lfs_vnops.c revision 1.109.2.6
1 /* $NetBSD: lfs_vnops.c,v 1.109.2.6 2004/10/30 06:46:50 skrll 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 1.109.2.6 2004/10/30 06:46:50 skrll 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 lwp *a_l; 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 (wait && !VPISEMPTY(vp)) 321 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 322 323 return error; 324 } 325 326 /* 327 * Take IN_ADIROP off, then call ufs_inactive. 328 */ 329 int 330 lfs_inactive(void *v) 331 { 332 struct vop_inactive_args /* { 333 struct vnode *a_vp; 334 struct lwp *a_l; 335 } */ *ap = v; 336 337 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 338 339 lfs_unmark_vnode(ap->a_vp); 340 341 /* 342 * The Ifile is only ever inactivated on unmount. 343 * Streamline this process by not giving it more dirty blocks. 344 */ 345 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 346 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 347 VOP_UNLOCK(ap->a_vp, 0); 348 return 0; 349 } 350 351 return ufs_inactive(v); 352 } 353 354 /* 355 * These macros are used to bracket UFS directory ops, so that we can 356 * identify all the pages touched during directory ops which need to 357 * be ordered and flushed atomically, so that they may be recovered. 358 */ 359 /* 360 * XXX KS - Because we have to mark nodes VDIROP in order to prevent 361 * the cache from reclaiming them while a dirop is in progress, we must 362 * also manage the number of nodes so marked (otherwise we can run out). 363 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 364 * is decremented during segment write, when VDIROP is taken off. 365 */ 366 #define SET_DIROP(vp) SET_DIROP2((vp), NULL) 367 #define SET_DIROP2(vp, vp2) lfs_set_dirop((vp), (vp2)) 368 static int lfs_set_dirop(struct vnode *, struct vnode *); 369 370 #define NRESERVE(fs) (btofsb(fs, (NIADDR + 3 + (2 * NIADDR + 3)) << fs->lfs_bshift)) 371 372 static int 373 lfs_set_dirop(struct vnode *vp, struct vnode *vp2) 374 { 375 struct lfs *fs; 376 int error; 377 378 KASSERT(VOP_ISLOCKED(vp)); 379 KASSERT(vp2 == NULL || VOP_ISLOCKED(vp2)); 380 381 fs = VTOI(vp)->i_lfs; 382 /* 383 * We might need one directory block plus supporting indirect blocks, 384 * plus an inode block and ifile page for the new vnode. 385 */ 386 if ((error = lfs_reserve(fs, vp, vp2, NRESERVE(fs))) != 0) 387 return (error); 388 389 if (fs->lfs_dirops == 0) 390 lfs_check(vp, LFS_UNUSED_LBN, 0); 391 restart: 392 simple_lock(&fs->lfs_interlock); 393 if (fs->lfs_writer) { 394 ltsleep(&fs->lfs_dirops, (PRIBIO + 1) | PNORELOCK, 395 "lfs_sdirop", 0, &fs->lfs_interlock); 396 goto restart; 397 } 398 simple_lock(&lfs_subsys_lock); 399 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 400 wakeup(&lfs_writer_daemon); 401 simple_unlock(&lfs_subsys_lock); 402 simple_unlock(&fs->lfs_interlock); 403 preempt(1); 404 goto restart; 405 } 406 407 if (lfs_dirvcount > LFS_MAX_DIROP) { 408 simple_unlock(&fs->lfs_interlock); 409 #ifdef DEBUG_LFS 410 printf("lfs_set_dirop: sleeping with dirops=%d, " 411 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount); 412 #endif 413 if ((error = ltsleep(&lfs_dirvcount, 414 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 415 &lfs_subsys_lock)) != 0) { 416 goto unreserve; 417 } 418 goto restart; 419 } 420 simple_unlock(&lfs_subsys_lock); 421 422 ++fs->lfs_dirops; 423 fs->lfs_doifile = 1; 424 simple_unlock(&fs->lfs_interlock); 425 426 /* Hold a reference so SET_ENDOP will be happy */ 427 vref(vp); 428 if (vp2) 429 vref(vp2); 430 431 return 0; 432 433 unreserve: 434 lfs_reserve(fs, vp, vp2, -NRESERVE(fs)); 435 return error; 436 } 437 438 #define SET_ENDOP(fs, vp, str) SET_ENDOP2((fs), (vp), NULL, (str)) 439 #define SET_ENDOP2(fs, vp, vp2, str) { \ 440 --(fs)->lfs_dirops; \ 441 if (!(fs)->lfs_dirops) { \ 442 if ((fs)->lfs_nadirop) { \ 443 panic("SET_ENDOP: %s: no dirops but nadirop=%d", \ 444 (str), (fs)->lfs_nadirop); \ 445 } \ 446 wakeup(&(fs)->lfs_writer); \ 447 lfs_check((vp),LFS_UNUSED_LBN,0); \ 448 } \ 449 lfs_reserve((fs), vp, vp2, -NRESERVE(fs)); /* XXX */ \ 450 vrele(vp); \ 451 if (vp2) \ 452 vrele(vp2); \ 453 } 454 455 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 456 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 457 458 void 459 lfs_mark_vnode(struct vnode *vp) 460 { 461 struct inode *ip = VTOI(vp); 462 struct lfs *fs = ip->i_lfs; 463 464 if (!(ip->i_flag & IN_ADIROP)) { 465 if (!(vp->v_flag & VDIROP)) { 466 (void)lfs_vref(vp); 467 ++lfs_dirvcount; 468 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 469 vp->v_flag |= VDIROP; 470 } 471 ++fs->lfs_nadirop; 472 ip->i_flag |= IN_ADIROP; 473 } else 474 KASSERT(vp->v_flag & VDIROP); 475 } 476 477 void 478 lfs_unmark_vnode(struct vnode *vp) 479 { 480 struct inode *ip = VTOI(vp); 481 482 if (ip->i_flag & IN_ADIROP) { 483 KASSERT(vp->v_flag & VDIROP); 484 --ip->i_lfs->lfs_nadirop; 485 ip->i_flag &= ~IN_ADIROP; 486 } 487 } 488 489 int 490 lfs_symlink(void *v) 491 { 492 struct vop_symlink_args /* { 493 struct vnode *a_dvp; 494 struct vnode **a_vpp; 495 struct componentname *a_cnp; 496 struct vattr *a_vap; 497 char *a_target; 498 } */ *ap = v; 499 int error; 500 501 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 502 vput(ap->a_dvp); 503 return error; 504 } 505 MARK_VNODE(ap->a_dvp); 506 error = ufs_symlink(ap); 507 UNMARK_VNODE(ap->a_dvp); 508 if (*(ap->a_vpp)) 509 UNMARK_VNODE(*(ap->a_vpp)); 510 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"symlink"); 511 return (error); 512 } 513 514 int 515 lfs_mknod(void *v) 516 { 517 struct vop_mknod_args /* { 518 struct vnode *a_dvp; 519 struct vnode **a_vpp; 520 struct componentname *a_cnp; 521 struct vattr *a_vap; 522 } */ *ap = v; 523 struct