1 /*- 2 * Copyright (c) 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * %sccs.include.redist.c% 6 * 7 * @(#)lfs_syscalls.c 8.4 (Berkeley) 01/13/94 8 */ 9 10 #include <sys/param.h> 11 #include <sys/proc.h> 12 #include <sys/buf.h> 13 #include <sys/mount.h> 14 #include <sys/vnode.h> 15 #include <sys/malloc.h> 16 #include <sys/kernel.h> 17 18 #include <ufs/ufs/quota.h> 19 #include <ufs/ufs/inode.h> 20 #include <ufs/ufs/ufsmount.h> 21 #include <ufs/ufs/ufs_extern.h> 22 23 #include <ufs/lfs/lfs.h> 24 #include <ufs/lfs/lfs_extern.h> 25 #define BUMP_FIP(SP) \ 26 (SP)->fip = (FINFO *) (&(SP)->fip->fi_blocks[(SP)->fip->fi_nblocks]) 27 28 #define INC_FINFO(SP) ++((SEGSUM *)((SP)->segsum))->ss_nfinfo 29 #define DEC_FINFO(SP) --((SEGSUM *)((SP)->segsum))->ss_nfinfo 30 31 /* 32 * Before committing to add something to a segment summary, make sure there 33 * is enough room. S is the bytes added to the summary. 34 */ 35 #define CHECK_SEG(s) \ 36 if (sp->sum_bytes_left < (s)) { \ 37 (void) lfs_writeseg(fs, sp); \ 38 } 39 struct buf *lfs_fakebuf __P((struct vnode *, int, size_t, caddr_t)); 40 41 /* 42 * lfs_markv: 43 * 44 * This will mark inodes and blocks dirty, so they are written into the log. 45 * It will block until all the blocks have been written. The segment create 46 * time passed in the block_info and inode_info structures is used to decide 47 * if the data is valid for each block (in case some process dirtied a block 48 * or inode that is being cleaned between the determination that a block is 49 * live and the lfs_markv call). 50 * 51 * 0 on success 52 * -1/errno is return on error. 53 */ 54 struct lfs_markv_args { 55 fsid_t *fsidp; /* file system */ 56 BLOCK_INFO *blkiov; /* block array */ 57 int blkcnt; /* count of block array entries */ 58 }; 59 int 60 lfs_markv(p, uap, retval) 61 struct proc *p; 62 struct lfs_markv_args *uap; 63 int *retval; 64 { 65 struct segment *sp; 66 BLOCK_INFO *blkp; 67 IFILE *ifp; 68 struct buf *bp, **bpp; 69 struct inode *ip; 70 struct lfs *fs; 71 struct mount *mntp; 72 struct vnode *vp; 73 fsid_t fsid; 74 void *start; 75 ino_t lastino; 76 daddr_t b_daddr, v_daddr; 77 u_long bsize; 78 int cnt, error; 79 80 if (error = suser(p->p_ucred, &p->p_acflag)) 81 return (error); 82 83 if (error = copyin(uap->fsidp, &fsid, sizeof(fsid_t))) 84 return (error); 85 if ((mntp = getvfs(&fsid)) == NULL) 86 return (EINVAL); 87 88 cnt = uap->blkcnt; 89 start = malloc(cnt * sizeof(BLOCK_INFO), M_SEGMENT, M_WAITOK); 90 if (error = copyin(uap->blkiov, start, cnt * sizeof(BLOCK_INFO))) 91 goto err1; 92 93 /* Mark blocks/inodes dirty. */ 94 fs = VFSTOUFS(mntp)->um_lfs; 95 bsize = fs->lfs_bsize; 96 error = 0; 97 98 lfs_seglock(fs, SEGM_SYNC | SEGM_CLEAN); 99 sp = fs->lfs_sp; 100 for (v_daddr = LFS_UNUSED_DADDR, lastino = LFS_UNUSED_INUM, 101 blkp = start; cnt--; ++blkp) { 102 /* 103 * Get the IFILE entry (only once) and see if the file still 104 * exists. 