1 /* 2 * Copyright (c) 2004,2013 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * Copyright (c) 1989, 1993 35 * The Regents of the University of California. All rights reserved. 36 * (c) UNIX System Laboratories, Inc. 37 * All or some portions of this file are derived from material licensed 38 * to the University of California by American Telephone and Telegraph 39 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 40 * the permission of UNIX System Laboratories, Inc. 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 67 /* 68 * External virtual filesystem routines 69 */ 70 71 #include <sys/param.h> 72 #include <sys/systm.h> 73 #include <sys/kernel.h> 74 #include <sys/malloc.h> 75 #include <sys/mount.h> 76 #include <sys/proc.h> 77 #include <sys/vnode.h> 78 #include <sys/buf.h> 79 #include <sys/eventhandler.h> 80 #include <sys/kthread.h> 81 #include <sys/sysctl.h> 82 83 #include <machine/limits.h> 84 85 #include <sys/buf2.h> 86 #include <sys/thread2.h> 87 #include <sys/sysref2.h> 88 89 #include <vm/vm.h> 90 #include <vm/vm_object.h> 91 92 struct mountscan_info { 93 TAILQ_ENTRY(mountscan_info) msi_entry; 94 int msi_how; 95 struct mount *msi_node; 96 }; 97 98 struct vmntvnodescan_info { 99 TAILQ_ENTRY(vmntvnodescan_info) entry; 100 struct vnode *vp; 101 }; 102 103 struct vnlru_info { 104 int pass; 105 }; 106 107 static int vnlru_nowhere = 0; 108 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RD, 109 &vnlru_nowhere, 0, 110 "Number of times the vnlru process ran without success"); 111 112 113 static struct lwkt_token mntid_token; 114 static struct mount dummymount; 115 116 /* note: mountlist exported to pstat */ 117 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 118 static TAILQ_HEAD(,mountscan_info) mountscan_list; 119 static struct lwkt_token mountlist_token; 120 121 static TAILQ_HEAD(,bio_ops) bio_ops_list = TAILQ_HEAD_INITIALIZER(bio_ops_list); 122 123 /* 124 * Called from vfsinit() 125 */ 126 void 127 vfs_mount_init(void) 128 { 129 lwkt_token_init(&mountlist_token, "mntlist"); 130 lwkt_token_init(&mntid_token, "mntid"); 131 TAILQ_INIT(&mountscan_list); 132 mount_init(&dummymount); 133 dummymount.mnt_flag |= MNT_RDONLY; 134 dummymount.mnt_kern_flag |= MNTK_ALL_MPSAFE; 135 } 136 137 /* 138 * Support function called to remove a vnode from the mountlist and 139 * deal with side effects for scans in progress. 140 * 141 * Target mnt_token is held on call. 142 */ 143 static void 144 vremovevnodemnt(struct vnode *vp) 145 { 146 struct vmntvnodescan_info *info; 147 struct mount *mp = vp->v_mount; 148 149 TAILQ_FOREACH(info, &mp->mnt_vnodescan_list, entry) { 150 if (info->vp == vp) 151 info->vp = TAILQ_NEXT(vp, v_nmntvnodes); 152 } 153 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); 154 } 155 156 /* 157 * Allocate a new vnode and associate it with a tag, mount point, and 158 * operations vector. 159 * 160 * A VX locked and refd vnode is returned. The caller should setup the 161 * remaining fields and vx_put() or, if he wishes to leave a vref, 162 * vx_unlock() the vnode. 163 */ 164 int 165 getnewvnode(enum vtagtype tag, struct mount *mp, 166 struct vnode **vpp, int lktimeout, int lkflags) 167 { 168 struct vnode *vp; 169 170 KKASSERT(mp != NULL); 171 172 vp = allocvnode(lktimeout, lkflags); 173 vp->v_tag = tag; 174 vp->v_data = NULL; 175 176 /* 177 * By default the vnode is assigned the mount point's normal 178 * operations vector. 179 */ 180 vp->v_ops = &mp->mnt_vn_use_ops; 181 182 /* 183 * Placing the vnode on the mount point's queue makes it visible. 184 * VNON prevents it from being messed with, however. 185 */ 186 insmntque(vp, mp); 187 188 /* 189 * A VX locked & refd vnode is returned. 