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 vp->v_pbuf_count = nswbuf_kva / NSWBUF_SPLIT; 182 183 /* 184 * Placing the vnode on the mount point's queue makes it visible. 185 * VNON prevents it from being messed with, however. 186 */ 187 insmntque(vp, mp); 188 189 /* 190 * A VX locked & refd vnode is returned. 191 */ 192 *vpp = vp; 193 return (0); 194 } 195 196 /* 197 * This function creates vnodes with special operations vectors. The 198 * mount point is optional. 199 * 200 * This routine is being phased out but is still used by vfs_conf to 201 * create vnodes for devices prior to the root mount (with mp == NULL). 202 */ 203 int 204 getspecialvnode(enum vtagtype tag, struct mount *mp, 205 struct vop_ops **ops, 206 struct vnode **vpp, int lktimeout, int lkflags) 207 { 208 struct vnode *vp; 209 210 vp = allocvnode(lktimeout, lkflags); 211 vp->v_tag = tag; 212 vp->v_data = NULL; 213 vp->v_ops = ops; 214 215 if (mp == NULL) 216 mp = &dummymount; 217 218 /* 219 * Placing the vnode on the mount point's queue makes it visible. 220 * VNON prevents it from being messed with, however. 221 */ 222 insmntque(vp, mp); 223 224 /* 225 * A VX locked & refd vnode is returned. 226 */ 227 *vpp = vp; 228 return (0); 229 } 230 231 /* 232 * Interlock against an unmount, return 0 on success, non-zero on failure. 233 * 234 * The passed flag may be 0 or LK_NOWAIT and is only used if an unmount 235 * is in-progress. 236 * 237 * If no unmount is in-progress LK_NOWAIT is ignored. No other flag bits 238 * are used. A shared locked will be obtained and the filesystem will not 239 * be unmountable until the lock is released. 240 */ 241 int 242 vfs_busy(struct mount *mp, int flags) 243 { 244 int lkflags; 245 246 atomic_add_int(&mp->mnt_refs, 1); 247 lwkt_gettoken(&mp->mnt_token); 248 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 249 if (flags & LK_NOWAIT) { 250 lwkt_reltoken(&mp->mnt_token); 251 atomic_add_int(&mp->mnt_refs, -1); 252 return (ENOENT); 253 } 254 /* XXX not MP safe */ 255 mp->mnt_kern_flag |= MNTK_MWAIT; 256 /* 257 * Since all busy locks are shared except the exclusive 258 * lock granted when unmounting, the only place that a 259 * wakeup needs to be done is at the release of the 260 * exclusive lock at the end of dounmount. 261 */ 262 tsleep((caddr_t)mp, 0, "vfs_busy", 0); 263 lwkt_reltoken(&mp->mnt_token); 264 atomic_add_int(&mp->mnt_refs, -1); 265 return (ENOENT); 266 } 267 lkflags = LK_SHARED; 268 if (lockmgr(&mp->mnt_lock, lkflags)) 269 panic("vfs_busy: unexpected lock failure"); 270 lwkt_reltoken(&mp->mnt_token); 271 return (0); 272 } 273 274 /* 275 * Free a busy filesystem. 276 * 277 * Decrement refs before releasing the lock so e.g. a pending umount 278 * doesn't give us an unexpected busy error. 279 */ 280 void 281 vfs_unbusy(struct mount *mp) 282 { 283 atomic_add_int(&mp->mnt_refs, -1); 284 lockmgr(&mp->mnt_lock, LK_RELEASE); 285 } 286 287 /* 288 * Lookup a filesystem type, and if found allocate and initialize 289 * a mount structure for it. 290 * 291 * Devname is usually updated by mount(8) after booting. 292 */ 293 int 294 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp) 295 { 296 struct vfsconf *vfsp; 297 struct mount *mp; 298 299 if (fstypename == NULL) 300 return (ENODEV); 301 302 vfsp = vfsconf_find_by_name(fstypename); 303 if (vfsp == NULL) 304 return (ENODEV); 305 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO); 306 mount_init(mp); 307 lockinit(&mp->mnt_lock, "vfslock", VLKTIMEOUT, 0); 308 309 vfs_busy(mp, 0); 310 mp->mnt_vfc = vfsp; 311 mp->mnt_op = vfsp->vfc_vfsops; 312 mp->mnt_pbuf_count = nswbuf_kva / NSWBUF_SPLIT; 313 vfsp->vfc_refcount++; 314 mp->mnt_stat.f_type = vfsp->vfc_typenum; 315 mp->mnt_flag |= MNT_RDONLY; 316 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 317 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 318 copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 319 *mpp = mp; 320 return (0); 321 } 322 323 /* 324 * Basic mount structure initialization 325 */ 326 void 327 mount_init(struct mount *mp) 328 { 329 lockinit(&mp->mnt_lock, "vfslock", hz*5, 0); 330 lwkt_token_init(&mp->mnt_token, "permnt"); 331 332 TAILQ_INIT(&mp->mnt_vnodescan_list); 333 TAILQ_INIT(&mp->mnt_nvnodelist); 334 TAILQ_INIT(&mp->mnt_reservedvnlist); 335 TAILQ_INIT(&mp->mnt_jlist); 336 mp->mnt_nvnodelistsize = 0; 337 mp->mnt_flag = 0; 338 mp->mnt_hold = 1; 339 mp->mnt_iosize_max = MAXPHYS; 340 vn_syncer_thr_create(mp); 341 } 342 343 void 344 mount_hold(struct mount *mp) 345 { 346 atomic_add_int(&mp->mnt_hold, 1); 347 } 348 349 void 350 mount_drop(struct mount *mp) 351 { 352 if (atomic_fetchadd_int(&mp->mnt_hold, -1) == 1) 353 kfree(mp, M_MOUNT); 354 } 355 356 /* 357 * Lookup a mount point by filesystem identifier. 358 */ 359 struct mount * 360 vfs_getvfs(fsid_t *fsid) 361 { 362 struct mount *mp; 363 364 lwkt_gettoken(&mountlist_token); 365 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 366 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 367 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 368 break; 369 } 370 } 371 lwkt_reltoken(&mountlist_token); 372 return (mp); 373 } 374 375 /* 376 * Get a new unique fsid. Try to make its val[0] unique, since this value 377 * will be used to create fake device numbers for stat(). Also try (but 378 * not so hard) make its val[0] unique mod 2^16, since some emulators only 379 * support 16-bit device numbers. We end up with unique val[0]'s for the 380 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 381 * 382 * Keep in mind that several mounts may be running in parallel. Starting 383 * the search one past where the previous search terminated is both a 384 * micro-optimization and a defense against returning the same fsid to 385 * different mounts. 386 */ 387 void 388 vfs_getnewfsid(struct mount *mp) 389 { 390 static u_int16_t mntid_base; 391 fsid_t tfsid; 392 int mtype; 393 394 lwkt_gettoken(&mntid_token); 395 mtype = mp->mnt_vfc->vfc_typenum; 396 tfsid.val[1] = mtype; 397 mtype = (mtype & 0xFF) << 24; 398 for (;;) { 399 tfsid.val[0] = makeudev(255, 400 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 401 mntid_base++; 402 if (vfs_getvfs(&tfsid) == NULL) 403 break; 404 } 405 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 406 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 407 lwkt_reltoken(&mntid_token); 408 } 409 410 /* 411 * Set the FSID for a new mount point to the template. Adjust 412 * the FSID to avoid collisions. 413 */ 414 int 415 vfs_setfsid(struct mount *mp, fsid_t *template) 416 { 417 int didmunge = 0; 418 419 bzero(&mp->mnt_stat.f_fsid, sizeof(mp->mnt_stat.f_fsid)); 420 for (;;) { 421 if (vfs_getvfs(template) == NULL) 422 break; 423 didmunge = 1; 424 ++template->val[1]; 425 } 426 mp->mnt_stat.f_fsid = *template; 427 return(didmunge); 428 } 429 430 /* 431 * This routine is called when we have too many vnodes. It attempts 432 * to free <count> vnodes and will potentially free vnodes that still 433 * have VM backing store (VM backing store is typically the cause 434 * of a vnode blowout so we want to do this). Therefore, this operation 435 * is not considered cheap. 436 * 437 * A number of conditions may prevent a vnode from being reclaimed. 