lfs_vnops.c revision 1.153
1 /* $NetBSD: lfs_vnops.c,v 1.153 2005/08/19 02:04:09 christos 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 1.153 2005/08/19 02:04:09 christos 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 %llu)", 639 (unsigned long long)ino); 640 /* return (error); */ 641 } 642 /* 643 * Remove vnode so that it will be reloaded by VFS_VGET and 644 * checked to see if it is an alias of an existing entry in 645 * the inode cache. 646 */ 647 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 648 649 VOP_UNLOCK(*vpp, 0); 650 lfs_vunref(*vpp); 651 (*vpp)->v_type = VNON; 652 vgone(*vpp); 653 error = VFS_VGET(mp, ino, vpp); 654 655 if (error != 0) { 656 *vpp = NULL; 657 return (error); 658 } 659 return (0); 660 } 661 662 int 663 lfs_create(void *v) 664 { 665 struct vop_create_args /* { 666 struct vnode *a_dvp; 667 struct vnode **a_vpp; 668 struct componentname *a_cnp; 669 struct vattr *a_vap; 670 } */ *ap = v; 671 int error; 672 673 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 674 vput(ap->a_dvp); 675 return error; 676 } 677 error = ufs_create(ap); 678 SET_ENDOP_CREATE_AP(ap, "create"); 679 return (error); 680 } 681 682 int 683 lfs_mkdir(void *v) 684 { 685 struct vop_mkdir_args /* { 686 struct vnode *a_dvp; 687 struct vnode **a_vpp; 688 struct componentname *a_cnp; 689 struct vattr *a_vap; 690 } */ *ap = v; 691 int error; 692 693 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 694 vput(ap->a_dvp); 695 return error; 696 } 697 error = ufs_mkdir(ap); 698 SET_ENDOP_CREATE_AP(ap, "mkdir"); 699 return (error); 700 } 701 702 int 703 lfs_remove(void *v) 704 { 705 struct vop_remove_args /* { 706 struct vnode *a_dvp; 707 struct vnode *a_vp; 708 struct componentname *a_cnp; 709 } */ *ap = v; 710 struct vnode *dvp, *vp; 711 int error; 712 713 dvp = ap->a_dvp; 714 vp = ap->a_vp; 715 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 716 if (dvp == vp) 717 vrele(vp); 718 else 719 vput(vp); 720 vput(dvp); 721 return error; 722 } 723 error = ufs_remove(ap); 724 SET_ENDOP_REMOVE(VTOI(dvp)->i_lfs, dvp, vp, "remove"); 725 return (error); 726 } 727 728 int 729 lfs_rmdir(void *v) 730 { 731 struct vop_rmdir_args /* { 732 struct vnodeop_desc *a_desc; 733 struct vnode *a_dvp; 734 struct vnode *a_vp; 735 struct componentname *a_cnp; 736 } */ *ap = v; 737 struct vnode *vp; 738 int error; 739 740 vp = ap->a_vp; 741 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 742 vrele(ap->a_dvp); 743 if (ap->a_vp != ap->a_dvp) 744 VOP_UNLOCK(ap->a_dvp, 0); 745 vput(vp); 746 return error; 747 } 748 error = ufs_rmdir(ap); 749 SET_ENDOP_REMOVE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vp, "rmdir"); 750 return (error); 751 } 752 753 int 754 lfs_link(void *v) 755 { 756 struct vop_link_args /* { 757 struct vnode *a_dvp; 758 struct vnode *a_vp; 759 struct componentname *a_cnp; 760 } */ *ap = v; 761 int error; 762 struct vnode **vpp = NULL; 763 764 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 765 vput(ap->a_dvp); 766 return error; 767 } 768 error = ufs_link(ap); 769 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 770 return (error); 771 } 772 773 int 774 lfs_rename(void *v) 775 { 776 struct vop_rename_args /* { 777 struct vnode *a_fdvp; 778 struct vnode *a_fvp; 779 struct componentname *a_fcnp; 780 struct vnode *a_tdvp; 781 struct vnode *a_tvp; 782 struct componentname *a_tcnp; 783 } */ *ap = v; 784 struct vnode *tvp, *fvp, *tdvp, *fdvp; 785 struct componentname *tcnp, *fcnp; 786 int error; 787 struct lfs *fs; 788 789 fs = VTOI(ap->a_fdvp)->i_lfs; 790 tvp = ap->a_tvp; 791 tdvp = ap->a_tdvp; 792 tcnp = ap->a_tcnp; 793 fvp = ap->a_fvp; 794 fdvp = ap->a_fdvp; 795 fcnp = ap->a_fcnp; 796 797 /* 798 * Check for cross-device rename. 799 * If it is, we don't want to set dirops, just error out. 800 * (In particular note that MARK_VNODE(tdvp) will DTWT on 801 * a cross-device rename.) 802 * 803 * Copied from ufs_rename. 804 */ 805 if ((fvp->v_mount != tdvp->v_mount) || 806 (tvp && (fvp->v_mount != tvp->v_mount))) { 807 error = EXDEV; 808 goto errout; 809 } 810 811 /* 812 * Check to make sure we're not renaming a vnode onto itself 813 * (deleting a hard link by renaming one name onto another); 814 * if we are we can't recursively call VOP_REMOVE since that 815 * would leave us with an unaccounted-for number of live dirops. 816 * 817 * Inline the relevant section of ufs_rename here, *before* 818 * calling SET_DIROP_REMOVE. 