lfs_vnops.c revision 1.150
1 /* $NetBSD: lfs_vnops.c,v 1.150 2005/04/27 20:35:10 perseant 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 1.150 2005/04/27 20:35:10 perseant 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/pool.h> 84 #include <sys/signalvar.h> 85 86 #include <miscfs/fifofs/fifo.h> 87 #include <miscfs/genfs/genfs.h> 88 #include <miscfs/specfs/specdev.h> 89 90 #include <ufs/ufs/inode.h> 91 #include <ufs/ufs/dir.h> 92 #include <ufs/ufs/ufsmount.h> 93 #include <ufs/ufs/ufs_extern.h> 94 95 #include <uvm/uvm.h> 96 #include <uvm/uvm_pmap.h> 97 #include <uvm/uvm_stat.h> 98 #include <uvm/uvm_pager.h> 99 100 #include <ufs/lfs/lfs.h> 101 #include <ufs/lfs/lfs_extern.h> 102 103 extern pid_t lfs_writer_daemon; 104 105 /* Global vfs data structures for lfs. */ 106 int (**lfs_vnodeop_p)(void *); 107 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 108 { &vop_default_desc, vn_default_error }, 109 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 110 { &vop_create_desc, lfs_create }, /* create */ 111 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 112 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 113 { &vop_open_desc, ufs_open }, /* open */ 114 { &vop_close_desc, lfs_close }, /* close */ 115 { &vop_access_desc, ufs_access }, /* access */ 116 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 117 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 118 { &vop_read_desc, lfs_read }, /* read */ 119 { &vop_write_desc, lfs_write }, /* write */ 120 { &vop_lease_desc, ufs_lease_check }, /* lease */ 121 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 122 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 123 { &vop_poll_desc, ufs_poll }, /* poll */ 124 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 125 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 126 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 127 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 128 { &vop_seek_desc, ufs_seek }, /* seek */ 129 { &vop_remove_desc, lfs_remove }, /* remove */ 130 { &vop_link_desc, lfs_link }, /* link */ 131 { &vop_rename_desc, lfs_rename }, /* rename */ 132 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 133 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 134 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 135 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 136 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 137 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 138 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 139 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 140 { &vop_lock_desc, ufs_lock }, /* lock */ 141 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 142 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 143 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 144 { &vop_print_desc, ufs_print }, /* print */ 145 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 146 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 147 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 148 { &vop_blkatoff_desc, lfs_blkatoff }, /* blkatoff */ 149 { &vop_valloc_desc, lfs_valloc }, /* valloc */ 150 { &vop_balloc_desc, lfs_balloc }, /* balloc */ 151 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 152 { &vop_truncate_desc, lfs_truncate }, /* truncate */ 153 { &vop_update_desc, lfs_update }, /* update */ 154 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 155 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 156 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 157 { NULL, NULL } 158 }; 159 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 160 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 161 162 int (**lfs_specop_p)(void *); 163 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 164 { &vop_default_desc, vn_default_error }, 165 { &vop_lookup_desc, spec_lookup }, /* lookup */ 166 { &vop_create_desc, spec_create }, /* create */ 167 { &vop_mknod_desc, spec_mknod }, /* mknod */ 168 { &vop_open_desc, spec_open }, /* open */ 169 { &vop_close_desc, lfsspec_close }, /* close */ 170 { &vop_access_desc, ufs_access }, /* access */ 171 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 172 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 173 { &vop_read_desc, ufsspec_read }, /* read */ 174 { &vop_write_desc, ufsspec_write }, /* write */ 175 { &vop_lease_desc, spec_lease_check }, /* lease */ 176 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 177 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 178 { &vop_poll_desc, spec_poll }, /* poll */ 179 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 180 { &vop_revoke_desc, spec_revoke }, /* revoke */ 181 { &vop_mmap_desc, spec_mmap }, /* mmap */ 182 { &vop_fsync_desc, spec_fsync }, /* fsync */ 183 { &vop_seek_desc, spec_seek }, /* seek */ 184 { &vop_remove_desc, spec_remove }, /* remove */ 185 { &vop_link_desc, spec_link }, /* link */ 186 { &vop_rename_desc, spec_rename }, /* rename */ 187 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 188 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 189 { &vop_symlink_desc, spec_symlink }, /* symlink */ 190 { &vop_readdir_desc, spec_readdir }, /* readdir */ 191 { &vop_readlink_desc, spec_readlink }, /* readlink */ 192 { &vop_abortop_desc, spec_abortop }, /* abortop */ 193 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 194 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 195 { &vop_lock_desc, ufs_lock }, /* lock */ 196 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 197 { &vop_bmap_desc, spec_bmap }, /* bmap */ 198 { &vop_strategy_desc, spec_strategy }, /* strategy */ 199 { &vop_print_desc, ufs_print }, /* print */ 200 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 201 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 202 { &vop_advlock_desc, spec_advlock }, /* advlock */ 203 { &vop_blkatoff_desc, spec_blkatoff }, /* blkatoff */ 204 { &vop_valloc_desc, spec_valloc }, /* valloc */ 205 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 206 { &vop_truncate_desc, spec_truncate }, /* truncate */ 207 { &vop_update_desc, lfs_update }, /* update */ 208 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 209 { &vop_getpages_desc, spec_getpages }, /* getpages */ 210 { &vop_putpages_desc, spec_putpages }, /* putpages */ 211 { NULL, NULL } 212 }; 213 const struct vnodeopv_desc lfs_specop_opv_desc = 214 { &lfs_specop_p, lfs_specop_entries }; 215 216 int (**lfs_fifoop_p)(void *); 217 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 218 { &vop_default_desc, vn_default_error }, 219 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 220 { &vop_create_desc, fifo_create }, /* create */ 221 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 222 { &vop_open_desc, fifo_open }, /* open */ 223 { &vop_close_desc, lfsfifo_close }, /* close */ 224 { &vop_access_desc, ufs_access }, /* access */ 225 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 226 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 227 { &vop_read_desc, ufsfifo_read }, /* read */ 228 { &vop_write_desc, ufsfifo_write }, /* write */ 229 { &vop_lease_desc, fifo_lease_check }, /* lease */ 230 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 231 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 232 { &vop_poll_desc, fifo_poll }, /* poll */ 233 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 234 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 