lfs_vnops.c revision 1.201.2.10
1 /* $NetBSD: lfs_vnops.c,v 1.201.2.10 2007/08/19 19:25:02 ad 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 1.201.2.10 2007/08/19 19:25:02 ad Exp $"); 71 72 #ifdef _KERNEL_OPT 73 #include "opt_compat_netbsd.h" 74 #endif 75 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/namei.h> 79 #include <sys/resourcevar.h> 80 #include <sys/kernel.h> 81 #include <sys/file.h> 82 #include <sys/stat.h> 83 #include <sys/buf.h> 84 #include <sys/proc.h> 85 #include <sys/mount.h> 86 #include <sys/vnode.h> 87 #include <sys/pool.h> 88 #include <sys/signalvar.h> 89 #include <sys/kauth.h> 90 #include <sys/syslog.h> 91 #include <sys/fstrans.h> 92 93 #include <miscfs/fifofs/fifo.h> 94 #include <miscfs/genfs/genfs.h> 95 #include <miscfs/specfs/specdev.h> 96 97 #include <ufs/ufs/inode.h> 98 #include <ufs/ufs/dir.h> 99 #include <ufs/ufs/ufsmount.h> 100 #include <ufs/ufs/ufs_extern.h> 101 102 #include <uvm/uvm.h> 103 #include <uvm/uvm_pmap.h> 104 #include <uvm/uvm_stat.h> 105 #include <uvm/uvm_pager.h> 106 107 #include <ufs/lfs/lfs.h> 108 #include <ufs/lfs/lfs_extern.h> 109 110 extern pid_t lfs_writer_daemon; 111 int lfs_ignore_lazy_sync = 1; 112 113 /* Global vfs data structures for lfs. */ 114 int (**lfs_vnodeop_p)(void *); 115 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 116 { &vop_default_desc, vn_default_error }, 117 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 118 { &vop_create_desc, lfs_create }, /* create */ 119 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 120 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 121 { &vop_open_desc, ufs_open }, /* open */ 122 { &vop_close_desc, lfs_close }, /* close */ 123 { &vop_access_desc, ufs_access }, /* access */ 124 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 125 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 126 { &vop_read_desc, lfs_read }, /* read */ 127 { &vop_write_desc, lfs_write }, /* write */ 128 { &vop_lease_desc, ufs_lease_check }, /* lease */ 129 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 130 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 131 { &vop_poll_desc, ufs_poll }, /* poll */ 132 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 133 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 134 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 135 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 136 { &vop_seek_desc, ufs_seek }, /* seek */ 137 { &vop_remove_desc, lfs_remove }, /* remove */ 138 { &vop_link_desc, lfs_link }, /* link */ 139 { &vop_rename_desc, lfs_rename }, /* rename */ 140 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 141 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 142 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 143 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 144 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 145 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 146 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 147 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 148 { &vop_lock_desc, ufs_lock }, /* lock */ 149 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 150 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 151 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 152 { &vop_print_desc, ufs_print }, /* print */ 153 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 154 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 155 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 156 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 157 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 158 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 159 { NULL, NULL } 160 }; 161 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 162 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 163 164 int (**lfs_specop_p)(void *); 165 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 166 { &vop_default_desc, vn_default_error }, 167 { &vop_lookup_desc, spec_lookup }, /* lookup */ 168 { &vop_create_desc, spec_create }, /* create */ 169 { &vop_mknod_desc, spec_mknod }, /* mknod */ 170 { &vop_open_desc, spec_open }, /* open */ 171 { &vop_close_desc, lfsspec_close }, /* close */ 172 { &vop_access_desc, ufs_access }, /* access */ 173 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 174 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 175 { &vop_read_desc, ufsspec_read }, /* read */ 176 { &vop_write_desc, ufsspec_write }, /* write */ 177 { &vop_lease_desc, spec_lease_check }, /* lease */ 178 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 179 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 180 { &vop_poll_desc, spec_poll }, /* poll */ 181 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 182 { &vop_revoke_desc, spec_revoke }, /* revoke */ 183 { &vop_mmap_desc, spec_mmap }, /* mmap */ 184 { &vop_fsync_desc, spec_fsync }, /* fsync */ 185 { &vop_seek_desc, spec_seek }, /* seek */ 186 { &vop_remove_desc, spec_remove }, /* remove */ 187 { &vop_link_desc, spec_link }, /* link */ 188 { &vop_rename_desc, spec_rename }, /* rename */ 189 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 190 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 191 { &vop_symlink_desc, spec_symlink }, /* symlink */ 192 { &vop_readdir_desc, spec_readdir }, /* readdir */ 193 { &vop_readlink_desc, spec_readlink }, /* readlink */ 194 { &vop_abortop_desc, spec_abortop }, /* abortop */ 195 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 196 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 197 { &vop_lock_desc, ufs_lock }, /* lock */ 198 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 199 { &vop_bmap_desc, spec_bmap }, /* bmap */ 200 { &vop_strategy_desc, spec_strategy }, /* strategy */ 201 { &vop_print_desc, ufs_print }, /* print */ 202 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 203 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 204 { &vop_advlock_desc, spec_advlock }, /* advlock */ 205 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 206 { &vop_getpages_desc, spec_getpages }, /* getpages */ 207 { &vop_putpages_desc, spec_putpages }, /* putpages */ 208 { NULL, NULL } 209 }; 210 const struct vnodeopv_desc lfs_specop_opv_desc = 211 { &lfs_specop_p, lfs_specop_entries }; 212 213 int (**lfs_fifoop_p)(void *); 214 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 215 { &vop_default_desc, vn_default_error }, 216 { &vop_lookup_desc, fifo_lookup }, /* lookup */ 217 { &vop_create_desc, fifo_create }, /* create */ 218 { &vop_mknod_desc, fifo_mknod }, /* mknod */ 219 { &vop_open_desc, fifo_open }, /* open */ 220 { &vop_close_desc, lfsfifo_close }, /* close */ 221 { &vop_access_desc, ufs_access }, /* access */ 222 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 223 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 224 { &vop_read_desc, ufsfifo_read }, /* read */ 225 { &vop_write_desc, ufsfifo_write }, /* write */ 226 { &vop_lease_desc, fifo_lease_check }, /* lease */ 227 { &vop_ioctl_desc, fifo_ioctl }, /* ioctl */ 228 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 229 { &vop_poll_desc, fifo_poll }, /* poll */ 230 { &vop_kqfilter_desc, fifo_kqfilter }, /* kqfilter */ 231 { &vop_revoke_desc, fifo_revoke }, /* revoke */ 232 { &vop_mmap_desc, fifo_mmap }, /* mmap */ 233 { &vop_fsync_desc, fifo_fsync }, /* fsync */ 234 { &vop_seek_desc, fifo_seek }, /* seek */ 235 { &vop_remove_desc, fifo_remove }, /* remove */ 236 { &vop_link_desc, fifo_link }, /* link */ 237 { &vop_rename_desc, fifo_rename }, /* rename */ 238 { &vop_mkdir_desc, fifo_mkdir }, /* mkdir */ 239 { &vop_rmdir_desc, fifo_rmdir }, /* rmdir */ 240 { &vop_symlink_desc, fifo_symlink }, /* symlink */ 241 { &vop_readdir_desc, fifo_readdir }, /* readdir */ 242 { &vop_readlink_desc, fifo_readlink }, /* readlink */ 243 { &vop_abortop_desc, fifo_abortop }, /* abortop */ 244 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 245 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 246 { &vop_lock_desc, ufs_lock }, /* lock */ 247 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 248 { &vop_bmap_desc, fifo_bmap }, /* bmap */ 249 { &vop_strategy_desc, fifo_strategy }, /* strategy */ 250 { &vop_print_desc, ufs_print }, /* print */ 251 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 252 { &vop_pathconf_desc, fifo_pathconf }, /* pathconf */ 253 { &vop_advlock_desc, fifo_advlock }, /* advlock */ 254 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 255 { &vop_putpages_desc, fifo_putpages }, /* putpages */ 256 { NULL, NULL } 257 }; 258 const struct vnodeopv_desc lfs_fifoop_opv_desc = 259 { &lfs_fifoop_p, lfs_fifoop_entries }; 260 261 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int, struct vm_page **); 262 263 #define LFS_READWRITE 264 #include <ufs/ufs/ufs_readwrite.c> 265 #undef LFS_READWRITE 266 267 /* 268 * Synch an open file. 269 */ 270 /* ARGSUSED */ 271 int 272 lfs_fsync(void *v) 273 { 274 struct vop_fsync_args /* { 275 struct vnode *a_vp; 276 kauth_cred_t a_cred; 277 int a_flags; 278 off_t offlo; 279 off_t offhi; 280 struct lwp *a_l; 281 } */ *ap = v; 282 struct vnode *vp = ap->a_vp; 283 int error, wait; 284 struct inode *ip = VTOI(vp); 285 struct lfs *fs = ip->i_lfs; 286 287 /* If we're mounted read-only, don't try to sync. */ 288 if (fs->lfs_ronly) 289 return 0; 290 291 /* 292 * Trickle sync simply adds this vnode to the pager list, as if 293 * the pagedaemon had requested a pageout. 