819 */ 820 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 821 (VTOI(tdvp)->i_flags & APPEND))) { 822 error = EPERM; 823 goto errout; 824 } 825 if (fvp == tvp) { 826 if (fvp->v_type == VDIR) { 827 error = EINVAL; 828 goto errout; 829 } 830 831 /* Release destination completely. */ 832 VOP_ABORTOP(tdvp, tcnp); 833 vput(tdvp); 834 vput(tvp); 835 836 /* Delete source. */ 837 vrele(fvp); 838 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 839 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 840 fcnp->cn_nameiop = DELETE; 841 if ((error = relookup(fdvp, &fvp, fcnp))){ 842 /* relookup blew away fdvp */ 843 return (error); 844 } 845 return (VOP_REMOVE(fdvp, fvp, fcnp)); 846 } 847 848 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 849 goto errout; 850 MARK_VNODE(fdvp); 851 MARK_VNODE(fvp); 852 853 error = ufs_rename(ap); 854 UNMARK_VNODE(fdvp); 855 UNMARK_VNODE(fvp); 856 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 857 return (error); 858 859 errout: 860 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 861 if (tdvp == tvp) 862 vrele(tdvp); 863 else 864 vput(tdvp); 865 if (tvp) 866 vput(tvp); 867 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 868 vrele(fdvp); 869 vrele(fvp); 870 return (error); 871 } 872 873 /* XXX hack to avoid calling ITIMES in getattr */ 874 int 875 lfs_getattr(void *v) 876 { 877 struct vop_getattr_args /* { 878 struct vnode *a_vp; 879 struct vattr *a_vap; 880 struct ucred *a_cred; 881 struct proc *a_p; 882 } */ *ap = v; 883 struct vnode *vp = ap->a_vp; 884 struct inode *ip = VTOI(vp); 885 struct vattr *vap = ap->a_vap; 886 struct lfs *fs = ip->i_lfs; 887 /* 888 * Copy from inode table 889 */ 890 vap->va_fsid = ip->i_dev; 891 vap->va_fileid = ip->i_number; 892 vap->va_mode = ip->i_mode & ~IFMT; 893 vap->va_nlink = ip->i_nlink; 894 vap->va_uid = ip->i_uid; 895 vap->va_gid = ip->i_gid; 896 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 897 vap->va_size = vp->v_size; 898 vap->va_atime.tv_sec = ip->i_ffs1_atime; 899 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 900 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 901 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 902 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 903 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 904 vap->va_flags = ip->i_flags; 905 vap->va_gen = ip->i_gen; 906 /* this doesn't belong here */ 907 if (vp->v_type == VBLK) 908 vap->va_blocksize = BLKDEV_IOSIZE; 909 else if (vp->v_type == VCHR) 910 vap->va_blocksize = MAXBSIZE; 911 else 912 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 913 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 914 vap->va_type = vp->v_type; 915 vap->va_filerev = ip->i_modrev; 916 return (0); 917 } 918 919 /* 920 * Check to make sure the inode blocks won't choke the buffer 921 * cache, then call ufs_setattr as usual. 922 */ 923 int 924 lfs_setattr(void *v) 925 { 926 struct vop_setattr_args /* { 927 struct vnode *a_vp; 928 struct vattr *a_vap; 929 struct ucred *a_cred; 930 struct proc *a_p; 931 } */ *ap = v; 932 struct vnode *vp = ap->a_vp; 933 934 lfs_check(vp, LFS_UNUSED_LBN, 0); 935 return ufs_setattr(v); 936 } 937 938 /* 939 * Close called 940 * 941 * XXX -- we were using ufs_close, but since it updates the 942 * times on the inode, we might need to bump the uinodes 943 * count. 944 */ 945 /* ARGSUSED */ 946 int 947 lfs_close(void *v) 948 { 949 struct vop_close_args /* { 950 struct vnode *a_vp; 951 int a_fflag; 952 struct ucred *a_cred; 953 struct proc *a_p; 954 } */ *ap = v; 955 struct vnode *vp = ap->a_vp; 956 struct inode *ip = VTOI(vp); 957 struct timespec ts; 958 959 if (vp == ip->i_lfs->lfs_ivnode && 960 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 961 return 0; 962 963 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 964 TIMEVAL_TO_TIMESPEC(&time, &ts); 965 LFS_ITIMES(ip, &ts, &ts, &ts); 966 } 967 return (0); 968 } 969 970 /* 971 * Close wrapper for special devices. 972 * 973 * Update the times on the inode then do device close. 974 */ 975 int 976 lfsspec_close(void *v) 977 { 978 struct vop_close_args /* { 979 struct vnode *a_vp; 980 int a_fflag; 981 struct ucred *a_cred; 982 struct proc *a_p; 983 } */ *ap = v; 984 struct vnode *vp; 985 struct inode *ip; 986 struct timespec ts; 987 988 vp = ap->a_vp; 989 ip = VTOI(vp); 990 if (vp->v_usecount > 1) { 991 TIMEVAL_TO_TIMESPEC(&time, &ts); 992 LFS_ITIMES(ip, &ts, &ts, &ts); 993 } 994 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 995 } 996 997 /* 998 * Close wrapper for fifo's. 999 * 1000 * Update the times on the inode then do device close. 1001 */ 1002 int 1003 lfsfifo_close(void *v) 1004 { 1005 struct vop_close_args /* { 1006 struct vnode *a_vp; 1007 int a_fflag; 1008 struct ucred *a_cred; 1009 struct proc *a_p; 1010 } */ *ap = v; 1011 struct vnode *vp; 1012 struct inode *ip; 1013 struct timespec ts; 1014 1015 vp = ap->a_vp; 1016 ip = VTOI(vp); 1017 if (ap->a_vp->v_usecount > 1) { 1018 TIMEVAL_TO_TIMESPEC(&time, &ts); 1019 LFS_ITIMES(ip, &ts, &ts, &ts); 1020 } 1021 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1022 } 1023 1024 /* 1025 * Reclaim an inode so that it can be used for other purposes. 1026 */ 1027 1028 int 1029 lfs_reclaim(void *v) 1030 { 1031 struct vop_reclaim_args /* { 1032 struct vnode *a_vp; 1033 struct proc *a_p; 1034 } */ *ap = v; 1035 struct vnode *vp = ap->a_vp; 1036 struct inode *ip = VTOI(vp); 1037 int error; 1038 1039 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1040 1041 LFS_CLR_UINO(ip, IN_ALLMOD); 1042 if ((error = ufs_reclaim(vp, ap->a_p))) 1043 return (error); 1044 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1045 lfs_deregister_all(vp); 1046 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1047 ip->inode_ext.lfs = NULL; 1048 pool_put(&lfs_inode_pool, vp->v_data); 1049 vp->v_data = NULL; 1050 return (0); 1051 } 1052 1053 /* 1054 * Read a block from a storage device. 