->fip->fi_lastlength = sp->start_bpp[nblocks - 1]->b_bcount; 957 1.32.2.2 perseant 958 1.32.2.2 perseant /* 959 1.32.2.2 perseant * Assign disk addresses, and update references to the logical 960 1.32.2.2 perseant * block and the segment usage information. 961 1.32.2.2 perseant */ 962 1.32.2.2 perseant fs = sp->fs; 963 1.32.2.2 perseant for (i = nblocks; i--; ++sp->start_bpp) { 964 1.32.2.2 perseant lbn = *sp->start_lbp++; 965 1.32.2.2 perseant 966 1.32.2.2 perseant (*sp->start_bpp)->b_blkno = off = fs->lfs_offset; 967 1.32.2.2 perseant if((*sp->start_bpp)->b_blkno == (*sp->start_bpp)->b_lblkno) { 968 1.32.2.2 perseant printf("lfs_updatemeta: ino %d blk %d has same lbn and daddr\n", VTOI(vp)->i_number, off); 969 1.32.2.2 perseant } 970 1.32.2.2 perseant fs->lfs_offset += 971 1.32.2.2 perseant fragstodb(fs, numfrags(fs, (*sp->start_bpp)->b_bcount)); 972 1.32.2.2 perseant error = ufs_bmaparray(vp, lbn, &daddr, a, &num, NULL); 973 1.32.2.2 perseant if (error) 974 1.32.2.2 perseant panic("lfs_updatemeta: ufs_bmaparray %d", error); 975 1.32.2.2 perseant ip = VTOI(vp); 976 1.32.2.2 perseant switch (num) { 977 1.32.2.2 perseant case 0: 978 1.32.2.2 perseant ip->i_ffs_db[lbn] = off; 979 1.32.2.2 perseant break; 980 1.32.2.2 perseant case 1: 981 1.32.2.2 perseant ip->i_ffs_ib[a[0].in_off] = off; 982 1.32.2.2 perseant break; 983 1.32.2.2 perseant default: 984 1.32.2.2 perseant ap = &a[num - 1]; 985 1.32.2.2 perseant if (bread(vp, ap->in_lbn, fs->lfs_bsize, NOCRED, &bp)) 986 1.32.2.2 perseant panic("lfs_updatemeta: bread bno %d", 987 1.32.2.2 perseant ap->in_lbn); 988 1.32.2.2 perseant /* 989 1.32.2.2 perseant * Bread may create a new (indirect) block which needs 990 1.32.2.2 perseant * to get counted for the inode. 991 1.32.2.2 perseant */ 992 1.32.2.2 perseant if (/* bp->b_blkno == -1 && */ 993 1.32.2.2 perseant !(bp->b_flags & (B_DELWRI|B_DONE))) { 994 1.32.2.2 perseant ip->i_ffs_blocks += fsbtodb(fs, 1); 995 1.32.2.2 perseant fs->lfs_bfree -= fragstodb(fs, fs->lfs_frag); 996 1.32.2.2 perseant } 997 1.32.2.2 perseant ((ufs_daddr_t *)bp->b_data)[ap->in_off] = off; 998 1.32.2.2 perseant VOP_BWRITE(bp); 999 1.32.2.2 perseant } 1000 1.32.2.2 perseant /* Update segment usage information. */ 1001 1.32.2.2 perseant if (daddr != UNASSIGNED && !(daddr >= fs->lfs_lastpseg && daddr <= off)) { 1002 1.32.2.2 perseant LFS_SEGENTRY(sup, fs, datosn(fs, daddr), bp); 1003 1.32.2.2 perseant #ifdef DIAGNOSTIC 1004 1.32.2.2 perseant if (sup->su_nbytes < (*sp->start_bpp)->b_bcount) { 1005 1.32.2.2 perseant /* XXX -- Change to a panic. */ 1006 1.32.2.2 perseant printf("lfs_updatemeta: negative bytes (segment %d short by %ld)\n", 1007 1.32.2.2 perseant datosn(fs, daddr), (*sp->start_bpp)->b_bcount - sup->su_nbytes); 1008 1.32.2.2 perseant printf("lfs_updatemeta: ino %d, lbn %d, addr = %x\n", 1009 1.32.2.2 perseant VTOI(sp->vp)->i_number, (*sp->start_bpp)->b_lblkno, daddr); 1010 1.32.2.2 perseant panic("lfs_updatemeta: negative bytes"); 1011 1.32.2.2 perseant sup->su_nbytes = (*sp->start_bpp)->b_bcount; 1012 1.32.2.2 perseant } 1013 1.32.2.2 perseant #endif 1014 1.32.2.2 perseant sup->su_nbytes -= (*sp->start_bpp)->b_bcount; 1015 1.32.2.2 perseant error = VOP_BWRITE(bp); 1016 1.32.2.2 perseant } 1017 1.32.2.2 perseant } 1018 1.32.2.2 perseant } 1019 1.32.2.2 perseant 1020 1.32.2.2 perseant /* 1021 1.32.2.2 perseant * Start a new segment. 1022 1.32.2.2 perseant */ 1023 1.32.2.2 perseant int 1024 1.32.2.2 perseant lfs_initseg(fs) 1025 1.32.2.2 perseant struct lfs *fs; 1026 1.32.2.2 perseant { 1027 1.32.2.2 perseant struct segment *sp; 1028 1.32.2.2 perseant SEGUSE *sup; 1029 1.32.2.2 perseant SEGSUM *ssp; 1030 1.32.2.2 perseant struct buf *bp; 1031 1.32.2.2 perseant int repeat; 1032 1.32.2.2 perseant 1033 1.32.2.2 perseant sp = fs->lfs_sp; 1034 1.32.2.2 perseant 1035 1.32.2.2 perseant repeat = 0; 1036 1.32.2.2 perseant /* Advance to the next segment. */ 1037 1.32.2.2 perseant if (!LFS_PARTIAL_FITS(fs)) { 1038 1.32.2.2 perseant /* Wake up any cleaning procs waiting on this file system. */ 1039 1.32.2.2 perseant wakeup(&lfs_allclean_wakeup); 1040 1.32.2.2 perseant wakeup(&fs->lfs_nextseg); 1041 1.32.2.2 perseant lfs_newseg(fs); 1042 1.32.2.2 perseant repeat = 1; 1043 1.32.2.2 perseant fs->lfs_offset = fs->lfs_curseg; 1044 1.32.2.2 perseant sp->seg_number = datosn(fs, fs->lfs_curseg); 1045 1.32.2.2 perseant sp->seg_bytes_left = fs->lfs_dbpseg * DEV_BSIZE; 1046 1.32.2.2 perseant /* 1047 1.32.2.2 perseant * If the segment contains