190 */ 191 *vpp = vp; 192 return (0); 193 } 194 195 /* 196 * This function creates vnodes with special operations vectors. The 197 * mount point is optional. 198 * 199 * This routine is being phased out but is still used by vfs_conf to 200 * create vnodes for devices prior to the root mount (with mp == NULL). 201 */ 202 int 203 getspecialvnode(enum vtagtype tag, struct mount *mp, 204 struct vop_ops **ops, 205 struct vnode **vpp, int lktimeout, int lkflags) 206 { 207 struct vnode *vp; 208 209 vp = allocvnode(lktimeout, lkflags); 210 vp->v_tag = tag; 211 vp->v_data = NULL; 212 vp->v_ops = ops; 213 214 if (mp == NULL) 215 mp = &dummymount; 216 217 /* 218 * Placing the vnode on the mount point's queue makes it visible. 219 * VNON prevents it from being messed with, however. 220 */ 221 insmntque(vp, mp); 222 223 /* 224 * A VX locked & refd vnode is returned. 225 */ 226 *vpp = vp; 227 return (0); 228 } 229 230 /* 231 * Interlock against an unmount, return 0 on success, non-zero on failure. 232 * 233 * The passed flag may be 0 or LK_NOWAIT and is only used if an unmount 234 * is in-progress. 235 * 236 * If no unmount is in-progress LK_NOWAIT is ignored. No other flag bits 237 * are used. A shared locked will be obtained and the filesystem will not 238 * be unmountable until the lock is released. 239 */ 240 int 241 vfs_busy(struct mount *mp, int flags) 242 { 243 int lkflags; 244 245 atomic_add_int(&mp->mnt_refs, 1); 246 lwkt_gettoken(&mp->mnt_token); 247 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 248 if (flags & LK_NOWAIT) { 249 lwkt_reltoken(&mp->mnt_token); 250 atomic_add_int(&mp->mnt_refs, -1); 251 return (ENOENT); 252 } 253 /* XXX not MP safe */ 254 mp->mnt_kern_flag |= MNTK_MWAIT; 255 /* 256 * Since all busy locks are shared except the exclusive 257 * lock granted when unmounting, the only place that a 258 * wakeup needs to be done is at the release of the 259 * exclusive lock at the end of dounmount. 260 */ 261 tsleep((caddr_t)mp, 0, "vfs_busy", 0); 262 lwkt_reltoken(&mp->mnt_token); 263 atomic_add_int(&mp->mnt_refs, -1); 264 return (ENOENT); 265 } 266 lkflags = LK_SHARED; 267 if (lockmgr(&mp->mnt_lock, lkflags)) 268 panic("vfs_busy: unexpected lock failure"); 269 lwkt_reltoken(&mp->mnt_token); 270 return (0); 271 } 272 273 /* 274 * Free a busy filesystem. 275 * 276 * Decrement refs before releasing the lock so e.g. a pending umount 277 * doesn't give us an unexpected busy error. 278 */ 279 void 280 vfs_unbusy(struct mount *mp) 281 { 282 atomic_add_int(&mp->mnt_refs, -1); 283 lockmgr(&mp->mnt_lock, LK_RELEASE); 284 } 285 286 /* 287 * Lookup a filesystem type, and if found allocate and initialize 288 * a mount structure for it. 289 * 290 * Devname is usually updated by mount(8) after booting. 291 */ 292 int 293 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp) 294 { 295 struct vfsconf *vfsp; 296 struct mount *mp; 297 298 if (fstypename == NULL) 299 return (ENODEV); 300 301 vfsp = vfsconf_find_by_name(fstypename); 302 if (vfsp == NULL) 303 return (ENODEV); 304 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO); 305 mount_init(mp); 306 lockinit(&mp->mnt_lock, "vfslock", VLKTIMEOUT, 0); 307 308 vfs_busy(mp, 0); 309 mp->mnt_vfc = vfsp; 310 mp->mnt_op = vfsp->vfc_vfsops; 311 vfsp->vfc_refcount++; 312 mp->mnt_stat.f_type = vfsp->vfc_typenum; 313 mp->mnt_flag |= MNT_RDONLY; 314 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 315 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 316 copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 317 *mpp = mp; 318 return (0); 319 } 320 321 /* 322 * Basic mount structure initialization 323 */ 324 void 325 mount_init(struct mount *mp) 326 { 327 lockinit(&mp->mnt_lock, "vfslock", hz*5, 0); 328 lwkt_token_init(&mp->mnt_token, "permnt"); 329 330 TAILQ_INIT(&mp->mnt_vnodescan_list); 331 TAILQ_INIT(&mp->mnt_nvnodelist); 332 TAILQ_INIT(&mp->mnt_reservedvnlist); 333 TAILQ_INIT(&mp->mnt_jlist); 334 mp->mnt_nvnodelistsize = 0; 335 mp->mnt_flag = 0; 336 mp->mnt_iosize_max = MAXPHYS; 337 vn_syncer_thr_create(mp); 338 } 339 340 /* 341 * Lookup a mount point by filesystem identifier. 