lfs_vnops.c revision 1.112
1 /* $NetBSD: lfs_vnops.c,v 1.112 2003/07/12 16:17:09 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. All advertising materials mentioning features or use of this software 51 * must display the following acknowledgement: 52 * This product includes software developed by the University of 53 * California, Berkeley and its contributors. 54 * 4. Neither the name of the University nor the names of its contributors 55 * may be used to endorse or promote products derived from this software 56 * without specific prior written permission. 57 * 58 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 59 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 60 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 61 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 62 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 63 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 64 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 65 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 66 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 67 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 68 * SUCH DAMAGE. 69 * 70 * @(#)lfs_vnops.c 8.13 (Berkeley) 6/10/95 71 */ 72 73 #include <sys/cdefs.h> 74 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.112 2003/07/12 16:17:09 yamt Exp $"); 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/namei.h> 79 #include <sys/resourcevar.h> 80 #include <sys/kernel.h> 81 #include <sys/file.h> 82 #include <sys/stat.h> 83 #include <sys/buf.h> 84 #include <sys/proc.h> 85 #include <sys/mount.h> 86 #include <sys/vnode.h> 87 #include <sys/malloc.h> 88 #include <sys/pool.h> 89 #include <sys/signalvar.h> 90 91 #include <miscfs/fifofs/fifo.h> 92 #include <miscfs/genfs/genfs.h> 93 #include <miscfs/specfs/specdev.h> 94 95 #include <ufs/ufs/inode.h> 96 #include <ufs/ufs/dir.h> 97 #include <ufs/ufs/ufsmount.h> 98 #include <ufs/ufs/ufs_extern.h> 99 100 #include <uvm/uvm.h> 101 #include <uvm/uvm_pmap.h> 102 #include <uvm/uvm_stat.h> 103 #include <uvm/uvm_pager.h> 104 105 #include <ufs/lfs/lfs.h> 106 #include <ufs/lfs/lfs_extern.h> 107 108 extern pid_t lfs_writer_daemon; 109 110 /* Global vfs data structures for lfs. */ 111 int (**lfs_vnodeop_p)(void *); 112 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 113 { &vop_default_desc, vn_default_error }, 114 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 115 { &vop_create_desc, lfs_create }, /* create */ 116 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 117 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 118 { &vop_open_desc, ufs_open }, /* open */ 119 { &vop_close_desc, lfs_close }, /* close */ 120 { &vop_access_desc, ufs_access }, /* access */ 121 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 122 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 123 { &vop_read_desc, lfs_read }, /* read */ 124 { &vop_write_desc, lfs_write }, /* write */ 125 { &vop_lease_desc, ufs_lease_check }, /* lease */ 126 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 127 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 128 { &vop_poll_desc, ufs_poll }, /* poll */ 129 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 130 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 131 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 132 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 133 { &vop_seek_desc, ufs_seek }, /* seek */ 134 { &vop_remove_desc, lfs_remove }, /* remove */ 135 { &vop_link_desc, lfs_link }, /* link */ 136 { &vop_rename_desc, lfs_rename }, /* rename */ 137 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 138 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 139 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 140 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 141 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 142 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 143 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 144 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 145 { &vop_lock_desc, ufs_lock }, /* lock */ 146 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 147 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 148 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 149 { &vop_print_desc, ufs_print }, /* print */ 150 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 151 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 152 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 153 { &vop_blkatoff_desc, lfs_blkatoff }, /* blkatoff */ 154 { &vop_valloc_desc, lfs_valloc }, /* valloc */ 155 { &vop_balloc_desc, lfs_balloc }, /* balloc */ 156 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 157 { &vop_truncate_desc, lfs_truncate }, /* truncate */ 158 { &vop_update_desc, lfs_update }, /* update */ 159 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 160 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 161 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 162 { NULL, NULL } 163 }; 164 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 165 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 166 167 int (**lfs_specop_p)(void *); 168 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 169 { &vop_default_desc, vn_default_error }, 170 { &vop_lookup_desc, spec_lookup }, /* lookup */ 171 { &vop_create_desc, spec_create }, /* create */ 172 { &vop_mknod_desc, spec_mknod }, /* mknod */ 173 { &vop_open_desc, spec_open }, /* open */ 174 { &vop_close_desc, lfsspec_close }, /* close */ 175 { &vop_access_desc, ufs_access }, /* access */ 176 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 177 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 178 { &vop_read_desc, ufsspec_read }, /* read */ 179 { &vop_write_desc, ufsspec_write }, /* write */ 180 { &vop_lease_desc, spec_lease_check }, /* lease */ 181 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 182 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 183 { &vop_poll_desc, spec_poll }, /* poll */ 184 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 185 { &vop_revoke_desc, spec_revoke }, /* revoke */ 186 { &vop_mmap_desc, spec_mmap }, /* mmap */ 187 { &vop_fsync_desc, spec_fsync }, /* fsync */ 188 { &vop_seek_desc, spec_seek }, /* seek */ 189 { &vop_remove_desc, spec_remove }, /* remove */ 190 { &vop_link_desc, spec_link }, /* link */ 191 { &vop_rename_desc, spec_rename }, /* rename */ 192 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 193 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 194 { &vop_symlink_desc, spec_symlink }, /* symlink */ 195 { &vop_readdir_desc, spec_readdir }, /* readdir */ 196 { &vop_readlink_desc, spec_readlink }, /* readlink */ 197 { &vop_abortop_desc, spec_abortop }, /* abortop */ 198 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 199 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 