294 */ 295 if (ap->a_flags & FSYNC_LAZY) { 296 if (lfs_ignore_lazy_sync == 0) { 297 mutex_enter(&fs->lfs_interlock); 298 if (!(ip->i_flags & IN_PAGING)) { 299 ip->i_flags |= IN_PAGING; 300 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, 301 i_lfs_pchain); 302 } 303 mutex_exit(&fs->lfs_interlock); 304 mutex_enter(&lfs_subsys_lock); 305 wakeup(&lfs_writer_daemon); 306 mutex_exit(&lfs_subsys_lock); 307 } 308 return 0; 309 } 310 311 /* 312 * If a vnode is bring cleaned, flush it out before we try to 313 * reuse it. This prevents the cleaner from writing files twice 314 * in the same partial segment, causing an accounting underflow. 315 */ 316 if (ap->a_flags & FSYNC_RECLAIM && ip->i_flags & IN_CLEANING) { 317 lfs_vflush(vp); 318 } 319 320 wait = (ap->a_flags & FSYNC_WAIT); 321 do { 322 mutex_enter(&vp->v_interlock); 323 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 324 round_page(ap->a_offhi), 325 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 326 if (error == EAGAIN) { 327 mutex_enter(&fs->lfs_interlock); 328 mtsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_fsync", 329 hz / 100 + 1, &fs->lfs_interlock); 330 mutex_exit(&fs->lfs_interlock); 331 } 332 } while (error == EAGAIN); 333 if (error) 334 return error; 335 336 if ((ap->a_flags & FSYNC_DATAONLY) == 0) 337 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 338 339 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 340 int l = 0; 341 error = VOP_IOCTL(ip->i_devvp, DIOCCACHESYNC, &l, FWRITE, 342 ap->a_l->l_cred, ap->a_l); 343 } 344 if (wait && !VPISEMPTY(vp)) 345 LFS_SET_UINO(ip, IN_MODIFIED); 346 347 return error; 348 } 349 350 /* 351 * Take IN_ADIROP off, then call ufs_inactive. 352 */ 353 int 354 lfs_inactive(void *v) 355 { 356 struct vop_inactive_args /* { 357 struct vnode *a_vp; 358 struct lwp *a_l; 359 } */ *ap = v; 360 361 KASSERT(VTOI(ap->a_vp)->i_nlink == VTOI(ap->a_vp)->i_ffs_effnlink); 362 363 lfs_unmark_vnode(ap->a_vp); 364 365 /* 366 * The Ifile is only ever inactivated on unmount. 367 * Streamline this process by not giving it more dirty blocks. 368 */ 369 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 370 mutex_enter(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 371 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 372 mutex_exit(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 373 VOP_UNLOCK(ap->a_vp, 0); 374 return 0; 375 } 376 377 return ufs_inactive(v); 378 } 379 380 /* 381 * These macros are used to bracket UFS directory ops, so that we can 382 * identify all the pages touched during directory ops which need to 383 * be ordered and flushed atomically, so that they may be recovered. 384 * 385 * Because we have to mark nodes VDIROP in order to prevent 386 * the cache from reclaiming them while a dirop is in progress, we must 387 * also manage the number of nodes so marked (otherwise we can run out). 388 * We do this by setting lfs_dirvcount to the number of marked vnodes; it 389 * is decremented during segment write, when VDIROP is taken off. 390 */ 391 #define MARK_VNODE(vp) lfs_mark_vnode(vp) 392 #define UNMARK_VNODE(vp) lfs_unmark_vnode(vp) 393 #define SET_DIROP_CREATE(dvp, vpp) lfs_set_dirop_create((dvp), (vpp)) 394 #define SET_DIROP_REMOVE(dvp, vp) lfs_set_dirop((dvp), (vp)) 395 static int lfs_set_dirop_create(struct vnode *, struct vnode **); 396 static int lfs_set_dirop(struct vnode *, struct vnode *); 397 398 static int 399 lfs_set_dirop(struct vnode *dvp, struct vnode *vp) 400 { 401 struct lfs *fs; 402 int error; 403 404 KASSERT(VOP_ISLOCKED(dvp)); 405 KASSERT(vp == NULL || VOP_ISLOCKED(vp)); 406 407 fs = VTOI(dvp)->i_lfs; 408 409 ASSERT_NO_SEGLOCK(fs); 410 /* 411 * LFS_NRESERVE calculates direct and indirect blocks as well 412 * as an inode block; an overestimate in most cases. 413 */ 414 if ((error = lfs_reserve(fs, dvp, vp, LFS_NRESERVE(fs))) != 0) 415 return (error); 416 417 restart: 418 mutex_enter(&fs->lfs_interlock); 419 if (fs->lfs_dirops == 0) { 420 mutex_exit(&fs->lfs_interlock); 421 lfs_check(dvp, LFS_UNUSED_LBN, 0); 422 mutex_enter(&fs->lfs_interlock); 423 } 424 while (fs->lfs_writer) { 425 error = mtsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH, 426 "lfs_sdirop", 0, &fs->lfs_interlock); 427 if (error == EINTR) { 428 mutex_exit(&fs->lfs_interlock); 429 goto unreserve; 430 } 431 } 432 mutex_enter(&lfs_subsys_lock); 433 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 434 wakeup(&lfs_writer_daemon); 435 mutex_exit(&lfs_subsys_lock); 436 mutex_exit(&fs->lfs_interlock); 437 preempt(); 438 goto restart; 439 } 440 441 if (lfs_dirvcount > LFS_MAX_DIROP) { 442 mutex_exit(&fs->lfs_interlock); 443 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 444 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 445 if ((error = mtsleep(&lfs_dirvcount, 446 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 447 &lfs_subsys_lock)) != 0) { 448 goto unreserve; 449 } 450 goto restart; 451 } 452 mutex_exit(&lfs_subsys_lock); 453 454 ++fs->lfs_dirops; 455 fs->lfs_doifile = 1; 456 mutex_exit(&fs->lfs_interlock); 457 458 /* Hold a reference so SET_ENDOP will be happy */ 459 vref(dvp); 460 if (vp) { 461 vref(vp); 462 MARK_VNODE(vp); 463 } 464 465 MARK_VNODE(dvp); 466 return 0; 467 468 unreserve: 469 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 470 return error; 471 } 472 473 /* 474 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 475 * in getnewvnode(), if we have a stacked filesystem mounted on top 476 * of us. 477 * 478 * NB: this means we have to clear the new vnodes on error. Fortunately 479 * SET_ENDOP is there to do that for us. 480 */ 481 static int 482 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 483 { 484 int error; 485 struct lfs *fs; 486 487 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 488 ASSERT_NO_SEGLOCK(fs); 489 if (fs->lfs_ronly) 490 return EROFS; 491 if (vpp && (error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, vpp))) { 492 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 493 dvp, error)); 494 return error; 495 } 496 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 497 if (vpp) { 498 ungetnewvnode(*vpp); 499 *vpp = NULL; 500 } 501 return error; 502 } 503 return 0; 504 } 505 506 #define SET_ENDOP_BASE(fs, dvp, str) \ 507 do { \ 508 mutex_enter(&(fs)->lfs_interlock); \ 509 --(fs)->lfs_dirops; \ 510 if (!(fs)->lfs_dirops) { \ 511 if ((fs)->lfs_nadirop) { \ 512 panic("SET_ENDOP: %s: no dirops but " \ 513 " nadirop=%d", (str), \ 514 (fs)->lfs_nadirop); \ 515 } \ 516 wakeup(&(fs)->lfs_writer); \ 517 mutex_exit(&(fs)->lfs_interlock); \ 518 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 519 } else \ 520 mutex_exit(&(fs)->lfs_interlock); \ 521 } while(0) 522 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 523 do { \ 524 UNMARK_VNODE(dvp); \ 525 if (nvpp && *nvpp) \ 526 UNMARK_VNODE(*nvpp); \ 527 /* Check for error return to stem vnode leakage */ \ 528 if (nvpp && *nvpp && !((*nvpp)->v_uflag & VU_DIROP)) \ 529 ungetnewvnode(*(nvpp)); \ 530 SET_ENDOP_BASE((fs), (dvp), (str)); \ 531 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 532 vrele(dvp); \ 533 } while(0) 534 #define SET_ENDOP_CREATE_AP(ap, str) \ 535 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 536 (ap)->a_vpp, (str)) 537 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 538 do { \ 539 UNMARK_VNODE(dvp); \ 540 if (ovp) \ 541 UNMARK_VNODE(ovp); \ 542 SET_ENDOP_BASE((fs), (dvp), (str)); \ 543 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 544 vrele(dvp); \ 545 if (ovp) \ 546 vrele(ovp); \ 547 } while(0) 548 549 void 550 lfs_mark_vnode(struct vnode *vp) 551 { 552 struct inode *ip = VTOI(vp); 553 struct lfs *fs = ip->i_lfs; 554 555 mutex_enter(&fs->lfs_interlock); 556 if (!(ip->i_flag & IN_ADIROP)) { 557 if (!