lfs_vnops.c revision 1.154
1 /* $NetBSD: lfs_vnops.c,v 1.154 2005/09/12 16:24:41 christos Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant (at) hhhh.org>. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 /* 39 * Copyright (c) 1986, 1989, 1991, 1993, 1995 40 * The Regents of the University of California. All rights reserved. 41 * 42 * Redistribution and use in source and binary forms, with or without 43 * modification, are permitted provided that the following conditions 44 * are met: 45 * 1. Redistributions of source code must retain the above copyright 46 * notice, this list of conditions and the following disclaimer. 47 * 2. Redistributions in binary form must reproduce the above copyright 48 * notice, this list of conditions and the following disclaimer in the 49 * documentation and/or other materials provided with the distribution. 50 * 3. Neither the name of the University nor the names of its contributors 51 * may be used to endorse or promote products derived from this software 52 * without specific prior written permission. 53 * 54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 64 * SUCH DAMAGE. 65 * 66 * @(#)lfs_vnops.c 8.13 (Berkeley) 6/10/95 67 */ 68 69 #include <sys/cdefs.h> 70 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.154 2005/09/12 16:24:41 christos Exp $"); 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/namei.h> 75 #include <sys/resourcevar.h> 76 #include <sys/kernel.h> 77 #include <sys/file.h> 78 #include <sys/stat.h> 79 #include <sys/buf.h> 80 #include <sys/proc.h> 81 #include <sys/mount.h> 82 #include <sys/vnode.h> 83 #include <sys/pool.h> 84 #include <sys/signalvar.h> 85 86 #include <miscfs/fifofs/fifo.h> 87 #include <miscfs/genfs/genfs.h> 88 #include <miscfs/specfs/specdev.h> 89 90 #include <ufs/ufs/inode.h> 91 #include <ufs/ufs/dir.h> 92 #include <ufs/ufs/ufsmount.h> 93 #include <ufs/ufs/ufs_extern.h> 94 95 #include <uvm/uvm.h> 96 #include <uvm/uvm_pmap.h> 97 #include <uvm/uvm_stat.h> 98 #include <uvm/uvm_pager.h> 99 100 #include <ufs/lfs/lfs.h> 101 #include <ufs/lfs/lfs_extern.h> 102 103 extern pid_t lfs_writer_daemon; 104 105 /* Global vfs data structures for lfs. */ 106 int (**lfs_vnodeop_p)(void *); 107 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 108 { &vop_default_desc, vn_default_error }, 109 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 110 { &vop_create_desc, lfs_create }, /* create */ 111 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 112 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 113 { &vop_open_desc, ufs_open }, /* open */ 114 { &vop_close_desc, lfs_close }, /* close */ 115 { &vop_access_desc, ufs_access }, /* access */ 116 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 117 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 118 { &vop_read_desc, lfs_read }, /* read */ 119 { &vop_write_desc, lfs_write }, /* write */ 120 { &vop_lease_desc, ufs_lease_check }, /* lease */ 121 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 122 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 123 { &vop_poll_desc, ufs_poll }, /* poll */ 124 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 125 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 126 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 127 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 128 { &vop_seek_desc, ufs_seek }, /* seek */ 129 { &vop_remove_desc, lfs_remove }, /* remove */ 130 { &vop_link_desc, lfs_link }, /* link */ 131 { &vop_rename_desc, lfs_rename }, /* rename */ 132 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 133 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 134 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 135 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 136 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 137 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 138 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 139 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 140 { &vop_lock_desc, ufs_lock }, /* lock */ 141 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 142 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 143 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 144 { &vop_print_desc, ufs_print }, /* print */ 145 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 146 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 147 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 148 { &vop_blkatoff_desc, lfs_blkatoff }, /* blkatoff */ 149 { &vop_valloc_desc, lfs_valloc }, /* valloc */ 150 { &vop_balloc_desc, lfs_balloc }, /* balloc */ 151 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 152 { &vop_truncate_desc, lfs_truncate }, /* truncate */ 153 { &vop_update_desc, lfs_update }, /* update */ 154 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 155 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 156 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 157 { NULL, NULL } 158 }; 159 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 160 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 161 162 int (**lfs_specop_p)(void *); 163 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 164 { &vop_default_desc, vn_default_error }, 165 { &vop_lookup_desc, spec_lookup }, /* lookup */ 166 { &vop_create_desc, spec_create }, /* create */ 167 { &vop_mknod_desc, spec_mknod }, /* mknod */ 168 { &vop_open_desc, spec_open }, /* open */ 169 { &vop_close_desc, lfsspec_close }, /* close */ 170 { &vop_access_desc, ufs_access }, /* access */ 171 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 172 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 173 { &vop_read_desc, ufsspec_read }, /* read */ 174 { &vop_write_desc, ufsspec_write }, /* write */ 175 { &vop_lease_desc, spec_lease_check }, /* lease */ 176 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 177 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 178 { &vop_poll_desc, spec_poll }, /* poll */ 179 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 180 { &vop_revoke_desc, spec_revoke }, /* revoke */ 181 { &vop_mmap_desc, spec_mmap }, /* mmap */ 182 { &vop_fsync_desc, spec_fsync }, /* fsync */ 183 { &vop_seek_desc, spec_seek }, /* seek */ 184 { &vop_remove_desc, spec_remove }, /* remove */ 185 { &vop_link_desc, spec_link }, /* link */ 186 { &vop_rename_desc, spec_rename }, /* rename */ 187 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 188 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 189 { &vop_symlink_desc, spec_symlink }, /* symlink */ 190 { &vop_readdir_desc, spec_readdir }, /* readdir */ 191 { &vop_readlink_desc, spec_readlink }, /* readlink */ 192 { &vop_abortop_desc, spec_abortop }, /* abortop */ 193 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 194 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 195 { &vop_lock_desc, ufs_lock }, /* lock */ 196 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 197 { &vop_bmap_desc, spec_bmap }, /* bmap */ 198 { &vop_strategy_desc, spec_strategy }, /* strategy */ 199 { &vop_print_desc, ufs_print }, /* print */ 200 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 201 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 202 { &vop_advlock_desc, spec_advlock }, /* advlock */ 203 { &vop_blkatoff_desc, spec_blkatoff }, /* blkatoff */ 204 { &vop_valloc_desc, spec_valloc }, /* valloc */ 205 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 206 { &vop_truncate_desc, spec_truncate }, /* truncate */ 207 { &vop_update_desc, lfs_update }, /* update */ 208 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 209 { &vop_getpages_desc, spec_getpages }, /* getpages */ 210 { &vop_putpages_desc, spec_putpages }, /* putpages */ 211 { NULL, NULL } 212 }; 213 const struct