lfs_vnops.c revision 1.157.10.5
1 /* $NetBSD: lfs_vnops.c,v 1.157.10.5 2006/05/11 23:32:03 elad 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 1.157.10.5 2006/05/11 23:32:03 elad 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 #include <sys/kauth.h> 86 87 #include <miscfs/fifofs/fifo.h> 88 #include <miscfs/genfs/genfs.h> 89 #include <miscfs/specfs/specdev.h> 90 91 #include <ufs/ufs/inode.h> 92 #include <ufs/ufs/dir.h> 93 #include <ufs/ufs/ufsmount.h> 94 #include <ufs/ufs/ufs_extern.h> 95 96 #include <uvm/uvm.h> 97 #include <uvm/uvm_pmap.h> 98 #include <uvm/uvm_stat.h> 99 #include <uvm/uvm_pager.h> 100 101 #include <ufs/lfs/lfs.h> 102 #include <ufs/lfs/lfs_extern.h> 103 104 extern pid_t lfs_writer_daemon; 105 106 /* Global vfs data structures for lfs. */ 107 int (**lfs_vnodeop_p)(void *); 108 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 109 { &vop_default_desc, vn_default_error }, 110 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 111 { &vop_create_desc, lfs_create }, /* create */ 112 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 113 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 114 { &vop_open_desc, ufs_open }, /* open */ 115 { &vop_close_desc, lfs_close }, /* close */ 116 { &vop_access_desc, ufs_access }, /* access */ 117 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 118 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 119 { &vop_read_desc, lfs_read }, /* read */ 120 { &vop_write_desc, lfs_write }, /* write */ 121 { &vop_lease_desc, ufs_lease_check }, /* lease */ 122 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 123 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 124 { &vop_poll_desc, ufs_poll }, /* poll */ 125 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 126 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 127 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 128 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 129 { &vop_seek_desc, ufs_seek }, /* seek */ 130 { &vop_remove_desc, lfs_remove }, /* remove */ 131 { &vop_link_desc, lfs_link }, /* link */ 132 { &vop_rename_desc, lfs_rename }, /* rename */ 133 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 134 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 135 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 136 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 137 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 138 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 139 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 140 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 141 { &vop_lock_desc, ufs_lock }, /* lock */ 142 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 143 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 144 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 145 { &vop_print_desc, ufs_print }, /* print */ 146 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 147 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 148 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 149 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 150 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 151 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 152 { NULL, NULL } 153 }; 154 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 155 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 156 157 int (**lfs_specop_p)(void *); 158 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 159 { &vop_default_desc, vn_default_error }, 160 { &vop_lookup_desc, spec_lookup }, /* lookup */ 161 { &vop_create_desc, spec_create }, /* create */ 162 { &vop_mknod_desc, spec_mknod }, /* mknod */ 163 { &vop_open_desc, spec_open }, /* open */ 164 { &vop_close_desc, lfsspec_close }, /* close */ 165 { &vop_access_desc, ufs_access }, /* access */ 166 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 167 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 168 { &vop_read_desc, ufsspec_read }, /* read */ 169 { &vop_write_desc, ufsspec_write }, /* write */ 170 { &vop_lease_desc, spec_lease_check }, /* lease */ 171 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 172 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 173 { &vop_poll_desc, spec_poll }, /* poll */ 174 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 175 { &vop_revoke_desc, spec_revoke }, /* revoke */ 176 { &vop_mmap_desc, spec_mmap }, /* mmap */ 177 { &vop_fsync_desc, spec_fsync }, /* fsync */ 178 { &vop_seek_desc, spec_seek }, /* seek */ 179 { &vop_remove_desc, spec_remove }, /* remove */ 180 { &vop_link_desc, spec_link }, /* link */ 181 { &vop_rename_desc, spec_rename }, /* rename */ 182 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 183 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 184 { &vop_symlink_desc, spec_symlink }, /* symlink */ 185 { &vop_readdir_desc, spec_readdir }, /* readdir */ 186 { &vop_readlink_desc, spec_readlink }, /* readlink */ 187 { &vop_abortop_desc, spec_abortop }, /* abortop */ 188 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 189 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 190 { &vop_lock_desc, ufs_lock }, /* lock */ 191 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 192 { &vop_bmap_desc, spec_bmap }, /* bmap */ 193 { &vop_strategy_desc, spec_strategy }, /* strategy */ 194 { &vop_print_desc, ufs_print }, /* print */ 195 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 196 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 197 { &vop_advlock_desc, spec_advlock }, /* advlock */ 198 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 199 { &vop_getpages_desc, spec_getpages }, /* getpages */ 200 { &vop_putpages_desc, spec_putpages }, /* putpages */ 201 { NULL, NULL } 202 }; 203 const struct vnodeopv_desc lfs_specop_opv_desc = 204 { &lfs_specop_p, lfs_specop_entries }; 205 206 int (**lfs_fifoop_p)(void *); 207 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 208 { &vop_default_desc, vn_default_error }, 209 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 210 { &vop_create_desc, fifo_create }, /* create */ 211 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 212 { &vop_open_desc, fifo_open }, /* open */ 213 { &vop_close_desc, lfsfifo_close }, /* close */ 214 { &vop_access_desc, ufs_access }, /* access */ 215 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 216 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 217 { &vop_read_desc, ufsfifo_read }, /* read */ 218 { &vop_write_desc, ufsfifo_write }, /* write */ 219 { &vop_lease_desc, fifo_lease_check }, /* lease */ 220 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 221 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 222 { &vop_poll_desc, fifo_poll }, /* poll */ 223 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 224 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 225 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 226 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 227 { &vop_seek_desc, fifo_seek }, /* seek */ 228 { &vop_remove_desc, fifo_remove }, /* remove */ 229 { &vop_link_desc, fifo_link }, /* link */ 230 { &vop_rename_desc, fifo_rename }, /* rename */ 231 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 232 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 233 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 234 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 235 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 236 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 237 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 238 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 239 { &vop_lock_desc, ufs_lock }, /* lock */ 240 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 241 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 242 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 243 { &vop_print_desc, ufs_print }, /* print */ 244 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 245 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 246 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 247 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 248 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 249 { NULL, NULL } 250 }; 251 const struct vnodeopv_desc lfs_fifoop_opv_desc = 252 { &lfs_fifoop_p, lfs_fifoop_entries }; 253 254 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int); 255 256 #define LFS_READWRITE 257 #include <ufs/ufs/ufs_readwrite.c> 258 #undef LFS_READWRITE 259 260 /* 261 * Synch an open file. 262 */ 263 /* ARGSUSED */ 264 int 265 lfs_fsync(void *v) 266 { 267 struct vop_fsync_args /* { 268 struct vnode *a_vp; 269 kauth_cred_t a_cred; 270 int a_flags; 271 off_t offlo; 272 off_t offhi; 273 struct lwp *a_l; 274 } */ *ap = v; 275 struct vnode *vp = ap->a_vp; 276 int error, wait; 277 278 /* If we're mounted read-only, don't try to sync. */ 279 if (VTOI(vp)->i_lfs->lfs_ronly) 280 return 0; 281 282 /* 283 * Trickle sync checks for need to do a checkpoint after possible 284 * activity from the pagedaemon. 