342 */ 343 struct mount * 344 vfs_getvfs(fsid_t *fsid) 345 { 346 struct mount *mp; 347 348 lwkt_gettoken(&mountlist_token); 349 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 350 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 351 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 352 break; 353 } 354 } 355 lwkt_reltoken(&mountlist_token); 356 return (mp); 357 } 358 359 /* 360 * Get a new unique fsid. Try to make its val[0] unique, since this value 361 * will be used to create fake device numbers for stat(). Also try (but 362 * not so hard) make its val[0] unique mod 2^16, since some emulators only 363 * support 16-bit device numbers. We end up with unique val[0]'s for the 364 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 365 * 366 * Keep in mind that several mounts may be running in parallel. Starting 367 * the search one past where the previous search terminated is both a 368 * micro-optimization and a defense against returning the same fsid to 369 * different mounts. 370 */ 371 void 372 vfs_getnewfsid(struct mount *mp) 373 { 374 static u_int16_t mntid_base; 375 fsid_t tfsid; 376 int mtype; 377 378 lwkt_gettoken(&mntid_token); 379 mtype = mp->mnt_vfc->vfc_typenum; 380 tfsid.val[1] = mtype; 381 mtype = (mtype & 0xFF) << 24; 382 for (;;) { 383 tfsid.val[0] = makeudev(255, 384 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 385 mntid_base++; 386 if (vfs_getvfs(&tfsid) == NULL) 387 break; 388 } 389 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 390 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 391 lwkt_reltoken(&mntid_token); 392 } 393 394 /* 395 * Set the FSID for a new mount point to the template. Adjust 396 * the FSID to avoid collisions. 397 */ 398 int 399 vfs_setfsid(struct mount *mp, fsid_t *template) 400 { 401 int didmunge = 0; 402 403 bzero(&mp->mnt_stat.f_fsid, sizeof(mp->mnt_stat.f_fsid)); 404 for (;;) { 405 if (vfs_getvfs(template) == NULL) 406 break; 407 didmunge = 1; 408 ++template->val[1]; 409 } 410 mp->mnt_stat.f_fsid = *template; 411 return(didmunge); 412 } 413 414 /* 415 * This routine is called when we have too many vnodes. It attempts 416 * to free <count> vnodes and will potentially free vnodes that still 417 * have VM backing store (VM backing store is typically the cause 418 * of a vnode blowout so we want to do this). Therefore, this operation 419 * is not considered cheap. 420 * 421 * A number of conditions may prevent a vnode from being reclaimed. 422 * the buffer cache may have references on the vnode, a directory 423 * vnode may still have references due to the namei cache representing 424 * underlying files, or the vnode may be in active use. It is not 425 * desireable to reuse such vnodes. These conditions may cause the 426 * number of vnodes to reach some minimum value regardless of what 427 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 428 */ 429 430 /* 431 * Attempt to recycle vnodes in a context that is always safe to block. 432 * Calling vlrurecycle() from the bowels of file system code has some 433 * interesting deadlock problems. 