200 { &vop_lock_desc, ufs_lock }, /* lock */ 201 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 202 { &vop_bmap_desc, spec_bmap }, /* bmap */ 203 { &vop_strategy_desc, spec_strategy }, /* strategy */ 204 { &vop_print_desc, ufs_print }, /* print */ 205 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 206 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 207 { &vop_advlock_desc, spec_advlock }, /* advlock */ 208 { &vop_blkatoff_desc, spec_blkatoff }, /* blkatoff */ 209 { &vop_valloc_desc, spec_valloc }, /* valloc */ 210 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 211 { &vop_truncate_desc, spec_truncate }, /* truncate */ 212 { &vop_update_desc, lfs_update }, /* update */ 213 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 214 { &vop_getpages_desc, spec_getpages }, /* getpages */ 215 { &vop_putpages_desc, spec_putpages }, /* putpages */ 216 { NULL, NULL } 217 }; 218 const struct vnodeopv_desc lfs_specop_opv_desc = 219 { &lfs_specop_p, lfs_specop_entries }; 220 221 int (**lfs_fifoop_p)(void *); 222 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 223 { &vop_default_desc, vn_default_error }, 224 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 225 { &vop_create_desc, fifo_create }, /* create */ 226 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 227 { &vop_open_desc, fifo_open }, /* open */ 228 { &vop_close_desc, lfsfifo_close }, /* close */ 229 { &vop_access_desc, ufs_access }, /* access */ 230 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 231 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 232 { &vop_read_desc, ufsfifo_read }, /* read */ 233 { &vop_write_desc, ufsfifo_write }, /* write */ 234 { &vop_lease_desc, fifo_lease_check }, /* lease */ 235 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 236 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 237 { &vop_poll_desc, fifo_poll }, /* poll */ 238 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 239 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 240 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 241 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 242 { &vop_seek_desc, fifo_seek }, /* seek */ 243 { &vop_remove_desc, fifo_remove }, /* remove */ 244 { &vop_link_desc, fifo_link }, /* link */ 245 { &vop_rename_desc, fifo_rename }, /* rename */ 246 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 247 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 248 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 249 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 250 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 251 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 252 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 253 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 254 { &vop_lock_desc, ufs_lock }, /* lock */ 255 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 256 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 257 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 258 { &vop_print_desc, ufs_print }, /* print */ 259 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 260 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 261 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 262 { &vop_blkatoff_desc, fifo_blkatoff }, /* blkatoff */ 263 { &vop_valloc_desc, fifo_valloc }, /* valloc */ 264 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 265 { &vop_truncate_desc, fifo_truncate }, /* truncate */ 266 { &vop_update_desc, lfs_update }, /* update */ 267 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 268 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 269 { NULL, NULL } 270 }; 271 const struct vnodeopv_desc lfs_fifoop_opv_desc = 272 { &lfs_fifoop_p, lfs_fifoop_entries }; 273 274 /* 275 * A function version of LFS_ITIMES, for the UFS functions which call ITIMES 276 */ 277 void 278 lfs_itimes(struct inode *ip, struct timespec *acc, struct timespec *mod, struct timespec *cre) 279 { 280 LFS_ITIMES(ip, acc, mod, cre); 281 } 282 283 #define LFS_READWRITE 284 #include <ufs/ufs/ufs_readwrite.c> 285 #undef LFS_READWRITE 286 287 /* 288 * Synch an open file. 289 */ 290 /* ARGSUSED */ 291 int 292 lfs_fsync(void *v) 293 { 294 struct vop_fsync_args /* { 295 struct vnode *a_vp; 296 struct ucred *a_cred; 297 int a_flags; 298 off_t offlo; 299 off_t offhi; 300 struct proc *a_p; 301 } */ *ap = v; 302 struct vnode *vp = ap->a_vp; 303 int error, wait; 304 305 /* 306 * Trickle sync checks for need to do a checkpoint after possible 307 * activity from the pagedaemon. 308 */ 309 if (ap->a_flags & FSYNC_LAZY) { 310 wakeup(&lfs_writer_daemon); 311 return 0; 312 } 313 314 wait = (ap->a_flags & FSYNC_WAIT); 315 simple_lock(&vp->v_interlock); 316 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 317 round_page(ap->a_offhi), 318 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 319 if (error) 320 return error; 321 error = VOP_UPDATE(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 322 if (wait && !VPISEMPTY(vp)) 323 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 324 325 return error; 326 } 327 328 /* 329 * Take IN_ADIROP off, then call ufs_inactive. 330 */ 331 int 332 lfs_inactive(void *v) 333 { 334 struct vop_inactive_args /* { 335 struct vnode *a_vp; 336 struct proc *a_p; 337 } */ *ap = v; 338 339 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 340 341 lfs_unmark_vnode(ap->a_vp); 342 343 /* 344 * The Ifile is only ever inactivated on unmount. 345 * Streamline this process by not giving it more dirty blocks. 346 */ 347 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 348 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 349 VOP_UNLOCK(ap->a_vp, 0); 350 return 0; 351 } 352 353 return ufs_inactive(v); 354 } 355 356 /* 357 * These macros are used to bracket UFS directory ops, so that we can 358 * identify all the pages touched during directory ops which need to 359 * be ordered and flushed atomically, so that they may be recovered. 360 */ 361 /* 362 * XXX KS - Because we have to mark nodes VDIROP in order to prevent 363 * the cache from reclaiming them while a dirop is in progress, we must 364 * also manage the number of nodes so marked (otherwise we can run out). 365 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 366 * is decremented during segment write, when VDIROP is taken off. 367 */ 368 #define SET_DIROP(vp) SET_DIROP2((vp), NULL) 369 #define SET_DIROP2(vp, vp2) lfs_set_dirop((vp), (vp2)) 370 static int lfs_set_dirop(struct vnode *, struct vnode *); 371 extern int lfs_do_flush; 372 373 #define NRESERVE(fs) (btofsb(fs, (NIADDR + 3 + (2 * NIADDR + 3)) << fs->lfs_bshift)) 374 375 static int 376 lfs_set_dirop(struct vnode *vp, struct vnode *vp2) 377 { 378 struct lfs *fs; 379 int error; 380 381 KASSERT(VOP_ISLOCKED(vp)); 382 KASSERT(vp2 == NULL || VOP_ISLOCKED(vp2)); 383 384 fs = VTOI(vp)->i_lfs; 385 /* 386 * We might need one directory block plus supporting indirect blocks, 387 * plus an inode block and ifile page for the new vnode. 