lfs_vnops.c revision 1.137.2.20
1 /* $NetBSD: lfs_vnops.c,v 1.137.2.20 2006/05/20 22:41:31 riz 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 1.137.2.20 2006/05/20 22:41:31 riz 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 /* If we're mounted read-only, don't try to sync. */ 303 if (VTOI(vp)->i_lfs->lfs_ronly) 304 return 0; 305 306 /* 307 * Trickle sync checks for need to do a checkpoint after possible 308 * activity from the pagedaemon. 309 */ 310 if (ap->a_flags & FSYNC_LAZY) { 311 simple_lock(&lfs_subsys_lock); 312 wakeup(&lfs_writer_daemon); 313 simple_unlock(&lfs_subsys_lock); 314 return 0; 315 } 316 317 wait = (ap->a_flags & FSYNC_WAIT); 318 simple_lock(&vp->v_interlock); 319 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 320 round_page(ap->a_offhi), 321 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 322 if (error) 323 return error; 324 error = VOP_UPDATE(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 325 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 326 int l = 0; 327 error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE, 328 ap->a_p->p_ucred, ap->a_p); 329 } 330 if (wait && !VPISEMPTY(vp)) 331 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 332 333 return error; 334 } 335 336 /* 337 * Take IN_ADIROP off, then call ufs_inactive. 338 */ 339 int 340 lfs_inactive(void *v) 341 { 342 struct vop_inactive_args /* { 343 struct vnode *a_vp; 344 struct proc *a_p; 345 } */ *ap = v; 346 347 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 348 349 lfs_unmark_vnode(ap->a_vp); 350 351 /* 352 * The Ifile is only ever inactivated on unmount. 353 * Streamline this process by not giving it more dirty blocks. 354 */ 355 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 356 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 357 VOP_UNLOCK(ap->a_vp, 0); 358 return 0; 359 } 360 361 return ufs_inactive(v); 362 } 363 364 /* 365 * These macros are used to bracket UFS directory ops, so that we can 366 * identify all the pages touched during directory ops which need to 367 * be ordered and flushed atomically, so that they may be recovered. 368 * 369 * Because we have to mark nodes VDIROP in order to prevent 370 * the cache from reclaiming them while a dirop is in progress, we must 371 * also manage the number of nodes so marked (otherwise we can run out). 372 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 373 * is decremented during segment write, when VDIROP is taken off. 374 */ 375 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 376 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 377 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 378 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 379 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 380 static int lfs_set_dirop(struct vnode *, struct vnode *); 381 382 static int 383 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 384 { 385 struct lfs *fs; 386 int error; 387 388 KASSERT(VOP_ISLOCKED(dvp)); 389 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 390 391 fs = VTOI(dvp)->i_lfs; 392 393 ASSERT_NO_SEGLOCK(fs); 394 /* 395 * LFS_NRESERVE calculates direct and indirect blocks as well 396 * as an inode block; an overestimate in most cases. 397 */ 398 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 399 return (error); 400 401 restart: 402 simple_lock(&fs->lfs_interlock); 403 if (fs->lfs_dirops == 0) { 404 simple_unlock(&fs->lfs_interlock); 405 lfs_check(dvp, LFS_UNUSED_LBN, 0); 406 simple_lock(&fs->lfs_interlock); 407 } 408 while (fs->lfs_writer) 409 ltsleep(&fs->lfs_dirops, (PRIBIO + 1), "lfs_sdirop", 0, 410 &fs->lfs_interlock); 411 simple_lock(&lfs_subsys_lock); 412 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 413 wakeup(&lfs_writer_daemon); 414 simple_unlock(&lfs_subsys_lock); 415 simple_unlock(&fs->lfs_interlock); 416 preempt(1); 417 goto restart; 418 } 419 420 if (lfs_dirvcount > LFS_MAX_DIROP) { 421 simple_unlock(&fs->lfs_interlock); 422 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 423 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 424 if ((error = ltsleep(&lfs_dirvcount, 425 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 426 &lfs_subsys_lock)) != 0) { 427 goto unreserve; 428 } 429 goto restart; 430 } 431 simple_unlock(&lfs_subsys_lock); 432 433 ++fs->lfs_dirops; 434 fs->lfs_doifile = 1; 435 simple_unlock(&fs->lfs_interlock); 436 437 /* Hold a reference so SET_ENDOP will be happy */ 438 vref(dvp); 439 if (vp) { 440 vref(vp); 441 MARK_VNODE(vp); 442 } 443 444 MARK_VNODE(dvp); 445 return 0; 446 447 unreserve: 448 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 449 return error; 450 } 451 452 /* 453 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 454 * in getnewvnode(), if we have a stacked filesystem mounted on top 455 * of us. 456 * 457 * NB: this means we have to clear the new vnodes on error. Fortunately 458 * SET_ENDOP is there to do that for us. 459 */ 460 static int 461 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 462 { 463 int error; 464 struct lfs *fs; 465 466 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 467 ASSERT_NO_SEGLOCK(fs); 468 if (fs->lfs_ronly) 469 return EROFS; 470 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 471 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 472 dvp, error)); 473 return error; 474 } 475 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 476 if (vpp) { 477 ungetnewvnode(*vpp); 478 *vpp = NULL; 479 } 480 return error; 481 } 482 return 0; 483 } 484 485 #define SET_ENDOP_BASE(fs, dvp, str) \ 486 do { \ 487 simple_lock(&(fs)->lfs_interlock); \ 488 --(fs)->lfs_dirops; \ 489 if (!(fs)->lfs_dirops) { \ 490 if ((fs)->lfs_nadirop) { \ 491 panic("SET_ENDOP: %s: no dirops but " \ 492 " nadirop=%d", (str), \ 493 (fs)->lfs_nadirop); \ 494 } \ 495 wakeup(&(fs)->lfs_writer); \ 496 simple_unlock(&(fs)->lfs_interlock); \ 497 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 498 } else \ 499 simple_unlock(&(fs)->lfs_interlock); \ 500 } while(0) 501 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 502 do { \ 503 UNMARK_VNODE(dvp); \ 504 if (nvpp && *nvpp) \ 505 UNMARK_VNODE(*nvpp); \ 506 /* Check for error return to stem vnode leakage */ \ 507 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 508 ungetnewvnode(*(nvpp)); \ 509 SET_ENDOP_BASE((fs), (dvp), (str)); \ 510 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 511 vrele(dvp); \ 512 } while(0) 513 #define SET_ENDOP_CREATE_AP(ap, str) \ 514 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 515 (ap)->a_vpp, (str)) 516 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 517 do { \ 518 UNMARK_VNODE(dvp); \ 519 if (ovp) \ 520 UNMARK_VNODE(ovp); \ 521 SET_ENDOP_BASE((fs), (dvp), (str)); \ 522 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 523 vrele(dvp); \ 524 if (ovp) \ 525 vrele(ovp); \ 526 } while(0) 527 528 void 529 lfs_mark_vnode(struct vnode *vp) 530 { 531 struct inode *ip = VTOI(vp); 532 struct lfs *fs = ip->i_lfs; 533 534 simple_lock(&fs->lfs_interlock); 535 if (!