105 */ 106 if (lastino != blkp->bi_inode) { 107 if (lastino != LFS_UNUSED_INUM) { 108 /* Finish up last file */ 109 lfs_updatemeta(sp); 110 lfs_writeinode(fs, sp, ip); 111 lfs_vunref(vp); 112 if (sp->fip->fi_nblocks) 113 BUMP_FIP(sp); 114 else { 115 DEC_FINFO(sp); 116 sp->sum_bytes_left += 117 sizeof(FINFO) - sizeof(daddr_t); 118 119 } 120 } 121 122 /* Start a new file */ 123 CHECK_SEG(sizeof(FINFO)); 124 sp->sum_bytes_left -= sizeof(FINFO) - sizeof(daddr_t); 125 INC_FINFO(sp); 126 sp->start_lbp = &sp->fip->fi_blocks[0]; 127 sp->vp = NULL; 128 sp->fip->fi_version = blkp->bi_version; 129 sp->fip->fi_nblocks = 0; 130 sp->fip->fi_ino = blkp->bi_inode; 131 lastino = blkp->bi_inode; 132 if (blkp->bi_inode == LFS_IFILE_INUM) 133 v_daddr = fs->lfs_idaddr; 134 else { 135 LFS_IENTRY(ifp, fs, blkp->bi_inode, bp); 136 v_daddr = ifp->if_daddr; 137 brelse(bp); 138 } 139 if (v_daddr == LFS_UNUSED_DADDR) 140 continue; 141 142 /* Get the vnode/inode. */ 143 if (lfs_fastvget(mntp, blkp->bi_inode, v_daddr, &vp, 144 blkp->bi_lbn == LFS_UNUSED_LBN ? 145 blkp->bi_bp : NULL)) { 146 #ifdef DIAGNOSTIC 147 printf("lfs_markv: VFS_VGET failed (%d)\n", 148 blkp->bi_inode); 149 #endif 150 lastino = LFS_UNUSED_INUM; 151 v_daddr = LFS_UNUSED_DADDR; 152 continue; 153 } 154 sp->vp = vp; 155 ip = VTOI(vp); 156 } else if (v_daddr == LFS_UNUSED_DADDR) 157 continue; 158 159 /* If this BLOCK_INFO didn't contain a block, keep going. */ 160 if (blkp->bi_lbn == LFS_UNUSED_LBN) 161 continue; 162 if (VOP_BMAP(vp, blkp->bi_lbn, NULL, &b_daddr, NULL) || 163 b_daddr != blkp->bi_daddr) 164 continue; 165 /* 166 * If we got to here, then we are keeping the block. If it 167 * is an indirect block, we want to actually put it in the 168 * buffer cache so that it can be updated in the finish_meta 169 * section. If it's not, we need to allocate a fake buffer 170 * so that writeseg can perform the copyin and write the buffer. 171 */ 172 if (blkp->bi_lbn >= 0) /* Data Block */ 173 bp = lfs_fakebuf(vp, blkp->bi_lbn, bsize, 174 blkp->bi_bp); 175 else { 176 bp = getblk(vp, blkp->bi_lbn, bsize, 0, 0); 177 if (!(bp->b_flags & (B_DELWRI | B_DONE | B_CACHE)) && 178 (error = copyin(blkp->bi_bp, bp->b_data, 179 bsize))) 180 goto err2; 181 if (error = VOP_BWRITE(bp)) 182 goto err2; 183 } 184 while (lfs_gatherblock(sp, bp, NULL)); 185 } 186 if (sp->vp) { 187 lfs_updatemeta(sp); 188 lfs_writeinode(fs, sp, ip); 189 lfs_vunref(vp); 190 if (!sp->fip->fi_nblocks) { 191 DEC_FINFO(sp); 192 sp->sum_bytes_left += sizeof(FINFO) - sizeof(daddr_t); 193 } 194 } 195 (void) lfs_writeseg(fs, sp); 196 lfs_segunlock(fs); 197 free(start, M_SEGMENT); 198 return (error); 199 /* 200 * XXX If we come in to error 2, we might have indirect blocks that were 201 * updated and