vattr *vap = ap->a_vap; 524 struct vnode **vpp = ap->a_vpp; 525 struct componentname *cnp = ap->a_cnp; 526 struct inode *ip; 527 int error; 528 struct mount *mp; 529 ino_t ino; 530 531 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 532 vput(ap->a_dvp); 533 return error; 534 } 535 MARK_VNODE(ap->a_dvp); 536 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 537 ap->a_dvp, vpp, cnp); 538 UNMARK_VNODE(ap->a_dvp); 539 if (*(ap->a_vpp)) 540 UNMARK_VNODE(*(ap->a_vpp)); 541 542 /* Either way we're done with the dirop at this point */ 543 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"mknod"); 544 545 if (error) 546 return (error); 547 548 ip = VTOI(*vpp); 549 mp = (*vpp)->v_mount; 550 ino = ip->i_number; 551 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 552 if (vap->va_rdev != VNOVAL) { 553 /* 554 * Want to be able to use this to make badblock 555 * inodes, so don't truncate the dev number. 556 */ 557 #if 0 558 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 559 UFS_MPNEEDSWAP((*vpp)->v_mount)); 560 #else 561 ip->i_ffs1_rdev = vap->va_rdev; 562 #endif 563 } 564 /* 565 * Call fsync to write the vnode so that we don't have to deal with 566 * flushing it when it's marked VDIROP|VXLOCK. 567 * 568 * XXX KS - If we can't flush we also can't call vgone(), so must 569 * return. But, that leaves this vnode in limbo, also not good. 570 * Can this ever happen (barring hardware failure)? 571 */ 572 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 573 cnp->cn_lwp)) != 0) { 574 printf("Couldn't fsync in mknod (ino %d)---what do I do?\n", 575 VTOI(*vpp)->i_number); 576 return (error); 577 } 578 /* 579 * Remove vnode so that it will be reloaded by VFS_VGET and 580 * checked to see if it is an alias of an existing entry in 581 * the inode cache. 582 */ 583 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 584 VOP_UNLOCK(*vpp, 0); 585 lfs_vunref(*vpp); 586 (*vpp)->v_type = VNON; 587 vgone(*vpp); 588 error = VFS_VGET(mp, ino, vpp); 589 if (error != 0) { 590 *vpp = NULL; 591 return (error); 592 } 593 return (0); 594 } 595 596 int 597 lfs_create(void *v) 598 { 599 struct vop_create_args /* { 600 struct vnode *a_dvp; 601 struct vnode **a_vpp; 602 struct componentname *a_cnp; 603 struct vattr *a_vap; 604 } */ *ap = v; 605 int error; 606 607 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 608 vput(ap->a_dvp); 609 return error; 610 } 611 MARK_VNODE(ap->a_dvp); 612 error = ufs_create(ap); 613 UNMARK_VNODE(ap->a_dvp); 614 if (*(ap->a_vpp)) 615 UNMARK_VNODE(*(ap->a_vpp)); 616 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"create"); 617 return (error); 618 } 619 620 int 621 lfs_mkdir(void *v) 622 { 623 struct vop_mkdir_args /* { 624 struct vnode *a_dvp; 625 struct vnode **a_vpp; 626 struct componentname *a_cnp; 627 struct vattr *a_vap; 628 } */ *ap = v; 629 int error; 630 631 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 632 vput(ap->a_dvp); 633 return error; 634 } 635 MARK_VNODE(ap->a_dvp); 636 error = ufs_mkdir(ap); 637 UNMARK_VNODE(ap->a_dvp); 638 if (*(ap->a_vpp)) 639 UNMARK_VNODE(*(ap->a_vpp)); 640 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"mkdir"); 641 return (error); 642 } 643 644 int 645 lfs_remove(void *v) 646 { 647 struct vop_remove_args /* { 648 struct vnode *a_dvp; 649 struct vnode *a_vp; 650 struct componentname *a_cnp; 651 } */ *ap = v; 652 struct vnode *dvp, *vp; 653 int error; 654 655 dvp = ap->a_dvp; 656 vp = ap->a_vp; 657 if ((error = SET_DIROP2(dvp, vp)) != 0) { 658 if (dvp == vp) 659 vrele(vp); 660 else 661 vput(vp); 662 vput(dvp); 663 return error; 664 } 665 MARK_VNODE(dvp); 666 MARK_VNODE(vp); 667 error = ufs_remove(ap); 668 UNMARK_VNODE(dvp); 669 UNMARK_VNODE(vp); 670 671 SET_ENDOP2(VTOI(dvp)->i_lfs, dvp, vp, "remove"); 672 return (error); 673 } 674 675 int 676 lfs_rmdir(void *v) 677 { 678 struct vop_rmdir_args /* { 679 struct vnodeop_desc *a_desc; 680 struct vnode *a_dvp; 681 struct vnode *a_vp; 682 struct componentname *a_cnp; 683 } */ *ap = v; 684 struct vnode *vp; 685 int error; 686 687 vp = ap->a_vp; 688 if ((error = SET_DIROP2(ap->a_dvp, ap->a_vp)) != 0) { 689 vrele(ap->a_dvp); 690 if (ap->a_vp != ap->a_dvp) 691 VOP_UNLOCK(ap->a_dvp, 0); 692 vput(vp); 693 return error; 694 } 695 MARK_VNODE(ap->a_dvp); 696 MARK_VNODE(vp); 697 error = ufs_rmdir(ap); 698 UNMARK_VNODE(ap->a_dvp); 699 UNMARK_VNODE(vp); 700 701 SET_ENDOP2(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vp, "rmdir"); 702 return (error); 703 } 704 705 int 706 lfs_link(void *v) 707 { 708 struct vop_link_args /* { 709 struct vnode *a_dvp; 710 struct vnode *a_vp; 711 struct componentname *a_cnp; 712 } */ *ap = v; 713 int error; 714 715 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 716 vput(ap->a_dvp); 717 return error; 718 } 719 MARK_VNODE(ap->a_dvp); 720 error = ufs_link(ap); 721 UNMARK_VNODE(ap->a_dvp); 722 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"link"); 723 return (error); 724 } 725 726 int 727 lfs_rename(void *v) 728 { 729 struct vop_rename_args /* { 730 struct vnode *a_fdvp; 731 struct vnode *a_fvp; 732 struct componentname *a_fcnp; 733 struct vnode *a_tdvp; 734 struct vnode *a_tvp; 735 struct componentname *a_tcnp; 736 } */ *ap = v; 737 struct vnode *tvp, *fvp, *tdvp, *fdvp; 738 struct componentname *tcnp, *fcnp; 739 int error; 740 struct lfs *fs; 741 742 fs = VTOI(ap->a_fdvp)->i_lfs; 743 tvp = ap->a_tvp; 744 tdvp = ap->a_tdvp; 745 tcnp = ap->a_tcnp; 746 fvp = ap->a_fvp; 747 fdvp = ap->a_fdvp; 748 fcnp = ap->a_fcnp; 749 750 /* 751 * Check for cross-device rename. 752 * If it is, we don't want to set dirops, just error out. 753 * (In particular note that MARK_VNODE(tdvp) will DTWT on 754 * a cross-device rename.) 755 * 756 * Copied from ufs_rename. 