388 */ 389 if ((error = lfs_reserve(fs, vp, vp2, NRESERVE(fs))) != 0) 390 return (error); 391 392 if (fs->lfs_dirops == 0) 393 lfs_check(vp, LFS_UNUSED_LBN, 0); 394 while (fs->lfs_writer || lfs_dirvcount > LFS_MAX_DIROP) { 395 if (fs->lfs_writer) 396 tsleep(&fs->lfs_dirops, PRIBIO + 1, "lfs_sdirop", 0); 397 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 398 wakeup(&lfs_writer_daemon); 399 preempt(NULL); 400 } 401 402 if (lfs_dirvcount > LFS_MAX_DIROP) { 403 #ifdef DEBUG_LFS 404 printf("lfs_set_dirop: sleeping with dirops=%d, " 405 "dirvcount=%d\n", fs->lfs_dirops, 406 lfs_dirvcount); 407 #endif 408 if ((error = tsleep(&lfs_dirvcount, PCATCH|PUSER, 409 "lfs_maxdirop", 0)) != 0) { 410 goto unreserve; 411 } 412 } 413 } 414 ++fs->lfs_dirops; 415 fs->lfs_doifile = 1; 416 417 /* Hold a reference so SET_ENDOP will be happy */ 418 vref(vp); 419 if (vp2) 420 vref(vp2); 421 422 return 0; 423 424 unreserve: 425 lfs_reserve(fs, vp, vp2, -NRESERVE(fs)); 426 return error; 427 } 428 429 #define SET_ENDOP(fs, vp, str) SET_ENDOP2((fs), (vp), NULL, (str)) 430 #define SET_ENDOP2(fs, vp, vp2, str) { \ 431 --(fs)->lfs_dirops; \ 432 if (!(fs)->lfs_dirops) { \ 433 if ((fs)->lfs_nadirop) { \ 434 panic("SET_ENDOP: %s: no dirops but nadirop=%d", \ 435 (str), (fs)->lfs_nadirop); \ 436 } \ 437 wakeup(&(fs)->lfs_writer); \ 438 lfs_check((vp),LFS_UNUSED_LBN,0); \ 439 } \ 440 lfs_reserve((fs), vp, vp2, -NRESERVE(fs)); /* XXX */ \ 441 vrele(vp); \ 442 if (vp2) \ 443 vrele(vp2); \ 444 } 445 446 #define MARK_VNODE(dvp) do { \ 447 struct inode *_ip = VTOI(dvp); \ 448 struct lfs *_fs = _ip->i_lfs; \ 449 \ 450 if (!((dvp)->v_flag & VDIROP)) { \ 451 (void)lfs_vref(dvp); \ 452 ++lfs_dirvcount; \ 453 TAILQ_INSERT_TAIL(&_fs->lfs_dchainhd, _ip, i_lfs_dchain); \ 454 } \ 455 (dvp)->v_flag |= VDIROP; \ 456 if (!(_ip->i_flag & IN_ADIROP)) { \ 457 ++_fs->lfs_nadirop; \ 458 } \ 459 _ip->i_flag |= IN_ADIROP; \ 460 } while (0) 461 462 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 463 464 void lfs_unmark_vnode(struct vnode *vp) 465 { 466 struct inode *ip; 467 468 ip = VTOI(vp); 469 470 if (ip->i_flag & IN_ADIROP) 471 --ip->i_lfs->lfs_nadirop; 472 ip->i_flag &= ~IN_ADIROP; 473 } 474 475 int 476 lfs_symlink(void *v) 477 { 478 struct vop_symlink_args /* { 479 struct vnode *a_dvp; 480 struct vnode **a_vpp; 481 struct componentname *a_cnp; 482 struct vattr *a_vap; 483 char *a_target; 484 } */ *ap = v; 485 int error; 486 487 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 488 vput(ap->a_dvp); 489 return error; 490 } 491 MARK_VNODE(ap->a_dvp); 492 error = ufs_symlink(ap); 493 UNMARK_VNODE(ap->a_dvp); 494 if (*(ap->a_vpp)) 495 UNMARK_VNODE(*(ap->a_vpp)); 496 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"symlink"); 497 return (error); 498 } 499 500 int 501 lfs_mknod(void *v) 502 { 503 struct vop_mknod_args /* { 504 struct vnode *a_dvp; 505 struct vnode **a_vpp; 506 struct componentname *a_cnp; 507 struct vattr *a_vap; 508 } */ *ap = v; 509 struct vattr *vap = ap->a_vap; 510 struct vnode **vpp = ap->a_vpp; 511 struct inode *ip; 512 int error; 513 struct mount *mp; 514 ino_t ino; 515 516 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 517 vput(ap->a_dvp); 518 return error; 519 } 520 MARK_VNODE(ap->a_dvp); 521 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 522 ap->a_dvp, vpp, ap->a_cnp); 523 UNMARK_VNODE(ap->a_dvp); 524 if (*(ap->a_vpp)) 525 UNMARK_VNODE(*(ap->a_vpp)); 526 527 /* Either way we're done with the dirop at this point */ 528 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"mknod"); 529 530 if (error) 531 return (error); 532 533 ip = VTOI(*vpp); 534 mp = (*vpp)->v_mount; 535 ino = ip->i_number; 536 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 537 if (vap->va_rdev != VNOVAL) { 538 /* 539 * Want to be able to use this to make badblock 540 * inodes, so don't truncate the dev number. 541 */ 542 #if 0 543 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 544 UFS_MPNEEDSWAP((*vpp)->v_mount)); 545 #else 546 ip->i_ffs1_rdev = vap->va_rdev; 547 #endif 548 } 549 /* 550 * Call fsync to write the vnode so that we don't have to deal with 551 * flushing it when it's marked VDIROP|VXLOCK. 552 * 553 * XXX KS - If we can't flush we also can't call vgone(), so must 554 * return. But, that leaves this vnode in limbo, also not good. 555 * Can this ever happen (barring hardware failure)? 556 */ 557 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 558 curproc)) != 0) { 559 printf("Couldn't fsync in mknod (ino %d)---what do I do?\n", 560 VTOI(*vpp)->i_number); 561 return (error); 562 } 563 /* 564 * Remove vnode so that it will be reloaded by VFS_VGET and 565 * checked to see if it is an alias of an existing entry in 566 * the inode cache. 567 */ 568 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 569 VOP_UNLOCK(*vpp, 0); 570 lfs_vunref(*vpp); 571 (*vpp)->v_type = VNON; 572 vgone(*vpp); 573 error = VFS_VGET(mp, ino, vpp); 574 if (error != 0) { 575 *vpp = NULL; 576 return (error); 577 } 578 return (0); 579 } 580 581 int 582 lfs_create(void *v) 583 { 584 struct vop_create_args /* { 585 struct vnode *a_dvp; 586 struct vnode **a_vpp; 587 struct componentname *a_cnp; 588 struct vattr *a_vap; 589 } */ *ap = v; 590 int error; 591 592 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 593 vput(ap->a_dvp); 594 return error; 595 } 596 MARK_VNODE(ap->a_dvp); 597 error = ufs_create(ap); 598 UNMARK_VNODE(ap->a_dvp); 599 if (*(ap->a_vpp)) 600 UNMARK_VNODE(*(ap->a_vpp)); 601 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"create"); 602 return (error); 603 } 604 605 int 606 lfs_mkdir(void *v) 607 { 608 struct vop_mkdir_args /* { 609 struct vnode *a_dvp; 610 struct vnode **a_vpp; 611 struct componentname *a_cnp; 612 struct vattr *a_vap; 613 } */ *ap = v; 614 int error; 615 616 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 617 vput(ap->a_dvp); 618 return error; 619 } 620 MARK_VNODE(ap->a_dvp); 621 error = ufs_mkdir(ap); 622 UNMARK_VNODE(ap->a_dvp); 623 if (*(ap->a_vpp)) 624 UNMARK_VNODE(*(ap->a_vpp)); 625 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"mkdir"); 626 return (error); 627 } 628 629 int 630 lfs_remove(void *v) 631 { 632 struct vop_remove_args /* { 633 struct vnode *a_dvp; 634 struct vnode *a_vp; 635 struct componentname *a_cnp; 636 } */ *ap = v; 637 struct vnode *dvp, *vp; 638 int error; 639 640 dvp = ap->a_dvp; 641 vp = ap->a_vp; 642 if ((error = SET_DIROP2(dvp, vp)) != 0) { 643 if (dvp == vp) 644 vrele(vp); 645 else 646 vput(vp); 647 vput(dvp); 648 return error; 649 } 650 MARK_VNODE(dvp); 651 MARK_VNODE(vp); 652 error = ufs_remove(ap); 653 UNMARK_VNODE(dvp); 654 UNMARK_VNODE(vp); 655 656 SET_ENDOP2(VTOI(dvp)->i_lfs, dvp, vp, "remove"); 657 return (error); 658 } 