(ip->i_flag & IN_ADIROP)) { 536 if (!(vp->v_flag & VDIROP)) { 537 (void)lfs_vref(vp); 538 simple_lock(&lfs_subsys_lock); 539 ++lfs_dirvcount; 540 simple_unlock(&lfs_subsys_lock); 541 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 542 vp->v_flag |= VDIROP; 543 } 544 ++fs->lfs_nadirop; 545 ip->i_flag |= IN_ADIROP; 546 } else 547 KASSERT(vp->v_flag & VDIROP); 548 simple_unlock(&fs->lfs_interlock); 549 } 550 551 void 552 lfs_unmark_vnode(struct vnode *vp) 553 { 554 struct inode *ip = VTOI(vp); 555 556 if (ip && (ip->i_flag & IN_ADIROP)) { 557 KASSERT(vp->v_flag & VDIROP); 558 simple_lock(&ip->i_lfs->lfs_interlock); 559 --ip->i_lfs->lfs_nadirop; 560 simple_unlock(&ip->i_lfs->lfs_interlock); 561 ip->i_flag &= ~IN_ADIROP; 562 } 563 } 564 565 int 566 lfs_symlink(void *v) 567 { 568 struct vop_symlink_args /* { 569 struct vnode *a_dvp; 570 struct vnode **a_vpp; 571 struct componentname *a_cnp; 572 struct vattr *a_vap; 573 char *a_target; 574 } */ *ap = v; 575 int error; 576 577 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 578 vput(ap->a_dvp); 579 return error; 580 } 581 error = ufs_symlink(ap); 582 SET_ENDOP_CREATE_AP(ap, "symlink"); 583 return (error); 584 } 585 586 int 587 lfs_mknod(void *v) 588 { 589 struct vop_mknod_args /* { 590 struct vnode *a_dvp; 591 struct vnode **a_vpp; 592 struct componentname *a_cnp; 593 struct vattr *a_vap; 594 } */ *ap = v; 595 struct vattr *vap = ap->a_vap; 596 struct vnode **vpp = ap->a_vpp; 597 struct inode *ip; 598 int error; 599 struct mount *mp; 600 ino_t ino; 601 602 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 603 vput(ap->a_dvp); 604 return error; 605 } 606 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 607 ap->a_dvp, vpp, ap->a_cnp); 608 609 /* Either way we're done with the dirop at this point */ 610 SET_ENDOP_CREATE_AP(ap, "mknod"); 611 612 if (error) 613 return (error); 614 615 ip = VTOI(*vpp); 616 mp = (*vpp)->v_mount; 617 ino = ip->i_number; 618 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 619 if (vap->va_rdev != VNOVAL) { 620 /* 621 * Want to be able to use this to make badblock 622 * inodes, so don't truncate the dev number. 623 */ 624 #if 0 625 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 626 UFS_MPNEEDSWAP((*vpp)->v_mount)); 627 #else 628 ip->i_ffs1_rdev = vap->va_rdev; 629 #endif 630 } 631 632 /* 633 * Call fsync to write the vnode so that we don't have to deal with 634 * flushing it when it's marked VDIROP|VXLOCK. 635 * 636 * XXX KS - If we can't flush we also can't call vgone(), so must 637 * return. But, that leaves this vnode in limbo, also not good. 638 * Can this ever happen (barring hardware failure)? 639 */ 640 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 641 curproc)) != 0) { 642 panic("lfs_mknod: couldn't fsync (ino %llu)", 643 (unsigned long long)ino); 644 /* return (error); */ 645 } 646 /* 647 * Remove vnode so that it will be reloaded by VFS_VGET and 648 * checked to see if it is an alias of an existing entry in 649 * the inode cache. 650 */ 651 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 652 653 VOP_UNLOCK(*vpp, 0); 654 lfs_vunref(*vpp); 655 (*vpp)->v_type = VNON; 656 vgone(*vpp); 657 error = VFS_VGET(mp, ino, vpp); 658 659 if (error != 0) { 660 *vpp = NULL; 661 return (error); 662 } 663 return (0); 664 } 665 666 int 667 lfs_create(void *v) 668 { 669 struct vop_create_args /* { 670 struct vnode *a_dvp; 671 struct vnode **a_vpp; 672 struct componentname *a_cnp; 673 struct vattr *a_vap; 674 } */ *ap = v; 675 int error; 676 677 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 678 vput(ap->a_dvp); 679 return error; 680 } 681 error = ufs_create(ap); 682 SET_ENDOP_CREATE_AP(ap, "create"); 683 return (error); 684 } 685 686 int 687 lfs_mkdir(void *v) 688 { 689 struct vop_mkdir_args /* { 690 struct vnode *a_dvp; 691 struct vnode **a_vpp; 692 struct componentname *a_cnp; 693 struct vattr *a_vap; 694 } */ *ap = v; 695 int error; 696 697 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 698 vput(ap->a_dvp); 699 return error; 700 } 701 error = ufs_mkdir(ap); 702 SET_ENDOP_CREATE_AP(ap, "mkdir"); 703 return (error); 704 } 705 706 int 707 lfs_remove(void *v) 708 { 709 struct vop_remove_args /* { 710 struct vnode *a_dvp; 711 struct vnode *a_vp; 712 struct componentname *a_cnp; 713 } */ *ap = v; 714 struct vnode *dvp, *vp; 715 int error; 716 717 dvp = ap->a_dvp; 718 vp = ap->a_vp; 719 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 720 if (dvp == vp) 721 vrele(vp); 722 else 723 vput(vp); 724 vput(dvp); 725 return error; 726 } 727 error = ufs_remove(ap); 728 SET_ENDOP_REMOVE(VTOI(dvp)->i_lfs, dvp, ap->a_vp, "remove"); 729 return (error); 730 } 731 732 int 733 lfs_rmdir(void *v) 734 { 735 struct vop_rmdir_args /* { 736 struct vnodeop_desc *a_desc; 737 struct vnode *a_dvp; 738 struct vnode *a_vp; 739 struct componentname *a_cnp; 740 } */ *ap = v; 741 struct vnode *vp; 742 int error; 743 744 vp = ap->a_vp; 745 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 746 vrele(ap->a_dvp); 747 if (ap->a_vp != ap->a_dvp) 748 VOP_UNLOCK(ap->a_dvp, 0); 749 vput(vp); 750 return error; 751 } 752 error = ufs_rmdir(ap); 753 SET_ENDOP_REMOVE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 754 return (error); 755 } 756 757 int 758 lfs_link(void *v) 759 { 760 struct vop_link_args /* { 761 struct vnode *a_dvp; 762 struct vnode *a_vp; 763 struct componentname *a_cnp; 764 } */ *ap = v; 765 int error; 766 struct vnode **vpp = NULL; 767 768 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 769 vput(ap->a_dvp); 770 return error; 771 } 772 error = ufs_link(ap); 773 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 774 return (error); 775 } 776 777 int 778 lfs_rename(void *v) 779 { 780 struct vop_rename_args /* { 781 struct vnode *a_fdvp; 782 struct vnode *a_fvp; 783 struct componentname *a_fcnp; 784 struct vnode *a_tdvp; 785 struct vnode *a_tvp; 786 struct componentname *a_tcnp; 787 } */ *ap = v; 788 struct vnode *tvp, *fvp, *tdvp, *fdvp; 789 struct componentname *tcnp, *fcnp; 790 int error; 791 struct lfs *fs; 792 793 fs = VTOI(ap->a_fdvp)->i_lfs; 794 tvp = ap->a_tvp; 795 tdvp = ap->a_tdvp; 796 tcnp = ap->a_tcnp; 797 fvp = ap->a_fvp; 798 fdvp = ap->a_fdvp; 799 fcnp = ap->a_fcnp; 800 801 /* 802 * Check for cross-device rename. 803 * If it is, we don't want to set dirops, just error out. 804 * (In particular note that MARK_VNODE(tdvp) will DTWT on 805 * a cross-device rename.) 806 * 807 * Copied from ufs_rename. 808 */ 809 if ((fvp->v_mount != tdvp->v_mount) || 810 (tvp && (fvp->v_mount != tvp->v_mount))) { 811 error = EXDEV; 812 goto errout; 813 } 814 815 /* 816 * Check to make sure we're not renaming a vnode onto itself 817 * (deleting a hard link by renaming one name onto another); 818 * if we are we can't recursively call VOP_REMOVE since that 819 * would leave us with an unaccounted-for number of live dirops. 