a superblock, update the offset 1048 1.32.2.2 perseant * and summary address to skip over it. 1049 1.32.2.2 perseant */ 1050 1.32.2.2 perseant LFS_SEGENTRY(sup, fs, sp->seg_number, bp); 1051 1.32.2.2 perseant if (sup->su_flags & SEGUSE_SUPERBLOCK) { 1052 1.32.2.2 perseant fs->lfs_offset += LFS_SBPAD / DEV_BSIZE; 1053 1.32.2.2 perseant sp->seg_bytes_left -= LFS_SBPAD; 1054 1.32.2.2 perseant } 1055 1.32.2.2 perseant brelse(bp); 1056 1.32.2.2 perseant } else { 1057 1.32.2.2 perseant sp->seg_number = datosn(fs, fs->lfs_curseg); 1058 1.32.2.2 perseant sp->seg_bytes_left = (fs->lfs_dbpseg - 1059 1.32.2.2 perseant (fs->lfs_offset - fs->lfs_curseg)) * DEV_BSIZE; 1060 1.32.2.2 perseant } 1061 1.32.2.2 perseant fs->lfs_lastpseg = fs->lfs_offset; 1062 1.32.2.2 perseant 1063 1.32.2.2 perseant sp->fs = fs; 1064 1.32.2.2 perseant sp->ibp = NULL; 1065 1.32.2.2 perseant sp->idp = NULL; 1066 1.32.2.2 perseant sp->ninodes = 0; 1067 1.32.2.2 perseant 1068 1.32.2.2 perseant /* Get a new buffer for SEGSUM and enter it into the buffer list. */ 1069 1.32.2.2 perseant sp->cbpp = sp->bpp; 1070 1.32.2.2 perseant *sp->cbpp = lfs_newbuf(VTOI(fs->lfs_ivnode)->i_devvp, 1071 1.32.2.2 perseant fs->lfs_offset, LFS_SUMMARY_SIZE); 1072 1.32.2.2 perseant sp->segsum = (*sp->cbpp)->b_data; 1073 1.32.2.2 perseant bzero(sp->segsum, LFS_SUMMARY_SIZE); 1074 1.32.2.2 perseant sp->start_bpp = ++sp->cbpp; 1075 1.32.2.2 perseant fs->lfs_offset += LFS_SUMMARY_SIZE / DEV_BSIZE; 1076 1.32.2.2 perseant 1077 1.32.2.2 perseant /* Set point to SEGSUM, initialize it. */ 1078 1.32.2.2 perseant ssp = sp->segsum; 1079 1.32.2.2 perseant ssp->ss_next = fs->lfs_nextseg; 1080 1.32.2.2 perseant ssp->ss_nfinfo = ssp->ss_ninos = 0; 1081 1.32.2.2 perseant ssp->ss_magic = SS_MAGIC; 1082 1.32.2.2 perseant 1083 1.32.2.2 perseant /* Set pointer to first FINFO, initialize it. */ 1084 1.32.2.2 perseant sp->fip = (struct finfo *)((caddr_t)sp->segsum + sizeof(SEGSUM)); 1085 1.32.2.2 perseant sp->fip->fi_nblocks = 0; 1086 1.32.2.2 perseant sp->start_lbp = &sp->fip->fi_blocks[0]; 1087 1.32.2.2 perseant sp->fip->fi_lastlength = 0; 1088 1.32.2.2 perseant 1089 1.32.2.2 perseant sp->seg_bytes_left -= LFS_SUMMARY_SIZE; 1090 1.32.2.2 perseant sp->sum_bytes_left = LFS_SUMMARY_SIZE - sizeof(SEGSUM); 1091 1.32.2.2 perseant 1092 1.32.2.2 perseant return(repeat); 1093 1.32.2.2 perseant } 1094 1.32.2.2 perseant 1095 1.32.2.2 perseant /* 1096 1.32.2.2 perseant * Return the next segment to write. 1097 1.32.2.2 perseant */ 1098 1.32.2.2 perseant void 1099 1.32.2.2 perseant lfs_newseg(fs) 1100 1.32.2.2 perseant struct lfs *fs; 1101 1.32.2.2 perseant { 1102 1.32.2.2 perseant CLEANERINFO *cip; 1103 1.32.2.2 perseant SEGUSE *sup; 1104 1.32.2.2 perseant struct buf *bp; 1105 1.32.2.2 perseant int curseg, isdirty, sn; 1106 1.32.2.2 perseant 1107 1.32.2.2 perseant LFS_SEGENTRY(sup, fs, datosn(fs, fs->lfs_nextseg), bp); 1108 1.32.2.2 perseant sup->su_flags |= SEGUSE_DIRTY | SEGUSE_ACTIVE; 1109 1.32.2.2 perseant sup->su_nbytes = 0; 1110 1.32.2.2 perseant sup->su_nsums = 0; 1111 1.32.2.2 perseant sup->su_ninos = 0; 1112 1.32.2.2 perseant (void) VOP_BWRITE(bp); 1113 1.32.2.2 perseant 1114 1.32.2.2 perseant LFS_CLEANERINFO(cip, fs, bp); 1115 1.32.2.2 perseant --cip->clean; 1116 1.32.2.2 perseant ++cip->dirty; 1117 1.32.2.2 perseant fs->lfs_nclean = cip->clean; 1118 1.32.2.2 perseant (void) VOP_BWRITE(bp); 1119 1.32.2.2 perseant 1120 1.32.2.2 perseant fs->lfs_lastseg = fs->lfs_curseg; 1121 1.32.2.2 perseant fs->lfs_curseg = fs->lfs_nextseg; 1122 1.32.2.2 perseant for (sn = curseg = datosn(fs, fs->lfs_curseg);;) { 1123 1.32.2.2 perseant sn = (sn + 1) % fs->lfs_nseg; 1124 1.32.2.2 perseant if (sn == curseg) 1125 1.32.2.2 perseant panic("lfs_nextseg: no clean segments"); 1126 1.32.2.2 perseant LFS_SEGENTRY(sup, fs, sn, bp); 1127 1.32.2.2 perseant isdirty = sup->su_flags & SEGUSE_DIRTY; 1128 1.32.2.2 perseant brelse(bp); 1129 1.32.2.2 perseant if (!isdirty) 1130 1.32.2.2 perseant break; 1131 1.32.2.2 perseant } 1132 1.32.2.2 perseant 1133 1.32.2.2 perseant ++fs->lfs_nactive; 1134 1.32.2.2 perseant fs->lfs_nextseg = sntoda(fs, sn); 1135 1.32.2.2 perseant if(lfs_dostats) { 1136 1.32.2.2 perseant ++lfs_stats.segsused; 1137 1.32.2.2 perseant } 1138 1.32.2.2 perseant } 1139 1.32.2.2 perseant 1140 1.32.2.2 perseant int 1141 1.32.2.2 perseant lfs_writeseg(fs, sp) 1142 1.32.2.2 perseant struct lfs *fs; 1143 1.32.2.2 perseant struct segment *sp; 1144 1.32.2.2 perseant { 1145 1.32.2.2 perseant extern