vnodeopv_desc lfs_specop_opv_desc = 214 { &lfs_specop_p, lfs_specop_entries }; 215 216 int (**lfs_fifoop_p)(void *); 217 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 218 { &vop_default_desc, vn_default_error }, 219 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 220 { &vop_create_desc, fifo_create }, /* create */ 221 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 222 { &vop_open_desc, fifo_open }, /* open */ 223 { &vop_close_desc, lfsfifo_close }, /* close */ 224 { &vop_access_desc, ufs_access }, /* access */ 225 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 226 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 227 { &vop_read_desc, ufsfifo_read }, /* read */ 228 { &vop_write_desc, ufsfifo_write }, /* write */ 229 { &vop_lease_desc, fifo_lease_check }, /* lease */ 230 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 231 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 232 { &vop_poll_desc, fifo_poll }, /* poll */ 233 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 234 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 235 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 236 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 237 { &vop_seek_desc, fifo_seek }, /* seek */ 238 { &vop_remove_desc, fifo_remove }, /* remove */ 239 { &vop_link_desc, fifo_link }, /* link */ 240 { &vop_rename_desc, fifo_rename }, /* rename */ 241 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 242 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 243 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 244 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 245 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 246 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 247 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 248 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 249 { &vop_lock_desc, ufs_lock }, /* lock */ 250 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 251 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 252 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 253 { &vop_print_desc, ufs_print }, /* print */ 254 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 255 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 256 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 257 { &vop_blkatoff_desc, fifo_blkatoff }, /* blkatoff */ 258 { &vop_valloc_desc, fifo_valloc }, /* valloc */ 259 { &vop_vfree_desc, lfs_vfree }, /* vfree */ 260 { &vop_truncate_desc, fifo_truncate }, /* truncate */ 261 { &vop_update_desc, lfs_update }, /* update */ 262 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 263 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 264 { NULL, NULL } 265 }; 266 const struct vnodeopv_desc lfs_fifoop_opv_desc = 267 { &lfs_fifoop_p, lfs_fifoop_entries }; 268 269 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int); 270 271 void 272 lfs_itimes(struct inode *ip, const struct timespec *acc, 273 const struct timespec *mod, const struct timespec *cre) 274 { 275 struct timespec *ts = NULL, tsb; 276 277 KASSERT(ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY)); 278 279 if (ip->i_flag & IN_ACCESS) { 280 if (acc == NULL) 281 acc = ts == NULL ? (ts = nanotime(&tsb)) : ts; 282 ip->i_ffs1_atime = acc->tv_sec; 283 ip->i_ffs1_atimensec = acc->tv_nsec; 284 if (ip->i_lfs->lfs_version > 1) { 285 struct lfs *fs = ip->i_lfs; 286 struct buf *ibp; 287 IFILE *ifp; 288 289 LFS_IENTRY(ifp, ip->i_lfs, ip->i_number, ibp); 290 ifp->if_atime_sec = acc->tv_sec; 291 ifp->if_atime_nsec = acc->tv_nsec; 292 LFS_BWRITE_LOG(ibp); 293 simple_lock(&fs->lfs_interlock); 294 fs->lfs_flags |= LFS_IFDIRTY; 295 simple_unlock(&fs->lfs_interlock); 296 } else { 297 LFS_SET_UINO(ip, IN_ACCESSED); 298 } 299 } 300 if (ip->i_flag & (IN_CHANGE | IN_UPDATE | IN_MODIFY)) { 301 if (ip->i_flag & (IN_UPDATE | IN_MODIFY)) { 302 if (mod == NULL) 303 mod = ts == NULL ? (ts = nanotime(&tsb)) : ts; 304 ip->i_ffs1_mtime = mod->tv_sec; 305 ip->i_ffs1_mtimensec = mod->tv_nsec; 306 ip->i_modrev++; 307 } 308 if (ip->i_flag & (IN_CHANGE | IN_MODIFY)) { 309 if (cre == NULL) 310 cre = ts == NULL ? (ts = nanotime(&tsb)) : ts; 311 ip->i_ffs1_ctime = cre->tv_sec; 312 ip->i_ffs1_ctimensec = cre->tv_nsec; 313 } 314 if (ip->i_flag & (IN_CHANGE | IN_UPDATE)) 315 LFS_SET_UINO(ip, IN_MODIFIED); 316 if (ip->i_flag & IN_MODIFY) 317 LFS_SET_UINO(ip, IN_ACCESSED); 318 } 319 ip->i_flag &= ~(IN_ACCESS | IN_CHANGE | IN_UPDATE | IN_MODIFY); 320 } 321 322 #define LFS_READWRITE 323 #include <ufs/ufs/ufs_readwrite.c> 324 #undef LFS_READWRITE 325 326 /* 327 * Synch an open file. 328 */ 329 /* ARGSUSED */ 330 int 331 lfs_fsync(void *v) 332 { 333 struct vop_fsync_args /* { 334 struct vnode *a_vp; 335 struct ucred *a_cred; 336 int a_flags; 337 off_t offlo; 338 off_t offhi; 339 struct proc *a_p; 340 } */ *ap = v; 341 struct vnode *vp = ap->a_vp; 342 int error, wait; 343 344 /* 345 * Trickle sync checks for need to do a checkpoint after possible 346 * activity from the pagedaemon. 347 */ 348 if (ap->a_flags & FSYNC_LAZY) { 349 simple_lock(&lfs_subsys_lock); 350 wakeup(&lfs_writer_daemon); 351 simple_unlock(&lfs_subsys_lock); 352 return 0; 353 } 354 355 wait = (ap->a_flags & FSYNC_WAIT); 356 simple_lock(&vp->v_interlock); 357 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 358 round_page(ap->a_offhi), 359 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 360 if (error) 361 return error; 362 error = VOP_UPDATE(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 363 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 364 int l = 0; 365 error = VOP_IOCTL(VTOI(vp)->i_devvp, DIOCCACHESYNC, &l, FWRITE, 366 ap->a_p->p_ucred, ap->a_p); 367 } 368 if (wait && !VPISEMPTY(vp)) 369 LFS_SET_UINO(VTOI(vp), IN_MODIFIED); 370 371 return error; 372 } 373 374 /* 375 * Take IN_ADIROP off, then call ufs_inactive. 376 */ 377 int 378 lfs_inactive(void *v) 379 { 380 struct vop_inactive_args /* { 381 struct vnode *a_vp; 382 struct proc *a_p; 383 } */ *ap = v; 384 385 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 386 387 lfs_unmark_vnode(ap->a_vp); 388 389 /* 390 * The Ifile is only ever inactivated on unmount. 391 * Streamline this process by not giving it more dirty blocks. 392 */ 393 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 394 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 395 VOP_UNLOCK(ap->a_vp, 0); 396 return 0; 397 } 398 399 return ufs_inactive(v); 400 } 401 402 /* 403 * These macros are used to bracket UFS directory ops, so that we can 404 * identify all the pages touched during directory ops which need to 405 * be ordered and flushed atomically, so that they may be recovered. 406 * 407 * Because we have to mark nodes VDIROP in order to prevent 408 * the cache from reclaiming them while a dirop is in progress, we must 409 * also manage the number of nodes so marked (otherwise we can run out). 410 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 411 * is decremented during segment write, when VDIROP is taken off. 412 */ 413 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 414 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 415 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 416 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 417 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 418 static int lfs_set_dirop(struct vnode *, struct vnode *); 419 420 static int 421 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 422 { 423 struct lfs *fs; 424 int error; 425 426 KASSERT(VOP_ISLOCKED(dvp)); 427 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 428 429 fs = VTOI(dvp)->i_lfs; 430 431 ASSERT_NO_SEGLOCK(fs); 432 /* 433 * LFS_NRESERVE calculates direct and indirect blocks as well 434 * as an inode block; an overestimate in most cases. 