820 * 821 * Inline the relevant section of ufs_rename here, *before* 822 * calling SET_DIROP_REMOVE. 823 */ 824 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 825 (VTOI(tdvp)->i_flags & APPEND))) { 826 error = EPERM; 827 goto errout; 828 } 829 if (fvp == tvp) { 830 if (fvp->v_type == VDIR) { 831 error = EINVAL; 832 goto errout; 833 } 834 835 /* Release destination completely. */ 836 VOP_ABORTOP(tdvp, tcnp); 837 vput(tdvp); 838 vput(tvp); 839 840 /* Delete source. */ 841 vrele(fvp); 842 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 843 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 844 fcnp->cn_nameiop = DELETE; 845 if ((error = relookup(fdvp, &fvp, fcnp))){ 846 /* relookup blew away fdvp */ 847 return (error); 848 } 849 return (VOP_REMOVE(fdvp, fvp, fcnp)); 850 } 851 852 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 853 goto errout; 854 MARK_VNODE(fdvp); 855 MARK_VNODE(fvp); 856 857 error = ufs_rename(ap); 858 UNMARK_VNODE(fdvp); 859 UNMARK_VNODE(fvp); 860 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 861 return (error); 862 863 errout: 864 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 865 if (tdvp == tvp) 866 vrele(tdvp); 867 else 868 vput(tdvp); 869 if (tvp) 870 vput(tvp); 871 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 872 vrele(fdvp); 873 vrele(fvp); 874 return (error); 875 } 876 877 /* XXX hack to avoid calling ITIMES in getattr */ 878 int 879 lfs_getattr(void *v) 880 { 881 struct vop_getattr_args /* { 882 struct vnode *a_vp; 883 struct vattr *a_vap; 884 struct ucred *a_cred; 885 struct proc *a_p; 886 } */ *ap = v; 887 struct vnode *vp = ap->a_vp; 888 struct inode *ip = VTOI(vp); 889 struct vattr *vap = ap->a_vap; 890 struct lfs *fs = ip->i_lfs; 891 /* 892 * Copy from inode table 893 */ 894 vap->va_fsid = ip->i_dev; 895 vap->va_fileid = ip->i_number; 896 vap->va_mode = ip->i_mode & ~IFMT; 897 vap->va_nlink = ip->i_nlink; 898 vap->va_uid = ip->i_uid; 899 vap->va_gid = ip->i_gid; 900 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 901 vap->va_size = vp->v_size; 902 vap->va_atime.tv_sec = ip->i_ffs1_atime; 903 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 904 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 905 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 906 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 907 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 908 vap->va_flags = ip->i_flags; 909 vap->va_gen = ip->i_gen; 910 /* this doesn't belong here */ 911 if (vp->v_type == VBLK) 912 vap->va_blocksize = BLKDEV_IOSIZE; 913 else if (vp->v_type == VCHR) 914 vap->va_blocksize = MAXBSIZE; 915 else 916 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 917 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 918 vap->va_type = vp->v_type; 919 vap->va_filerev = ip->i_modrev; 920 return (0); 921 } 922 923 /* 924 * Check to make sure the inode blocks won't choke the buffer 925 * cache, then call ufs_setattr as usual. 926 */ 927 int 928 lfs_setattr(void *v) 929 { 930 struct vop_setattr_args /* { 931 struct vnode *a_vp; 932 struct vattr *a_vap; 933 struct ucred *a_cred; 934 struct proc *a_p; 935 } */ *ap = v; 936 struct vnode *vp = ap->a_vp; 937 938 lfs_check(vp, LFS_UNUSED_LBN, 0); 939 return ufs_setattr(v); 940 } 941 942 /* 943 * Close called 944 * 945 * XXX -- we were using ufs_close, but since it updates the 946 * times on the inode, we might need to bump the uinodes 947 * count. 948 */ 949 /* ARGSUSED */ 950 int 951 lfs_close(void *v) 952 { 953 struct vop_close_args /* { 954 struct vnode *a_vp; 955 int a_fflag; 956 struct ucred *a_cred; 957 struct proc *a_p; 958 } */ *ap = v; 959 struct vnode *vp = ap->a_vp; 960 struct inode *ip = VTOI(vp); 961 struct timespec ts; 962 963 if (vp == ip->i_lfs->lfs_ivnode && 964 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 965 return 0; 966 967 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 968 TIMEVAL_TO_TIMESPEC(&time, &ts); 969 LFS_ITIMES(ip, &ts, &ts, &ts); 970 } 971 return (0); 972 } 973 974 /* 975 * Close wrapper for special devices. 976 * 977 * Update the times on the inode then do device close. 978 */ 979 int 980 lfsspec_close(void *v) 981 { 982 struct vop_close_args /* { 983 struct vnode *a_vp; 984 int a_fflag; 985 struct ucred *a_cred; 986 struct proc *a_p; 987 } */ *ap = v; 988 struct vnode *vp; 989 struct inode *ip; 990 struct timespec ts; 991 992 vp = ap->a_vp; 993 ip = VTOI(vp); 994 if (vp->v_usecount > 1) { 995 TIMEVAL_TO_TIMESPEC(&time, &ts); 996 LFS_ITIMES(ip, &ts, &ts, &ts); 997 } 998 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 999 } 1000 1001 /* 1002 * Close wrapper for fifo's. 1003 * 1004 * Update the times on the inode then do device close. 1005 */ 1006 int 1007 lfsfifo_close(void *v) 1008 { 1009 struct vop_close_args /* { 1010 struct vnode *a_vp; 1011 int a_fflag; 1012 struct ucred *a_cred; 1013 struct proc *a_p; 1014 } */ *ap = v; 1015 struct vnode *vp; 1016 struct inode *ip; 1017 struct timespec ts; 1018 1019 vp = ap->a_vp; 1020 ip = VTOI(vp); 1021 if (ap->a_vp->v_usecount > 1) { 1022 TIMEVAL_TO_TIMESPEC(&time, &ts); 1023 LFS_ITIMES(ip, &ts, &ts, &ts); 1024 } 1025 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1026 } 1027 1028 /* 1029 * Reclaim an inode so that it can be used for other purposes. 1030 */ 1031 1032 int 1033 lfs_reclaim(void *v) 1034 { 1035 struct vop_reclaim_args /* { 1036 struct vnode *a_vp; 1037 struct proc *a_p; 1038 } */ *ap = v; 1039 struct vnode *vp = ap->a_vp; 1040 struct inode *ip = VTOI(vp); 1041 int error; 1042 1043 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1044 1045 LFS_CLR_UINO(ip, IN_ALLMOD); 1046 if ((error = ufs_reclaim(vp, ap->a_p))) 1047 return (error); 1048 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1049 lfs_deregister_all(vp); 1050 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1051 ip->inode_ext.lfs = NULL; 1052 pool_put(&lfs_inode_pool, vp->v_data); 1053 vp->v_data = NULL; 1054 return (0); 1055 } 1056 1057 /* 1058 * Read a block from a storage device. 1059 * In order to avoid reading blocks that are in the process of being 1060 * written by the cleaner---and hence are not mutexed by the normal 1061 * buffer cache / page cache mechanisms---check for collisions before 1062 * reading. 