lfs_vnops.c revision 1.132.6.1
1 /* $NetBSD: lfs_vnops.c,v 1.132.6.1 2005/02/12 18:17:56 yamt Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant (at) hhhh.org>. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 /* 39 * Copyright (c) 1986, 1989, 1991, 1993, 1995 40 * The Regents of the University of California. All rights reserved. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)lfs_vnops.c 8.13 (Berkeley) 6/10/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.132.6.1 2005/02/12 18:17:56 yamt Exp $"); 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/namei.h> 75 #include <sys/resourcevar.h> 76 #include <sys/kernel.h> 77 #include <sys/file.h> 78 #include <sys/stat.h> 79 #include <sys/buf.h> 80 #include <sys/proc.h> 81 #include <sys/mount.h> 82 #include <sys/vnode.h> 83 #include <sys/malloc.h> 84 #include <sys/pool.h> 85 #include <sys/signalvar.h> 86 87 #include <miscfs/fifofs/fifo.h> 88 #include <miscfs/genfs/genfs.h> 89 #include <miscfs/specfs/specdev.h> 90 91 #include <ufs/ufs/inode.h> 92 #include <ufs/ufs/dir.h> 93 #include <ufs/ufs/ufsmount.h> 94 #include <ufs/ufs/ufs_extern.h> 95 96 #include <uvm/uvm.h> 97 #include <uvm/uvm_pmap.h> 98 #include <uvm/uvm_stat.h> 99 #include <uvm/uvm_pager.h> 100 101 #include <ufs/lfs/lfs.h> 102 #include <ufs/lfs/lfs_extern.h> 103 104 extern pid_t lfs_writer_daemon; 105 106 /* Global vfs data structures for lfs. */ 107 int (**lfs_vnodeop_p)(void *); 108 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 109 { &vop_default_desc, vn_default_error }, 110 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 111 { &vop_create_desc, lfs_create }, /* create */ 112 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 113 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 114 { &vop_open_desc, ufs_open }, /* open */ 115 { &vop_close_desc, lfs_close }, /* close */ 116 { &vop_access_desc, ufs_access }, /* access */ 117 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 118 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 119 { &vop_read_desc, lfs_read }, /* read */ 120 { &vop_write_desc, lfs_write }, /* write */ 121 { &vop_lease_desc, ufs_lease_check }, /* lease */ 122 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 123 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 124 { &vop_poll_desc, ufs_poll }, /* poll */ 125 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 126 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 127 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 128 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 129 { &vop_seek_desc, ufs_seek }, /* seek */ 130 { &vop_remove_desc, lfs_remove }, /* remove */ 131 { &vop_link_desc, lfs_link }, /* link */ 132 { &vop_rename_desc, lfs_rename }, /* rename */ 133 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 134 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 135 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 136 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 137 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 138 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 139 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 140 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 141 { &vop_lock_desc, ufs_lock }, /* lock */ 142 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 143 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 144 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 145 { &vop_print_desc, ufs_print }, /* print */ 146 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 147 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 148 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 149 { &vop_blkatoff_desc, lfs_blkatoff }, /* blkatoff */ 150 { &vop_valloc_desc, lfs_valloc }, /* valloc */ 151 { &vop_balloc_desc, lfs_balloc }, /* balloc */ 152 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 153 { &vop_truncate_desc, lfs_truncate }, /* truncate */ 154 { &vop_update_desc, lfs_update }, /* update */ 155 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 156 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 157 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 158 { NULL, NULL } 159 }; 160 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 161 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 162 163 int (**lfs_specop_p)(void *); 164 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 165 { &vop_default_desc, vn_default_error }, 166 { &vop_lookup_desc, spec_lookup }, /* lookup */ 167 { &vop_create_desc, spec_create }, /* create */ 168 { &vop_mknod_desc, spec_mknod }, /* mknod */ 169 { &vop_open_desc, spec_open }, /* open */ 170 { &vop_close_desc, lfsspec_close }, /* close */ 171 { &vop_access_desc, ufs_access }, /* access */ 172 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 173 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 174 { &vop_read_desc, ufsspec_read }, /* read */ 175 { &vop_write_desc, ufsspec_write }, /* write */ 176 { &vop_lease_desc, spec_lease_check }, /* lease */ 177 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 178 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 179 { &vop_poll_desc, spec_poll }, /* poll */ 180 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 181 { &vop_revoke_desc, spec_revoke }, /* revoke */ 182 { &vop_mmap_desc, spec_mmap }, /* mmap */ 183 { &vop_fsync_desc, spec_fsync }, /* fsync */ 184 { &vop_seek_desc, spec_seek }, /* seek */ 185 { &vop_remove_desc, spec_remove }, /* remove */ 186 { &vop_link_desc, spec_link }, /* link */ 187 { &vop_rename_desc, spec_rename }, /* rename */ 188 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 189 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 190 { &vop_symlink_desc, spec_symlink }, /* symlink */ 191 { &vop_readdir_desc, spec_readdir }, /* readdir */ 192 { &vop_readlink_desc, spec_readlink }, /* readlink */ 193 { &vop_abortop_desc, spec_abortop }, /* abortop */ 194 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 195 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 196 { &vop_lock_desc, ufs_lock }, /* lock */ 197 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 198 { &vop_bmap_desc, spec_bmap }, /* bmap */ 199 { &vop_strategy_desc, spec_strategy }, /* strategy */ 200 { &vop_print_desc, ufs_print }, /* print */ 201 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 202 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 203 { &vop_advlock_desc, spec_advlock }, /* advlock */ 204 { &vop_blkatoff_desc, spec_blkatoff }, /* blkatoff */ 205 { &vop_valloc_desc, spec_valloc }, /* valloc */ 206 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 207 { &vop_truncate_desc, spec_truncate }, /* truncate */ 208 { &vop_update_desc, lfs_update }, /* update */ 209 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 210 { &vop_getpages_desc, spec_getpages }, /* getpages */ 211 { &vop_putpages_desc, spec_putpages }, /* putpages */ 212 { NULL, NULL } 213 }; 214 const struct vnodeopv_desc lfs_specop_opv_desc = 215 { &lfs_specop_p, lfs_specop_entries }; 216 217 int (**lfs_fifoop_p)(void *); 218 