659 660 int 661 lfs_rmdir(void *v) 662 { 663 struct vop_rmdir_args /* { 664 struct vnodeop_desc *a_desc; 665 struct vnode *a_dvp; 666 struct vnode *a_vp; 667 struct componentname *a_cnp; 668 } */ *ap = v; 669 struct vnode *vp; 670 int error; 671 672 vp = ap->a_vp; 673 if ((error = SET_DIROP2(ap->a_dvp, ap->a_vp)) != 0) { 674 vrele(ap->a_dvp); 675 if (ap->a_vp != ap->a_dvp) 676 VOP_UNLOCK(ap->a_dvp, 0); 677 vput(vp); 678 return error; 679 } 680 MARK_VNODE(ap->a_dvp); 681 MARK_VNODE(vp); 682 error = ufs_rmdir(ap); 683 UNMARK_VNODE(ap->a_dvp); 684 UNMARK_VNODE(vp); 685 686 SET_ENDOP2(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vp, "rmdir"); 687 return (error); 688 } 689 690 int 691 lfs_link(void *v) 692 { 693 struct vop_link_args /* { 694 struct vnode *a_dvp; 695 struct vnode *a_vp; 696 struct componentname *a_cnp; 697 } */ *ap = v; 698 int error; 699 700 if ((error = SET_DIROP(ap->a_dvp)) != 0) { 701 vput(ap->a_dvp); 702 return error; 703 } 704 MARK_VNODE(ap->a_dvp); 705 error = ufs_link(ap); 706 UNMARK_VNODE(ap->a_dvp); 707 SET_ENDOP(VTOI(ap->a_dvp)->i_lfs,ap->a_dvp,"link"); 708 return (error); 709 } 710 711 int 712 lfs_rename(void *v) 713 { 714 struct vop_rename_args /* { 715 struct vnode *a_fdvp; 716 struct vnode *a_fvp; 717 struct componentname *a_fcnp; 718 struct vnode *a_tdvp; 719 struct vnode *a_tvp; 720 struct componentname *a_tcnp; 721 } */ *ap = v; 722 struct vnode *tvp, *fvp, *tdvp, *fdvp; 723 struct componentname *tcnp, *fcnp; 724 int error; 725 struct lfs *fs; 726 727 fs = VTOI(ap->a_fdvp)->i_lfs; 728 tvp = ap->a_tvp; 729 tdvp = ap->a_tdvp; 730 tcnp = ap->a_tcnp; 731 fvp = ap->a_fvp; 732 fdvp = ap->a_fdvp; 733 fcnp = ap->a_fcnp; 734 735 /* 736 * Check for cross-device rename. 737 * If it is, we don't want to set dirops, just error out. 738 * (In particular note that MARK_VNODE(tdvp) will DTWT on 739 * a cross-device rename.) 740 * 741 * Copied from ufs_rename. 742 */ 743 if ((fvp->v_mount != tdvp->v_mount) || 744 (tvp && (fvp->v_mount != tvp->v_mount))) { 745 error = EXDEV; 746 goto errout; 747 } 748 749 /* 750 * Check to make sure we're not renaming a vnode onto itself 751 * (deleting a hard link by renaming one name onto another); 752 * if we are we can't recursively call VOP_REMOVE since that 753 * would leave us with an unaccounted-for number of live dirops. 754 * 755 * Inline the relevant section of ufs_rename here, *before* 756 * calling SET_DIROP2. 757 */ 758 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 759 (VTOI(tdvp)->i_flags & APPEND))) { 760 error = EPERM; 761 goto errout; 762 } 763 if (fvp == tvp) { 764 if (fvp->v_type == VDIR) { 765 error = EINVAL; 766 goto errout; 767 } 768 769 /* Release destination completely. */ 770 VOP_ABORTOP(tdvp, tcnp); 771 vput(tdvp); 772 vput(tvp); 773 774 /* Delete source. */ 775 vrele(fvp); 776 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 777 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 778 fcnp->cn_nameiop = DELETE; 779 if ((error = relookup(fdvp, &fvp, fcnp))){ 780 /* relookup blew away fdvp */ 781 return (error); 782 } 783 return (VOP_REMOVE(fdvp, fvp, fcnp)); 784 } 785 786 if ((error = SET_DIROP2(tdvp, tvp)) != 0) 787 goto errout; 788 MARK_VNODE(fdvp); 789 MARK_VNODE(tdvp); 790 MARK_VNODE(fvp); 791 if (tvp) { 792 MARK_VNODE(tvp); 793 } 794 795 error = ufs_rename(ap); 796 UNMARK_VNODE(fdvp); 797 UNMARK_VNODE(tdvp); 798 UNMARK_VNODE(fvp); 799 if (tvp) { 800 UNMARK_VNODE(tvp); 801 } 802 SET_ENDOP2(fs, tdvp, tvp, "rename"); 803 return (error); 804 805 errout: 806 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 807 if (tdvp == tvp) 808 vrele(tdvp); 809 else 810 vput(tdvp); 811 if (tvp) 812 vput(tvp); 813 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 814 vrele(fdvp); 815 vrele(fvp); 816 return (error); 817 } 818 819 /* XXX hack to avoid calling ITIMES in getattr */ 820 int 821 lfs_getattr(void *v) 822 { 823 struct vop_getattr_args /* { 824 struct vnode *a_vp; 825 struct vattr *a_vap; 826 struct ucred *a_cred; 827 struct proc *a_p; 828 } */ *ap = v; 829 struct vnode *vp = ap->a_vp; 830 struct inode *ip = VTOI(vp); 831 struct vattr *vap = ap->a_vap; 832 struct lfs *fs = ip->i_lfs; 833 /* 834 * Copy from inode table 835 */ 836 vap->va_fsid = ip->i_dev; 837 vap->va_fileid = ip->i_number; 838 vap->va_mode = ip->i_mode & ~IFMT; 839 vap->va_nlink = ip->i_nlink; 840 vap->va_uid = ip->i_uid; 841 vap->va_gid = ip->i_gid; 842 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 843 vap->va_size = vp->v_size; 844 vap->va_atime.tv_sec = ip->i_ffs1_atime; 845 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 846 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 847 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 848 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 849 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 850 vap->va_flags = ip->i_flags; 851 vap->va_gen = ip->i_gen; 852 /* this doesn't belong here */ 853 if (vp->v_type == VBLK) 854 vap->va_blocksize = BLKDEV_IOSIZE; 855 else if (vp->v_type == VCHR) 856 vap->va_blocksize = MAXBSIZE; 857 else 858 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 859 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 860 vap->va_type = vp->v_type; 861 vap->va_filerev = ip->i_modrev; 862 return (0); 863 } 864 865 /* 866 * Check to make sure the inode blocks won't choke the buffer 867 * cache, then call ufs_setattr as usual. 868 */ 869 int 870 lfs_setattr(void *v) 871 { 872 struct vop_getattr_args /* { 873 struct vnode *a_vp; 874 struct vattr *a_vap; 875 struct ucred *a_cred; 876 struct proc *a_p; 877 } */ *ap = v; 878 struct vnode *vp = ap->a_vp; 879 880 lfs_check(vp, LFS_UNUSED_LBN, 0); 881 return ufs_setattr(v); 882 } 883 884 /* 885 * Close called 886 * 887 * XXX -- we were using ufs_close, but since it updates the 888 * times on the inode, we might need to bump the uinodes 889 * count. 890 */ 891 /* ARGSUSED */ 892 int 893 lfs_close(void *v) 894 { 895 struct vop_close_args /* { 896 struct vnode *a_vp; 897 int a_fflag; 898 struct ucred *a_cred; 899 struct proc *a_p; 900 } */ *ap = v; 901 struct vnode *vp = ap->a_vp; 902 struct inode *ip = VTOI(vp); 903 struct timespec ts; 904 905 if (vp == ip->i_lfs->lfs_ivnode && 906 vp->v_mount->mnt_flag & MNT_UNMOUNT) 907 return 0; 908 909 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 910 TIMEVAL_TO_TIMESPEC(&time, &ts); 911 LFS_ITIMES(ip, &ts, &ts, &ts); 912 } 913 return (0); 914 } 915 916 /* 917 * Close wrapper for special devices. 918 * 919 * Update the times on the inode then do device close. 920 */ 921 int 922 lfsspec_close(void *v) 923 { 924 struct vop_close_args /* { 925 struct vnode *a_vp; 926 int a_fflag; 927 struct ucred *a_cred; 928 struct proc *a_p; 929 } */ *ap = v; 930 struct vnode *vp; 931 struct inode *ip; 932 struct timespec ts; 933 934 vp = ap->a_vp; 935 ip = VTOI(vp); 936 if (vp->v_usecount > 1) { 937 TIMEVAL_TO_TIMESPEC(&time, &ts); 938 LFS_ITIMES(ip, &ts, &ts, &ts); 939 } 940 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 941 } 942 943 /* 944 * Close wrapper for fifo's. 945 * 946 * Update the times on the inode then do device close. 