1063 * 1064 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1065 * the active cleaner test. 1066 * 1067 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1068 */ 1069 int 1070 lfs_strategy(void *v) 1071 { 1072 struct vop_strategy_args /* { 1073 struct vnode *a_vp; 1074 struct buf *a_bp; 1075 } */ *ap = v; 1076 struct buf *bp; 1077 struct lfs *fs; 1078 struct vnode *vp; 1079 struct inode *ip; 1080 daddr_t tbn; 1081 int i, sn, error, slept; 1082 1083 bp = ap->a_bp; 1084 vp = ap->a_vp; 1085 ip = VTOI(vp); 1086 fs = ip->i_lfs; 1087 1088 /* lfs uses its strategy routine only for read */ 1089 KASSERT(bp->b_flags & B_READ); 1090 1091 if (vp->v_type == VBLK || vp->v_type == VCHR) 1092 panic("lfs_strategy: spec"); 1093 KASSERT(bp->b_bcount != 0); 1094 if (bp->b_blkno == bp->b_lblkno) { 1095 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1096 NULL); 1097 if (error) { 1098 bp->b_error = error; 1099 bp->b_flags |= B_ERROR; 1100 biodone(bp); 1101 return (error); 1102 } 1103 if ((long)bp->b_blkno == -1) /* no valid data */ 1104 clrbuf(bp); 1105 } 1106 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1107 biodone(bp); 1108 return (0); 1109 } 1110 1111 slept = 1; 1112 simple_lock(&fs->lfs_interlock); 1113 while (slept && fs->lfs_seglock) { 1114 simple_unlock(&fs->lfs_interlock); 1115 /* 1116 * Look through list of intervals. 1117 * There will only be intervals to look through 1118 * if the cleaner holds the seglock. 1119 * Since the cleaner is synchronous, we can trust 1120 * the list of intervals to be current. 1121 */ 1122 tbn = dbtofsb(fs, bp->b_blkno); 1123 sn = dtosn(fs, tbn); 1124 slept = 0; 1125 for (i = 0; i < fs->lfs_cleanind; i++) { 1126 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1127 tbn >= fs->lfs_cleanint[i]) { 1128 DLOG((DLOG_CLEAN, 1129 "lfs_strategy: ino %d lbn %" PRId64 1130 " ind %d sn %d fsb %" PRIx32 1131 " given sn %d fsb %" PRIx64 "\n", 1132 ip->i_number, bp->b_lblkno, i, 1133 dtosn(fs, fs->lfs_cleanint[i]), 1134 fs->lfs_cleanint[i], sn, tbn)); 1135 DLOG((DLOG_CLEAN, 1136 "lfs_strategy: sleeping on ino %d lbn %" 1137 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1138 simple_lock(&fs->lfs_interlock); 1139 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1140 /* Cleaner can't wait for itself */ 1141 ltsleep(&fs->lfs_iocount, 1142 (PRIBIO + 1) | PNORELOCK, 1143 "clean2", 0, 1144 &fs->lfs_interlock); 1145 slept = 1; 1146 break; 1147 } else if (fs->lfs_seglock) { 1148 ltsleep(&fs->lfs_seglock, 1149 (PRIBIO + 1) | PNORELOCK, 1150 "clean1", 0, 1151 &fs->lfs_interlock); 1152 slept = 1; 1153 break; 1154 } 1155 simple_unlock(&fs->lfs_interlock); 1156 } 1157 } 1158 simple_lock(&fs->lfs_interlock); 1159 } 1160 simple_unlock(&fs->lfs_interlock); 1161 1162 vp = ip->i_devvp; 1163 VOP_STRATEGY(vp, bp); 1164 return (0); 1165 } 1166 1167 void 1168 lfs_flush_dirops(struct lfs *fs) 1169 { 1170 struct inode *ip, *nip; 1171 struct vnode *vp; 1172 extern int lfs_dostats; 1173 struct segment *sp; 1174 int needunlock; 1175 1176 ASSERT_MAYBE_SEGLOCK(fs); 1177 KASSERT(fs->lfs_nadirop == 0); 1178 1179 if (fs->lfs_ronly) 1180 return; 1181 1182 simple_lock(&fs->lfs_interlock); 1183 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1184 simple_unlock(&fs->lfs_interlock); 1185 return; 1186 } else 1187 simple_unlock(&fs->lfs_interlock); 1188 1189 if (lfs_dostats) 1190 ++lfs_stats.flush_invoked; 1191 1192 /* 1193 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1194 * Technically this is a checkpoint (the on-disk state is valid) 1195 * even though we are leaving out all the file data. 1196 */ 1197 lfs_imtime(fs); 1198 lfs_seglock(fs, SEGM_CKP); 1199 sp = fs->lfs_sp; 1200 1201 /* 1202 * lfs_writevnodes, optimized to get dirops out of the way. 1203 * Only write dirops, and don't flush files' pages, only 1204 * blocks from the directories. 1205 * 1206 * We don't need to vref these files because they are 1207 * dirops and so hold an extra reference until the 1208 * segunlock clears them of that status. 1209 * 1210 * We don't need to check for IN_ADIROP because we know that 1211 * no dirops are active. 1212 * 1213 */ 1214 simple_lock(&fs->lfs_interlock); 1215 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1216 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1217 simple_unlock(&fs->lfs_interlock); 1218 vp = ITOV(ip); 1219 1220 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1221 1222 /* 1223 * All writes to directories come from dirops; all 1224 * writes to files' direct blocks go through the page 1225 * cache, which we're not touching. Reads to files 1226 * and/or directories will not be affected by writing 1227 * directory blocks inodes and file inodes. So we don't 1228 * really need to lock. If we don't lock, though, 1229 * make sure that we don't clear IN_MODIFIED 1230 * unnecessarily. 1231 */ 1232 if (vp->v_flag & VXLOCK) { 1233 simple_lock(&fs->lfs_interlock); 1234 continue; 1235 } 1236 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) { 1237 needunlock = 1; 1238 } else { 1239 DLOG((DLOG_VNODE, "lfs_flush_dirops: flushing locked ino %d\n", 1240 VTOI(vp)->i_number)); 1241 needunlock = 0; 1242 } 1243 if (vp->v_type != VREG && 1244 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1245 lfs_writefile(fs, sp, vp); 1246 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1247 !(ip->i_flag & IN_ALLMOD)) { 1248 LFS_SET_UINO(ip, IN_MODIFIED); 1249 } 1250 } 1251 (void) lfs_writeinode(fs, sp, ip); 1252 if (needunlock) 1253 VOP_UNLOCK(vp, 0); 1254 else 1255 LFS_SET_UINO(ip, IN_MODIFIED); 1256 simple_lock(&fs->lfs_interlock); 1257 } 1258 simple_unlock(&fs->lfs_interlock); 1259 /* We've written all the dirops there are */ 1260 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1261 lfs_finalize_fs_seguse(fs); 1262 (void) lfs_writeseg(fs, sp); 1263 lfs_segunlock(fs); 1264 } 1265 1266 /* 1267 * Flush all vnodes for which the pagedaemon has requested pageouts. 1268 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1269 * has just run, this would be an error). If we have to skip a vnode 1270 * for any reason, just skip it; if we have to wait for the cleaner, 1271 * abort. The writer daemon will call us again later. 1272 */ 1273 void 1274 lfs_flush_pchain(struct lfs *fs) 1275 { 1276 struct inode *ip, *nip; 1277 struct vnode *vp; 1278 extern int lfs_dostats; 1279 struct segment *sp; 1280 int error; 1281 1282 ASSERT_NO_SEGLOCK(fs); 1283 1284 if (fs->lfs_ronly) 1285 return; 1286 1287 simple_lock(&fs->lfs_interlock); 1288 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1289 simple_unlock(&fs->lfs_interlock); 1290 return; 1291 } else 1292 simple_unlock(&fs->lfs_interlock); 1293 1294 /* Get dirops out of the way */ 1295 lfs_flush_dirops(fs); 1296 1297 if (lfs_dostats) 1298 ++lfs_stats.flush_invoked; 1299 1300 /* 1301 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1302 */ 1303 lfs_imtime(fs); 1304 lfs_seglock(fs, 0); 1305 sp = fs->lfs_sp; 1306 1307 /* 1308 * lfs_writevnodes, optimized to clear pageout requests. 1309 * Only write non-dirop files that are in the pageout queue. 1310 * We're very conservative about what we write; we want to be 1311 * fast and async. 1312 */ 1313 simple_lock(&fs->lfs_interlock); 1314 top: 1315 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1316 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1317 vp = ITOV(ip); 1318 1319 if (!