757 */ 758 if ((fvp->v_mount != tdvp->v_mount) || 759 (tvp && (fvp->v_mount != tvp->v_mount))) { 760 error = EXDEV; 761 goto errout; 762 } 763 764 /* 765 * Check to make sure we're not renaming a vnode onto itself 766 * (deleting a hard link by renaming one name onto another); 767 * if we are we can't recursively call VOP_REMOVE since that 768 * would leave us with an unaccounted-for number of live dirops. 769 * 770 * Inline the relevant section of ufs_rename here, *before* 771 * calling SET_DIROP2. 772 */ 773 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 774 (VTOI(tdvp)->i_flags & APPEND))) { 775 error = EPERM; 776 goto errout; 777 } 778 if (fvp == tvp) { 779 if (fvp->v_type == VDIR) { 780 error = EINVAL; 781 goto errout; 782 } 783 784 /* Release destination completely. */ 785 VOP_ABORTOP(tdvp, tcnp); 786 vput(tdvp); 787 vput(tvp); 788 789 /* Delete source. */ 790 vrele(fvp); 791 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 792 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 793 fcnp->cn_nameiop = DELETE; 794 if ((error = relookup(fdvp, &fvp, fcnp))){ 795 /* relookup blew away fdvp */ 796 return (error); 797 } 798 return (VOP_REMOVE(fdvp, fvp, fcnp)); 799 } 800 801 if ((error = SET_DIROP2(tdvp, tvp)) != 0) 802 goto errout; 803 MARK_VNODE(fdvp); 804 MARK_VNODE(tdvp); 805 MARK_VNODE(fvp); 806 if (tvp) { 807 MARK_VNODE(tvp); 808 } 809 810 error = ufs_rename(ap); 811 UNMARK_VNODE(fdvp); 812 UNMARK_VNODE(tdvp); 813 UNMARK_VNODE(fvp); 814 if (tvp) { 815 UNMARK_VNODE(tvp); 816 } 817 SET_ENDOP2(fs, tdvp, tvp, "rename"); 818 return (error); 819 820 errout: 821 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 822 if (tdvp == tvp) 823 vrele(tdvp); 824 else 825 vput(tdvp); 826 if (tvp) 827 vput(tvp); 828 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 829 vrele(fdvp); 830 vrele(fvp); 831 return (error); 832 } 833 834 /* XXX hack to avoid calling ITIMES in getattr */ 835 int 836 lfs_getattr(void *v) 837 { 838 struct vop_getattr_args /* { 839 struct vnode *a_vp; 840 struct vattr *a_vap; 841 struct ucred *a_cred; 842 struct lwp *a_l; 843 } */ *ap = v; 844 struct vnode *vp = ap->a_vp; 845 struct inode *ip = VTOI(vp); 846 struct vattr *vap = ap->a_vap; 847 struct lfs *fs = ip->i_lfs; 848 /* 849 * Copy from inode table 850 */ 851 vap->va_fsid = ip->i_dev; 852 vap->va_fileid = ip->i_number; 853 vap->va_mode = ip->i_mode & ~IFMT; 854 vap->va_nlink = ip->i_nlink; 855 vap->va_uid = ip->i_uid; 856 vap->va_gid = ip->i_gid; 857 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 858 vap->va_size = vp->v_size; 859 vap->va_atime.tv_sec = ip->i_ffs1_atime; 860 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 861 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 862 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 863 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 864 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 865 vap->va_flags = ip->i_flags; 866 vap->va_gen = ip->i_gen; 867 /* this doesn't belong here */ 868 if (vp->v_type == VBLK) 869 vap->va_blocksize = BLKDEV_IOSIZE; 870 else if (vp->v_type == VCHR) 871 vap->va_blocksize = MAXBSIZE; 872 else 873 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 874 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 875 vap->va_type = vp->v_type; 876 vap->va_filerev = ip->i_modrev; 877 return (0); 878 } 879 880 /* 881 * Check to make sure the inode blocks won't choke the buffer 882 * cache, then call ufs_setattr as usual. 883 */ 884 int 885 lfs_setattr(void *v) 886 { 887 struct vop_getattr_args /* { 888 struct vnode *a_vp; 889 struct vattr *a_vap; 890 struct ucred *a_cred; 891 struct lwp *a_l; 892 } */ *ap = v; 893 struct vnode *vp = ap->a_vp; 894 895 lfs_check(vp, LFS_UNUSED_LBN, 0); 896 return ufs_setattr(v); 897 } 898 899 /* 900 * Close called 901 * 902 * XXX -- we were using ufs_close, but since it updates the 903 * times on the inode, we might need to bump the uinodes 904 * count. 905 */ 906 /* ARGSUSED */ 907 int 908 lfs_close(void *v) 909 { 910 struct vop_close_args /* { 911 struct vnode *a_vp; 912 int a_fflag; 913 struct ucred *a_cred; 914 struct lwp *a_l; 915 } */ *ap = v; 916 struct vnode *vp = ap->a_vp; 917 struct inode *ip = VTOI(vp); 918 struct timespec ts; 919 920 if (vp == ip->i_lfs->lfs_ivnode && 921 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 922 return 0; 923 924 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 925 TIMEVAL_TO_TIMESPEC(&time, &ts); 926 LFS_ITIMES(ip, &ts, &ts, &ts); 927 } 928 return (0); 929 } 930 931 /* 932 * Close wrapper for special devices. 933 * 934 * Update the times on the inode then do device close. 935 */ 936 int 937 lfsspec_close(void *v) 938 { 939 struct vop_close_args /* { 940 struct vnode *a_vp; 941 int a_fflag; 942 struct ucred *a_cred; 943 struct lwp *a_l; 944 } */ *ap = v; 945 struct vnode *vp; 946 struct inode *ip; 947 struct timespec ts; 948 949 vp = ap->a_vp; 950 ip = VTOI(vp); 951 if (vp->v_usecount > 1) { 952 TIMEVAL_TO_TIMESPEC(&time, &ts); 953 LFS_ITIMES(ip, &ts, &ts, &ts); 954 } 955 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 956 } 957 958 /* 959 * Close wrapper for fifo's. 960 * 961 * Update the times on the inode then do device close. 962 */ 963 int 964 lfsfifo_close(void *v) 965 { 966 struct vop_close_args /* { 967 struct vnode *a_vp; 968 int a_fflag; 969 struct ucred *a_cred; 970 struct lwp *a_l; 971 } */ *ap = v; 972 struct vnode *vp; 973 struct inode *ip; 974 struct timespec ts; 975 976 vp = ap->a_vp; 977 ip = VTOI(vp); 978 if (ap->a_vp->v_usecount > 1) { 979 TIMEVAL_TO_TIMESPEC(&time, &ts); 980 LFS_ITIMES(ip, &ts, &ts, &ts); 981 } 982 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 983 } 984 985 /* 986 * Reclaim an inode so that it can be used for other purposes. 