947 */ 948 int 949 lfsfifo_close(void *v) 950 { 951 struct vop_close_args /* { 952 struct vnode *a_vp; 953 int a_fflag; 954 struct ucred *a_cred; 955 struct proc *a_p; 956 } */ *ap = v; 957 struct vnode *vp; 958 struct inode *ip; 959 struct timespec ts; 960 961 vp = ap->a_vp; 962 ip = VTOI(vp); 963 if (ap->a_vp->v_usecount > 1) { 964 TIMEVAL_TO_TIMESPEC(&time, &ts); 965 LFS_ITIMES(ip, &ts, &ts, &ts); 966 } 967 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 968 } 969 970 /* 971 * Reclaim an inode so that it can be used for other purposes. 972 */ 973 int lfs_no_inactive = 0; 974 975 int 976 lfs_reclaim(void *v) 977 { 978 struct vop_reclaim_args /* { 979 struct vnode *a_vp; 980 struct proc *a_p; 981 } */ *ap = v; 982 struct vnode *vp = ap->a_vp; 983 struct inode *ip = VTOI(vp); 984 int error; 985 986 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 987 988 LFS_CLR_UINO(ip, IN_ALLMOD); 989 if ((error = ufs_reclaim(vp, ap->a_p))) 990 return (error); 991 pool_put(&lfs_dinode_pool, VTOI(vp)->i_din.ffs1_din); 992 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 993 ip->inode_ext.lfs = NULL; 994 pool_put(&lfs_inode_pool, vp->v_data); 995 vp->v_data = NULL; 996 return (0); 997 } 998 999 /* 1000 * Read a block from a storage device. 1001 * In order to avoid reading blocks that are in the process of being 1002 * written by the cleaner---and hence are not mutexed by the normal 1003 * buffer cache / page cache mechanisms---check for collisions before 1004 * reading. 1005 * 1006 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1007 * the active cleaner test. 1008 * 1009 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1010 */ 1011 int 1012 lfs_strategy(void *v) 1013 { 1014 struct vop_strategy_args /* { 1015 struct buf *a_bp; 1016 } */ *ap = v; 1017 struct buf *bp; 1018 struct lfs *fs; 1019 struct vnode *vp; 1020 struct inode *ip; 1021 daddr_t tbn; 1022 int i, sn, error, slept; 1023 1024 bp = ap->a_bp; 1025 vp = bp->b_vp; 1026 ip = VTOI(vp); 1027 fs = ip->i_lfs; 1028 1029 /* lfs uses its strategy routine only for read */ 1030 KASSERT(bp->b_flags & B_READ); 1031 1032 if (vp->v_type == VBLK || vp->v_type == VCHR) 1033 panic("lfs_strategy: spec"); 1034 KASSERT(bp->b_bcount != 0); 1035 if (bp->b_blkno == bp->b_lblkno) { 1036 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1037 NULL); 1038 if (error) { 1039 bp->b_error = error; 1040 bp->b_flags |= B_ERROR; 1041 biodone(bp); 1042 return (error); 1043 } 1044 if ((long)bp->b_blkno == -1) /* no valid data */ 1045 clrbuf(bp); 1046 } 1047 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1048 biodone(bp); 1049 return (0); 1050 } 1051 1052 slept = 1; 1053 simple_lock(&fs->lfs_interlock); 1054 while (slept && fs->lfs_seglock) { 1055 simple_unlock(&fs->lfs_interlock); 1056 /* 1057 * Look through list of intervals. 1058 * There will only be intervals to look through 1059 * if the cleaner holds the seglock. 1060 * Since the cleaner is synchronous, we can trust 1061 * the list of intervals to be current. 1062 */ 1063 tbn = dbtofsb(fs, bp->b_blkno); 1064 sn = dtosn(fs, tbn); 1065 slept = 0; 1066 for (i = 0; i < fs->lfs_cleanind; i++) { 1067 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1068 tbn >= fs->lfs_cleanint[i]) { 1069 #ifdef DEBUG_LFS 1070 printf("lfs_strategy: ino %d lbn %" PRId64 1071 " ind %d sn %d fsb %" PRIx32 1072 " given sn %d fsb %" PRIx64 "\n", 1073 ip->i_number, bp->b_lblkno, i, 1074 dtosn(fs, fs->lfs_cleanint[i]), 1075 fs->lfs_cleanint[i], sn, tbn); 1076 printf("lfs_strategy: sleeping on ino %d lbn %" 1077 PRId64 "\n", ip->i_number, bp->b_lblkno); 1078 #endif 1079 tsleep(&fs->lfs_seglock, PRIBIO+1, 1080 "lfs_strategy", 0); 1081 /* Things may be different now; start over. */ 1082 slept = 1; 1083 break; 1084 } 1085 } 1086 simple_lock(&fs->lfs_interlock); 1087 } 1088 simple_unlock(&fs->lfs_interlock); 1089 1090 vp = ip->i_devvp; 1091 bp->b_dev = vp->v_rdev; 1092 VOCALL (vp->v_op, VOFFSET(vop_strategy), ap); 1093 return (0); 1094 } 1095 1096 static void 1097 lfs_flush_dirops(struct lfs *fs) 1098 { 1099 struct inode *ip, *nip; 1100 struct vnode *vp; 1101 extern int lfs_dostats; 1102 struct segment *sp; 1103 int needunlock; 1104 1105 if (fs->lfs_ronly) 1106 return; 1107 1108 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) 1109 return; 1110 1111 if (lfs_dostats) 1112 ++lfs_stats.flush_invoked; 1113 1114 /* 1115 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1116 * Technically this is a checkpoint (the on-disk state is valid) 1117 * even though we are leaving out all the file data. 1118 */ 1119 lfs_imtime(fs); 1120 lfs_seglock(fs, SEGM_CKP); 1121 sp = fs->lfs_sp; 1122 1123 /* 1124 * lfs_writevnodes, optimized to get dirops out of the way. 1125 * Only write dirops, and don't flush files' pages, only 1126 * blocks from the directories. 1127 * 1128 * We don't need to vref these files because they are 1129 * dirops and so hold an extra reference until the 1130 * segunlock clears them of that status. 1131 * 1132 * We don't need to check for IN_ADIROP because we know that 1133 * no dirops are active. 1134 * 1135 */ 1136 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1137 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1138 vp = ITOV(ip); 1139 1140 /* 1141 * All writes to directories come from dirops; all 1142 * writes to files' direct blocks go through the page 1143 * cache, which we're not touching. Reads to files 1144 * and/or directories will not be affected by writing 1145 * directory blocks inodes and file inodes. So we don't 1146 * really need to lock. If we don't lock, though, 1147 * make sure that we don't clear IN_MODIFIED 1148 * unnecessarily. 1149 */ 1150 if (vp->v_flag & VXLOCK) 1151 continue; 1152 if (vn_lock(vp, LK_EXCLUSIVE | LK_CANRECURSE | 1153 LK_NOWAIT) == 0) { 1154 needunlock = 1; 1155 } else { 1156 printf("lfs_flush_dirops: flushing locked ino %d\n", 1157 VTOI(vp)->i_number); 1158 needunlock = 0; 1159 } 1160 if (vp->v_type != VREG && 1161 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1162 lfs_writefile(fs, sp, vp); 1163 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1164 !