435 */ 436 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 437 return (error); 438 439 restart: 440 simple_lock(&fs->lfs_interlock); 441 if (fs->lfs_dirops == 0) { 442 simple_unlock(&fs->lfs_interlock); 443 lfs_check(dvp, LFS_UNUSED_LBN, 0); 444 simple_lock(&fs->lfs_interlock); 445 } 446 while (fs->lfs_writer) 447 ltsleep(&fs->lfs_dirops, (PRIBIO + 1), "lfs_sdirop", 0, 448 &fs->lfs_interlock); 449 simple_lock(&lfs_subsys_lock); 450 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 451 wakeup(&lfs_writer_daemon); 452 simple_unlock(&lfs_subsys_lock); 453 simple_unlock(&fs->lfs_interlock); 454 preempt(1); 455 goto restart; 456 } 457 458 if (lfs_dirvcount > LFS_MAX_DIROP) { 459 simple_unlock(&fs->lfs_interlock); 460 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 461 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 462 if ((error = ltsleep(&lfs_dirvcount, 463 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 464 &lfs_subsys_lock)) != 0) { 465 goto unreserve; 466 } 467 goto restart; 468 } 469 simple_unlock(&lfs_subsys_lock); 470 471 ++fs->lfs_dirops; 472 fs->lfs_doifile = 1; 473 simple_unlock(&fs->lfs_interlock); 474 475 /* Hold a reference so SET_ENDOP will be happy */ 476 vref(dvp); 477 if (vp) { 478 vref(vp); 479 MARK_VNODE(vp); 480 } 481 482 MARK_VNODE(dvp); 483 return 0; 484 485 unreserve: 486 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 487 return error; 488 } 489 490 /* 491 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 492 * in getnewvnode(), if we have a stacked filesystem mounted on top 493 * of us. 494 * 495 * NB: this means we have to clear the new vnodes on error. Fortunately 496 * SET_ENDOP is there to do that for us. 497 */ 498 static int 499 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 500 { 501 int error; 502 struct lfs *fs; 503 504 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 505 ASSERT_NO_SEGLOCK(fs); 506 if (fs->lfs_ronly) 507 return EROFS; 508 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 509 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 510 dvp, error)); 511 return error; 512 } 513 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 514 if (vpp) { 515 ungetnewvnode(*vpp); 516 *vpp = NULL; 517 } 518 return error; 519 } 520 return 0; 521 } 522 523 #define SET_ENDOP_BASE(fs, dvp, str) \ 524 do { \ 525 simple_lock(&(fs)->lfs_interlock); \ 526 --(fs)->lfs_dirops; \ 527 if (!(fs)->lfs_dirops) { \ 528 if ((fs)->lfs_nadirop) { \ 529 panic("SET_ENDOP: %s: no dirops but " \ 530 " nadirop=%d", (str), \ 531 (fs)->lfs_nadirop); \ 532 } \ 533 wakeup(&(fs)->lfs_writer); \ 534 simple_unlock(&(fs)->lfs_interlock); \ 535 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 536 } else \ 537 simple_unlock(&(fs)->lfs_interlock); \ 538 } while(0) 539 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 540 do { \ 541 UNMARK_VNODE(dvp); \ 542 if (nvpp && *nvpp) \ 543 UNMARK_VNODE(*nvpp); \ 544 /* Check for error return to stem vnode leakage */ \ 545 if (nvpp && *nvpp && !((*nvpp)->v_flag & VDIROP)) \ 546 ungetnewvnode(*(nvpp)); \ 547 SET_ENDOP_BASE((fs), (dvp), (str)); \ 548 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 549 vrele(dvp); \ 550 } while(0) 551 #define SET_ENDOP_CREATE_AP(ap, str) \ 552 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 553 (ap)->a_vpp, (str)) 554 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 555 do { \ 556 UNMARK_VNODE(dvp); \ 557 if (ovp) \ 558 UNMARK_VNODE(ovp); \ 559 SET_ENDOP_BASE((fs), (dvp), (str)); \ 560 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 561 vrele(dvp); \ 562 if (ovp) \ 563 vrele(ovp); \ 564 } while(0) 565 566 void 567 lfs_mark_vnode(struct vnode *vp) 568 { 569 struct inode *ip = VTOI(vp); 570 struct lfs *fs = ip->i_lfs; 571 572 simple_lock(&fs->lfs_interlock); 573 if (!(ip->i_flag & IN_ADIROP)) { 574 if (!(vp->v_flag & VDIROP)) { 575 (void)lfs_vref(vp); 576 simple_lock(&lfs_subsys_lock); 577 ++lfs_dirvcount; 578 simple_unlock(&lfs_subsys_lock); 579 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 580 vp->v_flag |= VDIROP; 581 } 582 ++fs->lfs_nadirop; 583 ip->i_flag |= IN_ADIROP; 584 } else 585 KASSERT(vp->v_flag & VDIROP); 586 simple_unlock(&fs->lfs_interlock); 587 } 588 589 void 590 lfs_unmark_vnode(struct vnode *vp) 591 { 592 struct inode *ip = VTOI(vp); 593 594 if (ip && (ip->i_flag & IN_ADIROP)) { 595 KASSERT(vp->v_flag & VDIROP); 596 simple_lock(&ip->i_lfs->lfs_interlock); 597 --ip->i_lfs->lfs_nadirop; 598 simple_unlock(&ip->i_lfs->lfs_interlock); 599 ip->i_flag &= ~IN_ADIROP; 600 } 601 } 602 603 int 604 lfs_symlink(void *v) 605 { 606 struct vop_symlink_args /* { 607 struct vnode *a_dvp; 608 struct vnode **a_vpp; 609 struct componentname *a_cnp; 610 struct vattr *a_vap; 611 char *a_target; 612 } */ *ap = v; 613 int error; 614 615 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 616 vput(ap->a_dvp); 617 return error; 618 } 619 error = ufs_symlink(ap); 620 SET_ENDOP_CREATE_AP(ap, "symlink"); 621 return (error); 622 } 623 624 int 625 lfs_mknod(void *v) 626 { 627 struct vop_mknod_args /* { 628 struct vnode *a_dvp; 629 struct vnode **a_vpp; 630 struct componentname *a_cnp; 631 struct vattr *a_vap; 632 } */ *ap = v; 633 struct vattr *vap = ap->a_vap; 634 struct vnode **vpp = ap->a_vpp; 635 struct inode *ip; 636 int error; 637 struct mount *mp; 638 ino_t ino; 639 640 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 641 vput(ap->a_dvp); 642 return error; 643 } 644 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 645 ap->a_dvp, vpp, ap->a_cnp); 646 647 /* Either way we're done with the dirop at this point */ 648 SET_ENDOP_CREATE_AP(ap, "mknod"); 649 650 if (error) 651 return (error); 652 653 ip = VTOI(*vpp); 654 mp = (*vpp)->v_mount; 655 ino = ip->i_number; 656 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 657 if (vap->va_rdev != VNOVAL) { 658 /* 659 * Want to be able to use this to make badblock 660 * inodes, so don't truncate the dev number. 661 */ 662 #if 0 663 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 664 UFS_MPNEEDSWAP((*vpp)->v_mount)); 665 #else 666 ip->i_ffs1_rdev = vap->va_rdev; 667 #endif 668 } 669 670 /* 671 * Call fsync to write the vnode so that we don't have to deal with 672 * flushing it when it's marked VDIROP|VXLOCK. 673 * 674 * XXX KS - If we can't flush we also can't call vgone(), so must 675 * return. But, that leaves this vnode in limbo, also not good. 676 * Can this ever happen (barring hardware failure)? 677 */ 678 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 679 curproc)) != 0) { 680 panic("lfs_mknod: couldn't fsync (ino %llu)", 681 (unsigned long long)ino); 682 /* return (error); */ 683 } 684 /* 685 * Remove vnode so that it will be reloaded by VFS_VGET and 686 * checked to see if it is an alias of an existing entry in 687 * the inode cache. 