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 219 { &vop_default_desc, vn_default_error }, 220 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 221 { &vop_create_desc, fifo_create }, /* create */ 222 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 223 { &vop_open_desc, fifo_open }, /* open */ 224 { &vop_close_desc, lfsfifo_close }, /* close */ 225 { &vop_access_desc, ufs_access }, /* access */ 226 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 227 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 228 { &vop_read_desc, ufsfifo_read }, /* read */ 229 { &vop_write_desc, ufsfifo_write }, /* write */ 230 { &vop_lease_desc, fifo_lease_check }, /* lease */ 231 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 232 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 233 { &vop_poll_desc, fifo_poll }, /* poll */ 234 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 235 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 236 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 237 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 238 { &vop_seek_desc, fifo_seek }, /* seek */ 239 { &vop_remove_desc, fifo_remove }, /* remove */ 240 { &vop_link_desc, fifo_link }, /* link */ 241 { &vop_rename_desc, fifo_rename }, /* rename */ 242 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 243 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 244 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 245 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 246 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 247 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 248 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 249 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 250 { &vop_lock_desc, ufs_lock }, /* lock */ 251 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 252 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 253 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 254 { &vop_print_desc, ufs_print }, /* print */ 255 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 256 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 257 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 258 { &vop_blkatoff_desc, fifo_blkatoff }, /* blkatoff */ 259 { &vop_valloc_desc, fifo_valloc }, /* valloc */ 260 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 261 { &vop_truncate_desc, fifo_truncate }, /* truncate */ 262 { &vop_update_desc, lfs_update }, /* update */ 263 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 264 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 265 { NULL, NULL } 266 }; 267 const struct vnodeopv_desc lfs_fifoop_opv_desc = 268 { &lfs_fifoop_p, lfs_fifoop_entries }; 269 270 /* 271 * A function version of LFS_ITIMES, for the UFS functions which call ITIMES 272 */ 273 void 274 lfs_itimes(struct inode *ip, struct timespec *acc, struct timespec *mod, struct timespec *cre) 275 { 276 LFS_ITIMES(ip, acc, mod, cre); 277 } 278 279 #define LFS_READWRITE 280 #include <ufs/ufs/ufs_readwrite.c> 281 #undef LFS_READWRITE 282 283 /* 284 * Synch an open file. 285 */ 286 /* ARGSUSED */ 287 int 288 lfs_fsync(void *v) 289 { 290 struct vop_fsync_args /* { 291 struct vnode *a_vp; 292 struct ucred *a_cred; 293 int a_flags; 294 off_t offlo; 295 off_t offhi; 296 struct proc *a_p; 297 } */ *ap = v; 298 struct vnode *vp = ap->a_vp; 299 int error, wait; 300 301 /* 302 * Trickle sync checks for need to do a checkpoint after possible 303 * activity from the pagedaemon. 304 */ 305 if (ap->a_flags & FSYNC_LAZY) { 306 simple_lock(&lfs_subsys_lock); 307 wakeup(&lfs_writer_daemon); 308 simple_unlock(&lfs_subsys_lock); 309 return 0; 310 } 311 312 wait = (ap->a_flags & FSYNC_WAIT); 313 simple_lock(&vp->v_interlock); 314 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 315 round_page(ap->a_offhi), 316 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 317 if (error) 318 return error; 319 error = VOP_UPDATE(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 320 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 321 int l = 0; 322 error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE, 323 ap->a_p->p_ucred, ap->a_p); 324 } 325 if (wait && !VPISEMPTY(vp)) 326 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 327 328 return error; 329 } 330 331 /* 332 * Take IN_ADIROP off, then call ufs_inactive. 333 */ 334 int 335 lfs_inactive(void *v) 336 { 337 struct vop_inactive_args /* { 338 struct vnode *a_vp; 339 struct proc *a_p; 340 } */ *ap = v; 341 342 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 343 344 lfs_unmark_vnode(ap->a_vp); 345 346 /* 347 * The Ifile is only ever inactivated on unmount. 348 * Streamline this process by not giving it more dirty blocks. 349 */ 350 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 351 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 352 VOP_UNLOCK(ap->a_vp, 0); 353 return 0; 354 } 355 356 return ufs_inactive(v); 357 } 358 359 /* 360 * These macros are used to bracket UFS directory ops, so that we can 361 * identify all the pages touched during directory ops which need to 362 * be ordered and flushed atomically, so that they may be recovered. 363 */ 364 /* 365 * XXX KS - Because we have to mark nodes VDIROP in order to prevent 366 * the cache from reclaiming them while a dirop is in progress, we must 367 * also manage the number of nodes so marked (otherwise we can run out). 368 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 369 * is decremented during segment write, when VDIROP is taken off. 370 */ 371 #define SET_DIROP(vp) SET_DIROP2((vp), NULL) 372 #define SET_DIROP2(vp, vp2) lfs_set_dirop((vp), (vp2)) 373 static int lfs_set_dirop(struct vnode *, struct vnode *); 374 375 #define NRESERVE(fs) (btofsb(fs, (NIADDR + 3 + (2 * NIADDR + 3)) << fs->lfs_bshift)) 376 377 static int 378 lfs_set_dirop(struct vnode *vp, struct vnode *vp2) 379 { 380 struct lfs *fs; 381 int error; 382 383 KASSERT(VOP_ISLOCKED(vp)); 384 KASSERT(vp2 == NULL || VOP_ISLOCKED(vp2)); 385 386 fs = VTOI(vp)->i_lfs; 387 /* 388 * We might need one directory block plus supporting indirect blocks, 389 * plus an inode block and ifile page for the new vnode. 390 */ 391 if ((error = lfs_reserve(fs, vp, vp2, NRESERVE(fs))) != 0) 392 return (error); 393 394 if (fs->lfs_dirops == 0) 395 lfs_check(vp, LFS_UNUSED_LBN, 0); 396 restart: 397 simple_lock(&fs->lfs_interlock); 398 if (fs->lfs_writer) { 399 ltsleep(&fs->lfs_dirops, (PRIBIO + 1) | PNORELOCK, 400 "lfs_sdirop", 0, &fs->lfs_interlock); 401 goto restart; 402 } 403 simple_lock(&lfs_subsys_lock); 404 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 405 wakeup(&lfs_writer_daemon); 406 simple_unlock(&lfs_subsys_lock); 407 simple_unlock(&fs->lfs_interlock); 408 preempt(1); 409 goto restart; 410 } 411 412 if (lfs_dirvcount > LFS_MAX_DIROP) { 413 simple_unlock(&fs->lfs_interlock); 414 #ifdef DEBUG_LFS 415 printf("lfs_set_dirop: sleeping with dirops=%d, " 416 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount); 417 #endif 418 if ((error = ltsleep(&lfs_dirvcount, 419 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 420 &lfs_subsys_lock)) != 0) { 421 goto unreserve; 422 } 423 goto restart; 424 } 425 simple_unlock(&lfs_subsys_lock); 426 427 ++fs->lfs_dirops; 428 fs->lfs_doifile = 1; 429 simple_unlock(&fs->lfs_interlock); 430 431 /* Hold a reference so SET_ENDOP will be happy */ 432 vref(vp); 433 if (vp2) 434 vref(vp2); 435 436 return 0; 437 438 unreserve: 439 lfs_reserve(fs, vp, vp2, -NRESERVE(fs)); 440 return error; 441 } 442 443 #define SET_ENDOP(fs, vp, str) SET_ENDOP2((fs), (vp), NULL, (str)) 444 #define SET_ENDOP2(fs, vp, vp2, str) { \ 445 --(fs)->lfs_dirops; \ 446 if (!