(ip->i_flag & IN_ALLMOD)) { 1165 LFS_SET_UINO(ip, IN_MODIFIED); 1166 } 1167 } 1168 (void) lfs_writeinode(fs, sp, ip); 1169 if (needunlock) 1170 VOP_UNLOCK(vp, 0); 1171 else 1172 LFS_SET_UINO(ip, IN_MODIFIED); 1173 } 1174 /* We've written all the dirops there are */ 1175 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1176 (void) lfs_writeseg(fs, sp); 1177 lfs_segunlock(fs); 1178 } 1179 1180 /* 1181 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1182 */ 1183 int 1184 lfs_fcntl(void *v) 1185 { 1186 struct vop_fcntl_args /* { 1187 struct vnode *a_vp; 1188 u_long a_command; 1189 caddr_t a_data; 1190 int a_fflag; 1191 struct ucred *a_cred; 1192 struct proc *a_p; 1193 } */ *ap = v; 1194 struct timeval *tvp; 1195 BLOCK_INFO *blkiov; 1196 CLEANERINFO *cip; 1197 int blkcnt, error, oclean; 1198 struct lfs_fcntl_markv blkvp; 1199 fsid_t *fsidp; 1200 struct lfs *fs; 1201 struct buf *bp; 1202 daddr_t off; 1203 1204 /* Only respect LFS fcntls on fs root or Ifile */ 1205 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1206 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1207 return ufs_fcntl(v); 1208 } 1209 1210 /* Avoid locking a draining lock */ 1211 if (ap->a_vp->v_mount->mnt_flag & MNT_UNMOUNT) { 1212 return ESHUTDOWN; 1213 } 1214 1215 fs = VTOI(ap->a_vp)->i_lfs; 1216 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsid; 1217 1218 switch (ap->a_command) { 1219 case LFCNSEGWAITALL: 1220 fsidp = NULL; 1221 /* FALLSTHROUGH */ 1222 case LFCNSEGWAIT: 1223 tvp = (struct timeval *)ap->a_data; 1224 simple_lock(&fs->lfs_interlock); 1225 ++fs->lfs_sleepers; 1226 simple_unlock(&fs->lfs_interlock); 1227 VOP_UNLOCK(ap->a_vp, 0); 1228 1229 error = lfs_segwait(fsidp, tvp); 1230 1231 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1232 simple_lock(&fs->lfs_interlock); 1233 if (--fs->lfs_sleepers == 0) 1234 wakeup(&fs->lfs_sleepers); 1235 simple_unlock(&fs->lfs_interlock); 1236 return error; 1237 1238 case LFCNBMAPV: 1239 case LFCNMARKV: 1240 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1241 return (error); 1242 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1243 1244 blkcnt = blkvp.blkcnt; 1245 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1246 return (EINVAL); 1247 blkiov = malloc(blkcnt * sizeof(BLOCK_INFO), M_SEGMENT, M_WAITOK); 1248 if ((error = copyin(blkvp.blkiov, blkiov, 1249 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1250 free(blkiov, M_SEGMENT); 1251 return error; 1252 } 1253 1254 simple_lock(&fs->lfs_interlock); 1255 ++fs->lfs_sleepers; 1256 simple_unlock(&fs->lfs_interlock); 1257 VOP_UNLOCK(ap->a_vp, 0); 1258 if (ap->a_command == LFCNBMAPV) 1259 error = lfs_bmapv(ap->a_p, fsidp, blkiov, blkcnt); 1260 else /* LFCNMARKV */ 1261 error = lfs_markv(ap->a_p, fsidp, blkiov, blkcnt); 1262 if (error == 0) 1263 error = copyout(blkiov, blkvp.blkiov, 1264 blkcnt * sizeof(BLOCK_INFO)); 1265 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1266 simple_lock(&fs->lfs_interlock); 1267 if (--fs->lfs_sleepers == 0) 1268 wakeup(&fs->lfs_sleepers); 1269 simple_unlock(&fs->lfs_interlock); 1270 free(blkiov, M_SEGMENT); 1271 return error; 1272 1273 case LFCNRECLAIM: 1274 /* 1275 * Flush dirops and write Ifile, allowing empty segments 1276 * to be immediately reclaimed. 1277 */ 1278 lfs_writer_enter(fs, "pndirop"); 1279 off = fs->lfs_offset; 1280 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1281 lfs_flush_dirops(fs); 1282 LFS_CLEANERINFO(cip, fs, bp); 1283 oclean = cip->clean; 1284 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1285 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1286 lfs_segunlock(fs); 1287 lfs_writer_leave(fs); 1288 1289 #ifdef DEBUG_LFS 1290 LFS_CLEANERINFO(cip, fs, bp); 1291 oclean = cip->clean; 1292 printf("lfs_fcntl: reclaim wrote %" PRId64 " blocks, cleaned " 1293 "%" PRId32 " segments (activesb %d)\n", 1294 fs->lfs_offset - off, cip->clean - oclean, 1295 fs->lfs_activesb); 1296 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1297 #endif 1298 1299 return 0; 1300 1301 default: 1302 return ufs_fcntl(v); 1303 } 1304 return 0; 1305 } 1306 1307 int 1308 lfs_getpages(void *v) 1309 { 1310 struct vop_getpages_args /* { 1311 struct vnode *a_vp; 1312 voff_t a_offset; 1313 struct vm_page **a_m; 1314 int *a_count; 1315 int a_centeridx; 1316 vm_prot_t a_access_type; 1317 int a_advice; 1318 int a_flags; 1319 } */ *ap = v; 1320 1321 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1322 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1323 return EPERM; 1324 } 1325 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1326 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1327 } 1328 return genfs_getpages(v); 1329 } 1330 1331 /* 1332 * Make sure that for all pages in every block in the given range, 1333 * either all are dirty or all are clean. If any of the pages 1334 * we've seen so far are dirty, put the vnode on the paging chain, 1335 * and mark it IN_PAGING. 1336 * 1337 * If checkfirst != 0, don't check all the pages but return at the 1338 * first dirty page. 1339 */ 1340 static int 1341 check_dirty(struct lfs *fs, struct vnode *vp, 1342 off_t startoffset, off_t endoffset, off_t blkeof, 1343 int flags, int checkfirst) 1344 { 1345 int by_list; 1346 struct vm_page *curpg, *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1347 struct lwp *l = curlwp ? curlwp : &lwp0; 1348 off_t soff; 1349 voff_t off; 1350 int i, dirty, tdirty, nonexistent, any_dirty; 1351 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1352 1353 top: 1354 by_list = (vp->v_uobj.uo_npages <= 1355 ((endoffset - startoffset) >> PAGE_SHIFT) * 1356 UVM_PAGE_HASH_PENALTY); 1357 any_dirty = 0; 1358 1359 if (by_list) { 1360 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1361 PHOLD(l); 1362 } else { 1363 soff = startoffset; 1364 } 1365 while (by_list || soff < MIN(blkeof, endoffset)) { 1366 if (by_list) { 1367 if (pages_per_block > 1) { 1368 while (curpg && (curpg->offset & fs->lfs_bmask)) 1369 curpg = TAILQ_NEXT(curpg, listq); 1370 } 1371 if (curpg == NULL) 1372 break; 1373 soff = curpg->offset; 1374 } 1375 1376 /* 1377 * Mark all pages in extended range busy; find out if any 1378 * of them are dirty. 1379 */ 1380 nonexistent = dirty = 0; 1381 for (i = 0; i == 0 || i < pages_per_block; i++) { 1382 if (by_list && pages_per_block <= 1) { 1383 pgs[i] = pg = curpg; 1384 } else { 1385 off = soff + (i << PAGE_SHIFT); 1386 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1387 if (pg == NULL) { 1388 ++nonexistent; 1389 continue; 1390 } 1391 } 1392 KASSERT(pg != NULL); 1393 while (pg->flags & PG_BUSY) { 1394 pg->flags |= PG_WANTED; 1395 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1396 "lfsput", 0); 1397 simple_lock(&vp->v_interlock); 1398 if (by_list) { 1399 if (i > 0) 1400 uvm_page_unbusy(pgs, i); 1401 goto top; 1402 } 1403 } 1404 pg->flags |= PG_BUSY; 1405 UVM_PAGE_OWN(pg, "lfs_putpages"); 1406 1407 pmap_page_protect(pg, VM_PROT_NONE); 1408 tdirty = (pmap_clear_modify(pg) || 1409 (pg->flags & PG_CLEAN) == 0); 1410 dirty += tdirty; 1411 } 1412 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1413 if (by_list) { 1414 curpg = TAILQ_NEXT(curpg, listq); 1415 } else { 1416 soff += fs->lfs_bsize; 1417 } 1418 continue; 1419 } 1420 1421 any_dirty += dirty; 1422 KASSERT(nonexistent == 0); 1423 1424 /* 1425 * If any are dirty make all dirty; unbusy them, 1426 * but if we were asked to clean, wire them so that 1427 * the pagedaemon doesn't bother us about them while 1428 * they're on their way to disk. 1429 */ 1430 for (i = 0; i == 0 || i < pages_per_block; i++) { 1431 pg = pgs[i]; 1432 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1433 if (dirty) { 1434 pg->flags &= ~PG_CLEAN; 1435 if (flags & PGO_FREE) { 1436 /* XXXUBC need better way to update */ 1437 simple_lock(&lfs_subsys_lock); 1438 lfs_subsys_pages += MIN(1, pages_per_block); 1439 simple_unlock(&lfs_subsys_lock); 1440 /* 1441 * Wire the page so that 1442 * pdaemon doesn't see it again. 