lfs_vnops.c revision 1.143
1 /* $NetBSD: lfs_vnops.c,v 1.143 2005/04/14 00:58:26 perseant Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant (at) hhhh.org>. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 /* 39 * Copyright (c) 1986, 1989, 1991, 1993, 1995 40 * The Regents of the University of California. All rights reserved. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)lfs_vnops.c 8.13 (Berkeley) 6/10/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.143 2005/04/14 00:58:26 perseant Exp $"); 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/namei.h> 75 #include <sys/resourcevar.h> 76 #include <sys/kernel.h> 77 #include <sys/file.h> 78 #include <sys/stat.h> 79 #include <sys/buf.h> 80 #include <sys/proc.h> 81 #include <sys/mount.h> 82 #include <sys/vnode.h> 83 #include <sys/malloc.h> 84 #include <sys/pool.h> 85 #include <sys/signalvar.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_blkatoff_desc, lfs_blkatoff }, /* blkatoff */ 150 { &vop_valloc_desc, lfs_valloc }, /* valloc */ 151 { &vop_balloc_desc, lfs_balloc }, /* balloc */ 152 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 153 { &vop_truncate_desc, lfs_truncate }, /* truncate */ 154 { &vop_update_desc, lfs_update }, /* update */ 155 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 156 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 157 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 158 { NULL, NULL } 159 }; 160 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 161 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 162 163 int (**lfs_specop_p)(void *); 164 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 165 { &vop_default_desc, vn_default_error }, 166 { &vop_lookup_desc, spec_lookup }, /* lookup */ 167 { &vop_create_desc, spec_create }, /* create */ 168 { &vop_mknod_desc, spec_mknod }, /* mknod */ 169 { &vop_open_desc, spec_open }, /* open */ 170 { &vop_close_desc, lfsspec_close }, /* close */ 171 { &vop_access_desc, ufs_access }, /* access */ 172 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 173 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 174 { &vop_read_desc, ufsspec_read }, /* read */ 175 { &vop_write_desc, ufsspec_write }, /* write */ 176 { &vop_lease_desc, spec_lease_check }, /* lease */ 177 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 178 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 179 { &vop_poll_desc, spec_poll }, /* poll */ 180 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 181 { &vop_revoke_desc, spec_revoke }, /* revoke */ 182 { &vop_mmap_desc, spec_mmap }, /* mmap */ 183 { &vop_fsync_desc, spec_fsync }, /* fsync */ 184 { &vop_seek_desc, spec_seek }, /* seek */ 185 { &vop_remove_desc, spec_remove }, /* remove */ 186 { &vop_link_desc, spec_link }, /* link */ 187 { &vop_rename_desc, spec_rename }, /* rename */ 188 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 189 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 190 { &vop_symlink_desc, spec_symlink }, /* symlink */ 191 { &vop_readdir_desc, spec_readdir }, /* readdir */ 192 { &vop_readlink_desc, spec_readlink }, /* readlink */ 193 { &vop_abortop_desc, spec_abortop }, /* abortop */ 194 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 195 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 196 { &vop_lock_desc, ufs_lock }, /* lock */ 197 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 198 { &vop_bmap_desc, spec_bmap }, /* bmap */ 199 { &vop_strategy_desc, spec_strategy }, /* strategy */ 200 { &vop_print_desc, ufs_print }, /* print */ 201 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 202 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 203 { &vop_advlock_desc, spec_advlock }, /* advlock */ 204 { &vop_blkatoff_desc, spec_blkatoff }, /* blkatoff */ 205 { &vop_valloc_desc, spec_valloc }, /* valloc */ 206 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 207 { &vop_truncate_desc, spec_truncate }, /* truncate */ 208 { &vop_update_desc, lfs_update }, /* update */ 209 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 210 { &vop_getpages_desc, spec_getpages }, /* getpages */ 211 { &vop_putpages_desc, spec_putpages }, /* putpages */ 212 { NULL, NULL } 213 }; 214 const struct vnodeopv_desc lfs_specop_opv_desc = 215 { &lfs_specop_p, lfs_specop_entries }; 216 217 int (**lfs_fifoop_p)(void *); 218 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 219 { &vop_default_desc, vn_default_error }, 220 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 221 { &vop_create_desc, fifo_create }, /* create */ 222 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 223 { &vop_open_desc, fifo_open }, /* open */ 224 { &vop_close_desc, lfsfifo_close }, /* close */ 225 { &vop_access_desc, ufs_access }, /* access */ 226 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 227 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 228 { &vop_read_desc, ufsfifo_read }, /* read */ 229 { &vop_write_desc, ufsfifo_write }, /* write */ 230 { &vop_lease_desc, fifo_lease_check }, /* lease */ 231 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 232 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 233 { &vop_poll_desc, fifo_poll }, /* poll */ 234 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 235 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 236 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 237 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 238 { &vop_seek_desc, fifo_seek }, /* seek */ 239 { &vop_remove_desc, fifo_remove }, /* remove */ 240 { &vop_link_desc, fifo_link }, /* link */ 241 { &vop_rename_desc, fifo_rename }, /* rename */ 242 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 243 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 244 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 245 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 246 