235 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 236 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 237 { &vop_seek_desc, fifo_seek }, /* seek */ 238 { &vop_remove_desc, fifo_remove }, /* remove */ 239 { &vop_link_desc, fifo_link }, /* link */ 240 { &vop_rename_desc, fifo_rename }, /* rename */ 241 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 242 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 243 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 244 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 245 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 246 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 247 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 248 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 249 { &vop_lock_desc, ufs_lock }, /* lock */ 250 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 251 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 252 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 253 { &vop_print_desc, ufs_print }, /* print */ 254 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 255 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 256 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 257 { &vop_blkatoff_desc, fifo_blkatoff }, /* blkatoff */ 258 { &vop_valloc_desc, fifo_valloc }, /* valloc */ 259 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 260 { &vop_truncate_desc, fifo_truncate }, /* truncate */ 261 { &vop_update_desc, lfs_update }, /* update */ 262 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 263 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 264 { NULL, NULL } 265 }; 266 const struct vnodeopv_desc lfs_fifoop_opv_desc = 267 { &lfs_fifoop_p, lfs_fifoop_entries }; 268 269 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int); 270 271 /* 272 * A function version of LFS_ITIMES, for the UFS functions which call ITIMES 273 */ 274 void 275 lfs_itimes(struct inode *ip, struct timespec *acc, struct timespec *mod, struct timespec *cre) 276 { 277 LFS_ITIMES(ip, acc, mod, cre); 278 } 279 280 #define LFS_READWRITE 281 #include <ufs/ufs/ufs_readwrite.c> 282 #undef LFS_READWRITE 283 284 /* 285 * Synch an open file. 286 */ 287 /* ARGSUSED */ 288 int 289 lfs_fsync(void *v) 290 { 291 struct vop_fsync_args /* { 292 struct vnode *a_vp; 293 struct ucred *a_cred; 294 int a_flags; 295 off_t offlo; 296 off_t offhi; 297 struct proc *a_p; 298 } */ *ap = v; 299 struct vnode *vp = ap->a_vp; 300 int error, wait; 301 302 /* 303 * Trickle sync checks for need to do a checkpoint after possible 304 * activity from the pagedaemon. 305 */ 306 if (ap->a_flags & FSYNC_LAZY) { 307 simple_lock(&lfs_subsys_lock); 308 wakeup(&lfs_writer_daemon); 309 simple_unlock(&lfs_subsys_lock); 310 return 0; 311 } 312 313 wait = (ap->a_flags & FSYNC_WAIT); 314 simple_lock(&vp->v_interlock); 315 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 316 round_page(ap->a_offhi), 317 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 318 if (error) 319 return error; 320 error = VOP_UPDATE(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 321 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 322 int l = 0; 323 error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE, 324 ap->a_p->p_ucred, ap->a_p); 325 } 326 if (wait && !VPISEMPTY(vp)) 327 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 328 329 return error; 330 } 331 332 /* 333 * Take IN_ADIROP off, then call ufs_inactive. 334 */ 335 int 336 lfs_inactive(void *v) 337 { 338 struct vop_inactive_args /* { 339 struct vnode *a_vp; 340 struct proc *a_p; 341 } */ *ap = v; 342 343 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 344 345 lfs_unmark_vnode(ap->a_vp); 346 347 /* 348 * The Ifile is only ever inactivated on unmount. 349 * Streamline this process by not giving it more dirty blocks. 350 */ 351 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 352 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 353 VOP_UNLOCK(ap->a_vp, 0); 354 return 0; 355 } 356 357 return ufs_inactive(v); 358 } 359 360 /* 361 * These macros are used to bracket UFS directory ops, so that we can 362 * identify all the pages touched during directory ops which need to 363 * be ordered and flushed atomically, so that they may be recovered. 364 * 365 * Because we have to mark nodes VDIROP in order to prevent 366 * the cache from reclaiming them while a dirop is in progress, we must 367 * also manage the number of nodes so marked (otherwise we can run out). 368 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 369 * is decremented during segment write, when VDIROP is taken off. 370 */ 371 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 372 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 373 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 374 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 375 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 376 static int lfs_set_dirop(struct vnode *, struct vnode *); 377 378 static int 379 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 380 { 381 struct lfs *fs; 382 int error; 383 384 KASSERT(VOP_ISLOCKED(dvp)); 385 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 386 387 fs = VTOI(dvp)->i_lfs; 388 389 ASSERT_NO_SEGLOCK(fs); 390 /* 391 * LFS_NRESERVE calculates direct and indirect blocks as well 392 * as an inode block; an overestimate in most cases. 393 */ 394 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 395 return (error); 396 397 restart: 398 simple_lock(&fs->lfs_interlock); 399 if (fs->lfs_dirops == 0) { 400 simple_unlock(&fs->lfs_interlock); 401 lfs_check(dvp, LFS_UNUSED_LBN, 0); 402 simple_lock(&fs->lfs_interlock); 403 } 404 while (fs->lfs_writer) 405 ltsleep(&fs->lfs_dirops, (PRIBIO + 1), "lfs_sdirop", 0, 406 &fs->lfs_interlock); 407 simple_lock(&lfs_subsys_lock); 408 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 409 wakeup(&lfs_writer_daemon); 410 simple_unlock(&lfs_subsys_lock); 411 simple_unlock(&fs->lfs_interlock); 412 preempt(1); 413 goto restart; 414 } 415 416 if (lfs_dirvcount > LFS_MAX_DIROP) { 417 simple_unlock(&fs->lfs_interlock); 418 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 419 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 420 if ((error = ltsleep(&lfs_dirvcount, 421 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 422 &lfs_subsys_lock)) != 0) { 423 goto unreserve; 424 } 425 goto restart; 426 } 427 simple_unlock(&lfs_subsys_lock); 428 429 ++fs->lfs_dirops; 430 fs->lfs_doifile = 1; 431 simple_unlock(&fs->lfs_interlock); 432 433 /* Hold a reference so SET_ENDOP will be happy */ 434 vref(dvp); 435 if (vp) { 436 vref(vp); 437 MARK_VNODE(vp); 438 } 439 440 MARK_VNODE(dvp); 441 return 0; 442 443 unreserve: 444 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 445 return error; 446 } 447 448 /* 449 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 450 * in getnewvnode(), if we have a stacked filesystem mounted on top 451 * of us. 452 * 453 * NB: this means we have to clear the new vnodes on error. Fortunately 454 * SET_ENDOP is there to do that for us. 455 */ 456 static int 457 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 458 { 459 int error; 460 struct lfs *fs; 461 462 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 463 ASSERT_NO_SEGLOCK(fs); 464 if (fs->lfs_ronly) 465 return EROFS; 466 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 467 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 468 dvp, error)); 469 return error; 470 } 471 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 472 if (vpp) { 473 ungetnewvnode(*vpp); 474 *vpp = NULL; 475 } 476 return error; 477 } 478 return 0; 479 } 480 481 #define SET_ENDOP_BASE(fs, dvp, str) \ 482 do { \ 483 simple_lock(&(fs)->lfs_interlock); \ 484 --(fs)->lfs_dirops; \ 485 if (!