688 */ 689 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 690 691 VOP_UNLOCK(*vpp, 0); 692 lfs_vunref(*vpp); 693 (*vpp)->v_type = VNON; 694 vgone(*vpp); 695 error = VFS_VGET(mp, ino, vpp); 696 697 if (error != 0) { 698 *vpp = NULL; 699 return (error); 700 } 701 return (0); 702 } 703 704 int 705 lfs_create(void *v) 706 { 707 struct vop_create_args /* { 708 struct vnode *a_dvp; 709 struct vnode **a_vpp; 710 struct componentname *a_cnp; 711 struct vattr *a_vap; 712 } */ *ap = v; 713 int error; 714 715 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 716 vput(ap->a_dvp); 717 return error; 718 } 719 error = ufs_create(ap); 720 SET_ENDOP_CREATE_AP(ap, "create"); 721 return (error); 722 } 723 724 int 725 lfs_mkdir(void *v) 726 { 727 struct vop_mkdir_args /* { 728 struct vnode *a_dvp; 729 struct vnode **a_vpp; 730 struct componentname *a_cnp; 731 struct vattr *a_vap; 732 } */ *ap = v; 733 int error; 734 735 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 736 vput(ap->a_dvp); 737 return error; 738 } 739 error = ufs_mkdir(ap); 740 SET_ENDOP_CREATE_AP(ap, "mkdir"); 741 return (error); 742 } 743 744 int 745 lfs_remove(void *v) 746 { 747 struct vop_remove_args /* { 748 struct vnode *a_dvp; 749 struct vnode *a_vp; 750 struct componentname *a_cnp; 751 } */ *ap = v; 752 struct vnode *dvp, *vp; 753 int error; 754 755 dvp = ap->a_dvp; 756 vp = ap->a_vp; 757 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 758 if (dvp == vp) 759 vrele(vp); 760 else 761 vput(vp); 762 vput(dvp); 763 return error; 764 } 765 error = ufs_remove(ap); 766 SET_ENDOP_REMOVE(VTOI(dvp)->i_lfs, dvp, vp, "remove"); 767 return (error); 768 } 769 770 int 771 lfs_rmdir(void *v) 772 { 773 struct vop_rmdir_args /* { 774 struct vnodeop_desc *a_desc; 775 struct vnode *a_dvp; 776 struct vnode *a_vp; 777 struct componentname *a_cnp; 778 } */ *ap = v; 779 struct vnode *vp; 780 int error; 781 782 vp = ap->a_vp; 783 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 784 vrele(ap->a_dvp); 785 if (ap->a_vp != ap->a_dvp) 786 VOP_UNLOCK(ap->a_dvp, 0); 787 vput(vp); 788 return error; 789 } 790 error = ufs_rmdir(ap); 791 SET_ENDOP_REMOVE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vp, "rmdir"); 792 return (error); 793 } 794 795 int 796 lfs_link(void *v) 797 { 798 struct vop_link_args /* { 799 struct vnode *a_dvp; 800 struct vnode *a_vp; 801 struct componentname *a_cnp; 802 } */ *ap = v; 803 int error; 804 struct vnode **vpp = NULL; 805 806 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 807 vput(ap->a_dvp); 808 return error; 809 } 810 error = ufs_link(ap); 811 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 812 return (error); 813 } 814 815 int 816 lfs_rename(void *v) 817 { 818 struct vop_rename_args /* { 819 struct vnode *a_fdvp; 820 struct vnode *a_fvp; 821 struct componentname *a_fcnp; 822 struct vnode *a_tdvp; 823 struct vnode *a_tvp; 824 struct componentname *a_tcnp; 825 } */ *ap = v; 826 struct vnode *tvp, *fvp, *tdvp, *fdvp; 827 struct componentname *tcnp, *fcnp; 828 int error; 829 struct lfs *fs; 830 831 fs = VTOI(ap->a_fdvp)->i_lfs; 832 tvp = ap->a_tvp; 833 tdvp = ap->a_tdvp; 834 tcnp = ap->a_tcnp; 835 fvp = ap->a_fvp; 836 fdvp = ap->a_fdvp; 837 fcnp = ap->a_fcnp; 838 839 /* 840 * Check for cross-device rename. 841 * If it is, we don't want to set dirops, just error out. 842 * (In particular note that MARK_VNODE(tdvp) will DTWT on 843 * a cross-device rename.) 844 * 845 * Copied from ufs_rename. 846 */ 847 if ((fvp->v_mount != tdvp->v_mount) || 848 (tvp && (fvp->v_mount != tvp->v_mount))) { 849 error = EXDEV; 850 goto errout; 851 } 852 853 /* 854 * Check to make sure we're not renaming a vnode onto itself 855 * (deleting a hard link by renaming one name onto another); 856 * if we are we can't recursively call VOP_REMOVE since that 857 * would leave us with an unaccounted-for number of live dirops. 858 * 859 * Inline the relevant section of ufs_rename here, *before* 860 * calling SET_DIROP_REMOVE. 861 */ 862 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 863 (VTOI(tdvp)->i_flags & APPEND))) { 864 error = EPERM; 865 goto errout; 866 } 867 if (fvp == tvp) { 868 if (fvp->v_type == VDIR) { 869 error = EINVAL; 870 goto errout; 871 } 872 873 /* Release destination completely. */ 874 VOP_ABORTOP(tdvp, tcnp); 875 vput(tdvp); 876 vput(tvp); 877 878 /* Delete source. */ 879 vrele(fvp); 880 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 881 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 882 fcnp->cn_nameiop = DELETE; 883 if ((error = relookup(fdvp, &fvp, fcnp))){ 884 /* relookup blew away fdvp */ 885 return (error); 886 } 887 return (VOP_REMOVE(fdvp, fvp, fcnp)); 888 } 889 890 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 891 goto errout; 892 MARK_VNODE(fdvp); 893 MARK_VNODE(fvp); 894 895 error = ufs_rename(ap); 896 UNMARK_VNODE(fdvp); 897 UNMARK_VNODE(fvp); 898 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 899 return (error); 900 901 errout: 902 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 903 if (tdvp == tvp) 904 vrele(tdvp); 905 else 906 vput(tdvp); 907 if (tvp) 908 vput(tvp); 909 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 910 vrele(fdvp); 911 vrele(fvp); 912 return (error); 913 } 914 915 /* XXX hack to avoid calling ITIMES in getattr */ 916 int 917 lfs_getattr(void *v) 918 { 919 struct vop_getattr_args /* { 920 struct vnode *a_vp; 921 struct vattr *a_vap; 922 struct ucred *a_cred; 923 struct proc *a_p; 924 } */ *ap = v; 925 struct vnode *vp = ap->a_vp; 926 struct inode *ip = VTOI(vp); 927 struct vattr *vap = ap->a_vap; 928 struct lfs *fs = ip->i_lfs; 929 /* 930 * Copy from inode table 931 */ 932 vap->va_fsid = ip->i_dev; 933 vap->va_fileid = ip->i_number; 934 vap->va_mode = ip->i_mode & ~IFMT; 935 vap->va_nlink = ip->i_nlink; 936 vap->va_uid = ip->i_uid; 937 vap->va_gid = ip->i_gid; 938 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 939 vap->va_size = vp->v_size; 940 vap->va_atime.tv_sec = ip->i_ffs1_atime; 941 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 942 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 943 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 944 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 945 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 946 vap->va_flags = ip->i_flags; 947 vap->va_gen = ip->i_gen; 948 /* this doesn't belong here */ 949 if (vp->v_type == VBLK) 950 vap->va_blocksize = BLKDEV_IOSIZE; 951 else if (vp->v_type == VCHR) 952 vap->va_blocksize = MAXBSIZE; 953 else 954 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 955 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 956 vap->va_type = vp->v_type; 957 vap->va_filerev = ip->i_modrev; 958 return (0); 959 } 960 961 /* 962 * Check to make sure the inode blocks won't choke the buffer 963 * cache, then call ufs_setattr as usual. 964 */ 965 int 966 lfs_setattr(void *v) 967 { 968 struct vop_setattr_args /* { 969 struct vnode *a_vp; 970 struct vattr *a_vap; 971 struct ucred *a_cred; 972 struct proc *a_p; 973 } */ *ap = v; 974 struct vnode *vp = ap->a_vp; 975 976 lfs_check(vp, LFS_UNUSED_LBN, 0); 977 return ufs_setattr(v); 978 } 979 980 /* 981 * Close called 982 * 983 * XXX -- we were using ufs_close, but since it updates the 984 * times on the inode, we might need to bump the uinodes 985 * count. 986 */ 987 /* ARGSUSED */ 988 int 989 lfs_close(void *v) 990 { 991 struct vop_close_args /* { 992 struct vnode *a_vp; 993 int a_fflag; 994 struct ucred *a_cred; 995 struct proc *a_p; 996 } */ *ap = v; 997 struct vnode *vp = ap->a_vp; 998 struct inode *ip = VTOI(vp); 999 1000 if (vp == ip->i_lfs->lfs_ivnode && 1001 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1002 return 0; 1003 1004 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1005 LFS_ITIMES(ip, NULL, NULL, NULL); 1006 } 1007 return (0); 1008 } 1009 1010 /* 1011 * Close wrapper for special devices. 1012 * 1013 * Update the times on the inode then do device close. 1014 */ 1015 int 1016 lfsspec_close(void *v) 1017 { 1018 struct vop_close_args /* { 1019 struct vnode *a_vp; 1020 int a_fflag; 1021 struct ucred *a_cred; 1022 struct proc *a_p; 1023 } */ *ap = v; 1024 struct vnode *vp; 1025 struct inode *ip; 1026 1027 vp = ap->a_vp; 1028 ip = VTOI(vp); 1029 if (vp->v_usecount > 1) { 1030 LFS_ITIMES(ip, NULL, NULL, NULL); 1031 } 1032 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1033 } 1034 1035 /* 1036 * Close wrapper for fifo's. 1037 * 1038 * Update the times on the inode then do device close. 