(vp->v_uflag & VU_DIROP)) { 558 (void)lfs_vref(vp); 559 mutex_enter(&lfs_subsys_lock); 560 ++lfs_dirvcount; 561 ++fs->lfs_dirvcount; 562 mutex_exit(&lfs_subsys_lock); 563 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 564 vp->v_uflag |= VU_DIROP; 565 } 566 ++fs->lfs_nadirop; 567 ip->i_flag |= IN_ADIROP; 568 } else 569 KASSERT(vp->v_uflag & VU_DIROP); 570 mutex_exit(&fs->lfs_interlock); 571 } 572 573 void 574 lfs_unmark_vnode(struct vnode *vp) 575 { 576 struct inode *ip = VTOI(vp); 577 578 if (ip && (ip->i_flag & IN_ADIROP)) { 579 KASSERT(vp->v_uflag & VU_DIROP); 580 mutex_enter(&ip->i_lfs->lfs_interlock); 581 --ip->i_lfs->lfs_nadirop; 582 mutex_exit(&ip->i_lfs->lfs_interlock); 583 ip->i_flag &= ~IN_ADIROP; 584 } 585 } 586 587 int 588 lfs_symlink(void *v) 589 { 590 struct vop_symlink_args /* { 591 struct vnode *a_dvp; 592 struct vnode **a_vpp; 593 struct componentname *a_cnp; 594 struct vattr *a_vap; 595 char *a_target; 596 } */ *ap = v; 597 int error; 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_symlink(ap); 604 SET_ENDOP_CREATE_AP(ap, "symlink"); 605 return (error); 606 } 607 608 int 609 lfs_mknod(void *v) 610 { 611 struct vop_mknod_args /* { 612 struct vnode *a_dvp; 613 struct vnode **a_vpp; 614 struct componentname *a_cnp; 615 struct vattr *a_vap; 616 } */ *ap = v; 617 struct vattr *vap = ap->a_vap; 618 struct vnode **vpp = ap->a_vpp; 619 struct inode *ip; 620 int error; 621 struct mount *mp; 622 ino_t ino; 623 624 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 625 vput(ap->a_dvp); 626 return error; 627 } 628 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 629 ap->a_dvp, vpp, ap->a_cnp); 630 631 /* Either way we're done with the dirop at this point */ 632 SET_ENDOP_CREATE_AP(ap, "mknod"); 633 634 if (error) 635 return (error); 636 637 ip = VTOI(*vpp); 638 mp = (*vpp)->v_mount; 639 ino = ip->i_number; 640 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 641 if (vap->va_rdev != VNOVAL) { 642 /* 643 * Want to be able to use this to make badblock 644 * inodes, so don't truncate the dev number. 645 */ 646 #if 0 647 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 648 UFS_MPNEEDSWAP((*vpp)->v_mount)); 649 #else 650 ip->i_ffs1_rdev = vap->va_rdev; 651 #endif 652 } 653 654 /* 655 * Call fsync to write the vnode so that we don't have to deal with 656 * flushing it when it's marked VDIROP|VXLOCK. 657 * 658 * XXX KS - If we can't flush we also can't call vgone(), so must 659 * return. But, that leaves this vnode in limbo, also not good. 660 * Can this ever happen (barring hardware failure)? 661 */ 662 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0, 663 curlwp)) != 0) { 664 panic("lfs_mknod: couldn't fsync (ino %llu)", 665 (unsigned long long)ino); 666 /* return (error); */ 667 } 668 /* 669 * Remove vnode so that it will be reloaded by VFS_VGET and 670 * checked to see if it is an alias of an existing entry in 671 * the inode cache. 672 */ 673 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 674 675 VOP_UNLOCK(*vpp, 0); 676 lfs_vunref(*vpp); 677 (*vpp)->v_type = VNON; 678 vgone(*vpp); 679 error = VFS_VGET(mp, ino, vpp); 680 681 if (error != 0) { 682 *vpp = NULL; 683 return (error); 684 } 685 return (0); 686 } 687 688 int 689 lfs_create(void *v) 690 { 691 struct vop_create_args /* { 692 struct vnode *a_dvp; 693 struct vnode **a_vpp; 694 struct componentname *a_cnp; 695 struct vattr *a_vap; 696 } */ *ap = v; 697 int error; 698 699 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 700 vput(ap->a_dvp); 701 return error; 702 } 703 error = ufs_create(ap); 704 SET_ENDOP_CREATE_AP(ap, "create"); 705 return (error); 706 } 707 708 int 709 lfs_mkdir(void *v) 710 { 711 struct vop_mkdir_args /* { 712 struct vnode *a_dvp; 713 struct vnode **a_vpp; 714 struct componentname *a_cnp; 715 struct vattr *a_vap; 716 } */ *ap = v; 717 int error; 718 719 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 720 vput(ap->a_dvp); 721 return error; 722 } 723 error = ufs_mkdir(ap); 724 SET_ENDOP_CREATE_AP(ap, "mkdir"); 725 return (error); 726 } 727 728 int 729 lfs_remove(void *v) 730 { 731 struct vop_remove_args /* { 732 struct vnode *a_dvp; 733 struct vnode *a_vp; 734 struct componentname *a_cnp; 735 } */ *ap = v; 736 struct vnode *dvp, *vp; 737 struct inode *ip; 738 int error; 739 740 dvp = ap->a_dvp; 741 vp = ap->a_vp; 742 ip = VTOI(vp); 743 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 744 if (dvp == vp) 745 vrele(vp); 746 else 747 vput(vp); 748 vput(dvp); 749 return error; 750 } 751 error = ufs_remove(ap); 752 if (ip->i_nlink == 0) 753 lfs_orphan(ip->i_lfs, ip->i_number); 754 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 755 return (error); 756 } 757 758 int 759 lfs_rmdir(void *v) 760 { 761 struct vop_rmdir_args /* { 762 struct vnodeop_desc *a_desc; 763 struct vnode *a_dvp; 764 struct vnode *a_vp; 765 struct componentname *a_cnp; 766 } */ *ap = v; 767 struct vnode *vp; 768 struct inode *ip; 769 int error; 770 771 vp = ap->a_vp; 772 ip = VTOI(vp); 773 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 774 if (ap->a_dvp == vp) 775 vrele(ap->a_dvp); 776 else 777 vput(ap->a_dvp); 778 vput(vp); 779 return error; 780 } 781 error = ufs_rmdir(ap); 782 if (ip->i_nlink == 0) 783 lfs_orphan(ip->i_lfs, ip->i_number); 784 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 785 return (error); 786 } 787 788 int 789 lfs_link(void *v) 790 { 791 struct vop_link_args /* { 792 struct vnode *a_dvp; 793 struct vnode *a_vp; 794 struct componentname *a_cnp; 795 } */ *ap = v; 796 int error; 797 struct vnode **vpp = NULL; 798 799 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 800 vput(ap->a_dvp); 801 return error; 802 } 803 error = ufs_link(ap); 804 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 805 return (error); 806 } 807 808 int 809 lfs_rename(void *v) 810 { 811 struct vop_rename_args /* { 812 struct vnode *a_fdvp; 813 struct vnode *a_fvp; 814 struct componentname *a_fcnp; 815 struct vnode *a_tdvp; 816 struct vnode *a_tvp; 817 struct componentname *a_tcnp; 818 } */ *ap = v; 819 struct vnode *tvp, *fvp, *tdvp, *fdvp; 820 struct componentname *tcnp, *fcnp; 821 int error; 822 struct lfs *fs; 823 824 fs = VTOI(ap->a_fdvp)->i_lfs; 825 tvp = ap->a_tvp; 826 tdvp = ap->a_tdvp; 827 tcnp = ap->a_tcnp; 828 fvp = ap->a_fvp; 829 fdvp = ap->a_fdvp; 830 fcnp = ap->a_fcnp; 831 832 /* 833 * Check for cross-device rename. 834 * If it is, we don't want to set dirops, just error out. 835 * (In particular note that MARK_VNODE(tdvp) will DTWT on 836 * a cross-device rename.) 837 * 838 * Copied from ufs_rename. 839 */ 840 if ((fvp->v_mount != tdvp->v_mount) || 841 (tvp && (fvp->v_mount != tvp->v_mount))) { 842 error = EXDEV; 843 goto errout; 844 } 845 846 /* 847 * Check to make sure we're not renaming a vnode onto itself 848 * (deleting a hard link by renaming one name onto another); 849 * if we are we can't recursively call VOP_REMOVE since that 850 * would leave us with an unaccounted-for number of live dirops. 851 * 852 * Inline the relevant section of ufs_rename here, *before* 853 * calling SET_DIROP_REMOVE. 854 */ 855 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 856 (VTOI(tdvp)->i_flags & APPEND))) { 857 error = EPERM; 858 goto errout; 859 } 860 if (fvp == tvp) { 861 if (fvp->v_type == VDIR) { 862 error = EINVAL; 863 goto errout; 864 } 865 866 /* Release destination completely. */ 867 VOP_ABORTOP(tdvp, tcnp); 868 vput(tdvp); 869 vput(tvp); 870 871 /* Delete source. */ 872 vrele(fvp); 873 fcnp->cn_flags &= ~(MODMASK | SAVESTART); 874 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 875 fcnp->cn_nameiop = DELETE; 876 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 877 if ((error = relookup(fdvp, &fvp, fcnp))) { 878 vput(fdvp); 879 return (error); 880 } 881 return (VOP_REMOVE(fdvp, fvp, fcnp)); 882 } 883 884 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 885 goto errout; 886 MARK_VNODE(fdvp); 887 MARK_VNODE(fvp); 888 889 error = ufs_rename(ap); 890 UNMARK_VNODE(fdvp); 891 UNMARK_VNODE(fvp); 892 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 893 return (error); 894 895 errout: 896 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 897 if (tdvp == tvp) 898 vrele(tdvp); 899 else 900 vput(tdvp); 901 if (tvp) 902 vput(tvp); 903 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 904 vrele(fdvp); 905 vrele(fvp); 906 return (error); 907 } 908 909 /* XXX hack to avoid calling ITIMES in getattr */ 910 int 911 lfs_getattr(void *v) 912 { 913 struct vop_getattr_args /* { 914 struct vnode *a_vp; 915 struct vattr *a_vap; 916 kauth_cred_t a_cred; 917 struct lwp *a_l; 918 } */ *ap = v; 919 struct vnode *vp = ap->a_vp; 920 struct inode *ip = VTOI(vp); 921 struct vattr *vap = ap->a_vap; 922 struct lfs *fs = ip->i_lfs; 923 /* 924 * Copy from inode table 925 */ 926 vap->va_fsid = ip->i_dev; 927 vap->va_fileid = ip->i_number; 928 vap->va_mode = ip->i_mode & ~IFMT; 929 vap->va_nlink = ip->i_nlink; 930 vap->va_uid = ip->i_uid; 931 vap->va_gid = ip->i_gid; 932 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 933 vap->va_size = vp->v_size; 934 vap->va_atime.tv_sec = ip->i_ffs1_atime; 935 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 936 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 937 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 938 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 939 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 940 vap->va_flags = ip->i_flags; 941 vap->va_gen = ip->i_gen; 942 /* this doesn't belong here */ 943 if (vp->v_type == VBLK) 944 vap->va_blocksize = BLKDEV_IOSIZE; 945 else if (vp->v_type == VCHR) 946 vap->va_blocksize = MAXBSIZE; 947 else 948 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 949 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 950 vap->va_type = vp->v_type; 951 vap->va_filerev = ip->i_modrev; 952 return (0); 953 } 954 955 /* 956 * Check to make sure the inode blocks won't choke the buffer 957 * cache, then call ufs_setattr as usual. 