987 */ 988 989 int 990 lfs_reclaim(void *v) 991 { 992 struct vop_reclaim_args /* { 993 struct vnode *a_vp; 994 struct lwp *a_l; 995 } */ *ap = v; 996 struct vnode *vp = ap->a_vp; 997 struct inode *ip = VTOI(vp); 998 int error; 999 1000 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1001 1002 LFS_CLR_UINO(ip, IN_ALLMOD); 1003 if ((error = ufs_reclaim(vp, ap->a_l))) 1004 return (error); 1005 pool_put(&lfs_dinode_pool, VTOI(vp)->i_din.ffs1_din); 1006 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1007 ip->inode_ext.lfs = NULL; 1008 pool_put(&lfs_inode_pool, vp->v_data); 1009 vp->v_data = NULL; 1010 return (0); 1011 } 1012 1013 /* 1014 * Read a block from a storage device. 1015 * In order to avoid reading blocks that are in the process of being 1016 * written by the cleaner---and hence are not mutexed by the normal 1017 * buffer cache / page cache mechanisms---check for collisions before 1018 * reading. 1019 * 1020 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1021 * the active cleaner test. 1022 * 1023 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1024 */ 1025 int 1026 lfs_strategy(void *v) 1027 { 1028 struct vop_strategy_args /* { 1029 struct vnode *a_vp; 1030 struct buf *a_bp; 1031 } */ *ap = v; 1032 struct buf *bp; 1033 struct lfs *fs; 1034 struct vnode *vp; 1035 struct inode *ip; 1036 daddr_t tbn; 1037 int i, sn, error, slept; 1038 1039 bp = ap->a_bp; 1040 vp = ap->a_vp; 1041 ip = VTOI(vp); 1042 fs = ip->i_lfs; 1043 1044 /* lfs uses its strategy routine only for read */ 1045 KASSERT(bp->b_flags & B_READ); 1046 1047 if (vp->v_type == VBLK || vp->v_type == VCHR) 1048 panic("lfs_strategy: spec"); 1049 KASSERT(bp->b_bcount != 0); 1050 if (bp->b_blkno == bp->b_lblkno) { 1051 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1052 NULL); 1053 if (error) { 1054 bp->b_error = error; 1055 bp->b_flags |= B_ERROR; 1056 biodone(bp); 1057 return (error); 1058 } 1059 if ((long)bp->b_blkno == -1) /* no valid data */ 1060 clrbuf(bp); 1061 } 1062 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1063 biodone(bp); 1064 return (0); 1065 } 1066 1067 slept = 1; 1068 simple_lock(&fs->lfs_interlock); 1069 while (slept && fs->lfs_seglock) { 1070 simple_unlock(&fs->lfs_interlock); 1071 /* 1072 * Look through list of intervals. 1073 * There will only be intervals to look through 1074 * if the cleaner holds the seglock. 1075 * Since the cleaner is synchronous, we can trust 1076 * the list of intervals to be current. 1077 */ 1078 tbn = dbtofsb(fs, bp->b_blkno); 1079 sn = dtosn(fs, tbn); 1080 slept = 0; 1081 for (i = 0; i < fs->lfs_cleanind; i++) { 1082 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1083 tbn >= fs->lfs_cleanint[i]) { 1084 #ifdef DEBUG_LFS 1085 printf("lfs_strategy: ino %d lbn %" PRId64 1086 " ind %d sn %d fsb %" PRIx32 1087 " given sn %d fsb %" PRIx64 "\n", 1088 ip->i_number, bp->b_lblkno, i, 1089 dtosn(fs, fs->lfs_cleanint[i]), 1090 fs->lfs_cleanint[i], sn, tbn); 1091 printf("lfs_strategy: sleeping on ino %d lbn %" 1092 PRId64 "\n", ip->i_number, bp->b_lblkno); 1093 #endif 1094 tsleep(&fs->lfs_seglock, PRIBIO+1, 1095 "lfs_strategy", 0); 1096 /* Things may be different now; start over. */ 1097 slept = 1; 1098 break; 1099 } 1100 } 1101 simple_lock(&fs->lfs_interlock); 1102 } 1103 simple_unlock(&fs->lfs_interlock); 1104 1105 vp = ip->i_devvp; 1106 VOP_STRATEGY(vp, bp); 1107 return (0); 1108 } 1109 1110 static void 1111 lfs_flush_dirops(struct lfs *fs) 1112 { 1113 struct inode *ip, *nip; 1114 struct vnode *vp; 1115 extern int lfs_dostats; 1116 struct segment *sp; 1117 int needunlock; 1118 1119 if (fs->lfs_ronly) 1120 return; 1121 1122 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) 1123 return; 1124 1125 if (lfs_dostats) 1126 ++lfs_stats.flush_invoked; 1127 1128 /* 1129 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1130 * Technically this is a checkpoint (the on-disk state is valid) 1131 * even though we are leaving out all the file data. 1132 */ 1133 lfs_imtime(fs); 1134 lfs_seglock(fs, SEGM_CKP); 1135 sp = fs->lfs_sp; 1136 1137 /* 1138 * lfs_writevnodes, optimized to get dirops out of the way. 1139 * Only write dirops, and don't flush files' pages, only 1140 * blocks from the directories. 1141 * 1142 * We don't need to vref these files because they are 1143 * dirops and so hold an extra reference until the 1144 * segunlock clears them of that status. 1145 * 1146 * We don't need to check for IN_ADIROP because we know that 1147 * no dirops are active. 1148 * 1149 */ 1150 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1151 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1152 vp = ITOV(ip); 1153 1154 /* 1155 * All writes to directories come from dirops; all 1156 * writes to files' direct blocks go through the page 1157 * cache, which we're not touching. Reads to files 1158 * and/or directories will not be affected by writing 1159 * directory blocks inodes and file inodes. So we don't 1160 * really need to lock. If we don't lock, though, 1161 * make sure that we don't clear IN_MODIFIED 1162 * unnecessarily. 1163 */ 1164 if (vp->v_flag & VXLOCK) 1165 continue; 1166 if (vn_lock(vp, LK_EXCLUSIVE | LK_CANRECURSE | 1167 LK_NOWAIT) == 0) { 1168 needunlock = 1; 1169 } else { 1170 printf("lfs_flush_dirops: flushing locked ino %d\n", 1171 VTOI(vp)->i_number); 1172 needunlock = 0; 1173 } 1174 if (vp->v_type != VREG && 1175 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1176 lfs_writefile(fs, sp, vp); 1177 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1178 !