1039 */ 1040 int 1041 lfsfifo_close(void *v) 1042 { 1043 struct vop_close_args /* { 1044 struct vnode *a_vp; 1045 int a_fflag; 1046 struct ucred *a_cred; 1047 struct proc *a_p; 1048 } */ *ap = v; 1049 struct vnode *vp; 1050 struct inode *ip; 1051 1052 vp = ap->a_vp; 1053 ip = VTOI(vp); 1054 if (ap->a_vp->v_usecount > 1) { 1055 LFS_ITIMES(ip, NULL, NULL, NULL); 1056 } 1057 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1058 } 1059 1060 /* 1061 * Reclaim an inode so that it can be used for other purposes. 1062 */ 1063 1064 int 1065 lfs_reclaim(void *v) 1066 { 1067 struct vop_reclaim_args /* { 1068 struct vnode *a_vp; 1069 struct proc *a_p; 1070 } */ *ap = v; 1071 struct vnode *vp = ap->a_vp; 1072 struct inode *ip = VTOI(vp); 1073 int error; 1074 1075 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1076 1077 LFS_CLR_UINO(ip, IN_ALLMOD); 1078 if ((error = ufs_reclaim(vp, ap->a_p))) 1079 return (error); 1080 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1081 lfs_deregister_all(vp); 1082 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1083 ip->inode_ext.lfs = NULL; 1084 pool_put(&lfs_inode_pool, vp->v_data); 1085 vp->v_data = NULL; 1086 return (0); 1087 } 1088 1089 /* 1090 * Read a block from a storage device. 1091 * In order to avoid reading blocks that are in the process of being 1092 * written by the cleaner---and hence are not mutexed by the normal 1093 * buffer cache / page cache mechanisms---check for collisions before 1094 * reading. 1095 * 1096 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1097 * the active cleaner test. 1098 * 1099 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1100 */ 1101 int 1102 lfs_strategy(void *v) 1103 { 1104 struct vop_strategy_args /* { 1105 struct vnode *a_vp; 1106 struct buf *a_bp; 1107 } */ *ap = v; 1108 struct buf *bp; 1109 struct lfs *fs; 1110 struct vnode *vp; 1111 struct inode *ip; 1112 daddr_t tbn; 1113 int i, sn, error, slept; 1114 1115 bp = ap->a_bp; 1116 vp = ap->a_vp; 1117 ip = VTOI(vp); 1118 fs = ip->i_lfs; 1119 1120 /* lfs uses its strategy routine only for read */ 1121 KASSERT(bp->b_flags & B_READ); 1122 1123 if (vp->v_type == VBLK || vp->v_type == VCHR) 1124 panic("lfs_strategy: spec"); 1125 KASSERT(bp->b_bcount != 0); 1126 if (bp->b_blkno == bp->b_lblkno) { 1127 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1128 NULL); 1129 if (error) { 1130 bp->b_error = error; 1131 bp->b_flags |= B_ERROR; 1132 biodone(bp); 1133 return (error); 1134 } 1135 if ((long)bp->b_blkno == -1) /* no valid data */ 1136 clrbuf(bp); 1137 } 1138 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1139 biodone(bp); 1140 return (0); 1141 } 1142 1143 slept = 1; 1144 simple_lock(&fs->lfs_interlock); 1145 while (slept && fs->lfs_seglock) { 1146 simple_unlock(&fs->lfs_interlock); 1147 /* 1148 * Look through list of intervals. 1149 * There will only be intervals to look through 1150 * if the cleaner holds the seglock. 1151 * Since the cleaner is synchronous, we can trust 1152 * the list of intervals to be current. 1153 */ 1154 tbn = dbtofsb(fs, bp->b_blkno); 1155 sn = dtosn(fs, tbn); 1156 slept = 0; 1157 for (i = 0; i < fs->lfs_cleanind; i++) { 1158 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1159 tbn >= fs->lfs_cleanint[i]) { 1160 DLOG((DLOG_CLEAN, 1161 "lfs_strategy: ino %d lbn %" PRId64 1162 " ind %d sn %d fsb %" PRIx32 1163 " given sn %d fsb %" PRIx64 "\n", 1164 ip->i_number, bp->b_lblkno, i, 1165 dtosn(fs, fs->lfs_cleanint[i]), 1166 fs->lfs_cleanint[i], sn, tbn)); 1167 DLOG((DLOG_CLEAN, 1168 "lfs_strategy: sleeping on ino %d lbn %" 1169 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1170 simple_lock(&fs->lfs_interlock); 1171 if (fs->lfs_seglock) 1172 ltsleep(&fs->lfs_seglock, 1173 (PRIBIO + 1) | PNORELOCK, 1174 "lfs_strategy", 0, 1175 &fs->lfs_interlock); 1176 /* Things may be different now; start over. */ 1177 slept = 1; 1178 break; 1179 } 1180 } 1181 simple_lock(&fs->lfs_interlock); 1182 } 1183 simple_unlock(&fs->lfs_interlock); 1184 1185 vp = ip->i_devvp; 1186 VOP_STRATEGY(vp, bp); 1187 return (0); 1188 } 1189 1190 static void 1191 lfs_flush_dirops(struct lfs *fs) 1192 { 1193 struct inode *ip, *nip; 1194 struct vnode *vp; 1195 extern int lfs_dostats; 1196 struct segment *sp; 1197 int needunlock; 1198 1199 ASSERT_NO_SEGLOCK(fs); 1200 1201 if (fs->lfs_ronly) 1202 return; 1203 1204 simple_lock(&fs->lfs_interlock); 1205 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1206 simple_unlock(&fs->lfs_interlock); 1207 return; 1208 } else 1209 simple_unlock(&fs->lfs_interlock); 1210 1211 if (lfs_dostats) 1212 ++lfs_stats.flush_invoked; 1213 1214 /* 1215 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1216 * Technically this is a checkpoint (the on-disk state is valid) 1217 * even though we are leaving out all the file data. 1218 */ 1219 lfs_imtime(fs); 1220 lfs_seglock(fs, SEGM_CKP); 1221 sp = fs->lfs_sp; 1222 1223 /* 1224 * lfs_writevnodes, optimized to get dirops out of the way. 1225 * Only write dirops, and don't flush files' pages, only 1226 * blocks from the directories. 1227 * 1228 * We don't need to vref these files because they are 1229 * dirops and so hold an extra reference until the 1230 * segunlock clears them of that status. 1231 * 1232 * We don't need to check for IN_ADIROP because we know that 1233 * no dirops are active. 1234 * 1235 */ 1236 simple_lock(&fs->lfs_interlock); 1237 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1238 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1239 simple_unlock(&fs->lfs_interlock); 1240 vp = ITOV(ip); 1241 1242 /* 1243 * All writes to directories come from dirops; all 1244 * writes to files' direct blocks go through the page 1245 * cache, which we're not touching. Reads to files 1246 * and/or directories will not be affected by writing 1247 * directory blocks inodes and file inodes. So we don't 1248 * really need to lock. If we don't lock, though, 1249 * make sure that we don't clear IN_MODIFIED 1250 * unnecessarily. 1251 */ 1252 if (vp->v_flag & VXLOCK) 1253 continue; 1254 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) { 1255 needunlock = 1; 1256 } else { 1257 DLOG((DLOG_VNODE, "lfs_flush_dirops: flushing locked ino %d\n", 1258 VTOI(vp)->i_number)); 1259 needunlock = 0; 1260 } 1261 if (vp->v_type != VREG && 1262 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1263 lfs_writefile(fs, sp, vp); 1264 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1265 !(ip->i_flag & IN_ALLMOD)) { 1266 LFS_SET_UINO(ip, IN_MODIFIED); 1267 } 1268 } 1269 (void) lfs_writeinode(fs, sp, ip); 1270 if (needunlock) 1271 VOP_UNLOCK(vp, 0); 1272 else 1273 LFS_SET_UINO(ip, IN_MODIFIED); 1274 simple_lock(&fs->lfs_interlock); 1275 } 1276 simple_unlock(&fs->lfs_interlock); 1277 /* We've written all the dirops there are */ 1278 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1279 (void) lfs_writeseg(fs, sp); 1280 lfs_segunlock(fs); 1281 } 1282 1283 /* 1284 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1285 */ 1286 int 1287 lfs_fcntl(void *v) 1288 { 1289 struct vop_fcntl_args /* { 1290 struct vnode *a_vp; 1291 u_long a_command; 1292 caddr_t a_data; 1293 int a_fflag; 1294 struct ucred *a_cred; 1295 struct proc *a_p; 1296 } */ *ap = v; 1297 struct timeval *tvp; 1298 BLOCK_INFO *blkiov; 1299 CLEANERINFO *cip; 1300 SEGUSE *sup; 1301 int blkcnt, error, oclean; 1302 struct lfs_fcntl_markv blkvp; 1303 fsid_t *fsidp; 1304 struct lfs *fs; 1305 struct buf *bp; 1306 fhandle_t *fhp; 1307 daddr_t off; 1308 1309 /* Only respect LFS fcntls on fs root or Ifile */ 1310 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1311 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1312 return ufs_fcntl(v); 1313 } 1314 1315 /* Avoid locking a draining lock */ 1316 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1317 return ESHUTDOWN; 1318 } 1319 1320 fs = VTOI(ap->a_vp)->i_lfs; 1321 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1322 