285 */ 286 if (ap->a_flags & FSYNC_LAZY) { 287 simple_lock(&lfs_subsys_lock); 288 wakeup(&lfs_writer_daemon); 289 simple_unlock(&lfs_subsys_lock); 290 return 0; 291 } 292 293 wait = (ap->a_flags & FSYNC_WAIT); 294 simple_lock(&vp->v_interlock); 295 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 296 round_page(ap->a_offhi), 297 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 298 if (error) 299 return error; 300 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 301 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 302 int l = 0; 303 error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE, 304 ap->a_l->l_proc->p_cred, ap->a_l); 305 } 306 if (wait && !VPISEMPTY(vp)) 307 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 308 309 return error; 310 } 311 312 /* 313 * Take IN_ADIROP off, then call ufs_inactive. 314 */ 315 int 316 lfs_inactive(void *v) 317 { 318 struct vop_inactive_args /* { 319 struct vnode *a_vp; 320 struct lwp *a_l; 321 } */ *ap = v; 322 323 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 324 325 lfs_unmark_vnode(ap->a_vp); 326 327 /* 328 * The Ifile is only ever inactivated on unmount. 329 * Streamline this process by not giving it more dirty blocks. 330 */ 331 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 332 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 333 VOP_UNLOCK(ap->a_vp, 0); 334 return 0; 335 } 336 337 return ufs_inactive(v); 338 } 339 340 /* 341 * These macros are used to bracket UFS directory ops, so that we can 342 * identify all the pages touched during directory ops which need to 343 * be ordered and flushed atomically, so that they may be recovered. 344 * 345 * Because we have to mark nodes VDIROP in order to prevent 346 * the cache from reclaiming them while a dirop is in progress, we must 347 * also manage the number of nodes so marked (otherwise we can run out). 348 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 349 * is decremented during segment write, when VDIROP is taken off. 350 */ 351 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 352 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 353 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 354 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 355 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 356 static int lfs_set_dirop(struct vnode *, struct vnode *); 357 358 static int 359 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 360 { 361 struct lfs *fs; 362 int error; 363 364 KASSERT(VOP_ISLOCKED(dvp)); 365 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 366 367 fs = VTOI(dvp)->i_lfs; 368 369 ASSERT_NO_SEGLOCK(fs); 370 /* 371 * LFS_NRESERVE calculates direct and indirect blocks as well 372 * as an inode block; an overestimate in most cases. 373 */ 374 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 375 return (error); 376 377 restart: 378 simple_lock(&fs->lfs_interlock); 379 if (fs->lfs_dirops == 0) { 380 simple_unlock(&fs->lfs_interlock); 381 lfs_check(dvp, LFS_UNUSED_LBN, 0); 382 simple_lock(&fs->lfs_interlock); 383 } 384 while (fs->lfs_writer) 385 ltsleep(&fs->lfs_dirops, (PRIBIO + 1), "lfs_sdirop", 0, 386 &fs->lfs_interlock); 387 simple_lock(&lfs_subsys_lock); 388 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 389 wakeup(&lfs_writer_daemon); 390 simple_unlock(&lfs_subsys_lock); 391 simple_unlock(&fs->lfs_interlock); 392 preempt(1); 393 goto restart; 394 } 395 396 if (lfs_dirvcount > LFS_MAX_DIROP) { 397 simple_unlock(&fs->lfs_interlock); 398 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 399 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 400 if ((error = ltsleep(&lfs_dirvcount, 401 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 402 &lfs_subsys_lock)) != 0) { 403 goto unreserve; 404 } 405 goto restart; 406 } 407 simple_unlock(&lfs_subsys_lock); 408 409 ++fs->lfs_dirops; 410 fs->lfs_doifile = 1; 411 simple_unlock(&fs->lfs_interlock); 412 413 /* Hold a reference so SET_ENDOP will be happy */ 414 vref(dvp); 415 if (vp) { 416 vref(vp); 417 MARK_VNODE(vp); 418 } 419 420 MARK_VNODE(dvp); 421 return 0; 422 423 unreserve: 424 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 425 return error; 426 } 427 428 /* 429 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 430 * in getnewvnode(), if we have a stacked filesystem mounted on top 431 * of us. 432 * 433 * NB: this means we have to clear the new vnodes on error. Fortunately 434 * SET_ENDOP is there to do that for us. 435 */ 436 static int 437 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 438 { 439 int error; 440 struct lfs *fs; 441 442 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 443 ASSERT_NO_SEGLOCK(fs); 444 if (fs->lfs_ronly) 445 return EROFS; 446 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 447 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 448 dvp, error)); 449 return error; 450 } 451 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 452 if (vpp) { 453 ungetnewvnode(*vpp); 454 *vpp = NULL; 455 } 456 return error; 457 } 458 return 0; 459 } 460 461 #define SET_ENDOP_BASE(fs, dvp, str) \ 462 do { \ 463 simple_lock(&(fs)->lfs_interlock); \ 464 --(fs)->lfs_dirops; \ 465 if (!(fs)->lfs_dirops) { \ 466 if ((fs)->lfs_nadirop) { \ 467 panic("SET_ENDOP: %s: no dirops but " \ 468 " nadirop=%d", (str), \ 469 (fs)->lfs_nadirop); \ 470 } \ 471 wakeup(&(fs)->lfs_writer); \ 472 simple_unlock(&(fs)->lfs_interlock); \ 473 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 474 } else \ 475 simple_unlock(&(fs)->lfs_interlock); \ 476 } while(0) 477 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 478 do { \ 479 UNMARK_VNODE(dvp); \ 480 if (nvpp && *nvpp) \ 481 UNMARK_VNODE(*nvpp); \ 482 /* Check for error return to stem vnode leakage */ \ 483 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 484 ungetnewvnode(*(nvpp)); \ 485 SET_ENDOP_BASE((fs), (dvp), (str)); \ 486 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 487 vrele(dvp); \ 488 } while(0) 489 #define SET_ENDOP_CREATE_AP(ap, str) \ 490 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 491 (ap)->a_vpp, (str)) 492 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 493 do { \ 494 UNMARK_VNODE(dvp); \ 495 if (ovp) \ 496 UNMARK_VNODE(ovp); \ 497 SET_ENDOP_BASE((fs), (dvp), (str)); \ 498 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 499 vrele(dvp); \ 500 if (ovp) \ 501 vrele(ovp); \ 502 } while(0) 503 504 void 505 lfs_mark_vnode(struct vnode *vp) 506 { 507 struct inode *ip = VTOI(vp); 508 struct lfs *fs = ip->i_lfs; 509 510 simple_lock(&fs->lfs_interlock); 511 if (!(ip->i_flag & IN_ADIROP)) { 512 if (!