(fs)->lfs_dirops) { \ 486 if ((fs)->lfs_nadirop) { \ 487 panic("SET_ENDOP: %s: no dirops but " \ 488 " nadirop=%d", (str), \ 489 (fs)->lfs_nadirop); \ 490 } \ 491 wakeup(&(fs)->lfs_writer); \ 492 simple_unlock(&(fs)->lfs_interlock); \ 493 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 494 } else \ 495 simple_unlock(&(fs)->lfs_interlock); \ 496 } while(0) 497 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 498 do { \ 499 UNMARK_VNODE(dvp); \ 500 if (nvpp && *nvpp) \ 501 UNMARK_VNODE(*nvpp); \ 502 /* Check for error return to stem vnode leakage */ \ 503 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 504 ungetnewvnode(*(nvpp)); \ 505 SET_ENDOP_BASE((fs), (dvp), (str)); \ 506 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 507 vrele(dvp); \ 508 } while(0) 509 #define SET_ENDOP_CREATE_AP(ap, str) \ 510 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 511 (ap)->a_vpp, (str)) 512 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 513 do { \ 514 UNMARK_VNODE(dvp); \ 515 if (ovp) \ 516 UNMARK_VNODE(ovp); \ 517 SET_ENDOP_BASE((fs), (dvp), (str)); \ 518 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 519 vrele(dvp); \ 520 if (ovp) \ 521 vrele(ovp); \ 522 } while(0) 523 524 void 525 lfs_mark_vnode(struct vnode *vp) 526 { 527 struct inode *ip = VTOI(vp); 528 struct lfs *fs = ip->i_lfs; 529 530 simple_lock(&fs->lfs_interlock); 531 if (!(ip->i_flag & IN_ADIROP)) { 532 if (!(vp->v_flag & VDIROP)) { 533 (void)lfs_vref(vp); 534 simple_lock(&lfs_subsys_lock); 535 ++lfs_dirvcount; 536 simple_unlock(&lfs_subsys_lock); 537 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 538 vp->v_flag |= VDIROP; 539 } 540 ++fs->lfs_nadirop; 541 ip->i_flag |= IN_ADIROP; 542 } else 543 KASSERT(vp->v_flag & VDIROP); 544 simple_unlock(&fs->lfs_interlock); 545 } 546 547 void 548 lfs_unmark_vnode(struct vnode *vp) 549 { 550 struct inode *ip = VTOI(vp); 551 552 if (ip && (ip->i_flag & IN_ADIROP)) { 553 KASSERT(vp->v_flag & VDIROP); 554 simple_lock(&ip->i_lfs->lfs_interlock); 555 --ip->i_lfs->lfs_nadirop; 556 simple_unlock(&ip->i_lfs->lfs_interlock); 557 ip->i_flag &= ~IN_ADIROP; 558 } 559 } 560 561 int 562 lfs_symlink(void *v) 563 { 564 struct vop_symlink_args /* { 565 struct vnode *a_dvp; 566 struct vnode **a_vpp; 567 struct componentname *a_cnp; 568 struct vattr *a_vap; 569 char *a_target; 570 } */ *ap = v; 571 int error; 572 573 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 574 vput(ap->a_dvp); 575 return error; 576 } 577 error = ufs_symlink(ap); 578 SET_ENDOP_CREATE_AP(ap, "symlink"); 579 return (error); 580 } 581 582 int 583 lfs_mknod(void *v) 584 { 585 struct vop_mknod_args /* { 586 struct vnode *a_dvp; 587 struct vnode **a_vpp; 588 struct componentname *a_cnp; 589 struct vattr *a_vap; 590 } */ *ap = v; 591 struct vattr *vap = ap->a_vap; 592 struct vnode **vpp = ap->a_vpp; 593 struct inode *ip; 594 int error; 595 struct mount *mp; 596 ino_t ino; 597 598 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 599 vput(ap->a_dvp); 600 return error; 601 } 602 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 603 ap->a_dvp, vpp, ap->a_cnp); 604 605 /* Either way we're done with the dirop at this point */ 606 SET_ENDOP_CREATE_AP(ap, "mknod"); 607 608 if (error) 609 return (error); 610 611 ip = VTOI(*vpp); 612 mp = (*vpp)->v_mount; 613 ino = ip->i_number; 614 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 615 if (vap->va_rdev != VNOVAL) { 616 /* 617 * Want to be able to use this to make badblock 618 * inodes, so don't truncate the dev number. 619 */ 620 #if 0 621 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 622 UFS_MPNEEDSWAP((*vpp)->v_mount)); 623 #else 624 ip->i_ffs1_rdev = vap->va_rdev; 625 #endif 626 } 627 628 /* 629 * Call fsync to write the vnode so that we don't have to deal with 630 * flushing it when it's marked VDIROP|VXLOCK. 631 * 632 * XXX KS - If we can't flush we also can't call vgone(), so must 633 * return. But, that leaves this vnode in limbo, also not good. 634 * Can this ever happen (barring hardware failure)? 635 */ 636 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 637 curproc)) != 0) { 638 panic("lfs_mknod: couldn't fsync (ino %d)", ino); 639 /* return (error); */ 640 } 641 /* 642 * Remove vnode so that it will be reloaded by VFS_VGET and 643 * checked to see if it is an alias of an existing entry in 644 * the inode cache. 645 */ 646 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 647 648 VOP_UNLOCK(*vpp, 0); 649 lfs_vunref(*vpp); 650 (*vpp)->v_type = VNON; 651 vgone(*vpp); 652 error = VFS_VGET(mp, ino, vpp); 653 654 if (error != 0) { 655 *vpp = NULL; 656 return (error); 657 } 658 return (0); 659 } 660 661 int 662 lfs_create(void *v) 663 { 664 struct vop_create_args /* { 665 struct vnode *a_dvp; 666 struct vnode **a_vpp; 667 struct componentname *a_cnp; 668 struct vattr *a_vap; 669 } */ *ap = v; 670 int error; 671 672 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 673 vput(ap->a_dvp); 674 return error; 675 } 676 error = ufs_create(ap); 677 SET_ENDOP_CREATE_AP(ap, "create"); 678 return (error); 679 } 680 681 int 682 lfs_mkdir(void *v) 683 { 684 struct vop_mkdir_args /* { 685 struct vnode *a_dvp; 686 struct vnode **a_vpp; 687 struct componentname *a_cnp; 688 struct vattr *a_vap; 689 } */ *ap = v; 690 int error; 691 692 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 693 vput(ap->a_dvp); 694 return error; 695 } 696 error = ufs_mkdir(ap); 697 SET_ENDOP_CREATE_AP(ap, "mkdir"); 698 return (error); 699 } 700 701 int 702 lfs_remove(void *v) 703 { 704 struct vop_remove_args /* { 705 struct vnode *a_dvp; 706 struct vnode *a_vp; 707 struct componentname *a_cnp; 708 } */ *ap = v; 709 struct vnode *dvp, *vp; 710 int error; 711 712 dvp = ap->a_dvp; 713 vp = ap->a_vp; 714 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 715 if (dvp == vp) 716 vrele(vp); 717 else 718 vput(vp); 719 vput(dvp); 720 return error; 721 } 722 error = ufs_remove(ap); 723 SET_ENDOP_REMOVE(VTOI(dvp)->i_lfs, dvp, vp, "remove"); 724 return (error); 725 } 726 727 int 728 lfs_rmdir(void *v) 729 { 730 struct vop_rmdir_args /* { 731 struct vnodeop_desc *a_desc; 732 struct vnode *a_dvp; 733 struct vnode *a_vp; 734 struct componentname *a_cnp; 735 } */ *ap = v; 736 struct vnode *vp; 737 int error; 738 739 vp = ap->a_vp; 740 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 741 vrele(ap->a_dvp); 742 if (ap->a_vp != ap->a_dvp) 743 VOP_UNLOCK(ap->a_dvp, 0); 744 vput(vp); 745 return error; 746 } 747 error = ufs_rmdir(ap); 748 SET_ENDOP_REMOVE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vp, "rmdir"); 749 return (error); 750 } 751 752 int 753 lfs_link(void *v) 754 { 755 struct vop_link_args /* { 756 struct vnode *a_dvp; 757 struct vnode *a_vp; 758 struct componentname *a_cnp; 759 } */ *ap = v; 760 int error; 761 struct vnode **vpp = NULL; 762 763 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 764 vput(ap->a_dvp); 765 return error; 766 } 767 error = ufs_link(ap); 768 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 769 return (error); 770 } 771 772 int 773 lfs_rename(void *v) 774 { 775 struct vop_rename_args /* { 776 struct vnode *a_fdvp; 777 struct vnode *a_fvp; 778 struct componentname *a_fcnp; 779 struct vnode *a_tdvp; 780 struct vnode *a_tvp; 781 struct componentname *a_tcnp; 782 } */ *ap = v; 783 struct vnode *tvp, *fvp, *tdvp, *fdvp; 784 struct componentname *tcnp, *fcnp; 785 int error; 786 struct lfs *fs; 787 788 fs = VTOI(ap->a_fdvp)->i_lfs; 789 tvp = ap->a_tvp; 790 tdvp = ap->a_tdvp; 791 tcnp = ap->a_tcnp; 792 fvp = ap->a_fvp; 793 fdvp = ap->a_fdvp; 794 fcnp = ap->a_fcnp; 795 796 /* 797 * Check for cross-device rename. 