(fs)->lfs_dirops) { \ 447 if ((fs)->lfs_nadirop) { \ 448 panic("SET_ENDOP: %s: no dirops but nadirop=%d", \ 449 (str), (fs)->lfs_nadirop); \ 450 } \ 451 wakeup(&(fs)->lfs_writer); \ 452 lfs_check((vp),LFS_UNUSED_LBN,0); \ 453 } \ 454 lfs_reserve((fs), vp, vp2, -NRESERVE(fs)); /* XXX */ \ 455 vrele(vp); \ 456 if (vp2) \ 457 vrele(vp2); \ 458 } 459 460 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 461 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 462 463 void 464 lfs_mark_vnode(struct vnode *vp) 465 { 466 struct inode *ip = VTOI(vp); 467 struct lfs *fs = ip->i_lfs; 468 469 if (!(ip->i_flag & IN_ADIROP)) { 470 if (!(vp->v_flag & VDIROP)) { 471 (void)lfs_vref(vp); 472 ++lfs_dirvcount; 473 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 474 vp->v_flag |= VDIROP; 475 } 476 ++fs->lfs_nadirop; 477 ip->i_flag |= IN_ADIROP; 478 } else 479 KASSERT(vp->v_flag & VDIROP); 480 } 481 482 void 483 lfs_unmark_vnode(struct vnode *vp) 484 { 485 struct inode *ip = VTOI(vp); 486 487 if (ip->i_flag & IN_ADIROP) { 488 KASSERT(vp->v_flag & VDIROP); 489 --ip->i_lfs->lfs_nadirop; 490 ip->i_flag &= ~IN_ADIROP; 491 } 492 } 493 494 int 495 lfs_symlink(void *v) 496 { 497 struct vop_symlink_args /* { 498 struct vnode *a_dvp; 499 struct vnode **a_vpp; 500 struct componentname *a_cnp; 501 struct vattr *a_vap; 502 char *a_target; 503 } */ *ap = v; 504 int error; 505 506 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 507 vput(ap->a_dvp); 508 return error; 509 } 510 MARK_VNODE(ap->a_dvp); 511 error = ufs_symlink(ap); 512 UNMARK_VNODE(ap->a_dvp); 513 if (*(ap->a_vpp)) 514 UNMARK_VNODE(*(ap->a_vpp)); 515 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"symlink"); 516 return (error); 517 } 518 519 int 520 lfs_mknod(void *v) 521 { 522 struct vop_mknod_args /* { 523 struct vnode *a_dvp; 524 struct vnode **a_vpp; 525 struct componentname *a_cnp; 526 struct vattr *a_vap; 527 } */ *ap = v; 528 struct vattr *vap = ap->a_vap; 529 struct vnode **vpp = ap->a_vpp; 530 struct inode *ip; 531 int error; 532 struct mount *mp; 533 ino_t ino; 534 535 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 536 vput(ap->a_dvp); 537 return error; 538 } 539 MARK_VNODE(ap->a_dvp); 540 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 541 ap->a_dvp, vpp, ap->a_cnp); 542 UNMARK_VNODE(ap->a_dvp); 543 if (*(ap->a_vpp)) 544 UNMARK_VNODE(*(ap->a_vpp)); 545 546 /* Either way we're done with the dirop at this point */ 547 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"mknod"); 548 549 if (error) 550 return (error); 551 552 ip = VTOI(*vpp); 553 mp = (*vpp)->v_mount; 554 ino = ip->i_number; 555 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 556 if (vap->va_rdev != VNOVAL) { 557 /* 558 * Want to be able to use this to make badblock 559 * inodes, so don't truncate the dev number. 560 */ 561 #if 0 562 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 563 UFS_MPNEEDSWAP((*vpp)->v_mount)); 564 #else 565 ip->i_ffs1_rdev = vap->va_rdev; 566 #endif 567 } 568 /* 569 * Call fsync to write the vnode so that we don't have to deal with 570 * flushing it when it's marked VDIROP|VXLOCK. 571 * 572 * XXX KS - If we can't flush we also can't call vgone(), so must 573 * return. But, that leaves this vnode in limbo, also not good. 574 * Can this ever happen (barring hardware failure)? 575 */ 576 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 577 curproc)) != 0) { 578 printf("Couldn't fsync in mknod (ino %d)---what do I do?\n", 579 VTOI(*vpp)->i_number); 580 return (error); 581 } 582 /* 583 * Remove vnode so that it will be reloaded by VFS_VGET and 584 * checked to see if it is an alias of an existing entry in 585 * the inode cache. 586 */ 587 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 588 VOP_UNLOCK(*vpp, 0); 589 lfs_vunref(*vpp); 590 (*vpp)->v_type = VNON; 591 vgone(*vpp); 592 error = VFS_VGET(mp, ino, vpp); 593 if (error != 0) { 594 *vpp = NULL; 595 return (error); 596 } 597 return (0); 598 } 599 600 int 601 lfs_create(void *v) 602 { 603 struct vop_create_args /* { 604 struct vnode *a_dvp; 605 struct vnode **a_vpp; 606 struct componentname *a_cnp; 607 struct vattr *a_vap; 608 } */ *ap = v; 609 int error; 610 611 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 612 vput(ap->a_dvp); 613 return error; 614 } 615 MARK_VNODE(ap->a_dvp); 616 error = ufs_create(ap); 617 UNMARK_VNODE(ap->a_dvp); 618 if (*(ap->a_vpp)) 619 UNMARK_VNODE(*(ap->a_vpp)); 620 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"create"); 621 return (error); 622 } 623 624 int 625 lfs_mkdir(void *v) 626 { 627 struct vop_mkdir_args /* { 628 struct vnode *a_dvp; 629 struct vnode **a_vpp; 630 struct componentname *a_cnp; 631 struct vattr *a_vap; 632 } */ *ap = v; 633 int error; 634 635 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 636 vput(ap->a_dvp); 637 return error; 638 } 639 MARK_VNODE(ap->a_dvp); 640 error = ufs_mkdir(ap); 641 UNMARK_VNODE(ap->a_dvp); 642 if (*(ap->a_vpp)) 643 UNMARK_VNODE(*(ap->a_vpp)); 644 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"mkdir"); 645 return (error); 646 } 647 648 int 649 lfs_remove(void *v) 650 { 651 struct vop_remove_args /* { 652 struct vnode *a_dvp; 653 struct vnode *a_vp; 654 struct componentname *a_cnp; 655 } */ *ap = v; 656 struct vnode *dvp, *vp; 657 int error; 658 659 dvp = ap->a_dvp; 660 vp = ap->a_vp; 661 if ((error = SET_DIROP2(dvp, vp)) != 0) { 662 if (dvp == vp) 663 vrele(vp); 664 else 665 vput(vp); 666 vput(dvp); 667 return error; 668 } 669 MARK_VNODE(dvp); 670 MARK_VNODE(vp); 671 error = ufs_remove(ap); 672 UNMARK_VNODE(dvp); 673 UNMARK_VNODE(vp); 674 675 SET_ENDOP2(VTOI(dvp)->i_lfs, dvp, vp, "remove"); 676 return (error); 677 } 678 679 int 680 lfs_rmdir(void *v) 681 { 682 struct vop_rmdir_args /* { 683 struct vnodeop_desc *a_desc; 684 struct vnode *a_dvp; 685 struct vnode *a_vp; 686 struct componentname *a_cnp; 687 } */ *ap = v; 688 struct vnode *vp; 689 int error; 690 691 vp = ap->a_vp; 692 if ((error = SET_DIROP2(ap->a_dvp, ap->a_vp)) != 0) { 693 vrele(ap->a_dvp); 694 if (ap->a_vp != ap->a_dvp) 695 VOP_UNLOCK(ap->a_dvp, 0); 696 vput(vp); 697 return error; 698 } 699 MARK_VNODE(ap->a_dvp); 700 MARK_VNODE(vp); 701 error = ufs_rmdir(ap); 702 UNMARK_VNODE(ap->a_dvp); 703 UNMARK_VNODE(vp); 704 705 SET_ENDOP2(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vp, "rmdir"); 706 return (error); 707 } 708 709 int 710 lfs_link(void *v) 711 { 712 struct vop_link_args /* { 713 struct vnode *a_dvp; 714 struct vnode *a_vp; 715 struct componentname *a_cnp; 716 } */ *ap = v; 717 int error; 718 719 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 720 vput(ap->a_dvp); 721 return error; 722 } 723 MARK_VNODE(ap->a_dvp); 724 error = ufs_link(ap); 725 UNMARK_VNODE(ap->a_dvp); 726 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"link"); 727 return (error); 728 } 729 730 int 731 lfs_rename(void *v) 732 { 733 struct vop_rename_args /* { 734 struct vnode *a_fdvp; 735 struct vnode *a_fvp; 736 struct componentname *a_fcnp; 737 struct vnode *a_tdvp; 738 struct vnode *a_tvp; 739 struct componentname *a_tcnp; 740 } */ *ap = v; 741 struct vnode *tvp, *fvp, *tdvp, *fdvp; 742 struct componentname *tcnp, *fcnp; 743 int error; 744 struct lfs *fs; 745 746 fs = VTOI(ap->a_fdvp)->i_lfs; 747 tvp = ap->a_tvp; 748 tdvp = ap->a_tdvp; 749 tcnp = ap->a_tcnp; 750 fvp = ap->a_fvp; 751 fdvp = ap->a_fdvp; 752 fcnp = ap->a_fcnp; 753 754 /* 755 * Check for cross-device rename. 