(ip->i_flag & IN_ALLMOD)) { 1179 LFS_SET_UINO(ip, IN_MODIFIED); 1180 } 1181 } 1182 (void) lfs_writeinode(fs, sp, ip); 1183 if (needunlock) 1184 VOP_UNLOCK(vp, 0); 1185 else 1186 LFS_SET_UINO(ip, IN_MODIFIED); 1187 } 1188 /* We've written all the dirops there are */ 1189 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1190 (void) lfs_writeseg(fs, sp); 1191 lfs_segunlock(fs); 1192 } 1193 1194 /* 1195 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1196 */ 1197 int 1198 lfs_fcntl(void *v) 1199 { 1200 struct vop_fcntl_args /* { 1201 struct vnode *a_vp; 1202 u_long a_command; 1203 caddr_t a_data; 1204 int a_fflag; 1205 struct ucred *a_cred; 1206 struct lwp *a_l; 1207 } */ *ap = v; 1208 struct timeval *tvp; 1209 BLOCK_INFO *blkiov; 1210 CLEANERINFO *cip; 1211 int blkcnt, error, oclean; 1212 struct lfs_fcntl_markv blkvp; 1213 struct proc *p; 1214 fsid_t *fsidp; 1215 struct lfs *fs; 1216 struct buf *bp; 1217 daddr_t off; 1218 1219 /* Only respect LFS fcntls on fs root or Ifile */ 1220 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1221 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1222 return ufs_fcntl(v); 1223 } 1224 1225 /* Avoid locking a draining lock */ 1226 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1227 return ESHUTDOWN; 1228 } 1229 1230 p = ap->a_l->l_proc; 1231 fs = VTOI(ap->a_vp)->i_lfs; 1232 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1233 1234 switch (ap->a_command) { 1235 case LFCNSEGWAITALL: 1236 fsidp = NULL; 1237 /* FALLSTHROUGH */ 1238 case LFCNSEGWAIT: 1239 tvp = (struct timeval *)ap->a_data; 1240 simple_lock(&fs->lfs_interlock); 1241 ++fs->lfs_sleepers; 1242 simple_unlock(&fs->lfs_interlock); 1243 VOP_UNLOCK(ap->a_vp, 0); 1244 1245 error = lfs_segwait(fsidp, tvp); 1246 1247 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1248 simple_lock(&fs->lfs_interlock); 1249 if (--fs->lfs_sleepers == 0) 1250 wakeup(&fs->lfs_sleepers); 1251 simple_unlock(&fs->lfs_interlock); 1252 return error; 1253 1254 case LFCNBMAPV: 1255 case LFCNMARKV: 1256 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0) 1257 return (error); 1258 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1259 1260 blkcnt = blkvp.blkcnt; 1261 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1262 return (EINVAL); 1263 blkiov = malloc(blkcnt * sizeof(BLOCK_INFO), M_SEGMENT, M_WAITOK); 1264 if ((error = copyin(blkvp.blkiov, blkiov, 1265 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1266 free(blkiov, M_SEGMENT); 1267 return error; 1268 } 1269 1270 simple_lock(&fs->lfs_interlock); 1271 ++fs->lfs_sleepers; 1272 simple_unlock(&fs->lfs_interlock); 1273 VOP_UNLOCK(ap->a_vp, 0); 1274 if (ap->a_command == LFCNBMAPV) 1275 error = lfs_bmapv(p, fsidp, blkiov, blkcnt); 1276 else /* LFCNMARKV */ 1277 error = lfs_markv(p, fsidp, blkiov, blkcnt); 1278 if (error == 0) 1279 error = copyout(blkiov, blkvp.blkiov, 1280 blkcnt * sizeof(BLOCK_INFO)); 1281 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1282 simple_lock(&fs->lfs_interlock); 1283 if (--fs->lfs_sleepers == 0) 1284 wakeup(&fs->lfs_sleepers); 1285 simple_unlock(&fs->lfs_interlock); 1286 free(blkiov, M_SEGMENT); 1287 return error; 1288 1289 case LFCNRECLAIM: 1290 /* 1291 * Flush dirops and write Ifile, allowing empty segments 1292 * to be immediately reclaimed. 1293 */ 1294 lfs_writer_enter(fs, "pndirop"); 1295 off = fs->lfs_offset; 1296 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1297 lfs_flush_dirops(fs); 1298 LFS_CLEANERINFO(cip, fs, bp); 1299 oclean = cip->clean; 1300 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1301 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1302 lfs_segunlock(fs); 1303 lfs_writer_leave(fs); 1304 1305 #ifdef DEBUG_LFS 1306 LFS_CLEANERINFO(cip, fs, bp); 1307 oclean = cip->clean; 1308 printf("lfs_fcntl: reclaim wrote %" PRId64 " blocks, cleaned " 1309 "%" PRId32 " segments (activesb %d)\n", 1310 fs->lfs_offset - off, cip->clean - oclean, 1311 fs->lfs_activesb); 1312 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1313 #endif 1314 1315 return 0; 1316 1317 default: 1318 return ufs_fcntl(v); 1319 } 1320 return 0; 1321 } 1322 1323 int 1324 lfs_getpages(void *v) 1325 { 1326 struct vop_getpages_args /* { 1327 struct vnode *a_vp; 1328 voff_t a_offset; 1329 struct vm_page **a_m; 1330 int *a_count; 1331 int a_centeridx; 1332 vm_prot_t a_access_type; 1333 int a_advice; 1334 int a_flags; 1335 } */ *ap = v; 1336 1337 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1338 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1339 return EPERM; 1340 } 1341 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1342 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1343 } 1344 1345 /* 1346 * we're relying on the fact that genfs_getpages() always read in 1347 * entire filesystem blocks. 1348 */ 1349 return genfs_getpages(v); 1350 } 1351 1352 /* 1353 * Make sure that for all pages in every block in the given range, 1354 * either all are dirty or all are clean. If any of the pages 1355 * we've seen so far are dirty, put the vnode on the paging chain, 1356 * and mark it IN_PAGING. 1357 * 1358 * If checkfirst != 0, don't check all the pages but return at the 1359 * first dirty page. 1360 */ 1361 static int 1362 check_dirty(struct lfs *fs, struct vnode *vp, 1363 off_t startoffset, off_t endoffset, off_t blkeof, 1364 int flags, int checkfirst) 1365 { 1366 int by_list; 1367 struct vm_page *curpg = NULL; /* XXX: gcc */ 1368 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1369 off_t soff = 0; /* XXX: gcc */ 1370 voff_t off; 1371 int i; 1372 int nonexistent; 1373 int any_dirty; /* number of dirty pages */ 1374 int dirty; /* number of dirty pages in a block */ 1375 int tdirty; 1376 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1377 1378 top: 1379 by_list = (vp->v_uobj.uo_npages <= 1380 ((endoffset - startoffset) >> PAGE_SHIFT) * 1381 UVM_PAGE_HASH_PENALTY); 1382 any_dirty = 0; 1383 1384 if (by_list) { 1385 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1386 } else { 1387 soff = startoffset; 1388 } 1389 while (by_list || soff < MIN(blkeof, endoffset)) { 1390 if (by_list) { 1391 /* 1392 * find the first page in a block. 