798 * If it is, we don't want to set dirops, just error out. 799 * (In particular note that MARK_VNODE(tdvp) will DTWT on 800 * a cross-device rename.) 801 * 802 * Copied from ufs_rename. 803 */ 804 if ((fvp->v_mount != tdvp->v_mount) || 805 (tvp && (fvp->v_mount != tvp->v_mount))) { 806 error = EXDEV; 807 goto errout; 808 } 809 810 /* 811 * Check to make sure we're not renaming a vnode onto itself 812 * (deleting a hard link by renaming one name onto another); 813 * if we are we can't recursively call VOP_REMOVE since that 814 * would leave us with an unaccounted-for number of live dirops. 815 * 816 * Inline the relevant section of ufs_rename here, *before* 817 * calling SET_DIROP_REMOVE. 818 */ 819 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 820 (VTOI(tdvp)->i_flags & APPEND))) { 821 error = EPERM; 822 goto errout; 823 } 824 if (fvp == tvp) { 825 if (fvp->v_type == VDIR) { 826 error = EINVAL; 827 goto errout; 828 } 829 830 /* Release destination completely. */ 831 VOP_ABORTOP(tdvp, tcnp); 832 vput(tdvp); 833 vput(tvp); 834 835 /* Delete source. */ 836 vrele(fvp); 837 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 838 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 839 fcnp->cn_nameiop = DELETE; 840 if ((error = relookup(fdvp, &fvp, fcnp))){ 841 /* relookup blew away fdvp */ 842 return (error); 843 } 844 return (VOP_REMOVE(fdvp, fvp, fcnp)); 845 } 846 847 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 848 goto errout; 849 MARK_VNODE(fdvp); 850 MARK_VNODE(fvp); 851 852 error = ufs_rename(ap); 853 UNMARK_VNODE(fdvp); 854 UNMARK_VNODE(fvp); 855 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 856 return (error); 857 858 errout: 859 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 860 if (tdvp == tvp) 861 vrele(tdvp); 862 else 863 vput(tdvp); 864 if (tvp) 865 vput(tvp); 866 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 867 vrele(fdvp); 868 vrele(fvp); 869 return (error); 870 } 871 872 /* XXX hack to avoid calling ITIMES in getattr */ 873 int 874 lfs_getattr(void *v) 875 { 876 struct vop_getattr_args /* { 877 struct vnode *a_vp; 878 struct vattr *a_vap; 879 struct ucred *a_cred; 880 struct proc *a_p; 881 } */ *ap = v; 882 struct vnode *vp = ap->a_vp; 883 struct inode *ip = VTOI(vp); 884 struct vattr *vap = ap->a_vap; 885 struct lfs *fs = ip->i_lfs; 886 /* 887 * Copy from inode table 888 */ 889 vap->va_fsid = ip->i_dev; 890 vap->va_fileid = ip->i_number; 891 vap->va_mode = ip->i_mode & ~IFMT; 892 vap->va_nlink = ip->i_nlink; 893 vap->va_uid = ip->i_uid; 894 vap->va_gid = ip->i_gid; 895 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 896 vap->va_size = vp->v_size; 897 vap->va_atime.tv_sec = ip->i_ffs1_atime; 898 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 899 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 900 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 901 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 902 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 903 vap->va_flags = ip->i_flags; 904 vap->va_gen = ip->i_gen; 905 /* this doesn't belong here */ 906 if (vp->v_type == VBLK) 907 vap->va_blocksize = BLKDEV_IOSIZE; 908 else if (vp->v_type == VCHR) 909 vap->va_blocksize = MAXBSIZE; 910 else 911 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 912 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 913 vap->va_type = vp->v_type; 914 vap->va_filerev = ip->i_modrev; 915 return (0); 916 } 917 918 /* 919 * Check to make sure the inode blocks won't choke the buffer 920 * cache, then call ufs_setattr as usual. 921 */ 922 int 923 lfs_setattr(void *v) 924 { 925 struct vop_setattr_args /* { 926 struct vnode *a_vp; 927 struct vattr *a_vap; 928 struct ucred *a_cred; 929 struct proc *a_p; 930 } */ *ap = v; 931 struct vnode *vp = ap->a_vp; 932 933 lfs_check(vp, LFS_UNUSED_LBN, 0); 934 return ufs_setattr(v); 935 } 936 937 /* 938 * Close called 939 * 940 * XXX -- we were using ufs_close, but since it updates the 941 * times on the inode, we might need to bump the uinodes 942 * count. 943 */ 944 /* ARGSUSED */ 945 int 946 lfs_close(void *v) 947 { 948 struct vop_close_args /* { 949 struct vnode *a_vp; 950 int a_fflag; 951 struct ucred *a_cred; 952 struct proc *a_p; 953 } */ *ap = v; 954 struct vnode *vp = ap->a_vp; 955 struct inode *ip = VTOI(vp); 956 struct timespec ts; 957 958 if (vp == ip->i_lfs->lfs_ivnode && 959 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 960 return 0; 961 962 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 963 TIMEVAL_TO_TIMESPEC(&time, &ts); 964 LFS_ITIMES(ip, &ts, &ts, &ts); 965 } 966 return (0); 967 } 968 969 /* 970 * Close wrapper for special devices. 971 * 972 * Update the times on the inode then do device close. 973 */ 974 int 975 lfsspec_close(void *v) 976 { 977 struct vop_close_args /* { 978 struct vnode *a_vp; 979 int a_fflag; 980 struct ucred *a_cred; 981 struct proc *a_p; 982 } */ *ap = v; 983 struct vnode *vp; 984 struct inode *ip; 985 struct timespec ts; 986 987 vp = ap->a_vp; 988 ip = VTOI(vp); 989 if (vp->v_usecount > 1) { 990 TIMEVAL_TO_TIMESPEC(&time, &ts); 991 LFS_ITIMES(ip, &ts, &ts, &ts); 992 } 993 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 994 } 995 996 /* 997 * Close wrapper for fifo's. 998 * 999 * Update the times on the inode then do device close. 1000 */ 1001 int 1002 lfsfifo_close(void *v) 1003 { 1004 struct vop_close_args /* { 1005 struct vnode *a_vp; 1006 int a_fflag; 1007 struct ucred *a_cred; 1008 struct proc *a_p; 1009 } */ *ap = v; 1010 struct vnode *vp; 1011 struct inode *ip; 1012 struct timespec ts; 1013 1014 vp = ap->a_vp; 1015 ip = VTOI(vp); 1016 if (ap->a_vp->v_usecount > 1) { 1017 TIMEVAL_TO_TIMESPEC(&time, &ts); 1018 LFS_ITIMES(ip, &ts, &ts, &ts); 1019 } 1020 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1021 } 1022 1023 /* 1024 * Reclaim an inode so that it can be used for other purposes. 1025 */ 1026 1027 int 1028 lfs_reclaim(void *v) 1029 { 1030 struct vop_reclaim_args /* { 1031 struct vnode *a_vp; 1032 struct proc *a_p; 1033 } */ *ap = v; 1034 struct vnode *vp = ap->a_vp; 1035 struct inode *ip = VTOI(vp); 1036 int error; 1037 1038 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1039 1040 LFS_CLR_UINO(ip, IN_ALLMOD); 1041 if ((error = ufs_reclaim(vp, ap->a_p))) 1042 return (error); 1043 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1044 lfs_deregister_all(vp); 1045 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1046 ip->inode_ext.lfs = NULL; 1047 pool_put(&lfs_inode_pool, vp->v_data); 1048 vp->v_data = NULL; 1049 return (0); 1050 } 1051 1052 /* 1053 * Read a block from a storage device. 1054 * In order to avoid reading blocks that are in the process of being 1055 * written by the cleaner---and hence are not mutexed by the normal 1056 * buffer cache / page cache mechanisms---check for collisions before 1057 * reading. 