958 */ 959 int 960 lfs_setattr(void *v) 961 { 962 struct vop_setattr_args /* { 963 struct vnode *a_vp; 964 struct vattr *a_vap; 965 kauth_cred_t a_cred; 966 struct lwp *a_l; 967 } */ *ap = v; 968 struct vnode *vp = ap->a_vp; 969 970 lfs_check(vp, LFS_UNUSED_LBN, 0); 971 return ufs_setattr(v); 972 } 973 974 /* 975 * Release the block we hold on lfs_newseg wrapping. Called on file close, 976 * or explicitly from LFCNWRAPGO. Called with the interlock held. 977 */ 978 static int 979 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 980 { 981 if (fs->lfs_stoplwp != curlwp) 982 return EBUSY; 983 984 fs->lfs_stoplwp = NULL; 985 cv_signal(&fs->lfs_stopcv); 986 987 KASSERT(fs->lfs_nowrap > 0); 988 if (fs->lfs_nowrap <= 0) { 989 return 0; 990 } 991 992 if (--fs->lfs_nowrap == 0) { 993 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 994 wakeup(&fs->lfs_wrappass); 995 lfs_wakeup_cleaner(fs); 996 } 997 if (waitfor) { 998 mtsleep(&fs->lfs_nextseg, PCATCH | PUSER, 999 "segment", 0, &fs->lfs_interlock); 1000 } 1001 1002 return 0; 1003 } 1004 1005 /* 1006 * Close called 1007 */ 1008 /* ARGSUSED */ 1009 int 1010 lfs_close(void *v) 1011 { 1012 struct vop_close_args /* { 1013 struct vnode *a_vp; 1014 int a_fflag; 1015 kauth_cred_t a_cred; 1016 struct lwp *a_l; 1017 } */ *ap = v; 1018 struct vnode *vp = ap->a_vp; 1019 struct inode *ip = VTOI(vp); 1020 struct lfs *fs = ip->i_lfs; 1021 1022 if ((ip->i_number == ROOTINO || ip->i_number == LFS_IFILE_INUM) && 1023 fs->lfs_stoplwp == curlwp) { 1024 mutex_enter(&fs->lfs_interlock); 1025 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1026 lfs_wrapgo(fs, ip, 0); 1027 mutex_exit(&fs->lfs_interlock); 1028 } 1029 1030 if (vp == ip->i_lfs->lfs_ivnode && 1031 vp->v_mount->mnt_iflag & IMNT_UNMOUNT) 1032 return 0; 1033 1034 if (vp->v_usecount > 1 && vp != ip->i_lfs->lfs_ivnode) { 1035 LFS_ITIMES(ip, NULL, NULL, NULL); 1036 } 1037 return (0); 1038 } 1039 1040 /* 1041 * Close wrapper for special devices. 1042 * 1043 * Update the times on the inode then do device close. 1044 */ 1045 int 1046 lfsspec_close(void *v) 1047 { 1048 struct vop_close_args /* { 1049 struct vnode *a_vp; 1050 int a_fflag; 1051 kauth_cred_t a_cred; 1052 struct lwp *a_l; 1053 } */ *ap = v; 1054 struct vnode *vp; 1055 struct inode *ip; 1056 1057 vp = ap->a_vp; 1058 ip = VTOI(vp); 1059 if (vp->v_usecount > 1) { 1060 LFS_ITIMES(ip, NULL, NULL, NULL); 1061 } 1062 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1063 } 1064 1065 /* 1066 * Close wrapper for fifo's. 1067 * 1068 * Update the times on the inode then do device close. 1069 */ 1070 int 1071 lfsfifo_close(void *v) 1072 { 1073 struct vop_close_args /* { 1074 struct vnode *a_vp; 1075 int a_fflag; 1076 kauth_cred_ a_cred; 1077 struct lwp *a_l; 1078 } */ *ap = v; 1079 struct vnode *vp; 1080 struct inode *ip; 1081 1082 vp = ap->a_vp; 1083 ip = VTOI(vp); 1084 if (ap->a_vp->v_usecount > 1) { 1085 LFS_ITIMES(ip, NULL, NULL, NULL); 1086 } 1087 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1088 } 1089 1090 /* 1091 * Reclaim an inode so that it can be used for other purposes. 1092 */ 1093 1094 int 1095 lfs_reclaim(void *v) 1096 { 1097 struct vop_reclaim_args /* { 1098 struct vnode *a_vp; 1099 struct lwp *a_l; 1100 } */ *ap = v; 1101 struct vnode *vp = ap->a_vp; 1102 struct inode *ip = VTOI(vp); 1103 struct lfs *fs = ip->i_lfs; 1104 int error; 1105 1106 KASSERT(ip->i_nlink == ip->i_ffs_effnlink); 1107 1108 mutex_enter(&ip->i_lfs->lfs_interlock); 1109 LFS_CLR_UINO(ip, IN_ALLMOD); 1110 mutex_exit(&ip->i_lfs->lfs_interlock); 1111 if ((error = ufs_reclaim(vp, ap->a_l))) 1112 return (error); 1113 1114 /* 1115 * Take us off the paging and/or dirop queues if we were on them. 1116 * We shouldn't be on them. 1117 */ 1118 mutex_enter(&fs->lfs_interlock); 1119 if (ip->i_flags & IN_PAGING) { 1120 log(LOG_WARNING, "%s: reclaimed vnode is IN_PAGING\n", 1121 fs->lfs_fsmnt); 1122 ip->i_flags &= ~IN_PAGING; 1123 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1124 } 1125 if (vp->v_uflag & VU_DIROP) { 1126 panic("reclaimed vnode is VDIROP"); 1127 vp->v_uflag &= ~VU_DIROP; 1128 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 1129 } 1130 mutex_exit(&fs->lfs_interlock); 1131 1132 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1133 lfs_deregister_all(vp); 1134 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1135 ip->inode_ext.lfs = NULL; 1136 genfs_node_destroy(vp); 1137 pool_put(&lfs_inode_pool, vp->v_data); 1138 vp->v_data = NULL; 1139 return (0); 1140 } 1141 1142 /* 1143 * Read a block from a storage device. 1144 * In order to avoid reading blocks that are in the process of being 1145 * written by the cleaner---and hence are not mutexed by the normal 1146 * buffer cache / page cache mechanisms---check for collisions before 1147 * reading. 1148 * 1149 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1150 * the active cleaner test. 1151 * 1152 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1153 */ 1154 int 1155 lfs_strategy(void *v) 1156 { 1157 struct vop_strategy_args /* { 1158 struct vnode *a_vp; 1159 struct buf *a_bp; 1160 } */ *ap = v; 1161 struct buf *bp; 1162 struct lfs *fs; 1163 struct vnode *vp; 1164 struct inode *ip; 1165 daddr_t tbn; 1166 int i, sn, error, slept; 1167 1168 bp = ap->a_bp; 1169 vp = ap->a_vp; 1170 ip = VTOI(vp); 1171 fs = ip->i_lfs; 1172 1173 /* lfs uses its strategy routine only for read */ 1174 KASSERT(bp->b_flags & B_READ); 1175 1176 if (vp->v_type == VBLK || vp->v_type == VCHR) 1177 panic("lfs_strategy: spec"); 1178 KASSERT(bp->b_bcount != 0); 1179 if (bp->b_blkno == bp->b_lblkno) { 1180 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1181 NULL); 1182 if (error) { 1183 bp->b_error = error; 1184 bp->b_resid = bp->b_bcount; 1185 biodone(bp); 1186 return (error); 1187 } 1188 if ((long)bp->b_blkno == -1) /* no valid data */ 1189 clrbuf(bp); 1190 } 1191 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1192 bp->b_resid = bp->b_bcount; 1193 biodone(bp); 1194 return (0); 1195 } 1196 1197 slept = 1; 1198 mutex_enter(&fs->lfs_interlock); 1199 while (slept && fs->lfs_seglock) { 1200 mutex_exit(&fs->lfs_interlock); 1201 /* 1202 * Look through list of intervals. 1203 * There will only be intervals to look through 1204 * if the cleaner holds the seglock. 1205 * Since the cleaner is synchronous, we can trust 1206 * the list of intervals to be current. 1207 */ 1208 tbn = dbtofsb(fs, bp->b_blkno); 1209 sn = dtosn(fs, tbn); 1210 slept = 0; 1211 for (i = 0; i < fs->lfs_cleanind; i++) { 1212 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1213 tbn >= fs->lfs_cleanint[i]) { 1214 DLOG((DLOG_CLEAN, 1215 "lfs_strategy: ino %d lbn %" PRId64 1216 " ind %d sn %d fsb %" PRIx32 1217 " given sn %d fsb %" PRIx64 "\n", 1218 ip->i_number, bp->b_lblkno, i, 1219 dtosn(fs, fs->lfs_cleanint[i]), 1220 fs->lfs_cleanint[i], sn, tbn)); 1221 DLOG((DLOG_CLEAN, 1222 "lfs_strategy: sleeping on ino %d lbn %" 1223 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1224 mutex_enter(&fs->lfs_interlock); 1225 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1226 /* Cleaner can't wait for itself */ 1227 mtsleep(&fs->lfs_iocount, 1228 (PRIBIO + 1) | PNORELOCK, 1229 "clean2", 0, 1230 &fs->lfs_interlock); 1231 slept = 1; 1232 break; 1233 } else if (fs->lfs_seglock) { 1234 mtsleep(&fs->lfs_seglock, 1235 (PRIBIO + 1) | PNORELOCK, 1236 "clean1", 0, 1237 &fs->lfs_interlock); 1238 slept = 1; 1239 break; 1240 } 1241 mutex_exit(&fs->lfs_interlock); 1242 } 1243 } 1244 mutex_enter(&fs->lfs_interlock); 1245 } 1246 mutex_exit(&fs->lfs_interlock); 1247 1248 vp = ip->i_devvp; 1249 VOP_STRATEGY(vp, bp); 1250 return (0); 1251 } 1252 1253 void 1254 lfs_flush_dirops(struct lfs *fs) 1255 { 1256 struct inode *ip, *nip; 1257 struct vnode *vp; 1258 extern int lfs_dostats; 1259 struct segment *sp; 1260 int waslocked; 1261 1262 ASSERT_MAYBE_SEGLOCK(fs); 1263 KASSERT(fs->lfs_nadirop == 0); 1264 1265 if (fs->lfs_ronly) 1266 return; 1267 1268 mutex_enter(&fs->lfs_interlock); 1269 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1270 mutex_exit(&fs->lfs_interlock); 1271 return; 1272 } else 1273 mutex_exit(&fs->lfs_interlock); 1274 1275 if (lfs_dostats) 1276 ++lfs_stats.flush_invoked; 1277 1278 /* 1279 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1280 * Technically this is a checkpoint (the on-disk state is valid) 1281 * even though we are leaving out all the file data. 1282 */ 1283 lfs_imtime(fs); 1284 lfs_seglock(fs, SEGM_CKP); 1285 sp = fs->lfs_sp; 1286 1287 /* 1288 * lfs_writevnodes, optimized to get dirops out of the way. 1289 * Only write dirops, and don't flush files' pages, only 1290 * blocks from the directories. 