1393 */ 1394 if (pages_per_block > 1) { 1395 while (curpg && (curpg->offset & fs->lfs_bmask)) 1396 curpg = TAILQ_NEXT(curpg, listq); 1397 } 1398 if (curpg == NULL) 1399 break; 1400 soff = curpg->offset; 1401 } 1402 1403 /* 1404 * Mark all pages in extended range busy; find out if any 1405 * of them are dirty. 1406 */ 1407 nonexistent = dirty = 0; 1408 for (i = 0; i == 0 || i < pages_per_block; i++) { 1409 if (by_list && pages_per_block <= 1) { 1410 pgs[i] = pg = curpg; 1411 } else { 1412 off = soff + (i << PAGE_SHIFT); 1413 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1414 if (pg == NULL) { 1415 ++nonexistent; 1416 continue; 1417 } 1418 } 1419 KASSERT(pg != NULL); 1420 while (pg->flags & PG_BUSY) { 1421 pg->flags |= PG_WANTED; 1422 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1423 "lfsput", 0); 1424 simple_lock(&vp->v_interlock); 1425 if (by_list) { 1426 if (i > 0) 1427 uvm_page_unbusy(pgs, i); 1428 goto top; 1429 } 1430 } 1431 pg->flags |= PG_BUSY; 1432 UVM_PAGE_OWN(pg, "lfs_putpages"); 1433 1434 pmap_page_protect(pg, VM_PROT_NONE); 1435 tdirty = (pmap_clear_modify(pg) || 1436 (pg->flags & PG_CLEAN) == 0); 1437 dirty += tdirty; 1438 } 1439 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1440 if (by_list) { 1441 curpg = TAILQ_NEXT(curpg, listq); 1442 } else { 1443 soff += fs->lfs_bsize; 1444 } 1445 continue; 1446 } 1447 1448 any_dirty += dirty; 1449 KASSERT(nonexistent == 0); 1450 1451 /* 1452 * If any are dirty make all dirty; unbusy them, 1453 * but if we were asked to clean, wire them so that 1454 * the pagedaemon doesn't bother us about them while 1455 * they're on their way to disk. 1456 */ 1457 for (i = 0; i == 0 || i < pages_per_block; i++) { 1458 pg = pgs[i]; 1459 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1460 if (dirty) { 1461 pg->flags &= ~PG_CLEAN; 1462 if (flags & PGO_FREE) { 1463 /* XXXUBC need better way to update */ 1464 simple_lock(&lfs_subsys_lock); 1465 lfs_subsys_pages += MIN(1, pages_per_block); 1466 simple_unlock(&lfs_subsys_lock); 1467 /* 1468 * Wire the page so that 1469 * pdaemon doesn't see it again. 1470 */ 1471 uvm_lock_pageq(); 1472 uvm_pagewire(pg); 1473 uvm_unlock_pageq(); 1474 1475 /* Suspended write flag */ 1476 pg->flags |= PG_DELWRI; 1477 } 1478 } 1479 if (pg->flags & PG_WANTED) 1480 wakeup(pg); 1481 pg->flags &= ~(PG_WANTED|PG_BUSY); 1482 UVM_PAGE_OWN(pg, NULL); 1483 } 1484 1485 if (checkfirst && any_dirty) 1486 break; 1487 1488 if (by_list) { 1489 curpg = TAILQ_NEXT(curpg, listq); 1490 } else { 1491 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1492 } 1493 } 1494 1495 /* 1496 * If any pages were dirty, mark this inode as "pageout requested", 1497 * and put it on the paging queue. 1498 * XXXUBC locking (check locking on dchainhd too) 1499 */ 1500 #ifdef notyet 1501 if (any_dirty) { 1502 if (!(ip->i_flags & IN_PAGING)) { 1503 ip->i_flags |= IN_PAGING; 1504 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1505 } 1506 } 1507 #endif 1508 return any_dirty; 1509 } 1510 1511 /* 1512 * lfs_putpages functions like genfs_putpages except that 1513 * 1514 * (1) It needs to bounds-check the incoming requests to ensure that 1515 * they are block-aligned; if they are not, expand the range and 1516 * do the right thing in case, e.g., the requested range is clean 1517 * but the expanded range is dirty. 1518 * (2) It needs to explicitly send blocks to be written when it is done. 1519 * VOP_PUTPAGES is not ever called with the seglock held, so 1520 * we simply take the seglock and let lfs_segunlock wait for us. 1521 * XXX Actually we can be called with the seglock held, if we have 1522 * XXX to flush a vnode while lfs_markv is in operation. As of this 1523 * XXX writing we panic in this case. 1524 * 1525 * Assumptions: 1526 * 1527 * (1) The caller does not hold any pages in this vnode busy. If it does, 1528 * there is a danger that when we expand the page range and busy the 1529 * pages we will deadlock. 1530 * (2) We are called with vp->v_interlock held; we must return with it 1531 * released. 1532 * (3) We don't absolutely have to free pages right away, provided that 1533 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1534 * us a request with PGO_FREE, we take the pages out of the paging 1535 * queue and wake up the writer, which will handle freeing them for us. 1536 * 1537 * We ensure that for any filesystem block, all pages for that 1538 * block are either resident or not, even if those pages are higher 1539 * than EOF; that means that we will be getting requests to free 1540 * "unused" pages above EOF all the time, and should ignore them. 1541 * 1542 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1543 */ 1544 1545 int 1546 lfs_putpages(void *v) 1547 { 1548 int error; 1549 struct vop_putpages_args /* { 1550 struct vnode *a_vp; 1551 voff_t a_offlo; 1552 voff_t a_offhi; 1553 int a_flags; 1554 } */ *ap = v; 1555 struct vnode *vp; 1556 struct inode *ip; 1557 struct lfs *fs; 1558 struct segment *sp; 1559 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1560 off_t off, max_endoffset; 1561 int s; 1562 boolean_t seglocked, sync, pagedaemon; 1563 struct vm_page *pg; 1564 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1565 1566 vp = ap->a_vp; 1567 ip = VTOI(vp); 1568 fs = ip->i_lfs; 1569 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1570 pagedaemon = (curproc == uvm.pagedaemon_proc); 1571 1572 /* Putpages does nothing for metadata. */ 1573 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1574 simple_unlock(&vp->v_interlock); 1575 return 0; 1576 } 1577 1578 /* 1579 * If there are no pages, don't do anything. 