1058 * 1059 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1060 * the active cleaner test. 1061 * 1062 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1063 */ 1064 int 1065 lfs_strategy(void *v) 1066 { 1067 struct vop_strategy_args /* { 1068 struct vnode *a_vp; 1069 struct buf *a_bp; 1070 } */ *ap = v; 1071 struct buf *bp; 1072 struct lfs *fs; 1073 struct vnode *vp; 1074 struct inode *ip; 1075 daddr_t tbn; 1076 int i, sn, error, slept; 1077 1078 bp = ap->a_bp; 1079 vp = ap->a_vp; 1080 ip = VTOI(vp); 1081 fs = ip->i_lfs; 1082 1083 /* lfs uses its strategy routine only for read */ 1084 KASSERT(bp->b_flags & B_READ); 1085 1086 if (vp->v_type == VBLK || vp->v_type == VCHR) 1087 panic("lfs_strategy: spec"); 1088 KASSERT(bp->b_bcount != 0); 1089 if (bp->b_blkno == bp->b_lblkno) { 1090 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1091 NULL); 1092 if (error) { 1093 bp->b_error = error; 1094 bp->b_flags |= B_ERROR; 1095 biodone(bp); 1096 return (error); 1097 } 1098 if ((long)bp->b_blkno == -1) /* no valid data */ 1099 clrbuf(bp); 1100 } 1101 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1102 biodone(bp); 1103 return (0); 1104 } 1105 1106 slept = 1; 1107 simple_lock(&fs->lfs_interlock); 1108 while (slept && fs->lfs_seglock) { 1109 simple_unlock(&fs->lfs_interlock); 1110 /* 1111 * Look through list of intervals. 1112 * There will only be intervals to look through 1113 * if the cleaner holds the seglock. 1114 * Since the cleaner is synchronous, we can trust 1115 * the list of intervals to be current. 1116 */ 1117 tbn = dbtofsb(fs, bp->b_blkno); 1118 sn = dtosn(fs, tbn); 1119 slept = 0; 1120 for (i = 0; i < fs->lfs_cleanind; i++) { 1121 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1122 tbn >= fs->lfs_cleanint[i]) { 1123 DLOG((DLOG_CLEAN, 1124 "lfs_strategy: ino %d lbn %" PRId64 1125 " ind %d sn %d fsb %" PRIx32 1126 " given sn %d fsb %" PRIx64 "\n", 1127 ip->i_number, bp->b_lblkno, i, 1128 dtosn(fs, fs->lfs_cleanint[i]), 1129 fs->lfs_cleanint[i], sn, tbn)); 1130 DLOG((DLOG_CLEAN, 1131 "lfs_strategy: sleeping on ino %d lbn %" 1132 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1133 simple_lock(&fs->lfs_interlock); 1134 if (fs->lfs_seglock) 1135 ltsleep(&fs->lfs_seglock, 1136 (PRIBIO + 1) | PNORELOCK, 1137 "lfs_strategy", 0, 1138 &fs->lfs_interlock); 1139 /* Things may be different now; start over. */ 1140 slept = 1; 1141 break; 1142 } 1143 } 1144 simple_lock(&fs->lfs_interlock); 1145 } 1146 simple_unlock(&fs->lfs_interlock); 1147 1148 vp = ip->i_devvp; 1149 VOP_STRATEGY(vp, bp); 1150 return (0); 1151 } 1152 1153 static void 1154 lfs_flush_dirops(struct lfs *fs) 1155 { 1156 struct inode *ip, *nip; 1157 struct vnode *vp; 1158 extern int lfs_dostats; 1159 struct segment *sp; 1160 int needunlock; 1161 1162 ASSERT_NO_SEGLOCK(fs); 1163 1164 if (fs->lfs_ronly) 1165 return; 1166 1167 simple_lock(&fs->lfs_interlock); 1168 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1169 simple_unlock(&fs->lfs_interlock); 1170 return; 1171 } else 1172 simple_unlock(&fs->lfs_interlock); 1173 1174 if (lfs_dostats) 1175 ++lfs_stats.flush_invoked; 1176 1177 /* 1178 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1179 * Technically this is a checkpoint (the on-disk state is valid) 1180 * even though we are leaving out all the file data. 1181 */ 1182 lfs_imtime(fs); 1183 lfs_seglock(fs, SEGM_CKP); 1184 sp = fs->lfs_sp; 1185 1186 /* 1187 * lfs_writevnodes, optimized to get dirops out of the way. 1188 * Only write dirops, and don't flush files' pages, only 1189 * blocks from the directories. 1190 * 1191 * We don't need to vref these files because they are 1192 * dirops and so hold an extra reference until the 1193 * segunlock clears them of that status. 1194 * 1195 * We don't need to check for IN_ADIROP because we know that 1196 * no dirops are active. 1197 * 1198 */ 1199 simple_lock(&fs->lfs_interlock); 1200 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1201 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1202 simple_unlock(&fs->lfs_interlock); 1203 vp = ITOV(ip); 1204 1205 /* 1206 * All writes to directories come from dirops; all 1207 * writes to files' direct blocks go through the page 1208 * cache, which we're not touching. Reads to files 1209 * and/or directories will not be affected by writing 1210 * directory blocks inodes and file inodes. So we don't 1211 * really need to lock. If we don't lock, though, 1212 * make sure that we don't clear IN_MODIFIED 1213 * unnecessarily. 1214 */ 1215 if (vp->v_flag & VXLOCK) 1216 continue; 1217 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) { 1218 needunlock = 1; 1219 } else { 1220 DLOG((DLOG_VNODE, "lfs_flush_dirops: flushing locked ino %d\n", 1221 VTOI(vp)->i_number)); 1222 needunlock = 0; 1223 } 1224 if (vp->v_type != VREG && 1225 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1226 lfs_writefile(fs, sp, vp); 1227 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1228 !(ip->i_flag & IN_ALLMOD)) { 1229 LFS_SET_UINO(ip, IN_MODIFIED); 1230 } 1231 } 1232 (void) lfs_writeinode(fs, sp, ip); 1233 if (needunlock) 1234 VOP_UNLOCK(vp, 0); 1235 else 1236 LFS_SET_UINO(ip, IN_MODIFIED); 1237 simple_lock(&fs->lfs_interlock); 1238 } 1239 simple_unlock(&fs->lfs_interlock); 1240 /* We've written all the dirops there are */ 1241 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1242 (void) lfs_writeseg(fs, sp); 1243 lfs_segunlock(fs); 1244 } 1245 1246 /* 1247 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1248 */ 1249 int 1250 lfs_fcntl(void *v) 1251 { 1252 struct vop_fcntl_args /* { 1253 struct vnode *a_vp; 1254 u_long a_command; 1255 caddr_t a_data; 1256 int a_fflag; 1257 struct ucred *a_cred; 1258 struct proc *a_p; 1259 } */ *ap = v; 1260 struct timeval *tvp; 1261 BLOCK_INFO *blkiov; 1262 CLEANERINFO *cip; 1263 SEGUSE *sup; 1264 int blkcnt, error, oclean; 1265 struct lfs_fcntl_markv blkvp; 1266 fsid_t *fsidp; 1267 struct lfs *fs; 1268 struct buf *bp; 1269 fhandle_t *fhp; 1270 daddr_t off; 1271 1272 /* Only respect LFS fcntls on fs root or Ifile */ 1273 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1274 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1275 return ufs_fcntl(v); 1276 } 1277 1278 /* Avoid locking a draining lock */ 1279 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1280 return ESHUTDOWN; 1281 } 1282 1283 fs = VTOI(ap->a_vp)->i_lfs; 1284 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1285 1286 switch (ap->a_command) { 1287 case LFCNSEGWAITALL: 1288 case LFCNSEGWAITALL_COMPAT: 1289 fsidp = NULL; 1290 /* FALLSTHROUGH */ 1291 case LFCNSEGWAIT: 1292 case LFCNSEGWAIT_COMPAT: 1293 tvp = (struct timeval *)ap->a_data; 1294 simple_lock(&fs->lfs_interlock); 1295 ++fs->lfs_sleepers; 1296 simple_unlock(&fs->lfs_interlock); 1297 VOP_UNLOCK(ap->a_vp, 0); 1298 1299 error = lfs_segwait(fsidp, tvp); 1300 1301 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1302 simple_lock(&fs->lfs_interlock); 1303 if (--fs->lfs_sleepers == 0) 1304 wakeup(&fs->lfs_sleepers); 1305 simple_unlock(&fs->lfs_interlock); 1306 return error; 1307 1308 case LFCNBMAPV: 1309 case LFCNMARKV: 1310 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1311 return (error); 1312 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1313 1314 blkcnt = blkvp.blkcnt; 1315 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1316 return (EINVAL); 1317 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1318 if ((error = copyin(blkvp.blkiov, blkiov, 1319 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1320 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1321 return error; 1322 } 1323 1324 simple_lock(&fs->lfs_interlock); 1325 ++fs->lfs_sleepers; 1326 simple_unlock(&fs->lfs_interlock); 1327 VOP_UNLOCK(ap->a_vp, 0); 1328 if (ap->a_command == LFCNBMAPV) 1329 error = lfs_bmapv(ap->a_p, fsidp, blkiov, blkcnt); 1330 else /* LFCNMARKV */ 1331 error = lfs_markv(ap->a_p, fsidp, blkiov, blkcnt); 1332 if (error == 0) 1333 error = copyout(blkiov, blkvp.blkiov, 1334 blkcnt * sizeof(BLOCK_INFO)); 1335 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1336 simple_lock(&fs->lfs_interlock); 1337 if (--fs->lfs_sleepers == 0) 1338 wakeup(&fs->lfs_sleepers); 1339 simple_unlock(&fs->lfs_interlock); 1340 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1341 return error; 1342 1343 case LFCNRECLAIM: 1344 /* 1345 * Flush dirops and write Ifile, allowing empty segments 1346 * to be immediately reclaimed. 