1291 * 1292 * We don't need to vref these files because they are 1293 * dirops and so hold an extra reference until the 1294 * segunlock clears them of that status. 1295 * 1296 * We don't need to check for IN_ADIROP because we know that 1297 * no dirops are active. 1298 * 1299 */ 1300 mutex_enter(&fs->lfs_interlock); 1301 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1302 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1303 mutex_exit(&fs->lfs_interlock); 1304 vp = ITOV(ip); 1305 1306 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1307 1308 /* 1309 * All writes to directories come from dirops; all 1310 * writes to files' direct blocks go through the page 1311 * cache, which we're not touching. Reads to files 1312 * and/or directories will not be affected by writing 1313 * directory blocks inodes and file inodes. So we don't 1314 * really need to lock. If we don't lock, though, 1315 * make sure that we don't clear IN_MODIFIED 1316 * unnecessarily. 1317 */ 1318 if (vp->v_iflag & (VI_XLOCK | VI_FREEING)) { 1319 mutex_enter(&fs->lfs_interlock); 1320 continue; 1321 } 1322 waslocked = VOP_ISLOCKED(vp); 1323 if (vp->v_type != VREG && 1324 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1325 lfs_writefile(fs, sp, vp); 1326 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1327 !(ip->i_flag & IN_ALLMOD)) { 1328 mutex_enter(&fs->lfs_interlock); 1329 LFS_SET_UINO(ip, IN_MODIFIED); 1330 mutex_exit(&fs->lfs_interlock); 1331 } 1332 } 1333 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1334 (void) lfs_writeinode(fs, sp, ip); 1335 mutex_enter(&fs->lfs_interlock); 1336 if (waslocked == LK_EXCLOTHER) 1337 LFS_SET_UINO(ip, IN_MODIFIED); 1338 } 1339 mutex_exit(&fs->lfs_interlock); 1340 /* We've written all the dirops there are */ 1341 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1342 lfs_finalize_fs_seguse(fs); 1343 (void) lfs_writeseg(fs, sp); 1344 lfs_segunlock(fs); 1345 } 1346 1347 /* 1348 * Flush all vnodes for which the pagedaemon has requested pageouts. 1349 * Skip over any files that are marked VDIROP (since lfs_flush_dirop() 1350 * has just run, this would be an error). If we have to skip a vnode 1351 * for any reason, just skip it; if we have to wait for the cleaner, 1352 * abort. The writer daemon will call us again later. 1353 */ 1354 void 1355 lfs_flush_pchain(struct lfs *fs) 1356 { 1357 struct inode *ip, *nip; 1358 struct vnode *vp; 1359 extern int lfs_dostats; 1360 struct segment *sp; 1361 int error; 1362 1363 ASSERT_NO_SEGLOCK(fs); 1364 1365 if (fs->lfs_ronly) 1366 return; 1367 1368 mutex_enter(&fs->lfs_interlock); 1369 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1370 mutex_exit(&fs->lfs_interlock); 1371 return; 1372 } else 1373 mutex_exit(&fs->lfs_interlock); 1374 1375 /* Get dirops out of the way */ 1376 lfs_flush_dirops(fs); 1377 1378 if (lfs_dostats) 1379 ++lfs_stats.flush_invoked; 1380 1381 /* 1382 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1383 */ 1384 lfs_imtime(fs); 1385 lfs_seglock(fs, 0); 1386 sp = fs->lfs_sp; 1387 1388 /* 1389 * lfs_writevnodes, optimized to clear pageout requests. 1390 * Only write non-dirop files that are in the pageout queue. 1391 * We're very conservative about what we write; we want to be 1392 * fast and async. 1393 */ 1394 mutex_enter(&fs->lfs_interlock); 1395 top: 1396 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1397 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1398 vp = ITOV(ip); 1399 1400 if (!(ip->i_flags & IN_PAGING)) 1401 goto top; 1402 1403 if ((vp->v_iflag & VI_XLOCK) || (vp->v_uflag & VU_DIROP) != 0) 1404 continue; 1405 if (vp->v_type != VREG) 1406 continue; 1407 if (lfs_vref(vp)) 1408 continue; 1409 mutex_exit(&fs->lfs_interlock); 1410 1411 if (VOP_ISLOCKED(vp)) { 1412 lfs_vunref(vp); 1413 mutex_enter(&fs->lfs_interlock); 1414 continue; 1415 } 1416 1417 error = lfs_writefile(fs, sp, vp); 1418 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1419 !(ip->i_flag & IN_ALLMOD)) { 1420 mutex_enter(&fs->lfs_interlock); 1421 LFS_SET_UINO(ip, IN_MODIFIED); 1422 mutex_exit(&fs->lfs_interlock); 1423 } 1424 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1425 (void) lfs_writeinode(fs, sp, ip); 1426 1427 lfs_vunref(vp); 1428 1429 if (error == EAGAIN) { 1430 lfs_writeseg(fs, sp); 1431 mutex_enter(&fs->lfs_interlock); 1432 break; 1433 } 1434 mutex_enter(&fs->lfs_interlock); 1435 } 1436 mutex_exit(&fs->lfs_interlock); 1437 (void) lfs_writeseg(fs, sp); 1438 lfs_segunlock(fs); 1439 } 1440 1441 /* 1442 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1443 */ 1444 int 1445 lfs_fcntl(void *v) 1446 { 1447 struct vop_fcntl_args /* { 1448 struct vnode *a_vp; 1449 u_long a_command; 1450 void * a_data; 1451 int a_fflag; 1452 kauth_cred_t a_cred; 1453 struct lwp *a_l; 1454 } */ *ap = v; 1455 struct timeval *tvp; 1456 BLOCK_INFO *blkiov; 1457 CLEANERINFO *cip; 1458 SEGUSE *sup; 1459 int blkcnt, error, oclean; 1460 size_t fh_size; 1461 struct lfs_fcntl_markv blkvp; 1462 struct lwp *l; 1463 fsid_t *fsidp; 1464 struct lfs *fs; 1465 struct buf *bp; 1466 fhandle_t *fhp; 1467 daddr_t off; 1468 1469 /* Only respect LFS fcntls on fs root or Ifile */ 1470 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1471 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1472 return ufs_fcntl(v); 1473 } 1474 1475 /* Avoid locking a draining lock */ 1476 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1477 return ESHUTDOWN; 1478 } 1479 1480 /* LFS control and monitoring fcntls are available only to root */ 1481 l = ap->a_l; 1482 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1483 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1484 NULL)) != 0) 1485 return (error); 1486 1487 fs = VTOI(ap->a_vp)->i_lfs; 1488 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1489 1490 error = 0; 1491 switch (ap->a_command) { 1492 case LFCNSEGWAITALL: 1493 case LFCNSEGWAITALL_COMPAT: 1494 fsidp = NULL; 1495 /* FALLSTHROUGH */ 1496 case LFCNSEGWAIT: 1497 case LFCNSEGWAIT_COMPAT: 1498 tvp = (struct timeval *)ap->a_data; 1499 mutex_enter(&fs->lfs_interlock); 1500 ++fs->lfs_sleepers; 1501 mutex_exit(&fs->lfs_interlock); 1502 1503 error = lfs_segwait(fsidp, tvp); 1504 1505 mutex_enter(&fs->lfs_interlock); 1506 if (--fs->lfs_sleepers == 0) 1507 wakeup(&fs->lfs_sleepers); 1508 mutex_exit(&fs->lfs_interlock); 1509 return error; 1510 1511 case LFCNBMAPV: 1512 case LFCNMARKV: 1513 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1514 1515 blkcnt = blkvp.blkcnt; 1516 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1517 return (EINVAL); 1518 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1519 if ((error = copyin(blkvp.blkiov, blkiov, 1520 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1521 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1522 return error; 1523 } 1524 1525 mutex_enter(&fs->lfs_interlock); 1526 ++fs->lfs_sleepers; 1527 mutex_exit(&fs->lfs_interlock); 1528 if (ap->a_command == LFCNBMAPV) 1529 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1530 else /* LFCNMARKV */ 1531 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1532 if (error == 0) 1533 error = copyout(blkiov, blkvp.blkiov, 1534 blkcnt * sizeof(BLOCK_INFO)); 1535 mutex_enter(&fs->lfs_interlock); 1536 if (--fs->lfs_sleepers == 0) 1537 wakeup(&fs->lfs_sleepers); 1538 mutex_exit(&fs->lfs_interlock); 1539 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1540 return error; 1541 1542 case LFCNRECLAIM: 1543 /* 1544 * Flush dirops and write Ifile, allowing empty segments 1545 * to be immediately reclaimed. 