434 */ 435 static struct thread *vnlruthread; 436 437 static void 438 vnlru_proc(void) 439 { 440 struct thread *td = curthread; 441 442 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td, 443 SHUTDOWN_PRI_FIRST); 444 445 for (;;) { 446 kproc_suspend_loop(); 447 448 /* 449 * Try to free some vnodes if we have too many. Trigger based 450 * on potentially freeable vnodes but calculate the count 451 * based on total vnodes. 452 * 453 * (long) -> deal with 64 bit machines, intermediate overflow 454 */ 455 if (numvnodes >= desiredvnodes * 9 / 10 && 456 cachedvnodes + inactivevnodes >= desiredvnodes * 5 / 10) { 457 int count = numvnodes - desiredvnodes * 9 / 10; 458 459 if (count > (cachedvnodes + inactivevnodes) / 100) 460 count = (cachedvnodes + inactivevnodes) / 100; 461 if (count < 5) 462 count = 5; 463 freesomevnodes(count); 464 } 465 466 /* 467 * Do non-critical-path (more robust) cache cleaning, 468 * even if vnode counts are nominal, to try to avoid 469 * having to do it in the critical path. 470 */ 471 cache_hysteresis(0); 472 473 /* 474 * Nothing to do if most of our vnodes are already on 475 * the free list. 476 */ 477 if (numvnodes <= desiredvnodes * 9 / 10 || 478 cachedvnodes + inactivevnodes <= desiredvnodes * 5 / 10) { 479 tsleep(vnlruthread, 0, "vlruwt", hz); 480 continue; 481 } 482 } 483 } 484 485 /* 486 * MOUNTLIST FUNCTIONS 487 */ 488 489 /* 490 * mountlist_insert (MP SAFE) 491 * 492 * Add a new mount point to the mount list. 493 */ 494 void 495 mountlist_insert(struct mount *mp, int how) 496 { 497 lwkt_gettoken(&mountlist_token); 498 if (how == MNTINS_FIRST) 499 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 500 else 501 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 502 lwkt_reltoken(&mountlist_token); 503 } 504 505 /* 506 * mountlist_interlock (MP SAFE) 507 * 508 * Execute the specified interlock function with the mountlist token 509 * held. The function will be called in a serialized fashion verses 510 * other functions called through this mechanism. 511 */ 512 int 513 mountlist_interlock(int (*callback)(struct mount *), struct mount *mp) 514 { 515 int error; 516 517 lwkt_gettoken(&mountlist_token); 518 error = callback(mp); 519 lwkt_reltoken(&mountlist_token); 520 return (error); 521 } 522 523 /* 524 * mountlist_boot_getfirst (DURING BOOT ONLY) 525 * 526 * This function returns the first mount on the mountlist, which is 527 * expected to be the root mount. Since no interlocks are obtained 528 * this function is only safe to use during booting. 529 */ 530 531 struct mount * 532 mountlist_boot_getfirst(void) 533 { 534 return(TAILQ_FIRST(&mountlist)); 535 } 536 537 /* 538 * mountlist_remove (MP SAFE) 539 * 540 * Remove a node from the mountlist. If this node is the next scan node 541 * for any active mountlist scans, the active mountlist scan will be 542 * adjusted to skip the node, thus allowing removals during mountlist 543 * scans. 544 */ 545 void 546 mountlist_remove(struct mount *mp) 547 { 548 struct mountscan_info *msi; 549 550 lwkt_gettoken(&mountlist_token); 551 TAILQ_FOREACH(msi, &mountscan_list, msi_entry) { 552 if (msi->msi_node == mp) { 553 if (msi->msi_how & MNTSCAN_FORWARD) 554 msi->msi_node = TAILQ_NEXT(mp, mnt_list); 555 else 556 msi->msi_node = TAILQ_PREV(mp, mntlist, mnt_list); 557 } 558 } 559 TAILQ_REMOVE(&mountlist, mp, mnt_list); 560 lwkt_reltoken(&mountlist_token); 561 } 562 563 /* 564 * mountlist_exists (MP SAFE) 565 * 566 * Checks if a node exists in the mountlist. 