1347 */ 1348 VOP_UNLOCK(ap->a_vp, 0); 1349 lfs_writer_enter(fs, "pndirop"); 1350 off = fs->lfs_offset; 1351 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1352 lfs_flush_dirops(fs); 1353 LFS_CLEANERINFO(cip, fs, bp); 1354 oclean = cip->clean; 1355 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1356 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1357 lfs_segunlock(fs); 1358 lfs_writer_leave(fs); 1359 1360 #ifdef DEBUG 1361 LFS_CLEANERINFO(cip, fs, bp); 1362 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1363 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1364 fs->lfs_offset - off, cip->clean - oclean, 1365 fs->lfs_activesb)); 1366 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1367 #endif 1368 1369 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1370 return 0; 1371 1372 case LFCNIFILEFH: 1373 /* Return the filehandle of the Ifile */ 1374 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1375 return (error); 1376 fhp = (struct fhandle *)ap->a_data; 1377 fhp->fh_fsid = *fsidp; 1378 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid)); 1379 1380 case LFCNREWIND: 1381 /* Move lfs_offset to the lowest-numbered segment */ 1382 return lfs_rewind(fs, *(int *)ap->a_data); 1383 1384 case LFCNINVAL: 1385 /* Mark a segment SEGUSE_INVAL */ 1386 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1387 if (sup->su_nbytes > 0) { 1388 brelse(bp); 1389 lfs_unset_inval_all(fs); 1390 return EBUSY; 1391 } 1392 sup->su_flags |= SEGUSE_INVAL; 1393 VOP_BWRITE(bp); 1394 return 0; 1395 1396 case LFCNRESIZE: 1397 /* Resize the filesystem */ 1398 return lfs_resize_fs(fs, *(int *)ap->a_data); 1399 1400 default: 1401 return ufs_fcntl(v); 1402 } 1403 return 0; 1404 } 1405 1406 int 1407 lfs_getpages(void *v) 1408 { 1409 struct vop_getpages_args /* { 1410 struct vnode *a_vp; 1411 voff_t a_offset; 1412 struct vm_page **a_m; 1413 int *a_count; 1414 int a_centeridx; 1415 vm_prot_t a_access_type; 1416 int a_advice; 1417 int a_flags; 1418 } */ *ap = v; 1419 1420 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1421 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1422 return EPERM; 1423 } 1424 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1425 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1426 } 1427 1428 /* 1429 * we're relying on the fact that genfs_getpages() always read in 1430 * entire filesystem blocks. 1431 */ 1432 return genfs_getpages(v); 1433 } 1434 1435 /* 1436 * Make sure that for all pages in every block in the given range, 1437 * either all are dirty or all are clean. If any of the pages 1438 * we've seen so far are dirty, put the vnode on the paging chain, 1439 * and mark it IN_PAGING. 1440 * 1441 * If checkfirst != 0, don't check all the pages but return at the 1442 * first dirty page. 1443 */ 1444 static int 1445 check_dirty(struct lfs *fs, struct vnode *vp, 1446 off_t startoffset, off_t endoffset, off_t blkeof, 1447 int flags, int checkfirst) 1448 { 1449 int by_list; 1450 struct vm_page *curpg = NULL; /* XXX: gcc */ 1451 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1452 off_t soff = 0; /* XXX: gcc */ 1453 voff_t off; 1454 int i; 1455 int nonexistent; 1456 int any_dirty; /* number of dirty pages */ 1457 int dirty; /* number of dirty pages in a block */ 1458 int tdirty; 1459 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1460 1461 ASSERT_MAYBE_SEGLOCK(fs); 1462 top: 1463 by_list = (vp->v_uobj.uo_npages <= 1464 ((endoffset - startoffset) >> PAGE_SHIFT) * 1465 UVM_PAGE_HASH_PENALTY); 1466 any_dirty = 0; 1467 1468 if (by_list) { 1469 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1470 } else { 1471 soff = startoffset; 1472 } 1473 while (by_list || soff < MIN(blkeof, endoffset)) { 1474 if (by_list) { 1475 /* 1476 * Find the first page in a block. Skip 1477 * blocks outside our area of interest or beyond 1478 * the end of file. 1479 */ 1480 if (pages_per_block > 1) { 1481 while (curpg && 1482 ((curpg->offset & fs->lfs_bmask) || 1483 curpg->offset >= vp->v_size || 1484 curpg->offset >= endoffset)) 1485 curpg = TAILQ_NEXT(curpg, listq); 1486 } 1487 if (curpg == NULL) 1488 break; 1489 soff = curpg->offset; 1490 } 1491 1492 /* 1493 * Mark all pages in extended range busy; find out if any 1494 * of them are dirty. 1495 */ 1496 nonexistent = dirty = 0; 1497 for (i = 0; i == 0 || i < pages_per_block; i++) { 1498 if (by_list && pages_per_block <= 1) { 1499 pgs[i] = pg = curpg; 1500 } else { 1501 off = soff + (i << PAGE_SHIFT); 1502 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1503 if (pg == NULL) { 1504 ++nonexistent; 1505 continue; 1506 } 1507 } 1508 KASSERT(pg != NULL); 1509 while (pg->flags & PG_BUSY) { 1510 pg->flags |= PG_WANTED; 1511 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1512 "lfsput", 0); 1513 simple_lock(&vp->v_interlock); 1514 if (by_list) { 1515 if (i > 0) 1516 uvm_page_unbusy(pgs, i); 1517 goto top; 1518 } 1519 } 1520 pg->flags |= PG_BUSY; 1521 UVM_PAGE_OWN(pg, "lfs_putpages"); 1522 1523 pmap_page_protect(pg, VM_PROT_NONE); 1524 tdirty = (pmap_clear_modify(pg) || 1525 (pg->flags & PG_CLEAN) == 0); 1526 dirty += tdirty; 1527 } 1528 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1529 if (by_list) { 1530 curpg = TAILQ_NEXT(curpg, listq); 1531 } else { 1532 soff += fs->lfs_bsize; 1533 } 1534 continue; 1535 } 1536 1537 any_dirty += dirty; 1538 KASSERT(nonexistent == 0); 1539 1540 /* 1541 * If any are dirty make all dirty; unbusy them, 1542 * but if we were asked to clean, wire them so that 1543 * the pagedaemon doesn't bother us about them while 1544 * they're on their way to disk. 1545 */ 1546 for (i = 0; i == 0 || i < pages_per_block; i++) { 1547 pg = pgs[i]; 1548 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1549 if (dirty) { 1550 pg->flags &= ~PG_CLEAN; 1551 if (flags & PGO_FREE) { 1552 /* 1553 * Wire the page so that 1554 * pdaemon doesn't see it again. 1555 */ 1556 uvm_lock_pageq(); 1557 uvm_pagewire(pg); 1558 uvm_unlock_pageq(); 1559 1560 /* Suspended write flag */ 1561 pg->flags |= PG_DELWRI; 1562 } 1563 } 1564 if (pg->flags & PG_WANTED) 1565 wakeup(pg); 1566 pg->flags &= ~(PG_WANTED|PG_BUSY); 1567 UVM_PAGE_OWN(pg, NULL); 1568 } 1569 1570 if (checkfirst && any_dirty) 1571 break; 1572 1573 if (by_list) { 1574 curpg = TAILQ_NEXT(curpg, listq); 1575 } else { 1576 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1577 } 1578 } 1579 1580 /* 1581 * If any pages were dirty, mark this inode as "pageout requested", 1582 * and put it on the paging queue. 