1546 */ 1547 lfs_writer_enter(fs, "pndirop"); 1548 off = fs->lfs_offset; 1549 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1550 lfs_flush_dirops(fs); 1551 LFS_CLEANERINFO(cip, fs, bp); 1552 oclean = cip->clean; 1553 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1554 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1555 fs->lfs_sp->seg_flags |= SEGM_PROT; 1556 lfs_segunlock(fs); 1557 lfs_writer_leave(fs); 1558 1559 #ifdef DEBUG 1560 LFS_CLEANERINFO(cip, fs, bp); 1561 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1562 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1563 fs->lfs_offset - off, cip->clean - oclean, 1564 fs->lfs_activesb)); 1565 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1566 #endif 1567 1568 return 0; 1569 1570 #ifdef COMPAT_30 1571 case LFCNIFILEFH_COMPAT: 1572 /* Return the filehandle of the Ifile */ 1573 if ((error = kauth_authorize_generic(l->l_cred, 1574 KAUTH_GENERIC_ISSUSER, NULL)) != 0) 1575 return (error); 1576 fhp = (struct fhandle *)ap->a_data; 1577 fhp->fh_fsid = *fsidp; 1578 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1579 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1580 #endif 1581 1582 case LFCNIFILEFH_COMPAT2: 1583 case LFCNIFILEFH: 1584 /* Return the filehandle of the Ifile */ 1585 fhp = (struct fhandle *)ap->a_data; 1586 fhp->fh_fsid = *fsidp; 1587 fh_size = sizeof(struct lfs_fhandle) - 1588 offsetof(fhandle_t, fh_fid); 1589 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1590 1591 case LFCNREWIND: 1592 /* Move lfs_offset to the lowest-numbered segment */ 1593 return lfs_rewind(fs, *(int *)ap->a_data); 1594 1595 case LFCNINVAL: 1596 /* Mark a segment SEGUSE_INVAL */ 1597 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1598 if (sup->su_nbytes > 0) { 1599 brelse(bp, 0); 1600 lfs_unset_inval_all(fs); 1601 return EBUSY; 1602 } 1603 sup->su_flags |= SEGUSE_INVAL; 1604 VOP_BWRITE(bp); 1605 return 0; 1606 1607 case LFCNRESIZE: 1608 /* Resize the filesystem */ 1609 return lfs_resize_fs(fs, *(int *)ap->a_data); 1610 1611 case LFCNWRAPSTOP: 1612 case LFCNWRAPSTOP_COMPAT: 1613 /* 1614 * Hold lfs_newseg at segment 0; if requested, sleep until 1615 * the filesystem wraps around. To support external agents 1616 * (dump, fsck-based regression test) that need to look at 1617 * a snapshot of the filesystem, without necessarily 1618 * requiring that all fs activity stops. 1619 */ 1620 if (fs->lfs_stoplwp == curlwp) 1621 return EALREADY; 1622 1623 mutex_enter(&fs->lfs_interlock); 1624 while (fs->lfs_stoplwp != NULL) 1625 cv_wait(&fs->lfs_stopcv, &fs->lfs_interlock); 1626 fs->lfs_stoplwp = curlwp; 1627 if (fs->lfs_nowrap == 0) 1628 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1629 ++fs->lfs_nowrap; 1630 if (*(int *)ap->a_data == 1 || 1631 ap->a_command == LFCNWRAPSTOP_COMPAT) { 1632 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1633 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1634 "segwrap", 0, &fs->lfs_interlock); 1635 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1636 if (error) { 1637 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1638 } 1639 } 1640 mutex_exit(&fs->lfs_interlock); 1641 return 0; 1642 1643 case LFCNWRAPGO: 1644 case LFCNWRAPGO_COMPAT: 1645 /* 1646 * Having done its work, the agent wakes up the writer. 1647 * If the argument is 1, it sleeps until a new segment 1648 * is selected. 1649 */ 1650 mutex_enter(&fs->lfs_interlock); 1651 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1652 (ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1653 *((int *)ap->a_data))); 1654 mutex_exit(&fs->lfs_interlock); 1655 return error; 1656 1657 case LFCNWRAPPASS: 1658 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1659 return EALREADY; 1660 mutex_enter(&fs->lfs_interlock); 1661 if (fs->lfs_stoplwp != curlwp) { 1662 mutex_exit(&fs->lfs_interlock); 1663 return EALREADY; 1664 } 1665 if (fs->lfs_nowrap == 0) { 1666 mutex_exit(&fs->lfs_interlock); 1667 return EBUSY; 1668 } 1669 fs->lfs_wrappass = 1; 1670 wakeup(&fs->lfs_wrappass); 1671 /* Wait for the log to wrap, if asked */ 1672 if (*(int *)ap->a_data) { 1673 lfs_vref(ap->a_vp); 1674 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1675 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1676 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1677 "segwrap", 0, &fs->lfs_interlock); 1678 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1679 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1680 lfs_vunref(ap->a_vp); 1681 } 1682 mutex_exit(&fs->lfs_interlock); 1683 return error; 1684 1685 case LFCNWRAPSTATUS: 1686 mutex_enter(&fs->lfs_interlock); 1687 *(int *)ap->a_data = fs->lfs_wrapstatus; 1688 mutex_exit(&fs->lfs_interlock); 1689 return 0; 1690 1691 default: 1692 return ufs_fcntl(v); 1693 } 1694 return 0; 1695 } 1696 1697 int 1698 lfs_getpages(void *v) 1699 { 1700 struct vop_getpages_args /* { 1701 struct vnode *a_vp; 1702 voff_t a_offset; 1703 struct vm_page **a_m; 1704 int *a_count; 1705 int a_centeridx; 1706 vm_prot_t a_access_type; 1707 int a_advice; 1708 int a_flags; 1709 } */ *ap = v; 1710 1711 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1712 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1713 return EPERM; 1714 } 1715 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1716 mutex_enter(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 1717 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1718 mutex_exit(&VTOI(ap->a_vp)->i_lfs->lfs_interlock); 1719 } 1720 1721 /* 1722 * we're relying on the fact that genfs_getpages() always read in 1723 * entire filesystem blocks. 1724 */ 1725 return genfs_getpages(v); 1726 } 1727 1728 /* 1729 * Wait for a page to become unbusy, possibly printing diagnostic messages 1730 * as well. 1731 * 1732 * Called with vp->v_interlock held; return with it held. 1733 */ 1734 static void 1735 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label) 1736 { 1737 if ((pg->flags & PG_BUSY) == 0) 1738 return; /* Nothing to wait for! */ 1739 1740 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN) 1741 static struct vm_page *lastpg; 1742 1743 if (label != NULL && pg != lastpg) { 1744 if (pg->owner_tag) { 1745 printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n", 1746 curproc->p_pid, curlwp->l_lid, label, 1747 pg, pg->owner, pg->lowner, pg->owner_tag); 1748 } else { 1749 printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n", 1750 curproc->p_pid, curlwp->l_lid, label, pg); 1751 } 1752 } 1753 lastpg = pg; 1754 #endif 1755 1756 pg->flags |= PG_WANTED; 1757 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, "lfsput", 0); 1758 mutex_enter(&vp->v_interlock); 1759 } 1760 1761 /* 1762 * This routine is called by lfs_putpages() when it can't complete the 1763 * write because a page is busy. This means that either (1) someone, 1764 * possibly the pagedaemon, is looking at this page, and will give it up 1765 * presently; or (2) we ourselves are holding the page busy in the 1766 * process of being written (either gathered or actually on its way to 1767 * disk). We don't need to give up the segment lock, but we might need 1768 * to call lfs_writeseg() to expedite the page's journey to disk. 1769 * 1770 * Called with vp->v_interlock held; return with it held. 1771 */ 1772 /* #define BUSYWAIT */ 1773 static void 1774 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg, 1775 int seglocked, const char *label) 1776 { 1777 #ifndef BUSYWAIT 1778 struct inode *ip = VTOI(vp); 1779 struct segment *sp = fs->lfs_sp; 1780 int count = 0; 1781 1782 if (pg == NULL) 1783 return; 1784 1785 while (pg->flags & PG_BUSY) { 1786 mutex_exit(&vp->v_interlock); 1787 if (sp->cbpp - sp->bpp > 1) { 1788 /* Write gathered pages */ 1789 lfs_updatemeta(sp); 1790 lfs_release_finfo(fs); 1791 (void) lfs_writeseg(fs, sp); 1792 1793 /* 1794 * Reinitialize FIP 1795 */ 1796 KASSERT(sp->vp == vp); 1797 lfs_acquire_finfo(fs, ip->i_number, 1798 ip->i_gen); 1799 } 1800 ++count; 1801 mutex_enter(&vp->v_interlock); 1802 wait_for_page(vp, pg, label); 1803 } 1804 if (label != NULL && count > 1) 1805 printf("lfs_putpages[%d]: %s: %sn = %d\n", curproc->p_pid, 1806 label, (count > 0 ? "looping, " : ""), count); 1807 #else 1808 preempt(1); 1809 #endif 1810 } 1811 1812 /* 1813 * Make sure that for all pages in every block in the given range, 1814 * either all are dirty or all are clean. If any of the pages 1815 * we've seen so far are dirty, put the vnode on the paging chain, 1816 * and mark it IN_PAGING. 1817 * 1818 * If checkfirst != 0, don't check all the pages but return at the 1819 * first dirty page. 1820 */ 1821 static int 1822 check_dirty(struct lfs *fs, struct vnode *vp, 1823 off_t startoffset, off_t endoffset, off_t blkeof, 1824 int flags, int checkfirst, struct vm_page **pgp) 1825 { 1826 int by_list; 1827 struct vm_page *curpg = NULL; /* XXX: gcc */ 1828 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1829 off_t soff = 0; /* XXX: gcc */ 1830 voff_t off; 1831 int i; 1832 int nonexistent; 1833 int any_dirty; /* number of dirty pages */ 1834 int dirty; /* number of dirty pages in a block */ 1835 int tdirty; 1836 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1837 int pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1838 1839 ASSERT_MAYBE_SEGLOCK(fs); 1840 top: 1841 by_list = (vp->v_uobj.uo_npages <= 1842 ((endoffset - startoffset) >> PAGE_SHIFT) * 1843 UVM_PAGE_HASH_PENALTY); 1844 any_dirty = 0; 1845 1846 if (by_list) { 1847 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1848 } else { 1849 soff = startoffset; 1850 } 1851 while (by_list || soff < MIN(blkeof, endoffset)) { 1852 if (by_list) { 1853 /* 1854 * Find the first page in a block. Skip 1855 * blocks outside our area of interest or beyond 1856 * the end of file. 1857 */ 1858 if (pages_per_block > 1) { 1859 while (curpg && 1860 ((curpg->offset & fs->lfs_bmask) || 1861 curpg->offset >= vp->v_size || 1862 curpg->offset >= endoffset)) 1863 curpg = TAILQ_NEXT(curpg, listq); 1864 } 1865 if (curpg == NULL) 1866 break; 1867 soff = curpg->offset; 1868 } 1869 1870 /* 1871 * Mark all pages in extended range busy; find out if any 1872 * of them are dirty. 