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 247 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 248 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 249 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 250 { &vop_lock_desc, ufs_lock }, /* lock */ 251 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 252 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 253 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 254 { &vop_print_desc, ufs_print }, /* print */ 255 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 256 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 257 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 258 { &vop_blkatoff_desc, fifo_blkatoff }, /* blkatoff */ 259 { &vop_valloc_desc, fifo_valloc }, /* valloc */ 260 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 261 { &vop_truncate_desc, fifo_truncate }, /* truncate */ 262 { &vop_update_desc, lfs_update }, /* update */ 263 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 264 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 265 { NULL, NULL } 266 }; 267 const struct vnodeopv_desc lfs_fifoop_opv_desc = 268 { &lfs_fifoop_p, lfs_fifoop_entries }; 269 270 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int); 271 272 /* 273 * A function version of LFS_ITIMES, for the UFS functions which call ITIMES 274 */ 275 void 276 lfs_itimes(struct inode *ip, struct timespec *acc, struct timespec *mod, struct timespec *cre) 277 { 278 LFS_ITIMES(ip, acc, mod, cre); 279 } 280 281 #define LFS_READWRITE 282 #include <ufs/ufs/ufs_readwrite.c> 283 #undef LFS_READWRITE 284 285 /* 286 * Synch an open file. 287 */ 288 /* ARGSUSED */ 289 int 290 lfs_fsync(void *v) 291 { 292 struct vop_fsync_args /* { 293 struct vnode *a_vp; 294 struct ucred *a_cred; 295 int a_flags; 296 off_t offlo; 297 off_t offhi; 298 struct proc *a_p; 299 } */ *ap = v; 300 struct vnode *vp = ap->a_vp; 301 int error, wait; 302 303 /* 304 * Trickle sync checks for need to do a checkpoint after possible 305 * activity from the pagedaemon. 306 */ 307 if (ap->a_flags & FSYNC_LAZY) { 308 simple_lock(&lfs_subsys_lock); 309 wakeup(&lfs_writer_daemon); 310 simple_unlock(&lfs_subsys_lock); 311 return 0; 312 } 313 314 wait = (ap->a_flags & FSYNC_WAIT); 315 simple_lock(&vp->v_interlock); 316 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 317 round_page(ap->a_offhi), 318 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 319 if (error) 320 return error; 321 error = VOP_UPDATE(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 322 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 323 int l = 0; 324 error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE, 325 ap->a_p->p_ucred, ap->a_p); 326 } 327 if (wait && !VPISEMPTY(vp)) 328 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 329 330 return error; 331 } 332 333 /* 334 * Take IN_ADIROP off, then call ufs_inactive. 335 */ 336 int 337 lfs_inactive(void *v) 338 { 339 struct vop_inactive_args /* { 340 struct vnode *a_vp; 341 struct proc *a_p; 342 } */ *ap = v; 343 344 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 345 346 lfs_unmark_vnode(ap->a_vp); 347 348 /* 349 * The Ifile is only ever inactivated on unmount. 350 * Streamline this process by not giving it more dirty blocks. 351 */ 352 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 353 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 354 VOP_UNLOCK(ap->a_vp, 0); 355 return 0; 356 } 357 358 return ufs_inactive(v); 359 } 360 361 /* 362 * These macros are used to bracket UFS directory ops, so that we can 363 * identify all the pages touched during directory ops which need to 364 * be ordered and flushed atomically, so that they may be recovered. 365 * 366 * Because we have to mark nodes VDIROP in order to prevent 367 * the cache from reclaiming them while a dirop is in progress, we must 368 * also manage the number of nodes so marked (otherwise we can run out). 369 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 370 * is decremented during segment write, when VDIROP is taken off. 371 */ 372 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 373 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 374 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 375 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 376 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 377 static int lfs_set_dirop(struct vnode *, struct vnode *); 378 379 static int 380 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 381 { 382 struct lfs *fs; 383 int error; 384 385 KASSERT(VOP_ISLOCKED(dvp)); 386 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 387 388 fs = VTOI(dvp)->i_lfs; 389 390 ASSERT_NO_SEGLOCK(fs); 391 /* 392 * LFS_NRESERVE calculates direct and indirect blocks as well 393 * as an inode block; an overestimate in most cases. 394 */ 395 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 396 return (error); 397 398 restart: 399 simple_lock(&fs->lfs_interlock); 400 if (fs->lfs_dirops == 0) { 401 simple_unlock(&fs->lfs_interlock); 402 lfs_check(dvp, LFS_UNUSED_LBN, 0); 403 simple_lock(&fs->lfs_interlock); 404 } 405 while (fs->lfs_writer) 406 ltsleep(&fs->lfs_dirops, (PRIBIO + 1), "lfs_sdirop", 0, 407 &fs->lfs_interlock); 408 simple_lock(&lfs_subsys_lock); 409 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 410 wakeup(&lfs_writer_daemon); 411 simple_unlock(&lfs_subsys_lock); 412 simple_unlock(&fs->lfs_interlock); 413 preempt(1); 414 goto restart; 415 } 416 417 if (lfs_dirvcount > LFS_MAX_DIROP) { 418 simple_unlock(&fs->lfs_interlock); 419 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 420 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 421 if ((error = ltsleep(&lfs_dirvcount, 422 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 423 &lfs_subsys_lock)) != 0) { 424 goto unreserve; 425 } 426 goto restart; 427 } 428 simple_unlock(&lfs_subsys_lock); 429 430 ++fs->lfs_dirops; 