1583 * XXXUBC locking (check locking on dchainhd too) 1584 */ 1585 #ifdef notyet 1586 if (any_dirty) { 1587 if (!(ip->i_flags & IN_PAGING)) { 1588 ip->i_flags |= IN_PAGING; 1589 simple_lock(&fs->lfs_interlock); 1590 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1591 simple_unlock(&fs->lfs_interlock); 1592 } 1593 } 1594 #endif 1595 return any_dirty; 1596 } 1597 1598 /* 1599 * lfs_putpages functions like genfs_putpages except that 1600 * 1601 * (1) It needs to bounds-check the incoming requests to ensure that 1602 * they are block-aligned; if they are not, expand the range and 1603 * do the right thing in case, e.g., the requested range is clean 1604 * but the expanded range is dirty. 1605 * (2) It needs to explicitly send blocks to be written when it is done. 1606 * VOP_PUTPAGES is not ever called with the seglock held, so 1607 * we simply take the seglock and let lfs_segunlock wait for us. 1608 * XXX Actually we can be called with the seglock held, if we have 1609 * XXX to flush a vnode while lfs_markv is in operation. As of this 1610 * XXX writing we panic in this case. 1611 * 1612 * Assumptions: 1613 * 1614 * (1) The caller does not hold any pages in this vnode busy. If it does, 1615 * there is a danger that when we expand the page range and busy the 1616 * pages we will deadlock. 1617 * (2) We are called with vp->v_interlock held; we must return with it 1618 * released. 1619 * (3) We don't absolutely have to free pages right away, provided that 1620 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1621 * us a request with PGO_FREE, we take the pages out of the paging 1622 * queue and wake up the writer, which will handle freeing them for us. 1623 * 1624 * We ensure that for any filesystem block, all pages for that 1625 * block are either resident or not, even if those pages are higher 1626 * than EOF; that means that we will be getting requests to free 1627 * "unused" pages above EOF all the time, and should ignore them. 1628 * 1629 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1630 */ 1631 1632 int 1633 lfs_putpages(void *v) 1634 { 1635 int error; 1636 struct vop_putpages_args /* { 1637 struct vnode *a_vp; 1638 voff_t a_offlo; 1639 voff_t a_offhi; 1640 int a_flags; 1641 } */ *ap = v; 1642 struct vnode *vp; 1643 struct inode *ip; 1644 struct lfs *fs; 1645 struct segment *sp; 1646 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1647 off_t off, max_endoffset; 1648 int s; 1649 boolean_t seglocked, sync, pagedaemon; 1650 struct vm_page *pg; 1651 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1652 1653 vp = ap->a_vp; 1654 ip = VTOI(vp); 1655 fs = ip->i_lfs; 1656 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1657 pagedaemon = (curproc == uvm.pagedaemon_proc); 1658 1659 /* Putpages does nothing for metadata. */ 1660 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1661 simple_unlock(&vp->v_interlock); 1662 return 0; 1663 } 1664 1665 /* 1666 * If there are no pages, don't do anything. 1667 */ 1668 if (vp->v_uobj.uo_npages == 0) { 1669 s = splbio(); 1670 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1671 (vp->v_flag & VONWORKLST)) { 1672 vp->v_flag &= ~VONWORKLST; 1673 LIST_REMOVE(vp, v_synclist); 1674 } 1675 splx(s); 1676 simple_unlock(&vp->v_interlock); 1677 return 0; 1678 } 1679 1680 blkeof = blkroundup(fs, ip->i_size); 1681 1682 /* 1683 * Ignore requests to free pages past EOF but in the same block 1684 * as EOF, unless the request is synchronous. (XXX why sync?) 1685 * XXXUBC Make these pages look "active" so the pagedaemon won't 1686 * XXXUBC bother us with them again. 1687 */ 1688 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1689 origoffset = ap->a_offlo; 1690 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1691 pg = uvm_pagelookup(&vp->v_uobj, off); 1692 KASSERT(pg != NULL); 1693 while (pg->flags & PG_BUSY) { 1694 pg->flags |= PG_WANTED; 1695 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1696 "lfsput2", 0); 1697 simple_lock(&vp->v_interlock); 1698 } 1699 uvm_lock_pageq(); 1700 uvm_pageactivate(pg); 1701 uvm_unlock_pageq(); 1702 } 1703 ap->a_offlo = blkeof; 1704 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1705 simple_unlock(&vp->v_interlock); 1706 return 0; 1707 } 1708 } 1709 1710 /* 1711 * Extend page range to start and end at block boundaries. 1712 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1713 */ 1714 origoffset = ap->a_offlo; 1715 origendoffset = ap->a_offhi; 1716 startoffset = origoffset & ~(fs->lfs_bmask); 1717 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1718 << fs->lfs_bshift; 1719 1720 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1721 endoffset = max_endoffset; 1722 origendoffset = endoffset; 1723 } else { 1724 origendoffset = round_page(ap->a_offhi); 1725 endoffset = round_page(blkroundup(fs, origendoffset)); 1726 } 1727 1728 KASSERT(startoffset > 0 || endoffset >= startoffset); 1729 if (startoffset == endoffset) { 1730 /* Nothing to do, why were we called? */ 1731 simple_unlock(&vp->v_interlock); 1732 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1733 PRId64 "\n", startoffset)); 1734 return 0; 1735 } 1736 1737 ap->a_offlo = startoffset; 1738 ap->a_offhi = endoffset; 1739 1740 if (!(ap->a_flags & PGO_CLEANIT)) 1741 return genfs_putpages(v); 1742 1743 /* 1744 * If there are more than one page per block, we don't want 1745 * to get caught locking them backwards; so set PGO_BUSYFAIL 1746 * to avoid deadlocks. 1747 */ 1748 ap->a_flags |= PGO_BUSYFAIL; 1749 1750 do { 1751 int r; 1752 1753 /* If no pages are dirty, we can just use genfs_putpages. */ 1754 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 1755 ap->a_flags, 1) != 0) 1756 break; 1757 1758 /* 1759 * Sometimes pages are dirtied between the time that 1760 * we check and the time we try to clean them. 1761 * Instruct lfs_gop_write to return EDEADLK in this case 1762 * so we can write them properly. 1763 */ 1764 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 1765 r = genfs_putpages(v); 1766 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 1767 if (r != EDEADLK) 1768 return r; 1769 1770 /* Start over. */ 1771 preempt(1); 1772 simple_lock(&vp->v_interlock); 1773 } while(1); 1774 1775 /* 1776 * Dirty and asked to clean. 1777 * 1778 * Pagedaemon can't actually write LFS pages; wake up 1779 * the writer to take care of that. The writer will 1780 * notice the pager inode queue and act on that. 1781 */ 1782 if (pagedaemon) { 1783 simple_lock(&fs->lfs_interlock); 1784 ++fs->lfs_pdflush; 1785 simple_unlock(&fs->lfs_interlock); 1786 wakeup(&lfs_writer_daemon); 1787 simple_unlock(&vp->v_interlock); 1788 return EWOULDBLOCK; 1789 } 1790 1791 /* 1792 * If this is a file created in a recent dirop, we can't flush its 1793 * inode until the dirop is complete. Drain dirops, then flush the 1794 * filesystem (taking care of any other pending dirops while we're 1795 * at it). 1796 */ 1797 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1798 (vp->v_flag & VDIROP)) { 1799 int locked; 1800 1801 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 1802 locked = VOP_ISLOCKED(vp) && /* XXX */ 1803 vp->v_lock.lk_lockholder == curproc->p_pid; 1804 simple_unlock(&vp->v_interlock); 1805 lfs_writer_enter(fs, "ppdirop"); 1806 if (locked) 1807 VOP_UNLOCK(vp, 0); 1808 1809 simple_lock(&fs->lfs_interlock); 1810 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1811 simple_unlock(&fs->lfs_interlock); 1812 1813 simple_lock(&vp->v_interlock); 1814 if (locked) 1815 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 1816 lfs_writer_leave(fs); 1817 1818 /* XXX the flush should have taken care of this one too! */ 1819 } 1820 1821 /* 1822 * This is it. We are going to write some pages. From here on 1823 * down it's all just mechanics. 