1055 * In order to avoid reading blocks that are in the process of being 1056 * written by the cleaner---and hence are not mutexed by the normal 1057 * buffer cache / page cache mechanisms---check for collisions before 1058 * reading. 1059 * 1060 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1061 * the active cleaner test. 1062 * 1063 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1064 */ 1065 int 1066 lfs_strategy(void *v) 1067 { 1068 struct vop_strategy_args /* { 1069 struct vnode *a_vp; 1070 struct buf *a_bp; 1071 } */ *ap = v; 1072 struct buf *bp; 1073 struct lfs *fs; 1074 struct vnode *vp; 1075 struct inode *ip; 1076 daddr_t tbn; 1077 int i, sn, error, slept; 1078 1079 bp = ap->a_bp; 1080 vp = ap->a_vp; 1081 ip = VTOI(vp); 1082 fs = ip->i_lfs; 1083 1084 /* lfs uses its strategy routine only for read */ 1085 KASSERT(bp->b_flags & B_READ); 1086 1087 if (vp->v_type == VBLK || vp->v_type == VCHR) 1088 panic("lfs_strategy: spec"); 1089 KASSERT(bp->b_bcount != 0); 1090 if (bp->b_blkno == bp->b_lblkno) { 1091 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1092 NULL); 1093 if (error) { 1094 bp->b_error = error; 1095 bp->b_flags |= B_ERROR; 1096 biodone(bp); 1097 return (error); 1098 } 1099 if ((long)bp->b_blkno == -1) /* no valid data */ 1100 clrbuf(bp); 1101 } 1102 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1103 biodone(bp); 1104 return (0); 1105 } 1106 1107 slept = 1; 1108 simple_lock(&fs->lfs_interlock); 1109 while (slept && fs->lfs_seglock) { 1110 simple_unlock(&fs->lfs_interlock); 1111 /* 1112 * Look through list of intervals. 1113 * There will only be intervals to look through 1114 * if the cleaner holds the seglock. 1115 * Since the cleaner is synchronous, we can trust 1116 * the list of intervals to be current. 1117 */ 1118 tbn = dbtofsb(fs, bp->b_blkno); 1119 sn = dtosn(fs, tbn); 1120 slept = 0; 1121 for (i = 0; i < fs->lfs_cleanind; i++) { 1122 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1123 tbn >= fs->lfs_cleanint[i]) { 1124 DLOG((DLOG_CLEAN, 1125 "lfs_strategy: ino %d lbn %" PRId64 1126 " ind %d sn %d fsb %" PRIx32 1127 " given sn %d fsb %" PRIx64 "\n", 1128 ip->i_number, bp->b_lblkno, i, 1129 dtosn(fs, fs->lfs_cleanint[i]), 1130 fs->lfs_cleanint[i], sn, tbn)); 1131 DLOG((DLOG_CLEAN, 1132 "lfs_strategy: sleeping on ino %d lbn %" 1133 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1134 simple_lock(&fs->lfs_interlock); 1135 if (fs->lfs_seglock) 1136 ltsleep(&fs->lfs_seglock, 1137 (PRIBIO + 1) | PNORELOCK, 1138 "lfs_strategy", 0, 1139 &fs->lfs_interlock); 1140 /* Things may be different now; start over. */ 1141 slept = 1; 1142 break; 1143 } 1144 } 1145 simple_lock(&fs->lfs_interlock); 1146 } 1147 simple_unlock(&fs->lfs_interlock); 1148 1149 vp = ip->i_devvp; 1150 VOP_STRATEGY(vp, bp); 1151 return (0); 1152 } 1153 1154 static void 1155 lfs_flush_dirops(struct lfs *fs) 1156 { 1157 struct inode *ip, *nip; 1158 struct vnode *vp; 1159 extern int lfs_dostats; 1160 struct segment *sp; 1161 int needunlock; 1162 1163 ASSERT_NO_SEGLOCK(fs); 1164 1165 if (fs->lfs_ronly) 1166 return; 1167 1168 simple_lock(&fs->lfs_interlock); 1169 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1170 simple_unlock(&fs->lfs_interlock); 1171 return; 1172 } else 1173 simple_unlock(&fs->lfs_interlock); 1174 1175 if (lfs_dostats) 1176 ++lfs_stats.flush_invoked; 1177 1178 /* 1179 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1180 * Technically this is a checkpoint (the on-disk state is valid) 1181 * even though we are leaving out all the file data. 1182 */ 1183 lfs_imtime(fs); 1184 lfs_seglock(fs, SEGM_CKP); 1185 sp = fs->lfs_sp; 1186 1187 /* 1188 * lfs_writevnodes, optimized to get dirops out of the way. 1189 * Only write dirops, and don't flush files' pages, only 1190 * blocks from the directories. 1191 * 1192 * We don't need to vref these files because they are 1193 * dirops and so hold an extra reference until the 1194 * segunlock clears them of that status. 1195 * 1196 * We don't need to check for IN_ADIROP because we know that 1197 * no dirops are active. 1198 * 1199 */ 1200 simple_lock(&fs->lfs_interlock); 1201 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1202 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1203 simple_unlock(&fs->lfs_interlock); 1204 vp = ITOV(ip); 1205 1206 /* 1207 * All writes to directories come from dirops; all 1208 * writes to files' direct blocks go through the page 1209 * cache, which we're not touching. Reads to files 1210 * and/or directories will not be affected by writing 1211 * directory blocks inodes and file inodes. So we don't 1212 * really need to lock. If we don't lock, though, 1213 * make sure that we don't clear IN_MODIFIED 1214 * unnecessarily. 