431 fs->lfs_doifile = 1; 432 simple_unlock(&fs->lfs_interlock); 433 434 /* Hold a reference so SET_ENDOP will be happy */ 435 vref(dvp); 436 if (vp) { 437 vref(vp); 438 MARK_VNODE(vp); 439 } 440 441 MARK_VNODE(dvp); 442 return 0; 443 444 unreserve: 445 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 446 return error; 447 } 448 449 /* 450 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 451 * in getnewvnode(), if we have a stacked filesystem mounted on top 452 * of us. 453 * 454 * NB: this means we have to clear the new vnodes on error. Fortunately 455 * SET_ENDOP is there to do that for us. 456 */ 457 static int 458 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 459 { 460 int error; 461 struct lfs *fs; 462 463 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 464 ASSERT_NO_SEGLOCK(fs); 465 if (fs->lfs_ronly) 466 return EROFS; 467 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 468 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 469 dvp, error)); 470 return error; 471 } 472 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 473 if (vpp) { 474 ungetnewvnode(*vpp); 475 *vpp = NULL; 476 } 477 return error; 478 } 479 return 0; 480 } 481 482 #define SET_ENDOP_BASE(fs, dvp, str) \ 483 do { \ 484 simple_lock(&(fs)->lfs_interlock); \ 485 --(fs)->lfs_dirops; \ 486 if (!(fs)->lfs_dirops) { \ 487 if ((fs)->lfs_nadirop) { \ 488 panic("SET_ENDOP: %s: no dirops but " \ 489 " nadirop=%d", (str), \ 490 (fs)->lfs_nadirop); \ 491 } \ 492 wakeup(&(fs)->lfs_writer); \ 493 simple_unlock(&(fs)->lfs_interlock); \ 494 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 495 } else \ 496 simple_unlock(&(fs)->lfs_interlock); \ 497 } while(0) 498 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 499 do { \ 500 UNMARK_VNODE(dvp); \ 501 if (nvpp && *nvpp) \ 502 UNMARK_VNODE(*nvpp); \ 503 /* Check for error return to stem vnode leakage */ \ 504 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 505 ungetnewvnode(*(nvpp)); \ 506 SET_ENDOP_BASE((fs), (dvp), (str)); \ 507 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 508 vrele(dvp); \ 509 } while(0) 510 #define SET_ENDOP_CREATE_AP(ap, str) \ 511 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 512 (ap)->a_vpp, (str)) 513 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 514 do { \ 515 UNMARK_VNODE(dvp); \ 516 if (ovp) \ 517 UNMARK_VNODE(ovp); \ 518 SET_ENDOP_BASE((fs), (dvp), (str)); \ 519 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 520 vrele(dvp); \ 521 if (ovp) \ 522 vrele(ovp); \ 523 } while(0) 524 525 void 526 lfs_mark_vnode(struct vnode *vp) 527 { 528 struct inode *ip = VTOI(vp); 529 struct lfs *fs = ip->i_lfs; 530 531 simple_lock(&fs->lfs_interlock); 532 if (!(ip->i_flag & IN_ADIROP)) { 533 if (!(vp->v_flag & VDIROP)) { 534 (void)lfs_vref(vp); 535 simple_lock(&lfs_subsys_lock); 536 ++lfs_dirvcount; 537 simple_unlock(&lfs_subsys_lock); 538 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 539 vp->v_flag |= VDIROP; 540 } 541 ++fs->lfs_nadirop; 542 ip->i_flag |= IN_ADIROP; 543 } else 544 KASSERT(vp->v_flag & VDIROP); 545 simple_unlock(&fs->lfs_interlock); 546 } 547 548 void 549 lfs_unmark_vnode(struct vnode *vp) 550 { 551 struct inode *ip = VTOI(vp); 552 553 if (ip->i_flag & IN_ADIROP) { 554 KASSERT(vp->v_flag & VDIROP); 555 simple_lock(&ip->i_lfs->lfs_interlock); 556 --ip->i_lfs->lfs_nadirop; 557 simple_unlock(&ip->i_lfs->lfs_interlock); 558 ip->i_flag &= ~IN_ADIROP; 559 } 560 } 561 562 int 563 lfs_symlink(void *v) 564 { 565 struct vop_symlink_args /* { 566 struct vnode *a_dvp; 567 struct vnode **a_vpp; 568 struct componentname *a_cnp; 569 struct vattr *a_vap; 570 char *a_target; 571 } */ *ap = v; 572 int error; 573 574 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 575 vput(ap->a_dvp); 576 return error; 577 } 578 error = ufs_symlink(ap); 579 SET_ENDOP_CREATE_AP(ap, "symlink"); 580 return (error); 581 } 582 583 int 584 lfs_mknod(void *v) 585 { 586 struct vop_mknod_args /* { 587 struct vnode *a_dvp; 588 struct vnode **a_vpp; 589 struct componentname *a_cnp; 590 struct vattr *a_vap; 591 } */ *ap = v; 592 struct vattr *vap = ap->a_vap; 593 struct vnode **vpp = ap->a_vpp; 594 struct inode *ip; 595 int error; 596 struct mount *mp; 597 ino_t ino; 598 599 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 600 vput(ap->a_dvp); 601 return error; 602 } 603 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 604 ap->a_dvp, vpp, ap->a_cnp); 605 606 /* Either way we're done with the dirop at this point */ 607 SET_ENDOP_CREATE_AP(ap, "mknod"); 608 609 if (error) 610 return (error); 611 612 ip = VTOI(*vpp); 613 mp = (*vpp)->v_mount; 614 ino = ip->i_number; 615 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 616 if (vap->va_rdev != VNOVAL) { 617 /* 618 * Want to be able to use this to make badblock 619 * inodes, so don't truncate the dev number. 620 */ 621 #if 0 622 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 623 UFS_MPNEEDSWAP((*vpp)->v_mount)); 624 #else 625 ip->i_ffs1_rdev = vap->va_rdev; 626 #endif 627 } 628 629 /* 630 * Call fsync to write the vnode so that we don't have to deal with 631 * flushing it when it's marked VDIROP|VXLOCK. 632 * 633 * XXX KS - If we can't flush we also can't call vgone(), so must 634 * return. But, that leaves this vnode in limbo, also not good. 635 * Can this ever happen (barring hardware failure)? 636 */ 637 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 638 curproc)) != 0) { 639 panic("lfs_mknod: couldn't fsync (ino %d)", 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_getattr_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 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1046 ip->inode_ext.lfs = NULL; 1047 pool_put(&lfs_inode_pool, vp->v_data); 1048 vp->v_data = NULL; 1049 return (0); 1050 } 1051 1052 /* 1053 * Read a block from a storage device. 