int locked_queue_count; 1146 1.32.2.2 perseant extern long locked_queue_bytes; 1147 1.32.2.2 perseant struct buf **bpp, *bp, *cbp; 1148 1.32.2.2 perseant SEGUSE *sup; 1149 1.32.2.2 perseant SEGSUM *ssp; 1150 1.32.2.2 perseant dev_t i_dev; 1151 1.32.2.2 perseant u_long *datap, *dp; 1152 1.32.2.2 perseant int do_again, i, nblocks, s; 1153 1.32.2.2 perseant #ifdef LFS_TRACK_IOS 1154 1.32.2.2 perseant int j; 1155 1.32.2.2 perseant #endif 1156 1.32.2.2 perseant int (*strategy)__P((void *)); 1157 1.32.2.2 perseant struct vop_strategy_args vop_strategy_a; 1158 1.32.2.2 perseant u_short ninos; 1159 1.32.2.2 perseant struct vnode *devvp; 1160 1.32.2.2 perseant char *p; 1161 1.32.2.2 perseant struct vnode *vn; 1162 1.32.2.2 perseant struct inode *ip; 1163 1.32.2.2 perseant #if defined(DEBUG) && defined(LFS_PROPELLER) 1164 1.32.2.2 perseant static int propeller; 1165 1.32.2.2 perseant char propstring[4] = "-\\|/"; 1166 1.32.2.2 perseant 1167 1.32.2.2 perseant printf("%c\b",propstring[propeller++]); 1168 1.32.2.2 perseant if(propeller==4) 1169 1.32.2.2 perseant propeller = 0; 1170 1.32.2.2 perseant #endif 1171 1.32.2.2 perseant 1172 1.32.2.2 perseant /* 1173 1.32.2.2 perseant * If there are no buffers other than the segment summary to write 1174 1.32.2.2 perseant * and it is not a checkpoint, don't do anything. On a checkpoint, 1175 1.32.2.2 perseant * even if there aren't any buffers, you need to write the superblock. 1176 1.32.2.2 perseant */ 1177 1.32.2.2 perseant if ((nblocks = sp->cbpp - sp->bpp) == 1) 1178 1.32.2.2 perseant return (0); 1179 1.32.2.2 perseant 1180 1.32.2.2 perseant #ifdef DEBUG_LFS 1181 1.32.2.2 perseant lfs_check_bpp(fs,sp,__FILE__,__LINE__); 1182 1.32.2.2 perseant #endif 1183 1.32.2.2 perseant i_dev = VTOI(fs->lfs_ivnode)->i_dev; 1184 1.32.2.2 perseant devvp = VTOI(fs->lfs_ivnode)->i_devvp; 1185 1.32.2.2 perseant 1186 1.32.2.2 perseant /* Update the segment usage information. */ 1187 1.32.2.2 perseant LFS_SEGENTRY(sup, fs, sp->seg_number, bp); 1188 1.32.2.2 perseant 1189 1.32.2.2 perseant /* Loop through all blocks, except the segment summary. */ 1190 1.32.2.2 perseant for (bpp = sp->bpp; ++bpp < sp->cbpp; ) { 1191 1.32.2.2 perseant if((*bpp)->b_vp != devvp) 1192 1.32.2.2 perseant sup->su_nbytes += (*bpp)->b_bcount; 1193 1.32.2.2 perseant } 1194 1.32.2.2 perseant 1195 1.32.2.2 perseant ssp = (SEGSUM *)sp->segsum; 1196 1.32.2.2 perseant 1197 1.32.2.2 perseant ninos = (ssp->ss_ninos + INOPB(fs) - 1) / INOPB(fs); 1198 1.32.2.2 perseant sup->su_nbytes += ssp->ss_ninos * DINODE_SIZE; 1199 1.32.2.2 perseant /* sup->su_nbytes += LFS_SUMMARY_SIZE; */ 1200 1.32.2.2 perseant sup->su_lastmod = time.tv_sec; 1201 1.32.2.2 perseant sup->su_ninos += ninos; 1202 1.32.2.2 perseant ++sup->su_nsums; 1203 1.32.2.2 perseant 1204 1.32.2.2 perseant do_again = !(bp->b_flags & B_GATHERED); 1205 1.32.2.2 perseant (void)VOP_BWRITE(bp); 1206 1.32.2.2 perseant /* 1207 1.32.2.2 perseant * Compute checksum across data and then across summary; the first 1208 1.32.2.2 perseant * block (the summary block) is skipped. Set the create time here 1209 1.32.2.2 perseant * so that it's guaranteed to be later than the inode mod times. 1210 1.32.2.2 perseant * 1211 1.32.2.2 perseant * XXX 1212 1.32.2.2 perseant * Fix this to do it inline, instead of malloc/copy. 1213 1.32.2.2 perseant */ 1214 1.32.2.2 perseant datap = dp = malloc(nblocks * sizeof(u_long), M_SEGMENT, M_WAITOK); 1215 1.32.2.2 perseant for (bpp = sp->bpp, i = nblocks - 1; i--;) { 1216 1.32.2.2 perseant if (((*++bpp)->b_flags & (B_CALL|B_INVAL)) == (B_CALL|B_INVAL)) { 1217 1.32.2.2 perseant if (copyin((*bpp)->b_saveaddr, dp++, sizeof(u_long))) 1218 1.32.2.2 perseant panic("lfs_writeseg: copyin failed [1]: ino %d blk %d", VTOI((*bpp)->b_vp)->i_number, (*bpp)->b_lblkno); 1219 1.32.2.2 perseant } else { 1220 1.32.2.2 perseant if( !((*bpp)->b_flags & B_CALL) ) { 1221 1.32.2.2 perseant /* 1222 1.32.2.2 perseant * Before we record data for a checksm, 1223 1.32.2.2 perseant * make sure the data won't change in between 1224 1.32.2.2 perseant * the checksum calculation and the write, 1225 1.32.2.2 perseant * by marking the buffer B_BUSY. It will 1226 1.32.2.2 perseant * be freed later by brelse(). 