1580 */ 1581 if (vp->v_uobj.uo_npages == 0) { 1582 s = splbio(); 1583 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1584 (vp->v_flag & VONWORKLST)) { 1585 vp->v_flag &= ~VONWORKLST; 1586 LIST_REMOVE(vp, v_synclist); 1587 } 1588 splx(s); 1589 simple_unlock(&vp->v_interlock); 1590 return 0; 1591 } 1592 1593 blkeof = blkroundup(fs, ip->i_size); 1594 1595 /* 1596 * Ignore requests to free pages past EOF but in the same block 1597 * as EOF, unless the request is synchronous. (XXX why sync?) 1598 * XXXUBC Make these pages look "active" so the pagedaemon won't 1599 * XXXUBC bother us with them again. 1600 */ 1601 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1602 origoffset = ap->a_offlo; 1603 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1604 pg = uvm_pagelookup(&vp->v_uobj, off); 1605 KASSERT(pg != NULL); 1606 while (pg->flags & PG_BUSY) { 1607 pg->flags |= PG_WANTED; 1608 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1609 "lfsput2", 0); 1610 simple_lock(&vp->v_interlock); 1611 } 1612 uvm_lock_pageq(); 1613 uvm_pageactivate(pg); 1614 uvm_unlock_pageq(); 1615 } 1616 ap->a_offlo = blkeof; 1617 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1618 simple_unlock(&vp->v_interlock); 1619 return 0; 1620 } 1621 } 1622 1623 /* 1624 * Extend page range to start and end at block boundaries. 1625 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1626 */ 1627 origoffset = ap->a_offlo; 1628 origendoffset = ap->a_offhi; 1629 startoffset = origoffset & ~(fs->lfs_bmask); 1630 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1631 << fs->lfs_bshift; 1632 1633 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1634 endoffset = max_endoffset; 1635 origendoffset = endoffset; 1636 } else { 1637 origendoffset = round_page(ap->a_offhi); 1638 endoffset = round_page(blkroundup(fs, origendoffset)); 1639 } 1640 1641 KASSERT(startoffset > 0 || endoffset >= startoffset); 1642 if (startoffset == endoffset) { 1643 /* Nothing to do, why were we called? */ 1644 simple_unlock(&vp->v_interlock); 1645 #ifdef DEBUG 1646 printf("lfs_putpages: startoffset = endoffset = %" PRId64 "\n", 1647 startoffset); 1648 #endif 1649 return 0; 1650 } 1651 1652 ap->a_offlo = startoffset; 1653 ap->a_offhi = endoffset; 1654 1655 if (!(ap->a_flags & PGO_CLEANIT)) 1656 return genfs_putpages(v); 1657 1658 /* 1659 * If there are more than one page per block, we don't want 1660 * to get caught locking them backwards; so set PGO_BUSYFAIL 1661 * to avoid deadlocks. 1662 */ 1663 ap->a_flags |= PGO_BUSYFAIL; 1664 1665 do { 1666 int r; 1667 1668 /* If no pages are dirty, we can just use genfs_putpages. */ 1669 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 1670 ap->a_flags, 1) != 0) 1671 break; 1672 1673 if ((r = genfs_putpages(v)) != EDEADLK) 1674 return r; 1675 1676 /* Start over. */ 1677 preempt(1); 1678 simple_lock(&vp->v_interlock); 1679 } while(1); 1680 1681 /* 1682 * Dirty and asked to clean. 1683 * 1684 * Pagedaemon can't actually write LFS pages; wake up 1685 * the writer to take care of that. The writer will 1686 * notice the pager inode queue and act on that. 1687 */ 1688 if (pagedaemon) { 1689 ++fs->lfs_pdflush; 1690 wakeup(&lfs_writer_daemon); 1691 simple_unlock(&vp->v_interlock); 1692 return EWOULDBLOCK; 1693 } 1694 1695 /* 1696 * If this is a file created in a recent dirop, we can't flush its 1697 * inode until the dirop is complete. Drain dirops, then flush the 1698 * filesystem (taking care of any other pending dirops while we're 1699 * at it). 1700 */ 1701 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1702 (vp->v_flag & VDIROP)) { 1703 int locked; 1704 1705 /* printf("putpages to clean VDIROP, flushing\n"); */ 1706 lfs_writer_enter(fs, "ppdirop"); 1707 locked = VOP_ISLOCKED(vp) && /* XXX */ 1708 vp->v_lock.lk_lockholder == curproc->p_pid; 1709 if (locked) 1710 VOP_UNLOCK(vp, 0); 1711 simple_unlock(&vp->v_interlock); 1712 1713 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1714 1715 simple_lock(&vp->v_interlock); 1716 if (locked) 1717 VOP_LOCK(vp, LK_EXCLUSIVE); 1718 lfs_writer_leave(fs); 1719 1720 /* XXX the flush should have taken care of this one too! */ 1721 } 1722 1723 /* 1724 * This is it. We are going to write some pages. From here on 1725 * down it's all just mechanics. 1726 * 1727 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1728 */ 1729 ap->a_flags &= ~PGO_SYNCIO; 1730 1731 /* 1732 * If we've already got the seglock, flush the node and return. 1733 * The FIP has already been set up for us by lfs_writefile, 1734 * and FIP cleanup and lfs_updatemeta will also be done there, 1735 * unless genfs_putpages returns EDEADLK; then we must flush 1736 * what we have, and correct FIP and segment header accounting. 1737 */ 1738 1739 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 1740 if (!seglocked) { 1741 simple_unlock(&vp->v_interlock); 1742 /* 1743 * Take the seglock, because we are going to be writing pages. 1744 */ 1745 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 1746 if (error != 0) 1747 return error; 1748 simple_lock(&vp->v_interlock); 1749 } 1750 1751 /* 1752 * VOP_PUTPAGES should not be called while holding the seglock. 1753 * XXXUBC fix lfs_markv, or do this properly. 1754 */ 1755 /* KASSERT(fs->lfs_seglock == 1); */ 1756 1757 /* 1758 * We assume we're being called with sp->fip pointing at blank space. 1759 * Account for a new FIP in the segment header, and set sp->vp. 1760 * (This should duplicate the setup at the top of lfs_writefile().) 