1054 * In order to avoid reading blocks that are in the process of being 1055 * written by the cleaner---and hence are not mutexed by the normal 1056 * buffer cache / page cache mechanisms---check for collisions before 1057 * reading. 1058 * 1059 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1060 * the active cleaner test. 1061 * 1062 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1063 */ 1064 int 1065 lfs_strategy(void *v) 1066 { 1067 struct vop_strategy_args /* { 1068 struct vnode *a_vp; 1069 struct buf *a_bp; 1070 } */ *ap = v; 1071 struct buf *bp; 1072 struct lfs *fs; 1073 struct vnode *vp; 1074 struct inode *ip; 1075 daddr_t tbn; 1076 int i, sn, error, slept; 1077 1078 bp = ap->a_bp; 1079 vp = ap->a_vp; 1080 ip = VTOI(vp); 1081 fs = ip->i_lfs; 1082 1083 /* lfs uses its strategy routine only for read */ 1084 KASSERT(bp->b_flags & B_READ); 1085 1086 if (vp->v_type == VBLK || vp->v_type == VCHR) 1087 panic("lfs_strategy: spec"); 1088 KASSERT(bp->b_bcount != 0); 1089 if (bp->b_blkno == bp->b_lblkno) { 1090 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1091 NULL); 1092 if (error) { 1093 bp->b_error = error; 1094 bp->b_flags |= B_ERROR; 1095 biodone(bp); 1096 return (error); 1097 } 1098 if ((long)bp->b_blkno == -1) /* no valid data */ 1099 clrbuf(bp); 1100 } 1101 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1102 biodone(bp); 1103 return (0); 1104 } 1105 1106 slept = 1; 1107 simple_lock(&fs->lfs_interlock); 1108 while (slept && fs->lfs_seglock) { 1109 simple_unlock(&fs->lfs_interlock); 1110 /* 1111 * Look through list of intervals. 1112 * There will only be intervals to look through 1113 * if the cleaner holds the seglock. 1114 * Since the cleaner is synchronous, we can trust 1115 * the list of intervals to be current. 1116 */ 1117 tbn = dbtofsb(fs, bp->b_blkno); 1118 sn = dtosn(fs, tbn); 1119 slept = 0; 1120 for (i = 0; i < fs->lfs_cleanind; i++) { 1121 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1122 tbn >= fs->lfs_cleanint[i]) { 1123 DLOG((DLOG_CLEAN, 1124 "lfs_strategy: ino %d lbn %" PRId64 1125 " ind %d sn %d fsb %" PRIx32 1126 " given sn %d fsb %" PRIx64 "\n", 1127 ip->i_number, bp->b_lblkno, i, 1128 dtosn(fs, fs->lfs_cleanint[i]), 1129 fs->lfs_cleanint[i], sn, tbn)); 1130 DLOG((DLOG_CLEAN, 1131 "lfs_strategy: sleeping on ino %d lbn %" 1132 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1133 simple_lock(&fs->lfs_interlock); 1134 if (fs->lfs_seglock) 1135 ltsleep(&fs->lfs_seglock, 1136 (PRIBIO + 1) | PNORELOCK, 1137 "lfs_strategy", 0, 1138 &fs->lfs_interlock); 1139 /* Things may be different now; start over. */ 1140 slept = 1; 1141 break; 1142 } 1143 } 1144 simple_lock(&fs->lfs_interlock); 1145 } 1146 simple_unlock(&fs->lfs_interlock); 1147 1148 vp = ip->i_devvp; 1149 VOP_STRATEGY(vp, bp); 1150 return (0); 1151 } 1152 1153 static void 1154 lfs_flush_dirops(struct lfs *fs) 1155 { 1156 struct inode *ip, *nip; 1157 struct vnode *vp; 1158 extern int lfs_dostats; 1159 struct segment *sp; 1160 int needunlock; 1161 1162 ASSERT_NO_SEGLOCK(fs); 1163 1164 if (fs->lfs_ronly) 1165 return; 1166 1167 simple_lock(&fs->lfs_interlock); 1168 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1169 simple_unlock(&fs->lfs_interlock); 1170 return; 1171 } else 1172 simple_unlock(&fs->lfs_interlock); 1173 1174 if (lfs_dostats) 1175 ++lfs_stats.flush_invoked; 1176 1177 /* 1178 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1179 * Technically this is a checkpoint (the on-disk state is valid) 1180 * even though we are leaving out all the file data. 1181 */ 1182 lfs_imtime(fs); 1183 lfs_seglock(fs, SEGM_CKP); 1184 sp = fs->lfs_sp; 1185 1186 /* 1187 * lfs_writevnodes, optimized to get dirops out of the way. 1188 * Only write dirops, and don't flush files' pages, only 1189 * blocks from the directories. 1190 * 1191 * We don't need to vref these files because they are 1192 * dirops and so hold an extra reference until the 1193 * segunlock clears them of that status. 1194 * 1195 * We don't need to check for IN_ADIROP because we know that 1196 * no dirops are active. 1197 * 1198 */ 1199 simple_lock(&fs->lfs_interlock); 1200 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1201 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1202 simple_unlock(&fs->lfs_interlock); 1203 vp = ITOV(ip); 1204 1205 /* 1206 * All writes to directories come from dirops; all 1207 * writes to files' direct blocks go through the page 1208 * cache, which we're not touching. Reads to files 1209 * and/or directories will not be affected by writing 1210 * directory blocks inodes and file inodes. So we don't 1211 * really need to lock. If we don't lock, though, 1212 * make sure that we don't clear IN_MODIFIED 1213 * unnecessarily. 1214 */ 1215 if (vp->v_flag & VXLOCK) 1216 continue; 1217 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) { 1218 needunlock = 1; 1219 } else { 1220 DLOG((DLOG_VNODE, "lfs_flush_dirops: flushing locked ino %d\n", 1221 VTOI(vp)->i_number)); 1222 needunlock = 0; 1223 } 1224 if (vp->v_type != VREG && 1225 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1226 lfs_writefile(fs, sp, vp); 1227 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1228 !