(ip->i_flags & IN_PAGING)) 1320 goto top; 1321 1322 if (vp->v_flag & (VXLOCK|VDIROP)) 1323 continue; 1324 if (vp->v_type != VREG) 1325 continue; 1326 if (lfs_vref(vp)) 1327 continue; 1328 simple_unlock(&fs->lfs_interlock); 1329 1330 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 1331 lfs_vunref(vp); 1332 simple_lock(&fs->lfs_interlock); 1333 continue; 1334 } 1335 1336 error = lfs_writefile(fs, sp, vp); 1337 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1338 !(ip->i_flag & IN_ALLMOD)) { 1339 LFS_SET_UINO(ip, IN_MODIFIED); 1340 } 1341 (void) lfs_writeinode(fs, sp, ip); 1342 1343 VOP_UNLOCK(vp, 0); 1344 lfs_vunref(vp); 1345 1346 if (error == EAGAIN) { 1347 lfs_writeseg(fs, sp); 1348 simple_lock(&fs->lfs_interlock); 1349 break; 1350 } 1351 simple_lock(&fs->lfs_interlock); 1352 } 1353 simple_unlock(&fs->lfs_interlock); 1354 (void) lfs_writeseg(fs, sp); 1355 lfs_segunlock(fs); 1356 } 1357 1358 /* 1359 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1360 */ 1361 int 1362 lfs_fcntl(void *v) 1363 { 1364 struct vop_fcntl_args /* { 1365 struct vnode *a_vp; 1366 u_long a_command; 1367 caddr_t a_data; 1368 int a_fflag; 1369 struct ucred *a_cred; 1370 struct proc *a_p; 1371 } */ *ap = v; 1372 struct timeval *tvp; 1373 BLOCK_INFO *blkiov; 1374 CLEANERINFO *cip; 1375 SEGUSE *sup; 1376 int blkcnt, error, oclean; 1377 struct lfs_fcntl_markv blkvp; 1378 fsid_t *fsidp; 1379 struct lfs *fs; 1380 struct buf *bp; 1381 fhandle_t *fhp; 1382 daddr_t off; 1383 1384 /* Only respect LFS fcntls on fs root or Ifile */ 1385 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1386 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1387 return ufs_fcntl(v); 1388 } 1389 1390 /* Avoid locking a draining lock */ 1391 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1392 return ESHUTDOWN; 1393 } 1394 1395 fs = VTOI(ap->a_vp)->i_lfs; 1396 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1397 1398 switch (ap->a_command) { 1399 case LFCNSEGWAITALL: 1400 case LFCNSEGWAITALL_COMPAT: 1401 fsidp = NULL; 1402 /* FALLSTHROUGH */ 1403 case LFCNSEGWAIT: 1404 case LFCNSEGWAIT_COMPAT: 1405 tvp = (struct timeval *)ap->a_data; 1406 simple_lock(&fs->lfs_interlock); 1407 ++fs->lfs_sleepers; 1408 simple_unlock(&fs->lfs_interlock); 1409 VOP_UNLOCK(ap->a_vp, 0); 1410 1411 error = lfs_segwait(fsidp, tvp); 1412 1413 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1414 simple_lock(&fs->lfs_interlock); 1415 if (--fs->lfs_sleepers == 0) 1416 wakeup(&fs->lfs_sleepers); 1417 simple_unlock(&fs->lfs_interlock); 1418 return error; 1419 1420 case LFCNBMAPV: 1421 case LFCNMARKV: 1422 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1423 return (error); 1424 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1425 1426 blkcnt = blkvp.blkcnt; 1427 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1428 return (EINVAL); 1429 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1430 if ((error = copyin(blkvp.blkiov, blkiov, 1431 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1432 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1433 return error; 1434 } 1435 1436 simple_lock(&fs->lfs_interlock); 1437 ++fs->lfs_sleepers; 1438 simple_unlock(&fs->lfs_interlock); 1439 VOP_UNLOCK(ap->a_vp, 0); 1440 if (ap->a_command == LFCNBMAPV) 1441 error = lfs_bmapv(ap->a_p, fsidp, blkiov, blkcnt); 1442 else /* LFCNMARKV */ 1443 error = lfs_markv(ap->a_p, fsidp, blkiov, blkcnt); 1444 if (error == 0) 1445 error = copyout(blkiov, blkvp.blkiov, 1446 blkcnt * sizeof(BLOCK_INFO)); 1447 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1448 simple_lock(&fs->lfs_interlock); 1449 if (--fs->lfs_sleepers == 0) 1450 wakeup(&fs->lfs_sleepers); 1451 simple_unlock(&fs->lfs_interlock); 1452 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1453 return error; 1454 1455 case LFCNRECLAIM: 1456 /* 1457 * Flush dirops and write Ifile, allowing empty segments 1458 * to be immediately reclaimed. 1459 */ 1460 VOP_UNLOCK(ap->a_vp, 0); 1461 lfs_writer_enter(fs, "pndirop"); 1462 off = fs->lfs_offset; 1463 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1464 lfs_flush_dirops(fs); 1465 LFS_CLEANERINFO(cip, fs, bp); 1466 oclean = cip->clean; 1467 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1468 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1469 fs->lfs_sp->seg_flags |= SEGM_PROT; 1470 lfs_segunlock(fs); 1471 lfs_writer_leave(fs); 1472 1473 #ifdef DEBUG 1474 LFS_CLEANERINFO(cip, fs, bp); 1475 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1476 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1477 fs->lfs_offset - off, cip->clean - oclean, 1478 fs->lfs_activesb)); 1479 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1480 #endif 1481 1482 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1483 return 0; 1484 1485 case LFCNIFILEFH: 1486 /* Return the filehandle of the Ifile */ 1487 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1488 return (error); 1489 fhp = (struct fhandle *)ap->a_data; 1490 fhp->fh_fsid = *fsidp; 1491 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid)); 1492 1493 case LFCNREWIND: 1494 /* Move lfs_offset to the lowest-numbered segment */ 1495 return lfs_rewind(fs, *(int *)ap->a_data); 1496 1497 case LFCNINVAL: 1498 /* Mark a segment SEGUSE_INVAL */ 1499 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1500 if (sup->su_nbytes > 0) { 1501 brelse(bp); 1502 lfs_unset_inval_all(fs); 1503 return EBUSY; 1504 } 1505 sup->su_flags |= SEGUSE_INVAL; 1506 VOP_BWRITE(bp); 1507 return 0; 1508 1509 case LFCNRESIZE: 1510 /* Resize the filesystem */ 1511 return lfs_resize_fs(fs, *(int *)ap->a_data); 1512 1513 case LFCNWRAPSTOP: 1514 /* 1515 * Hold lfs_newseg at segment 0; sleep until the filesystem 1516 * wraps around. For debugging purposes, so an external 1517 * agent can log every segment in the filesystem as it 1518 * was written, and we can regression-test checkpoint 1519 * validity in the general case. 1520 */ 1521 VOP_UNLOCK(ap->a_vp, 0); 1522 simple_lock(&fs->lfs_interlock); 1523 fs->lfs_nowrap = 1; 1524 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER | PNORELOCK, 1525 "segwrap", 0, &fs->lfs_interlock); 1526 if (error) { 1527 fs->lfs_nowrap = 0; 1528 wakeup(&fs->lfs_nowrap); 1529 } 1530 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1531 return 0; 1532 1533 case LFCNWRAPGO: 1534 /* 1535 * Having done its work, the agent wakes up the writer. 1536 * It sleeps until a new segment is selected. 