(vp->v_flag & VDIROP)) { 513 (void)lfs_vref(vp); 514 simple_lock(&lfs_subsys_lock); 515 ++lfs_dirvcount; 516 ++fs->lfs_dirvcount; 517 simple_unlock(&lfs_subsys_lock); 518 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 519 vp->v_flag |= VDIROP; 520 } 521 ++fs->lfs_nadirop; 522 ip->i_flag |= IN_ADIROP; 523 } else 524 KASSERT(vp->v_flag & VDIROP); 525 simple_unlock(&fs->lfs_interlock); 526 } 527 528 void 529 lfs_unmark_vnode(struct vnode *vp) 530 { 531 struct inode *ip = VTOI(vp); 532 533 if (ip && (ip->i_flag & IN_ADIROP)) { 534 KASSERT(vp->v_flag & VDIROP); 535 simple_lock(&ip->i_lfs->lfs_interlock); 536 --ip->i_lfs->lfs_nadirop; 537 simple_unlock(&ip->i_lfs->lfs_interlock); 538 ip->i_flag &= ~IN_ADIROP; 539 } 540 } 541 542 int 543 lfs_symlink(void *v) 544 { 545 struct vop_symlink_args /* { 546 struct vnode *a_dvp; 547 struct vnode **a_vpp; 548 struct componentname *a_cnp; 549 struct vattr *a_vap; 550 char *a_target; 551 } */ *ap = v; 552 int error; 553 554 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 555 vput(ap->a_dvp); 556 return error; 557 } 558 error = ufs_symlink(ap); 559 SET_ENDOP_CREATE_AP(ap, "symlink"); 560 return (error); 561 } 562 563 int 564 lfs_mknod(void *v) 565 { 566 struct vop_mknod_args /* { 567 struct vnode *a_dvp; 568 struct vnode **a_vpp; 569 struct componentname *a_cnp; 570 struct vattr *a_vap; 571 } */ *ap = v; 572 struct vattr *vap = ap->a_vap; 573 struct vnode **vpp = ap->a_vpp; 574 struct inode *ip; 575 int error; 576 struct mount *mp; 577 ino_t ino; 578 579 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 580 vput(ap->a_dvp); 581 return error; 582 } 583 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 584 ap->a_dvp, vpp, ap->a_cnp); 585 586 /* Either way we're done with the dirop at this point */ 587 SET_ENDOP_CREATE_AP(ap, "mknod"); 588 589 if (error) 590 return (error); 591 592 ip = VTOI(*vpp); 593 mp = (*vpp)->v_mount; 594 ino = ip->i_number; 595 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 596 if (vap->va_rdev != VNOVAL) { 597 /* 598 * Want to be able to use this to make badblock 599 * inodes, so don't truncate the dev number. 600 */ 601 #if 0 602 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 603 UFS_MPNEEDSWAP((*vpp)->v_mount)); 604 #else 605 ip->i_ffs1_rdev = vap->va_rdev; 606 #endif 607 } 608 609 /* 610 * Call fsync to write the vnode so that we don't have to deal with 611 * flushing it when it's marked VDIROP|VXLOCK. 612 * 613 * XXX KS - If we can't flush we also can't call vgone(), so must 614 * return. But, that leaves this vnode in limbo, also not good. 615 * Can this ever happen (barring hardware failure)? 616 */ 617 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 618 curlwp)) != 0) { 619 panic("lfs_mknod: couldn't fsync (ino %llu)", 620 (unsigned long long)ino); 621 /* return (error); */ 622 } 623 /* 624 * Remove vnode so that it will be reloaded by VFS_VGET and 625 * checked to see if it is an alias of an existing entry in 626 * the inode cache. 627 */ 628 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 629 630 VOP_UNLOCK(*vpp, 0); 631 lfs_vunref(*vpp); 632 (*vpp)->v_type = VNON; 633 vgone(*vpp); 634 error = VFS_VGET(mp, ino, vpp); 635 636 if (error != 0) { 637 *vpp = NULL; 638 return (error); 639 } 640 return (0); 641 } 642 643 int 644 lfs_create(void *v) 645 { 646 struct vop_create_args /* { 647 struct vnode *a_dvp; 648 struct vnode **a_vpp; 649 struct componentname *a_cnp; 650 struct vattr *a_vap; 651 } */ *ap = v; 652 int error; 653 654 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 655 vput(ap->a_dvp); 656 return error; 657 } 658 error = ufs_create(ap); 659 SET_ENDOP_CREATE_AP(ap, "create"); 660 return (error); 661 } 662 663 int 664 lfs_mkdir(void *v) 665 { 666 struct vop_mkdir_args /* { 667 struct vnode *a_dvp; 668 struct vnode **a_vpp; 669 struct componentname *a_cnp; 670 struct vattr *a_vap; 671 } */ *ap = v; 672 int error; 673 674 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 675 vput(ap->a_dvp); 676 return error; 677 } 678 error = ufs_mkdir(ap); 679 SET_ENDOP_CREATE_AP(ap, "mkdir"); 680 return (error); 681 } 682 683 int 684 lfs_remove(void *v) 685 { 686 struct vop_remove_args /* { 687 struct vnode *a_dvp; 688 struct vnode *a_vp; 689 struct componentname *a_cnp; 690 } */ *ap = v; 691 struct vnode *dvp, *vp; 692 int error; 693 694 dvp = ap->a_dvp; 695 vp = ap->a_vp; 696 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 697 if (dvp == vp) 698 vrele(vp); 699 else 700 vput(vp); 701 vput(dvp); 702 return error; 703 } 704 error = ufs_remove(ap); 705 SET_ENDOP_REMOVE(VTOI(dvp)->i_lfs, dvp, ap->a_vp, "remove"); 706 return (error); 707 } 708 709 int 710 lfs_rmdir(void *v) 711 { 712 struct vop_rmdir_args /* { 713 struct vnodeop_desc *a_desc; 714 struct vnode *a_dvp; 715 struct vnode *a_vp; 716 struct componentname *a_cnp; 717 } */ *ap = v; 718 struct vnode *vp; 719 int error; 720 721 vp = ap->a_vp; 722 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 723 vrele(ap->a_dvp); 724 if (ap->a_vp != ap->a_dvp) 725 VOP_UNLOCK(ap->a_dvp, 0); 726 vput(vp); 727 return error; 728 } 729 error = ufs_rmdir(ap); 730 SET_ENDOP_REMOVE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 731 return (error); 732 } 733 734 int 735 lfs_link(void *v) 736 { 737 struct vop_link_args /* { 738 struct vnode *a_dvp; 739 struct vnode *a_vp; 740 struct componentname *a_cnp; 741 } */ *ap = v; 742 int error; 743 struct vnode **vpp = NULL; 744 745 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 746 vput(ap->a_dvp); 747 return error; 748 } 749 error = ufs_link(ap); 750 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 751 return (error); 752 } 753 754 int 755 lfs_rename(void *v) 756 { 757 struct vop_rename_args /* { 758 struct vnode *a_fdvp; 759 struct vnode *a_fvp; 760 struct componentname *a_fcnp; 761 struct vnode *a_tdvp; 762 struct vnode *a_tvp; 763 struct componentname *a_tcnp; 764 } */ *ap = v; 765 struct vnode *tvp, *fvp, *tdvp, *fdvp; 766 struct componentname *tcnp, *fcnp; 767 int error; 768 struct lfs *fs; 769 770 fs = VTOI(ap->a_fdvp)->i_lfs; 771 tvp = ap->a_tvp; 772 tdvp = ap->a_tdvp; 773 tcnp = ap->a_tcnp; 774 fvp = ap->a_fvp; 775 fdvp = ap->a_fdvp; 776 fcnp = ap->a_fcnp; 777 778 /* 779 * Check for cross-device rename. 780 * If it is, we don't want to set dirops, just error out. 781 * (In particular note that MARK_VNODE(tdvp) will DTWT on 782 * a cross-device rename.) 783 * 784 * Copied from ufs_rename. 785 */ 786 if ((fvp->v_mount != tdvp->v_mount) || 787 (tvp && (fvp->v_mount != tvp->v_mount))) { 788 error = EXDEV; 789 goto errout; 790 } 791 792 /* 793 * Check to make sure we're not renaming a vnode onto itself 794 * (deleting a hard link by renaming one name onto another); 795 * if we are we can't recursively call VOP_REMOVE since that 796 * would leave us with an unaccounted-for number of live dirops. 797 * 798 * Inline the relevant section of ufs_rename here, *before* 799 * calling SET_DIROP_REMOVE. 