(ip->i_flag & IN_ALLMOD)) { 1229 LFS_SET_UINO(ip, IN_MODIFIED); 1230 } 1231 } 1232 (void) lfs_writeinode(fs, sp, ip); 1233 if (needunlock) 1234 VOP_UNLOCK(vp, 0); 1235 else 1236 LFS_SET_UINO(ip, IN_MODIFIED); 1237 simple_lock(&fs->lfs_interlock); 1238 } 1239 simple_unlock(&fs->lfs_interlock); 1240 /* We've written all the dirops there are */ 1241 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1242 (void) lfs_writeseg(fs, sp); 1243 lfs_segunlock(fs); 1244 } 1245 1246 /* 1247 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1248 */ 1249 int 1250 lfs_fcntl(void *v) 1251 { 1252 struct vop_fcntl_args /* { 1253 struct vnode *a_vp; 1254 u_long a_command; 1255 caddr_t a_data; 1256 int a_fflag; 1257 struct ucred *a_cred; 1258 struct proc *a_p; 1259 } */ *ap = v; 1260 struct timeval *tvp; 1261 BLOCK_INFO *blkiov; 1262 CLEANERINFO *cip; 1263 int blkcnt, error, oclean; 1264 struct lfs_fcntl_markv blkvp; 1265 fsid_t *fsidp; 1266 struct lfs *fs; 1267 struct buf *bp; 1268 fhandle_t *fhp; 1269 daddr_t off; 1270 1271 /* Only respect LFS fcntls on fs root or Ifile */ 1272 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1273 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1274 return ufs_fcntl(v); 1275 } 1276 1277 /* Avoid locking a draining lock */ 1278 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1279 return ESHUTDOWN; 1280 } 1281 1282 fs = VTOI(ap->a_vp)->i_lfs; 1283 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1284 1285 switch (ap->a_command) { 1286 case LFCNSEGWAITALL: 1287 case LFCNSEGWAITALL_COMPAT: 1288 fsidp = NULL; 1289 /* FALLSTHROUGH */ 1290 case LFCNSEGWAIT: 1291 case LFCNSEGWAIT_COMPAT: 1292 tvp = (struct timeval *)ap->a_data; 1293 simple_lock(&fs->lfs_interlock); 1294 ++fs->lfs_sleepers; 1295 simple_unlock(&fs->lfs_interlock); 1296 VOP_UNLOCK(ap->a_vp, 0); 1297 1298 error = lfs_segwait(fsidp, tvp); 1299 1300 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1301 simple_lock(&fs->lfs_interlock); 1302 if (--fs->lfs_sleepers == 0) 1303 wakeup(&fs->lfs_sleepers); 1304 simple_unlock(&fs->lfs_interlock); 1305 return error; 1306 1307 case LFCNBMAPV: 1308 case LFCNMARKV: 1309 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1310 return (error); 1311 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1312 1313 blkcnt = blkvp.blkcnt; 1314 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1315 return (EINVAL); 1316 blkiov = malloc(blkcnt * sizeof(BLOCK_INFO), M_SEGMENT, M_WAITOK); 1317 if ((error = copyin(blkvp.blkiov, blkiov, 1318 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1319 free(blkiov, M_SEGMENT); 1320 return error; 1321 } 1322 1323 simple_lock(&fs->lfs_interlock); 1324 ++fs->lfs_sleepers; 1325 simple_unlock(&fs->lfs_interlock); 1326 VOP_UNLOCK(ap->a_vp, 0); 1327 if (ap->a_command == LFCNBMAPV) 1328 error = lfs_bmapv(ap->a_p, fsidp, blkiov, blkcnt); 1329 else /* LFCNMARKV */ 1330 error = lfs_markv(ap->a_p, fsidp, blkiov, blkcnt); 1331 if (error == 0) 1332 error = copyout(blkiov, blkvp.blkiov, 1333 blkcnt * sizeof(BLOCK_INFO)); 1334 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1335 simple_lock(&fs->lfs_interlock); 1336 if (--fs->lfs_sleepers == 0) 1337 wakeup(&fs->lfs_sleepers); 1338 simple_unlock(&fs->lfs_interlock); 1339 free(blkiov, M_SEGMENT); 1340 return error; 1341 1342 case LFCNRECLAIM: 1343 /* 1344 * Flush dirops and write Ifile, allowing empty segments 1345 * to be immediately reclaimed. 1346 */ 1347 VOP_UNLOCK(ap->a_vp, 0); 1348 lfs_writer_enter(fs, "pndirop"); 1349 off = fs->lfs_offset; 1350 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1351 lfs_flush_dirops(fs); 1352 LFS_CLEANERINFO(cip, fs, bp); 1353 oclean = cip->clean; 1354 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1355 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1356 lfs_segunlock(fs); 1357 lfs_writer_leave(fs); 1358 1359 #ifdef DEBUG 1360 LFS_CLEANERINFO(cip, fs, bp); 1361 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1362 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1363 fs->lfs_offset - off, cip->clean - oclean, 1364 fs->lfs_activesb)); 1365 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1366 #endif 1367 1368 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1369 return 0; 1370 1371 case LFCNIFILEFH: 1372 /* Return the filehandle of the Ifile */ 1373 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1374 return (error); 1375 fhp = (struct fhandle *)ap->a_data; 1376 fhp->fh_fsid = *fsidp; 1377 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid)); 1378 1379 default: 1380 return ufs_fcntl(v); 1381 } 1382 return 0; 1383 } 1384 1385 int 1386 lfs_getpages(void *v) 1387 { 1388 struct vop_getpages_args /* { 1389 struct vnode *a_vp; 1390 voff_t a_offset; 1391 struct vm_page **a_m; 1392 int *a_count; 1393 int a_centeridx; 1394 vm_prot_t a_access_type; 1395 int a_advice; 1396 int a_flags; 1397 } */ *ap = v; 1398 1399 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1400 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1401 return EPERM; 1402 } 1403 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1404 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1405 } 1406 1407 /* 1408 * we're relying on the fact that genfs_getpages() always read in 1409 * entire filesystem blocks. 1410 */ 1411 return genfs_getpages(v); 1412 } 1413 1414 /* 1415 * Make sure that for all pages in every block in the given range, 1416 * either all are dirty or all are clean. If any of the pages 1417 * we've seen so far are dirty, put the vnode on the paging chain, 1418 * and mark it IN_PAGING. 1419 * 1420 * If checkfirst != 0, don't check all the pages but return at the 1421 * first dirty page. 1422 */ 1423 static int 1424 check_dirty(struct lfs *fs, struct vnode *vp, 1425 off_t startoffset, off_t endoffset, off_t blkeof, 1426 int flags, int checkfirst) 1427 { 1428 int by_list; 1429 struct vm_page *curpg = NULL; /* XXX: gcc */ 1430 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1431 off_t soff = 0; /* XXX: gcc */ 1432 voff_t off; 1433 int i; 1434 int nonexistent; 1435 int any_dirty; /* number of dirty pages */ 1436 int dirty; /* number of dirty pages in a block */ 1437 int tdirty; 1438 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1439 1440 ASSERT_MAYBE_SEGLOCK(fs); 1441 top: 1442 by_list = (vp->v_uobj.uo_npages <= 1443 ((endoffset - startoffset) >> PAGE_SHIFT) * 1444 UVM_PAGE_HASH_PENALTY); 1445 any_dirty = 0; 1446 1447 if (by_list) { 1448 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1449 } else { 1450 soff = startoffset; 1451 } 1452 while (by_list || soff < MIN(blkeof, endoffset)) { 1453 if (by_list) { 1454 /* 1455 * Find the first page in a block. Skip 1456 * blocks outside our area of interest or beyond 1457 * the end of file. 1458 */ 1459 if (pages_per_block > 1) { 1460 while (curpg && 1461 ((curpg->offset & fs->lfs_bmask) || 1462 curpg->offset >= vp->v_size || 1463 curpg->offset >= endoffset)) 1464 curpg = TAILQ_NEXT(curpg, listq); 1465 } 1466 if (curpg == NULL) 1467 break; 1468 soff = curpg->offset; 1469 } 1470 1471 /* 1472 * Mark all pages in extended range busy; find out if any 1473 * of them are dirty. 