lfs_vnops.c revision 1.239
1 /* $NetBSD: lfs_vnops.c,v 1.239 2012/01/02 22:10:45 perseant Exp $ */ 2 3 /*- 4 * Copyright (c) 1999, 2000, 2001, 2002, 2003 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Konrad E. Schroder <perseant (at) hhhh.org>. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 /* 32 * Copyright (c) 1986, 1989, 1991, 1993, 1995 33 * The Regents of the University of California. All rights reserved. 34 * 35 * Redistribution and use in source and binary forms, with or without 36 * modification, are permitted provided that the following conditions 37 * are met: 38 * 1. Redistributions of source code must retain the above copyright 39 * notice, this list of conditions and the following disclaimer. 40 * 2. Redistributions in binary form must reproduce the above copyright 41 * notice, this list of conditions and the following disclaimer in the 42 * documentation and/or other materials provided with the distribution. 43 * 3. Neither the name of the University nor the names of its contributors 44 * may be used to endorse or promote products derived from this software 45 * without specific prior written permission. 46 * 47 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 48 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 49 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 50 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 51 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 52 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 53 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 54 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 55 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 56 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 57 * SUCH DAMAGE. 58 * 59 * @(#)lfs_vnops.c 8.13 (Berkeley) 6/10/95 60 */ 61 62 #include <sys/cdefs.h> 63 __KERNEL_RCSID(0, "$NetBSD: lfs_vnops.c,v 1.239 2012/01/02 22:10:45 perseant Exp $"); 64 65 #ifdef _KERNEL_OPT 66 #include "opt_compat_netbsd.h" 67 #include "opt_uvm_page_trkown.h" 68 #endif 69 70 #include <sys/param.h> 71 #include <sys/systm.h> 72 #include <sys/namei.h> 73 #include <sys/resourcevar.h> 74 #include <sys/kernel.h> 75 #include <sys/file.h> 76 #include <sys/stat.h> 77 #include <sys/buf.h> 78 #include <sys/proc.h> 79 #include <sys/mount.h> 80 #include <sys/vnode.h> 81 #include <sys/pool.h> 82 #include <sys/signalvar.h> 83 #include <sys/kauth.h> 84 #include <sys/syslog.h> 85 #include <sys/fstrans.h> 86 87 #include <miscfs/fifofs/fifo.h> 88 #include <miscfs/genfs/genfs.h> 89 #include <miscfs/specfs/specdev.h> 90 91 #include <ufs/ufs/inode.h> 92 #include <ufs/ufs/dir.h> 93 #include <ufs/ufs/ufsmount.h> 94 #include <ufs/ufs/ufs_extern.h> 95 96 #include <uvm/uvm.h> 97 #include <uvm/uvm_pmap.h> 98 #include <uvm/uvm_stat.h> 99 #include <uvm/uvm_pager.h> 100 101 #include <ufs/lfs/lfs.h> 102 #include <ufs/lfs/lfs_extern.h> 103 104 extern pid_t lfs_writer_daemon; 105 int lfs_ignore_lazy_sync = 1; 106 107 /* Global vfs data structures for lfs. */ 108 int (**lfs_vnodeop_p)(void *); 109 const struct vnodeopv_entry_desc lfs_vnodeop_entries[] = { 110 { &vop_default_desc, vn_default_error }, 111 { &vop_lookup_desc, ufs_lookup }, /* lookup */ 112 { &vop_create_desc, lfs_create }, /* create */ 113 { &vop_whiteout_desc, ufs_whiteout }, /* whiteout */ 114 { &vop_mknod_desc, lfs_mknod }, /* mknod */ 115 { &vop_open_desc, ufs_open }, /* open */ 116 { &vop_close_desc, lfs_close }, /* close */ 117 { &vop_access_desc, ufs_access }, /* access */ 118 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 119 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 120 { &vop_read_desc, lfs_read }, /* read */ 121 { &vop_write_desc, lfs_write }, /* write */ 122 { &vop_ioctl_desc, ufs_ioctl }, /* ioctl */ 123 { &vop_fcntl_desc, lfs_fcntl }, /* fcntl */ 124 { &vop_poll_desc, ufs_poll }, /* poll */ 125 { &vop_kqfilter_desc, genfs_kqfilter }, /* kqfilter */ 126 { &vop_revoke_desc, ufs_revoke }, /* revoke */ 127 { &vop_mmap_desc, lfs_mmap }, /* mmap */ 128 { &vop_fsync_desc, lfs_fsync }, /* fsync */ 129 { &vop_seek_desc, ufs_seek }, /* seek */ 130 { &vop_remove_desc, lfs_remove }, /* remove */ 131 { &vop_link_desc, lfs_link }, /* link */ 132 { &vop_rename_desc, lfs_rename }, /* rename */ 133 { &vop_mkdir_desc, lfs_mkdir }, /* mkdir */ 134 { &vop_rmdir_desc, lfs_rmdir }, /* rmdir */ 135 { &vop_symlink_desc, lfs_symlink }, /* symlink */ 136 { &vop_readdir_desc, ufs_readdir }, /* readdir */ 137 { &vop_readlink_desc, ufs_readlink }, /* readlink */ 138 { &vop_abortop_desc, ufs_abortop }, /* abortop */ 139 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 140 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 141 { &vop_lock_desc, ufs_lock }, /* lock */ 142 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 143 { &vop_bmap_desc, ufs_bmap }, /* bmap */ 144 { &vop_strategy_desc, lfs_strategy }, /* strategy */ 145 { &vop_print_desc, ufs_print }, /* print */ 146 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 147 { &vop_pathconf_desc, ufs_pathconf }, /* pathconf */ 148 { &vop_advlock_desc, ufs_advlock }, /* advlock */ 149 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 150 { &vop_getpages_desc, lfs_getpages }, /* getpages */ 151 { &vop_putpages_desc, lfs_putpages }, /* putpages */ 152 { NULL, NULL } 153 }; 154 const struct vnodeopv_desc lfs_vnodeop_opv_desc = 155 { &lfs_vnodeop_p, lfs_vnodeop_entries }; 156 157 int (**lfs_specop_p)(void *); 158 const struct vnodeopv_entry_desc lfs_specop_entries[] = { 159 { &vop_default_desc, vn_default_error }, 160 { &vop_lookup_desc, spec_lookup }, /* lookup */ 161 { &vop_create_desc, spec_create }, /* create */ 162 { &vop_mknod_desc, spec_mknod }, /* mknod */ 163 { &vop_open_desc, spec_open }, /* open */ 164 { &vop_close_desc, lfsspec_close }, /* close */ 165 { &vop_access_desc, ufs_access }, /* access */ 166 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 167 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 168 { &vop_read_desc, ufsspec_read }, /* read */ 169 { &vop_write_desc, ufsspec_write }, /* write */ 170 { &vop_ioctl_desc, spec_ioctl }, /* ioctl */ 171 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 172 { &vop_poll_desc, spec_poll }, /* poll */ 173 { &vop_kqfilter_desc, spec_kqfilter }, /* kqfilter */ 174 { &vop_revoke_desc, spec_revoke }, /* revoke */ 175 { &vop_mmap_desc, spec_mmap }, /* mmap */ 176 { &vop_fsync_desc, spec_fsync }, /* fsync */ 177 { &vop_seek_desc, spec_seek }, /* seek */ 178 { &vop_remove_desc, spec_remove }, /* remove */ 179 { &vop_link_desc, spec_link }, /* link */ 180 { &vop_rename_desc, spec_rename }, /* rename */ 181 { &vop_mkdir_desc, spec_mkdir }, /* mkdir */ 182 { &vop_rmdir_desc, spec_rmdir }, /* rmdir */ 183 { &vop_symlink_desc, spec_symlink }, /* symlink */ 184 { &vop_readdir_desc, spec_readdir }, /* readdir */ 185 { &vop_readlink_desc, spec_readlink }, /* readlink */ 186 { &vop_abortop_desc, spec_abortop }, /* abortop */ 187 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 188 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 189 { &vop_lock_desc, ufs_lock }, /* lock */ 190 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 191 { &vop_bmap_desc, spec_bmap }, /* bmap */ 192 { &vop_strategy_desc, spec_strategy }, /* strategy */ 193 { &vop_print_desc, ufs_print }, /* print */ 194 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 195 { &vop_pathconf_desc, spec_pathconf }, /* pathconf */ 196 { &vop_advlock_desc, spec_advlock }, /* advlock */ 197 { &vop_bwrite_desc, vn_bwrite }, /* bwrite */ 198 { &vop_getpages_desc, spec_getpages }, /* getpages */ 199 { &vop_putpages_desc, spec_putpages }, /* putpages */ 200 { NULL, NULL } 201 }; 202 const struct vnodeopv_desc lfs_specop_opv_desc = 203 { &lfs_specop_p, lfs_specop_entries }; 204 205 int (**lfs_fifoop_p)(void *); 206 const struct vnodeopv_entry_desc lfs_fifoop_entries[] = { 207 { &vop_default_desc, vn_default_error }, 208 { &vop_lookup_desc, vn_fifo_bypass }, /* lookup */ 209 { &vop_create_desc, vn_fifo_bypass }, /* create */ 210 { &vop_mknod_desc, vn_fifo_bypass }, /* mknod */ 211 { &vop_open_desc, vn_fifo_bypass }, /* open */ 212 { &vop_close_desc, lfsfifo_close }, /* close */ 213 { &vop_access_desc, ufs_access }, /* access */ 214 { &vop_getattr_desc, lfs_getattr }, /* getattr */ 215 { &vop_setattr_desc, lfs_setattr }, /* setattr */ 216 { &vop_read_desc, ufsfifo_read }, /* read */ 217 { &vop_write_desc, ufsfifo_write }, /* write */ 218 { &vop_ioctl_desc, vn_fifo_bypass }, /* ioctl */ 219 { &vop_fcntl_desc, ufs_fcntl }, /* fcntl */ 220 { &vop_poll_desc, vn_fifo_bypass }, /* poll */ 221 { &vop_kqfilter_desc, vn_fifo_bypass }, /* kqfilter */ 222 { &vop_revoke_desc, vn_fifo_bypass }, /* revoke */ 223 { &vop_mmap_desc, vn_fifo_bypass }, /* mmap */ 224 { &vop_fsync_desc, vn_fifo_bypass }, /* fsync */ 225 { &vop_seek_desc, vn_fifo_bypass }, /* seek */ 226 { &vop_remove_desc, vn_fifo_bypass }, /* remove */ 227 { &vop_link_desc, vn_fifo_bypass }, /* link */ 228 { &vop_rename_desc, vn_fifo_bypass }, /* rename */ 229 { &vop_mkdir_desc, vn_fifo_bypass }, /* mkdir */ 230 { &vop_rmdir_desc, vn_fifo_bypass }, /* rmdir */ 231 { &vop_symlink_desc, vn_fifo_bypass }, /* symlink */ 232 { &vop_readdir_desc, vn_fifo_bypass }, /* readdir */ 233 { &vop_readlink_desc, vn_fifo_bypass }, /* readlink */ 234 { &vop_abortop_desc, vn_fifo_bypass }, /* abortop */ 235 { &vop_inactive_desc, lfs_inactive }, /* inactive */ 236 { &vop_reclaim_desc, lfs_reclaim }, /* reclaim */ 237 { &vop_lock_desc, ufs_lock }, /* lock */ 238 { &vop_unlock_desc, ufs_unlock }, /* unlock */ 239 { &vop_bmap_desc, vn_fifo_bypass }, /* bmap */ 240 { &vop_strategy_desc, vn_fifo_bypass }, /* strategy */ 241 { &vop_print_desc, ufs_print }, /* print */ 242 { &vop_islocked_desc, ufs_islocked }, /* islocked */ 243 { &vop_pathconf_desc, vn_fifo_bypass }, /* pathconf */ 244 { &vop_advlock_desc, vn_fifo_bypass }, /* advlock */ 245 { &vop_bwrite_desc, lfs_bwrite }, /* bwrite */ 246 { &vop_putpages_desc, vn_fifo_bypass }, /* putpages */ 247 { NULL, NULL } 248 }; 249 const struct vnodeopv_desc lfs_fifoop_opv_desc = 250 { &lfs_fifoop_p, lfs_fifoop_entries }; 251 252 static int check_dirty(struct lfs *, struct vnode *, off_t, off_t, off_t, int, int, struct vm_page **); 253 254 #define LFS_READWRITE 255 #include <ufs/ufs/ufs_readwrite.c> 256 #undef LFS_READWRITE 257 258 /* 259 * Synch an open file. 260 */ 261 /* ARGSUSED */ 262 int 263 lfs_fsync(void *v) 264 { 265 struct vop_fsync_args /* { 266 struct vnode *a_vp; 267 kauth_cred_t a_cred; 268 int a_flags; 269 off_t offlo; 270 off_t offhi; 271 } */ *ap = v; 272 struct vnode *vp = ap->a_vp; 273 int error, wait; 274 struct inode *ip = VTOI(vp); 275 struct lfs *fs = ip->i_lfs; 276 277 /* If we're mounted read-only, don't try to sync. */ 278 if (fs->lfs_ronly) 279 return 0; 280 281 /* If a removed vnode is being cleaned, no need to sync here. */ 282 if ((ap->a_flags & FSYNC_RECLAIM) != 0 && ip->i_mode == 0) 283 return 0; 284 285 /* 286 * Trickle sync simply adds this vnode to the pager list, as if 287 * the pagedaemon had requested a pageout. 288 */ 289 if (ap->a_flags & FSYNC_LAZY) { 290 if (lfs_ignore_lazy_sync == 0) { 291 mutex_enter(&lfs_lock); 292 if (!(ip->i_flags & IN_PAGING)) { 293 ip->i_flags |= IN_PAGING; 294 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, 295 i_lfs_pchain); 296 } 297 wakeup(&lfs_writer_daemon); 298 mutex_exit(&lfs_lock); 299 } 300 return 0; 301 } 302 303 /* 304 * If a vnode is bring cleaned, flush it out before we try to 305 * reuse it. This prevents the cleaner from writing files twice 306 * in the same partial segment, causing an accounting underflow. 307 */ 308 if (ap->a_flags & FSYNC_RECLAIM && ip->i_flags & IN_CLEANING) { 309 lfs_vflush(vp); 310 } 311 312 wait = (ap->a_flags & FSYNC_WAIT); 313 do { 314 mutex_enter(vp->v_interlock); 315 error = VOP_PUTPAGES(vp, trunc_page(ap->a_offlo), 316 round_page(ap->a_offhi), 317 PGO_CLEANIT | (wait ? PGO_SYNCIO : 0)); 318 if (error == EAGAIN) { 319 mutex_enter(&lfs_lock); 320 mtsleep(&fs->lfs_avail, PCATCH | PUSER, "lfs_fsync", 321 hz / 100 + 1, &lfs_lock); 322 mutex_exit(&lfs_lock); 323 } 324 } while (error == EAGAIN); 325 if (error) 326 return error; 327 328 if ((ap->a_flags & FSYNC_DATAONLY) == 0) 329 error = lfs_update(vp, NULL, NULL, wait ? UPDATE_WAIT : 0); 330 331 if (error == 0 && ap->a_flags & FSYNC_CACHE) { 332 int l = 0; 333 error = VOP_IOCTL(ip->i_devvp, DIOCCACHESYNC, &l, FWRITE, 334 curlwp->l_cred); 335 } 336 if (wait && !VPISEMPTY(vp)) 337 LFS_SET_UINO(ip, IN_MODIFIED); 338 339 return error; 340 } 341 342 /* 343 * Take IN_ADIROP off, then call ufs_inactive. 344 */ 345 int 346 lfs_inactive(void *v) 347 { 348 struct vop_inactive_args /* { 349 struct vnode *a_vp; 350 } */ *ap = v; 351 352 lfs_unmark_vnode(ap->a_vp); 353 354 /* 355 * The Ifile is only ever inactivated on unmount. 356 * Streamline this process by not giving it more dirty blocks. 357 */ 358 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) { 359 mutex_enter(&lfs_lock); 360 LFS_CLR_UINO(VTOI(ap->a_vp), IN_ALLMOD); 361 mutex_exit(&lfs_lock); 362 VOP_UNLOCK(ap->a_vp); 363 return 0; 364 } 365 366 #ifdef DEBUG 367 /* 368 * This might happen on unmount. 369 * XXX If it happens at any other time, it should be a panic. 370 */ 371 if (ap->a_vp->v_uflag & VU_DIROP) { 372 struct inode *ip = VTOI(ap->a_vp); 373 printf("lfs_inactive: inactivating VU_DIROP? ino = %d\n", (int)ip->i_number); 374 } 375 #endif /* DIAGNOSTIC */ 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 VU_DIROP 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 VU_DIROP 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(&lfs_lock); 419 if (fs->lfs_dirops == 0) { 420 mutex_exit(&lfs_lock); 421 lfs_check(dvp, LFS_UNUSED_LBN, 0); 422 mutex_enter(&lfs_lock); 423 } 424 while (fs->lfs_writer) { 425 error = mtsleep(&fs->lfs_dirops, (PRIBIO + 1) | PCATCH, 426 "lfs_sdirop", 0, &lfs_lock); 427 if (error == EINTR) { 428 mutex_exit(&lfs_lock); 429 goto unreserve; 430 } 431 } 432 if (lfs_dirvcount > LFS_MAX_DIROP && fs->lfs_dirops == 0) { 433 wakeup(&lfs_writer_daemon); 434 mutex_exit(&lfs_lock); 435 preempt(); 436 goto restart; 437 } 438 439 if (lfs_dirvcount > LFS_MAX_DIROP) { 440 mutex_exit(&lfs_lock); 441 DLOG((DLOG_DIROP, "lfs_set_dirop: sleeping with dirops=%d, " 442 "dirvcount=%d\n", fs->lfs_dirops, lfs_dirvcount)); 443 if ((error = mtsleep(&lfs_dirvcount, 444 PCATCH | PUSER | PNORELOCK, "lfs_maxdirop", 0, 445 &lfs_lock)) != 0) { 446 goto unreserve; 447 } 448 goto restart; 449 } 450 451 ++fs->lfs_dirops; 452 /* fs->lfs_doifile = 1; */ /* XXX why? --ks */ 453 mutex_exit(&lfs_lock); 454 455 /* Hold a reference so SET_ENDOP will be happy */ 456 vref(dvp); 457 if (vp) { 458 vref(vp); 459 MARK_VNODE(vp); 460 } 461 462 MARK_VNODE(dvp); 463 return 0; 464 465 unreserve: 466 lfs_reserve(fs, dvp, vp, -LFS_NRESERVE(fs)); 467 return error; 468 } 469 470 /* 471 * Get a new vnode *before* adjusting the dirop count, to avoid a deadlock 472 * in getnewvnode(), if we have a stacked filesystem mounted on top 473 * of us. 474 * 475 * NB: this means we have to clear the new vnodes on error. Fortunately 476 * SET_ENDOP is there to do that for us. 477 */ 478 static int 479 lfs_set_dirop_create(struct vnode *dvp, struct vnode **vpp) 480 { 481 int error; 482 struct lfs *fs; 483 484 fs = VFSTOUFS(dvp->v_mount)->um_lfs; 485 ASSERT_NO_SEGLOCK(fs); 486 if (fs->lfs_ronly) 487 return EROFS; 488 if (vpp == NULL) { 489 return lfs_set_dirop(dvp, NULL); 490 } 491 error = getnewvnode(VT_LFS, dvp->v_mount, lfs_vnodeop_p, NULL, vpp); 492 if (error) { 493 DLOG((DLOG_ALLOC, "lfs_set_dirop_create: dvp %p error %d\n", 494 dvp, error)); 495 return error; 496 } 497 if ((error = lfs_set_dirop(dvp, NULL)) != 0) { 498 ungetnewvnode(*vpp); 499 *vpp = NULL; 500 return error; 501 } 502 return 0; 503 } 504 505 #define SET_ENDOP_BASE(fs, dvp, str) \ 506 do { \ 507 mutex_enter(&lfs_lock); \ 508 --(fs)->lfs_dirops; \ 509 if (!(fs)->lfs_dirops) { \ 510 if ((fs)->lfs_nadirop) { \ 511 panic("SET_ENDOP: %s: no dirops but " \ 512 " nadirop=%d", (str), \ 513 (fs)->lfs_nadirop); \ 514 } \ 515 wakeup(&(fs)->lfs_writer); \ 516 mutex_exit(&lfs_lock); \ 517 lfs_check((dvp), LFS_UNUSED_LBN, 0); \ 518 } else \ 519 mutex_exit(&lfs_lock); \ 520 } while(0) 521 #define SET_ENDOP_CREATE(fs, dvp, nvpp, str) \ 522 do { \ 523 UNMARK_VNODE(dvp); \ 524 if (nvpp && *nvpp) \ 525 UNMARK_VNODE(*nvpp); \ 526 /* Check for error return to stem vnode leakage */ \ 527 if (nvpp && *nvpp && !((*nvpp)->v_uflag & VU_DIROP)) \ 528 ungetnewvnode(*(nvpp)); \ 529 SET_ENDOP_BASE((fs), (dvp), (str)); \ 530 lfs_reserve((fs), (dvp), NULL, -LFS_NRESERVE(fs)); \ 531 vrele(dvp); \ 532 } while(0) 533 #define SET_ENDOP_CREATE_AP(ap, str) \ 534 SET_ENDOP_CREATE(VTOI((ap)->a_dvp)->i_lfs, (ap)->a_dvp, \ 535 (ap)->a_vpp, (str)) 536 #define SET_ENDOP_REMOVE(fs, dvp, ovp, str) \ 537 do { \ 538 UNMARK_VNODE(dvp); \ 539 if (ovp) \ 540 UNMARK_VNODE(ovp); \ 541 SET_ENDOP_BASE((fs), (dvp), (str)); \ 542 lfs_reserve((fs), (dvp), (ovp), -LFS_NRESERVE(fs)); \ 543 vrele(dvp); \ 544 if (ovp) \ 545 vrele(ovp); \ 546 } while(0) 547 548 void 549 lfs_mark_vnode(struct vnode *vp) 550 { 551 struct inode *ip = VTOI(vp); 552 struct lfs *fs = ip->i_lfs; 553 554 mutex_enter(&lfs_lock); 555 if (!(ip->i_flag & IN_ADIROP)) { 556 if (!(vp->v_uflag & VU_DIROP)) { 557 mutex_enter(vp->v_interlock); 558 if (lfs_vref(vp) != 0) 559 panic("lfs_mark_vnode: could not vref"); 560 ++lfs_dirvcount; 561 ++fs->lfs_dirvcount; 562 TAILQ_INSERT_TAIL(&fs->lfs_dchainhd, ip, i_lfs_dchain); 563 vp->v_uflag |= VU_DIROP; 564 } 565 ++fs->lfs_nadirop; 566 ip->i_flag &= ~IN_CDIROP; 567 ip->i_flag |= IN_ADIROP; 568 } else 569 KASSERT(vp->v_uflag & VU_DIROP); 570 mutex_exit(&lfs_lock); 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(&lfs_lock); 581 --ip->i_lfs->lfs_nadirop; 582 mutex_exit(&lfs_lock); 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 struct ufs_lookup_results *ulr; 624 625 /* XXX should handle this material another way */ 626 ulr = &VTOI(ap->a_dvp)->i_crap; 627 UFS_CHECK_CRAPCOUNTER(VTOI(ap->a_dvp)); 628 629 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 630 vput(ap->a_dvp); 631 return error; 632 } 633 error = ufs_makeinode(MAKEIMODE(vap->va_type, vap->va_mode), 634 ap->a_dvp, ulr, vpp, ap->a_cnp); 635 636 /* Either way we're done with the dirop at this point */ 637 SET_ENDOP_CREATE_AP(ap, "mknod"); 638 639 if (error) 640 return (error); 641 642 ip = VTOI(*vpp); 643 mp = (*vpp)->v_mount; 644 ino = ip->i_number; 645 ip->i_flag |= IN_ACCESS | IN_CHANGE | IN_UPDATE; 646 if (vap->va_rdev != VNOVAL) { 647 /* 648 * Want to be able to use this to make badblock 649 * inodes, so don't truncate the dev number. 