1323 switch (ap->a_command) { 1324 case LFCNSEGWAITALL: 1325 case LFCNSEGWAITALL_COMPAT: 1326 fsidp = NULL; 1327 /* FALLSTHROUGH */ 1328 case LFCNSEGWAIT: 1329 case LFCNSEGWAIT_COMPAT: 1330 tvp = (struct timeval *)ap->a_data; 1331 simple_lock(&fs->lfs_interlock); 1332 ++fs->lfs_sleepers; 1333 simple_unlock(&fs->lfs_interlock); 1334 VOP_UNLOCK(ap->a_vp, 0); 1335 1336 error = lfs_segwait(fsidp, tvp); 1337 1338 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1339 simple_lock(&fs->lfs_interlock); 1340 if (--fs->lfs_sleepers == 0) 1341 wakeup(&fs->lfs_sleepers); 1342 simple_unlock(&fs->lfs_interlock); 1343 return error; 1344 1345 case LFCNBMAPV: 1346 case LFCNMARKV: 1347 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1348 return (error); 1349 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1350 1351 blkcnt = blkvp.blkcnt; 1352 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1353 return (EINVAL); 1354 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1355 if ((error = copyin(blkvp.blkiov, blkiov, 1356 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1357 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1358 return error; 1359 } 1360 1361 simple_lock(&fs->lfs_interlock); 1362 ++fs->lfs_sleepers; 1363 simple_unlock(&fs->lfs_interlock); 1364 VOP_UNLOCK(ap->a_vp, 0); 1365 if (ap->a_command == LFCNBMAPV) 1366 error = lfs_bmapv(ap->a_p, fsidp, blkiov, blkcnt); 1367 else /* LFCNMARKV */ 1368 error = lfs_markv(ap->a_p, fsidp, blkiov, blkcnt); 1369 if (error == 0) 1370 error = copyout(blkiov, blkvp.blkiov, 1371 blkcnt * sizeof(BLOCK_INFO)); 1372 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1373 simple_lock(&fs->lfs_interlock); 1374 if (--fs->lfs_sleepers == 0) 1375 wakeup(&fs->lfs_sleepers); 1376 simple_unlock(&fs->lfs_interlock); 1377 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1378 return error; 1379 1380 case LFCNRECLAIM: 1381 /* 1382 * Flush dirops and write Ifile, allowing empty segments 1383 * to be immediately reclaimed. 1384 */ 1385 VOP_UNLOCK(ap->a_vp, 0); 1386 lfs_writer_enter(fs, "pndirop"); 1387 off = fs->lfs_offset; 1388 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1389 lfs_flush_dirops(fs); 1390 LFS_CLEANERINFO(cip, fs, bp); 1391 oclean = cip->clean; 1392 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1393 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1394 lfs_segunlock(fs); 1395 lfs_writer_leave(fs); 1396 1397 #ifdef DEBUG 1398 LFS_CLEANERINFO(cip, fs, bp); 1399 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1400 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1401 fs->lfs_offset - off, cip->clean - oclean, 1402 fs->lfs_activesb)); 1403 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1404 #endif 1405 1406 VOP_LOCK(ap->a_vp, LK_EXCLUSIVE); 1407 return 0; 1408 1409 case LFCNIFILEFH: 1410 /* Return the filehandle of the Ifile */ 1411 if ((error = suser(ap->a_p->p_ucred, &ap->a_p->p_acflag)) != 0) 1412 return (error); 1413 fhp = (struct fhandle *)ap->a_data; 1414 fhp->fh_fsid = *fsidp; 1415 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid)); 1416 1417 case LFCNREWIND: 1418 /* Move lfs_offset to the lowest-numbered segment */ 1419 return lfs_rewind(fs, *(int *)ap->a_data); 1420 1421 case LFCNINVAL: 1422 /* Mark a segment SEGUSE_INVAL */ 1423 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1424 if (sup->su_nbytes > 0) { 1425 brelse(bp); 1426 lfs_unset_inval_all(fs); 1427 return EBUSY; 1428 } 1429 sup->su_flags |= SEGUSE_INVAL; 1430 VOP_BWRITE(bp); 1431 return 0; 1432 1433 case LFCNRESIZE: 1434 /* Resize the filesystem */ 1435 return lfs_resize_fs(fs, *(int *)ap->a_data); 1436 1437 default: 1438 return ufs_fcntl(v); 1439 } 1440 return 0; 1441 } 1442 1443 int 1444 lfs_getpages(void *v) 1445 { 1446 struct vop_getpages_args /* { 1447 struct vnode *a_vp; 1448 voff_t a_offset; 1449 struct vm_page **a_m; 1450 int *a_count; 1451 int a_centeridx; 1452 vm_prot_t a_access_type; 1453 int a_advice; 1454 int a_flags; 1455 } */ *ap = v; 1456 1457 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1458 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1459 return EPERM; 1460 } 1461 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1462 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1463 } 1464 1465 /* 1466 * we're relying on the fact that genfs_getpages() always read in 1467 * entire filesystem blocks. 1468 */ 1469 return genfs_getpages(v); 1470 } 1471 1472 /* 1473 * Make sure that for all pages in every block in the given range, 1474 * either all are dirty or all are clean. If any of the pages 1475 * we've seen so far are dirty, put the vnode on the paging chain, 1476 * and mark it IN_PAGING. 1477 * 1478 * If checkfirst != 0, don't check all the pages but return at the 1479 * first dirty page. 1480 */ 1481 static int 1482 check_dirty(struct lfs *fs, struct vnode *vp, 1483 off_t startoffset, off_t endoffset, off_t blkeof, 1484 int flags, int checkfirst) 1485 { 1486 int by_list; 1487 struct vm_page *curpg = NULL; /* XXX: gcc */ 1488 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1489 off_t soff = 0; /* XXX: gcc */ 1490 voff_t off; 1491 int i; 1492 int nonexistent; 1493 int any_dirty; /* number of dirty pages */ 1494 int dirty; /* number of dirty pages in a block */ 1495 int tdirty; 1496 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1497 1498 ASSERT_MAYBE_SEGLOCK(fs); 1499 top: 1500 by_list = (vp->v_uobj.uo_npages <= 1501 ((endoffset - startoffset) >> PAGE_SHIFT) * 1502 UVM_PAGE_HASH_PENALTY); 1503 any_dirty = 0; 1504 1505 if (by_list) { 1506 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1507 } else { 1508 soff = startoffset; 1509 } 1510 while (by_list || soff < MIN(blkeof, endoffset)) { 1511 if (by_list) { 1512 /* 1513 * Find the first page in a block. Skip 1514 * blocks outside our area of interest or beyond 1515 * the end of file. 1516 */ 1517 if (pages_per_block > 1) { 1518 while (curpg && 1519 ((curpg->offset & fs->lfs_bmask) || 1520 curpg->offset >= vp->v_size || 1521 curpg->offset >= endoffset)) 1522 curpg = TAILQ_NEXT(curpg, listq); 1523 } 1524 if (curpg == NULL) 1525 break; 1526 soff = curpg->offset; 1527 } 1528 1529 /* 1530 * Mark all pages in extended range busy; find out if any 1531 * of them are dirty. 1532 */ 1533 nonexistent = dirty = 0; 1534 for (i = 0; i == 0 || i < pages_per_block; i++) { 1535 if (by_list && pages_per_block <= 1) { 1536 pgs[i] = pg = curpg; 1537 } else { 1538 off = soff + (i << PAGE_SHIFT); 1539 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1540 if (pg == NULL) { 1541 ++nonexistent; 1542 continue; 1543 } 1544 } 1545 KASSERT(pg != NULL); 1546 while (pg->flags & PG_BUSY) { 1547 pg->flags |= PG_WANTED; 1548 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1549 "lfsput", 0); 1550 simple_lock(&vp->v_interlock); 1551 if (by_list) { 1552 if (i > 0) 1553 uvm_page_unbusy(pgs, i); 1554 goto top; 1555 } 1556 } 1557 pg->flags |= PG_BUSY; 1558 UVM_PAGE_OWN(pg, "lfs_putpages"); 1559 1560 pmap_page_protect(pg, VM_PROT_NONE); 1561 tdirty = (pmap_clear_modify(pg) || 1562 (pg->flags & PG_CLEAN) == 0); 1563 dirty += tdirty; 1564 } 1565 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1566 if (by_list) { 1567 curpg = TAILQ_NEXT(curpg, listq); 1568 } else { 1569 soff += fs->lfs_bsize; 1570 } 1571 continue; 1572 } 1573 1574 any_dirty += dirty; 1575 KASSERT(nonexistent == 0); 1576 1577 /* 1578 * If any are dirty make all dirty; unbusy them, 1579 * but if we were asked to clean, wire them so that 1580 * the pagedaemon doesn't bother us about them while 1581 * they're on their way to disk. 1582 */ 1583 for (i = 0; i == 0 || i < pages_per_block; i++) { 1584 pg = pgs[i]; 1585 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1586 if (dirty) { 1587 pg->flags &= ~PG_CLEAN; 1588 if (flags & PGO_FREE) { 1589 /* 1590 * Wire the page so that 1591 * pdaemon doesn't see it again. 