now have bad block pointers. I don't know what to do 202 * about this. 203 */ 204 205 err2: lfs_vunref(vp); 206 /* Free up fakebuffers */ 207 for (bpp = --sp->cbpp; bpp >= sp->bpp; --bpp) 208 if ((*bpp)->b_flags & B_CALL) { 209 brelvp(*bpp); 210 free(*bpp, M_SEGMENT); 211 } else 212 brelse(*bpp); 213 lfs_segunlock(fs); 214 err1: 215 free(start, M_SEGMENT); 216 return(error); 217 } 218 219 /* 220 * lfs_bmapv: 221 * 222 * This will fill in the current disk address for arrays of blocks. 223 * 224 * 0 on success 225 * -1/errno is return on error. 226 */ 227 struct lfs_bmapv_args { 228 fsid_t *fsidp; /* file system */ 229 BLOCK_INFO *blkiov; /* block array */ 230 int blkcnt; /* count of block array entries */ 231 }; 232 int 233 lfs_bmapv(p, uap, retval) 234 struct proc *p; 235 struct lfs_bmapv_args *uap; 236 int *retval; 237 { 238 BLOCK_INFO *blkp; 239 struct mount *mntp; 240 struct vnode *vp; 241 fsid_t fsid; 242 void *start; 243 daddr_t daddr; 244 int cnt, error, step; 245 246 if (error = suser(p->p_ucred, &p->p_acflag)) 247 return (error); 248 249 if (error = copyin(uap->fsidp, &fsid, sizeof(fsid_t))) 250 return (error); 251 if ((mntp = getvfs(&fsid)) == NULL) 252 return (EINVAL); 253 254 cnt = uap->blkcnt; 255 start = blkp = malloc(cnt * sizeof(BLOCK_INFO), M_SEGMENT, M_WAITOK); 256 if (error = copyin(uap->blkiov, blkp, cnt * sizeof(BLOCK_INFO))) { 257 free(blkp, M_SEGMENT); 258 return (error); 259 } 260 261 for (step = cnt; step--; ++blkp) { 262 if (blkp->bi_lbn == LFS_UNUSED_LBN) 263 continue; 264 /* Could be a deadlock ? */ 265 if (VFS_VGET(mntp, blkp->bi_inode, &vp)) 266 daddr = LFS_UNUSED_DADDR; 267 else { 268 if (VOP_BMAP(vp, blkp->bi_lbn, NULL, &daddr, NULL)) 269 daddr = LFS_UNUSED_DADDR; 270 vput(vp); 271 } 272 blkp->bi_daddr = daddr; 273 } 274 copyout(start, uap->blkiov, cnt * sizeof(BLOCK_INFO)); 275 free(start, M_SEGMENT); 276 return (0); 277 } 278 279 /* 280 * lfs_segclean: 281 * 282 * Mark the segment clean. 283 * 284 * 0 on success 285 * -1/errno is return on error. 286 */ 287 struct lfs_segclean_args { 288 fsid_t *fsidp; /* file system */ 289 u_long segment; /* segment number */ 290 }; 291 int 292 lfs_segclean(p, uap, retval) 293 struct proc *p; 294 struct lfs_segclean_args *uap; 295 int *retval; 296 { 297 CLEANERINFO *cip; 298 SEGUSE *sup; 299 struct buf *bp; 300 struct mount *mntp; 301 struct lfs *fs; 302 fsid_t fsid; 303 int error; 304 305 if (error = suser(p->p_ucred, &p->p_acflag)) 306 return (error); 307 308 if (error = copyin(uap->fsidp, &fsid, sizeof(fsid_t))) 309 return (error); 310 if ((mntp = getvfs(&fsid)) == NULL) 311 return (EINVAL); 312 313 fs = VFSTOUFS(mntp)->um_lfs; 314 315 if (datosn(fs, fs->lfs_curseg) == uap->segment) 316 return (EBUSY); 317 318 LFS_SEGENTRY(sup, fs, uap->segment, bp); 319 if (sup->su_flags & SEGUSE_ACTIVE) { 320 brelse(bp); 