800 */ 801 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 802 (VTOI(tdvp)->i_flags & APPEND))) { 803 error = EPERM; 804 goto errout; 805 } 806 if (fvp == tvp) { 807 if (fvp->v_type == VDIR) { 808 error = EINVAL; 809 goto errout; 810 } 811 812 /* Release destination completely. */ 813 VOP_ABORTOP(tdvp, tcnp); 814 vput(tdvp); 815 vput(tvp); 816 817 /* Delete source. */ 818 vrele(fvp); 819 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 820 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 821 fcnp->cn_nameiop = DELETE; 822 if ((error = relookup(fdvp, &fvp, fcnp))){ 823 /* relookup blew away fdvp */ 824 return (error); 825 } 826 return (VOP_REMOVE(fdvp, fvp, fcnp)); 827 } 828 829 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 830 goto errout; 831 MARK_VNODE(fdvp); 832 MARK_VNODE(fvp); 833 834 error = ufs_rename(ap); 835 UNMARK_VNODE(fdvp); 836 UNMARK_VNODE(fvp); 837 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 838 return (error); 839 840 errout: 841 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 842 if (tdvp == tvp) 843 vrele(tdvp); 844 else 845 vput(tdvp); 846 if (tvp) 847 vput(tvp); 848 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 849 vrele(fdvp); 850 vrele(fvp); 851 return (error); 852 } 853 854 /* XXX hack to avoid calling ITIMES in getattr */ 855 int 856 lfs_getattr(void *v) 857 { 858 struct vop_getattr_args /* { 859 struct vnode *a_vp; 860 struct vattr *a_vap; 861 kauth_cred_t a_cred; 862 struct lwp *a_l; 863 } */ *ap = v; 864 struct vnode *vp = ap->a_vp; 865 struct inode *ip = VTOI(vp); 866 struct vattr *vap = ap->a_vap; 867 struct lfs *fs = ip->i_lfs; 868 /* 869 * Copy from inode table 870 */ 871 vap->va_fsid = ip->i_dev; 872 vap->va_fileid = ip->i_number; 873 vap->va_mode = ip->i_mode & ~IFMT; 874 vap->va_nlink = ip->i_nlink; 875 vap->va_uid = ip->i_uid; 876 vap->va_gid = ip->i_gid; 877 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 878 vap->va_size = vp->v_size; 879 vap->va_atime.tv_sec = ip->i_ffs1_atime; 880 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 881 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 882 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 883 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 884 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 885 vap->va_flags = ip->i_flags; 886 vap->va_gen = ip->i_gen; 887 /* this doesn't belong here */ 888 if (vp->v_type == VBLK) 889 vap->va_blocksize = BLKDEV_IOSIZE; 890 else if (vp->v_type == VCHR) 891 vap->va_blocksize = MAXBSIZE; 892 else 893 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 894 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 895 vap->va_type = vp->v_type; 896 vap->va_filerev = ip->i_modrev; 897 return (0); 898 } 899 900 /* 901 * Check to make sure the inode blocks won't choke the buffer 902 * cache, then call ufs_setattr as usual. 903 */ 904 int 905 lfs_setattr(void *v) 906 { 907 struct vop_setattr_args /* { 908 struct vnode *a_vp; 909 struct vattr *a_vap; 910 kauth_cred_t a_cred; 911 struct lwp *a_l; 912 } */ *ap = v; 913 struct vnode *vp = ap->a_vp; 914 915 lfs_check(vp, LFS_UNUSED_LBN, 0); 916 return ufs_setattr(v); 917 } 918 919 /* 920 * Close called 921 * 922 * XXX -- we were using ufs_close, but since it updates the 923 * times on the inode, we might need to bump the uinodes 924 * count. 925 */ 926 /* ARGSUSED */ 927 int 928 lfs_close(void *v) 929 { 930 struct vop_close_args /* { 931 struct vnode *a_vp; 932 int a_fflag; 933 kauth_cred_t a_cred; 934 struct lwp *a_l; 935 } */ *ap = v; 936 struct vnode *vp = ap->a_vp; 937 struct inode *ip = VTOI(vp); 938 939 if (vp == ip->i_lfs->lfs_ivnode && 940 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 941 return 0; 942 943 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 944 LFS_ITIMES(ip, NULL, NULL, NULL); 945 } 946 return (0); 947 } 948 949 /* 950 * Close wrapper for special devices. 951 * 952 * Update the times on the inode then do device close. 953 */ 954 int 955 lfsspec_close(void *v) 956 { 957 struct vop_close_args /* { 958 struct vnode *a_vp; 959 int a_fflag; 960 kauth_cred_t a_cred; 961 struct lwp *a_l; 962 } */ *ap = v; 963 struct vnode *vp; 964 struct inode *ip; 965 966 vp = ap->a_vp; 967 ip = VTOI(vp); 968 if (vp->v_usecount > 1) { 969 LFS_ITIMES(ip, NULL, NULL, NULL); 970 } 971 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 972 } 973 974 /* 975 * Close wrapper for fifo's. 976 * 977 * Update the times on the inode then do device close. 978 */ 979 int 980 lfsfifo_close(void *v) 981 { 982 struct vop_close_args /* { 983 struct vnode *a_vp; 984 int a_fflag; 985 kauth_cred_ a_cred; 986 struct lwp *a_l; 987 } */ *ap = v; 988 struct vnode *vp; 989 struct inode *ip; 990 991 vp = ap->a_vp; 992 ip = VTOI(vp); 993 if (ap->a_vp->v_usecount > 1) { 994 LFS_ITIMES(ip, NULL, NULL, NULL); 995 } 996 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 997 } 998 999 /* 1000 * Reclaim an inode so that it can be used for other purposes. 1001 */ 1002 1003 int 1004 lfs_reclaim(void *v) 1005 { 1006 struct vop_reclaim_args /* { 1007 struct vnode *a_vp; 1008 struct lwp *a_l; 1009 } */ *ap = v; 1010 struct vnode *vp = ap->a_vp; 1011 struct inode *ip = VTOI(vp); 1012 int error; 1013 1014 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1015 1016 LFS_CLR_UINO(ip, IN_ALLMOD); 1017 if ((error = ufs_reclaim(vp, ap->a_l))) 1018 return (error); 1019 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1020 lfs_deregister_all(vp); 1021 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1022 ip->inode_ext.lfs = NULL; 1023 pool_put(&lfs_inode_pool, vp->v_data); 1024 vp->v_data = NULL; 1025 return (0); 1026 } 1027 1028 /* 1029 * Read a block from a storage device. 1030 * In order to avoid reading blocks that are in the process of being 1031 * written by the cleaner---and hence are not mutexed by the normal 1032 * buffer cache / page cache mechanisms---check for collisions before 1033 * reading. 1034 * 1035 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1036 * the active cleaner test. 1037 * 1038 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1039 */ 1040 int 1041 lfs_strategy(void *v) 1042 { 1043 struct vop_strategy_args /* { 1044 struct vnode *a_vp; 1045 struct buf *a_bp; 1046 } */ *ap = v; 1047 struct buf *bp; 1048 struct lfs *fs; 1049 struct vnode *vp; 1050 struct inode *ip; 1051 daddr_t tbn; 1052 int i, sn, error, slept; 1053 1054 bp = ap->a_bp; 1055 vp = ap->a_vp; 1056 ip = VTOI(vp); 1057 fs = ip->i_lfs; 1058 1059 /* lfs uses its strategy routine only for read */ 1060 KASSERT(bp->b_flags & B_READ); 1061 1062 if (vp->v_type == VBLK || vp->v_type == VCHR) 1063 panic("lfs_strategy: spec"); 1064 KASSERT(bp->b_bcount != 0); 1065 if (bp->b_blkno == bp->b_lblkno) { 1066 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1067 NULL); 1068 if (error) { 1069 bp->b_error = error; 1070 bp->b_flags |= B_ERROR; 1071 biodone(bp); 1072 return (error); 1073 } 1074 if ((long)bp->b_blkno == -1) /* no valid data */ 1075 clrbuf(bp); 1076 } 1077 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1078 biodone(bp); 1079 return (0); 1080 } 1081 1082 slept = 1; 1083 simple_lock(&fs->lfs_interlock); 1084 while (slept && fs->lfs_seglock) { 1085 simple_unlock(&fs->lfs_interlock); 1086 /* 1087 * Look through list of intervals. 1088 * There will only be intervals to look through 1089 * if the cleaner holds the seglock. 1090 * Since the cleaner is synchronous, we can trust 1091 * the list of intervals to be current. 1092 */ 1093 tbn = dbtofsb(fs, bp->b_blkno); 1094 sn = dtosn(fs, tbn); 1095 slept = 0; 1096 for (i = 0; i < fs->lfs_cleanind; i++) { 1097 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1098 tbn >= fs->lfs_cleanint[i]) { 1099 DLOG((DLOG_CLEAN, 1100 "lfs_strategy: ino %d lbn %" PRId64 1101 " ind %d sn %d fsb %" PRIx32 1102 " given sn %d fsb %" PRIx64 "\n", 1103 ip->i_number, bp->b_lblkno, i, 1104 dtosn(fs, fs->lfs_cleanint[i]), 1105 fs->lfs_cleanint[i], sn, tbn)); 1106 DLOG((DLOG_CLEAN, 1107 "lfs_strategy: sleeping on ino %d lbn %" 1108 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1109 simple_lock(&fs->lfs_interlock); 1110 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1111 /* Cleaner can't wait for itself */ 1112 ltsleep(&fs->lfs_iocount, 1113 (PRIBIO + 1) | PNORELOCK, 1114 "clean2", 0, 1115 &fs->lfs_interlock); 1116 slept = 1; 1117 break; 1118 } else if (fs->lfs_seglock) { 1119 ltsleep(&fs->lfs_seglock, 1120 (PRIBIO + 1) | PNORELOCK, 1121 "clean1", 0, 1122 &fs->lfs_interlock); 1123 slept = 1; 1124 break; 1125 } 1126 simple_unlock(&fs->lfs_interlock); 1127 } 1128 } 1129 simple_lock(&fs->lfs_interlock); 1130 } 1131 simple_unlock(&fs->lfs_interlock); 1132 1133 vp = ip->i_devvp; 1134 VOP_STRATEGY(vp, bp); 1135 return (0); 1136 } 1137 1138 void 1139 lfs_flush_dirops(struct lfs *fs) 1140 { 1141 struct inode *ip, *nip; 1142 struct vnode *vp; 1143 extern int lfs_dostats; 1144 struct segment *sp; 1145 int waslocked; 1146 1147 ASSERT_MAYBE_SEGLOCK(fs); 1148 KASSERT(fs->lfs_nadirop == 0); 1149 1150 if (fs->lfs_ronly) 1151 return; 1152 1153 simple_lock(&fs->lfs_interlock); 1154 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1155 simple_unlock(&fs->lfs_interlock); 1156 return; 1157 } else 1158 simple_unlock(&fs->lfs_interlock); 1159 1160 if (lfs_dostats) 1161 ++lfs_stats.flush_invoked; 1162 1163 /* 1164 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1165 * Technically this is a checkpoint (the on-disk state is valid) 1166 * even though we are leaving out all the file data. 1167 */ 1168 lfs_imtime(fs); 1169 lfs_seglock(fs, SEGM_CKP); 1170 sp = fs->lfs_sp; 1171 1172 /* 1173 * lfs_writevnodes, optimized to get dirops out of the way. 1174 * Only write dirops, and don't flush files' pages, only 1175 * blocks from the directories. 