1215 */ 1216 if (vp->v_flag & VXLOCK) 1217 continue; 1218 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) { 1219 needunlock = 1; 1220 } else { 1221 DLOG((DLOG_VNODE, "lfs_flush_dirops: flushing locked ino %d\n", 1222 VTOI(vp)->i_number)); 1223 needunlock = 0; 1224 } 1225 if (vp->v_type != VREG && 1226 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1227 lfs_writefile(fs, sp, vp); 1228 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1229 !(ip->i_flag & IN_ALLMOD)) { 1230 LFS_SET_UINO(ip, IN_MODIFIED); 1231 } 1232 } 1233 (void) lfs_writeinode(fs, sp, ip); 1234 if (needunlock) 1235 VOP_UNLOCK(vp, 0); 1236 else 1237 LFS_SET_UINO(ip, IN_MODIFIED); 1238 simple_lock(&fs->lfs_interlock); 1239 } 1240 simple_unlock(&fs->lfs_interlock); 1241 /* We've written all the dirops there are */ 1242 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1243 (void) lfs_writeseg(fs, sp); 1244 lfs_segunlock(fs); 1245 } 1246 1247 /* 1248 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1249 */ 1250 int 1251 lfs_fcntl(void *v) 1252 { 1253 struct vop_fcntl_args /* { 1254 struct vnode *a_vp; 1255 u_long a_command; 1256 caddr_t a_data; 1257 int a_fflag; 1258 struct ucred *a_cred; 1259 struct proc *a_p; 1260 } */ *ap = v; 1261 struct timeval *tvp; 1262 BLOCK_INFO *blkiov; 1263 CLEANERINFO *cip; 1264 SEGUSE *sup; 1265 int blkcnt, error, oclean; 1266 struct lfs_fcntl_markv blkvp; 1267 fsid_t *fsidp; 1268 struct lfs *fs; 1269 struct buf *bp; 1270 fhandle_t *fhp; 1271 daddr_t off; 1272 1273 /* Only respect LFS fcntls on fs root or Ifile */ 1274 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1275 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1276 return ufs_fcntl(v); 1277 } 1278 1279 /* Avoid locking a draining lock */ 1280 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1281 return ESHUTDOWN; 1282 } 1283 1284 fs = VTOI(ap->a_vp)->i_lfs; 1285 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1286 1287 switch (ap->a_command) { 1288 case LFCNSEGWAITALL: 1289 case LFCNSEGWAITALL_COMPAT: 1290 fsidp = NULL; 1291 /* FALLSTHROUGH */ 1292 case LFCNSEGWAIT: 1293 case LFCNSEGWAIT_COMPAT: 1294 tvp = (struct timeval *)ap->a_data; 1295 simple_lock(&fs->lfs_interlock); 1296 ++fs->lfs_sleepers; 1297 simple_unlock(&fs->lfs_interlock); 1298 VOP_UNLOCK(ap->a_vp, 0); 1299 1300 error = lfs_segwait(fsidp, tvp); 1301 1302 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1303 simple_lock(&fs->lfs_interlock); 1304 if (--fs->lfs_sleepers == 0) 1305 wakeup(&fs->lfs_sleepers); 1306 simple_unlock(&fs->lfs_interlock); 1307 return error; 1308 1309 case LFCNBMAPV: 1310 case LFCNMARKV: 1311 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1312 return (error); 1313 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1314 1315 blkcnt = blkvp.blkcnt; 1316 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1317 return (EINVAL); 1318 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1319 if ((error = copyin(blkvp.blkiov, blkiov, 1320 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1321 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1322 return error; 1323 } 1324 1325 simple_lock(&fs->lfs_interlock); 1326 ++fs->lfs_sleepers; 1327 simple_unlock(&fs->lfs_interlock); 1328 VOP_UNLOCK(ap->a_vp, 0); 1329 if (ap->a_command == LFCNBMAPV) 1330 error = lfs_bmapv(ap->a_p, fsidp, blkiov, blkcnt); 1331 else /* LFCNMARKV */ 1332 error = lfs_markv(ap->a_p, fsidp, blkiov, blkcnt); 1333 if (error == 0) 1334 error = copyout(blkiov, blkvp.blkiov, 1335 blkcnt * sizeof(BLOCK_INFO)); 1336 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1337 simple_lock(&fs->lfs_interlock); 1338 if (--fs->lfs_sleepers == 0) 1339 wakeup(&fs->lfs_sleepers); 1340 simple_unlock(&fs->lfs_interlock); 1341 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1342 return error; 1343 1344 case LFCNRECLAIM: 1345 /* 1346 * Flush dirops and write Ifile, allowing empty segments 1347 * to be immediately reclaimed. 1348 */ 1349 VOP_UNLOCK(ap->a_vp, 0); 1350 lfs_writer_enter(fs, "pndirop"); 1351 off = fs->lfs_offset; 1352 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1353 lfs_flush_dirops(fs); 1354 LFS_CLEANERINFO(cip, fs, bp); 1355 oclean = cip->clean; 1356 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1357 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1358 lfs_segunlock(fs); 1359 lfs_writer_leave(fs); 1360 1361 #ifdef DEBUG 1362 LFS_CLEANERINFO(cip, fs, bp); 1363 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1364 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1365 fs->lfs_offset - off, cip->clean - oclean, 1366 fs->lfs_activesb)); 1367 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1368 #endif 1369 1370 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1371 return 0; 1372 1373 case LFCNIFILEFH: 1374 /* Return the filehandle of the Ifile */ 1375 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1376 return (error); 1377 fhp = (struct fhandle *)ap->a_data; 1378 fhp->fh_fsid = *fsidp; 1379 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid)); 1380 1381 case