567 * This function is mainly used by VFS quota code to check if a 568 * cached nullfs struct mount pointer is still valid at use time 569 * 570 * FIXME: there is no warranty the mp passed to that function 571 * will be the same one used by VFS_ACCOUNT() later 572 */ 573 int 574 mountlist_exists(struct mount *mp) 575 { 576 int node_exists = 0; 577 struct mount* lmp; 578 579 lwkt_gettoken(&mountlist_token); 580 TAILQ_FOREACH(lmp, &mountlist, mnt_list) { 581 if (lmp == mp) { 582 node_exists = 1; 583 break; 584 } 585 } 586 lwkt_reltoken(&mountlist_token); 587 return(node_exists); 588 } 589 590 /* 591 * mountlist_scan (MP SAFE) 592 * 593 * Safely scan the mount points on the mount list. Unless otherwise 594 * specified each mount point will be busied prior to the callback and 595 * unbusied afterwords. The callback may safely remove any mount point 596 * without interfering with the scan. If the current callback 597 * mount is removed the scanner will not attempt to unbusy it. 598 * 599 * If a mount node cannot be busied it is silently skipped. 600 * 601 * The callback return value is aggregated and a total is returned. A return 602 * value of < 0 is not aggregated and will terminate the scan. 603 * 604 * MNTSCAN_FORWARD - the mountlist is scanned in the forward direction 605 * MNTSCAN_REVERSE - the mountlist is scanned in reverse 606 * MNTSCAN_NOBUSY - the scanner will make the callback without busying 607 * the mount node. 608 */ 609 int 610 mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how) 611 { 612 struct mountscan_info info; 613 struct mount *mp; 614 int count; 615 int res; 616 617 lwkt_gettoken(&mountlist_token); 618 619 info.msi_how = how; 620 info.msi_node = NULL; /* paranoia */ 621 TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry); 622 623 res = 0; 624 625 if (how & MNTSCAN_FORWARD) { 626 info.msi_node = TAILQ_FIRST(&mountlist); 627 while ((mp = info.msi_node) != NULL) { 628 if (how & MNTSCAN_NOBUSY) { 629 count = callback(mp, data); 630 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 631 count = callback(mp, data); 632 if (mp == info.msi_node) 633 vfs_unbusy(mp); 634 } else { 635 count = 0; 636 } 637 if (count < 0) 638 break; 639 res += count; 640 if (mp == info.msi_node) 641 info.msi_node = TAILQ_NEXT(mp, mnt_list); 642 } 643 } else if (how & MNTSCAN_REVERSE) { 644 info.msi_node = TAILQ_LAST(&mountlist, mntlist); 645 while ((mp = info.msi_node) != NULL) { 646 if (how & MNTSCAN_NOBUSY) { 647 count = callback(mp, data); 648 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 649 count = callback(mp, data); 650 if (mp == info.msi_node) 651 vfs_unbusy(mp); 652 } else { 653 count = 0; 654 } 655 if (count < 0) 656 break; 657 res += count; 658 if (mp == info.msi_node) 659 info.msi_node = TAILQ_PREV(mp, mntlist, mnt_list); 660 } 661 } 662 TAILQ_REMOVE(&mountscan_list, &info, msi_entry); 663 lwkt_reltoken(&mountlist_token); 664 return(res); 665 } 666 667 /* 668 * MOUNT RELATED VNODE FUNCTIONS 669 */ 670 671 static struct kproc_desc vnlru_kp = { 672 "vnlru", 673 vnlru_proc, 674 &vnlruthread 675 }; 676 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp) 677 678 /* 679 * Move a vnode from one mount queue to another. 680 */ 681 void 682 insmntque(struct vnode *vp, struct mount *mp) 683 { 684 struct mount *omp; 685 686 /* 687 * Delete from old mount point vnode list, if on one. 688 */ 689 if ((omp = vp->v_mount) != NULL) { 690 lwkt_gettoken(&omp->mnt_token); 691 KKASSERT(omp == vp->v_mount); 692 KASSERT(omp->mnt_nvnodelistsize > 0, 693 ("bad mount point vnode list size")); 694 vremovevnodemnt(vp); 695 omp->mnt_nvnodelistsize--; 696 lwkt_reltoken(&omp->mnt_token); 697 } 698 699 /* 700 * Insert into list of vnodes for the new mount point, if available. 