1592 */ 1593 uvm_lock_pageq(); 1594 uvm_pagewire(pg); 1595 uvm_unlock_pageq(); 1596 1597 /* Suspended write flag */ 1598 pg->flags |= PG_DELWRI; 1599 } 1600 } 1601 if (pg->flags & PG_WANTED) 1602 wakeup(pg); 1603 pg->flags &= ~(PG_WANTED|PG_BUSY); 1604 UVM_PAGE_OWN(pg, NULL); 1605 } 1606 1607 if (checkfirst && any_dirty) 1608 break; 1609 1610 if (by_list) { 1611 curpg = TAILQ_NEXT(curpg, listq); 1612 } else { 1613 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1614 } 1615 } 1616 1617 /* 1618 * If any pages were dirty, mark this inode as "pageout requested", 1619 * and put it on the paging queue. 1620 * XXXUBC locking (check locking on dchainhd too) 1621 */ 1622 #ifdef notyet 1623 if (any_dirty) { 1624 if (!(ip->i_flags & IN_PAGING)) { 1625 ip->i_flags |= IN_PAGING; 1626 simple_lock(&fs->lfs_interlock); 1627 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1628 simple_unlock(&fs->lfs_interlock); 1629 } 1630 } 1631 #endif 1632 return any_dirty; 1633 } 1634 1635 /* 1636 * lfs_putpages functions like genfs_putpages except that 1637 * 1638 * (1) It needs to bounds-check the incoming requests to ensure that 1639 * they are block-aligned; if they are not, expand the range and 1640 * do the right thing in case, e.g., the requested range is clean 1641 * but the expanded range is dirty. 1642 * (2) It needs to explicitly send blocks to be written when it is done. 1643 * VOP_PUTPAGES is not ever called with the seglock held, so 1644 * we simply take the seglock and let lfs_segunlock wait for us. 1645 * XXX Actually we can be called with the seglock held, if we have 1646 * XXX to flush a vnode while lfs_markv is in operation. As of this 1647 * XXX writing we panic in this case. 1648 * 1649 * Assumptions: 1650 * 1651 * (1) The caller does not hold any pages in this vnode busy. If it does, 1652 * there is a danger that when we expand the page range and busy the 1653 * pages we will deadlock. 1654 * (2) We are called with vp->v_interlock held; we must return with it 1655 * released. 1656 * (3) We don't absolutely have to free pages right away, provided that 1657 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1658 * us a request with PGO_FREE, we take the pages out of the paging 1659 * queue and wake up the writer, which will handle freeing them for us. 1660 * 1661 * We ensure that for any filesystem block, all pages for that 1662 * block are either resident or not, even if those pages are higher 1663 * than EOF; that means that we will be getting requests to free 1664 * "unused" pages above EOF all the time, and should ignore them. 1665 * 1666 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 1667 */ 1668 1669 int 1670 lfs_putpages(void *v) 1671 { 1672 int error; 1673 struct vop_putpages_args /* { 1674 struct vnode *a_vp; 1675 voff_t a_offlo; 1676 voff_t a_offhi; 1677 int a_flags; 1678 } */ *ap = v; 1679 struct vnode *vp; 1680 struct inode *ip; 1681 struct lfs *fs; 1682 struct segment *sp; 1683 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 1684 off_t off, max_endoffset; 1685 int s; 1686 boolean_t seglocked, sync, pagedaemon; 1687 struct vm_page *pg; 1688 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 1689 1690 vp = ap->a_vp; 1691 ip = VTOI(vp); 1692 fs = ip->i_lfs; 1693 sync = (ap->a_flags & PGO_SYNCIO) != 0; 1694 pagedaemon = (curproc == uvm.pagedaemon_proc); 1695 1696 /* Putpages does nothing for metadata. */ 1697 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 1698 simple_unlock(&vp->v_interlock); 1699 return 0; 1700 } 1701 1702 /* 1703 * If there are no pages, don't do anything. 1704 */ 1705 if (vp->v_uobj.uo_npages == 0) { 1706 s = splbio(); 1707 if (LIST_FIRST(&vp->v_dirtyblkhd) == NULL && 1708 (vp->v_flag & VONWORKLST)) { 1709 vp->v_flag &= ~VONWORKLST; 1710 LIST_REMOVE(vp, v_synclist); 1711 } 1712 splx(s); 1713 simple_unlock(&vp->v_interlock); 1714 return 0; 1715 } 1716 1717 blkeof = blkroundup(fs, ip->i_size); 1718 1719 /* 1720 * Ignore requests to free pages past EOF but in the same block 1721 * as EOF, unless the request is synchronous. (XXX why sync?) 1722 * XXXUBC Make these pages look "active" so the pagedaemon won't 1723 * XXXUBC bother us with them again. 1724 */ 1725 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 1726 origoffset = ap->a_offlo; 1727 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 1728 pg = uvm_pagelookup(&vp->v_uobj, off); 1729 KASSERT(pg != NULL); 1730 while (pg->flags & PG_BUSY) { 1731 pg->flags |= PG_WANTED; 1732 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 1733 "lfsput2", 0); 1734 simple_lock(&vp->v_interlock); 1735 } 1736 uvm_lock_pageq(); 1737 uvm_pageactivate(pg); 1738 uvm_unlock_pageq(); 1739 } 1740 ap->a_offlo = blkeof; 1741 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 1742 simple_unlock(&vp->v_interlock); 1743 return 0; 1744 } 1745 } 1746 1747 /* 1748 * Extend page range to start and end at block boundaries. 1749 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 1750 */ 1751 origoffset = ap->a_offlo; 1752 origendoffset = ap->a_offhi; 1753 startoffset = origoffset & ~(fs->lfs_bmask); 1754 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 1755 << fs->lfs_bshift; 1756 1757 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 1758 endoffset = max_endoffset; 1759 origendoffset = endoffset; 1760 } else { 1761 origendoffset = round_page(ap->a_offhi); 1762 endoffset = round_page(blkroundup(fs, origendoffset)); 1763 } 1764 1765 KASSERT(startoffset > 0 || endoffset >= startoffset); 1766 if (startoffset == endoffset) { 1767 /* Nothing to do, why were we called? */ 1768 simple_unlock(&vp->v_interlock); 1769 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 1770 PRId64 "\n", startoffset)); 1771 return 0; 1772 } 1773 1774 ap->a_offlo = startoffset; 1775 ap->a_offhi = endoffset; 1776 1777 if (!(ap->a_flags & PGO_CLEANIT)) 1778 return genfs_putpages(v); 1779 1780 /* 1781 * If there are more than one page per block, we don't want 1782 * to get caught locking them backwards; so set PGO_BUSYFAIL 1783 * to avoid deadlocks. 1784 */ 1785 ap->a_flags |= PGO_BUSYFAIL; 1786 1787 do { 1788 int r; 1789 1790 /* If no pages are dirty, we can just use genfs_putpages. */ 1791 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 1792 ap->a_flags, 1) != 0) 1793 break; 1794 1795 /* 1796 * Sometimes pages are dirtied between the time that 1797 * we check and the time we try to clean them. 1798 * Instruct lfs_gop_write to return EDEADLK in this case 1799 * so we can write them properly. 1800 */ 1801 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 1802 r = genfs_putpages(v); 1803 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 1804 if (r != EDEADLK) 1805 return r; 1806 1807 /* Start over. */ 1808 preempt(1); 1809 simple_lock(&vp->v_interlock); 1810 } while(1); 1811 1812 /* 1813 * Dirty and asked to clean. 1814 * 1815 * Pagedaemon can't actually write LFS pages; wake up 1816 * the writer to take care of that. The writer will 1817 * notice the pager inode queue and act on that. 1818 */ 1819 if (pagedaemon) { 1820 simple_lock(&fs->lfs_interlock); 1821 ++fs->lfs_pdflush; 1822 simple_unlock(&fs->lfs_interlock); 1823 wakeup(&lfs_writer_daemon); 1824 simple_unlock(&vp->v_interlock); 1825 return EWOULDBLOCK; 1826 } 1827 1828 /* 1829 * If this is a file created in a recent dirop, we can't flush its 1830 * inode until the dirop is complete. Drain dirops, then flush the 1831 * filesystem (taking care of any other pending dirops while we're 1832 * at it). 