LFCNREWIND: 1382 /* Move lfs_offset to the lowest-numbered segment */ 1383 return lfs_rewind(fs, *(int *)ap->a_data); 1384 1385 case LFCNINVAL: 1386 /* Mark a segment SEGUSE_INVAL */ 1387 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1388 if (sup->su_nbytes > 0) { 1389 brelse(bp); 1390 lfs_unset_inval_all(fs); 1391 return EBUSY; 1392 } 1393 sup->su_flags |= SEGUSE_INVAL; 1394 VOP_BWRITE(bp); 1395 return 0; 1396 1397 case LFCNRESIZE: 1398 /* Resize the filesystem */ 1399 return lfs_resize_fs(fs, *(int *)ap->a_data); 1400 1401 default: 1402 return ufs_fcntl(v); 1403 } 1404 return 0; 1405 } 1406 1407 int 1408 lfs_getpages(void *v) 1409 { 1410 struct vop_getpages_args /* { 1411 struct vnode *a_vp; 1412 voff_t a_offset; 1413 struct vm_page **a_m; 1414 int *a_count; 1415 int a_centeridx; 1416 vm_prot_t a_access_type; 1417 int a_advice; 1418 int a_flags; 1419 } */ *ap = v; 1420 1421 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1422 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1423 return EPERM; 1424 } 1425 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1426 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1427 } 1428 1429 /* 1430 * we're relying on the fact that genfs_getpages() always read in 1431 * entire filesystem blocks. 1432 */ 1433 return genfs_getpages(v); 1434 } 1435 1436 /* 1437 * Make sure that for all pages in every block in the given range, 1438 * either all are dirty or all are clean. If any of the pages 1439 * we've seen so far are dirty, put the vnode on the paging chain, 1440 * and mark it IN_PAGING. 1441 * 1442 * If checkfirst != 0, don't check all the pages but return at the 1443 * first dirty page. 1444 */ 1445 static int 1446 check_dirty(struct lfs *fs, struct vnode *vp, 1447 off_t startoffset, off_t endoffset, off_t blkeof, 1448 int flags, int checkfirst) 1449 { 1450 int by_list; 1451 struct vm_page *curpg = NULL; /* XXX: gcc */ 1452 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1453 off_t soff = 0; /* XXX: gcc */ 1454 voff_t off; 1455 int i; 1456 int nonexistent; 1457 int any_dirty; /* number of dirty pages */ 1458 int dirty; /* number of dirty pages in a block */ 1459 int tdirty; 1460 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1461 1462 ASSERT_MAYBE_SEGLOCK(fs); 1463 top: 1464 by_list = (vp->v_uobj.uo_npages <= 1465 ((endoffset - startoffset) >> PAGE_SHIFT) * 1466 UVM_PAGE_HASH_PENALTY); 1467 any_dirty = 0; 1468 1469 if (by_list) { 1470 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1471 } else { 1472 soff = startoffset; 1473 } 1474 while (by_list || soff < MIN(blkeof, endoffset)) { 1475 if (by_list) { 1476 /* 1477 * Find the first page in a block. Skip 1478 * blocks outside our area of interest or beyond 1479 * the end of file. 1480 */ 1481 if (pages_per_block > 1) { 1482 while (curpg && 1483 ((curpg->offset & fs->lfs_bmask) || 1484 curpg->offset >= vp->v_size || 1485 curpg->offset >= endoffset)) 1486 curpg = TAILQ_NEXT(curpg, listq); 1487 } 1488 if (curpg == NULL) 1489 break; 1490 soff = curpg->offset; 1491 } 1492 1493 /* 1494 * Mark all pages in extended range busy; find out if any 1495 * of them are dirty. 1496 */ 1497 nonexistent = dirty = 0; 1498 for (i = 0; i == 0 || i < pages_per_block; i++) { 1499 if (by_list && pages_per_block <= 1) { 1500 pgs[i] = pg = curpg; 1501 } else { 1502 off = soff + (i << PAGE_SHIFT); 1503 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1504 if (pg == NULL) { 1505 ++nonexistent; 1506 continue; 1507 } 1508 } 1509 KASSERT(pg != NULL); 1510 while (pg->flags & PG_BUSY) { 1511 pg->flags |= PG_WANTED; 1512 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1513 "lfsput", 0); 1514 simple_lock(&vp->v_interlock); 1515 if (by_list) { 1516 if (i > 0) 1517 uvm_page_unbusy(pgs, i); 1518 goto top; 1519 } 1520 } 1521 pg->flags |= PG_BUSY; 1522 UVM_PAGE_OWN(pg, "lfs_putpages"); 1523 1524 pmap_page_protect(pg, VM_PROT_NONE); 1525 tdirty = (pmap_clear_modify(pg) || 1526 (pg->flags & PG_CLEAN) == 0); 1527 dirty += tdirty; 1528 } 1529 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1530 if (by_list) { 1531 curpg = TAILQ_NEXT(curpg, listq); 1532 } else { 1533 soff += fs->lfs_bsize; 1534 } 1535 continue; 1536 } 1537 1538 any_dirty += dirty; 1539 KASSERT(nonexistent == 0); 1540 1541 /* 1542 * If any are dirty make all dirty; unbusy them, 1543 * but if we were asked to clean, wire them so that 1544 * the pagedaemon doesn't bother us about them while 1545 * they're on their way to disk. 1546 */ 1547 for (i = 0; i == 0 || i < pages_per_block; i++) { 1548 pg = pgs[i]; 1549 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1550 if (dirty) { 1551 pg->flags &= ~PG_CLEAN; 1552 if (flags & PGO_FREE) { 1553 /* 1554 * Wire the page so that 1555 * pdaemon doesn't see it again. 