650 */ 651 #if 0 652 ip->i_ffs1_rdev = ufs_rw32(vap->va_rdev, 653 UFS_MPNEEDSWAP((*vpp)->v_mount)); 654 #else 655 ip->i_ffs1_rdev = vap->va_rdev; 656 #endif 657 } 658 659 /* 660 * Call fsync to write the vnode so that we don't have to deal with 661 * flushing it when it's marked VU_DIROP|VI_XLOCK. 662 * 663 * XXX KS - If we can't flush we also can't call vgone(), so must 664 * return. But, that leaves this vnode in limbo, also not good. 665 * Can this ever happen (barring hardware failure)? 666 */ 667 if ((error = VOP_FSYNC(*vpp, NOCRED, FSYNC_WAIT, 0, 0)) != 0) { 668 panic("lfs_mknod: couldn't fsync (ino %llu)", 669 (unsigned long long)ino); 670 /* return (error); */ 671 } 672 /* 673 * Remove vnode so that it will be reloaded by VFS_VGET and 674 * checked to see if it is an alias of an existing entry in 675 * the inode cache. 676 */ 677 /* Used to be vput, but that causes us to call VOP_INACTIVE twice. */ 678 679 VOP_UNLOCK(*vpp); 680 (*vpp)->v_type = VNON; 681 vgone(*vpp); 682 error = VFS_VGET(mp, ino, vpp); 683 684 if (error != 0) { 685 *vpp = NULL; 686 return (error); 687 } 688 return (0); 689 } 690 691 int 692 lfs_create(void *v) 693 { 694 struct vop_create_args /* { 695 struct vnode *a_dvp; 696 struct vnode **a_vpp; 697 struct componentname *a_cnp; 698 struct vattr *a_vap; 699 } */ *ap = v; 700 int error; 701 702 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 703 vput(ap->a_dvp); 704 return error; 705 } 706 error = ufs_create(ap); 707 SET_ENDOP_CREATE_AP(ap, "create"); 708 return (error); 709 } 710 711 int 712 lfs_mkdir(void *v) 713 { 714 struct vop_mkdir_args /* { 715 struct vnode *a_dvp; 716 struct vnode **a_vpp; 717 struct componentname *a_cnp; 718 struct vattr *a_vap; 719 } */ *ap = v; 720 int error; 721 722 if ((error = SET_DIROP_CREATE(ap->a_dvp, ap->a_vpp)) != 0) { 723 vput(ap->a_dvp); 724 return error; 725 } 726 error = ufs_mkdir(ap); 727 SET_ENDOP_CREATE_AP(ap, "mkdir"); 728 return (error); 729 } 730 731 int 732 lfs_remove(void *v) 733 { 734 struct vop_remove_args /* { 735 struct vnode *a_dvp; 736 struct vnode *a_vp; 737 struct componentname *a_cnp; 738 } */ *ap = v; 739 struct vnode *dvp, *vp; 740 struct inode *ip; 741 int error; 742 743 dvp = ap->a_dvp; 744 vp = ap->a_vp; 745 ip = VTOI(vp); 746 if ((error = SET_DIROP_REMOVE(dvp, vp)) != 0) { 747 if (dvp == vp) 748 vrele(vp); 749 else 750 vput(vp); 751 vput(dvp); 752 return error; 753 } 754 error = ufs_remove(ap); 755 if (ip->i_nlink == 0) 756 lfs_orphan(ip->i_lfs, ip->i_number); 757 SET_ENDOP_REMOVE(ip->i_lfs, dvp, ap->a_vp, "remove"); 758 return (error); 759 } 760 761 int 762 lfs_rmdir(void *v) 763 { 764 struct vop_rmdir_args /* { 765 struct vnodeop_desc *a_desc; 766 struct vnode *a_dvp; 767 struct vnode *a_vp; 768 struct componentname *a_cnp; 769 } */ *ap = v; 770 struct vnode *vp; 771 struct inode *ip; 772 int error; 773 774 vp = ap->a_vp; 775 ip = VTOI(vp); 776 if ((error = SET_DIROP_REMOVE(ap->a_dvp, ap->a_vp)) != 0) { 777 if (ap->a_dvp == vp) 778 vrele(ap->a_dvp); 779 else 780 vput(ap->a_dvp); 781 vput(vp); 782 return error; 783 } 784 error = ufs_rmdir(ap); 785 if (ip->i_nlink == 0) 786 lfs_orphan(ip->i_lfs, ip->i_number); 787 SET_ENDOP_REMOVE(ip->i_lfs, ap->a_dvp, ap->a_vp, "rmdir"); 788 return (error); 789 } 790 791 int 792 lfs_link(void *v) 793 { 794 struct vop_link_args /* { 795 struct vnode *a_dvp; 796 struct vnode *a_vp; 797 struct componentname *a_cnp; 798 } */ *ap = v; 799 int error; 800 struct vnode **vpp = NULL; 801 802 if ((error = SET_DIROP_CREATE(ap->a_dvp, vpp)) != 0) { 803 vput(ap->a_dvp); 804 return error; 805 } 806 error = ufs_link(ap); 807 SET_ENDOP_CREATE(VTOI(ap->a_dvp)->i_lfs, ap->a_dvp, vpp, "link"); 808 return (error); 809 } 810 811 int 812 lfs_rename(void *v) 813 { 814 struct vop_rename_args /* { 815 struct vnode *a_fdvp; 816 struct vnode *a_fvp; 817 struct componentname *a_fcnp; 818 struct vnode *a_tdvp; 819 struct vnode *a_tvp; 820 struct componentname *a_tcnp; 821 } */ *ap = v; 822 struct vnode *tvp, *fvp, *tdvp, *fdvp; 823 struct componentname *tcnp, *fcnp; 824 int error; 825 struct lfs *fs; 826 827 fs = VTOI(ap->a_fdvp)->i_lfs; 828 tvp = ap->a_tvp; 829 tdvp = ap->a_tdvp; 830 tcnp = ap->a_tcnp; 831 fvp = ap->a_fvp; 832 fdvp = ap->a_fdvp; 833 fcnp = ap->a_fcnp; 834 835 /* 836 * Check for cross-device rename. 837 * If it is, we don't want to set dirops, just error out. 838 * (In particular note that MARK_VNODE(tdvp) will DTWT on 839 * a cross-device rename.) 840 * 841 * Copied from ufs_rename. 842 */ 843 if ((fvp->v_mount != tdvp->v_mount) || 844 (tvp && (fvp->v_mount != tvp->v_mount))) { 845 error = EXDEV; 846 goto errout; 847 } 848 849 /* 850 * Check to make sure we're not renaming a vnode onto itself 851 * (deleting a hard link by renaming one name onto another); 852 * if we are we can't recursively call VOP_REMOVE since that 853 * would leave us with an unaccounted-for number of live dirops. 854 * 855 * Inline the relevant section of ufs_rename here, *before* 856 * calling SET_DIROP_REMOVE. 857 */ 858 if (tvp && ((VTOI(tvp)->i_flags & (IMMUTABLE | APPEND)) || 859 (VTOI(tdvp)->i_flags & APPEND))) { 860 error = EPERM; 861 goto errout; 862 } 863 if (fvp == tvp) { 864 if (fvp->v_type == VDIR) { 865 error = EINVAL; 866 goto errout; 867 } 868 869 /* Release destination completely. */ 870 VOP_ABORTOP(tdvp, tcnp); 871 vput(tdvp); 872 vput(tvp); 873 874 /* Delete source. */ 875 vrele(fvp); 876 fcnp->cn_flags &= ~(MODMASK); 877 fcnp->cn_flags |= LOCKPARENT | LOCKLEAF; 878 fcnp->cn_nameiop = DELETE; 879 vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY); 880 if ((error = relookup(fdvp, &fvp, fcnp, 0))) { 881 vput(fdvp); 882 return (error); 883 } 884 return (VOP_REMOVE(fdvp, fvp, fcnp)); 885 } 886 887 if ((error = SET_DIROP_REMOVE(tdvp, tvp)) != 0) 888 goto errout; 889 MARK_VNODE(fdvp); 890 MARK_VNODE(fvp); 891 892 error = ufs_rename(ap); 893 UNMARK_VNODE(fdvp); 894 UNMARK_VNODE(fvp); 895 SET_ENDOP_REMOVE(fs, tdvp, tvp, "rename"); 896 return (error); 897 898 errout: 899 VOP_ABORTOP(tdvp, ap->a_tcnp); /* XXX, why not in NFS? */ 900 if (tdvp == tvp) 901 vrele(tdvp); 902 else 903 vput(tdvp); 904 if (tvp) 905 vput(tvp); 906 VOP_ABORTOP(fdvp, ap->a_fcnp); /* XXX, why not in NFS? */ 907 vrele(fdvp); 908 vrele(fvp); 909 return (error); 910 } 911 912 /* XXX hack to avoid calling ITIMES in getattr */ 913 int 914 lfs_getattr(void *v) 915 { 916 struct vop_getattr_args /* { 917 struct vnode *a_vp; 918 struct vattr *a_vap; 919 kauth_cred_t a_cred; 920 } */ *ap = v; 921 struct vnode *vp = ap->a_vp; 922 struct inode *ip = VTOI(vp); 923 struct vattr *vap = ap->a_vap; 924 struct lfs *fs = ip->i_lfs; 925 /* 926 * Copy from inode table 927 */ 928 vap->va_fsid = ip->i_dev; 929 vap->va_fileid = ip->i_number; 930 vap->va_mode = ip->i_mode & ~IFMT; 931 vap->va_nlink = ip->i_nlink; 932 vap->va_uid = ip->i_uid; 933 vap->va_gid = ip->i_gid; 934 vap->va_rdev = (dev_t)ip->i_ffs1_rdev; 935 vap->va_size = vp->v_size; 936 vap->va_atime.tv_sec = ip->i_ffs1_atime; 937 vap->va_atime.tv_nsec = ip->i_ffs1_atimensec; 938 vap->va_mtime.tv_sec = ip->i_ffs1_mtime; 939 vap->va_mtime.tv_nsec = ip->i_ffs1_mtimensec; 940 vap->va_ctime.tv_sec = ip->i_ffs1_ctime; 941 vap->va_ctime.tv_nsec = ip->i_ffs1_ctimensec; 942 vap->va_flags = ip->i_flags; 943 vap->va_gen = ip->i_gen; 944 /* this doesn't belong here */ 945 if (vp->v_type == VBLK) 946 vap->va_blocksize = BLKDEV_IOSIZE; 947 else if (vp->v_type == VCHR) 948 vap->va_blocksize = MAXBSIZE; 949 else 950 vap->va_blocksize = vp->v_mount->mnt_stat.f_iosize; 951 vap->va_bytes = fsbtob(fs, (u_quad_t)ip->i_lfs_effnblks); 952 vap->va_type = vp->v_type; 953 vap->va_filerev = ip->i_modrev; 954 return (0); 955 } 956 957 /* 958 * Check to make sure the inode blocks won't choke the buffer 959 * cache, then call ufs_setattr as usual. 960 */ 961 int 962 lfs_setattr(void *v) 963 { 964 struct vop_setattr_args /* { 965 struct vnode *a_vp; 966 struct vattr *a_vap; 967 kauth_cred_t a_cred; 968 } */ *ap = v; 969 struct vnode *vp = ap->a_vp; 970 971 lfs_check(vp, LFS_UNUSED_LBN, 0); 972 return ufs_setattr(v); 973 } 974 975 /* 976 * Release the block we hold on lfs_newseg wrapping. Called on file close, 977 * or explicitly from LFCNWRAPGO. Called with the interlock held. 978 */ 979 static int 980 lfs_wrapgo(struct lfs *fs, struct inode *ip, int waitfor) 981 { 982 if (fs->lfs_stoplwp != curlwp) 983 return EBUSY; 984 985 fs->lfs_stoplwp = NULL; 986 cv_signal(&fs->lfs_stopcv); 987 988 KASSERT(fs->lfs_nowrap > 0); 989 if (fs->lfs_nowrap <= 0) { 990 return 0; 991 } 992 993 if (--fs->lfs_nowrap == 0) { 994 log(LOG_NOTICE, "%s: re-enabled log wrap\n", fs->lfs_fsmnt); 995 wakeup(&fs->lfs_wrappass); 996 lfs_wakeup_cleaner(fs); 997 } 998 if (waitfor) { 999 mtsleep(&fs->lfs_nextseg, PCATCH | PUSER, "segment", 1000 0, &lfs_lock); 1001 } 1002 1003 return 0; 1004 } 1005 1006 /* 1007 * Close called 1008 */ 1009 /* ARGSUSED */ 1010 int 1011 lfs_close(void *v) 1012 { 1013 struct vop_close_args /* { 1014 struct vnode *a_vp; 1015 int a_fflag; 1016 kauth_cred_t a_cred; 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(&lfs_lock); 1025 log(LOG_NOTICE, "lfs_close: releasing log wrap control\n"); 1026 lfs_wrapgo(fs, ip, 0); 1027 mutex_exit(&lfs_lock); 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 } */ *ap = v; 1053 struct vnode *vp; 1054 struct inode *ip; 1055 1056 vp = ap->a_vp; 1057 ip = VTOI(vp); 1058 if (vp->v_usecount > 1) { 1059 LFS_ITIMES(ip, NULL, NULL, NULL); 1060 } 1061 return (VOCALL (spec_vnodeop_p, VOFFSET(vop_close), ap)); 1062 } 1063 1064 /* 1065 * Close wrapper for fifo's. 1066 * 1067 * Update the times on the inode then do device close. 