1474 */ 1475 nonexistent = dirty = 0; 1476 for (i = 0; i == 0 || i < pages_per_block; i++) { 1477 if (by_list && pages_per_block <= 1) { 1478 pgs[i] = pg = curpg; 1479 } else { 1480 off = soff + (i << PAGE_SHIFT); 1481 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1482 if (pg == NULL) { 1483 ++nonexistent; 1484 continue; 1485 } 1486 } 1487 KASSERT(pg != NULL); 1488 while (pg->flags & PG_BUSY) { 1489 pg->flags |= PG_WANTED; 1490 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1491 "lfsput", 0); 1492 simple_lock(&vp->v_interlock); 1493 if (by_list) { 1494 if (i > 0) 1495 uvm_page_unbusy(pgs, i); 1496 goto top; 1497 } 1498 } 1499 pg->flags |= PG_BUSY; 1500 UVM_PAGE_OWN(pg, "lfs_putpages"); 1501 1502 pmap_page_protect(pg, VM_PROT_NONE); 1503 tdirty = (pmap_clear_modify(pg) || 1504 (pg->flags & PG_CLEAN) == 0); 1505 dirty += tdirty; 1506 } 1507 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1508 if (by_list) { 1509 curpg = TAILQ_NEXT(curpg, listq); 1510 } else { 1511 soff += fs->lfs_bsize; 1512 } 1513 continue; 1514 } 1515 1516 any_dirty += dirty; 1517 KASSERT(nonexistent == 0); 1518 1519 /* 1520 * If any are dirty make all dirty; unbusy them, 1521 * but if we were asked to clean, wire them so that 1522 * the pagedaemon doesn't bother us about them while 1523 * they're on their way to disk. 1524 */ 1525 for (i = 0; i == 0 || i < pages_per_block; i++) { 1526 pg = pgs[i]; 1527 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1528 if (dirty) { 1529 pg->flags &= ~PG_CLEAN; 1530 if (flags & PGO_FREE) { 1531 /* XXXUBC need better way to update */ 1532 simple_lock(&lfs_subsys_lock); 1533 lfs_subsys_pages += MIN(1, pages_per_block); 1534 simple_unlock(&lfs_subsys_lock); 1535 /* 1536 * Wire the page so that 1537 * pdaemon doesn't see it again. 1538 */ 1539 uvm_lock_pageq(); 1540 uvm_pagewire(pg); 1541 uvm_unlock_pageq(); 1542 1543 /* Suspended write flag */ 1544 pg->flags |= PG_DELWRI; 1545 } 1546 } 1547 if (pg->flags & PG_WANTED) 1548 wakeup(pg); 1549 pg->flags &= ~(PG_WANTED|PG_BUSY); 1550 UVM_PAGE_OWN(pg, NULL); 1551 } 1552 1553 if (checkfirst && any_dirty) 1554 break; 1555 1556 if (by_list) { 1557 curpg = TAILQ_NEXT(curpg, listq); 1558 } else { 1559 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1560 } 1561 } 1562 1563 /* 1564 * If any pages were dirty, mark this inode as "pageout requested", 1565 * and put it on the paging queue. 1566 * XXXUBC locking (check locking on dchainhd too) 1567 */ 1568 #ifdef notyet 1569 if (any_dirty) { 1570 if (!(ip->i_flags & IN_PAGING)) { 1571 ip->i_flags |= IN_PAGING; 1572 simple_lock(&fs->lfs_interlock); 1573 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1574 simple_unlock(&fs->lfs_interlock); 1575 } 1576 } 1577 #endif 1578 return any_dirty; 1579 } 1580 1581 /* 1582 * lfs_putpages functions like genfs_putpages except that 1583 * 1584 * (1) It needs to bounds-check the incoming requests to ensure that 1585 * they are block-aligned; if they are not, expand the range and 1586 * do the right thing in case, e.g., the requested range is clean 1587 * but the expanded range is dirty. 1588 * (2) It needs to explicitly send blocks to be written when it is done. 1589 * VOP_PUTPAGES is not ever called with the seglock held, so 1590 * we simply take the seglock and let lfs_segunlock wait for us. 1591 * XXX Actually we can be called with the seglock held, if we have 1592 * XXX to flush a vnode while lfs_markv is in operation. As of this 1593 * XXX writing we panic in this case. 1594 * 1595 * Assumptions: 1596 * 1597 * (1) The caller does not hold any pages in this vnode busy. If it does, 1598 * there is a danger that when we expand the page range and busy the 1599 * pages we will deadlock. 1600 * (2) We are called with vp->v_interlock held; we must return with it 1601 * released. 1602 * (3) We don't absolutely have to free pages right away, provided that 1603 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1604 * us a request with PGO_FREE, we take the pages out of the paging 1605 * queue and wake up the writer, which will handle freeing them for us. 1606 * 1607 * We ensure that for any filesystem block, all pages for that 1608 * block are either resident or not, even if those pages are higher 1609 * than EOF; that means that we will be getting requests to free 1610 * "unused" pages above EOF all the time, and should ignore them. 1611 * 1612 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1613 */ 1614 1615 int 1616 lfs_putpages(void *v) 1617 { 1618 int error; 1619 struct vop_putpages_args /* { 1620 struct vnode *a_vp; 1621 voff_t a_offlo; 1622 voff_t a_offhi; 1623 int a_flags; 1624 } */ *ap = v; 1625 struct vnode *vp; 1626 struct inode *ip; 1627 struct lfs *fs; 1628 struct segment *sp; 1629 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1630 off_t off, max_endoffset; 1631 int s; 1632 boolean_t seglocked, sync, pagedaemon; 1633 struct vm_page *pg; 1634 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1635 1636 vp = ap->a_vp; 1637 ip = VTOI(vp); 1638 fs = ip->i_lfs; 1639 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1640 pagedaemon = (curproc == uvm.pagedaemon_proc); 1641 1642 /* Putpages does nothing for metadata. */ 1643 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1644 simple_unlock(&vp->v_interlock); 1645 return 0; 1646 } 1647 1648 /* 1649 * If there are no pages, don't do anything. 