1556 */ 1557 uvm_lock_pageq(); 1558 uvm_pagewire(pg); 1559 uvm_unlock_pageq(); 1560 1561 /* Suspended write flag */ 1562 pg->flags |= PG_DELWRI; 1563 } 1564 } 1565 if (pg->flags & PG_WANTED) 1566 wakeup(pg); 1567 pg->flags &= ~(PG_WANTED|PG_BUSY); 1568 UVM_PAGE_OWN(pg, NULL); 1569 } 1570 1571 if (checkfirst && any_dirty) 1572 break; 1573 1574 if (by_list) { 1575 curpg = TAILQ_NEXT(curpg, listq); 1576 } else { 1577 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1578 } 1579 } 1580 1581 /* 1582 * If any pages were dirty, mark this inode as "pageout requested", 1583 * and put it on the paging queue. 1584 * XXXUBC locking (check locking on dchainhd too) 1585 */ 1586 #ifdef notyet 1587 if (any_dirty) { 1588 if (!(ip->i_flags & IN_PAGING)) { 1589 ip->i_flags |= IN_PAGING; 1590 simple_lock(&fs->lfs_interlock); 1591 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1592 simple_unlock(&fs->lfs_interlock); 1593 } 1594 } 1595 #endif 1596 return any_dirty; 1597 } 1598 1599 /* 1600 * lfs_putpages functions like genfs_putpages except that 1601 * 1602 * (1) It needs to bounds-check the incoming requests to ensure that 1603 * they are block-aligned; if they are not, expand the range and 1604 * do the right thing in case, e.g., the requested range is clean 1605 * but the expanded range is dirty. 1606 * (2) It needs to explicitly send blocks to be written when it is done. 1607 * VOP_PUTPAGES is not ever called with the seglock held, so 1608 * we simply take the seglock and let lfs_segunlock wait for us. 1609 * XXX Actually we can be called with the seglock held, if we have 1610 * XXX to flush a vnode while lfs_markv is in operation. As of this 1611 * XXX writing we panic in this case. 1612 * 1613 * Assumptions: 1614 * 1615 * (1) The caller does not hold any pages in this vnode busy. If it does, 1616 * there is a danger that when we expand the page range and busy the 1617 * pages we will deadlock. 1618 * (2) We are called with vp->v_interlock held; we must return with it 1619 * released. 1620 * (3) We don't absolutely have to free pages right away, provided that 1621 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1622 * us a request with PGO_FREE, we take the pages out of the paging 1623 * queue and wake up the writer, which will handle freeing them for us. 1624 * 1625 * We ensure that for any filesystem block, all pages for that 1626 * block are either resident or not, even if those pages are higher 1627 * than EOF; that means that we will be getting requests to free 1628 * "unused" pages above EOF all the time, and should ignore them. 1629 * 1630 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1631 */ 1632 1633 int 1634 lfs_putpages(void *v) 1635 { 1636 int error; 1637 struct vop_putpages_args /* { 1638 struct vnode *a_vp; 1639 voff_t a_offlo; 1640 voff_t a_offhi; 1641 int a_flags; 1642 } */ *ap = v; 1643 struct vnode *vp; 1644 struct inode *ip; 1645 struct lfs *fs; 1646 struct segment *sp; 1647 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1648 off_t off, max_endoffset; 1649 int s; 1650 boolean_t seglocked, sync, pagedaemon; 1651 struct vm_page *pg; 1652 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1653 1654 vp = ap->a_vp; 1655 ip = VTOI(vp); 1656 fs = ip->i_lfs; 1657 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1658 pagedaemon = (curproc == uvm.pagedaemon_proc); 1659 1660 /* Putpages does nothing for metadata. */ 1661 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1662 simple_unlock(&vp->v_interlock); 1663 return 0; 1664 } 1665 1666 /* 1667 * If there are no pages, don't do anything. 1668 */ 1669 if (vp->v_uobj.uo_npages == 0) { 1670 s = splbio(); 1671 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1672 (vp->v_flag & VONWORKLST)) { 1673 vp->v_flag &= ~VONWORKLST; 1674 LIST_REMOVE(vp, v_synclist); 1675 } 1676 splx(s); 1677 simple_unlock(&vp->v_interlock); 1678 return 0; 1679 } 1680 1681 blkeof = blkroundup(fs, ip->i_size); 1682 1683 /* 1684 * Ignore requests to free pages past EOF but in the same block 1685 * as EOF, unless the request is synchronous. (XXX why sync?) 1686 * XXXUBC Make these pages look "active" so the pagedaemon won't 1687 * XXXUBC bother us with them again. 1688 */ 1689 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1690 origoffset = ap->a_offlo; 1691 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1692 pg = uvm_pagelookup(&vp->v_uobj, off); 1693 KASSERT(pg != NULL); 1694 while (pg->flags & PG_BUSY) { 1695 pg->flags |= PG_WANTED; 1696 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1697 "lfsput2", 0); 1698 simple_lock(&vp->v_interlock); 1699 } 1700 uvm_lock_pageq(); 1701 uvm_pageactivate(pg); 1702 uvm_unlock_pageq(); 1703 } 1704 ap->a_offlo = blkeof; 1705 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1706 simple_unlock(&vp->v_interlock); 1707 return 0; 1708 } 1709 } 1710 1711 /* 1712 * Extend page range to start and end at block boundaries. 