1068 */ 1069 int 1070 lfsfifo_close(void *v) 1071 { 1072 struct vop_close_args /* { 1073 struct vnode *a_vp; 1074 int a_fflag; 1075 kauth_cred_ a_cred; 1076 } */ *ap = v; 1077 struct vnode *vp; 1078 struct inode *ip; 1079 1080 vp = ap->a_vp; 1081 ip = VTOI(vp); 1082 if (ap->a_vp->v_usecount > 1) { 1083 LFS_ITIMES(ip, NULL, NULL, NULL); 1084 } 1085 return (VOCALL (fifo_vnodeop_p, VOFFSET(vop_close), ap)); 1086 } 1087 1088 /* 1089 * Reclaim an inode so that it can be used for other purposes. 1090 */ 1091 1092 int 1093 lfs_reclaim(void *v) 1094 { 1095 struct vop_reclaim_args /* { 1096 struct vnode *a_vp; 1097 } */ *ap = v; 1098 struct vnode *vp = ap->a_vp; 1099 struct inode *ip = VTOI(vp); 1100 struct lfs *fs = ip->i_lfs; 1101 int error; 1102 1103 /* 1104 * The inode must be freed and updated before being removed 1105 * from its hash chain. Other threads trying to gain a hold 1106 * on the inode will be stalled because it is locked (VI_XLOCK). 1107 */ 1108 if (ip->i_nlink <= 0 && (vp->v_mount->mnt_flag & MNT_RDONLY) == 0) 1109 lfs_vfree(vp, ip->i_number, ip->i_omode); 1110 1111 mutex_enter(&lfs_lock); 1112 LFS_CLR_UINO(ip, IN_ALLMOD); 1113 mutex_exit(&lfs_lock); 1114 if ((error = ufs_reclaim(vp))) 1115 return (error); 1116 1117 /* 1118 * Take us off the paging and/or dirop queues if we were on them. 1119 * We shouldn't be on them. 1120 */ 1121 mutex_enter(&lfs_lock); 1122 if (ip->i_flags & IN_PAGING) { 1123 log(LOG_WARNING, "%s: reclaimed vnode is IN_PAGING\n", 1124 fs->lfs_fsmnt); 1125 ip->i_flags &= ~IN_PAGING; 1126 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 1127 } 1128 if (vp->v_uflag & VU_DIROP) { 1129 panic("reclaimed vnode is VU_DIROP"); 1130 vp->v_uflag &= ~VU_DIROP; 1131 TAILQ_REMOVE(&fs->lfs_dchainhd, ip, i_lfs_dchain); 1132 } 1133 mutex_exit(&lfs_lock); 1134 1135 pool_put(&lfs_dinode_pool, ip->i_din.ffs1_din); 1136 lfs_deregister_all(vp); 1137 pool_put(&lfs_inoext_pool, ip->inode_ext.lfs); 1138 ip->inode_ext.lfs = NULL; 1139 genfs_node_destroy(vp); 1140 pool_put(&lfs_inode_pool, vp->v_data); 1141 vp->v_data = NULL; 1142 return (0); 1143 } 1144 1145 /* 1146 * Read a block from a storage device. 1147 * In order to avoid reading blocks that are in the process of being 1148 * written by the cleaner---and hence are not mutexed by the normal 1149 * buffer cache / page cache mechanisms---check for collisions before 1150 * reading. 1151 * 1152 * We inline ufs_strategy to make sure that the VOP_BMAP occurs *before* 1153 * the active cleaner test. 1154 * 1155 * XXX This code assumes that lfs_markv makes synchronous checkpoints. 1156 */ 1157 int 1158 lfs_strategy(void *v) 1159 { 1160 struct vop_strategy_args /* { 1161 struct vnode *a_vp; 1162 struct buf *a_bp; 1163 } */ *ap = v; 1164 struct buf *bp; 1165 struct lfs *fs; 1166 struct vnode *vp; 1167 struct inode *ip; 1168 daddr_t tbn; 1169 #define MAXLOOP 25 1170 int i, sn, error, slept, loopcount; 1171 1172 bp = ap->a_bp; 1173 vp = ap->a_vp; 1174 ip = VTOI(vp); 1175 fs = ip->i_lfs; 1176 1177 /* lfs uses its strategy routine only for read */ 1178 KASSERT(bp->b_flags & B_READ); 1179 1180 if (vp->v_type == VBLK || vp->v_type == VCHR) 1181 panic("lfs_strategy: spec"); 1182 KASSERT(bp->b_bcount != 0); 1183 if (bp->b_blkno == bp->b_lblkno) { 1184 error = VOP_BMAP(vp, bp->b_lblkno, NULL, &bp->b_blkno, 1185 NULL); 1186 if (error) { 1187 bp->b_error = error; 1188 bp->b_resid = bp->b_bcount; 1189 biodone(bp); 1190 return (error); 1191 } 1192 if ((long)bp->b_blkno == -1) /* no valid data */ 1193 clrbuf(bp); 1194 } 1195 if ((long)bp->b_blkno < 0) { /* block is not on disk */ 1196 bp->b_resid = bp->b_bcount; 1197 biodone(bp); 1198 return (0); 1199 } 1200 1201 slept = 1; 1202 loopcount = 0; 1203 mutex_enter(&lfs_lock); 1204 while (slept && fs->lfs_seglock) { 1205 mutex_exit(&lfs_lock); 1206 /* 1207 * Look through list of intervals. 1208 * There will only be intervals to look through 1209 * if the cleaner holds the seglock. 1210 * Since the cleaner is synchronous, we can trust 1211 * the list of intervals to be current. 1212 */ 1213 tbn = dbtofsb(fs, bp->b_blkno); 1214 sn = dtosn(fs, tbn); 1215 slept = 0; 1216 for (i = 0; i < fs->lfs_cleanind; i++) { 1217 if (sn == dtosn(fs, fs->lfs_cleanint[i]) && 1218 tbn >= fs->lfs_cleanint[i]) { 1219 DLOG((DLOG_CLEAN, 1220 "lfs_strategy: ino %d lbn %" PRId64 1221 " ind %d sn %d fsb %" PRIx32 1222 " given sn %d fsb %" PRIx64 "\n", 1223 ip->i_number, bp->b_lblkno, i, 1224 dtosn(fs, fs->lfs_cleanint[i]), 1225 fs->lfs_cleanint[i], sn, tbn)); 1226 DLOG((DLOG_CLEAN, 1227 "lfs_strategy: sleeping on ino %d lbn %" 1228 PRId64 "\n", ip->i_number, bp->b_lblkno)); 1229 mutex_enter(&lfs_lock); 1230 if (LFS_SEGLOCK_HELD(fs) && fs->lfs_iocount) { 1231 /* 1232 * Cleaner can't wait for itself. 1233 * Instead, wait for the blocks 1234 * to be written to disk. 1235 * XXX we need pribio in the test 1236 * XXX here. 1237 */ 1238 mtsleep(&fs->lfs_iocount, 1239 (PRIBIO + 1) | PNORELOCK, 1240 "clean2", hz/10 + 1, 1241 &lfs_lock); 1242 slept = 1; 1243 ++loopcount; 1244 break; 1245 } else if (fs->lfs_seglock) { 1246 mtsleep(&fs->lfs_seglock, 1247 (PRIBIO + 1) | PNORELOCK, 1248 "clean1", 0, 1249 &lfs_lock); 1250 slept = 1; 1251 break; 1252 } 1253 mutex_exit(&lfs_lock); 1254 } 1255 } 1256 mutex_enter(&lfs_lock); 1257 if (loopcount > MAXLOOP) { 1258 printf("lfs_strategy: breaking out of clean2 loop\n"); 1259 break; 1260 } 1261 } 1262 mutex_exit(&lfs_lock); 1263 1264 vp = ip->i_devvp; 1265 VOP_STRATEGY(vp, bp); 1266 return (0); 1267 } 1268 1269 /* 1270 * Inline lfs_segwrite/lfs_writevnodes, but just for dirops. 1271 * Technically this is a checkpoint (the on-disk state is valid) 1272 * even though we are leaving out all the file data. 1273 */ 1274 int 1275 lfs_flush_dirops(struct lfs *fs) 1276 { 1277 struct inode *ip, *nip; 1278 struct vnode *vp; 1279 extern int lfs_dostats; 1280 struct segment *sp; 1281 int flags = 0; 1282 int error = 0; 1283 1284 ASSERT_MAYBE_SEGLOCK(fs); 1285 KASSERT(fs->lfs_nadirop == 0); 1286 1287 if (fs->lfs_ronly) 1288 return EROFS; 1289 1290 mutex_enter(&lfs_lock); 1291 if (TAILQ_FIRST(&fs->lfs_dchainhd) == NULL) { 1292 mutex_exit(&lfs_lock); 1293 return 0; 1294 } else 1295 mutex_exit(&lfs_lock); 1296 1297 if (lfs_dostats) 1298 ++lfs_stats.flush_invoked; 1299 1300 lfs_imtime(fs); 1301 lfs_seglock(fs, flags); 1302 sp = fs->lfs_sp; 1303 1304 /* 1305 * lfs_writevnodes, optimized to get dirops out of the way. 1306 * Only write dirops, and don't flush files' pages, only 1307 * blocks from the directories. 1308 * 1309 * We don't need to vref these files because they are 1310 * dirops and so hold an extra reference until the 1311 * segunlock clears them of that status. 1312 * 1313 * We don't need to check for IN_ADIROP because we know that 1314 * no dirops are active. 1315 * 1316 */ 1317 mutex_enter(&lfs_lock); 1318 for (ip = TAILQ_FIRST(&fs->lfs_dchainhd); ip != NULL; ip = nip) { 1319 nip = TAILQ_NEXT(ip, i_lfs_dchain); 1320 mutex_exit(&lfs_lock); 1321 vp = ITOV(ip); 1322 1323 KASSERT((ip->i_flag & IN_ADIROP) == 0); 1324 KASSERT(vp->v_uflag & VU_DIROP); 1325 KASSERT(!(vp->v_iflag & VI_XLOCK)); 1326 1327 /* 1328 * All writes to directories come from dirops; all 1329 * writes to files' direct blocks go through the page 1330 * cache, which we're not touching. Reads to files 1331 * and/or directories will not be affected by writing 1332 * directory blocks inodes and file inodes. So we don't 1333 * really need to lock. 1334 */ 1335 if (vp->v_iflag & VI_XLOCK) { 1336 mutex_enter(&lfs_lock); 1337 continue; 1338 } 1339 /* XXX see below 1340 * waslocked = VOP_ISLOCKED(vp); 1341 */ 1342 if (vp->v_type != VREG && 1343 ((ip->i_flag & IN_ALLMOD) || !VPISEMPTY(vp))) { 1344 error = lfs_writefile(fs, sp, vp); 1345 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1346 !(ip->i_flag & IN_ALLMOD)) { 1347 mutex_enter(&lfs_lock); 1348 LFS_SET_UINO(ip, IN_MODIFIED); 1349 mutex_exit(&lfs_lock); 1350 } 1351 if (error && (sp->seg_flags & SEGM_SINGLE)) { 1352 mutex_enter(&lfs_lock); 1353 error = EAGAIN; 1354 break; 1355 } 1356 } 1357 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1358 error = lfs_writeinode(fs, sp, ip); 1359 mutex_enter(&lfs_lock); 1360 if (error && (sp->seg_flags & SEGM_SINGLE)) { 1361 error = EAGAIN; 1362 break; 1363 } 1364 1365 /* 1366 * We might need to update these inodes again, 1367 * for example, if they have data blocks to write. 