1176 * 1177 * We don't need to vref these files because they are 1178 * dirops and so hold an extra reference until the 1179 * segunlock clears them of that status. 1180 * 1181 * We don't need to check for IN_ADIROP because we know that 1182 * no dirops are active. 1183 * 1184 */ 1185 simple_lock(&fs->lfs_interlock); 1186 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1187 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1188 simple_unlock(&fs->lfs_interlock); 1189 vp = ITOV(ip); 1190 1191 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1192 1193 /* 1194 * All writes to directories come from dirops; all 1195 * writes to files' direct blocks go through the page 1196 * cache, which we're not touching. Reads to files 1197 * and/or directories will not be affected by writing 1198 * directory blocks inodes and file inodes. So we don't 1199 * really need to lock. If we don't lock, though, 1200 * make sure that we don't clear IN_MODIFIED 1201 * unnecessarily. 1202 */ 1203 if (vp->v_flag & (VXLOCK | VFREEING)) { 1204 simple_lock(&fs->lfs_interlock); 1205 continue; 1206 } 1207 waslocked = VOP_ISLOCKED(vp); 1208 if (vp->v_type != VREG && 1209 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1210 lfs_writefile(fs, sp, vp); 1211 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1212 !(ip->i_flag & IN_ALLMOD)) { 1213 LFS_SET_UINO(ip, IN_MODIFIED); 1214 } 1215 } 1216 (void) lfs_writeinode(fs, sp, ip); 1217 if (waslocked) 1218 LFS_SET_UINO(ip, IN_MODIFIED); 1219 simple_lock(&fs->lfs_interlock); 1220 } 1221 simple_unlock(&fs->lfs_interlock); 1222 /* We've written all the dirops there are */ 1223 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1224 lfs_finalize_fs_seguse(fs); 1225 (void) lfs_writeseg(fs, sp); 1226 lfs_segunlock(fs); 1227 } 1228 1229 /* 1230 * Flush all vnodes for which the pagedaemon has requested pageouts. 1231 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1232 * has just run, this would be an error). If we have to skip a vnode 1233 * for any reason, just skip it; if we have to wait for the cleaner, 1234 * abort. The writer daemon will call us again later. 1235 */ 1236 void 1237 lfs_flush_pchain(struct lfs *fs) 1238 { 1239 struct inode *ip, *nip; 1240 struct vnode *vp; 1241 extern int lfs_dostats; 1242 struct segment *sp; 1243 int error; 1244 1245 ASSERT_NO_SEGLOCK(fs); 1246 1247 if (fs->lfs_ronly) 1248 return; 1249 1250 simple_lock(&fs->lfs_interlock); 1251 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1252 simple_unlock(&fs->lfs_interlock); 1253 return; 1254 } else 1255 simple_unlock(&fs->lfs_interlock); 1256 1257 /* Get dirops out of the way */ 1258 lfs_flush_dirops(fs); 1259 1260 if (lfs_dostats) 1261 ++lfs_stats.flush_invoked; 1262 1263 /* 1264 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1265 */ 1266 lfs_imtime(fs); 1267 lfs_seglock(fs, 0); 1268 sp = fs->lfs_sp; 1269 1270 /* 1271 * lfs_writevnodes, optimized to clear pageout requests. 1272 * Only write non-dirop files that are in the pageout queue. 1273 * We're very conservative about what we write; we want to be 1274 * fast and async. 1275 */ 1276 simple_lock(&fs->lfs_interlock); 1277 top: 1278 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1279 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1280 vp = ITOV(ip); 1281 1282 if (!(ip->i_flags & IN_PAGING)) 1283 goto top; 1284 1285 if (vp->v_flag & (VXLOCK|VDIROP)) 1286 continue; 1287 if (vp->v_type != VREG) 1288 continue; 1289 if (lfs_vref(vp)) 1290 continue; 1291 simple_unlock(&fs->lfs_interlock); 1292 1293 if (VOP_ISLOCKED(vp)) { 1294 lfs_vunref(vp); 1295 simple_lock(&fs->lfs_interlock); 1296 continue; 1297 } 1298 1299 error = lfs_writefile(fs, sp, vp); 1300 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1301 !(ip->i_flag & IN_ALLMOD)) { 1302 LFS_SET_UINO(ip, IN_MODIFIED); 1303 } 1304 (void) lfs_writeinode(fs, sp, ip); 1305 1306 lfs_vunref(vp); 1307 1308 if (error == EAGAIN) { 1309 lfs_writeseg(fs, sp); 1310 simple_lock(&fs->lfs_interlock); 1311 break; 1312 } 1313 simple_lock(&fs->lfs_interlock); 1314 } 1315 simple_unlock(&fs->lfs_interlock); 1316 (void) lfs_writeseg(fs, sp); 1317 lfs_segunlock(fs); 1318 } 1319 1320 /* 1321 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1322 */ 1323 int 1324 lfs_fcntl(void *v) 1325 { 1326 struct vop_fcntl_args /* { 1327 struct vnode *a_vp; 1328 u_long a_command; 1329 caddr_t a_data; 1330 int a_fflag; 1331 kauth_cred_t a_cred; 1332 struct lwp *a_l; 1333 } */ *ap = v; 1334 struct timeval *tvp; 1335 BLOCK_INFO *blkiov; 1336 CLEANERINFO *cip; 1337 SEGUSE *sup; 1338 int blkcnt, error, oclean; 1339 struct lfs_fcntl_markv blkvp; 1340 struct proc *p; 1341 fsid_t *fsidp; 1342 struct lfs *fs; 1343 struct buf *bp; 1344 fhandle_t *fhp; 1345 daddr_t off; 1346 1347 /* Only respect LFS fcntls on fs root or Ifile */ 1348 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1349 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1350 return ufs_fcntl(v); 1351 } 1352 1353 /* Avoid locking a draining lock */ 1354 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1355 return ESHUTDOWN; 1356 } 1357 1358 p = ap->a_l->l_proc; 1359 fs = VTOI(ap->a_vp)->i_lfs; 1360 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1361 1362 switch (ap->a_command) { 1363 case LFCNSEGWAITALL: 1364 case LFCNSEGWAITALL_COMPAT: 1365 fsidp = NULL; 1366 /* FALLSTHROUGH */ 1367 case LFCNSEGWAIT: 1368 case LFCNSEGWAIT_COMPAT: 1369 tvp = (struct timeval *)ap->a_data; 1370 simple_lock(&fs->lfs_interlock); 1371 ++fs->lfs_sleepers; 1372 simple_unlock(&fs->lfs_interlock); 1373 1374 error = lfs_segwait(fsidp, tvp); 1375 1376 simple_lock(&fs->lfs_interlock); 1377 if (--fs->lfs_sleepers == 0) 1378 wakeup(&fs->lfs_sleepers); 1379 simple_unlock(&fs->lfs_interlock); 1380 return error; 1381 1382 case LFCNBMAPV: 1383 case LFCNMARKV: 1384 if ((error = kauth_authorize_generic(p->p_cred, KAUTH_GENERIC_ISSUSER, 1385 &p->p_acflag)) != 0) 1386 return (error); 1387 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1388 1389 blkcnt = blkvp.blkcnt; 1390 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1391 return (EINVAL); 1392 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1393 if ((error = copyin(blkvp.blkiov, blkiov, 1394 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1395 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1396 return error; 1397 } 1398 1399 simple_lock(&fs->lfs_interlock); 1400 ++fs->lfs_sleepers; 1401 simple_unlock(&fs->lfs_interlock); 1402 if (ap->a_command == LFCNBMAPV) 1403 error = lfs_bmapv(p, fsidp, blkiov, blkcnt); 1404 else /* LFCNMARKV */ 1405 error = lfs_markv(p, fsidp, blkiov, blkcnt); 1406 if (error == 0) 1407 error = copyout(blkiov, blkvp.blkiov, 1408 blkcnt * sizeof(BLOCK_INFO)); 1409 simple_lock(&fs->lfs_interlock); 1410 if (--fs->lfs_sleepers == 0) 1411 wakeup(&fs->lfs_sleepers); 1412 simple_unlock(&fs->lfs_interlock); 1413 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1414 return error; 1415 1416 case LFCNRECLAIM: 1417 /* 1418 * Flush dirops and write Ifile, allowing empty segments 1419 * to be immediately reclaimed. 