1227 1.32.2.2 perseant */ 1228 1.32.2.2 perseant again: 1229 1.32.2.2 perseant s = splbio(); 1230 1.32.2.2 perseant if((*bpp)->b_flags & B_BUSY) { 1231 1.32.2.2 perseant #ifdef DEBUG 1232 1.32.2.2 perseant printf("lfs_writeseg: avoiding potential data summary corruption for ino %d, lbn %d\n", 1233 1.32.2.2 perseant VTOI((*bpp)->b_vp)->i_number, 1234 1.32.2.2 perseant bp->b_lblkno); 1235 1.32.2.2 perseant #endif 1236 1.32.2.2 perseant (*bpp)->b_flags |= B_WANTED; 1237 1.32.2.2 perseant tsleep((*bpp), (PRIBIO + 1), 1238 1.32.2.2 perseant "lfs_writeseg", 0); 1239 1.32.2.2 perseant splx(s); 1240 1.32.2.2 perseant goto again; 1241 1.32.2.2 perseant } 1242 1.32.2.2 perseant (*bpp)->b_flags |= B_BUSY; 1243 1.32.2.2 perseant splx(s); 1244 1.32.2.2 perseant } 1245 1.32.2.2 perseant *dp++ = ((u_long *)(*bpp)->b_data)[0]; 1246 1.32.2.2 perseant } 1247 1.32.2.2 perseant } 1248 1.32.2.2 perseant ssp->ss_create = time.tv_sec; 1249 1.32.2.2 perseant ssp->ss_datasum = cksum(datap, (nblocks - 1) * sizeof(u_long)); 1250 1.32.2.2 perseant ssp->ss_sumsum = 1251 1.32.2.2 perseant cksum(&ssp->ss_datasum, LFS_SUMMARY_SIZE - sizeof(ssp->ss_sumsum)); 1252 1.32.2.2 perseant free(datap, M_SEGMENT); 1253 1.32.2.2 perseant #ifdef DIAGNOSTIC 1254 1.32.2.2 perseant if (fs->lfs_bfree < fsbtodb(fs, ninos) + LFS_SUMMARY_SIZE / DEV_BSIZE) 1255 1.32.2.2 perseant panic("lfs_writeseg: No diskspace for summary"); 1256 1.32.2.2 perseant #endif 1257 1.32.2.2 perseant fs->lfs_bfree -= (fsbtodb(fs, ninos) + LFS_SUMMARY_SIZE / DEV_BSIZE); 1258 1.32.2.2 perseant 1259 1.32.2.2 perseant strategy = devvp->v_op[VOFFSET(vop_strategy)]; 1260 1.32.2.2 perseant 1261 1.32.2.2 perseant /* 1262 1.32.2.2 perseant * When we simply write the blocks we lose a rotation for every block 1263 1.32.2.2 perseant * written. To avoid this problem, we allocate memory in chunks, copy 1264 1.32.2.2 perseant * the buffers into the chunk and write the chunk. CHUNKSIZE is the 1265 1.32.2.2 perseant * largest size I/O devices can handle. 1266 1.32.2.2 perseant * When the data is copied to the chunk, turn off the the B_LOCKED bit 1267 1.32.2.2 perseant * and brelse the buffer (which will move them to the LRU list). Add 1268 1.32.2.2 perseant * the B_CALL flag to the buffer header so we can count I/O's for the 1269 1.32.2.2 perseant * checkpoints and so we can release the allocated memory. 1270 1.32.2.2 perseant * 1271 1.32.2.2 perseant * XXX 1272 1.32.2.2 perseant * This should be removed if the new virtual memory system allows us to 1273 1.32.2.2 perseant * easily make the buffers contiguous in kernel memory and if that's 1274 1.32.2.2 perseant * fast enough. 1275 1.32.2.2 perseant */ 1276 1.32.2.2 perseant 1277 1.32.2.2 perseant #define CHUNKSIZE MAXPHYS 1278 1.32.2.2 perseant 1279 1.32.2.2 perseant if(devvp==NULL) 1280 1.32.2.2 perseant panic("devvp is NULL"); 1281 1.32.2.2 perseant for (bpp = sp->bpp,i = nblocks; i;) { 1282 1.32.2.2 perseant cbp = lfs_newbuf(devvp, (*bpp)->b_blkno, CHUNKSIZE); 1283 1.32.2.2 perseant cbp->b_dev = i_dev; 1284 1.32.2.2 perseant cbp->b_flags |= B_ASYNC | B_BUSY; 1285 1.32.2.2 perseant cbp->b_bcount = 0; 1286 1.32.2.2 perseant 1287 1.32.2.2 perseant #ifdef DIAGNOSTIC 1288 1.32.2.2 perseant if(datosn(fs,(*bpp)->b_blkno + ((*bpp)->b_bcount - 1)/DEV_BSIZE) != datosn(fs,cbp->b_blkno)) { 1289 1.32.2.2 perseant panic("lfs_writeseg: Segment overwrite"); 1290 1.32.2.2 perseant } 1291 1.32.2.2 perseant #endif 1292 1.32.2.2 perseant 1293 1.32.2.2 perseant if(fs->lfs_iocount >= LFS_THROTTLE) { 1294 1.32.2.2 perseant tsleep(&fs->lfs_iocount, PRIBIO+1, "lfs throttle", 0); 1295 1.32.2.2 perseant } 1296 1.32.2.2 perseant s = splbio(); 1297 1.32.2.2 perseant ++fs->lfs_iocount; 1298 1.32.2.2 perseant #ifdef LFS_TRACK_IOS 1299 1.32.2.2 perseant for(j=0;j<LFS_THROTTLE;j++) { 1300 1.32.2.2 perseant if(fs->lfs_pending[j]==LFS_UNUSED_DADDR) { 1301 1.32.2.2 perseant fs->lfs_pending[j] = cbp->b_blkno; 1302 1.32.2.2 perseant break; 1303 1.32.2.2 perseant } 1304 1.32.2.2 perseant } 1305 1.32.2.2 perseant #endif /* LFS_TRACK_IOS */ 1306 1.32.2.2 perseant for (p = cbp->b_data; i && cbp->b_bcount < CHUNKSIZE; i--) { 1307 1.32.2.2 perseant bp = *bpp; 1308 1.32.2.2 perseant 1309 1.32.2.2 perseant if (bp->b_bcount > (CHUNKSIZE - cbp->b_bcount)) 1310 1.32.2.2 perseant break; 1311 1.32.2.2 perseant 1312 1.32.2.2 perseant /* 1313 1.32.2.2 perseant * Fake buffers from the cleaner are marked as B_INVAL. 1314 1.32.2.2 perseant * We need to copy the data from user space rather than 1315 1.32.2.2 perseant * from the buffer indicated. 1316 1.32.2.2 perseant * XXX == what do I do on an error? 