1368 * Make sure that after this flush, they are still 1369 * marked IN_MODIFIED so that we don't forget to 1370 * write them. 1371 */ 1372 /* XXX only for non-directories? --KS */ 1373 LFS_SET_UINO(ip, IN_MODIFIED); 1374 } 1375 mutex_exit(&lfs_lock); 1376 /* We've written all the dirops there are */ 1377 ((SEGSUM *)(sp->segsum))->ss_flags &= ~(SS_CONT); 1378 lfs_finalize_fs_seguse(fs); 1379 (void) lfs_writeseg(fs, sp); 1380 lfs_segunlock(fs); 1381 1382 return error; 1383 } 1384 1385 /* 1386 * Flush all vnodes for which the pagedaemon has requested pageouts. 1387 * Skip over any files that are marked VU_DIROP (since lfs_flush_dirop() 1388 * has just run, this would be an error). If we have to skip a vnode 1389 * for any reason, just skip it; if we have to wait for the cleaner, 1390 * abort. The writer daemon will call us again later. 1391 */ 1392 int 1393 lfs_flush_pchain(struct lfs *fs) 1394 { 1395 struct inode *ip, *nip; 1396 struct vnode *vp; 1397 extern int lfs_dostats; 1398 struct segment *sp; 1399 int error, error2; 1400 1401 ASSERT_NO_SEGLOCK(fs); 1402 1403 if (fs->lfs_ronly) 1404 return EROFS; 1405 1406 mutex_enter(&lfs_lock); 1407 if (TAILQ_FIRST(&fs->lfs_pchainhd) == NULL) { 1408 mutex_exit(&lfs_lock); 1409 return 0; 1410 } else 1411 mutex_exit(&lfs_lock); 1412 1413 /* Get dirops out of the way */ 1414 if ((error = lfs_flush_dirops(fs)) != 0) 1415 return error; 1416 1417 if (lfs_dostats) 1418 ++lfs_stats.flush_invoked; 1419 1420 /* 1421 * Inline lfs_segwrite/lfs_writevnodes, but just for pageouts. 1422 */ 1423 lfs_imtime(fs); 1424 lfs_seglock(fs, 0); 1425 sp = fs->lfs_sp; 1426 1427 /* 1428 * lfs_writevnodes, optimized to clear pageout requests. 1429 * Only write non-dirop files that are in the pageout queue. 1430 * We're very conservative about what we write; we want to be 1431 * fast and async. 1432 */ 1433 mutex_enter(&lfs_lock); 1434 top: 1435 for (ip = TAILQ_FIRST(&fs->lfs_pchainhd); ip != NULL; ip = nip) { 1436 nip = TAILQ_NEXT(ip, i_lfs_pchain); 1437 vp = ITOV(ip); 1438 1439 if (!(ip->i_flags & IN_PAGING)) 1440 goto top; 1441 1442 mutex_enter(vp->v_interlock); 1443 if ((vp->v_iflag & VI_XLOCK) || (vp->v_uflag & VU_DIROP) != 0) { 1444 mutex_exit(vp->v_interlock); 1445 continue; 1446 } 1447 if (vp->v_type != VREG) { 1448 mutex_exit(vp->v_interlock); 1449 continue; 1450 } 1451 if (lfs_vref(vp)) 1452 continue; 1453 mutex_exit(&lfs_lock); 1454 1455 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_RETRY) != 0) { 1456 lfs_vunref(vp); 1457 mutex_enter(&lfs_lock); 1458 continue; 1459 } 1460 1461 error = lfs_writefile(fs, sp, vp); 1462 if (!VPISEMPTY(vp) && !WRITEINPROG(vp) && 1463 !(ip->i_flag & IN_ALLMOD)) { 1464 mutex_enter(&lfs_lock); 1465 LFS_SET_UINO(ip, IN_MODIFIED); 1466 mutex_exit(&lfs_lock); 1467 } 1468 KDASSERT(ip->i_number != LFS_IFILE_INUM); 1469 error2 = lfs_writeinode(fs, sp, ip); 1470 1471 VOP_UNLOCK(vp); 1472 lfs_vunref(vp); 1473 1474 if (error == EAGAIN || error2 == EAGAIN) { 1475 lfs_writeseg(fs, sp); 1476 mutex_enter(&lfs_lock); 1477 break; 1478 } 1479 mutex_enter(&lfs_lock); 1480 } 1481 mutex_exit(&lfs_lock); 1482 (void) lfs_writeseg(fs, sp); 1483 lfs_segunlock(fs); 1484 1485 return 0; 1486 } 1487 1488 /* 1489 * Provide a fcntl interface to sys_lfs_{segwait,bmapv,markv}. 1490 */ 1491 int 1492 lfs_fcntl(void *v) 1493 { 1494 struct vop_fcntl_args /* { 1495 struct vnode *a_vp; 1496 u_int a_command; 1497 void * a_data; 1498 int a_fflag; 1499 kauth_cred_t a_cred; 1500 } */ *ap = v; 1501 struct timeval tv; 1502 struct timeval *tvp; 1503 BLOCK_INFO *blkiov; 1504 CLEANERINFO *cip; 1505 SEGUSE *sup; 1506 int blkcnt, error, oclean; 1507 size_t fh_size; 1508 struct lfs_fcntl_markv blkvp; 1509 struct lwp *l; 1510 fsid_t *fsidp; 1511 struct lfs *fs; 1512 struct buf *bp; 1513 fhandle_t *fhp; 1514 daddr_t off; 1515 1516 /* Only respect LFS fcntls on fs root or Ifile */ 1517 if (VTOI(ap->a_vp)->i_number != ROOTINO && 1518 VTOI(ap->a_vp)->i_number != LFS_IFILE_INUM) { 1519 return ufs_fcntl(v); 1520 } 1521 1522 /* Avoid locking a draining lock */ 1523 if (ap->a_vp->v_mount->mnt_iflag & IMNT_UNMOUNT) { 1524 return ESHUTDOWN; 1525 } 1526 1527 /* LFS control and monitoring fcntls are available only to root */ 1528 l = curlwp; 1529 if (((ap->a_command & 0xff00) >> 8) == 'L' && 1530 (error = kauth_authorize_generic(l->l_cred, KAUTH_GENERIC_ISSUSER, 1531 NULL)) != 0) 1532 return (error); 1533 1534 fs = VTOI(ap->a_vp)->i_lfs; 1535 fsidp = &ap->a_vp->v_mount->mnt_stat.f_fsidx; 1536 1537 error = 0; 1538 switch ((int)ap->a_command) { 1539 case LFCNSEGWAITALL_COMPAT_50: 1540 case LFCNSEGWAITALL_COMPAT: 1541 fsidp = NULL; 1542 /* FALLSTHROUGH */ 1543 case LFCNSEGWAIT_COMPAT_50: 1544 case LFCNSEGWAIT_COMPAT: 1545 { 1546 struct timeval50 *tvp50 1547 = (struct timeval50 *)ap->a_data; 1548 timeval50_to_timeval(tvp50, &tv); 1549 tvp = &tv; 1550 } 1551 goto segwait_common; 1552 case LFCNSEGWAITALL: 1553 fsidp = NULL; 1554 /* FALLSTHROUGH */ 1555 case LFCNSEGWAIT: 1556 tvp = (struct timeval *)ap->a_data; 1557 segwait_common: 1558 mutex_enter(&lfs_lock); 1559 ++fs->lfs_sleepers; 1560 mutex_exit(&lfs_lock); 1561 1562 error = lfs_segwait(fsidp, tvp); 1563 1564 mutex_enter(&lfs_lock); 1565 if (--fs->lfs_sleepers == 0) 1566 wakeup(&fs->lfs_sleepers); 1567 mutex_exit(&lfs_lock); 1568 return error; 1569 1570 case LFCNBMAPV: 1571 case LFCNMARKV: 1572 blkvp = *(struct lfs_fcntl_markv *)ap->a_data; 1573 1574 blkcnt = blkvp.blkcnt; 1575 if ((u_int) blkcnt > LFS_MARKV_MAXBLKCNT) 1576 return (EINVAL); 1577 blkiov = lfs_malloc(fs, blkcnt * sizeof(BLOCK_INFO), LFS_NB_BLKIOV); 1578 if ((error = copyin(blkvp.blkiov, blkiov, 1579 blkcnt * sizeof(BLOCK_INFO))) != 0) { 1580 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1581 return error; 1582 } 1583 1584 mutex_enter(&lfs_lock); 1585 ++fs->lfs_sleepers; 1586 mutex_exit(&lfs_lock); 1587 if (ap->a_command == LFCNBMAPV) 1588 error = lfs_bmapv(l->l_proc, fsidp, blkiov, blkcnt); 1589 else /* LFCNMARKV */ 1590 error = lfs_markv(l->l_proc, fsidp, blkiov, blkcnt); 1591 if (error == 0) 1592 error = copyout(blkiov, blkvp.blkiov, 1593 blkcnt * sizeof(BLOCK_INFO)); 1594 mutex_enter(&lfs_lock); 1595 if (--fs->lfs_sleepers == 0) 1596 wakeup(&fs->lfs_sleepers); 1597 mutex_exit(&lfs_lock); 1598 lfs_free(fs, blkiov, LFS_NB_BLKIOV); 1599 return error; 1600 1601 case LFCNRECLAIM: 1602 /* 1603 * Flush dirops and write Ifile, allowing empty segments 1604 * to be immediately reclaimed. 1605 */ 1606 lfs_writer_enter(fs, "pndirop"); 1607 off = fs->lfs_offset; 1608 lfs_seglock(fs, SEGM_FORCE_CKP | SEGM_CKP); 1609 lfs_flush_dirops(fs); 1610 LFS_CLEANERINFO(cip, fs, bp); 1611 oclean = cip->clean; 1612 LFS_SYNC_CLEANERINFO(cip, fs, bp, 1); 1613 lfs_segwrite(ap->a_vp->v_mount, SEGM_FORCE_CKP); 1614 fs->lfs_sp->seg_flags |= SEGM_PROT; 1615 lfs_segunlock(fs); 1616 lfs_writer_leave(fs); 1617 1618 #ifdef DEBUG 1619 LFS_CLEANERINFO(cip, fs, bp); 1620 DLOG((DLOG_CLEAN, "lfs_fcntl: reclaim wrote %" PRId64 1621 " blocks, cleaned %" PRId32 " segments (activesb %d)\n", 1622 fs->lfs_offset - off, cip->clean - oclean, 1623 fs->lfs_activesb)); 1624 LFS_SYNC_CLEANERINFO(cip, fs, bp, 0); 1625 #endif 1626 1627 return 0; 1628 1629 case LFCNIFILEFH_COMPAT: 1630 /* Return the filehandle of the Ifile */ 1631 if ((error = kauth_authorize_system(l->l_cred, 1632 KAUTH_SYSTEM_FILEHANDLE, 0, NULL, NULL, NULL)) != 0) 1633 return (error); 1634 fhp = (struct fhandle *)ap->a_data; 1635 fhp->fh_fsid = *fsidp; 1636 fh_size = 16; /* former VFS_MAXFIDSIZ */ 1637 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1638 1639 case LFCNIFILEFH_COMPAT2: 1640 case LFCNIFILEFH: 1641 /* Return the filehandle of the Ifile */ 1642 fhp = (struct fhandle *)ap->a_data; 1643 fhp->fh_fsid = *fsidp; 1644 fh_size = sizeof(struct lfs_fhandle) - 1645 offsetof(fhandle_t, fh_fid); 1646 return lfs_vptofh(fs->lfs_ivnode, &(fhp->fh_fid), &fh_size); 1647 1648 case LFCNREWIND: 1649 /* Move lfs_offset to the lowest-numbered segment */ 1650 return lfs_rewind(fs, *(int *)ap->a_data); 1651 1652 case LFCNINVAL: 1653 /* Mark a segment SEGUSE_INVAL */ 1654 LFS_SEGENTRY(sup, fs, *(int *)ap->a_data, bp); 1655 if (sup->su_nbytes > 0) { 1656 brelse(bp, 0); 1657 lfs_unset_inval_all(fs); 1658 return EBUSY; 1659 } 1660 sup->su_flags |= SEGUSE_INVAL; 1661 VOP_BWRITE(bp->b_vp, bp); 1662 return 0; 1663 1664 case LFCNRESIZE: 1665 /* Resize the filesystem */ 1666 return lfs_resize_fs(fs, *(int *)ap->a_data); 1667 1668 case LFCNWRAPSTOP: 1669 case LFCNWRAPSTOP_COMPAT: 1670 /* 1671 * Hold lfs_newseg at segment 0; if requested, sleep until 1672 * the filesystem wraps around. To support external agents 1673 * (dump, fsck-based regression test) that need to look at 1674 * a snapshot of the filesystem, without necessarily 1675 * requiring that all fs activity stops. 1676 */ 1677 if (fs->lfs_stoplwp == curlwp) 1678 return EALREADY; 1679 1680 mutex_enter(&lfs_lock); 1681 while (fs->lfs_stoplwp != NULL) 1682 cv_wait(&fs->lfs_stopcv, &lfs_lock); 1683 fs->lfs_stoplwp = curlwp; 1684 if (fs->lfs_nowrap == 0) 1685 log(LOG_NOTICE, "%s: disabled log wrap\n", fs->lfs_fsmnt); 1686 ++fs->lfs_nowrap; 1687 if (*(int *)ap->a_data == 1 1688 || ap->a_command == LFCNWRAPSTOP_COMPAT) { 1689 log(LOG_NOTICE, "LFCNSTOPWRAP waiting for log wrap\n"); 1690 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1691 "segwrap", 0, &lfs_lock); 1692 log(LOG_NOTICE, "LFCNSTOPWRAP done waiting\n"); 1693 if (error) { 1694 lfs_wrapgo(fs, VTOI(ap->a_vp), 0); 1695 } 1696 } 1697 mutex_exit(&lfs_lock); 1698 return 0; 1699 1700 case LFCNWRAPGO: 1701 case LFCNWRAPGO_COMPAT: 1702 /* 1703 * Having done its work, the agent wakes up the writer. 