1420 */ 1421 lfs_writer_enter(fs, "pndirop"); 1422 off = fs->lfs_offset; 1423 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1424 lfs_flush_dirops(fs); 1425 LFS_CLEANERINFO(cip, fs, bp); 1426 oclean = cip->clean; 1427 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1428 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1429 fs->lfs_sp->seg_flags |= SEGM_PROT; 1430 lfs_segunlock(fs); 1431 lfs_writer_leave(fs); 1432 1433 #ifdef DEBUG 1434 LFS_CLEANERINFO(cip, fs, bp); 1435 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1436 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1437 fs->lfs_offset - off, cip->clean - oclean, 1438 fs->lfs_activesb)); 1439 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1440 #endif 1441 1442 return 0; 1443 1444 case LFCNIFILEFH: 1445 /* Return the filehandle of the Ifile */ 1446 if ((error = kauth_authorize_generic(ap->a_l->l_proc->p_cred, 1447 KAUTH_GENERIC_ISSUSER, 1448 &ap->a_l->l_proc->p_acflag)) != 0) 1449 return (error); 1450 fhp = (struct fhandle *)ap->a_data; 1451 fhp->fh_fsid = *fsidp; 1452 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid)); 1453 1454 case LFCNREWIND: 1455 /* Move lfs_offset to the lowest-numbered segment */ 1456 return lfs_rewind(fs, *(int *)ap->a_data); 1457 1458 case LFCNINVAL: 1459 /* Mark a segment SEGUSE_INVAL */ 1460 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1461 if (sup->su_nbytes > 0) { 1462 brelse(bp); 1463 lfs_unset_inval_all(fs); 1464 return EBUSY; 1465 } 1466 sup->su_flags |= SEGUSE_INVAL; 1467 VOP_BWRITE(bp); 1468 return 0; 1469 1470 case LFCNRESIZE: 1471 /* Resize the filesystem */ 1472 return lfs_resize_fs(fs, *(int *)ap->a_data); 1473 1474 case LFCNWRAPSTOP: 1475 /* 1476 * Hold lfs_newseg at segment 0; sleep until the filesystem 1477 * wraps around. For debugging purposes, so an external 1478 * agent can log every segment in the filesystem as it 1479 * was written, and we can regression-test checkpoint 1480 * validity in the general case. 1481 */ 1482 simple_lock(&fs->lfs_interlock); 1483 fs->lfs_nowrap = 1; 1484 error = ltsleep(&fs->lfs_nowrap, PCATCH | PUSER | PNORELOCK, 1485 "segwrap", 0, &fs->lfs_interlock); 1486 if (error) { 1487 fs->lfs_nowrap = 0; 1488 wakeup(&fs->lfs_nowrap); 1489 } 1490 return 0; 1491 1492 case LFCNWRAPGO: 1493 /* 1494 * Having done its work, the agent wakes up the writer. 1495 * It sleeps until a new segment is selected. 1496 */ 1497 simple_lock(&fs->lfs_interlock); 1498 fs->lfs_nowrap = 0; 1499 wakeup(&fs->lfs_nowrap); 1500 ltsleep(&fs->lfs_nextseg, PCATCH | PUSER | PNORELOCK, 1501 "segment", 0, &fs->lfs_interlock); 1502 return 0; 1503 1504 default: 1505 return ufs_fcntl(v); 1506 } 1507 return 0; 1508 } 1509 1510 int 1511 lfs_getpages(void *v) 1512 { 1513 struct vop_getpages_args /* { 1514 struct vnode *a_vp; 1515 voff_t a_offset; 1516 struct vm_page **a_m; 1517 int *a_count; 1518 int a_centeridx; 1519 vm_prot_t a_access_type; 1520 int a_advice; 1521 int a_flags; 1522 } */ *ap = v; 1523 1524 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1525 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1526 return EPERM; 1527 } 1528 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1529 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1530 } 1531 1532 /* 1533 * we're relying on the fact that genfs_getpages() always read in 1534 * entire filesystem blocks. 1535 */ 1536 return genfs_getpages(v); 1537 } 1538 1539 /* 1540 * Make sure that for all pages in every block in the given range, 1541 * either all are dirty or all are clean. If any of the pages 1542 * we've seen so far are dirty, put the vnode on the paging chain, 1543 * and mark it IN_PAGING. 1544 * 1545 * If checkfirst != 0, don't check all the pages but return at the 1546 * first dirty page. 1547 */ 1548 static int 1549 check_dirty(struct lfs *fs, struct vnode *vp, 1550 off_t startoffset, off_t endoffset, off_t blkeof, 1551 int flags, int checkfirst) 1552 { 1553 int by_list; 1554 struct vm_page *curpg = NULL; /* XXX: gcc */ 1555 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1556 off_t soff = 0; /* XXX: gcc */ 1557 voff_t off; 1558 int i; 1559 int nonexistent; 1560 int any_dirty; /* number of dirty pages */ 1561 int dirty; /* number of dirty pages in a block */ 1562 int tdirty; 1563 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1564 int pagedaemon = (curproc == uvm.pagedaemon_proc); 1565 1566 ASSERT_MAYBE_SEGLOCK(fs); 1567 top: 1568 by_list = (vp->v_uobj.uo_npages <= 1569 ((endoffset - startoffset) >> PAGE_SHIFT) * 1570 UVM_PAGE_HASH_PENALTY); 1571 any_dirty = 0; 1572 1573 if (by_list) { 1574 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1575 } else { 1576 soff = startoffset; 1577 } 1578 while (by_list || soff < MIN(blkeof, endoffset)) { 1579 if (by_list) { 1580 /* 1581 * Find the first page in a block. Skip 1582 * blocks outside our area of interest or beyond 1583 * the end of file. 1584 */ 1585 if (pages_per_block > 1) { 1586 while (curpg && 1587 ((curpg->offset & fs->lfs_bmask) || 1588 curpg->offset >= vp->v_size || 1589 curpg->offset >= endoffset)) 1590 curpg = TAILQ_NEXT(curpg, listq); 1591 } 1592 if (curpg == NULL) 1593 break; 1594 soff = curpg->offset; 1595 } 1596 1597 /* 1598 * Mark all pages in extended range busy; find out if any 1599 * of them are dirty. 1600 */ 1601 nonexistent = dirty = 0; 1602 for (i = 0; i == 0 || i < pages_per_block; i++) { 1603 if (by_list && pages_per_block <= 1) { 1604 pgs[i] = pg = curpg; 1605 } else { 1606 off = soff + (i << PAGE_SHIFT); 1607 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1608 if (pg == NULL) { 1609 ++nonexistent; 1610 continue; 1611 } 1612 } 1613 KASSERT(pg != NULL); 1614 1615 /* 1616 * If we're holding the segment lock, we can deadlocked 1617 * against a process that has our page and is waiting 1618 * for the cleaner, while the cleaner waits for the 1619 * segment lock. Just bail in that case. 1620 */ 1621 if ((pg->flags & PG_BUSY) && 1622 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1623 if (by_list && i > 0) 1624 uvm_page_unbusy(pgs, i); 1625 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1626 return -1; 1627 } 1628 1629 while (pg->flags & PG_BUSY) { 1630 pg->flags |= PG_WANTED; 1631 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1632 "lfsput", 0); 1633 simple_lock(&vp->v_interlock); 1634 if (by_list) { 1635 if (i > 0) 1636 uvm_page_unbusy(pgs, i); 1637 goto top; 1638 } 1639 } 1640 pg->flags |= PG_BUSY; 1641 UVM_PAGE_OWN(pg, "lfs_putpages"); 1642 1643 pmap_page_protect(pg, VM_PROT_NONE); 1644 tdirty = (pmap_clear_modify(pg) || 1645 (pg->flags & PG_CLEAN) == 0); 1646 dirty += tdirty; 1647 } 1648 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1649 if (by_list) { 1650 curpg = TAILQ_NEXT(curpg, listq); 1651 } else { 1652 soff += fs->lfs_bsize; 1653 } 1654 continue; 1655 } 1656 1657 any_dirty += dirty; 1658 KASSERT(nonexistent == 0); 1659 1660 /* 1661 * If any are dirty make all dirty; unbusy them, 1662 * but if we were asked to clean, wire them so that 1663 * the pagedaemon doesn't bother us about them while 1664 * they're on their way to disk. 1665 */ 1666 for (i = 0; i == 0 || i < pages_per_block; i++) { 1667 pg = pgs[i]; 1668 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1669 if (dirty) { 1670 pg->flags &= ~PG_CLEAN; 1671 if (flags & PGO_FREE) { 1672 /* 1673 * Wire the page so that 1674 * pdaemon doesn't see it again. 