1537 */ 1538 VOP_UNLOCK(ap->a_vp, 0); 1539 simple_lock(&fs->lfs_interlock); 1540 fs->lfs_nowrap = 0; 1541 wakeup(&fs->lfs_nowrap); 1542 ltsleep(&fs->lfs_nextseg, PCATCH | PUSER | PNORELOCK, 1543 "segment", 0, &fs->lfs_interlock); 1544 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1545 return 0; 1546 1547 default: 1548 return ufs_fcntl(v); 1549 } 1550 return 0; 1551 } 1552 1553 int 1554 lfs_getpages(void *v) 1555 { 1556 struct vop_getpages_args /* { 1557 struct vnode *a_vp; 1558 voff_t a_offset; 1559 struct vm_page **a_m; 1560 int *a_count; 1561 int a_centeridx; 1562 vm_prot_t a_access_type; 1563 int a_advice; 1564 int a_flags; 1565 } */ *ap = v; 1566 1567 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1568 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1569 return EPERM; 1570 } 1571 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1572 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1573 } 1574 1575 /* 1576 * we're relying on the fact that genfs_getpages() always read in 1577 * entire filesystem blocks. 1578 */ 1579 return genfs_getpages(v); 1580 } 1581 1582 /* 1583 * Make sure that for all pages in every block in the given range, 1584 * either all are dirty or all are clean. If any of the pages 1585 * we've seen so far are dirty, put the vnode on the paging chain, 1586 * and mark it IN_PAGING. 1587 * 1588 * If checkfirst != 0, don't check all the pages but return at the 1589 * first dirty page. 1590 */ 1591 static int 1592 check_dirty(struct lfs *fs, struct vnode *vp, 1593 off_t startoffset, off_t endoffset, off_t blkeof, 1594 int flags, int checkfirst) 1595 { 1596 int by_list; 1597 struct vm_page *curpg = NULL; /* XXX: gcc */ 1598 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1599 off_t soff = 0; /* XXX: gcc */ 1600 voff_t off; 1601 int i; 1602 int nonexistent; 1603 int any_dirty; /* number of dirty pages */ 1604 int dirty; /* number of dirty pages in a block */ 1605 int tdirty; 1606 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1607 int pagedaemon = (curproc == uvm.pagedaemon_proc); 1608 1609 ASSERT_MAYBE_SEGLOCK(fs); 1610 top: 1611 by_list = (vp->v_uobj.uo_npages <= 1612 ((endoffset - startoffset) >> PAGE_SHIFT) * 1613 UVM_PAGE_HASH_PENALTY); 1614 any_dirty = 0; 1615 1616 if (by_list) { 1617 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1618 } else { 1619 soff = startoffset; 1620 } 1621 while (by_list || soff < MIN(blkeof, endoffset)) { 1622 if (by_list) { 1623 /* 1624 * Find the first page in a block. Skip 1625 * blocks outside our area of interest or beyond 1626 * the end of file. 1627 */ 1628 if (pages_per_block > 1) { 1629 while (curpg && 1630 ((curpg->offset & fs->lfs_bmask) || 1631 curpg->offset >= vp->v_size || 1632 curpg->offset >= endoffset)) 1633 curpg = TAILQ_NEXT(curpg, listq); 1634 } 1635 if (curpg == NULL) 1636 break; 1637 soff = curpg->offset; 1638 } 1639 1640 /* 1641 * Mark all pages in extended range busy; find out if any 1642 * of them are dirty. 1643 */ 1644 nonexistent = dirty = 0; 1645 for (i = 0; i == 0 || i < pages_per_block; i++) { 1646 if (by_list && pages_per_block <= 1) { 1647 pgs[i] = pg = curpg; 1648 } else { 1649 off = soff + (i << PAGE_SHIFT); 1650 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1651 if (pg == NULL) { 1652 ++nonexistent; 1653 continue; 1654 } 1655 } 1656 KASSERT(pg != NULL); 1657 1658 /* 1659 * If we're holding the segment lock, we can deadlocked 1660 * against a process that has our page and is waiting 1661 * for the cleaner, while the cleaner waits for the 1662 * segment lock. Just bail in that case. 1663 */ 1664 if ((pg->flags & PG_BUSY) && 1665 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1666 if (by_list && i > 0) 1667 uvm_page_unbusy(pgs, i); 1668 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1669 return -1; 1670 } 1671 1672 while (pg->flags & PG_BUSY) { 1673 pg->flags |= PG_WANTED; 1674 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1675 "lfsput", 0); 1676 simple_lock(&vp->v_interlock); 1677 if (by_list) { 1678 if (i > 0) 1679 uvm_page_unbusy(pgs, i); 1680 goto top; 1681 } 1682 } 1683 pg->flags |= PG_BUSY; 1684 UVM_PAGE_OWN(pg, "lfs_putpages"); 1685 1686 pmap_page_protect(pg, VM_PROT_NONE); 1687 tdirty = (pmap_clear_modify(pg) || 1688 (pg->flags & PG_CLEAN) == 0); 1689 dirty += tdirty; 1690 } 1691 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1692 if (by_list) { 1693 curpg = TAILQ_NEXT(curpg, listq); 1694 } else { 1695 soff += fs->lfs_bsize; 1696 } 1697 continue; 1698 } 1699 1700 any_dirty += dirty; 1701 KASSERT(nonexistent == 0); 1702 1703 /* 1704 * If any are dirty make all dirty; unbusy them, 1705 * but if we were asked to clean, wire them so that 1706 * the pagedaemon doesn't bother us about them while 1707 * they're on their way to disk. 1708 */ 1709 for (i = 0; i == 0 || i < pages_per_block; i++) { 1710 pg = pgs[i]; 1711 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1712 if (dirty) { 1713 pg->flags &= ~PG_CLEAN; 1714 if (flags & PGO_FREE) { 1715 /* 1716 * Wire the page so that 1717 * pdaemon doesn't see it again. 1718 */ 1719 uvm_lock_pageq(); 1720 uvm_pagewire(pg); 1721 uvm_unlock_pageq(); 1722 1723 /* Suspended write flag */ 1724 pg->flags |= PG_DELWRI; 1725 } 1726 } 1727 if (pg->flags & PG_WANTED) 1728 wakeup(pg); 1729 pg->flags &= ~(PG_WANTED|PG_BUSY); 1730 UVM_PAGE_OWN(pg, NULL); 1731 } 1732 1733 if (checkfirst && any_dirty) 1734 break; 1735 1736 if (by_list) { 1737 curpg = TAILQ_NEXT(curpg, listq); 1738 } else { 1739 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1740 } 1741 } 1742 1743 return any_dirty; 1744 } 1745 1746 /* 1747 * lfs_putpages functions like genfs_putpages except that 1748 * 1749 * (1) It needs to bounds-check the incoming requests to ensure that 1750 * they are block-aligned; if they are not, expand the range and 1751 * do the right thing in case, e.g., the requested range is clean 1752 * but the expanded range is dirty. 1753 * (2) It needs to explicitly send blocks to be written when it is done. 1754 * VOP_PUTPAGES is not ever called with the seglock held, so 1755 * we simply take the seglock and let lfs_segunlock wait for us. 1756 * XXX Actually we can be called with the seglock held, if we have 1757 * XXX to flush a vnode while lfs_markv is in operation. As of this 1758 * XXX writing we panic in this case. 1759 * 1760 * Assumptions: 1761 * 1762 * (1) The caller does not hold any pages in this vnode busy. If it does, 1763 * there is a danger that when we expand the page range and busy the 1764 * pages we will deadlock. 1765 * (2) We are called with vp->v_interlock held; we must return with it 1766 * released. 