1704 * If the argument is 1, it sleeps until a new segment 1705 * is selected. 1706 */ 1707 mutex_enter(&lfs_lock); 1708 error = lfs_wrapgo(fs, VTOI(ap->a_vp), 1709 ap->a_command == LFCNWRAPGO_COMPAT ? 1 : 1710 *((int *)ap->a_data)); 1711 mutex_exit(&lfs_lock); 1712 return error; 1713 1714 case LFCNWRAPPASS: 1715 if ((VTOI(ap->a_vp)->i_lfs_iflags & LFSI_WRAPWAIT)) 1716 return EALREADY; 1717 mutex_enter(&lfs_lock); 1718 if (fs->lfs_stoplwp != curlwp) { 1719 mutex_exit(&lfs_lock); 1720 return EALREADY; 1721 } 1722 if (fs->lfs_nowrap == 0) { 1723 mutex_exit(&lfs_lock); 1724 return EBUSY; 1725 } 1726 fs->lfs_wrappass = 1; 1727 wakeup(&fs->lfs_wrappass); 1728 /* Wait for the log to wrap, if asked */ 1729 if (*(int *)ap->a_data) { 1730 mutex_enter(ap->a_vp->v_interlock); 1731 if (lfs_vref(ap->a_vp) != 0) 1732 panic("LFCNWRAPPASS: lfs_vref failed"); 1733 VTOI(ap->a_vp)->i_lfs_iflags |= LFSI_WRAPWAIT; 1734 log(LOG_NOTICE, "LFCNPASS waiting for log wrap\n"); 1735 error = mtsleep(&fs->lfs_nowrap, PCATCH | PUSER, 1736 "segwrap", 0, &lfs_lock); 1737 log(LOG_NOTICE, "LFCNPASS done waiting\n"); 1738 VTOI(ap->a_vp)->i_lfs_iflags &= ~LFSI_WRAPWAIT; 1739 lfs_vunref(ap->a_vp); 1740 } 1741 mutex_exit(&lfs_lock); 1742 return error; 1743 1744 case LFCNWRAPSTATUS: 1745 mutex_enter(&lfs_lock); 1746 *(int *)ap->a_data = fs->lfs_wrapstatus; 1747 mutex_exit(&lfs_lock); 1748 return 0; 1749 1750 default: 1751 return ufs_fcntl(v); 1752 } 1753 return 0; 1754 } 1755 1756 int 1757 lfs_getpages(void *v) 1758 { 1759 struct vop_getpages_args /* { 1760 struct vnode *a_vp; 1761 voff_t a_offset; 1762 struct vm_page **a_m; 1763 int *a_count; 1764 int a_centeridx; 1765 vm_prot_t a_access_type; 1766 int a_advice; 1767 int a_flags; 1768 } */ *ap = v; 1769 1770 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM && 1771 (ap->a_access_type & VM_PROT_WRITE) != 0) { 1772 return EPERM; 1773 } 1774 if ((ap->a_access_type & VM_PROT_WRITE) != 0) { 1775 mutex_enter(&lfs_lock); 1776 LFS_SET_UINO(VTOI(ap->a_vp), IN_MODIFIED); 1777 mutex_exit(&lfs_lock); 1778 } 1779 1780 /* 1781 * we're relying on the fact that genfs_getpages() always read in 1782 * entire filesystem blocks. 1783 */ 1784 return genfs_getpages(v); 1785 } 1786 1787 /* 1788 * Wait for a page to become unbusy, possibly printing diagnostic messages 1789 * as well. 1790 * 1791 * Called with vp->v_interlock held; return with it held. 1792 */ 1793 static void 1794 wait_for_page(struct vnode *vp, struct vm_page *pg, const char *label) 1795 { 1796 KASSERT(mutex_owned(vp->v_interlock)); 1797 if ((pg->flags & PG_BUSY) == 0) 1798 return; /* Nothing to wait for! */ 1799 1800 #if defined(DEBUG) && defined(UVM_PAGE_TRKOWN) 1801 static struct vm_page *lastpg; 1802 1803 if (label != NULL && pg != lastpg) { 1804 if (pg->owner_tag) { 1805 printf("lfs_putpages[%d.%d]: %s: page %p owner %d.%d [%s]\n", 1806 curproc->p_pid, curlwp->l_lid, label, 1807 pg, pg->owner, pg->lowner, pg->owner_tag); 1808 } else { 1809 printf("lfs_putpages[%d.%d]: %s: page %p unowned?!\n", 1810 curproc->p_pid, curlwp->l_lid, label, pg); 1811 } 1812 } 1813 lastpg = pg; 1814 #endif 1815 1816 pg->flags |= PG_WANTED; 1817 UVM_UNLOCK_AND_WAIT(pg, vp->v_interlock, 0, "lfsput", 0); 1818 mutex_enter(vp->v_interlock); 1819 } 1820 1821 /* 1822 * This routine is called by lfs_putpages() when it can't complete the 1823 * write because a page is busy. This means that either (1) someone, 1824 * possibly the pagedaemon, is looking at this page, and will give it up 1825 * presently; or (2) we ourselves are holding the page busy in the 1826 * process of being written (either gathered or actually on its way to 1827 * disk). We don't need to give up the segment lock, but we might need 1828 * to call lfs_writeseg() to expedite the page's journey to disk. 1829 * 1830 * Called with vp->v_interlock held; return with it held. 1831 */ 1832 /* #define BUSYWAIT */ 1833 static void 1834 write_and_wait(struct lfs *fs, struct vnode *vp, struct vm_page *pg, 1835 int seglocked, const char *label) 1836 { 1837 KASSERT(mutex_owned(vp->v_interlock)); 1838 #ifndef BUSYWAIT 1839 struct inode *ip = VTOI(vp); 1840 struct segment *sp = fs->lfs_sp; 1841 int count = 0; 1842 1843 if (pg == NULL) 1844 return; 1845 1846 while (pg->flags & PG_BUSY && 1847 pg->uobject == &vp->v_uobj) { 1848 mutex_exit(vp->v_interlock); 1849 if (sp->cbpp - sp->bpp > 1) { 1850 /* Write gathered pages */ 1851 lfs_updatemeta(sp); 1852 lfs_release_finfo(fs); 1853 (void) lfs_writeseg(fs, sp); 1854 1855 /* 1856 * Reinitialize FIP 1857 */ 1858 KASSERT(sp->vp == vp); 1859 lfs_acquire_finfo(fs, ip->i_number, 1860 ip->i_gen); 1861 } 1862 ++count; 1863 mutex_enter(vp->v_interlock); 1864 wait_for_page(vp, pg, label); 1865 } 1866 if (label != NULL && count > 1) { 1867 DLOG((DLOG_PAGE, "lfs_putpages[%d]: %s: %sn = %d\n", 1868 curproc->p_pid, label, (count > 0 ? "looping, " : ""), 1869 count)); 1870 } 1871 #else 1872 preempt(1); 1873 #endif 1874 KASSERT(mutex_owned(vp->v_interlock)); 1875 } 1876 1877 /* 1878 * Make sure that for all pages in every block in the given range, 1879 * either all are dirty or all are clean. If any of the pages 1880 * we've seen so far are dirty, put the vnode on the paging chain, 1881 * and mark it IN_PAGING. 1882 * 1883 * If checkfirst != 0, don't check all the pages but return at the 1884 * first dirty page. 1885 */ 1886 static int 1887 check_dirty(struct lfs *fs, struct vnode *vp, 1888 off_t startoffset, off_t endoffset, off_t blkeof, 1889 int flags, int checkfirst, struct vm_page **pgp) 1890 { 1891 int by_list; 1892 struct vm_page *curpg = NULL; /* XXX: gcc */ 1893 struct vm_page *pgs[MAXBSIZE / PAGE_SIZE], *pg; 1894 off_t soff = 0; /* XXX: gcc */ 1895 voff_t off; 1896 int i; 1897 int nonexistent; 1898 int any_dirty; /* number of dirty pages */ 1899 int dirty; /* number of dirty pages in a block */ 1900 int tdirty; 1901 int pages_per_block = fs->lfs_bsize >> PAGE_SHIFT; 1902 int pagedaemon = (curlwp == uvm.pagedaemon_lwp); 1903 1904 KASSERT(mutex_owned(vp->v_interlock)); 1905 ASSERT_MAYBE_SEGLOCK(fs); 1906 top: 1907 by_list = (vp->v_uobj.uo_npages <= 1908 ((endoffset - startoffset) >> PAGE_SHIFT) * 1909 UVM_PAGE_TREE_PENALTY); 1910 any_dirty = 0; 1911 1912 if (by_list) { 1913 curpg = TAILQ_FIRST(&vp->v_uobj.memq); 1914 } else { 1915 soff = startoffset; 1916 } 1917 while (by_list || soff < MIN(blkeof, endoffset)) { 1918 if (by_list) { 1919 /* 1920 * Find the first page in a block. Skip 1921 * blocks outside our area of interest or beyond 1922 * the end of file. 1923 */ 1924 KASSERT(curpg == NULL 1925 || (curpg->flags & PG_MARKER) == 0); 1926 if (pages_per_block > 1) { 1927 while (curpg && 1928 ((curpg->offset & fs->lfs_bmask) || 1929 curpg->offset >= vp->v_size || 1930 curpg->offset >= endoffset)) { 1931 curpg = TAILQ_NEXT(curpg, listq.queue); 1932 KASSERT(curpg == NULL || 1933 (curpg->flags & PG_MARKER) == 0); 1934 } 1935 } 1936 if (curpg == NULL) 1937 break; 1938 soff = curpg->offset; 1939 } 1940 1941 /* 1942 * Mark all pages in extended range busy; find out if any 1943 * of them are dirty. 1944 */ 1945 nonexistent = dirty = 0; 1946 for (i = 0; i == 0 || i < pages_per_block; i++) { 1947 KASSERT(mutex_owned(vp->v_interlock)); 1948 if (by_list && pages_per_block <= 1) { 1949 pgs[i] = pg = curpg; 1950 } else { 1951 off = soff + (i << PAGE_SHIFT); 1952 pgs[i] = pg = uvm_pagelookup(&vp->v_uobj, off); 1953 if (pg == NULL) { 1954 ++nonexistent; 1955 continue; 1956 } 1957 } 1958 KASSERT(pg != NULL); 1959 1960 /* 1961 * If we're holding the segment lock, we can deadlock 1962 * against a process that has our page and is waiting 1963 * for the cleaner, while the cleaner waits for the 1964 * segment lock. Just bail in that case. 1965 */ 1966 if ((pg->flags & PG_BUSY) && 1967 (pagedaemon || LFS_SEGLOCK_HELD(fs))) { 1968 if (i > 0) 1969 uvm_page_unbusy(pgs, i); 1970 DLOG((DLOG_PAGE, "lfs_putpages: avoiding 3-way or pagedaemon deadlock\n")); 1971 if (pgp) 1972 *pgp = pg; 1973 KASSERT(mutex_owned(vp->v_interlock)); 1974 return -1; 1975 } 1976 1977 while (pg->flags & PG_BUSY) { 1978 wait_for_page(vp, pg, NULL); 1979 KASSERT(mutex_owned(vp->v_interlock)); 1980 if (i > 0) 1981 uvm_page_unbusy(pgs, i); 1982 KASSERT(mutex_owned(vp->v_interlock)); 1983 goto top; 1984 } 1985 pg->flags |= PG_BUSY; 1986 UVM_PAGE_OWN(pg, "lfs_putpages"); 1987 1988 pmap_page_protect(pg, VM_PROT_NONE); 1989 tdirty = (pmap_clear_modify(pg) || 1990 (pg->flags & PG_CLEAN) == 0); 1991 dirty += tdirty; 1992 } 1993 if (pages_per_block > 0 && nonexistent >= pages_per_block) { 1994 if (by_list) { 1995 curpg = TAILQ_NEXT(curpg, listq.queue); 1996 } else { 1997 soff += fs->lfs_bsize; 1998 } 1999 continue; 2000 } 2001 2002 any_dirty += dirty; 2003 KASSERT(nonexistent == 0); 2004 KASSERT(mutex_owned(vp->v_interlock)); 2005 2006 /* 2007 * If any are dirty make all dirty; unbusy them, 2008 * but if we were asked to clean, wire them so that 2009 * the pagedaemon doesn't bother us about them while 2010 * they're on their way to disk. 2011 */ 2012 for (i = 0; i == 0 || i < pages_per_block; i++) { 2013 KASSERT(mutex_owned(vp->v_interlock)); 2014 pg = pgs[i]; 2015 KASSERT(!