1650 */ 1651 if (vp->v_uobj.uo_npages == 0) { 1652 s = splbio(); 1653 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1654 (vp->v_flag & VONWORKLST)) { 1655 vp->v_flag &= ~VONWORKLST; 1656 LIST_REMOVE(vp, v_synclist); 1657 } 1658 splx(s); 1659 simple_unlock(&vp->v_interlock); 1660 return 0; 1661 } 1662 1663 blkeof = blkroundup(fs, ip->i_size); 1664 1665 /* 1666 * Ignore requests to free pages past EOF but in the same block 1667 * as EOF, unless the request is synchronous. (XXX why sync?) 1668 * XXXUBC Make these pages look "active" so the pagedaemon won't 1669 * XXXUBC bother us with them again. 1670 */ 1671 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1672 origoffset = ap->a_offlo; 1673 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1674 pg = uvm_pagelookup(&vp->v_uobj, off); 1675 KASSERT(pg != NULL); 1676 while (pg->flags & PG_BUSY) { 1677 pg->flags |= PG_WANTED; 1678 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1679 "lfsput2", 0); 1680 simple_lock(&vp->v_interlock); 1681 } 1682 uvm_lock_pageq(); 1683 uvm_pageactivate(pg); 1684 uvm_unlock_pageq(); 1685 } 1686 ap->a_offlo = blkeof; 1687 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1688 simple_unlock(&vp->v_interlock); 1689 return 0; 1690 } 1691 } 1692 1693 /* 1694 * Extend page range to start and end at block boundaries. 1695 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1696 */ 1697 origoffset = ap->a_offlo; 1698 origendoffset = ap->a_offhi; 1699 startoffset = origoffset & ~(fs->lfs_bmask); 1700 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1701 << fs->lfs_bshift; 1702 1703 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1704 endoffset = max_endoffset; 1705 origendoffset = endoffset; 1706 } else { 1707 origendoffset = round_page(ap->a_offhi); 1708 endoffset = round_page(blkroundup(fs, origendoffset)); 1709 } 1710 1711 KASSERT(startoffset > 0 || endoffset >= startoffset); 1712 if (startoffset == endoffset) { 1713 /* Nothing to do, why were we called? */ 1714 simple_unlock(&vp->v_interlock); 1715 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1716 PRId64 "\n", startoffset)); 1717 return 0; 1718 } 1719 1720 ap->a_offlo = startoffset; 1721 ap->a_offhi = endoffset; 1722 1723 if (!(ap->a_flags & PGO_CLEANIT)) 1724 return genfs_putpages(v); 1725 1726 /* 1727 * If there are more than one page per block, we don't want 1728 * to get caught locking them backwards; so set PGO_BUSYFAIL 1729 * to avoid deadlocks. 1730 */ 1731 ap->a_flags |= PGO_BUSYFAIL; 1732 1733 do { 1734 int r; 1735 1736 /* If no pages are dirty, we can just use genfs_putpages. */ 1737 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 1738 ap->a_flags, 1) != 0) 1739 break; 1740 1741 /* 1742 * Sometimes pages are dirtied between the time that 1743 * we check and the time we try to clean them. 1744 * Instruct lfs_gop_write to return EDEADLK in this case 1745 * so we can write them properly. 1746 */ 1747 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 1748 r = genfs_putpages(v); 1749 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 1750 if (r != EDEADLK) 1751 return r; 1752 1753 /* Start over. */ 1754 preempt(1); 1755 simple_lock(&vp->v_interlock); 1756 } while(1); 1757 1758 /* 1759 * Dirty and asked to clean. 1760 * 1761 * Pagedaemon can't actually write LFS pages; wake up 1762 * the writer to take care of that. The writer will 1763 * notice the pager inode queue and act on that. 1764 */ 1765 if (pagedaemon) { 1766 simple_lock(&fs->lfs_interlock); 1767 ++fs->lfs_pdflush; 1768 simple_unlock(&fs->lfs_interlock); 1769 wakeup(&lfs_writer_daemon); 1770 simple_unlock(&vp->v_interlock); 1771 return EWOULDBLOCK; 1772 } 1773 1774 /* 1775 * If this is a file created in a recent dirop, we can't flush its 1776 * inode until the dirop is complete. Drain dirops, then flush the 1777 * filesystem (taking care of any other pending dirops while we're 1778 * at it). 1779 */ 1780 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1781 (vp->v_flag & VDIROP)) { 1782 int locked; 1783 1784 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 1785 locked = VOP_ISLOCKED(vp) && /* XXX */ 1786 vp->v_lock.lk_lockholder == curproc->p_pid; 1787 simple_unlock(&vp->v_interlock); 1788 lfs_writer_enter(fs, "ppdirop"); 1789 if (locked) 1790 VOP_UNLOCK(vp, 0); 1791 1792 simple_lock(&fs->lfs_interlock); 1793 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1794 simple_unlock(&fs->lfs_interlock); 1795 1796 simple_lock(&vp->v_interlock); 1797 if (locked) 1798 VOP_LOCK(vp, LK_EXCLUSIVE); 1799 lfs_writer_leave(fs); 1800 1801 /* XXX the flush should have taken care of this one too! */ 1802 } 1803 1804 /* 1805 * This is it. We are going to write some pages. From here on 1806 * down it's all just mechanics. 1807 * 1808 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1809 */ 1810 ap->a_flags &= ~PGO_SYNCIO; 1811 1812 /* 1813 * If we've already got the seglock, flush the node and return. 1814 * The FIP has already been set up for us by lfs_writefile, 1815 * and FIP cleanup and lfs_updatemeta will also be done there, 1816 * unless genfs_putpages returns EDEADLK; then we must flush 1817 * what we have, and correct FIP and segment header accounting. 1818 */ 1819 1820 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 1821 if (!seglocked) { 1822 simple_unlock(&vp->v_interlock); 1823 /* 1824 * Take the seglock, because we are going to be writing pages. 1825 */ 1826 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 1827 if (error != 0) 1828 return error; 1829 simple_lock(&vp->v_interlock); 1830 } 1831 1832 /* 1833 * VOP_PUTPAGES should not be called while holding the seglock. 1834 * XXXUBC fix lfs_markv, or do this properly. 1835 */ 1836 #ifdef notyet 1837 KASSERT(fs->lfs_seglock == 1); 1838 #endif /* notyet */ 1839 1840 /* 1841 * We assume we're being called with sp->fip pointing at blank space. 