1443 */ 1444 uvm_lock_pageq(); 1445 uvm_pagewire(pg); 1446 uvm_unlock_pageq(); 1447 1448 /* Suspended write flag */ 1449 pg->flags |= PG_DELWRI; 1450 } 1451 } 1452 if (pg->flags & PG_WANTED) 1453 wakeup(pg); 1454 pg->flags &= ~(PG_WANTED|PG_BUSY); 1455 UVM_PAGE_OWN(pg, NULL); 1456 } 1457 1458 if (checkfirst && any_dirty) 1459 return any_dirty; 1460 1461 if (by_list) { 1462 curpg = TAILQ_NEXT(curpg, listq); 1463 } else { 1464 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1465 } 1466 } 1467 if (by_list) { 1468 PRELE(l); 1469 } 1470 1471 /* 1472 * If any pages were dirty, mark this inode as "pageout requested", 1473 * and put it on the paging queue. 1474 * XXXUBC locking (check locking on dchainhd too) 1475 */ 1476 #ifdef notyet 1477 if (any_dirty) { 1478 if (!(ip->i_flags & IN_PAGING)) { 1479 ip->i_flags |= IN_PAGING; 1480 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1481 } 1482 } 1483 #endif 1484 return any_dirty; 1485 } 1486 1487 /* 1488 * lfs_putpages functions like genfs_putpages except that 1489 * 1490 * (1) It needs to bounds-check the incoming requests to ensure that 1491 * they are block-aligned; if they are not, expand the range and 1492 * do the right thing in case, e.g., the requested range is clean 1493 * but the expanded range is dirty. 1494 * (2) It needs to explicitly send blocks to be written when it is done. 1495 * VOP_PUTPAGES is not ever called with the seglock held, so 1496 * we simply take the seglock and let lfs_segunlock wait for us. 1497 * XXX Actually we can be called with the seglock held, if we have 1498 * XXX to flush a vnode while lfs_markv is in operation. As of this 1499 * XXX writing we panic in this case. 1500 * 1501 * Assumptions: 1502 * 1503 * (1) The caller does not hold any pages in this vnode busy. If it does, 1504 * there is a danger that when we expand the page range and busy the 1505 * pages we will deadlock. 1506 * (2) We are called with vp->v_interlock held; we must return with it 1507 * released. 1508 * (3) We don't absolutely have to free pages right away, provided that 1509 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1510 * us a request with PGO_FREE, we take the pages out of the paging 1511 * queue and wake up the writer, which will handle freeing them for us. 1512 * 1513 * We ensure that for any filesystem block, all pages for that 1514 * block are either resident or not, even if those pages are higher 1515 * than EOF; that means that we will be getting requests to free 1516 * "unused" pages above EOF all the time, and should ignore them. 1517 */ 1518 1519 int 1520 lfs_putpages(void *v) 1521 { 1522 int error; 1523 struct vop_putpages_args /* { 1524 struct vnode *a_vp; 1525 voff_t a_offlo; 1526 voff_t a_offhi; 1527 int a_flags; 1528 } */ *ap = v; 1529 struct vnode *vp; 1530 struct inode *ip; 1531 struct lfs *fs; 1532 struct segment *sp; 1533 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1534 off_t off, max_endoffset; 1535 int pages_per_block; 1536 int s, sync, dirty, pagedaemon; 1537 struct vm_page *pg; 1538 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1539 1540 vp = ap->a_vp; 1541 ip = VTOI(vp); 1542 fs = ip->i_lfs; 1543 sync = (ap->a_flags & PGO_SYNCIO); 1544 pagedaemon = (curproc == uvm.pagedaemon_proc); 1545 1546 /* Putpages does nothing for metadata. */ 1547 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1548 simple_unlock(&vp->v_interlock); 1549 return 0; 1550 } 1551 1552 /* 1553 * If there are no pages, don't do anything. 1554 */ 1555 if (vp->v_uobj.uo_npages == 0) { 1556 s = splbio(); 1557 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1558 (vp->v_flag & VONWORKLST)) { 1559 vp->v_flag &= ~VONWORKLST; 1560 LIST_REMOVE(vp, v_synclist); 1561 } 1562 splx(s); 1563 simple_unlock(&vp->v_interlock); 1564 return 0; 1565 } 1566 1567 blkeof = blkroundup(fs, ip->i_size); 1568 1569 /* 1570 * Ignore requests to free pages past EOF but in the same block 1571 * as EOF, unless the request is synchronous. (XXX why sync?) 1572 * XXXUBC Make these pages look "active" so the pagedaemon won't 1573 * XXXUBC bother us with them again. 1574 */ 1575 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1576 origoffset = ap->a_offlo; 1577 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1578 pg = uvm_pagelookup(&vp->v_uobj, off); 1579 KASSERT(pg != NULL); 1580 while (pg->flags & PG_BUSY) { 1581 pg->flags |= PG_WANTED; 1582 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1583 "lfsput2", 0); 1584 simple_lock(&vp->v_interlock); 1585 } 1586 uvm_lock_pageq(); 1587 uvm_pageactivate(pg); 1588 uvm_unlock_pageq(); 1589 } 1590 ap->a_offlo = blkeof; 1591 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1592 simple_unlock(&vp->v_interlock); 1593 return 0; 1594 } 1595 } 1596 1597 /* 1598 * Extend page range to start and end at block boundaries. 1599 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1600 */ 1601 pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1602 origoffset = ap->a_offlo; 1603 origendoffset = ap->a_offhi; 1604 startoffset = origoffset & ~(fs->lfs_bmask); 1605 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1606 << fs->lfs_bshift; 1607 1608 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1609 endoffset = max_endoffset; 1610 origendoffset = endoffset; 1611 } else { 1612 origendoffset = round_page(ap->a_offhi); 1613 endoffset = round_page(blkroundup(fs, origendoffset)); 1614 } 1615 1616 KASSERT(startoffset > 0 || endoffset >= startoffset); 1617 if (startoffset == endoffset) { 1618 /* Nothing to do, why were we called? */ 1619 simple_unlock(&vp->v_interlock); 1620 #ifdef DEBUG 1621 printf("lfs_putpages: startoffset = endoffset = %" PRId64 "\n", 1622 startoffset); 1623 #endif 1624 return 0; 1625 } 1626 1627 ap->a_offlo = startoffset; 1628 ap->a_offhi = endoffset; 1629 1630 if (!(ap->a_flags & PGO_CLEANIT)) 1631 return genfs_putpages(v); 1632 1633 /* 1634 * If there are more than one page per block, we don't want 1635 * to get caught locking them backwards; so set PGO_BUSYFAIL 1636 * to avoid deadlocks. 