1873 */ 1874 nonexistent = dirty = 0; 1875 for (i = 0; i == 0 || i < pages_per_block; i++) { 1876 if (by_list && pages_per_block <= 1) { 1877 pgs[i] = pg = curpg; 1878 } else { 1879 off = soff + (i << PAGE_SHIFT); 1880 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1881 if (pg == NULL) { 1882 ++nonexistent; 1883 continue; 1884 } 1885 } 1886 KASSERT(pg != NULL); 1887 1888 /* 1889 * If we're holding the segment lock, we can deadlock 1890 * against a process that has our page and is waiting 1891 * for the cleaner, while the cleaner waits for the 1892 * segment lock. Just bail in that case. 1893 */ 1894 if ((pg->flags & PG_BUSY) && 1895 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1896 if (i > 0) 1897 uvm_page_unbusy(pgs, i); 1898 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1899 if (pgp) 1900 *pgp = pg; 1901 return -1; 1902 } 1903 1904 while (pg->flags & PG_BUSY) { 1905 wait_for_page(vp, pg, NULL); 1906 if (i > 0) 1907 uvm_page_unbusy(pgs, i); 1908 goto top; 1909 } 1910 pg->flags |= PG_BUSY; 1911 UVM_PAGE_OWN(pg, "lfs_putpages"); 1912 1913 pmap_page_protect(pg, VM_PROT_NONE); 1914 tdirty = (pmap_clear_modify(pg) || 1915 (pg->flags & PG_CLEAN) == 0); 1916 dirty += tdirty; 1917 } 1918 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1919 if (by_list) { 1920 curpg = TAILQ_NEXT(curpg, listq); 1921 } else { 1922 soff += fs->lfs_bsize; 1923 } 1924 continue; 1925 } 1926 1927 any_dirty += dirty; 1928 KASSERT(nonexistent == 0); 1929 1930 /* 1931 * If any are dirty make all dirty; unbusy them, 1932 * but if we were asked to clean, wire them so that 1933 * the pagedaemon doesn't bother us about them while 1934 * they're on their way to disk. 1935 */ 1936 for (i = 0; i == 0 || i < pages_per_block; i++) { 1937 pg = pgs[i]; 1938 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 1939 if (dirty) { 1940 pg->flags &= ~PG_CLEAN; 1941 if (flags & PGO_FREE) { 1942 /* 1943 * Wire the page so that 1944 * pdaemon doesn't see it again. 1945 */ 1946 mutex_enter(&uvm_pageqlock); 1947 uvm_pagewire(pg); 1948 mutex_exit(&uvm_pageqlock); 1949 1950 /* Suspended write flag */ 1951 pg->flags |= PG_DELWRI; 1952 } 1953 } 1954 if (pg->flags & PG_WANTED) 1955 wakeup(pg); 1956 pg->flags &= ~(PG_WANTED|PG_BUSY); 1957 UVM_PAGE_OWN(pg, NULL); 1958 } 1959 1960 if (checkfirst && any_dirty) 1961 break; 1962 1963 if (by_list) { 1964 curpg = TAILQ_NEXT(curpg, listq); 1965 } else { 1966 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 1967 } 1968 } 1969 1970 return any_dirty; 1971 } 1972 1973 /* 1974 * lfs_putpages functions like genfs_putpages except that 1975 * 1976 * (1) It needs to bounds-check the incoming requests to ensure that 1977 * they are block-aligned; if they are not, expand the range and 1978 * do the right thing in case, e.g., the requested range is clean 1979 * but the expanded range is dirty. 1980 * 1981 * (2) It needs to explicitly send blocks to be written when it is done. 1982 * If VOP_PUTPAGES is called without the seglock held, we simply take 1983 * the seglock and let lfs_segunlock wait for us. 1984 * XXX There might be a bad situation if we have to flush a vnode while 1985 * XXX lfs_markv is in operation. As of this writing we panic in this 1986 * XXX case. 1987 * 1988 * Assumptions: 1989 * 1990 * (1) The caller does not hold any pages in this vnode busy. If it does, 1991 * there is a danger that when we expand the page range and busy the 1992 * pages we will deadlock. 1993 * 1994 * (2) We are called with vp->v_interlock held; we must return with it 1995 * released. 1996 * 1997 * (3) We don't absolutely have to free pages right away, provided that 1998 * the request does not have PGO_SYNCIO. When the pagedaemon gives 1999 * us a request with PGO_FREE, we take the pages out of the paging 2000 * queue and wake up the writer, which will handle freeing them for us. 2001 * 2002 * We ensure that for any filesystem block, all pages for that 2003 * block are either resident or not, even if those pages are higher 2004 * than EOF; that means that we will be getting requests to free 2005 * "unused" pages above EOF all the time, and should ignore them. 2006 * 2007 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 2008 * into has been set up for us by lfs_writefile. If not, we will 2009 * have to handle allocating and/or freeing an finfo entry. 2010 * 2011 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 2012 */ 2013 2014 /* How many times to loop before we should start to worry */ 2015 #define TOOMANY 4 2016 2017 int 2018 lfs_putpages(void *v) 2019 { 2020 int error; 2021 struct vop_putpages_args /* { 2022 struct vnode *a_vp; 2023 voff_t a_offlo; 2024 voff_t a_offhi; 2025 int a_flags; 2026 } */ *ap = v; 2027 struct vnode *vp; 2028 struct inode *ip; 2029 struct lfs *fs; 2030 struct segment *sp; 2031 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 2032 off_t off, max_endoffset; 2033 bool seglocked, sync, pagedaemon; 2034 struct vm_page *pg, *busypg; 2035 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 2036 #ifdef DEBUG 2037 int debug_n_again, debug_n_dirtyclean; 2038 #endif 2039 2040 vp = ap->a_vp; 2041 ip = VTOI(vp); 2042 fs = ip->i_lfs; 2043 sync = (ap->a_flags & PGO_SYNCIO) != 0; 2044 pagedaemon = (curlwp == uvm.pagedaemon_lwp); 2045 2046 /* Putpages does nothing for metadata. */ 2047 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 2048 mutex_exit(&vp->v_interlock); 2049 return 0; 2050 } 2051 2052 /* 2053 * If there are no pages, don't do anything. 2054 */ 2055 if (vp->v_uobj.uo_npages == 0) { 2056 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 2057 (vp->v_iflag & VI_ONWORKLST) && 2058 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2059 vp->v_iflag &= ~VI_WRMAPDIRTY; 2060 vn_syncer_remove_from_worklist(vp); 2061 } 2062 mutex_exit(&vp->v_interlock); 2063 2064 /* Remove us from paging queue, if we were on it */ 2065 mutex_enter(&fs->lfs_interlock); 2066 if (ip->i_flags & IN_PAGING) { 2067 ip->i_flags &= ~IN_PAGING; 2068 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2069 } 2070 mutex_exit(&fs->lfs_interlock); 2071 return 0; 2072 } 2073 2074 blkeof = blkroundup(fs, ip->i_size); 2075 2076 /* 2077 * Ignore requests to free pages past EOF but in the same block 2078 * as EOF, unless the request is synchronous. (If the request is 2079 * sync, it comes from lfs_truncate.) 2080 * XXXUBC Make these pages look "active" so the pagedaemon won't 2081 * XXXUBC bother us with them again. 2082 */ 2083 if (!sync && ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 2084 origoffset = ap->a_offlo; 2085 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 2086 pg = uvm_pagelookup(&vp->v_uobj, off); 2087 KASSERT(pg != NULL); 2088 while (pg->flags & PG_BUSY) { 2089 pg->flags |= PG_WANTED; 2090 UVM_UNLOCK_AND_WAIT(pg, &vp->v_interlock, 0, 2091 "lfsput2", 0); 2092 mutex_enter(&vp->v_interlock); 2093 } 2094 mutex_enter(&uvm_pageqlock); 2095 uvm_pageactivate(pg); 2096 mutex_exit(&uvm_pageqlock); 2097 } 2098 ap->a_offlo = blkeof; 2099 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 2100 mutex_exit(&vp->v_interlock); 2101 return 0; 2102 } 2103 } 2104 2105 /* 2106 * Extend page range to start and end at block boundaries. 2107 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 2108 */ 2109 origoffset = ap->a_offlo; 2110 origendoffset = ap->a_offhi; 2111 startoffset = origoffset & ~(fs->lfs_bmask); 2112 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 2113 << fs->lfs_bshift; 2114 2115 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2116 endoffset = max_endoffset; 2117 origendoffset = endoffset; 2118 } else { 2119 origendoffset = round_page(ap->a_offhi); 2120 endoffset = round_page(blkroundup(fs, origendoffset)); 2121 } 2122 2123 KASSERT(startoffset > 0 || endoffset >= startoffset); 2124 if (startoffset == endoffset) { 2125 /* Nothing to do, why were we called? */ 2126 mutex_exit(&vp->v_interlock); 2127 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 2128 PRId64 "\n", startoffset)); 2129 return 0; 2130 } 2131 2132 ap->a_offlo = startoffset; 2133 ap->a_offhi = endoffset; 2134 2135 /* 2136 * If not cleaning, just send the pages through genfs_putpages 2137 * to be returned to the pool. 2138 */ 2139 if (!(ap->a_flags & PGO_CLEANIT)) 2140 return genfs_putpages(v); 2141 2142 /* Set PGO_BUSYFAIL to avoid deadlocks */ 2143 ap->a_flags |= PGO_BUSYFAIL; 2144 2145 /* 2146 * Likewise, if we are asked to clean but the pages are not 2147 * dirty, we can just free them using genfs_putpages. 2148 */ 2149 #ifdef DEBUG 2150 debug_n_dirtyclean = 0; 2151 #endif 2152 do { 2153 int r; 2154 2155 /* Count the number of dirty pages */ 2156 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2157 ap->a_flags, 1, NULL); 2158 if (r < 0) { 2159 /* Pages are busy with another process */ 2160 mutex_exit(&vp->v_interlock); 2161 return EDEADLK; 2162 } 2163 if (r > 0) /* Some pages are dirty */ 2164 break; 2165 2166 /* 2167 * Sometimes pages are dirtied between the time that 2168 * we check and the time we try to clean them. 2169 * Instruct lfs_gop_write to return EDEADLK in this case 2170 * so we can write them properly. 