701 * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer. 702 */ 703 if (mp == NULL) { 704 vp->v_mount = NULL; 705 return; 706 } 707 lwkt_gettoken(&mp->mnt_token); 708 vp->v_mount = mp; 709 if (mp->mnt_syncer) { 710 TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes); 711 } else { 712 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 713 } 714 mp->mnt_nvnodelistsize++; 715 lwkt_reltoken(&mp->mnt_token); 716 } 717 718 719 /* 720 * Scan the vnodes under a mount point and issue appropriate callbacks. 721 * 722 * The fastfunc() callback is called with just the mountlist token held 723 * (no vnode lock). It may not block and the vnode may be undergoing 724 * modifications while the caller is processing it. The vnode will 725 * not be entirely destroyed, however, due to the fact that the mountlist 726 * token is held. A return value < 0 skips to the next vnode without calling 727 * the slowfunc(), a return value > 0 terminates the loop. 728 * 729 * The slowfunc() callback is called after the vnode has been successfully 730 * locked based on passed flags. The vnode is skipped if it gets rearranged 731 * or destroyed while blocking on the lock. A non-zero return value from 732 * the slow function terminates the loop. The slow function is allowed to 733 * arbitrarily block. The scanning code guarentees consistency of operation 734 * even if the slow function deletes or moves the node, or blocks and some 735 * other thread deletes or moves the node. 736 * 737 * NOTE: We hold vmobj_token to prevent a VM object from being destroyed 738 * out from under the fastfunc()'s vnode test. It will not prevent 739 * v_object from getting NULL'd out but it will ensure that the 740 * pointer (if we race) will remain stable. Only needed when 741 * fastfunc is non-NULL. 742 */ 743 int 744 vmntvnodescan( 745 struct mount *mp, 746 int flags, 747 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data), 748 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data), 749 void *data 750 ) { 751 struct vmntvnodescan_info info; 752 struct vnode *vp; 753 int r = 0; 754 int maxcount = mp->mnt_nvnodelistsize * 2; 755 int stopcount = 0; 756 int count = 0; 757 758 lwkt_gettoken(&mp->mnt_token); 759 if (fastfunc) 760 lwkt_gettoken(&vmobj_token); 761 762 /* 763 * If asked to do one pass stop after iterating available vnodes. 764 * Under heavy loads new vnodes can be added while we are scanning, 765 * so this isn't perfect. Create a slop factor of 2x. 766 */ 767 if (flags & VMSC_ONEPASS) 768 stopcount = mp->mnt_nvnodelistsize; 769 770 info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 771 TAILQ_INSERT_TAIL(&mp->mnt_vnodescan_list, &info, entry); 772 773 while ((vp = info.vp) != NULL) { 774 if (--maxcount == 0) { 775 kprintf("Warning: excessive fssync iteration\n"); 776 maxcount = mp->mnt_nvnodelistsize * 2; 777 } 778 779 /* 780 * Skip if visible but not ready, or special (e.g. 781 * mp->mnt_syncer) 782 */ 783 if (vp->v_type == VNON) 784 goto next; 785 KKASSERT(vp->v_mount == mp); 786 787 /* 788 * Quick test. A negative return continues the loop without 789 * calling the slow test. 0 continues onto the slow test. 790 * A positive number aborts the loop. 791 */ 792 if (fastfunc) { 793 if ((r = fastfunc(mp, vp, data)) < 0) { 794 r = 0; 795 goto next; 796 } 797 if (r) 798 break; 799 } 800 801 /* 802 * Get a vxlock on the vnode, retry if it has moved or isn't 803 * in the mountlist where we expect it. 804 */ 805 if (slowfunc) { 806 int error; 807 808 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 809 case VMSC_GETVP: 810 error = vget(vp, LK_EXCLUSIVE); 811 break; 812 case VMSC_GETVP|VMSC_NOWAIT: 813 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT); 814 break; 815 case VMSC_GETVX: 816 vx_get(vp); 817 error = 0; 818 break; 819 default: 820 error = 0; 821 break; 822 } 823 if (error) 824 goto next; 825 /* 826 * Do not call the slow function if the vnode is 827 * invalid or if it was ripped out from under us 828 * while we (potentially) blocked. 