((pg->flags & PG_CLEAN) && (pg->flags & PG_DELWRI))); 2016 KASSERT(pg->flags & PG_BUSY); 2017 if (dirty) { 2018 pg->flags &= ~PG_CLEAN; 2019 if (flags & PGO_FREE) { 2020 /* 2021 * Wire the page so that 2022 * pdaemon doesn't see it again. 2023 */ 2024 mutex_enter(&uvm_pageqlock); 2025 uvm_pagewire(pg); 2026 mutex_exit(&uvm_pageqlock); 2027 2028 /* Suspended write flag */ 2029 pg->flags |= PG_DELWRI; 2030 } 2031 } 2032 if (pg->flags & PG_WANTED) 2033 wakeup(pg); 2034 pg->flags &= ~(PG_WANTED|PG_BUSY); 2035 UVM_PAGE_OWN(pg, NULL); 2036 } 2037 2038 if (checkfirst && any_dirty) 2039 break; 2040 2041 if (by_list) { 2042 curpg = TAILQ_NEXT(curpg, listq.queue); 2043 } else { 2044 soff += MAX(PAGE_SIZE, fs->lfs_bsize); 2045 } 2046 } 2047 2048 KASSERT(mutex_owned(vp->v_interlock)); 2049 return any_dirty; 2050 } 2051 2052 /* 2053 * lfs_putpages functions like genfs_putpages except that 2054 * 2055 * (1) It needs to bounds-check the incoming requests to ensure that 2056 * they are block-aligned; if they are not, expand the range and 2057 * do the right thing in case, e.g., the requested range is clean 2058 * but the expanded range is dirty. 2059 * 2060 * (2) It needs to explicitly send blocks to be written when it is done. 2061 * If VOP_PUTPAGES is called without the seglock held, we simply take 2062 * the seglock and let lfs_segunlock wait for us. 2063 * XXX There might be a bad situation if we have to flush a vnode while 2064 * XXX lfs_markv is in operation. As of this writing we panic in this 2065 * XXX case. 2066 * 2067 * Assumptions: 2068 * 2069 * (1) The caller does not hold any pages in this vnode busy. If it does, 2070 * there is a danger that when we expand the page range and busy the 2071 * pages we will deadlock. 2072 * 2073 * (2) We are called with vp->v_interlock held; we must return with it 2074 * released. 2075 * 2076 * (3) We don't absolutely have to free pages right away, provided that 2077 * the request does not have PGO_SYNCIO. When the pagedaemon gives 2078 * us a request with PGO_FREE, we take the pages out of the paging 2079 * queue and wake up the writer, which will handle freeing them for us. 2080 * 2081 * We ensure that for any filesystem block, all pages for that 2082 * block are either resident or not, even if those pages are higher 2083 * than EOF; that means that we will be getting requests to free 2084 * "unused" pages above EOF all the time, and should ignore them. 2085 * 2086 * (4) If we are called with PGO_LOCKED, the finfo array we are to write 2087 * into has been set up for us by lfs_writefile. If not, we will 2088 * have to handle allocating and/or freeing an finfo entry. 2089 * 2090 * XXX note that we're (ab)using PGO_LOCKED as "seglock held". 2091 */ 2092 2093 /* How many times to loop before we should start to worry */ 2094 #define TOOMANY 4 2095 2096 int 2097 lfs_putpages(void *v) 2098 { 2099 int error; 2100 struct vop_putpages_args /* { 2101 struct vnode *a_vp; 2102 voff_t a_offlo; 2103 voff_t a_offhi; 2104 int a_flags; 2105 } */ *ap = v; 2106 struct vnode *vp; 2107 struct inode *ip; 2108 struct lfs *fs; 2109 struct segment *sp; 2110 off_t origoffset, startoffset, endoffset, origendoffset, blkeof; 2111 off_t off, max_endoffset; 2112 bool seglocked, sync, pagedaemon, reclaim; 2113 struct vm_page *pg, *busypg; 2114 UVMHIST_FUNC("lfs_putpages"); UVMHIST_CALLED(ubchist); 2115 int oreclaim = 0; 2116 int donewriting = 0; 2117 #ifdef DEBUG 2118 int debug_n_again, debug_n_dirtyclean; 2119 #endif 2120 2121 vp = ap->a_vp; 2122 ip = VTOI(vp); 2123 fs = ip->i_lfs; 2124 sync = (ap->a_flags & PGO_SYNCIO) != 0; 2125 reclaim = (ap->a_flags & PGO_RECLAIM) != 0; 2126 pagedaemon = (curlwp == uvm.pagedaemon_lwp); 2127 2128 KASSERT(mutex_owned(vp->v_interlock)); 2129 2130 /* Putpages does nothing for metadata. */ 2131 if (vp == fs->lfs_ivnode || vp->v_type != VREG) { 2132 mutex_exit(vp->v_interlock); 2133 return 0; 2134 } 2135 2136 /* 2137 * If there are no pages, don't do anything. 2138 */ 2139 if (vp->v_uobj.uo_npages == 0) { 2140 if (TAILQ_EMPTY(&vp->v_uobj.memq) && 2141 (vp->v_iflag & VI_ONWORKLST) && 2142 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) { 2143 vp->v_iflag &= ~VI_WRMAPDIRTY; 2144 vn_syncer_remove_from_worklist(vp); 2145 } 2146 mutex_exit(vp->v_interlock); 2147 2148 /* Remove us from paging queue, if we were on it */ 2149 mutex_enter(&lfs_lock); 2150 if (ip->i_flags & IN_PAGING) { 2151 ip->i_flags &= ~IN_PAGING; 2152 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2153 } 2154 mutex_exit(&lfs_lock); 2155 2156 KASSERT(!mutex_owned(vp->v_interlock)); 2157 return 0; 2158 } 2159 2160 blkeof = blkroundup(fs, ip->i_size); 2161 2162 /* 2163 * Ignore requests to free pages past EOF but in the same block 2164 * as EOF, unless the vnode is being reclaimed or the request 2165 * is synchronous. (If the request is sync, it comes from 2166 * lfs_truncate.) 2167 * 2168 * To avoid being flooded with this request, make these pages 2169 * look "active". 2170 */ 2171 if (!sync && !reclaim && 2172 ap->a_offlo >= ip->i_size && ap->a_offlo < blkeof) { 2173 origoffset = ap->a_offlo; 2174 for (off = origoffset; off < blkeof; off += fs->lfs_bsize) { 2175 pg = uvm_pagelookup(&vp->v_uobj, off); 2176 KASSERT(pg != NULL); 2177 while (pg->flags & PG_BUSY) { 2178 pg->flags |= PG_WANTED; 2179 UVM_UNLOCK_AND_WAIT(pg, vp->v_interlock, 0, 2180 "lfsput2", 0); 2181 mutex_enter(vp->v_interlock); 2182 } 2183 mutex_enter(&uvm_pageqlock); 2184 uvm_pageactivate(pg); 2185 mutex_exit(&uvm_pageqlock); 2186 } 2187 ap->a_offlo = blkeof; 2188 if (ap->a_offhi > 0 && ap->a_offhi <= ap->a_offlo) { 2189 mutex_exit(vp->v_interlock); 2190 return 0; 2191 } 2192 } 2193 2194 /* 2195 * Extend page range to start and end at block boundaries. 2196 * (For the purposes of VOP_PUTPAGES, fragments don't exist.) 2197 */ 2198 origoffset = ap->a_offlo; 2199 origendoffset = ap->a_offhi; 2200 startoffset = origoffset & ~(fs->lfs_bmask); 2201 max_endoffset = (trunc_page(LLONG_MAX) >> fs->lfs_bshift) 2202 << fs->lfs_bshift; 2203 2204 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2205 endoffset = max_endoffset; 2206 origendoffset = endoffset; 2207 } else { 2208 origendoffset = round_page(ap->a_offhi); 2209 endoffset = round_page(blkroundup(fs, origendoffset)); 2210 } 2211 2212 KASSERT(startoffset > 0 || endoffset >= startoffset); 2213 if (startoffset == endoffset) { 2214 /* Nothing to do, why were we called? */ 2215 mutex_exit(vp->v_interlock); 2216 DLOG((DLOG_PAGE, "lfs_putpages: startoffset = endoffset = %" 2217 PRId64 "\n", startoffset)); 2218 return 0; 2219 } 2220 2221 ap->a_offlo = startoffset; 2222 ap->a_offhi = endoffset; 2223 2224 /* 2225 * If not cleaning, just send the pages through genfs_putpages 2226 * to be returned to the pool. 2227 */ 2228 if (!(ap->a_flags & PGO_CLEANIT)) { 2229 DLOG((DLOG_PAGE, "lfs_putpages: no cleanit vn %p ino %d (flags %x)\n", 2230 vp, (int)ip->i_number, ap->a_flags)); 2231 int r = genfs_putpages(v); 2232 KASSERT(!mutex_owned(vp->v_interlock)); 2233 return r; 2234 } 2235 2236 /* Set PGO_BUSYFAIL to avoid deadlocks */ 2237 ap->a_flags |= PGO_BUSYFAIL; 2238 2239 /* 2240 * Likewise, if we are asked to clean but the pages are not 2241 * dirty, we can just free them using genfs_putpages. 2242 */ 2243 #ifdef DEBUG 2244 debug_n_dirtyclean = 0; 2245 #endif 2246 do { 2247 int r; 2248 KASSERT(mutex_owned(vp->v_interlock)); 2249 2250 /* Count the number of dirty pages */ 2251 r = check_dirty(fs, vp, startoffset, endoffset, blkeof, 2252 ap->a_flags, 1, NULL); 2253 if (r < 0) { 2254 /* Pages are busy with another process */ 2255 mutex_exit(vp->v_interlock); 2256 return EDEADLK; 2257 } 2258 if (r > 0) /* Some pages are dirty */ 2259 break; 2260 2261 /* 2262 * Sometimes pages are dirtied between the time that 2263 * we check and the time we try to clean them. 2264 * Instruct lfs_gop_write to return EDEADLK in this case 2265 * so we can write them properly. 2266 */ 2267 ip->i_lfs_iflags |= LFSI_NO_GOP_WRITE; 2268 r = genfs_do_putpages(vp, startoffset, endoffset, 2269 ap->a_flags & ~PGO_SYNCIO, &busypg); 2270 ip->i_lfs_iflags &= ~LFSI_NO_GOP_WRITE; 2271 if (r != EDEADLK) { 2272 KASSERT(!mutex_owned(vp->v_interlock)); 2273 return r; 2274 } 2275 2276 /* One of the pages was busy. Start over. */ 2277 mutex_enter(vp->v_interlock); 2278 wait_for_page(vp, busypg, "dirtyclean"); 2279 #ifdef DEBUG 2280 ++debug_n_dirtyclean; 2281 #endif 2282 } while(1); 2283 2284 #ifdef DEBUG 2285 if (debug_n_dirtyclean > TOOMANY) 2286 DLOG((DLOG_PAGE, "lfs_putpages: dirtyclean: looping, n = %d\n", 2287 debug_n_dirtyclean)); 2288 #endif 2289 2290 /* 2291 * Dirty and asked to clean. 2292 * 2293 * Pagedaemon can't actually write LFS pages; wake up 2294 * the writer to take care of that. The writer will 2295 * notice the pager inode queue and act on that. 2296 * 2297 * XXX We must drop the vp->interlock before taking the lfs_lock or we 2298 * get a nasty deadlock with lfs_flush_pchain(). 