756 * If it is, we don't want to set dirops, just error out. 757 * (In particular note that MARK_VNODE(tdvp) will DTWT on 758 * a cross-device rename.) 759 * 760 * Copied from ufs_rename. 761 */ 762 if ((fvp->v_mount != tdvp->v_mount) || 763 (tvp && (fvp->v_mount != tvp->v_mount))) { 764 error = EXDEV; 765 goto errout; 766 } 767 768 /* 769 * Check to make sure we're not renaming a vnode onto itself 770 * (deleting a hard link by renaming one name onto another); 771 * if we are we can't recursively call VOP_REMOVE since that 772 * would leave us with an unaccounted-for number of live dirops. 773 * 774 * Inline the relevant section of ufs_rename here, *before* 775 * calling SET_DIROP2. 776 */ 777 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 778 (VTOI(tdvp)->i_flags & APPEND))) { 779 error = EPERM; 780 goto errout; 781 } 782 if (fvp == tvp) { 783 if (fvp->v_type == VDIR) { 784 error = EINVAL; 785 goto errout; 786 } 787 788 /* Release destination completely. */ 789 VOP_ABORTOP(tdvp, tcnp); 790 vput(tdvp); 791 vput(tvp); 792 793 /* Delete source. */ 794 vrele(fvp); 795 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 796 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 797 fcnp->cn_nameiop = DELETE; 798 if ((error = relookup(fdvp, &fvp, fcnp))){ 799 /* relookup blew away fdvp */ 800 return (error); 801 } 802 return (VOP_REMOVE(fdvp, fvp, fcnp)); 803 } 804 805 if ((error = SET_DIROP2(tdvp, tvp)) != 0) 806 goto errout; 807 MARK_VNODE(fdvp); 808 MARK_VNODE(tdvp); 809 MARK_VNODE(fvp); 810 if (tvp) { 811 MARK_VNODE(tvp); 812 } 813 814 error = ufs_rename(ap); 815 UNMARK_VNODE(fdvp); 816 UNMARK_VNODE(tdvp); 817 UNMARK_VNODE(fvp); 818 if (tvp) { 819 UNMARK_VNODE(tvp); 820 } 821 SET_ENDOP2(fs, tdvp, tvp, "rename"); 822 return (error); 823 824 errout: 825 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 826 if (tdvp == tvp) 827 vrele(tdvp); 828 else 829 vput(tdvp); 830 if (tvp) 831 vput(tvp); 832 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 833 vrele(fdvp); 834 vrele(fvp); 835 return (error); 836 } 837 838 /* XXX hack to avoid calling ITIMES in getattr */ 839 int 840 lfs_getattr(void *v) 841 { 842 struct vop_getattr_args /* { 843 struct vnode *a_vp; 844 struct vattr *a_vap; 845 struct ucred *a_cred; 846 struct proc *a_p; 847 } */ *ap = v; 848 struct vnode *vp = ap->a_vp; 849 struct inode *ip = VTOI(vp); 850 struct vattr *vap = ap->a_vap; 851 struct lfs *fs = ip->i_lfs; 852 /* 853 * Copy from inode table 854 */ 855 vap->va_fsid = ip->i_dev; 856 vap->va_fileid = ip->i_number; 857 vap->va_mode = ip->i_mode & ~IFMT; 858 vap->va_nlink = ip->i_nlink; 859 vap->va_uid = ip->i_uid; 860 vap->va_gid = ip->i_gid; 861 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 862 vap->va_size = vp->v_size; 863 vap->va_atime.tv_sec = ip->i_ffs1_atime; 864 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 865 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 866 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 867 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 868 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 869 vap->va_flags = ip->i_flags; 870 vap->va_gen = ip->i_gen; 871 /* this doesn't belong here */ 872 if (vp->v_type == VBLK) 873 vap->va_blocksize = BLKDEV_IOSIZE; 874 else if (vp->v_type == VCHR) 875 vap->va_blocksize = MAXBSIZE; 876 else 877 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 878 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 879 vap->va_type = vp->v_type; 880 vap->va_filerev = ip->i_modrev; 881 return (0); 882 } 883 884 /* 885 * Check to make sure the inode blocks won't choke the buffer 886 * cache, then call ufs_setattr as usual. 887 */ 888 int 889 lfs_setattr(void *v) 890 { 891 struct vop_getattr_args /* { 892 struct vnode *a_vp; 893 struct vattr *a_vap; 894 struct ucred *a_cred; 895 struct proc *a_p; 896 } */ *ap = v; 897 struct vnode *vp = ap->a_vp; 898 899 lfs_check(vp, LFS_UNUSED_LBN, 0); 900 return ufs_setattr(v); 901 } 902 903 /* 904 * Close called 905 * 906 * XXX -- we were using ufs_close, but since it updates the 907 * times on the inode, we might need to bump the uinodes 908 * count. 909 */ 910 /* ARGSUSED */ 911 int 912 lfs_close(void *v) 913 { 914 struct vop_close_args /* { 915 struct vnode *a_vp; 916 int a_fflag; 917 struct ucred *a_cred; 918 struct proc *a_p; 919 } */ *ap = v; 920 struct vnode *vp = ap->a_vp; 921 struct inode *ip = VTOI(vp); 922 struct timespec ts; 923 924 if (vp == ip->i_lfs->lfs_ivnode && 925 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 926 return 0; 927 928 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 929 TIMEVAL_TO_TIMESPEC(&time, &ts); 930 LFS_ITIMES(ip, &ts, &ts, &ts); 931 } 932 return (0); 933 } 934 935 /* 936 * Close wrapper for special devices. 937 * 938 * Update the times on the inode then do device close. 939 */ 940 int 941 lfsspec_close(void *v) 942 { 943 struct vop_close_args /* { 944 struct vnode *a_vp; 945 int a_fflag; 946 struct ucred *a_cred; 947 struct proc *a_p; 948 } */ *ap = v; 949 struct vnode *vp; 950 struct inode *ip; 951 struct timespec ts; 952 953 vp = ap->a_vp; 954 ip = VTOI(vp); 955 if (vp->v_usecount > 1) { 956 TIMEVAL_TO_TIMESPEC(&time, &ts); 957 LFS_ITIMES(ip, &ts, &ts, &ts); 958 } 959 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 960 } 961 962 /* 963 * Close wrapper for fifo's. 964 * 965 * Update the times on the inode then do device close. 966 */ 967 int 968 lfsfifo_close(void *v) 969 { 970 struct vop_close_args /* { 971 struct vnode *a_vp; 972 int a_fflag; 973 struct ucred *a_cred; 974 struct proc *a_p; 975 } */ *ap = v; 976 struct vnode *vp; 977 struct inode *ip; 978 struct timespec ts; 979 980 vp = ap->a_vp; 981 ip = VTOI(vp); 982 if (ap->a_vp->v_usecount > 1) { 983 TIMEVAL_TO_TIMESPEC(&time, &ts); 984 LFS_ITIMES(ip, &ts, &ts, &ts); 985 } 986 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 987 } 988 989 /* 990 * Reclaim an inode so that it can be used for other purposes. 