829 */ 830 if (info.vp == vp && vp->v_type != VNON) 831 r = slowfunc(mp, vp, data); 832 833 /* 834 * Cleanup 835 */ 836 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 837 case VMSC_GETVP: 838 case VMSC_GETVP|VMSC_NOWAIT: 839 vput(vp); 840 break; 841 case VMSC_GETVX: 842 vx_put(vp); 843 break; 844 default: 845 break; 846 } 847 if (r != 0) 848 break; 849 } 850 851 next: 852 /* 853 * Yield after some processing. Depending on the number 854 * of vnodes, we might wind up running for a long time. 855 * Because threads are not preemptable, time critical 856 * userland processes might starve. Give them a chance 857 * now and then. 858 */ 859 if (++count == 10000) { 860 /* 861 * We really want to yield a bit, so we simply 862 * sleep a tick 863 */ 864 tsleep(mp, 0, "vnodescn", 1); 865 count = 0; 866 } 867 868 /* 869 * If doing one pass this decrements to zero. If it starts 870 * at zero it is effectively unlimited for the purposes of 871 * this loop. 872 */ 873 if (--stopcount == 0) 874 break; 875 876 /* 877 * Iterate. If the vnode was ripped out from under us 878 * info.vp will already point to the next vnode, otherwise 879 * we have to obtain the next valid vnode ourselves. 880 */ 881 if (info.vp == vp) 882 info.vp = TAILQ_NEXT(vp, v_nmntvnodes); 883 } 884 885 TAILQ_REMOVE(&mp->mnt_vnodescan_list, &info, entry); 886 if (fastfunc) 887 lwkt_reltoken(&vmobj_token); 888 lwkt_reltoken(&mp->mnt_token); 889 return(r); 890 } 891 892 /* 893 * Remove any vnodes in the vnode table belonging to mount point mp. 894 * 895 * If FORCECLOSE is not specified, there should not be any active ones, 896 * return error if any are found (nb: this is a user error, not a 897 * system error). If FORCECLOSE is specified, detach any active vnodes 898 * that are found. 899 * 900 * If WRITECLOSE is set, only flush out regular file vnodes open for 901 * writing. 902 * 903 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped. 904 * 905 * `rootrefs' specifies the base reference count for the root vnode 906 * of this filesystem. The root vnode is considered busy if its 907 * v_refcnt exceeds this value. On a successful return, vflush() 908 * will call vrele() on the root vnode exactly rootrefs times. 909 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 910 * be zero. 911 */ 912 #ifdef DIAGNOSTIC 913 static int busyprt = 0; /* print out busy vnodes */ 914 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 915 #endif 916 917 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data); 918 919 struct vflush_info { 920 int flags; 921 int busy; 922 thread_t td; 923 }; 924 925 int 926 vflush(struct mount *mp, int rootrefs, int flags) 927 { 928 struct thread *td = curthread; /* XXX */ 929 struct vnode *rootvp = NULL; 930 int error; 931 struct vflush_info vflush_info; 932 933 if (rootrefs > 0) { 934 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 935 ("vflush: bad args")); 936 /* 937 * Get the filesystem root vnode. We can vput() it 938 * immediately, since with rootrefs > 0, it won't go away. 939 */ 940 if ((error = VFS_ROOT(mp, &rootvp)) != 0) { 941 if ((flags & FORCECLOSE) == 0) 942 return (error); 943 rootrefs = 0; 944 /* continue anyway */ 945 } 946 if (rootrefs) 947 vput(rootvp); 948 } 949 950 vflush_info.busy = 0; 951 vflush_info.flags = flags; 952 vflush_info.td = td; 953 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info); 954 955 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 