1675 */ 1676 uvm_lock_pageq(); 1677 uvm_pagewire(pg); 1678 uvm_unlock_pageq(); 1679 1680 /* Suspended write flag */ 1681 pg->flags |= PG_DELWRI; 1682 } 1683 } 1684 if (pg->flags & PG_WANTED) 1685 wakeup(pg); 1686 pg->flags &= ~(PG_WANTED|PG_BUSY); 1687 UVM_PAGE_OWN(pg, NULL); 1688 } 1689 1690 if (checkfirst && any_dirty) 1691 break; 1692 1693 if (by_list) { 1694 curpg = TAILQ_NEXT(curpg, listq); 1695 } else { 1696 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1697 } 1698 } 1699 1700 return any_dirty; 1701 } 1702 1703 /* 1704 * lfs_putpages functions like genfs_putpages except that 1705 * 1706 * (1) It needs to bounds-check the incoming requests to ensure that 1707 * they are block-aligned; if they are not, expand the range and 1708 * do the right thing in case, e.g., the requested range is clean 1709 * but the expanded range is dirty. 1710 * (2) It needs to explicitly send blocks to be written when it is done. 1711 * VOP_PUTPAGES is not ever called with the seglock held, so 1712 * we simply take the seglock and let lfs_segunlock wait for us. 1713 * XXX Actually we can be called with the seglock held, if we have 1714 * XXX to flush a vnode while lfs_markv is in operation. As of this 1715 * XXX writing we panic in this case. 1716 * 1717 * Assumptions: 1718 * 1719 * (1) The caller does not hold any pages in this vnode busy. If it does, 1720 * there is a danger that when we expand the page range and busy the 1721 * pages we will deadlock. 1722 * (2) We are called with vp->v_interlock held; we must return with it 1723 * released. 1724 * (3) We don't absolutely have to free pages right away, provided that 1725 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1726 * us a request with PGO_FREE, we take the pages out of the paging 1727 * queue and wake up the writer, which will handle freeing them for us. 1728 * 1729 * We ensure that for any filesystem block, all pages for that 1730 * block are either resident or not, even if those pages are higher 1731 * than EOF; that means that we will be getting requests to free 1732 * "unused" pages above EOF all the time, and should ignore them. 1733 * 1734 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1735 */ 1736 1737 int 1738 lfs_putpages(void *v) 1739 { 1740 int error; 1741 struct vop_putpages_args /* { 1742 struct vnode *a_vp; 1743 voff_t a_offlo; 1744 voff_t a_offhi; 1745 int a_flags; 1746 } */ *ap = v; 1747 struct vnode *vp; 1748 struct inode *ip; 1749 struct lfs *fs; 1750 struct segment *sp; 1751 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1752 off_t off, max_endoffset; 1753 int s; 1754 boolean_t seglocked, sync, pagedaemon; 1755 struct vm_page *pg; 1756 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1757 1758 vp = ap->a_vp; 1759 ip = VTOI(vp); 1760 fs = ip->i_lfs; 1761 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1762 pagedaemon = (curproc == uvm.pagedaemon_proc); 1763 1764 /* Putpages does nothing for metadata. */ 1765 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1766 simple_unlock(&vp->v_interlock); 1767 return 0; 1768 } 1769 1770 /* 1771 * If there are no pages, don't do anything. 1772 */ 1773 if (vp->v_uobj.uo_npages == 0) { 1774 s = splbio(); 1775 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1776 (vp->v_flag & VONWORKLST)) { 1777 vp->v_flag &= ~VONWORKLST; 1778 LIST_REMOVE(vp, v_synclist); 1779 } 1780 splx(s); 1781 simple_unlock(&vp->v_interlock); 1782 1783 /* Remove us from paging queue, if we were on it */ 1784 simple_lock(&fs->lfs_interlock); 1785 if (ip->i_flags & IN_PAGING) { 1786 ip->i_flags &= ~IN_PAGING; 1787 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1788 } 1789 simple_unlock(&fs->lfs_interlock); 1790 return 0; 1791 } 1792 1793 blkeof = blkroundup(fs, ip->i_size); 1794 1795 /* 1796 * Ignore requests to free pages past EOF but in the same block 1797 * as EOF, unless the request is synchronous. (If the request is 1798 * sync, it comes from lfs_truncate.) 1799 * XXXUBC Make these pages look "active" so the pagedaemon won't 1800 * XXXUBC bother us with them again. 1801 */ 1802 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1803 origoffset = ap->a_offlo; 1804 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1805 pg = uvm_pagelookup(&vp->v_uobj, off); 1806 KASSERT(pg != NULL); 1807 while (pg->flags & PG_BUSY) { 1808 pg->flags |= PG_WANTED; 1809 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1810 "lfsput2", 0); 1811 simple_lock(&vp->v_interlock); 1812 } 1813 uvm_lock_pageq(); 1814 uvm_pageactivate(pg); 1815 uvm_unlock_pageq(); 1816 } 1817 ap->a_offlo = blkeof; 1818 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1819 simple_unlock(&vp->v_interlock); 1820 return 0; 1821 } 1822 } 1823 1824 /* 1825 * Extend page range to start and end at block boundaries. 1826 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1827 */ 1828 origoffset = ap->a_offlo; 1829 origendoffset = ap->a_offhi; 1830 startoffset = origoffset & ~(fs->lfs_bmask); 1831 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1832 << fs->lfs_bshift; 1833 1834 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1835 endoffset = max_endoffset; 1836 origendoffset = endoffset; 1837 } else { 1838 origendoffset = round_page(ap->a_offhi); 1839 endoffset = round_page(blkroundup(fs, origendoffset)); 1840 } 1841 1842 KASSERT(startoffset > 0 || endoffset >= startoffset); 1843 if (startoffset == endoffset) { 1844 /* Nothing to do, why were we called? */ 1845 simple_unlock(&vp->v_interlock); 1846 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1847 PRId64 "\n", startoffset)); 1848 return 0; 1849 } 1850 1851 ap->a_offlo = startoffset; 1852 ap->a_offhi = endoffset; 1853 1854 if (!(ap->a_flags & PGO_CLEANIT)) 1855 return genfs_putpages(v); 1856 1857 /* 1858 * If there are more than one page per block, we don't want 1859 * to get caught locking them backwards; so set PGO_BUSYFAIL 1860 * to avoid deadlocks. 1861 */ 1862 ap->a_flags |= PGO_BUSYFAIL; 1863 1864 do { 1865 int r; 1866 1867 /* If no pages are dirty, we can just use genfs_putpages. */ 1868 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 1869 ap->a_flags, 1); 1870 if (r < 0) { 1871 simple_unlock(&vp->v_interlock); 1872 return EDEADLK; 1873 } 1874 if (r > 0) 1875 break; 1876 1877 /* 1878 * Sometimes pages are dirtied between the time that 1879 * we check and the time we try to clean them. 1880 * Instruct lfs_gop_write to return EDEADLK in this case 1881 * so we can write them properly. 1882 */ 1883 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 1884 r = genfs_putpages(v); 1885 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 1886 if (r != EDEADLK) 1887 return r; 1888 1889 /* Start over. */ 1890 preempt(1); 1891 simple_lock(&vp->v_interlock); 1892 } while(1); 1893 1894 /* 1895 * Dirty and asked to clean. 1896 * 1897 * Pagedaemon can't actually write LFS pages; wake up 1898 * the writer to take care of that. The writer will 1899 * notice the pager inode queue and act on that. 1900 */ 1901 if (pagedaemon) { 1902 simple_lock(&fs->lfs_interlock); 1903 if (!(ip->i_flags & IN_PAGING)) { 1904 ip->i_flags |= IN_PAGING; 1905 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1906 } 1907 simple_lock(&lfs_subsys_lock); 1908 wakeup(&lfs_writer_daemon); 1909 simple_unlock(&lfs_subsys_lock); 1910 simple_unlock(&fs->lfs_interlock); 1911 simple_unlock(&vp->v_interlock); 1912 preempt(1); 1913 return EWOULDBLOCK; 1914 } 1915 1916 /* 1917 * If this is a file created in a recent dirop, we can't flush its 1918 * inode until the dirop is complete. Drain dirops, then flush the 1919 * filesystem (taking care of any other pending dirops while we're 1920 * at it). 