991 */ 992 993 int 994 lfs_reclaim(void *v) 995 { 996 struct vop_reclaim_args /* { 997 struct vnode *a_vp; 998 struct proc *a_p; 999 } */ *ap = v; 1000 struct vnode *vp = ap->a_vp; 1001 struct inode *ip = VTOI(vp); 1002 int error; 1003 1004 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1005 1006 LFS_CLR_UINO(ip, IN_ALLMOD); 1007 if ((error = ufs_reclaim(vp, ap->a_p))) 1008 return (error); 1009 pool_put(&lfs_dinode_pool, VTOI(vp)->i_din.ffs1_din); 1010 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1011 ip->inode_ext.lfs = NULL; 1012 pool_put(&lfs_inode_pool, vp->v_data); 1013 vp->v_data = NULL; 1014 return (0); 1015 } 1016 1017 /* 1018 * Read a block from a storage device. 1019 * In order to avoid reading blocks that are in the process of being 1020 * written by the cleaner---and hence are not mutexed by the normal 1021 * buffer cache / page cache mechanisms---check for collisions before 1022 * reading. 1023 * 1024 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1025 * the active cleaner test. 1026 * 1027 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1028 */ 1029 int 1030 lfs_strategy(void *v) 1031 { 1032 struct vop_strategy_args /* { 1033 struct vnode *a_vp; 1034 struct buf *a_bp; 1035 } */ *ap = v; 1036 struct buf *bp; 1037 struct lfs *fs; 1038 struct vnode *vp; 1039 struct inode *ip; 1040 daddr_t tbn; 1041 int i, sn, error, slept; 1042 1043 bp = ap->a_bp; 1044 vp = ap->a_vp; 1045 ip = VTOI(vp); 1046 fs = ip->i_lfs; 1047 1048 /* lfs uses its strategy routine only for read */ 1049 KASSERT(bp->b_flags & B_READ); 1050 1051 if (vp->v_type == VBLK || vp->v_type == VCHR) 1052 panic("lfs_strategy: spec"); 1053 KASSERT(bp->b_bcount != 0); 1054 if (bp->b_blkno == bp->b_lblkno) { 1055 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1056 NULL); 1057 if (error) { 1058 bp->b_error = error; 1059 bp->b_flags |= B_ERROR; 1060 biodone(bp); 1061 return (error); 1062 } 1063 if ((long)bp->b_blkno == -1) /* no valid data */ 1064 clrbuf(bp); 1065 } 1066 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1067 biodone(bp); 1068 return (0); 1069 } 1070 1071 slept = 1; 1072 simple_lock(&fs->lfs_interlock); 1073 while (slept && fs->lfs_seglock) { 1074 simple_unlock(&fs->lfs_interlock); 1075 /* 1076 * Look through list of intervals. 1077 * There will only be intervals to look through 1078 * if the cleaner holds the seglock. 1079 * Since the cleaner is synchronous, we can trust 1080 * the list of intervals to be current. 1081 */ 1082 tbn = dbtofsb(fs, bp->b_blkno); 1083 sn = dtosn(fs, tbn); 1084 slept = 0; 1085 for (i = 0; i < fs->lfs_cleanind; i++) { 1086 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1087 tbn >= fs->lfs_cleanint[i]) { 1088 #ifdef DEBUG_LFS 1089 printf("lfs_strategy: ino %d lbn %" PRId64 1090 " ind %d sn %d fsb %" PRIx32 1091 " given sn %d fsb %" PRIx64 "\n", 1092 ip->i_number, bp->b_lblkno, i, 1093 dtosn(fs, fs->lfs_cleanint[i]), 1094 fs->lfs_cleanint[i], sn, tbn); 1095 printf("lfs_strategy: sleeping on ino %d lbn %" 1096 PRId64 "\n", ip->i_number, bp->b_lblkno); 1097 #endif 1098 tsleep(&fs->lfs_seglock, PRIBIO+1, 1099 "lfs_strategy", 0); 1100 /* Things may be different now; start over. */ 1101 slept = 1; 1102 break; 1103 } 1104 } 1105 simple_lock(&fs->lfs_interlock); 1106 } 1107 simple_unlock(&fs->lfs_interlock); 1108 1109 vp = ip->i_devvp; 1110 VOP_STRATEGY(vp, bp); 1111 return (0); 1112 } 1113 1114 static void 1115 lfs_flush_dirops(struct lfs *fs) 1116 { 1117 struct inode *ip, *nip; 1118 struct vnode *vp; 1119 extern int lfs_dostats; 1120 struct segment *sp; 1121 int needunlock; 1122 1123 if (fs->lfs_ronly) 1124 return; 1125 1126 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) 1127 return; 1128 1129 if (lfs_dostats) 1130 ++lfs_stats.flush_invoked; 1131 1132 /* 1133 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1134 * Technically this is a checkpoint (the on-disk state is valid) 1135 * even though we are leaving out all the file data. 1136 */ 1137 lfs_imtime(fs); 1138 lfs_seglock(fs, SEGM_CKP); 1139 sp = fs->lfs_sp; 1140 1141 /* 1142 * lfs_writevnodes, optimized to get dirops out of the way. 1143 * Only write dirops, and don't flush files' pages, only 1144 * blocks from the directories. 1145 * 1146 * We don't need to vref these files because they are 1147 * dirops and so hold an extra reference until the 1148 * segunlock clears them of that status. 1149 * 1150 * We don't need to check for IN_ADIROP because we know that 1151 * no dirops are active. 1152 * 1153 */ 1154 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1155 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1156 vp = ITOV(ip); 1157 1158 /* 1159 * All writes to directories come from dirops; all 1160 * writes to files' direct blocks go through the page 1161 * cache, which we're not touching. Reads to files 1162 * and/or directories will not be affected by writing 1163 * directory blocks inodes and file inodes. So we don't 1164 * really need to lock. If we don't lock, though, 1165 * make sure that we don't clear IN_MODIFIED 1166 * unnecessarily. 1167 */ 1168 if (vp->v_flag & VXLOCK) 1169 continue; 1170 if (vn_lock(vp, LK_EXCLUSIVE | LK_CANRECURSE | 1171 LK_NOWAIT) == 0) { 1172 needunlock = 1; 1173 } else { 1174 printf("lfs_flush_dirops: flushing locked ino %d\n", 1175 VTOI(vp)->i_number); 1176 needunlock = 0; 1177 } 1178 if (vp->v_type != VREG && 1179 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1180 lfs_writefile(fs, sp, vp); 1181 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1182 !(ip->i_flag & IN_ALLMOD)) { 1183 LFS_SET_UINO(ip, IN_MODIFIED); 1184 } 1185 } 1186 (void) lfs_writeinode(fs, sp, ip); 1187 if (needunlock) 1188 VOP_UNLOCK(vp, 0); 1189 else 1190 LFS_SET_UINO(ip, IN_MODIFIED); 1191 } 1192 /* We've written all the dirops there are */ 1193 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1194 (void) lfs_writeseg(fs, sp); 1195 lfs_segunlock(fs); 1196 } 1197 1198 /* 1199 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1200 */ 1201 int 1202 lfs_fcntl(void *v) 1203 { 1204 struct vop_fcntl_args /* { 1205 struct vnode *a_vp; 1206 u_long a_command; 1207 caddr_t a_data; 1208 int a_fflag; 1209 struct ucred *a_cred; 1210 struct proc *a_p; 1211 } */ *ap = v; 1212 struct timeval *tvp; 1213 BLOCK_INFO *blkiov; 1214 CLEANERINFO *cip; 1215 int blkcnt, error, oclean; 1216 struct lfs_fcntl_markv blkvp; 1217 fsid_t *fsidp; 1218 struct lfs *fs; 1219 struct buf *bp; 1220 daddr_t off; 1221 1222 /* Only respect LFS fcntls on fs root or Ifile */ 1223 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1224 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1225 return ufs_fcntl(v); 1226 } 1227 1228 /* Avoid locking a draining lock */ 1229 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1230 return ESHUTDOWN; 1231 } 1232 1233 fs = VTOI(ap->a_vp)->i_lfs; 1234 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1235 1236 switch (ap->a_command) { 1237 case LFCNSEGWAITALL: 1238 fsidp = NULL; 1239 /* FALLSTHROUGH */ 1240 case LFCNSEGWAIT: 1241 tvp = (struct timeval *)ap->a_data; 1242 simple_lock(&fs->lfs_interlock); 1243 ++fs->lfs_sleepers; 1244 simple_unlock(&fs->lfs_interlock); 1245 VOP_UNLOCK(ap->a_vp, 0); 1246 1247 error = lfs_segwait(fsidp, tvp); 1248 1249 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1250 