956 /* 957 * If just the root vnode is busy, and if its refcount 958 * is equal to `rootrefs', then go ahead and kill it. 959 */ 960 KASSERT(vflush_info.busy > 0, ("vflush: not busy")); 961 KASSERT(VREFCNT(rootvp) >= rootrefs, ("vflush: rootrefs")); 962 if (vflush_info.busy == 1 && VREFCNT(rootvp) == rootrefs) { 963 vx_lock(rootvp); 964 vgone_vxlocked(rootvp); 965 vx_unlock(rootvp); 966 vflush_info.busy = 0; 967 } 968 } 969 if (vflush_info.busy) 970 return (EBUSY); 971 for (; rootrefs > 0; rootrefs--) 972 vrele(rootvp); 973 return (0); 974 } 975 976 /* 977 * The scan callback is made with an VX locked vnode. 978 */ 979 static int 980 vflush_scan(struct mount *mp, struct vnode *vp, void *data) 981 { 982 struct vflush_info *info = data; 983 struct vattr vattr; 984 int flags = info->flags; 985 986 /* 987 * Generally speaking try to deactivate on 0 refs (catch-all) 988 */ 989 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 990 991 /* 992 * Skip over a vnodes marked VSYSTEM. 993 */ 994 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 995 return(0); 996 } 997 998 /* 999 * Do not force-close VCHR or VBLK vnodes 1000 */ 1001 if (vp->v_type == VCHR || vp->v_type == VBLK) 1002 flags &= ~(WRITECLOSE|FORCECLOSE); 1003 1004 /* 1005 * If WRITECLOSE is set, flush out unlinked but still open 1006 * files (even if open only for reading) and regular file 1007 * vnodes open for writing. 1008 */ 1009 if ((flags & WRITECLOSE) && 1010 (vp->v_type == VNON || 1011 (VOP_GETATTR(vp, &vattr) == 0 && 1012 vattr.va_nlink > 0)) && 1013 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1014 return(0); 1015 } 1016 1017 /* 1018 * If we are the only holder (refcnt of 1) or the vnode is in 1019 * termination (refcnt < 0), we can vgone the vnode. 1020 */ 1021 if (VREFCNT(vp) <= 1) { 1022 vgone_vxlocked(vp); 1023 return(0); 1024 } 1025 1026 /* 1027 * If FORCECLOSE is set, forcibly destroy the vnode and then move 1028 * it to a dummymount structure so vop_*() functions don't deref 1029 * a NULL pointer. 1030 */ 1031 if (flags & FORCECLOSE) { 1032 vhold(vp); 1033 vgone_vxlocked(vp); 1034 if (vp->v_mount == NULL) 1035 insmntque(vp, &dummymount); 1036 vdrop(vp); 1037 return(0); 1038 } 1039 if (vp->v_type == VCHR || vp->v_type == VBLK) 1040 kprintf("vflush: Warning, cannot destroy busy device vnode\n"); 1041 #ifdef DIAGNOSTIC 1042 if (busyprt) 1043 vprint("vflush: busy vnode", vp); 1044 #endif 1045 ++info->busy; 1046 return(0); 1047 } 1048 1049 void 1050 add_bio_ops(struct bio_ops *ops) 1051 { 1052 TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry); 1053 } 1054 1055 void 1056 rem_bio_ops(struct bio_ops *ops) 1057 { 1058 TAILQ_REMOVE(&bio_ops_list, ops, entry); 1059 } 1060 1061 /* 1062 * This calls the bio_ops io_sync function either for a mount point 1063 * or generally. 1064 * 1065 * WARNING: softdeps is weirdly coded and just isn't happy unless 1066 * io_sync is called with a NULL mount from the general syncing code. 1067 */ 1068 void 1069 bio_ops_sync(struct mount *mp) 1070 { 1071 struct bio_ops *ops; 1072 1073 if (mp) { 1074 if ((ops = mp->mnt_bioops) != NULL) 1075 ops->io_sync(mp); 1076 } else { 1077 TAILQ_FOREACH(ops, &bio_ops_list, entry) { 1078 ops->io_sync(NULL); 1079 } 1080 } 1081 } 1082 1083 /* 1084 * Lookup a mount point by nch 1085 */ 1086 struct mount * 1087 mount_get_by_nc(struct namecache *ncp) 1088 { 1089 struct mount *mp = NULL; 1090 1091 lwkt_gettoken(&mountlist_token); 1092 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1093 if (ncp == mp->mnt_ncmountpt.ncp) 1094 break; 1095 } 1096 lwkt_reltoken(&mountlist_token); 1097 return (mp); 1098 } 1099 1100