1713 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1714 */ 1715 origoffset = ap->a_offlo; 1716 origendoffset = ap->a_offhi; 1717 startoffset = origoffset & ~(fs->lfs_bmask); 1718 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1719 << fs->lfs_bshift; 1720 1721 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1722 endoffset = max_endoffset; 1723 origendoffset = endoffset; 1724 } else { 1725 origendoffset = round_page(ap->a_offhi); 1726 endoffset = round_page(blkroundup(fs, origendoffset)); 1727 } 1728 1729 KASSERT(startoffset > 0 || endoffset >= startoffset); 1730 if (startoffset == endoffset) { 1731 /* Nothing to do, why were we called? */ 1732 simple_unlock(&vp->v_interlock); 1733 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1734 PRId64 "\n", startoffset)); 1735 return 0; 1736 } 1737 1738 ap->a_offlo = startoffset; 1739 ap->a_offhi = endoffset; 1740 1741 if (!(ap->a_flags & PGO_CLEANIT)) 1742 return genfs_putpages(v); 1743 1744 /* 1745 * If there are more than one page per block, we don't want 1746 * to get caught locking them backwards; so set PGO_BUSYFAIL 1747 * to avoid deadlocks. 1748 */ 1749 ap->a_flags |= PGO_BUSYFAIL; 1750 1751 do { 1752 int r; 1753 1754 /* If no pages are dirty, we can just use genfs_putpages. */ 1755 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 1756 ap->a_flags, 1) != 0) 1757 break; 1758 1759 /* 1760 * Sometimes pages are dirtied between the time that 1761 * we check and the time we try to clean them. 1762 * Instruct lfs_gop_write to return EDEADLK in this case 1763 * so we can write them properly. 1764 */ 1765 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 1766 r = genfs_putpages(v); 1767 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 1768 if (r != EDEADLK) 1769 return r; 1770 1771 /* Start over. */ 1772 preempt(1); 1773 simple_lock(&vp->v_interlock); 1774 } while(1); 1775 1776 /* 1777 * Dirty and asked to clean. 1778 * 1779 * Pagedaemon can't actually write LFS pages; wake up 1780 * the writer to take care of that. The writer will 1781 * notice the pager inode queue and act on that. 1782 */ 1783 if (pagedaemon) { 1784 simple_lock(&fs->lfs_interlock); 1785 ++fs->lfs_pdflush; 1786 simple_unlock(&fs->lfs_interlock); 1787 wakeup(&lfs_writer_daemon); 1788 simple_unlock(&vp->v_interlock); 1789 return EWOULDBLOCK; 1790 } 1791 1792 /* 1793 * If this is a file created in a recent dirop, we can't flush its 1794 * inode until the dirop is complete. Drain dirops, then flush the 1795 * filesystem (taking care of any other pending dirops while we're 1796 * at it). 1797 */ 1798 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1799 (vp->v_flag & VDIROP)) { 1800 int locked; 1801 1802 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 1803 locked = VOP_ISLOCKED(vp) && /* XXX */ 1804 vp->v_lock.lk_lockholder == curproc->p_pid; 1805 simple_unlock(&vp->v_interlock); 1806 lfs_writer_enter(fs, "ppdirop"); 1807 if (locked) 1808 VOP_UNLOCK(vp, 0); 1809 1810 simple_lock(&fs->lfs_interlock); 1811 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1812 simple_unlock(&fs->lfs_interlock); 1813 1814 simple_lock(&vp->v_interlock); 1815 if (locked) { 1816 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 1817 simple_lock(&vp->v_interlock); 1818 } 1819 lfs_writer_leave(fs); 1820 1821 /* XXX the flush should have taken care of this one too! */ 1822 } 1823 1824 /* 1825 * This is it. We are going to write some pages. From here on 1826 * down it's all just mechanics. 1827 * 1828 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1829 */ 1830 ap->a_flags &= ~PGO_SYNCIO; 1831 1832 /* 1833 * If we've already got the seglock, flush the node and return. 1834 * The FIP has already been set up for us by lfs_writefile, 1835 * and FIP cleanup and lfs_updatemeta will also be done there, 1836 * unless genfs_putpages returns EDEADLK; then we must flush 1837 * what we have, and correct FIP and segment header accounting. 1838 */ 1839 1840 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 1841 if (!seglocked) { 1842 simple_unlock(&vp->v_interlock); 1843 /* 1844 * Take the seglock, because we are going to be writing pages. 1845 */ 1846 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 1847 if (error != 0) 1848 return error; 1849 simple_lock(&vp->v_interlock); 1850 } 1851 1852 /* 1853 * VOP_PUTPAGES should not be called while holding the seglock. 1854 * XXXUBC fix lfs_markv, or do this properly. 1855 */ 1856 #ifdef notyet 1857 KASSERT(fs->lfs_seglock == 1); 1858 #endif /* notyet */ 1859 1860 /* 1861 * We assume we're being called with sp->fip pointing at blank space. 1862 * Account for a new FIP in the segment header, and set sp->vp. 1863 * (This should duplicate the setup at the top of lfs_writefile().) 1864 */ 1865 sp = fs->lfs_sp; 1866 if (!seglocked) { 1867 if (sp->seg_bytes_left < fs->lfs_bsize || 1868 sp->sum_bytes_left < sizeof(struct finfo)) 1869 (void) lfs_writeseg(fs, fs->lfs_sp); 1870 1871 sp->sum_bytes_left -= FINFOSIZE; 1872 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1873 } 1874 KASSERT(sp->vp == NULL); 1875 sp->vp = vp; 1876 1877 if (!seglocked) { 1878 if (vp->v_flag & VDIROP) 1879 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 1880 } 1881 1882 sp->fip->fi_nblocks = 0; 1883 sp->fip->fi_ino = ip->i_number; 1884 sp->fip->fi_version = ip->i_gen; 1885 1886 /* 1887 * Loop through genfs_putpages until all pages are gathered. 