321 return(EBUSY); 322 } 323 fs->lfs_avail += fsbtodb(fs, fs->lfs_ssize) - 1; 324 fs->lfs_bfree += (sup->su_nsums * LFS_SUMMARY_SIZE / DEV_BSIZE) + 325 sup->su_ninos * btodb(fs->lfs_bsize); 326 sup->su_flags &= ~SEGUSE_DIRTY; 327 (void) VOP_BWRITE(bp); 328 329 LFS_CLEANERINFO(cip, fs, bp); 330 ++cip->clean; 331 --cip->dirty; 332 (void) VOP_BWRITE(bp); 333 wakeup(&fs->lfs_avail); 334 return (0); 335 } 336 337 /* 338 * lfs_segwait: 339 * 340 * This will block until a segment in file system fsid is written. A timeout 341 * in milliseconds may be specified which will awake the cleaner automatically. 342 * An fsid of -1 means any file system, and a timeout of 0 means forever. 343 * 344 * 0 on success 345 * 1 on timeout 346 * -1/errno is return on error. 347 */ 348 struct lfs_segwait_args { 349 fsid_t *fsidp; /* file system */ 350 struct timeval *tv; /* timeout */ 351 }; 352 int 353 lfs_segwait(p, uap, retval) 354 struct proc *p; 355 struct lfs_segwait_args *uap; 356 int *retval; 357 { 358 extern int lfs_allclean_wakeup; 359 struct mount *mntp; 360 struct timeval atv; 361 fsid_t fsid; 362 void *addr; 363 u_long timeout; 364 int error, s; 365 366 if (error = suser(p->p_ucred, &p->p_acflag)) { 367 return (error); 368 } 369 #ifdef WHEN_QUADS_WORK 370 if (error = copyin(uap->fsidp, &fsid, sizeof(fsid_t))) 371 return (error); 372 if (fsid == (fsid_t)-1) 373 addr = &lfs_allclean_wakeup; 374 else { 375 if ((mntp = getvfs(&fsid)) == NULL) 376 return (EINVAL); 377 addr = &VFSTOUFS(mntp)->um_lfs->lfs_nextseg; 378 } 379 #else 380 if (error = copyin(uap->fsidp, &fsid, sizeof(fsid_t))) 381 return (error); 382 if ((mntp = getvfs(&fsid)) == NULL) 383 addr = &lfs_allclean_wakeup; 384 else 385 addr = &VFSTOUFS(mntp)->um_lfs->lfs_nextseg; 386 #endif 387 388 if (uap->tv) { 389 if (error = copyin(uap->tv, &atv, sizeof(struct timeval))) 390 return (error); 391 if (itimerfix(&atv)) 392 return (EINVAL); 393 s = splclock(); 394 timevaladd(&atv, (struct timeval *)&time); 395 timeout = hzto(&atv); 396 splx(s); 397 } else 398 timeout = 0; 399 400 error = tsleep(addr, PCATCH | PUSER, "segment", timeout); 401 return (error == ERESTART ? EINTR : 0); 402 } 403 404 /* 405 * VFS_VGET call specialized for the cleaner. The cleaner already knows the 406 * daddr from the ifile, so don't look it up again. If the cleaner is 407 * processing IINFO structures, it may have the ondisk inode already, so 408 * don't go retrieving it again. 409 */ 410 int 411 lfs_fastvget(mp, ino, daddr, vpp, dinp) 412 struct mount *mp; 413 ino_t ino; 414 daddr_t daddr; 415 struct vnode **vpp; 416 struct dinode *dinp; 417 { 418 register struct inode *ip; 419 struct vnode *vp; 420 struct ufsmount *ump; 421 struct buf *bp; 422 dev_t dev; 423 int error; 424 425 ump = VFSTOUFS(mp); 426 dev = ump->um_dev; 427 /* 428 * This is playing fast and loose. Someone may have the inode 429 * locked, in which case they are going to be distinctly unhappy 430 * if we trash something. 