simple_lock(&fs->lfs_interlock); 1251 if (--fs->lfs_sleepers == 0) 1252 wakeup(&fs->lfs_sleepers); 1253 simple_unlock(&fs->lfs_interlock); 1254 return error; 1255 1256 case LFCNBMAPV: 1257 case LFCNMARKV: 1258 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1259 return (error); 1260 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1261 1262 blkcnt = blkvp.blkcnt; 1263 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1264 return (EINVAL); 1265 blkiov = malloc(blkcnt * sizeof(BLOCK_INFO), M_SEGMENT, M_WAITOK); 1266 if ((error = copyin(blkvp.blkiov, blkiov, 1267 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1268 free(blkiov, M_SEGMENT); 1269 return error; 1270 } 1271 1272 simple_lock(&fs->lfs_interlock); 1273 ++fs->lfs_sleepers; 1274 simple_unlock(&fs->lfs_interlock); 1275 VOP_UNLOCK(ap->a_vp, 0); 1276 if (ap->a_command == LFCNBMAPV) 1277 error = lfs_bmapv(ap->a_p, fsidp, blkiov, blkcnt); 1278 else /* LFCNMARKV */ 1279 error = lfs_markv(ap->a_p, fsidp, blkiov, blkcnt); 1280 if (error == 0) 1281 error = copyout(blkiov, blkvp.blkiov, 1282 blkcnt * sizeof(BLOCK_INFO)); 1283 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1284 simple_lock(&fs->lfs_interlock); 1285 if (--fs->lfs_sleepers == 0) 1286 wakeup(&fs->lfs_sleepers); 1287 simple_unlock(&fs->lfs_interlock); 1288 free(blkiov, M_SEGMENT); 1289 return error; 1290 1291 case LFCNRECLAIM: 1292 /* 1293 * Flush dirops and write Ifile, allowing empty segments 1294 * to be immediately reclaimed. 1295 */ 1296 lfs_writer_enter(fs, "pndirop"); 1297 off = fs->lfs_offset; 1298 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1299 lfs_flush_dirops(fs); 1300 LFS_CLEANERINFO(cip, fs, bp); 1301 oclean = cip->clean; 1302 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1303 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1304 lfs_segunlock(fs); 1305 lfs_writer_leave(fs); 1306 1307 #ifdef DEBUG_LFS 1308 LFS_CLEANERINFO(cip, fs, bp); 1309 oclean = cip->clean; 1310 printf("lfs_fcntl: reclaim wrote %" PRId64 " blocks, cleaned " 1311 "%" PRId32 " segments (activesb %d)\n", 1312 fs->lfs_offset - off, cip->clean - oclean, 1313 fs->lfs_activesb); 1314 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1315 #endif 1316 1317 return 0; 1318 1319 default: 1320 return ufs_fcntl(v); 1321 } 1322 return 0; 1323 } 1324 1325 int 1326 lfs_getpages(void *v) 1327 { 1328 struct vop_getpages_args /* { 1329 struct vnode *a_vp; 1330 voff_t a_offset; 1331 struct vm_page **a_m; 1332 int *a_count; 1333 int a_centeridx; 1334 vm_prot_t a_access_type; 1335 int a_advice; 1336 int a_flags; 1337 } */ *ap = v; 1338 1339 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1340 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1341 return EPERM; 1342 } 1343 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1344 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1345 } 1346 1347 /* 1348 * we're relying on the fact that genfs_getpages() always read in 1349 * entire filesystem blocks. 1350 */ 1351 return genfs_getpages(v); 1352 } 1353 1354 /* 1355 * Make sure that for all pages in every block in the given range, 1356 * either all are dirty or all are clean. If any of the pages 1357 * we've seen so far are dirty, put the vnode on the paging chain, 1358 * and mark it IN_PAGING. 1359 * 1360 * If checkfirst != 0, don't check all the pages but return at the 1361 * first dirty page. 1362 */ 1363 static int 1364 check_dirty(struct lfs *fs, struct vnode *vp, 1365 off_t startoffset, off_t endoffset, off_t blkeof, 1366 int flags, int checkfirst) 1367 { 1368 int by_list; 1369 struct vm_page *curpg = NULL; /* XXX: gcc */ 1370 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1371 off_t soff = 0; /* XXX: gcc */ 1372 voff_t off; 1373 int i; 1374 int nonexistent; 1375 int any_dirty; /* number of dirty pages */ 1376 int dirty; /* number of dirty pages in a block */ 1377 int tdirty; 1378 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1379 1380 top: 1381 by_list = (vp->v_uobj.uo_npages <= 1382 ((endoffset - startoffset) >> PAGE_SHIFT) * 1383 UVM_PAGE_HASH_PENALTY); 1384 any_dirty = 0; 1385 1386 if (by_list) { 1387 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1388 } else { 1389 soff = startoffset; 1390 } 1391 while (by_list || soff < MIN(blkeof, endoffset)) { 1392 if (by_list) { 1393 /* 1394 * find the first page in a block. 1395 */ 1396 if (pages_per_block > 1) { 1397 while (curpg && (curpg->offset & fs->lfs_bmask)) 1398 curpg = TAILQ_NEXT(curpg, listq); 1399 } 1400 if (curpg == NULL) 1401 break; 1402 soff = curpg->offset; 1403 } 1404 1405 /* 1406 * Mark all pages in extended range busy; find out if any 1407 * of them are dirty. 1408 */ 1409 nonexistent = dirty = 0; 1410 for (i = 0; i == 0 || i < pages_per_block; i++) { 1411 if (by_list && pages_per_block <= 1) { 1412 pgs[i] = pg = curpg; 1413 } else { 1414 off = soff + (i << PAGE_SHIFT); 1415 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1416 if (pg == NULL) { 1417 ++nonexistent; 1418 continue; 1419 } 1420 } 1421 KASSERT(pg != NULL); 1422 while (pg->flags & PG_BUSY) { 1423 pg->flags |= PG_WANTED; 1424 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1425 "lfsput", 0); 1426 simple_lock(&vp->v_interlock); 1427 if (by_list) { 1428 if (i > 0) 1429 uvm_page_unbusy(pgs, i); 1430 goto top; 1431 } 1432 } 1433 pg->flags |= PG_BUSY; 1434 UVM_PAGE_OWN(pg, "lfs_putpages"); 1435 1436 pmap_page_protect(pg, VM_PROT_NONE); 1437 tdirty = (pmap_clear_modify(pg) || 1438 (pg->flags & PG_CLEAN) == 0); 1439 dirty += tdirty; 1440 } 1441 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1442 if (by_list) { 1443 curpg = TAILQ_NEXT(curpg, listq); 1444 } else { 1445 soff += fs->lfs_bsize; 1446 } 1447 continue; 1448 } 1449 1450 any_dirty += dirty; 1451 KASSERT(nonexistent == 0); 1452 1453 /* 1454 * If any are dirty make all dirty; unbusy them, 1455 * but if we were asked to clean, wire them so that 1456 * the pagedaemon doesn't bother us about them while 1457 * they're on their way to disk. 1458 */ 1459 for (i = 0; i == 0 || i < pages_per_block; i++) { 1460 pg = pgs[i]; 1461 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1462 if (dirty) { 1463 pg->flags &= ~PG_CLEAN; 1464 if (flags & PGO_FREE) { 1465 /* XXXUBC need better way to update */ 1466 simple_lock(&lfs_subsys_lock); 1467 lfs_subsys_pages += MIN(1, pages_per_block); 1468 simple_unlock(&lfs_subsys_lock); 1469 /* 1470 * Wire the page so that 1471 * pdaemon doesn't see it again. 1472 */ 1473 uvm_lock_pageq(); 1474 uvm_pagewire(pg); 1475 uvm_unlock_pageq(); 1476 1477 /* Suspended write flag */ 1478 pg->flags |= PG_DELWRI; 1479 } 1480 } 1481 if (pg->flags & PG_WANTED) 1482 wakeup(pg); 1483 pg->flags &= ~(PG_WANTED|PG_BUSY); 1484 UVM_PAGE_OWN(pg, NULL); 1485 } 1486 1487 if (checkfirst && any_dirty) 1488 break; 1489 1490 if (by_list) { 1491 curpg = TAILQ_NEXT(curpg, listq); 1492 } else { 1493 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1494 } 1495 } 1496 1497 /* 1498 * If any pages were dirty, mark this inode as "pageout requested", 1499 * and put it on the paging queue. 