1761 */ 1762 sp = fs->lfs_sp; 1763 if (!seglocked) { 1764 if (sp->seg_bytes_left < fs->lfs_bsize || 1765 sp->sum_bytes_left < sizeof(struct finfo)) 1766 (void) lfs_writeseg(fs, fs->lfs_sp); 1767 1768 sp->sum_bytes_left -= FINFOSIZE; 1769 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1770 } 1771 KASSERT(sp->vp == NULL); 1772 sp->vp = vp; 1773 1774 if (!seglocked) { 1775 if (vp->v_flag & VDIROP) 1776 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 1777 } 1778 1779 sp->fip->fi_nblocks = 0; 1780 sp->fip->fi_ino = ip->i_number; 1781 sp->fip->fi_version = ip->i_gen; 1782 1783 /* 1784 * Loop through genfs_putpages until all pages are gathered. 1785 * genfs_putpages() drops the interlock, so reacquire it if necessary. 1786 * Whenever we lose the interlock we have to rerun check_dirty, as 1787 * well. 1788 */ 1789 again: 1790 check_dirty(fs, vp, startoffset, endoffset, blkeof, ap->a_flags, 0); 1791 1792 if ((error = genfs_putpages(v)) == EDEADLK) { 1793 #ifdef DEBUG_LFS 1794 printf("lfs_putpages: genfs_putpages returned EDEADLK [2]" 1795 " ino %d off %x (seg %d)\n", 1796 ip->i_number, fs->lfs_offset, 1797 dtosn(fs, fs->lfs_offset)); 1798 #endif 1799 /* If nothing to write, short-circuit */ 1800 if (sp->cbpp - sp->bpp > 1) { 1801 /* Write gathered pages */ 1802 lfs_updatemeta(sp); 1803 (void) lfs_writeseg(fs, sp); 1804 1805 /* 1806 * Reinitialize brand new FIP and add us to it. 1807 * (This should duplicate the fixup in 1808 * lfs_gatherpages().) 1809 */ 1810 KASSERT(sp->vp == vp); 1811 sp->fip->fi_version = ip->i_gen; 1812 sp->fip->fi_ino = ip->i_number; 1813 /* Add us to the new segment summary. */ 1814 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1815 sp->sum_bytes_left -= FINFOSIZE; 1816 } 1817 1818 /* Give the write a chance to complete */ 1819 preempt(1); 1820 1821 /* We've lost the interlock. Start over. */ 1822 simple_lock(&vp->v_interlock); 1823 goto again; 1824 } 1825 1826 KASSERT(sp->vp == vp); 1827 if (!seglocked) { 1828 sp->vp = NULL; /* XXX lfs_gather below will set this */ 1829 1830 /* Write indirect blocks as well */ 1831 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 1832 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 1833 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 1834 1835 KASSERT(sp->vp == NULL); 1836 sp->vp = vp; 1837 } 1838 1839 /* 1840 * Blocks are now gathered into a segment waiting to be written. 1841 * All that's left to do is update metadata, and write them. 1842 */ 1843 lfs_updatemeta(sp); 1844 KASSERT(sp->vp == vp); 1845 sp->vp = NULL; 1846 1847 if (seglocked) { 1848 /* we're called by lfs_writefile. */ 1849 return error; 1850 } 1851 1852 /* 1853 * Clean up FIP, since we're done writing this file. 1854 * This should duplicate cleanup at the end of lfs_writefile(). 1855 */ 1856 if (sp->fip->fi_nblocks != 0) { 1857 sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE + 1858 sizeof(int32_t) * sp->fip->fi_nblocks); 1859 sp->start_lbp = &sp->fip->fi_blocks[0]; 1860 } else { 1861 sp->sum_bytes_left += FINFOSIZE; 1862 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 1863 } 1864 lfs_writeseg(fs, fs->lfs_sp); 1865 1866 /* 1867 * XXX - with the malloc/copy writeseg, the pages are freed by now 1868 * even if we don't wait (e.g. if we hold a nested lock). This 1869 * will not be true if we stop using malloc/copy. 1870 */ 1871 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 1872 lfs_segunlock(fs); 1873 1874 /* 1875 * Wait for v_numoutput to drop to zero. The seglock should 1876 * take care of this, but there is a slight possibility that 1877 * aiodoned might not have got around to our buffers yet. 1878 */ 1879 if (sync) { 1880 int s; 1881 1882 s = splbio(); 1883 simple_lock(&global_v_numoutput_slock); 1884 while (vp->v_numoutput > 0) { 1885 #ifdef DEBUG 1886 printf("ino %d sleeping on num %d\n", 1887 ip->i_number, vp->v_numoutput); 1888 #endif 1889 vp->v_flag |= VBWAIT; 1890 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 1891 &global_v_numoutput_slock); 1892 } 1893 simple_unlock(&global_v_numoutput_slock); 1894 splx(s); 1895 } 1896 return error; 1897 } 1898 1899 /* 1900 * Return the last logical file offset that should be written for this file 1901 * if we're doing a write that ends at "size". If writing, we need to know 1902 * about sizes on disk, i.e. fragments if there are any; if reading, we need 1903 * to know about entire blocks. 1904 */ 1905 void 1906 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 1907 { 1908 struct inode *ip = VTOI(vp); 1909 struct lfs *fs = ip->i_lfs; 1910 daddr_t olbn, nlbn; 1911 1912 KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)); 1913 KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)) 1914 != (GOP_SIZE_READ | GOP_SIZE_WRITE)); 1915 1916 olbn = lblkno(fs, ip->i_size); 1917 nlbn = lblkno(fs, size); 1918 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 1919 *eobp = fragroundup(fs, size); 1920 } else { 1921 *eobp = blkroundup(fs, size); 1922 } 1923 } 1924 1925 #ifdef DEBUG 1926 void lfs_dump_vop(void *); 1927 1928 void 1929 lfs_dump_vop(void *v) 1930 { 1931 struct vop_putpages_args /* { 1932 struct vnode *a_vp; 1933 voff_t a_offlo; 1934 voff_t a_offhi; 1935 int a_flags; 1936 } */ *ap = v; 1937 1938 #ifdef DDB 1939 vfs_vnode_print(ap->a_vp, 0, printf); 1940 #endif 1941 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 1942 } 1943 #endif 1944 1945 int 1946 lfs_mmap(void *v) 1947 { 1948 struct vop_mmap_args /* { 1949 const struct vnodeop_desc *a_desc; 1950 struct vnode *a_vp; 1951 int a_fflags; 1952 struct ucred *a_cred; 1953 struct lwp *a_l; 1954 } */ *ap = v; 1955 1956 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 1957 return EOPNOTSUPP; 1958 return ufs_mmap(v); 1959 } 1960
Indexes created Sun Oct 19 02:09:48 GMT 2025