1317 1.32.2.2 perseant */ 1318 1.32.2.2 perseant if ((bp->b_flags & (B_CALL|B_INVAL)) == (B_CALL|B_INVAL)) { 1319 1.32.2.2 perseant if (copyin(bp->b_saveaddr, p, bp->b_bcount)) 1320 1.32.2.2 perseant panic("lfs_writeseg: copyin failed [2]"); 1321 1.32.2.2 perseant } else 1322 1.32.2.2 perseant bcopy(bp->b_data, p, bp->b_bcount); 1323 1.32.2.2 perseant p += bp->b_bcount; 1324 1.32.2.2 perseant cbp->b_bcount += bp->b_bcount; 1325 1.32.2.2 perseant if (bp->b_flags & B_LOCKED) { 1326 1.32.2.2 perseant --locked_queue_count; 1327 1.32.2.2 perseant locked_queue_bytes -= bp->b_bufsize; 1328 1.32.2.2 perseant } 1329 1.32.2.2 perseant bp->b_flags &= ~(B_ERROR | B_READ | B_DELWRI | 1330 1.32.2.2 perseant B_LOCKED | B_GATHERED); 1331 1.32.2.2 perseant vn = bp->b_vp; 1332 1.32.2.2 perseant if (bp->b_flags & B_CALL) { 1333 1.32.2.2 perseant /* if B_CALL, it was created with newbuf */ 1334 1.32.2.2 perseant lfs_freebuf(bp); 1335 1.32.2.2 perseant } else { 1336 1.32.2.2 perseant bremfree(bp); 1337 1.32.2.2 perseant bp->b_flags |= B_DONE; 1338 1.32.2.2 perseant if(vn) 1339 1.32.2.2 perseant reassignbuf(bp, vn); 1340 1.32.2.2 perseant brelse(bp); 1341 1.32.2.2 perseant } 1342 1.32.2.2 perseant if(bp->b_flags & B_NEEDCOMMIT) { /* XXX */ 1343 1.32.2.2 perseant bp->b_flags &= ~B_NEEDCOMMIT; 1344 1.32.2.2 perseant wakeup(bp); 1345 1.32.2.2 perseant } 1346 1.32.2.2 perseant 1347 1.32.2.2 perseant bpp++; 1348 1.32.2.2 perseant 1349 1.32.2.2 perseant /* 1350 1.32.2.2 perseant * If this is the last block for this vnode, but 1351 1.32.2.2 perseant * there are other blocks on its dirty list, 1352 1.32.2.2 perseant * set IN_MODIFIED/IN_CLEANING depending on what 1353 1.32.2.2 perseant * sort of block. Only do this for our mount point, 1354 1.32.2.2 perseant * not for, e.g., inode blocks that are attached to 1355 1.32.2.2 perseant * the devvp. 1356 1.32.2.2 perseant */ 1357 1.32.2.2 perseant if(i>1 && vn && *bpp && (*bpp)->b_vp != vn 1358 1.32.2.2 perseant && (*bpp)->b_vp && (bp=vn->v_dirtyblkhd.lh_first)!=NULL && 1359 1.32.2.2 perseant vn->v_mount == fs->lfs_ivnode->v_mount) 1360 1.32.2.2 perseant { 1361 1.32.2.2 perseant ip = VTOI(vn); 1362 1.32.2.2 perseant #ifdef DEBUG_LFS 1363 1.32.2.2 perseant printf("lfs_writeseg: marking ino %d\n",ip->i_number); 1364 1.32.2.2 perseant #endif 1365 1.32.2.2 perseant if(!(ip->i_flag & (IN_CLEANING|IN_MODIFIED))) { 1366 1.32.2.2 perseant fs->lfs_uinodes++; 1367 1.32.2.2 perseant if(bp->b_flags & B_CALL) 1368 1.32.2.2 perseant ip->i_flag |= IN_CLEANING; 1369 1.32.2.2 perseant else 1370 1.32.2.2 perseant ip->i_flag |= IN_MODIFIED; 1371 1.32.2.2 perseant } 1372 1.32.2.2 perseant } 1373 1.32.2.2 perseant /* if(vn->v_dirtyblkhd.lh_first == NULL) */ 1374 1.32.2.2 perseant wakeup(vn); 1375 1.32.2.2 perseant } 1376 1.32.2.2 perseant ++cbp->b_vp->v_numoutput; 1377 1.32.2.2 perseant splx(s); 1378 1.32.2.2 perseant /* 1379 1.32.2.2 perseant * XXXX This is a gross and disgusting hack. Since these 1380 1.32.2.2 perseant * buffers are physically addressed, they hang off the 1381 1.32.2.2 perseant * device vnode (devvp). As a result, they have no way 1382 1.32.2.2 perseant * of getting to the LFS superblock or lfs structure to 1383 1.32.2.2 perseant * keep track of the number of I/O's pending. So, I am 1384 1.32.2.2 perseant * going to stuff the fs into the saveaddr field of 1385 1.32.2.2 perseant * the buffer (yuk). 1386 1.32.2.2 perseant */ 1387 1.32.2.2 perseant cbp->b_saveaddr = (caddr_t)fs; 1388 1.32.2.2 perseant vop_strategy_a.a_desc = VDESC(vop_strategy); 1389 1.32.2.2 perseant vop_strategy_a.a_bp = cbp; 1390 1.32.2.2 perseant (strategy)(&vop_strategy_a); 1391 1.32.2.2 perseant } 1392 1.32.2.2 perseant /* 1393 1.32.2.2 perseant * XXX 1394 1.32.2.2 perseant * Vinvalbuf can move locked buffers off the locked queue 1395 1.32.2.2 perseant * and we have no way of knowing about this. So, after 1396 1.32.2.2 perseant * doing a big write, we recalculate how many buffers are 1397 1.32.2.2 perseant * really still left on the locked queue. 