1767 * (3) We don't absolutely have to free pages right away, provided that 1768 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1769 * us a request with PGO_FREE, we take the pages out of the paging 1770 * queue and wake up the writer, which will handle freeing them for us. 1771 * 1772 * We ensure that for any filesystem block, all pages for that 1773 * block are either resident or not, even if those pages are higher 1774 * than EOF; that means that we will be getting requests to free 1775 * "unused" pages above EOF all the time, and should ignore them. 1776 * 1777 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1778 */ 1779 1780 int 1781 lfs_putpages(void *v) 1782 { 1783 int error; 1784 struct vop_putpages_args /* { 1785 struct vnode *a_vp; 1786 voff_t a_offlo; 1787 voff_t a_offhi; 1788 int a_flags; 1789 } */ *ap = v; 1790 struct vnode *vp; 1791 struct inode *ip; 1792 struct lfs *fs; 1793 struct segment *sp; 1794 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1795 off_t off, max_endoffset; 1796 int s; 1797 boolean_t seglocked, sync, pagedaemon; 1798 struct vm_page *pg; 1799 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1800 1801 vp = ap->a_vp; 1802 ip = VTOI(vp); 1803 fs = ip->i_lfs; 1804 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1805 pagedaemon = (curproc == uvm.pagedaemon_proc); 1806 1807 /* Putpages does nothing for metadata. */ 1808 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1809 simple_unlock(&vp->v_interlock); 1810 return 0; 1811 } 1812 1813 /* 1814 * If there are no pages, don't do anything. 1815 */ 1816 if (vp->v_uobj.uo_npages == 0) { 1817 s = splbio(); 1818 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1819 (vp->v_flag & VONWORKLST)) { 1820 vp->v_flag &= ~VONWORKLST; 1821 LIST_REMOVE(vp, v_synclist); 1822 } 1823 splx(s); 1824 simple_unlock(&vp->v_interlock); 1825 1826 /* Remove us from paging queue, if we were on it */ 1827 simple_lock(&fs->lfs_interlock); 1828 if (ip->i_flags & IN_PAGING) { 1829 ip->i_flags &= ~IN_PAGING; 1830 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1831 } 1832 simple_unlock(&fs->lfs_interlock); 1833 return 0; 1834 } 1835 1836 blkeof = blkroundup(fs, ip->i_size); 1837 1838 /* 1839 * Ignore requests to free pages past EOF but in the same block 1840 * as EOF, unless the request is synchronous. (If the request is 1841 * sync, it comes from lfs_truncate.) 1842 * XXXUBC Make these pages look "active" so the pagedaemon won't 1843 * XXXUBC bother us with them again. 1844 */ 1845 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1846 origoffset = ap->a_offlo; 1847 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1848 pg = uvm_pagelookup(&vp->v_uobj, off); 1849 KASSERT(pg != NULL); 1850 while (pg->flags & PG_BUSY) { 1851 pg->flags |= PG_WANTED; 1852 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1853 "lfsput2", 0); 1854 simple_lock(&vp->v_interlock); 1855 } 1856 uvm_lock_pageq(); 1857 uvm_pageactivate(pg); 1858 uvm_unlock_pageq(); 1859 } 1860 ap->a_offlo = blkeof; 1861 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1862 simple_unlock(&vp->v_interlock); 1863 return 0; 1864 } 1865 } 1866 1867 /* 1868 * Extend page range to start and end at block boundaries. 1869 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1870 */ 1871 origoffset = ap->a_offlo; 1872 origendoffset = ap->a_offhi; 1873 startoffset = origoffset & ~(fs->lfs_bmask); 1874 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1875 << fs->lfs_bshift; 1876 1877 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1878 endoffset = max_endoffset; 1879 origendoffset = endoffset; 1880 } else { 1881 origendoffset = round_page(ap->a_offhi); 1882 endoffset = round_page(blkroundup(fs, origendoffset)); 1883 } 1884 1885 KASSERT(startoffset > 0 || endoffset >= startoffset); 1886 if (startoffset == endoffset) { 1887 /* Nothing to do, why were we called? */ 1888 simple_unlock(&vp->v_interlock); 1889 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1890 PRId64 "\n", startoffset)); 1891 return 0; 1892 } 1893 1894 ap->a_offlo = startoffset; 1895 ap->a_offhi = endoffset; 1896 1897 if (!(ap->a_flags & PGO_CLEANIT)) 1898 return genfs_putpages(v); 1899 1900 /* 1901 * If there are more than one page per block, we don't want 1902 * to get caught locking them backwards; so set PGO_BUSYFAIL 1903 * to avoid deadlocks. 1904 */ 1905 ap->a_flags |= PGO_BUSYFAIL; 1906 1907 do { 1908 int r; 1909 1910 /* If no pages are dirty, we can just use genfs_putpages. */ 1911 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 1912 ap->a_flags, 1); 1913 if (r < 0) { 1914 simple_unlock(&vp->v_interlock); 1915 return EDEADLK; 1916 } 1917 if (r > 0) 1918 break; 1919 1920 /* 1921 * Sometimes pages are dirtied between the time that 1922 * we check and the time we try to clean them. 1923 * Instruct lfs_gop_write to return EDEADLK in this case 1924 * so we can write them properly. 1925 */ 1926 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 1927 r = genfs_putpages(v); 1928 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 1929 if (r != EDEADLK) 1930 return r; 1931 1932 /* Start over. */ 1933 preempt(1); 1934 simple_lock(&vp->v_interlock); 1935 } while(1); 1936 1937 /* 1938 * Dirty and asked to clean. 1939 * 1940 * Pagedaemon can't actually write LFS pages; wake up 1941 * the writer to take care of that. The writer will 1942 * notice the pager inode queue and act on that. 1943 */ 1944 if (pagedaemon) { 1945 simple_lock(&fs->lfs_interlock); 1946 if (!(ip->i_flags & IN_PAGING)) { 1947 ip->i_flags |= IN_PAGING; 1948 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1949 } 1950 simple_lock(&lfs_subsys_lock); 1951 wakeup(&lfs_writer_daemon); 1952 simple_unlock(&lfs_subsys_lock); 1953 simple_unlock(&fs->lfs_interlock); 1954 simple_unlock(&vp->v_interlock); 1955 preempt(1); 1956 return EWOULDBLOCK; 1957 } 1958 1959 /* 1960 * If this is a file created in a recent dirop, we can't flush its 1961 * inode until the dirop is complete. Drain dirops, then flush the 1962 * filesystem (taking care of any other pending dirops while we're 1963 * at it). 1964 */ 1965 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1966 (vp->v_flag & VDIROP)) { 1967 int locked; 1968 1969 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 1970 locked = VOP_ISLOCKED(vp) && /* XXX */ 1971 vp->v_lock.lk_lockholder == curproc->p_pid; 1972 simple_unlock(&vp->v_interlock); 1973 lfs_writer_enter(fs, "ppdirop"); 1974 if (locked) 1975 VOP_UNLOCK(vp, 0); 1976 1977 simple_lock(&fs->lfs_interlock); 1978 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1979 simple_unlock(&fs->lfs_interlock); 1980 1981 simple_lock(&vp->v_interlock); 1982 if (locked) { 1983 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 1984 simple_lock(&vp->v_interlock); 1985 } 1986 lfs_writer_leave(fs); 1987 1988 /* XXX the flush should have taken care of this one too! */ 1989 } 1990 1991 /* 1992 * This is it. We are going to write some pages. From here on 1993 * down it's all just mechanics. 1994 * 1995 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1996 */ 1997 ap->a_flags &= ~PGO_SYNCIO; 1998 1999 /* 2000 * If we've already got the seglock, flush the node and return. 2001 * The FIP has already been set up for us by lfs_writefile, 2002 * and FIP cleanup and lfs_updatemeta will also be done there, 2003 * unless genfs_putpages returns EDEADLK; then we must flush 2004 * what we have, and correct FIP and segment header accounting. 2005 */ 2006 2007 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2008 if (!seglocked) { 2009 simple_unlock(&vp->v_interlock); 2010 /* 2011 * Take the seglock, because we are going to be writing pages. 