1833 */ 1834 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 1835 (vp->v_flag & VDIROP)) { 1836 int locked; 1837 1838 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 1839 locked = VOP_ISLOCKED(vp) && /* XXX */ 1840 vp->v_lock.lk_lockholder == curproc->p_pid; 1841 simple_unlock(&vp->v_interlock); 1842 lfs_writer_enter(fs, "ppdirop"); 1843 if (locked) 1844 VOP_UNLOCK(vp, 0); 1845 1846 simple_lock(&fs->lfs_interlock); 1847 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 1848 simple_unlock(&fs->lfs_interlock); 1849 1850 simple_lock(&vp->v_interlock); 1851 if (locked) { 1852 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 1853 simple_lock(&vp->v_interlock); 1854 } 1855 lfs_writer_leave(fs); 1856 1857 /* XXX the flush should have taken care of this one too! */ 1858 } 1859 1860 /* 1861 * This is it. We are going to write some pages. From here on 1862 * down it's all just mechanics. 1863 * 1864 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 1865 */ 1866 ap->a_flags &= ~PGO_SYNCIO; 1867 1868 /* 1869 * If we've already got the seglock, flush the node and return. 1870 * The FIP has already been set up for us by lfs_writefile, 1871 * and FIP cleanup and lfs_updatemeta will also be done there, 1872 * unless genfs_putpages returns EDEADLK; then we must flush 1873 * what we have, and correct FIP and segment header accounting. 1874 */ 1875 1876 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 1877 if (!seglocked) { 1878 simple_unlock(&vp->v_interlock); 1879 /* 1880 * Take the seglock, because we are going to be writing pages. 1881 */ 1882 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 1883 if (error != 0) 1884 return error; 1885 simple_lock(&vp->v_interlock); 1886 } 1887 1888 /* 1889 * VOP_PUTPAGES should not be called while holding the seglock. 1890 * XXXUBC fix lfs_markv, or do this properly. 1891 */ 1892 #ifdef notyet 1893 KASSERT(fs->lfs_seglock == 1); 1894 #endif /* notyet */ 1895 1896 /* 1897 * We assume we're being called with sp->fip pointing at blank space. 1898 * Account for a new FIP in the segment header, and set sp->vp. 1899 * (This should duplicate the setup at the top of lfs_writefile().) 1900 */ 1901 sp = fs->lfs_sp; 1902 if (!seglocked) { 1903 if (sp->seg_bytes_left < fs->lfs_bsize || 1904 sp->sum_bytes_left < sizeof(struct finfo)) 1905 (void) lfs_writeseg(fs, fs->lfs_sp); 1906 1907 sp->sum_bytes_left -= FINFOSIZE; 1908 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1909 } 1910 KASSERT(sp->vp == NULL); 1911 sp->vp = vp; 1912 1913 if (!seglocked) { 1914 if (vp->v_flag & VDIROP) 1915 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 1916 } 1917 1918 sp->fip->fi_nblocks = 0; 1919 sp->fip->fi_ino = ip->i_number; 1920 sp->fip->fi_version = ip->i_gen; 1921 1922 /* 1923 * Loop through genfs_putpages until all pages are gathered. 1924 * genfs_putpages() drops the interlock, so reacquire it if necessary. 1925 * Whenever we lose the interlock we have to rerun check_dirty, as 1926 * well. 1927 */ 1928 again: 1929 check_dirty(fs, vp, startoffset, endoffset, blkeof, ap->a_flags, 0); 1930 1931 if ((error = genfs_putpages(v)) == EDEADLK) { 1932 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 1933 " EDEADLK [2] ino %d off %x (seg %d)\n", 1934 ip->i_number, fs->lfs_offset, 1935 dtosn(fs, fs->lfs_offset))); 1936 /* If nothing to write, short-circuit */ 1937 if (sp->cbpp - sp->bpp > 1) { 1938 /* Write gathered pages */ 1939 lfs_updatemeta(sp); 1940 (void) lfs_writeseg(fs, sp); 1941 1942 /* 1943 * Reinitialize brand new FIP and add us to it. 1944 * (This should duplicate the fixup in 1945 * lfs_gatherpages().) 1946 */ 1947 KASSERT(sp->vp == vp); 1948 sp->fip->fi_version = ip->i_gen; 1949 sp->fip->fi_ino = ip->i_number; 1950 /* Add us to the new segment summary. */ 1951 ++((SEGSUM *)(sp->segsum))->ss_nfinfo; 1952 sp->sum_bytes_left -= FINFOSIZE; 1953 } 1954 1955 /* Give the write a chance to complete */ 1956 preempt(1); 1957 1958 /* We've lost the interlock. Start over. */ 1959 simple_lock(&vp->v_interlock); 1960 goto again; 1961 } 1962 1963 KASSERT(sp->vp == vp); 1964 if (!seglocked) { 1965 sp->vp = NULL; /* XXX lfs_gather below will set this */ 1966 1967 /* Write indirect blocks as well */ 1968 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 1969 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 1970 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 1971 1972 KASSERT(sp->vp == NULL); 1973 sp->vp = vp; 1974 } 1975 1976 /* 1977 * Blocks are now gathered into a segment waiting to be written. 1978 * All that's left to do is update metadata, and write them. 1979 */ 1980 lfs_updatemeta(sp); 1981 KASSERT(sp->vp == vp); 1982 sp->vp = NULL; 1983 1984 if (seglocked) { 1985 /* we're called by lfs_writefile. */ 1986 return error; 1987 } 1988 1989 /* 1990 * Clean up FIP, since we're done writing this file. 1991 * This should duplicate cleanup at the end of lfs_writefile(). 1992 */ 1993 if (sp->fip->fi_nblocks != 0) { 1994 sp->fip = (FINFO*)((caddr_t)sp->fip + FINFOSIZE + 1995 sizeof(int32_t) * sp->fip->fi_nblocks); 1996 sp->start_lbp = &sp->fip->fi_blocks[0]; 1997 } else { 1998 sp->sum_bytes_left += FINFOSIZE; 1999 --((SEGSUM *)(sp->segsum))->ss_nfinfo; 2000 } 2001 lfs_writeseg(fs, fs->lfs_sp); 2002 2003 /* 2004 * XXX - with the malloc/copy writeseg, the pages are freed by now 2005 * even if we don't wait (e.g. if we hold a nested lock). This 2006 * will not be true if we stop using malloc/copy. 2007 */ 2008 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2009 lfs_segunlock(fs); 2010 2011 /* 2012 * Wait for v_numoutput to drop to zero. The seglock should 2013 * take care of this, but there is a slight possibility that 2014 * aiodoned might not have got around to our buffers yet. 2015 */ 2016 if (sync) { 2017 s = splbio(); 2018 simple_lock(&global_v_numoutput_slock); 2019 while (vp->v_numoutput > 0) { 2020 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2021 " num %d\n", ip->i_number, vp->v_numoutput)); 2022 vp->v_flag |= VBWAIT; 2023 ltsleep(&vp->v_numoutput, PRIBIO + 1, "lfs_vn", 0, 2024 &global_v_numoutput_slock); 2025 } 2026 simple_unlock(&global_v_numoutput_slock); 2027 splx(s); 2028 } 2029 return error; 2030 } 2031 2032 /* 2033 * Return the last logical file offset that should be written for this file 2034 * if we're doing a write that ends at "size". If writing, we need to know 2035 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2036 * to know about entire blocks. 2037 */ 2038 void 2039 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2040 { 2041 struct inode *ip = VTOI(vp); 2042 struct lfs *fs = ip->i_lfs; 2043 daddr_t olbn, nlbn; 2044 2045 KASSERT(flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)); 2046 KASSERT((flags & (GOP_SIZE_READ | GOP_SIZE_WRITE)) 2047 != (GOP_SIZE_READ | GOP_SIZE_WRITE)); 2048 2049 olbn = lblkno(fs, ip->i_size); 2050 nlbn = lblkno(fs, size); 2051 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2052 *eobp = fragroundup(fs, size); 2053 } else { 2054 *eobp = blkroundup(fs, size); 2055 } 2056 } 2057 2058 #ifdef DEBUG 2059 void lfs_dump_vop(void *); 2060 2061 void 2062 lfs_dump_vop(void *v) 2063 { 2064 struct vop_putpages_args /* { 2065 struct vnode *a_vp; 2066 voff_t a_offlo; 2067 voff_t a_offhi; 2068 int a_flags; 2069 } */ *ap = v; 2070 2071 #ifdef DDB 2072 vfs_vnode_print(ap->a_vp, 0, printf); 2073 #endif 2074 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2075 } 2076 #endif 2077 2078 int 2079 lfs_mmap(void *v) 2080 { 2081 struct vop_mmap_args /* { 2082 const struct vnodeop_desc *a_desc; 2083 struct vnode *a_vp; 2084 int a_fflags; 2085 struct ucred *a_cred; 2086 struct proc *a_p; 2087 } */ *ap = v; 2088 2089 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2090 return EOPNOTSUPP; 2091 return ufs_mmap(v); 2092 } 2093
Indexes created Tue Sep 30 17:09:57 GMT 2025