1398 1.32.2.2 perseant */ 1399 1.32.2.2 perseant lfs_countlocked(&locked_queue_count,&locked_queue_bytes); 1400 1.32.2.2 perseant wakeup(&locked_queue_count); 1401 1.32.2.2 perseant if(lfs_dostats) { 1402 1.32.2.2 perseant ++lfs_stats.psegwrites; 1403 1.32.2.2 perseant lfs_stats.blocktot += nblocks - 1; 1404 1.32.2.2 perseant if (fs->lfs_sp->seg_flags & SEGM_SYNC) 1405 1.32.2.2 perseant ++lfs_stats.psyncwrites; 1406 1.32.2.2 perseant if (fs->lfs_sp->seg_flags & SEGM_CLEAN) { 1407 1.32.2.2 perseant ++lfs_stats.pcleanwrites; 1408 1.32.2.2 perseant lfs_stats.cleanblocks += nblocks - 1; 1409 1.32.2.2 perseant } 1410 1.32.2.2 perseant } 1411 1.32.2.2 perseant return (lfs_initseg(fs) || do_again); 1412 1.32.2.2 perseant } 1413 1.32.2.2 perseant 1414 1.32.2.2 perseant void 1415 1.32.2.2 perseant lfs_writesuper(fs, daddr) 1416 1.32.2.2 perseant struct lfs *fs; 1417 1.32.2.2 perseant daddr_t daddr; 1418 1.32.2.2 perseant { 1419 1.32.2.2 perseant struct buf *bp; 1420 1.32.2.2 perseant dev_t i_dev; 1421 1.32.2.2 perseant int (*strategy) __P((void *)); 1422 1.32.2.2 perseant int s; 1423 1.32.2.2 perseant struct vop_strategy_args vop_strategy_a; 1424 1.32.2.2 perseant 1425 1.32.2.2 perseant #ifdef LFS_CANNOT_ROLLFW 1426 1.32.2.2 perseant /* 1427 1.32.2.2 perseant * If we can write one superblock while another is in 1428 1.32.2.2 perseant * progress, we risk not having a complete checkpoint if we crash. 1429 1.32.2.2 perseant * So, block here if a superblock write is in progress. 1430 1.32.2.2 perseant * 1431 1.32.2.2 perseant * XXX - should be a proper lock, not this hack 1432 1.32.2.2 perseant */ 1433 1.32.2.2 perseant while(fs->lfs_sbactive) { 1434 1.32.2.2 perseant tsleep(&fs->lfs_sbactive, PRIBIO+1, "lfs sb", 0); 1435 1.32.2.2 perseant } 1436 1.32.2.2 perseant fs->lfs_sbactive = daddr; 1437 1.32.2.2 perseant #endif 1438 1.32.2.2 perseant i_dev = VTOI(fs->lfs_ivnode)->i_dev; 1439 1.32.2.2 perseant strategy = VTOI(fs->lfs_ivnode)->i_devvp->v_op[VOFFSET(vop_strategy)]; 1440 1.32.2.2 perseant 1441 1.32.2.2 perseant /* Set timestamp of this version of the superblock */ 1442 1.32.2.2 perseant fs->lfs_tstamp = time.tv_sec; 1443 1.32.2.2 perseant 1444 1.32.2.2 perseant /* Checksum the superblock and copy it into a buffer. */ 1445 1.32.2.2 perseant fs->lfs_cksum = lfs_sb_cksum(&(fs->lfs_dlfs)); 1446 1.32.2.2 perseant bp = lfs_newbuf(VTOI(fs->lfs_ivnode)->i_devvp, daddr, LFS_SBPAD); 1447 1.32.2.2 perseant *(struct dlfs *)bp->b_data = fs->lfs_dlfs; 1448 1.32.2.2 perseant 1449 1.32.2.2 perseant bp->b_dev = i_dev; 1450 1.32.2.2 perseant bp->b_flags |= B_BUSY | B_CALL | B_ASYNC; 1451 1.32.2.2 perseant bp->b_flags &= ~(B_DONE | B_ERROR | B_READ | B_DELWRI); 1452 1.32.2.2 perseant bp->b_iodone = lfs_supercallback; 1453 1.32.2.2 perseant /* XXX KS - same nasty hack as above */ 1454 1.32.2.2 perseant bp->b_saveaddr = (caddr_t)fs; 1455 1.32.2.2 perseant 1456 1.32.2.2 perseant vop_strategy_a.a_desc = VDESC(vop_strategy); 1457 1.32.2.2 perseant vop_strategy_a.a_bp = bp; 1458 1.32.2.2 perseant s = splbio(); 1459 1.32.2.2 perseant ++bp->b_vp->v_numoutput; 1460 1.32.2.2 perseant splx(s); 1461 1.32.2.2 perseant (strategy)(&vop_strategy_a); 1462 1.32.2.2 perseant } 1463 1.32.2.2 perseant 1464 1.32.2.2 perseant /* 1465 1.32.2.2 perseant * Logical block number match routines used when traversing the dirty block 1466 1.32.2.2 perseant * chain. 1467 1.32.2.2 perseant */ 1468 1.32.2.2 perseant int 1469 1.32.2.2 perseant lfs_match_fake(fs, bp) 1470 1.32.2.2 perseant struct lfs *fs; 1471 1.32.2.2 perseant struct buf *bp; 1472 1.32.2.2 perseant { 1473 1.32.2.2 perseant return (bp->b_flags & B_CALL); 1474 1.32.2.2 perseant } 1475 1.32.2.2 perseant 1476 1.32.2.2 perseant int 1477 1.32.2.2 perseant lfs_match_data(fs, bp) 1478 1.32.2.2 perseant struct lfs *fs; 1479 1.32.2.2 perseant struct buf *bp; 1480 1.32.2.2 perseant { 1481 1.32.2.2 perseant return (bp->b_lblkno >= 0); 1482 1.32.2.2 perseant } 1483 1.32.2.2 perseant 1484 1.32.2.2 perseant int 1485 1.32.2.2 perseant lfs_match_indir(fs, bp) 1486 1.32.2.2 perseant struct lfs *fs; 1487 1.32.2.2 perseant struct buf *bp; 1488 1.32.2.2 perseant { 1489 1.32.2.2 perseant int lbn; 1490 1.32.2.2 perseant 1491 1.32.2.2 perseant lbn = bp->b_lblkno; 1492 1.32.2.2 perseant return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 0); 1493 1.32.2.2 perseant } 1494 1.32.2.2 perseant 1495 1.32.2.2 perseant int 1496 1.32.2.2 perseant lfs_match_dindir(fs, bp) 1497 1.32.2.2 perseant struct lfs *fs; 1498 1.32.2.2 perseant struct buf *bp; 1499 1.32.2.2 perseant { 1500 1.32.2.2 perseant int lbn; 1501 1.32.2.2 perseant 1502 1.32.2.2 perseant lbn = bp->b_lblkno; 1503 1.32.2.2 perseant return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 1); 1504 1.32.2.2 perseant } 1505 1.32.2.2 perseant 1506 1.32.2.2 perseant int 1507 1.32.2.2 perseant lfs_match_tindir(fs, bp) 1508 1.32.2.2 perseant struct lfs *fs; 1509 1.32.2.2 perseant struct buf *bp; 1510 1.32.2.2 perseant { 1511 1.32.2.2 perseant int lbn; 1512 1.32.2.2 perseant 1513 1.32.2.2 perseant lbn = bp->b_lblkno; 1514 1.32.2.2 perseant return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 2); 1515 1.32.2.2 perseant } 1516 1.32.2.2 perseant 1517 1.32.2.2 perseant /* 1518 1.32.2.2 perseant * XXX - The only buffers that are going to hit these functions are the 1519 1.32.2.2 perseant * segment write blocks, or the segment summaries, or the superblocks. 