1921 */ 1922 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1923 (vp->v_flag & VDIROP)) { 1924 int locked; 1925 1926 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 1927 locked = VOP_ISLOCKED(vp) && /* XXX */ 1928 vp->v_lock.lk_lockholder == curproc->p_pid; 1929 simple_unlock(&vp->v_interlock); 1930 lfs_writer_enter(fs, "ppdirop"); 1931 if (locked) 1932 VOP_UNLOCK(vp, 0); 1933 1934 simple_lock(&fs->lfs_interlock); 1935 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1936 simple_unlock(&fs->lfs_interlock); 1937 1938 simple_lock(&vp->v_interlock); 1939 if (locked) { 1940 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 1941 simple_lock(&vp->v_interlock); 1942 } 1943 lfs_writer_leave(fs); 1944 1945 /* XXX the flush should have taken care of this one too! */ 1946 } 1947 1948 /* 1949 * This is it. We are going to write some pages. From here on 1950 * down it's all just mechanics. 1951 * 1952 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1953 */ 1954 ap->a_flags &= ~PGO_SYNCIO; 1955 1956 /* 1957 * If we've already got the seglock, flush the node and return. 1958 * The FIP has already been set up for us by lfs_writefile, 1959 * and FIP cleanup and lfs_updatemeta will also be done there, 1960 * unless genfs_putpages returns EDEADLK; then we must flush 1961 * what we have, and correct FIP and segment header accounting. 1962 */ 1963 1964 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 1965 if (!seglocked) { 1966 simple_unlock(&vp->v_interlock); 1967 /* 1968 * Take the seglock, because we are going to be writing pages. 1969 */ 1970 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 1971 if (error != 0) 1972 return error; 1973 simple_lock(&vp->v_interlock); 1974 } 1975 1976 /* 1977 * VOP_PUTPAGES should not be called while holding the seglock. 1978 * XXXUBC fix lfs_markv, or do this properly. 1979 */ 1980 #ifdef notyet 1981 KASSERT(fs->lfs_seglock == 1); 1982 #endif /* notyet */ 1983 1984 /* 1985 * We assume we're being called with sp->fip pointing at blank space. 1986 * Account for a new FIP in the segment header, and set sp->vp. 1987 * (This should duplicate the setup at the top of lfs_writefile().) 1988 */ 1989 sp = fs->lfs_sp; 1990 if (!seglocked) { 1991 if (sp->seg_bytes_left < fs->lfs_bsize || 1992 sp->sum_bytes_left < sizeof(struct finfo)) 1993 (void) lfs_writeseg(fs, fs->lfs_sp); 1994 1995 sp->sum_bytes_left -= FINFOSIZE; 1996 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1997 } 1998 KASSERT(sp->vp == NULL); 1999 sp->vp = vp; 2000 2001 if (!seglocked) { 2002 if (vp->v_flag & VDIROP) 2003 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2004 } 2005 2006 sp->fip->fi_nblocks = 0; 2007 sp->fip->fi_ino = ip->i_number; 2008 sp->fip->fi_version = ip->i_gen; 2009 2010 /* 2011 * Loop through genfs_putpages until all pages are gathered. 2012 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2013 * Whenever we lose the interlock we have to rerun check_dirty, as 2014 * well. 2015 */ 2016 again: 2017 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2018 ap->a_flags, 0) < 0) { 2019 simple_unlock(&vp->v_interlock); 2020 sp->vp = NULL; 2021 if (!seglocked) 2022 lfs_segunlock(fs); 2023 return EDEADLK; 2024 } 2025 2026 error = genfs_putpages(v); 2027 if (error == EDEADLK || error == EAGAIN) { 2028 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2029 " EDEADLK [2] ino %d off %x (seg %d)\n", 2030 ip->i_number, fs->lfs_offset, 2031 dtosn(fs, fs->lfs_offset))); 2032 /* If nothing to write, short-circuit */ 2033 if (sp->cbpp - sp->bpp > 1) { 2034 /* Write gathered pages */ 2035 lfs_updatemeta(sp); 2036 (void) lfs_writeseg(fs, sp); 2037 2038 /* 2039 * Reinitialize brand new FIP and add us to it. 2040 * (This should duplicate the fixup in 2041 * lfs_gatherpages().) 2042 */ 2043 KASSERT(sp->vp == vp); 2044 sp->fip->fi_version = ip->i_gen; 2045 sp->fip->fi_ino = ip->i_number; 2046 /* Add us to the new segment summary. */ 2047 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 2048 sp->sum_bytes_left -= FINFOSIZE; 2049 } 2050 2051 /* Give the write a chance to complete */ 2052 preempt(1); 2053 2054 /* We've lost the interlock. Start over. */ 2055 if (error == EDEADLK) { 2056 simple_lock(&vp->v_interlock); 2057 goto again; 2058 } 2059 } 2060 2061 KASSERT(sp->vp == vp); 2062 if (!seglocked) { 2063 sp->vp = NULL; /* XXX lfs_gather below will set this */ 2064 2065 /* Write indirect blocks as well */ 2066 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2067 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2068 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2069 2070 KASSERT(sp->vp == NULL); 2071 sp->vp = vp; 2072 } 2073 2074 /* 2075 * Blocks are now gathered into a segment waiting to be written. 2076 * All that's left to do is update metadata, and write them. 2077 */ 2078 lfs_updatemeta(sp); 2079 KASSERT(sp->vp == vp); 2080 sp->vp = NULL; 2081 2082 if (seglocked) { 2083 /* we're called by lfs_writefile. */ 2084 return error; 2085 } 2086 2087 /* 2088 * Clean up FIP, since we're done writing this file. 2089 * This should duplicate cleanup at the end of lfs_writefile(). 2090 */ 2091 if (sp->fip->fi_nblocks != 0) { 2092 sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE + 2093 sizeof(int32_t) * sp->fip->fi_nblocks); 2094 sp->start_lbp = &sp->fip->fi_blocks[0]; 2095 } else { 2096 sp->sum_bytes_left += FINFOSIZE; 2097 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 2098 } 2099 lfs_writeseg(fs, fs->lfs_sp); 2100 2101 /* 2102 * Remove us from paging queue, since we've now written all our 2103 * pages. 2104 */ 2105 simple_lock(&fs->lfs_interlock); 2106 if (ip->i_flags & IN_PAGING) { 2107 ip->i_flags &= ~IN_PAGING; 2108 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2109 } 2110 simple_unlock(&fs->lfs_interlock); 2111 2112 /* 2113 * XXX - with the malloc/copy writeseg, the pages are freed by now 2114 * even if we don't wait (e.g. if we hold a nested lock). This 2115 * will not be true if we stop using malloc/copy. 2116 */ 2117 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2118 lfs_segunlock(fs); 2119 2120 /* 2121 * Wait for v_numoutput to drop to zero. The seglock should 2122 * take care of this, but there is a slight possibility that 2123 * aiodoned might not have got around to our buffers yet. 2124 */ 2125 if (sync) { 2126 s = splbio(); 2127 simple_lock(&global_v_numoutput_slock); 2128 while (vp->v_numoutput > 0) { 2129 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2130 " num %d\n", ip->i_number, vp->v_numoutput)); 2131 vp->v_flag |= VBWAIT; 2132 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2133 &global_v_numoutput_slock); 2134 } 2135 simple_unlock(&global_v_numoutput_slock); 2136 splx(s); 2137 } 2138 return error; 2139 } 2140 2141 /* 2142 * Return the last logical file offset that should be written for this file 2143 * if we're doing a write that ends at "size". If writing, we need to know 2144 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2145 * to know about entire blocks. 2146 */ 2147 void 2148 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2149 { 2150 struct inode *ip = VTOI(vp); 2151 struct lfs *fs = ip->i_lfs; 2152 daddr_t olbn, nlbn; 2153 2154 olbn = lblkno(fs, ip->i_size); 2155 nlbn = lblkno(fs, size); 2156 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2157 *eobp = fragroundup(fs, size); 2158 } else { 2159 *eobp = blkroundup(fs, size); 2160 } 2161 } 2162 2163 #ifdef DEBUG 2164 void lfs_dump_vop(void *); 2165 2166 void 2167 lfs_dump_vop(void *v) 2168 { 2169 struct vop_putpages_args /* { 2170 struct vnode *a_vp; 2171 voff_t a_offlo; 2172 voff_t a_offhi; 2173 int a_flags; 2174 } */ *ap = v; 2175 2176 #ifdef DDB 2177 vfs_vnode_print(ap->a_vp, 0, printf); 2178 #endif 2179 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2180 } 2181 #endif 2182 2183 int 2184 lfs_mmap(void *v) 2185 { 2186 struct vop_mmap_args /* { 2187 const struct vnodeop_desc *a_desc; 2188 struct vnode *a_vp; 2189 int a_fflags; 2190 kauth_cred_t a_cred; 2191 struct lwp *a_l; 2192 } */ *ap = v; 2193 2194 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2195 return EOPNOTSUPP; 2196 return ufs_mmap(v); 2197 } 2198
Indexes created Sat Oct 18 23:09:54 GMT 2025