431 */ 432 if ((*vpp = ufs_ihashlookup(dev, ino)) != NULL) { 433 lfs_vref(*vpp); 434 if ((*vpp)->v_flag & VXLOCK) 435 printf ("Cleaned vnode VXLOCKED\n"); 436 ip = VTOI(*vpp); 437 if (ip->i_flags & IN_LOCKED) 438 printf("cleaned vnode locked\n"); 439 if (!(ip->i_flag & IN_MODIFIED)) { 440 ++ump->um_lfs->lfs_uinodes; 441 ip->i_flag |= IN_MODIFIED; 442 } 443 ip->i_flag |= IN_MODIFIED; 444 return (0); 445 } 446 447 /* Allocate new vnode/inode. */ 448 if (error = lfs_vcreate(mp, ino, &vp)) { 449 *vpp = NULL; 450 return (error); 451 } 452 453 /* 454 * Put it onto its hash chain and lock it so that other requests for 455 * this inode will block if they arrive while we are sleeping waiting 456 * for old data structures to be purged or for the contents of the 457 * disk portion of this inode to be read. 458 */ 459 ip = VTOI(vp); 460 ufs_ihashins(ip); 461 462 /* 463 * XXX 464 * This may not need to be here, logically it should go down with 465 * the i_devvp initialization. 466 * Ask Kirk. 467 */ 468 ip->i_lfs = ump->um_lfs; 469 470 /* Read in the disk contents for the inode, copy into the inode. */ 471 if (dinp) 472 if (error = copyin(dinp, &ip->i_din, sizeof(struct dinode))) 473 return (error); 474 else { 475 if (error = bread(ump->um_devvp, daddr, 476 (int)ump->um_lfs->lfs_bsize, NOCRED, &bp)) { 477 /* 478 * The inode does not contain anything useful, so it 479 * would be misleading to leave it on its hash chain. 480 * Iput() will return it to the free list. 481 */ 482 ufs_ihashrem(ip); 483 484 /* Unlock and discard unneeded inode. */ 485 lfs_vunref(vp); 486 brelse(bp); 487 *vpp = NULL; 488 return (error); 489 } 490 ip->i_din = 491 *lfs_ifind(ump->um_lfs, ino, (struct dinode *)bp->b_data); 492 brelse(bp); 493 } 494 495 /* Inode was just read from user space or disk, make sure it's locked */ 496 ip->i_flag |= IN_LOCKED; 497 498 /* 499 * Initialize the vnode from the inode, check for aliases. In all 500 * cases re-init ip, the underlying vnode/inode may have changed. 501 */ 502 if (error = ufs_vinit(mp, lfs_specop_p, LFS_FIFOOPS, &vp)) { 503 lfs_vunref(vp); 504 *vpp = NULL; 505 return (error); 506 } 507 /* 508 * Finish inode initialization now that aliasing has been resolved. 509 */ 510 ip->i_devvp = ump->um_devvp; 511 ip->i_flag |= IN_MODIFIED; 512 ++ump->um_lfs->lfs_uinodes; 513 VREF(ip->i_devvp); 514 *vpp = vp; 515 return (0); 516 } 517 struct buf * 518 lfs_fakebuf(vp, lbn, size, uaddr) 519 struct vnode *vp; 520 int lbn; 521 size_t size; 522 caddr_t uaddr; 523 { 524 struct buf *bp; 525 526 bp = lfs_newbuf(vp, lbn, 0); 527 bp->b_saveaddr = uaddr; 528 bp->b_bufsize = size; 529 bp->b_bcount = size; 530 bp->b_flags |= B_INVAL; 531 return(bp); 532 } 533