2171 */ 2172 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2173 r = genfs_do_putpages(vp, startoffset, endoffset, 2174 ap->a_flags, &busypg); 2175 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2176 if (r != EDEADLK) 2177 return r; 2178 2179 /* One of the pages was busy. Start over. */ 2180 mutex_enter(&vp->v_interlock); 2181 wait_for_page(vp, busypg, "dirtyclean"); 2182 #ifdef DEBUG 2183 ++debug_n_dirtyclean; 2184 #endif 2185 } while(1); 2186 2187 #ifdef DEBUG 2188 if (debug_n_dirtyclean > TOOMANY) 2189 printf("lfs_putpages: dirtyclean: looping, n = %d\n", 2190 debug_n_dirtyclean); 2191 #endif 2192 2193 /* 2194 * Dirty and asked to clean. 2195 * 2196 * Pagedaemon can't actually write LFS pages; wake up 2197 * the writer to take care of that. The writer will 2198 * notice the pager inode queue and act on that. 2199 */ 2200 if (pagedaemon) { 2201 mutex_enter(&fs->lfs_interlock); 2202 if (!(ip->i_flags & IN_PAGING)) { 2203 ip->i_flags |= IN_PAGING; 2204 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2205 } 2206 mutex_enter(&lfs_subsys_lock); 2207 wakeup(&lfs_writer_daemon); 2208 mutex_exit(&lfs_subsys_lock); 2209 mutex_exit(&fs->lfs_interlock); 2210 mutex_exit(&vp->v_interlock); 2211 preempt(); 2212 return EWOULDBLOCK; 2213 } 2214 2215 /* 2216 * If this is a file created in a recent dirop, we can't flush its 2217 * inode until the dirop is complete. Drain dirops, then flush the 2218 * filesystem (taking care of any other pending dirops while we're 2219 * at it). 2220 */ 2221 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2222 (vp->v_uflag & VU_DIROP)) { 2223 int locked; 2224 2225 DLOG((DLOG_PAGE, "lfs_putpages: flushing VDIROP\n")); 2226 locked = (VOP_ISLOCKED(vp) == LK_EXCLUSIVE); 2227 mutex_exit(&vp->v_interlock); 2228 lfs_writer_enter(fs, "ppdirop"); 2229 if (locked) 2230 VOP_UNLOCK(vp, 0); /* XXX why? */ 2231 2232 mutex_enter(&fs->lfs_interlock); 2233 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2234 mutex_exit(&fs->lfs_interlock); 2235 2236 mutex_enter(&vp->v_interlock); 2237 if (locked) { 2238 VOP_LOCK(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2239 mutex_enter(&vp->v_interlock); 2240 } 2241 lfs_writer_leave(fs); 2242 2243 /* XXX the flush should have taken care of this one too! */ 2244 } 2245 2246 /* 2247 * This is it. We are going to write some pages. From here on 2248 * down it's all just mechanics. 2249 * 2250 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2251 */ 2252 ap->a_flags &= ~PGO_SYNCIO; 2253 2254 /* 2255 * If we've already got the seglock, flush the node and return. 2256 * The FIP has already been set up for us by lfs_writefile, 2257 * and FIP cleanup and lfs_updatemeta will also be done there, 2258 * unless genfs_putpages returns EDEADLK; then we must flush 2259 * what we have, and correct FIP and segment header accounting. 2260 */ 2261 get_seglock: 2262 /* 2263 * If we are not called with the segment locked, lock it. 2264 * Account for a new FIP in the segment header, and set sp->vp. 2265 * (This should duplicate the setup at the top of lfs_writefile().) 2266 */ 2267 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2268 if (!seglocked) { 2269 mutex_exit(&vp->v_interlock); 2270 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2271 if (error != 0) 2272 return error; 2273 mutex_enter(&vp->v_interlock); 2274 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2275 } 2276 sp = fs->lfs_sp; 2277 KASSERT(sp->vp == NULL); 2278 sp->vp = vp; 2279 2280 /* 2281 * Ensure that the partial segment is marked SS_DIROP if this 2282 * vnode is a DIROP. 2283 */ 2284 if (!seglocked && vp->v_uflag & VU_DIROP) 2285 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2286 2287 /* 2288 * Loop over genfs_putpages until all pages are gathered. 2289 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2290 * Whenever we lose the interlock we have to rerun check_dirty, as 2291 * well, since more pages might have been dirtied in our absence. 2292 */ 2293 #ifdef DEBUG 2294 debug_n_again = 0; 2295 #endif 2296 do { 2297 busypg = NULL; 2298 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2299 ap->a_flags, 0, &busypg) < 0) { 2300 mutex_exit(&vp->v_interlock); 2301 2302 mutex_enter(&vp->v_interlock); 2303 write_and_wait(fs, vp, busypg, seglocked, NULL); 2304 if (!seglocked) { 2305 lfs_release_finfo(fs); 2306 lfs_segunlock(fs); 2307 } 2308 sp->vp = NULL; 2309 goto get_seglock; 2310 } 2311 2312 busypg = NULL; 2313 error = genfs_do_putpages(vp, startoffset, endoffset, 2314 ap->a_flags, &busypg); 2315 2316 if (error == EDEADLK || error == EAGAIN) { 2317 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2318 " %d ino %d off %x (seg %d)\n", error, 2319 ip->i_number, fs->lfs_offset, 2320 dtosn(fs, fs->lfs_offset))); 2321 2322 mutex_enter(&vp->v_interlock); 2323 write_and_wait(fs, vp, busypg, seglocked, "again"); 2324 } 2325 #ifdef DEBUG 2326 ++debug_n_again; 2327 #endif 2328 } while (error == EDEADLK); 2329 #ifdef DEBUG 2330 if (debug_n_again > TOOMANY) 2331 printf("lfs_putpages: again: looping, n = %d\n", debug_n_again); 2332 #endif 2333 2334 KASSERT(sp != NULL && sp->vp == vp); 2335 if (!seglocked) { 2336 sp->vp = NULL; 2337 2338 /* Write indirect blocks as well */ 2339 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2340 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2341 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2342 2343 KASSERT(sp->vp == NULL); 2344 sp->vp = vp; 2345 } 2346 2347 /* 2348 * Blocks are now gathered into a segment waiting to be written. 2349 * All that's left to do is update metadata, and write them. 2350 */ 2351 lfs_updatemeta(sp); 2352 KASSERT(sp->vp == vp); 2353 sp->vp = NULL; 2354 2355 /* 2356 * If we were called from lfs_writefile, we don't need to clean up 2357 * the FIP or unlock the segment lock. We're done. 2358 */ 2359 if (seglocked) 2360 return error; 2361 2362 /* Clean up FIP and send it to disk. */ 2363 lfs_release_finfo(fs); 2364 lfs_writeseg(fs, fs->lfs_sp); 2365 2366 /* 2367 * Remove us from paging queue if we wrote all our pages. 2368 */ 2369 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2370 mutex_enter(&fs->lfs_interlock); 2371 if (ip->i_flags & IN_PAGING) { 2372 ip->i_flags &= ~IN_PAGING; 2373 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2374 } 2375 mutex_exit(&fs->lfs_interlock); 2376 } 2377 2378 /* 2379 * XXX - with the malloc/copy writeseg, the pages are freed by now 2380 * even if we don't wait (e.g. if we hold a nested lock). This 2381 * will not be true if we stop using malloc/copy. 2382 */ 2383 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2384 lfs_segunlock(fs); 2385 2386 /* 2387 * Wait for v_numoutput to drop to zero. The seglock should 2388 * take care of this, but there is a slight possibility that 2389 * aiodoned might not have got around to our buffers yet. 2390 */ 2391 if (sync) { 2392 mutex_enter(&vp->v_interlock); 2393 while (vp->v_numoutput > 0) { 2394 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2395 " num %d\n", ip->i_number, vp->v_numoutput)); 2396 cv_wait(&vp->v_cv, &vp->v_interlock); 2397 } 2398 mutex_exit(&vp->v_interlock); 2399 } 2400 return error; 2401 } 2402 2403 /* 2404 * Return the last logical file offset that should be written for this file 2405 * if we're doing a write that ends at "size". If writing, we need to know 2406 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2407 * to know about entire blocks. 2408 */ 2409 void 2410 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2411 { 2412 struct inode *ip = VTOI(vp); 2413 struct lfs *fs = ip->i_lfs; 2414 daddr_t olbn, nlbn; 2415 2416 olbn = lblkno(fs, ip->i_size); 2417 nlbn = lblkno(fs, size); 2418 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2419 *eobp = fragroundup(fs, size); 2420 } else { 2421 *eobp = blkroundup(fs, size); 2422 } 2423 } 2424 2425 #ifdef DEBUG 2426 void lfs_dump_vop(void *); 2427 2428 void 2429 lfs_dump_vop(void *v) 2430 { 2431 struct vop_putpages_args /* { 2432 struct vnode *a_vp; 2433 voff_t a_offlo; 2434 voff_t a_offhi; 2435 int a_flags; 2436 } */ *ap = v; 2437 2438 #ifdef DDB 2439 vfs_vnode_print(ap->a_vp, 0, printf); 2440 #endif 2441 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2442 } 2443 #endif 2444 2445 int 2446 lfs_mmap(void *v) 2447 { 2448 struct vop_mmap_args /* { 2449 const struct vnodeop_desc *a_desc; 2450 struct vnode *a_vp; 2451 int a_fflags; 2452 kauth_cred_t a_cred; 2453 struct lwp *a_l; 2454 } */ *ap = v; 2455 2456 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2457 return EOPNOTSUPP; 2458 return ufs_mmap(v); 2459 } 2460
Indexes created Mon Oct 20 01:09:56 GMT 2025