1520 1.32.2.2 perseant * 1521 1.32.2.2 perseant * All of the above are created by lfs_newbuf, and so do not need to be 1522 1.32.2.2 perseant * released via brelse. 1523 1.32.2.2 perseant */ 1524 1.32.2.2 perseant void 1525 1.32.2.2 perseant lfs_callback(bp) 1526 1.32.2.2 perseant struct buf *bp; 1527 1.32.2.2 perseant { 1528 1.32.2.2 perseant struct lfs *fs; 1529 1.32.2.2 perseant #ifdef LFS_TRACK_IOS 1530 1.32.2.2 perseant int j; 1531 1.32.2.2 perseant #endif 1532 1.32.2.2 perseant 1533 1.32.2.2 perseant fs = (struct lfs *)bp->b_saveaddr; 1534 1.32.2.2 perseant #ifdef DIAGNOSTIC 1535 1.32.2.2 perseant if (fs->lfs_iocount == 0) 1536 1.32.2.2 perseant panic("lfs_callback: zero iocount\n"); 1537 1.32.2.2 perseant #endif 1538 1.32.2.2 perseant if (--fs->lfs_iocount < LFS_THROTTLE) 1539 1.32.2.2 perseant wakeup(&fs->lfs_iocount); 1540 1.32.2.2 perseant #ifdef LFS_TRACK_IOS 1541 1.32.2.2 perseant for(j=0;j<LFS_THROTTLE;j++) { 1542 1.32.2.2 perseant if(fs->lfs_pending[j]==bp->b_blkno) { 1543 1.32.2.2 perseant fs->lfs_pending[j] = LFS_UNUSED_DADDR; 1544 1.32.2.2 perseant wakeup(&(fs->lfs_pending[j])); 1545 1.32.2.2 perseant break; 1546 1.32.2.2 perseant } 1547 1.32.2.2 perseant } 1548 1.32.2.2 perseant #endif /* LFS_TRACK_IOS */ 1549 1.32.2.2 perseant 1550 1.32.2.2 perseant lfs_freebuf(bp); 1551 1.32.2.2 perseant } 1552 1.32.2.2 perseant 1553 1.32.2.2 perseant void 1554 1.32.2.2 perseant lfs_supercallback(bp) 1555 1.32.2.2 perseant struct buf *bp; 1556 1.32.2.2 perseant { 1557 1.32.2.2 perseant #ifdef LFS_CANNOT_ROLLFW 1558 1.32.2.2 perseant struct lfs *fs; 1559 1.32.2.2 perseant 1560 1.32.2.2 perseant fs = (struct lfs *)bp->b_saveaddr; 1561 1.32.2.2 perseant fs->lfs_sbactive=NULL; 1562 1.32.2.2 perseant wakeup(&fs->lfs_sbactive); 1563 1.32.2.2 perseant #endif 1564 1.32.2.2 perseant lfs_freebuf(bp); 1565 1.32.2.2 perseant } 1566 1.32.2.2 perseant 1567 1.32.2.2 perseant /* 1568 1.32.2.2 perseant * Shellsort (diminishing increment sort) from Data Structures and 1569 1.32.2.2 perseant * Algorithms, Aho, Hopcraft and Ullman, 1983 Edition, page 290; 1570 1.32.2.2 perseant * see also Knuth Vol. 3, page 84. The increments are selected from 1571 1.32.2.2 perseant * formula (8), page 95. Roughly O(N^3/2). 1572 1.32.2.2 perseant */ 1573 1.32.2.2 perseant /* 1574 1.32.2.2 perseant * This is our own private copy of shellsort because we want to sort 1575 1.32.2.2 perseant * two parallel arrays (the array of buffer pointers and the array of 1576 1.32.2.2 perseant * logical block numbers) simultaneously. Note that we cast the array 1577 1.32.2.2 perseant * of logical block numbers to a unsigned in this routine so that the 1578 1.32.2.2 perseant * negative block numbers (meta data blocks) sort AFTER the data blocks. 1579 1.32.2.2 perseant */ 1580 1.32.2.2 perseant 1581 1.32.2.2 perseant void 1582 1.32.2.2 perseant lfs_shellsort(bp_array, lb_array, nmemb) 1583 1.32.2.2 perseant struct buf **bp_array; 1584 1.32.2.2 perseant ufs_daddr_t *lb_array; 1585 1.32.2.2 perseant register int nmemb; 1586 1.32.2.2 perseant { 1587 1.32.2.2 perseant static int __rsshell_increments[] = { 4, 1, 0 }; 1588 1.32.2.2 perseant register int incr, *incrp, t1, t2; 1589 1.32.2.2 perseant struct buf *bp_temp; 1590 1.32.2.2 perseant u_long lb_temp; 1591 1.32.2.2 perseant 1592 1.32.2.2 perseant for (incrp = __rsshell_increments; (incr = *incrp++) != 0;) 1593 1.32.2.2 perseant for (t1 = incr; t1 < nmemb; ++t1) 1594 1.32.2.2 perseant for (t2 = t1 - incr; t2 >= 0;) 1595 1.32.2.2 perseant if (lb_array[t2] > lb_array[t2 + incr]) { 1596 1.32.2.2 perseant lb_temp = lb_array[t2]; 1597 1.32.2.2 perseant lb_array[t2] = lb_array[t2 + incr]; 1598 1.32.2.2 perseant lb_array[t2 + incr] = lb_temp; 1599 1.32.2.2 perseant bp_temp = bp_array[t2]; 1600 1.32.2.2 perseant bp_array[t2] = bp_array[t2 + incr]; 1601 1.32.2.2 perseant bp_array[t2 + incr] = bp_temp; 1602 1.32.2.2 perseant t2 -= incr; 1603 1.32.2.2 perseant } else 1604 1.32.2.2 perseant break; 1605 1.32.2.2 perseant } 1606 1.32.2.2 perseant 1607 1.32.2.2 perseant /* 1608 1.32.2.2 perseant * Check VXLOCK. Return 1 if the vnode is locked. Otherwise, vget it. 1609 1.32.2.2 perseant */ 1610 1.32.2.2 perseant int 1611 1.32.2.2 perseant lfs_vref(vp) 1612 1.32.2.2 perseant register struct vnode *vp; 1613 1.32.2.2 perseant { 1614 1.32.2.2 perseant /* 1615 1.32.2.2 perseant * If we return 1 here during a flush, we risk vinvalbuf() not 1616 1.32.2.2 perseant * being able to flush all of the pages from this vnode, which 1617 1.32.2.2 perseant * will cause it to panic. So, return 0 if a flush is in progress. 