1637 */ 1638 ap->a_flags |= PGO_BUSYFAIL; 1639 1640 do { 1641 int r; 1642 1643 /* If no pages are dirty, we can just use genfs_putpages. */ 1644 if ((dirty = check_dirty(fs, vp, startoffset, endoffset, blkeof, 1645 ap->a_flags, 1)) != 0) 1646 break; 1647 1648 if ((r = genfs_putpages(v)) != EDEADLK) 1649 return r; 1650 1651 /* Start over. */ 1652 preempt(NULL); 1653 simple_lock(&vp->v_interlock); 1654 } while(1); 1655 1656 /* 1657 * Dirty and asked to clean. 1658 * 1659 * Pagedaemon can't actually write LFS pages; wake up 1660 * the writer to take care of that. The writer will 1661 * notice the pager inode queue and act on that. 1662 */ 1663 if (pagedaemon) { 1664 ++fs->lfs_pdflush; 1665 wakeup(&lfs_writer_daemon); 1666 simple_unlock(&vp->v_interlock); 1667 return EWOULDBLOCK; 1668 } 1669 1670 /* 1671 * If this is a file created in a recent dirop, we can't flush its 1672 * inode until the dirop is complete. Drain dirops, then flush the 1673 * filesystem (taking care of any other pending dirops while we're 1674 * at it). 1675 */ 1676 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1677 (vp->v_flag & VDIROP)) { 1678 int locked; 1679 1680 /* printf("putpages to clean VDIROP, flushing\n"); */ 1681 lfs_writer_enter(fs, "ppdirop"); 1682 locked = VOP_ISLOCKED(vp) && /* XXX */ 1683 vp->v_lock.lk_lockholder == curproc->p_pid; 1684 if (locked) 1685 VOP_UNLOCK(vp, 0); 1686 simple_unlock(&vp->v_interlock); 1687 1688 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1689 1690 simple_lock(&vp->v_interlock); 1691 if (locked) 1692 VOP_LOCK(vp, LK_EXCLUSIVE); 1693 lfs_writer_leave(fs); 1694 1695 /* XXX the flush should have taken care of this one too! */ 1696 } 1697 1698 /* 1699 * This is it. We are going to write some pages. From here on 1700 * down it's all just mechanics. 1701 * 1702 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1703 */ 1704 ap->a_flags &= ~PGO_SYNCIO; 1705 1706 /* 1707 * If we've already got the seglock, flush the node and return. 1708 * The FIP has already been set up for us by lfs_writefile, 1709 * and FIP cleanup and lfs_updatemeta will also be done there, 1710 * unless genfs_putpages returns EDEADLK; then we must flush 1711 * what we have, and correct FIP and segment header accounting. 1712 */ 1713 if (ap->a_flags & PGO_LOCKED) { 1714 sp = fs->lfs_sp; 1715 sp->vp = vp; 1716 1717 /* 1718 * Make sure that all pages in any given block are dirty, or 1719 * none of them are. 1720 */ 1721 again: 1722 check_dirty(fs, vp, startoffset, endoffset, blkeof, 1723 ap->a_flags, 0); 1724 1725 if ((error = genfs_putpages(v)) == EDEADLK) { 1726 #ifdef DEBUG_LFS 1727 printf("lfs_putpages: genfs_putpages returned EDEADLK" 1728 " ino %d off %x (seg %d)\n", 1729 ip->i_number, fs->lfs_offset, 1730 dtosn(fs, fs->lfs_offset)); 1731 #endif 1732 /* If nothing to write, short-circuit */ 1733 if (sp->cbpp - sp->bpp == 1) { 1734 preempt(NULL); 1735 simple_lock(&vp->v_interlock); 1736 goto again; 1737 } 1738 /* Write gathered pages */ 1739 lfs_updatemeta(sp); 1740 (void) lfs_writeseg(fs, sp); 1741 1742 /* Reinitialize brand new FIP and add us to it */ 1743 sp->vp = vp; 1744 sp->fip->fi_version = ip->i_gen; 1745 sp->fip->fi_ino = ip->i_number; 1746 /* Add us to the new segment summary. */ 1747 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1748 sp->sum_bytes_left -= 1749 sizeof(struct finfo) - sizeof(int32_t); 1750 1751 /* Give the write a chance to complete */ 1752 preempt(NULL); 1753 1754 /* We've lost the interlock. Start over. */ 1755 simple_lock(&vp->v_interlock); 1756 goto again; 1757 } 1758 lfs_updatemeta(sp); 1759 return error; 1760 } 1761 1762 simple_unlock(&vp->v_interlock); 1763 /* 1764 * Take the seglock, because we are going to be writing pages. 1765 */ 1766 if ((error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0))) != 0) 1767 return error; 1768 1769 /* 1770 * VOP_PUTPAGES should not be called while holding the seglock. 1771 * XXXUBC fix lfs_markv, or do this properly. 1772 */ 1773 /* KASSERT(fs->lfs_seglock == 1); */ 1774 1775 /* 1776 * We assume we're being called with sp->fip pointing at blank space. 1777 * Account for a new FIP in the segment header, and set sp->vp. 1778 * (This should duplicate the setup at the top of lfs_writefile().) 1779 */ 1780 sp = fs->lfs_sp; 1781 if (sp->seg_bytes_left < fs->lfs_bsize || 1782 sp->sum_bytes_left < sizeof(struct finfo)) 1783 (void) lfs_writeseg(fs, fs->lfs_sp); 1784 1785 sp->sum_bytes_left -= sizeof(struct finfo) - sizeof(int32_t); 1786 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1787 sp->vp = vp; 1788 1789 if (vp->v_flag & VDIROP) 1790 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 1791 1792 sp->fip->fi_nblocks = 0; 1793 sp->fip->fi_ino = ip->i_number; 1794 sp->fip->fi_version = ip->i_gen; 1795 1796 /* 1797 * Loop through genfs_putpages until all pages are gathered. 1798 * genfs_putpages() drops the interlock, so reacquire it if necessary. 1799 * Whenever we lose the interlock we have to rerun check_dirty, as 1800 * well. 1801 */ 1802 again2: 1803 simple_lock(&vp->v_interlock); 1804 check_dirty(fs, vp, startoffset, endoffset, blkeof, ap->a_flags, 0); 1805 1806 if ((error = genfs_putpages(v)) == EDEADLK) { 1807 #ifdef DEBUG_LFS 1808 printf("lfs_putpages: genfs_putpages returned EDEADLK [2]" 1809 " ino %d off %x (seg %d)\n", 1810 ip->i_number, fs->lfs_offset, 1811 dtosn(fs, fs->lfs_offset)); 1812 #endif 1813 /* If nothing to write, short-circuit */ 1814 if (sp->cbpp - sp->bpp == 1) { 1815 preempt(NULL); 1816 goto again2; 1817 } 1818 /* Write gathered pages */ 1819 lfs_updatemeta(sp); 1820 (void) lfs_writeseg(fs, sp); 1821 1822 /* 1823 * Reinitialize brand new FIP and add us to it. 1824 * (This should duplicate the fixup in lfs_gatherpages().) 1825 */ 1826 sp->vp = vp; 1827 sp->fip->fi_version = ip->i_gen; 1828 sp->fip->fi_ino = ip->i_number; 1829 /* Add us to the new segment summary. */ 1830 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1831 sp->sum_bytes_left -= 1832 sizeof(struct finfo) - sizeof(int32_t); 1833 1834 /* Give the write a chance to complete */ 1835 preempt(NULL); 1836 1837 /* We've lost the interlock. Start over. */ 1838 goto again2; 1839 } 1840 1841 /* Write indirect blocks as well */ 1842 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 1843 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 1844 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 1845 1846 /* 1847 * Blocks are now gathered into a segment waiting to be written. 1848 * All that's left to do is update metadata, and write them. 1849 */ 1850 lfs_updatemeta(fs->lfs_sp); 1851 fs->lfs_sp->vp = NULL; 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 + sizeof(struct finfo) + 1858 sizeof(int32_t) * (sp->fip->fi_nblocks - 1)); 1859 sp->start_lbp = &sp->fip->fi_blocks[0]; 1860 } else { 1861 sp->sum_bytes_left += sizeof(FINFO) - sizeof(int32_t); 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_WRITE) && 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 proc *a_p; 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 Sat Oct 18 21:09:54 GMT 2025