1824 * 1825 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1826 */ 1827 ap->a_flags &= ~PGO_SYNCIO; 1828 1829 /* 1830 * If we've already got the seglock, flush the node and return. 1831 * The FIP has already been set up for us by lfs_writefile, 1832 * and FIP cleanup and lfs_updatemeta will also be done there, 1833 * unless genfs_putpages returns EDEADLK; then we must flush 1834 * what we have, and correct FIP and segment header accounting. 1835 */ 1836 1837 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 1838 if (!seglocked) { 1839 simple_unlock(&vp->v_interlock); 1840 /* 1841 * Take the seglock, because we are going to be writing pages. 1842 */ 1843 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 1844 if (error != 0) 1845 return error; 1846 simple_lock(&vp->v_interlock); 1847 } 1848 1849 /* 1850 * VOP_PUTPAGES should not be called while holding the seglock. 1851 * XXXUBC fix lfs_markv, or do this properly. 1852 */ 1853 #ifdef notyet 1854 KASSERT(fs->lfs_seglock == 1); 1855 #endif /* notyet */ 1856 1857 /* 1858 * We assume we're being called with sp->fip pointing at blank space. 1859 * Account for a new FIP in the segment header, and set sp->vp. 1860 * (This should duplicate the setup at the top of lfs_writefile().) 1861 */ 1862 sp = fs->lfs_sp; 1863 if (!seglocked) { 1864 if (sp->seg_bytes_left < fs->lfs_bsize || 1865 sp->sum_bytes_left < sizeof(struct finfo)) 1866 (void) lfs_writeseg(fs, fs->lfs_sp); 1867 1868 sp->sum_bytes_left -= FINFOSIZE; 1869 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1870 } 1871 KASSERT(sp->vp == NULL); 1872 sp->vp = vp; 1873 1874 if (!seglocked) { 1875 if (vp->v_flag & VDIROP) 1876 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 1877 } 1878 1879 sp->fip->fi_nblocks = 0; 1880 sp->fip->fi_ino = ip->i_number; 1881 sp->fip->fi_version = ip->i_gen; 1882 1883 /* 1884 * Loop through genfs_putpages until all pages are gathered. 1885 * genfs_putpages() drops the interlock, so reacquire it if necessary. 1886 * Whenever we lose the interlock we have to rerun check_dirty, as 1887 * well. 1888 */ 1889 again: 1890 check_dirty(fs, vp, startoffset, endoffset, blkeof, ap->a_flags, 0); 1891 1892 if ((error = genfs_putpages(v)) == EDEADLK) { 1893 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 1894 " EDEADLK [2] ino %d off %x (seg %d)\n", 1895 ip->i_number, fs->lfs_offset, 1896 dtosn(fs, fs->lfs_offset))); 1897 /* If nothing to write, short-circuit */ 1898 if (sp->cbpp - sp->bpp > 1) { 1899 /* Write gathered pages */ 1900 lfs_updatemeta(sp); 1901 (void) lfs_writeseg(fs, sp); 1902 1903 /* 1904 * Reinitialize brand new FIP and add us to it. 1905 * (This should duplicate the fixup in 1906 * lfs_gatherpages().) 1907 */ 1908 KASSERT(sp->vp == vp); 1909 sp->fip->fi_version = ip->i_gen; 1910 sp->fip->fi_ino = ip->i_number; 1911 /* Add us to the new segment summary. */ 1912 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1913 sp->sum_bytes_left -= FINFOSIZE; 1914 } 1915 1916 /* Give the write a chance to complete */ 1917 preempt(1); 1918 1919 /* We've lost the interlock. Start over. */ 1920 simple_lock(&vp->v_interlock); 1921 goto again; 1922 } 1923 1924 KASSERT(sp->vp == vp); 1925 if (!seglocked) { 1926 sp->vp = NULL; /* XXX lfs_gather below will set this */ 1927 1928 /* Write indirect blocks as well */ 1929 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 1930 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 1931 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 1932 1933 KASSERT(sp->vp == NULL); 1934 sp->vp = vp; 1935 } 1936 1937 /* 1938 * Blocks are now gathered into a segment waiting to be written. 1939 * All that's left to do is update metadata, and write them. 1940 */ 1941 lfs_updatemeta(sp); 1942 KASSERT(sp->vp == vp); 1943 sp->vp = NULL; 1944 1945 if (seglocked) { 1946 /* we're called by lfs_writefile. */ 1947 return error; 1948 } 1949 1950 /* 1951 * Clean up FIP, since we're done writing this file. 1952 * This should duplicate cleanup at the end of lfs_writefile(). 1953 */ 1954 if (sp->fip->fi_nblocks != 0) { 1955 sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE + 1956 sizeof(int32_t) * sp->fip->fi_nblocks); 1957 sp->start_lbp = &sp->fip->fi_blocks[0]; 1958 } else { 1959 sp->sum_bytes_left += FINFOSIZE; 1960 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 1961 } 1962 lfs_writeseg(fs, fs->lfs_sp); 1963 1964 /* 1965 * XXX - with the malloc/copy writeseg, the pages are freed by now 1966 * even if we don't wait (e.g. if we hold a nested lock). This 1967 * will not be true if we stop using malloc/copy. 1968 */ 1969 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 1970 lfs_segunlock(fs); 1971 1972 /* 1973 * Wait for v_numoutput to drop to zero. The seglock should 1974 * take care of this, but there is a slight possibility that 1975 * aiodoned might not have got around to our buffers yet. 1976 */ 1977 if (sync) { 1978 int s; 1979 1980 s = splbio(); 1981 simple_lock(&global_v_numoutput_slock); 1982 while (vp->v_numoutput > 0) { 1983 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 1984 " num %d\n", ip->i_number, vp->v_numoutput)); 1985 vp->v_flag |= VBWAIT; 1986 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 1987 &global_v_numoutput_slock); 1988 } 1989 simple_unlock(&global_v_numoutput_slock); 1990 splx(s); 1991 } 1992 return error; 1993 } 1994 1995 /* 1996 * Return the last logical file offset that should be written for this file 1997 * if we're doing a write that ends at "size". If writing, we need to know 1998 * about sizes on disk, i.e. fragments if there are any; if reading, we need 1999 * to know about entire blocks. 2000 */ 2001 void 2002 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2003 { 2004 struct inode *ip = VTOI(vp); 2005 struct lfs *fs = ip->i_lfs; 2006 daddr_t olbn, nlbn; 2007 2008 KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)); 2009 KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)) 2010 != (GOP_SIZE_READ | GOP_SIZE_WRITE)); 2011 2012 olbn = lblkno(fs, ip->i_size); 2013 nlbn = lblkno(fs, size); 2014 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2015 *eobp = fragroundup(fs, size); 2016 } else { 2017 *eobp = blkroundup(fs, size); 2018 } 2019 } 2020 2021 #ifdef DEBUG 2022 void lfs_dump_vop(void *); 2023 2024 void 2025 lfs_dump_vop(void *v) 2026 { 2027 struct vop_putpages_args /* { 2028 struct vnode *a_vp; 2029 voff_t a_offlo; 2030 voff_t a_offhi; 2031 int a_flags; 2032 } */ *ap = v; 2033 2034 #ifdef DDB 2035 vfs_vnode_print(ap->a_vp, 0, printf); 2036 #endif 2037 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2038 } 2039 #endif 2040 2041 int 2042 lfs_mmap(void *v) 2043 { 2044 struct vop_mmap_args /* { 2045 const struct vnodeop_desc *a_desc; 2046 struct vnode *a_vp; 2047 int a_fflags; 2048 struct ucred *a_cred; 2049 struct proc *a_p; 2050 } */ *ap = v; 2051 2052 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2053 return EOPNOTSUPP; 2054 return ufs_mmap(v); 2055 } 2056
Indexes created Tue Sep 30 20:09:53 GMT 2025