1618 1.32.2.2 perseant */ 1619 1.32.2.2 perseant if (vp->v_flag & VXLOCK) { 1620 1.32.2.2 perseant if(IS_FLUSHING(VTOI(vp)->i_lfs,vp)) { 1621 1.32.2.2 perseant return 0; 1622 1.32.2.2 perseant } 1623 1.32.2.2 perseant return(1); 1624 1.32.2.2 perseant } 1625 1.32.2.2 perseant return (vget(vp, 0)); 1626 1.32.2.2 perseant } 1627 1.32.2.2 perseant 1628 1.32.2.2 perseant /* 1629 1.32.2.2 perseant * This is vrele except that we do not want to VOP_INACTIVE this vnode. We 1630 1.32.2.2 perseant * inline vrele here to avoid the vn_lock and VOP_INACTIVE call at the end. 1631 1.32.2.2 perseant */ 1632 1.32.2.2 perseant void 1633 1.32.2.2 perseant lfs_vunref(vp) 1634 1.32.2.2 perseant register struct vnode *vp; 1635 1.32.2.2 perseant { 1636 1.32.2.2 perseant /* 1637 1.32.2.2 perseant * Analogous to lfs_vref, if the node is flushing, fake it. 1638 1.32.2.2 perseant */ 1639 1.32.2.2 perseant if((vp->v_flag & VXLOCK) && IS_FLUSHING(VTOI(vp)->i_lfs,vp)) { 1640 1.32.2.2 perseant return; 1641 1.32.2.2 perseant } 1642 1.32.2.2 perseant 1643 1.32.2.2 perseant simple_lock(&vp->v_interlock); 1644 1.32.2.2 perseant #ifdef DIAGNOSTIC 1645 1.32.2.2 perseant if(vp->v_usecount<=0) { 1646 1.32.2.2 perseant printf("lfs_vunref: flags are 0x%lx\n", vp->v_flag); 1647 1.32.2.2 perseant printf("lfs_vunref: usecount = %ld\n", vp->v_usecount); 1648 1.32.2.2 perseant panic("lfs_vunref: v_usecount<0"); 1649 1.32.2.2 perseant } 1650 1.32.2.2 perseant #endif 1651 1.32.2.2 perseant vp->v_usecount--; 1652 1.32.2.2 perseant if (vp->v_usecount > 0) { 1653 1.32.2.2 perseant simple_unlock(&vp->v_interlock); 1654 1.32.2.2 perseant return; 1655 1.32.2.2 perseant } 1656 1.32.2.2 perseant #ifdef DIAGNOSTIC 1657 1.32.2.2 perseant if(VOP_ISLOCKED(vp)) 1658 1.32.2.2 perseant panic("lfs_vunref: vnode locked"); 1659 1.32.2.2 perseant #endif 1660 1.32.2.2 perseant /* 1661 1.32.2.2 perseant * insert at tail of LRU list 1662 1.32.2.2 perseant */ 1663 1.32.2.2 perseant simple_lock(&vnode_free_list_slock); 1664 1.32.2.2 perseant TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 1665 1.32.2.2 perseant simple_unlock(&vnode_free_list_slock); 1666 1.32.2.2 perseant simple_unlock(&vp->v_interlock); 1667 1.32.2.2 perseant } 1668 1.32.2.2 perseant 1669 1.32.2.2 perseant /* 1670 1.32.2.2 perseant * We use this when we have vnodes that were loaded in solely for cleaning. 1671 1.32.2.2 perseant * There is no reason to believe that these vnodes will be referenced again 1672 1.32.2.2 perseant * soon, since the cleaning process is unrelated to normal filesystem 1673 1.32.2.2 perseant * activity. Putting cleaned vnodes at the tail of the list has the effect 1674 1.32.2.2 perseant * of flushing the vnode LRU. So, put vnodes that were loaded only for 1675 1.32.2.2 perseant * cleaning at the head of the list, instead. 1676 1.32.2.2 perseant */ 1677 1.32.2.2 perseant void 1678 1.32.2.2 perseant lfs_vunref_head(vp) 1679 1.32.2.2 perseant register struct vnode *vp; 1680 1.32.2.2 perseant { 1681 1.32.2.2 perseant simple_lock(&vp->v_interlock); 1682 1.32.2.2 perseant #ifdef DIAGNOSTIC 1683 1.32.2.2 perseant if(vp->v_usecount==0) { 1684 1.32.2.2 perseant panic("lfs_vunref: v_usecount<0"); 1685 1.32.2.2 perseant } 1686 1.32.2.2 perseant #endif 1687 1.32.2.2 perseant vp->v_usecount--; 1688 1.32.2.2 perseant if (vp->v_usecount > 0) { 1689 1.32.2.2 perseant simple_unlock(&vp->v_interlock); 1690 1.32.2.2 perseant return; 1691 1.32.2.2 perseant } 1692 1.32.2.2 perseant #ifdef DIAGNOSTIC 1693 1.32.2.2 perseant if(VOP_ISLOCKED(vp)) 1694 1.32.2.2 perseant panic("lfs_vunref_head: vnode locked"); 1695 1.32.2.2 perseant #endif 1696 1.32.2.2 perseant /* 1697 1.32.2.2 perseant * insert at head of LRU list 1698 1.32.2.2 perseant */ 1699 1.32.2.2 perseant simple_lock(&vnode_free_list_slock); 1700 1.32.2.2 perseant TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 1701 1.32.2.2 perseant simple_unlock(&vnode_free_list_slock); 1702 1.32.2.2 perseant simple_unlock(&vp->v_interlock); 1703 1.32.2.2 perseant } 1704 1.32.2.2 perseant 1705
Indexes created Fri Oct 03 02:09:57 GMT 2025