2012 */ 2013 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2014 if (error != 0) 2015 return error; 2016 simple_lock(&vp->v_interlock); 2017 } 2018 2019 /* 2020 * VOP_PUTPAGES should not be called while holding the seglock. 2021 * XXXUBC fix lfs_markv, or do this properly. 2022 */ 2023 #ifdef notyet 2024 KASSERT(fs->lfs_seglock == 1); 2025 #endif /* notyet */ 2026 2027 /* 2028 * We assume we're being called with sp->fip pointing at blank space. 2029 * Account for a new FIP in the segment header, and set sp->vp. 2030 * (This should duplicate the setup at the top of lfs_writefile().) 2031 */ 2032 sp = fs->lfs_sp; 2033 if (!seglocked) { 2034 if (sp->seg_bytes_left < fs->lfs_bsize || 2035 sp->sum_bytes_left < sizeof(struct finfo)) 2036 (void) lfs_writeseg(fs, fs->lfs_sp); 2037 2038 sp->sum_bytes_left -= FINFOSIZE; 2039 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 2040 } 2041 KASSERT(sp->vp == NULL); 2042 sp->vp = vp; 2043 2044 if (!seglocked) { 2045 if (vp->v_flag & VDIROP) 2046 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2047 } 2048 2049 sp->fip->fi_nblocks = 0; 2050 sp->fip->fi_ino = ip->i_number; 2051 sp->fip->fi_version = ip->i_gen; 2052 2053 /* 2054 * Loop through genfs_putpages until all pages are gathered. 2055 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2056 * Whenever we lose the interlock we have to rerun check_dirty, as 2057 * well. 2058 */ 2059 again: 2060 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2061 ap->a_flags, 0) < 0) { 2062 simple_unlock(&vp->v_interlock); 2063 sp->vp = NULL; 2064 if (!seglocked) 2065 lfs_segunlock(fs); 2066 return EDEADLK; 2067 } 2068 2069 error = genfs_putpages(v); 2070 if (error == EDEADLK || error == EAGAIN) { 2071 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2072 " EDEADLK [2] ino %d off %x (seg %d)\n", 2073 ip->i_number, fs->lfs_offset, 2074 dtosn(fs, fs->lfs_offset))); 2075 /* If nothing to write, short-circuit */ 2076 if (sp->cbpp - sp->bpp > 1) { 2077 /* Write gathered pages */ 2078 lfs_updatemeta(sp); 2079 (void) lfs_writeseg(fs, sp); 2080 2081 /* 2082 * Reinitialize brand new FIP and add us to it. 2083 * (This should duplicate the fixup in 2084 * lfs_gatherpages().) 2085 */ 2086 KASSERT(sp->vp == vp); 2087 sp->fip->fi_version = ip->i_gen; 2088 sp->fip->fi_ino = ip->i_number; 2089 /* Add us to the new segment summary. */ 2090 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 2091 sp->sum_bytes_left -= FINFOSIZE; 2092 } 2093 2094 /* Give the write a chance to complete */ 2095 preempt(1); 2096 2097 /* We've lost the interlock. Start over. */ 2098 if (error == EDEADLK) { 2099 simple_lock(&vp->v_interlock); 2100 goto again; 2101 } 2102 } 2103 2104 KASSERT(sp->vp == vp); 2105 if (!seglocked) { 2106 sp->vp = NULL; /* XXX lfs_gather below will set this */ 2107 2108 /* Write indirect blocks as well */ 2109 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2110 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2111 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2112 2113 KASSERT(sp->vp == NULL); 2114 sp->vp = vp; 2115 } 2116 2117 /* 2118 * Blocks are now gathered into a segment waiting to be written. 2119 * All that's left to do is update metadata, and write them. 2120 */ 2121 lfs_updatemeta(sp); 2122 KASSERT(sp->vp == vp); 2123 sp->vp = NULL; 2124 2125 if (seglocked) { 2126 /* we're called by lfs_writefile. */ 2127 return error; 2128 } 2129 2130 /* 2131 * Clean up FIP, since we're done writing this file. 2132 * This should duplicate cleanup at the end of lfs_writefile(). 2133 */ 2134 if (sp->fip->fi_nblocks != 0) { 2135 sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE + 2136 sizeof(int32_t) * sp->fip->fi_nblocks); 2137 sp->start_lbp = &sp->fip->fi_blocks[0]; 2138 } else { 2139 sp->sum_bytes_left += FINFOSIZE; 2140 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 2141 } 2142 lfs_writeseg(fs, fs->lfs_sp); 2143 2144 /* 2145 * Remove us from paging queue, since we've now written all our 2146 * pages. 2147 */ 2148 simple_lock(&fs->lfs_interlock); 2149 if (ip->i_flags & IN_PAGING) { 2150 ip->i_flags &= ~IN_PAGING; 2151 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2152 } 2153 simple_unlock(&fs->lfs_interlock); 2154 2155 /* 2156 * XXX - with the malloc/copy writeseg, the pages are freed by now 2157 * even if we don't wait (e.g. if we hold a nested lock). This 2158 * will not be true if we stop using malloc/copy. 2159 */ 2160 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2161 lfs_segunlock(fs); 2162 2163 /* 2164 * Wait for v_numoutput to drop to zero. The seglock should 2165 * take care of this, but there is a slight possibility that 2166 * aiodoned might not have got around to our buffers yet. 2167 */ 2168 if (sync) { 2169 s = splbio(); 2170 simple_lock(&global_v_numoutput_slock); 2171 while (vp->v_numoutput > 0) { 2172 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2173 " num %d\n", ip->i_number, vp->v_numoutput)); 2174 vp->v_flag |= VBWAIT; 2175 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2176 &global_v_numoutput_slock); 2177 } 2178 simple_unlock(&global_v_numoutput_slock); 2179 splx(s); 2180 } 2181 return error; 2182 } 2183 2184 /* 2185 * Return the last logical file offset that should be written for this file 2186 * if we're doing a write that ends at "size". If writing, we need to know 2187 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2188 * to know about entire blocks. 2189 */ 2190 void 2191 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2192 { 2193 struct inode *ip = VTOI(vp); 2194 struct lfs *fs = ip->i_lfs; 2195 daddr_t olbn, nlbn; 2196 2197 KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)); 2198 KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)) 2199 != (GOP_SIZE_READ | GOP_SIZE_WRITE)); 2200 2201 olbn = lblkno(fs, ip->i_size); 2202 nlbn = lblkno(fs, size); 2203 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2204 *eobp = fragroundup(fs, size); 2205 } else { 2206 *eobp = blkroundup(fs, size); 2207 } 2208 } 2209 2210 #ifdef DEBUG 2211 void lfs_dump_vop(void *); 2212 2213 void 2214 lfs_dump_vop(void *v) 2215 { 2216 struct vop_putpages_args /* { 2217 struct vnode *a_vp; 2218 voff_t a_offlo; 2219 voff_t a_offhi; 2220 int a_flags; 2221 } */ *ap = v; 2222 2223 #ifdef DDB 2224 vfs_vnode_print(ap->a_vp, 0, printf); 2225 #endif 2226 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2227 } 2228 #endif 2229 2230 int 2231 lfs_mmap(void *v) 2232 { 2233 struct vop_mmap_args /* { 2234 const struct vnodeop_desc *a_desc; 2235 struct vnode *a_vp; 2236 int a_fflags; 2237 struct ucred *a_cred; 2238 struct proc *a_p; 2239 } */ *ap = v; 2240 2241 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2242 return EOPNOTSUPP; 2243 return ufs_mmap(v); 2244 } 2245
Indexes created Sun Oct 19 18:09:56 GMT 2025