1500 * XXXUBC locking (check locking on dchainhd too) 1501 */ 1502 #ifdef notyet 1503 if (any_dirty) { 1504 if (!(ip->i_flags & IN_PAGING)) { 1505 ip->i_flags |= IN_PAGING; 1506 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1507 } 1508 } 1509 #endif 1510 return any_dirty; 1511 } 1512 1513 /* 1514 * lfs_putpages functions like genfs_putpages except that 1515 * 1516 * (1) It needs to bounds-check the incoming requests to ensure that 1517 * they are block-aligned; if they are not, expand the range and 1518 * do the right thing in case, e.g., the requested range is clean 1519 * but the expanded range is dirty. 1520 * (2) It needs to explicitly send blocks to be written when it is done. 1521 * VOP_PUTPAGES is not ever called with the seglock held, so 1522 * we simply take the seglock and let lfs_segunlock wait for us. 1523 * XXX Actually we can be called with the seglock held, if we have 1524 * XXX to flush a vnode while lfs_markv is in operation. As of this 1525 * XXX writing we panic in this case. 1526 * 1527 * Assumptions: 1528 * 1529 * (1) The caller does not hold any pages in this vnode busy. If it does, 1530 * there is a danger that when we expand the page range and busy the 1531 * pages we will deadlock. 1532 * (2) We are called with vp->v_interlock held; we must return with it 1533 * released. 1534 * (3) We don't absolutely have to free pages right away, provided that 1535 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1536 * us a request with PGO_FREE, we take the pages out of the paging 1537 * queue and wake up the writer, which will handle freeing them for us. 1538 * 1539 * We ensure that for any filesystem block, all pages for that 1540 * block are either resident or not, even if those pages are higher 1541 * than EOF; that means that we will be getting requests to free 1542 * "unused" pages above EOF all the time, and should ignore them. 1543 * 1544 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1545 */ 1546 1547 int 1548 lfs_putpages(void *v) 1549 { 1550 int error; 1551 struct vop_putpages_args /* { 1552 struct vnode *a_vp; 1553 voff_t a_offlo; 1554 voff_t a_offhi; 1555 int a_flags; 1556 } */ *ap = v; 1557 struct vnode *vp; 1558 struct inode *ip; 1559 struct lfs *fs; 1560 struct segment *sp; 1561 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1562 off_t off, max_endoffset; 1563 int s; 1564 boolean_t seglocked, sync, pagedaemon; 1565 struct vm_page *pg; 1566 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1567 1568 vp = ap->a_vp; 1569 ip = VTOI(vp); 1570 fs = ip->i_lfs; 1571 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1572 pagedaemon = (curproc == uvm.pagedaemon_proc); 1573 1574 /* Putpages does nothing for metadata. */ 1575 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1576 simple_unlock(&vp->v_interlock); 1577 return 0; 1578 } 1579 1580 /* 1581 * If there are no pages, don't do anything. 1582 */ 1583 if (vp->v_uobj.uo_npages == 0) { 1584 s = splbio(); 1585 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1586 (vp->v_flag & VONWORKLST)) { 1587 vp->v_flag &= ~VONWORKLST; 1588 LIST_REMOVE(vp, v_synclist); 1589 } 1590 splx(s); 1591 simple_unlock(&vp->v_interlock); 1592 return 0; 1593 } 1594 1595 blkeof = blkroundup(fs, ip->i_size); 1596 1597 /* 1598 * Ignore requests to free pages past EOF but in the same block 1599 * as EOF, unless the request is synchronous. (XXX why sync?) 1600 * XXXUBC Make these pages look "active" so the pagedaemon won't 1601 * XXXUBC bother us with them again. 1602 */ 1603 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1604 origoffset = ap->a_offlo; 1605 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1606 pg = uvm_pagelookup(&vp->v_uobj, off); 1607 KASSERT(pg != NULL); 1608 while (pg->flags & PG_BUSY) { 1609 pg->flags |= PG_WANTED; 1610 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1611 "lfsput2", 0); 1612 simple_lock(&vp->v_interlock); 1613 } 1614 uvm_lock_pageq(); 1615 uvm_pageactivate(pg); 1616 uvm_unlock_pageq(); 1617 } 1618 ap->a_offlo = blkeof; 1619 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1620 simple_unlock(&vp->v_interlock); 1621 return 0; 1622 } 1623 } 1624 1625 /* 1626 * Extend page range to start and end at block boundaries. 1627 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1628 */ 1629 origoffset = ap->a_offlo; 1630 origendoffset = ap->a_offhi; 1631 startoffset = origoffset & ~(fs->lfs_bmask); 1632 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1633 << fs->lfs_bshift; 1634 1635 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1636 endoffset = max_endoffset; 1637 origendoffset = endoffset; 1638 } else { 1639 origendoffset = round_page(ap->a_offhi); 1640 endoffset = round_page(blkroundup(fs, origendoffset)); 1641 } 1642 1643 KASSERT(startoffset > 0 || endoffset >= startoffset); 1644 if (startoffset == endoffset) { 1645 /* Nothing to do, why were we called? */ 1646 simple_unlock(&vp->v_interlock); 1647 #ifdef DEBUG 1648 printf("lfs_putpages: startoffset = endoffset = %" PRId64 "\n", 1649 startoffset); 1650 #endif 1651 return 0; 1652 } 1653 1654 ap->a_offlo = startoffset; 1655 ap->a_offhi = endoffset; 1656 1657 if (!(ap->a_flags & PGO_CLEANIT)) 1658 return genfs_putpages(v); 1659 1660 /* 1661 * If there are more than one page per block, we don't want 1662 * to get caught locking them backwards; so set PGO_BUSYFAIL 1663 * to avoid deadlocks. 1664 */ 1665 ap->a_flags |= PGO_BUSYFAIL; 1666 1667 do { 1668 int r; 1669 1670 /* If no pages are dirty, we can just use genfs_putpages. */ 1671 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 1672 ap->a_flags, 1) != 0) 1673 break; 1674 1675 if ((r = genfs_putpages(v)) != EDEADLK) 1676 return r; 1677 1678 /* Start over. */ 1679 preempt(1); 1680 simple_lock(&vp->v_interlock); 1681 } while(1); 1682 1683 /* 1684 * Dirty and asked to clean. 1685 * 1686 * Pagedaemon can't actually write LFS pages; wake up 1687 * the writer to take care of that. The writer will 1688 * notice the pager inode queue and act on that. 1689 */ 1690 if (pagedaemon) { 1691 ++fs->lfs_pdflush; 1692 wakeup(&lfs_writer_daemon); 1693 simple_unlock(&vp->v_interlock); 1694 return EWOULDBLOCK; 1695 } 1696 1697 /* 1698 * If this is a file created in a recent dirop, we can't flush its 1699 * inode until the dirop is complete. Drain dirops, then flush the 1700 * filesystem (taking care of any other pending dirops while we're 1701 * at it). 1702 */ 1703 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1704 (vp->v_flag & VDIROP)) { 1705 int locked; 1706 1707 /* printf("putpages to clean VDIROP, flushing\n"); */ 1708 lfs_writer_enter(fs, "ppdirop"); 1709 locked = VOP_ISLOCKED(vp) && /* XXX */ 1710 vp->v_lock.lk_lockholder == curproc->p_pid; 1711 if (locked) 1712 VOP_UNLOCK(vp, 0); 1713 simple_unlock(&vp->v_interlock); 1714 1715 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1716 1717 simple_lock(&vp->v_interlock); 1718 if (locked) 1719 VOP_LOCK(vp, LK_EXCLUSIVE); 1720 lfs_writer_leave(fs); 1721 1722 /* XXX the flush should have taken care of this one too! */ 1723 } 1724 1725 /* 1726 * This is it. We are going to write some pages. From here on 1727 * down it's all just mechanics. 