2299 */ 2300 if (pagedaemon) { 2301 mutex_exit(vp->v_interlock); 2302 mutex_enter(&lfs_lock); 2303 if (!(ip->i_flags & IN_PAGING)) { 2304 ip->i_flags |= IN_PAGING; 2305 TAILQ_INSERT_TAIL(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2306 } 2307 wakeup(&lfs_writer_daemon); 2308 mutex_exit(&lfs_lock); 2309 preempt(); 2310 KASSERT(!mutex_owned(vp->v_interlock)); 2311 return EWOULDBLOCK; 2312 } 2313 2314 /* 2315 * If this is a file created in a recent dirop, we can't flush its 2316 * inode until the dirop is complete. Drain dirops, then flush the 2317 * filesystem (taking care of any other pending dirops while we're 2318 * at it). 2319 */ 2320 if ((ap->a_flags & (PGO_CLEANIT|PGO_LOCKED)) == PGO_CLEANIT && 2321 (vp->v_uflag & VU_DIROP)) { 2322 DLOG((DLOG_PAGE, "lfs_putpages: flushing VU_DIROP\n")); 2323 2324 lfs_writer_enter(fs, "ppdirop"); 2325 2326 /* Note if we hold the vnode locked */ 2327 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE) 2328 { 2329 DLOG((DLOG_PAGE, "lfs_putpages: dirop inode already locked\n")); 2330 } else { 2331 DLOG((DLOG_PAGE, "lfs_putpages: dirop inode not locked\n")); 2332 } 2333 mutex_exit(vp->v_interlock); 2334 2335 mutex_enter(&lfs_lock); 2336 lfs_flush_fs(fs, sync ? SEGM_SYNC : 0); 2337 mutex_exit(&lfs_lock); 2338 2339 mutex_enter(vp->v_interlock); 2340 lfs_writer_leave(fs); 2341 2342 /* The flush will have cleaned out this vnode as well, 2343 no need to do more to it. */ 2344 } 2345 2346 /* 2347 * This is it. We are going to write some pages. From here on 2348 * down it's all just mechanics. 2349 * 2350 * Don't let genfs_putpages wait; lfs_segunlock will wait for us. 2351 */ 2352 ap->a_flags &= ~PGO_SYNCIO; 2353 2354 /* 2355 * If we've already got the seglock, flush the node and return. 2356 * The FIP has already been set up for us by lfs_writefile, 2357 * and FIP cleanup and lfs_updatemeta will also be done there, 2358 * unless genfs_putpages returns EDEADLK; then we must flush 2359 * what we have, and correct FIP and segment header accounting. 2360 */ 2361 get_seglock: 2362 /* 2363 * If we are not called with the segment locked, lock it. 2364 * Account for a new FIP in the segment header, and set sp->vp. 2365 * (This should duplicate the setup at the top of lfs_writefile().) 2366 */ 2367 seglocked = (ap->a_flags & PGO_LOCKED) != 0; 2368 if (!seglocked) { 2369 mutex_exit(vp->v_interlock); 2370 error = lfs_seglock(fs, SEGM_PROT | (sync ? SEGM_SYNC : 0)); 2371 if (error != 0) { 2372 KASSERT(!mutex_owned(vp->v_interlock)); 2373 return error; 2374 } 2375 mutex_enter(vp->v_interlock); 2376 lfs_acquire_finfo(fs, ip->i_number, ip->i_gen); 2377 } 2378 sp = fs->lfs_sp; 2379 KASSERT(sp->vp == NULL); 2380 sp->vp = vp; 2381 2382 /* Note segments written by reclaim; only for debugging */ 2383 if ((vp->v_iflag & VI_XLOCK) != 0) { 2384 sp->seg_flags |= SEGM_RECLAIM; 2385 fs->lfs_reclino = ip->i_number; 2386 } 2387 2388 /* 2389 * Ensure that the partial segment is marked SS_DIROP if this 2390 * vnode is a DIROP. 2391 */ 2392 if (!seglocked && vp->v_uflag & VU_DIROP) 2393 ((SEGSUM *)(sp->segsum))->ss_flags |= (SS_DIROP|SS_CONT); 2394 2395 /* 2396 * Loop over genfs_putpages until all pages are gathered. 2397 * genfs_putpages() drops the interlock, so reacquire it if necessary. 2398 * Whenever we lose the interlock we have to rerun check_dirty, as 2399 * well, since more pages might have been dirtied in our absence. 2400 */ 2401 #ifdef DEBUG 2402 debug_n_again = 0; 2403 #endif 2404 do { 2405 busypg = NULL; 2406 KASSERT(mutex_owned(vp->v_interlock)); 2407 if (check_dirty(fs, vp, startoffset, endoffset, blkeof, 2408 ap->a_flags, 0, &busypg) < 0) { 2409 mutex_exit(vp->v_interlock); 2410 /* XXX why? --ks */ 2411 mutex_enter(vp->v_interlock); 2412 write_and_wait(fs, vp, busypg, seglocked, NULL); 2413 if (!seglocked) { 2414 mutex_exit(vp->v_interlock); 2415 lfs_release_finfo(fs); 2416 lfs_segunlock(fs); 2417 mutex_enter(vp->v_interlock); 2418 } 2419 sp->vp = NULL; 2420 goto get_seglock; 2421 } 2422 2423 busypg = NULL; 2424 KASSERT(!mutex_owned(&uvm_pageqlock)); 2425 oreclaim = (ap->a_flags & PGO_RECLAIM); 2426 ap->a_flags &= ~PGO_RECLAIM; 2427 error = genfs_do_putpages(vp, startoffset, endoffset, 2428 ap->a_flags, &busypg); 2429 ap->a_flags |= oreclaim; 2430 2431 if (error == EDEADLK || error == EAGAIN) { 2432 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2433 " %d ino %d off %x (seg %d)\n", error, 2434 ip->i_number, fs->lfs_offset, 2435 dtosn(fs, fs->lfs_offset))); 2436 2437 if (oreclaim) { 2438 mutex_enter(vp->v_interlock); 2439 write_and_wait(fs, vp, busypg, seglocked, "again"); 2440 mutex_exit(vp->v_interlock); 2441 } else { 2442 if ((sp->seg_flags & SEGM_SINGLE) && 2443 fs->lfs_curseg != fs->lfs_startseg) 2444 donewriting = 1; 2445 } 2446 } else if (error) { 2447 DLOG((DLOG_PAGE, "lfs_putpages: genfs_putpages returned" 2448 " %d ino %d off %x (seg %d)\n", error, 2449 (int)ip->i_number, fs->lfs_offset, 2450 dtosn(fs, fs->lfs_offset))); 2451 } 2452 /* genfs_do_putpages loses the interlock */ 2453 #ifdef DEBUG 2454 ++debug_n_again; 2455 #endif 2456 if (oreclaim && error == EAGAIN) { 2457 DLOG((DLOG_PAGE, "vp %p ino %d vi_flags %x a_flags %x avoiding vclean panic\n", 2458 vp, (int)ip->i_number, vp->v_iflag, ap->a_flags)); 2459 mutex_enter(vp->v_interlock); 2460 } 2461 if (error == EDEADLK) 2462 mutex_enter(vp->v_interlock); 2463 } while (error == EDEADLK || (oreclaim && error == EAGAIN)); 2464 #ifdef DEBUG 2465 if (debug_n_again > TOOMANY) 2466 DLOG((DLOG_PAGE, "lfs_putpages: again: looping, n = %d\n", debug_n_again)); 2467 #endif 2468 2469 KASSERT(sp != NULL && sp->vp == vp); 2470 if (!seglocked && !donewriting) { 2471 sp->vp = NULL; 2472 2473 /* Write indirect blocks as well */ 2474 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_indir); 2475 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_dindir); 2476 lfs_gather(fs, fs->lfs_sp, vp, lfs_match_tindir); 2477 2478 KASSERT(sp->vp == NULL); 2479 sp->vp = vp; 2480 } 2481 2482 /* 2483 * Blocks are now gathered into a segment waiting to be written. 2484 * All that's left to do is update metadata, and write them. 2485 */ 2486 lfs_updatemeta(sp); 2487 KASSERT(sp->vp == vp); 2488 sp->vp = NULL; 2489 2490 /* 2491 * If we were called from lfs_writefile, we don't need to clean up 2492 * the FIP or unlock the segment lock. We're done. 2493 */ 2494 if (seglocked) { 2495 KASSERT(!mutex_owned(vp->v_interlock)); 2496 return error; 2497 } 2498 2499 /* Clean up FIP and send it to disk. */ 2500 lfs_release_finfo(fs); 2501 lfs_writeseg(fs, fs->lfs_sp); 2502 2503 /* 2504 * Remove us from paging queue if we wrote all our pages. 2505 */ 2506 if (origendoffset == 0 || ap->a_flags & PGO_ALLPAGES) { 2507 mutex_enter(&lfs_lock); 2508 if (ip->i_flags & IN_PAGING) { 2509 ip->i_flags &= ~IN_PAGING; 2510 TAILQ_REMOVE(&fs->lfs_pchainhd, ip, i_lfs_pchain); 2511 } 2512 mutex_exit(&lfs_lock); 2513 } 2514 2515 /* 2516 * XXX - with the malloc/copy writeseg, the pages are freed by now 2517 * even if we don't wait (e.g. if we hold a nested lock). This 2518 * will not be true if we stop using malloc/copy. 2519 */ 2520 KASSERT(fs->lfs_sp->seg_flags & SEGM_PROT); 2521 lfs_segunlock(fs); 2522 2523 /* 2524 * Wait for v_numoutput to drop to zero. The seglock should 2525 * take care of this, but there is a slight possibility that 2526 * aiodoned might not have got around to our buffers yet. 2527 */ 2528 if (sync) { 2529 mutex_enter(vp->v_interlock); 2530 while (vp->v_numoutput > 0) { 2531 DLOG((DLOG_PAGE, "lfs_putpages: ino %d sleeping on" 2532 " num %d\n", ip->i_number, vp->v_numoutput)); 2533 cv_wait(&vp->v_cv, vp->v_interlock); 2534 } 2535 mutex_exit(vp->v_interlock); 2536 } 2537 KASSERT(!mutex_owned(vp->v_interlock)); 2538 return error; 2539 } 2540 2541 /* 2542 * Return the last logical file offset that should be written for this file 2543 * if we're doing a write that ends at "size". If writing, we need to know 2544 * about sizes on disk, i.e. fragments if there are any; if reading, we need 2545 * to know about entire blocks. 2546 */ 2547 void 2548 lfs_gop_size(struct vnode *vp, off_t size, off_t *eobp, int flags) 2549 { 2550 struct inode *ip = VTOI(vp); 2551 struct lfs *fs = ip->i_lfs; 2552 daddr_t olbn, nlbn; 2553 2554 olbn = lblkno(fs, ip->i_size); 2555 nlbn = lblkno(fs, size); 2556 if (!(flags & GOP_SIZE_MEM) && nlbn < NDADDR && olbn <= nlbn) { 2557 *eobp = fragroundup(fs, size); 2558 } else { 2559 *eobp = blkroundup(fs, size); 2560 } 2561 } 2562 2563 #ifdef DEBUG 2564 void lfs_dump_vop(void *); 2565 2566 void 2567 lfs_dump_vop(void *v) 2568 { 2569 struct vop_putpages_args /* { 2570 struct vnode *a_vp; 2571 voff_t a_offlo; 2572 voff_t a_offhi; 2573 int a_flags; 2574 } */ *ap = v; 2575 2576 #ifdef DDB 2577 vfs_vnode_print(ap->a_vp, 0, printf); 2578 #endif 2579 lfs_dump_dinode(VTOI(ap->a_vp)->i_din.ffs1_din); 2580 } 2581 #endif 2582 2583 int 2584 lfs_mmap(void *v) 2585 { 2586 struct vop_mmap_args /* { 2587 const struct vnodeop_desc *a_desc; 2588 struct vnode *a_vp; 2589 vm_prot_t a_prot; 2590 kauth_cred_t a_cred; 2591 } */ *ap = v; 2592 2593 if (VTOI(ap->a_vp)->i_number == LFS_IFILE_INUM) 2594 return EOPNOTSUPP; 2595 return ufs_mmap(v); 2596 } 2597
Indexes created Tue Sep 30 20:09:53 GMT 2025