1842 * Account for a new FIP in the segment header, and set sp->vp. 1843 * (This should duplicate the setup at the top of lfs_writefile().) 1844 */ 1845 sp = fs->lfs_sp; 1846 if (!seglocked) { 1847 if (sp->seg_bytes_left < fs->lfs_bsize || 1848 sp->sum_bytes_left < sizeof(struct finfo)) 1849 (void) lfs_writeseg(fs, fs->lfs_sp); 1850 1851 sp->sum_bytes_left -= FINFOSIZE; 1852 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1853 } 1854 KASSERT(sp->vp == NULL); 1855 sp->vp = vp; 1856 1857 if (!seglocked) { 1858 if (vp->v_flag & VDIROP) 1859 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 1860 } 1861 1862 sp->fip->fi_nblocks = 0; 1863 sp->fip->fi_ino = ip->i_number; 1864 sp->fip->fi_version = ip->i_gen; 1865 1866 /* 1867 * Loop through genfs_putpages until all pages are gathered. 1868 * genfs_putpages() drops the interlock, so reacquire it if necessary. 1869 * Whenever we lose the interlock we have to rerun check_dirty, as 1870 * well. 1871 */ 1872 again: 1873 check_dirty(fs, vp, startoffset, endoffset, blkeof, ap->a_flags, 0); 1874 1875 if ((error = genfs_putpages(v)) == EDEADLK) { 1876 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 1877 " EDEADLK [2] ino %d off %x (seg %d)\n", 1878 ip->i_number, fs->lfs_offset, 1879 dtosn(fs, fs->lfs_offset))); 1880 /* If nothing to write, short-circuit */ 1881 if (sp->cbpp - sp->bpp > 1) { 1882 /* Write gathered pages */ 1883 lfs_updatemeta(sp); 1884 (void) lfs_writeseg(fs, sp); 1885 1886 /* 1887 * Reinitialize brand new FIP and add us to it. 1888 * (This should duplicate the fixup in 1889 * lfs_gatherpages().) 1890 */ 1891 KASSERT(sp->vp == vp); 1892 sp->fip->fi_version = ip->i_gen; 1893 sp->fip->fi_ino = ip->i_number; 1894 /* Add us to the new segment summary. */ 1895 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1896 sp->sum_bytes_left -= FINFOSIZE; 1897 } 1898 1899 /* Give the write a chance to complete */ 1900 preempt(1); 1901 1902 /* We've lost the interlock. Start over. */ 1903 simple_lock(&vp->v_interlock); 1904 goto again; 1905 } 1906 1907 KASSERT(sp->vp == vp); 1908 if (!seglocked) { 1909 sp->vp = NULL; /* XXX lfs_gather below will set this */ 1910 1911 /* Write indirect blocks as well */ 1912 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 1913 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 1914 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 1915 1916 KASSERT(sp->vp == NULL); 1917 sp->vp = vp; 1918 } 1919 1920 /* 1921 * Blocks are now gathered into a segment waiting to be written. 1922 * All that's left to do is update metadata, and write them. 1923 */ 1924 lfs_updatemeta(sp); 1925 KASSERT(sp->vp == vp); 1926 sp->vp = NULL; 1927 1928 if (seglocked) { 1929 /* we're called by lfs_writefile. */ 1930 return error; 1931 } 1932 1933 /* 1934 * Clean up FIP, since we're done writing this file. 1935 * This should duplicate cleanup at the end of lfs_writefile(). 1936 */ 1937 if (sp->fip->fi_nblocks != 0) { 1938 sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE + 1939 sizeof(int32_t) * sp->fip->fi_nblocks); 1940 sp->start_lbp = &sp->fip->fi_blocks[0]; 1941 } else { 1942 sp->sum_bytes_left += FINFOSIZE; 1943 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 1944 } 1945 lfs_writeseg(fs, fs->lfs_sp); 1946 1947 /* 1948 * XXX - with the malloc/copy writeseg, the pages are freed by now 1949 * even if we don't wait (e.g. if we hold a nested lock). This 1950 * will not be true if we stop using malloc/copy. 1951 */ 1952 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 1953 lfs_segunlock(fs); 1954 1955 /* 1956 * Wait for v_numoutput to drop to zero. The seglock should 1957 * take care of this, but there is a slight possibility that 1958 * aiodoned might not have got around to our buffers yet. 1959 */ 1960 if (sync) { 1961 int s; 1962 1963 s = splbio(); 1964 simple_lock(&global_v_numoutput_slock); 1965 while (vp->v_numoutput > 0) { 1966 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 1967 " num %d\n", ip->i_number, vp->v_numoutput)); 1968 vp->v_flag |= VBWAIT; 1969 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 1970 &global_v_numoutput_slock); 1971 } 1972 simple_unlock(&global_v_numoutput_slock); 1973 splx(s); 1974 } 1975 return error; 1976 } 1977 1978 /* 1979 * Return the last logical file offset that should be written for this file 1980 * if we're doing a write that ends at "size". If writing, we need to know 1981 * about sizes on disk, i.e. fragments if there are any; if reading, we need 1982 * to know about entire blocks. 1983 */ 1984 void 1985 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 1986 { 1987 struct inode *ip = VTOI(vp); 1988 struct lfs *fs = ip->i_lfs; 1989 daddr_t olbn, nlbn; 1990 1991 KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)); 1992 KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)) 1993 != (GOP_SIZE_READ | GOP_SIZE_WRITE)); 1994 1995 olbn = lblkno(fs, ip->i_size); 1996 nlbn = lblkno(fs, size); 1997 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 1998 *eobp = fragroundup(fs, size); 1999 } else { 2000 *eobp = blkroundup(fs, size); 2001 } 2002 } 2003 2004 #ifdef DEBUG 2005 void lfs_dump_vop(void *); 2006 2007 void 2008 lfs_dump_vop(void *v) 2009 { 2010 struct vop_putpages_args /* { 2011 struct vnode *a_vp; 2012 voff_t a_offlo; 2013 voff_t a_offhi; 2014 int a_flags; 2015 } */ *ap = v; 2016 2017 #ifdef DDB 2018 vfs_vnode_print(ap->a_vp, 0, printf); 2019 #endif 2020 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2021 } 2022 #endif 2023 2024 int 2025 lfs_mmap(void *v) 2026 { 2027 struct vop_mmap_args /* { 2028 const struct vnodeop_desc *a_desc; 2029 struct vnode *a_vp; 2030 int a_fflags; 2031 struct ucred *a_cred; 2032 struct proc *a_p; 2033 } */ *ap = v; 2034 2035 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2036 return EOPNOTSUPP; 2037 return ufs_mmap(v); 2038 } 2039
Indexes created Tue Sep 30 11:09:46 GMT 2025