1728 * 1729 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1730 */ 1731 ap->a_flags &= ~PGO_SYNCIO; 1732 1733 /* 1734 * If we've already got the seglock, flush the node and return. 1735 * The FIP has already been set up for us by lfs_writefile, 1736 * and FIP cleanup and lfs_updatemeta will also be done there, 1737 * unless genfs_putpages returns EDEADLK; then we must flush 1738 * what we have, and correct FIP and segment header accounting. 1739 */ 1740 1741 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 1742 if (!seglocked) { 1743 simple_unlock(&vp->v_interlock); 1744 /* 1745 * Take the seglock, because we are going to be writing pages. 1746 */ 1747 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 1748 if (error != 0) 1749 return error; 1750 simple_lock(&vp->v_interlock); 1751 } 1752 1753 /* 1754 * VOP_PUTPAGES should not be called while holding the seglock. 1755 * XXXUBC fix lfs_markv, or do this properly. 1756 */ 1757 /* KASSERT(fs->lfs_seglock == 1); */ 1758 1759 /* 1760 * We assume we're being called with sp->fip pointing at blank space. 1761 * Account for a new FIP in the segment header, and set sp->vp. 1762 * (This should duplicate the setup at the top of lfs_writefile().) 1763 */ 1764 sp = fs->lfs_sp; 1765 if (!seglocked) { 1766 if (sp->seg_bytes_left < fs->lfs_bsize || 1767 sp->sum_bytes_left < sizeof(struct finfo)) 1768 (void) lfs_writeseg(fs, fs->lfs_sp); 1769 1770 sp->sum_bytes_left -= FINFOSIZE; 1771 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1772 } 1773 KASSERT(sp->vp == NULL); 1774 sp->vp = vp; 1775 1776 if (!seglocked) { 1777 if (vp->v_flag & VDIROP) 1778 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 1779 } 1780 1781 sp->fip->fi_nblocks = 0; 1782 sp->fip->fi_ino = ip->i_number; 1783 sp->fip->fi_version = ip->i_gen; 1784 1785 /* 1786 * Loop through genfs_putpages until all pages are gathered. 1787 * genfs_putpages() drops the interlock, so reacquire it if necessary. 1788 * Whenever we lose the interlock we have to rerun check_dirty, as 1789 * well. 1790 */ 1791 again: 1792 check_dirty(fs, vp, startoffset, endoffset, blkeof, ap->a_flags, 0); 1793 1794 if ((error = genfs_putpages(v)) == EDEADLK) { 1795 #ifdef DEBUG_LFS 1796 printf("lfs_putpages: genfs_putpages returned EDEADLK [2]" 1797 " ino %d off %x (seg %d)\n", 1798 ip->i_number, fs->lfs_offset, 1799 dtosn(fs, fs->lfs_offset)); 1800 #endif 1801 /* If nothing to write, short-circuit */ 1802 if (sp->cbpp - sp->bpp > 1) { 1803 /* Write gathered pages */ 1804 lfs_updatemeta(sp); 1805 (void) lfs_writeseg(fs, sp); 1806 1807 /* 1808 * Reinitialize brand new FIP and add us to it. 1809 * (This should duplicate the fixup in 1810 * lfs_gatherpages().) 1811 */ 1812 KASSERT(sp->vp == vp); 1813 sp->fip->fi_version = ip->i_gen; 1814 sp->fip->fi_ino = ip->i_number; 1815 /* Add us to the new segment summary. */ 1816 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1817 sp->sum_bytes_left -= FINFOSIZE; 1818 } 1819 1820 /* Give the write a chance to complete */ 1821 preempt(1); 1822 1823 /* We've lost the interlock. Start over. */ 1824 simple_lock(&vp->v_interlock); 1825 goto again; 1826 } 1827 1828 KASSERT(sp->vp == vp); 1829 if (!seglocked) { 1830 sp->vp = NULL; /* XXX lfs_gather below will set this */ 1831 1832 /* Write indirect blocks as well */ 1833 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 1834 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 1835 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 1836 1837 KASSERT(sp->vp == NULL); 1838 sp->vp = vp; 1839 } 1840 1841 /* 1842 * Blocks are now gathered into a segment waiting to be written. 1843 * All that's left to do is update metadata, and write them. 1844 */ 1845 lfs_updatemeta(sp); 1846 KASSERT(sp->vp == vp); 1847 sp->vp = NULL; 1848 1849 if (seglocked) { 1850 /* we're called by lfs_writefile. */ 1851 return error; 1852 } 1853 1854 /* 1855 * Clean up FIP, since we're done writing this file. 1856 * This should duplicate cleanup at the end of lfs_writefile(). 1857 */ 1858 if (sp->fip->fi_nblocks != 0) { 1859 sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE + 1860 sizeof(int32_t) * sp->fip->fi_nblocks); 1861 sp->start_lbp = &sp->fip->fi_blocks[0]; 1862 } else { 1863 sp->sum_bytes_left += FINFOSIZE; 1864 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 1865 } 1866 lfs_writeseg(fs, fs->lfs_sp); 1867 1868 /* 1869 * XXX - with the malloc/copy writeseg, the pages are freed by now 1870 * even if we don't wait (e.g. if we hold a nested lock). This 1871 * will not be true if we stop using malloc/copy. 1872 */ 1873 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 1874 lfs_segunlock(fs); 1875 1876 /* 1877 * Wait for v_numoutput to drop to zero. The seglock should 1878 * take care of this, but there is a slight possibility that 1879 * aiodoned might not have got around to our buffers yet. 1880 */ 1881 if (sync) { 1882 int s; 1883 1884 s = splbio(); 1885 simple_lock(&global_v_numoutput_slock); 1886 while (vp->v_numoutput > 0) { 1887 #ifdef DEBUG 1888 printf("ino %d sleeping on num %d\n", 1889 ip->i_number, vp->v_numoutput); 1890 #endif 1891 vp->v_flag |= VBWAIT; 1892 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 1893 &global_v_numoutput_slock); 1894 } 1895 simple_unlock(&global_v_numoutput_slock); 1896 splx(s); 1897 } 1898 return error; 1899 } 1900 1901 /* 1902 * Return the last logical file offset that should be written for this file 1903 * if we're doing a write that ends at "size". If writing, we need to know 1904 * about sizes on disk, i.e. fragments if there are any; if reading, we need 1905 * to know about entire blocks. 1906 */ 1907 void 1908 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 1909 { 1910 struct inode *ip = VTOI(vp); 1911 struct lfs *fs = ip->i_lfs; 1912 daddr_t olbn, nlbn; 1913 1914 KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)); 1915 KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)) 1916 != (GOP_SIZE_READ | GOP_SIZE_WRITE)); 1917 1918 olbn = lblkno(fs, ip->i_size); 1919 nlbn = lblkno(fs, size); 1920 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 1921 *eobp = fragroundup(fs, size); 1922 } else { 1923 *eobp = blkroundup(fs, size); 1924 } 1925 } 1926 1927 #ifdef DEBUG 1928 void lfs_dump_vop(void *); 1929 1930 void 1931 lfs_dump_vop(void *v) 1932 { 1933 struct vop_putpages_args /* { 1934 struct vnode *a_vp; 1935 voff_t a_offlo; 1936 voff_t a_offhi; 1937 int a_flags; 1938 } */ *ap = v; 1939 1940 #ifdef DDB 1941 vfs_vnode_print(ap->a_vp, 0, printf); 1942 #endif 1943 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 1944 } 1945 #endif 1946 1947 int 1948 lfs_mmap(void *v) 1949 { 1950 struct vop_mmap_args /* { 1951 const struct vnodeop_desc *a_desc; 1952 struct vnode *a_vp; 1953 int a_fflags; 1954 struct ucred *a_cred; 1955 struct proc *a_p; 1956 } */ *ap = v; 1957 1958 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 1959 return EOPNOTSUPP; 1960 return ufs_mmap(v); 1961 } 1962
Indexes created Sun Oct 19 02:09:48 GMT 2025