1888 * genfs_putpages() drops the interlock, so reacquire it if necessary. 1889 * Whenever we lose the interlock we have to rerun check_dirty, as 1890 * well. 1891 */ 1892 again: 1893 check_dirty(fs, vp, startoffset, endoffset, blkeof, ap->a_flags, 0); 1894 1895 if ((error = genfs_putpages(v)) == EDEADLK) { 1896 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 1897 " EDEADLK [2] ino %d off %x (seg %d)\n", 1898 ip->i_number, fs->lfs_offset, 1899 dtosn(fs, fs->lfs_offset))); 1900 /* If nothing to write, short-circuit */ 1901 if (sp->cbpp - sp->bpp > 1) { 1902 /* Write gathered pages */ 1903 lfs_updatemeta(sp); 1904 (void) lfs_writeseg(fs, sp); 1905 1906 /* 1907 * Reinitialize brand new FIP and add us to it. 1908 * (This should duplicate the fixup in 1909 * lfs_gatherpages().) 1910 */ 1911 KASSERT(sp->vp == vp); 1912 sp->fip->fi_version = ip->i_gen; 1913 sp->fip->fi_ino = ip->i_number; 1914 /* Add us to the new segment summary. */ 1915 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1916 sp->sum_bytes_left -= FINFOSIZE; 1917 } 1918 1919 /* Give the write a chance to complete */ 1920 preempt(1); 1921 1922 /* We've lost the interlock. Start over. */ 1923 simple_lock(&vp->v_interlock); 1924 goto again; 1925 } 1926 1927 KASSERT(sp->vp == vp); 1928 if (!seglocked) { 1929 sp->vp = NULL; /* XXX lfs_gather below will set this */ 1930 1931 /* Write indirect blocks as well */ 1932 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 1933 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 1934 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 1935 1936 KASSERT(sp->vp == NULL); 1937 sp->vp = vp; 1938 } 1939 1940 /* 1941 * Blocks are now gathered into a segment waiting to be written. 1942 * All that's left to do is update metadata, and write them. 1943 */ 1944 lfs_updatemeta(sp); 1945 KASSERT(sp->vp == vp); 1946 sp->vp = NULL; 1947 1948 if (seglocked) { 1949 /* we're called by lfs_writefile. */ 1950 return error; 1951 } 1952 1953 /* 1954 * Clean up FIP, since we're done writing this file. 1955 * This should duplicate cleanup at the end of lfs_writefile(). 1956 */ 1957 if (sp->fip->fi_nblocks != 0) { 1958 sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE + 1959 sizeof(int32_t) * sp->fip->fi_nblocks); 1960 sp->start_lbp = &sp->fip->fi_blocks[0]; 1961 } else { 1962 sp->sum_bytes_left += FINFOSIZE; 1963 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 1964 } 1965 lfs_writeseg(fs, fs->lfs_sp); 1966 1967 /* 1968 * XXX - with the malloc/copy writeseg, the pages are freed by now 1969 * even if we don't wait (e.g. if we hold a nested lock). This 1970 * will not be true if we stop using malloc/copy. 1971 */ 1972 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 1973 lfs_segunlock(fs); 1974 1975 /* 1976 * Wait for v_numoutput to drop to zero. The seglock should 1977 * take care of this, but there is a slight possibility that 1978 * aiodoned might not have got around to our buffers yet. 1979 */ 1980 if (sync) { 1981 s = splbio(); 1982 simple_lock(&global_v_numoutput_slock); 1983 while (vp->v_numoutput > 0) { 1984 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 1985 " num %d\n", ip->i_number, vp->v_numoutput)); 1986 vp->v_flag |= VBWAIT; 1987 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 1988 &global_v_numoutput_slock); 1989 } 1990 simple_unlock(&global_v_numoutput_slock); 1991 splx(s); 1992 } 1993 return error; 1994 } 1995 1996 /* 1997 * Return the last logical file offset that should be written for this file 1998 * if we're doing a write that ends at "size". If writing, we need to know 1999 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2000 * to know about entire blocks. 2001 */ 2002 void 2003 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2004 { 2005 struct inode *ip = VTOI(vp); 2006 struct lfs *fs = ip->i_lfs; 2007 daddr_t olbn, nlbn; 2008 2009 KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)); 2010 KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)) 2011 != (GOP_SIZE_READ | GOP_SIZE_WRITE)); 2012 2013 olbn = lblkno(fs, ip->i_size); 2014 nlbn = lblkno(fs, size); 2015 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2016 *eobp = fragroundup(fs, size); 2017 } else { 2018 *eobp = blkroundup(fs, size); 2019 } 2020 } 2021 2022 #ifdef DEBUG 2023 void lfs_dump_vop(void *); 2024 2025 void 2026 lfs_dump_vop(void *v) 2027 { 2028 struct vop_putpages_args /* { 2029 struct vnode *a_vp; 2030 voff_t a_offlo; 2031 voff_t a_offhi; 2032 int a_flags; 2033 } */ *ap = v; 2034 2035 #ifdef DDB 2036 vfs_vnode_print(ap->a_vp, 0, printf); 2037 #endif 2038 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2039 } 2040 #endif 2041 2042 int 2043 lfs_mmap(void *v) 2044 { 2045 struct vop_mmap_args /* { 2046 const struct vnodeop_desc *a_desc; 2047 struct vnode *a_vp; 2048 int a_fflags; 2049 struct ucred *a_cred; 2050 struct proc *a_p; 2051 } */ *ap = v; 2052 2053 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2054 return EOPNOTSUPP; 2055 return ufs_mmap(v); 2056 } 2057
Indexes created Mon Sep 29 21:09:56 GMT 2025