438 * the buffer cache may have references on the vnode, a directory 439 * vnode may still have references due to the namei cache representing 440 * underlying files, or the vnode may be in active use. It is not 441 * desireable to reuse such vnodes. These conditions may cause the 442 * number of vnodes to reach some minimum value regardless of what 443 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 444 */ 445 446 /* 447 * Attempt to recycle vnodes in a context that is always safe to block. 448 * Calling vlrurecycle() from the bowels of file system code has some 449 * interesting deadlock problems. 450 */ 451 static struct thread *vnlruthread; 452 453 static void 454 vnlru_proc(void) 455 { 456 struct thread *td = curthread; 457 458 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td, 459 SHUTDOWN_PRI_FIRST); 460 461 for (;;) { 462 int ncached; 463 464 kproc_suspend_loop(); 465 466 /* 467 * Try to free some vnodes if we have too many. Trigger based 468 * on potentially freeable vnodes but calculate the count 469 * based on total vnodes. 470 * 471 * (long) -> deal with 64 bit machines, intermediate overflow 472 */ 473 ncached = countcachedvnodes(1); 474 if (numvnodes >= maxvnodes * 9 / 10 && 475 ncached + inactivevnodes >= maxvnodes * 5 / 10) { 476 int count = numvnodes - maxvnodes * 9 / 10; 477 478 if (count > (ncached + inactivevnodes) / 100) 479 count = (ncached + inactivevnodes) / 100; 480 if (count < 5) 481 count = 5; 482 freesomevnodes(count); 483 } 484 485 /* 486 * Do non-critical-path (more robust) cache cleaning, 487 * even if vnode counts are nominal, to try to avoid 488 * having to do it in the critical path. 489 */ 490 cache_hysteresis(0); 491 492 /* 493 * Nothing to do if most of our vnodes are already on 494 * the free list. 495 */ 496 ncached = countcachedvnodes(1); 497 if (numvnodes <= maxvnodes * 9 / 10 || 498 ncached + inactivevnodes <= maxvnodes * 5 / 10) { 499 tsleep(vnlruthread, 0, "vlruwt", hz); 500 continue; 501 } 502 } 503 } 504 505 /* 506 * MOUNTLIST FUNCTIONS 507 */ 508 509 /* 510 * mountlist_insert (MP SAFE) 511 * 512 * Add a new mount point to the mount list. 513 */ 514 void 515 mountlist_insert(struct mount *mp, int how) 516 { 517 lwkt_gettoken(&mountlist_token); 518 if (how == MNTINS_FIRST) 519 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 520 else 521 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 522 lwkt_reltoken(&mountlist_token); 523 } 524 525 /* 526 * mountlist_interlock (MP SAFE) 527 * 528 * Execute the specified interlock function with the mountlist token 529 * held. The function will be called in a serialized fashion verses 530 * other functions called through this mechanism. 531 */ 532 int 533 mountlist_interlock(int (*callback)(struct mount *), struct mount *mp) 534 { 535 int error; 536 537 lwkt_gettoken(&mountlist_token); 538 error = callback(mp); 539 lwkt_reltoken(&mountlist_token); 540 return (error); 541 } 542 543 /* 544 * mountlist_boot_getfirst (DURING BOOT ONLY) 545 * 546 * This function returns the first mount on the mountlist, which is 547 * expected to be the root mount. Since no interlocks are obtained 548 * this function is only safe to use during booting. 549 */ 550 551 struct mount * 552 mountlist_boot_getfirst(void) 553 { 554 return(TAILQ_FIRST(&mountlist)); 555 } 556 557 /* 558 * mountlist_remove (MP SAFE) 559 * 560 * Remove a node from the mountlist. If this node is the next scan node 561 * for any active mountlist scans, the active mountlist scan will be 562 * adjusted to skip the node, thus allowing removals during mountlist 563 * scans. 564 */ 565 void 566 mountlist_remove(struct mount *mp) 567 { 568 struct mountscan_info *msi; 569 570 lwkt_gettoken(&mountlist_token); 571 TAILQ_FOREACH(msi, &mountscan_list, msi_entry) { 572 if (msi->msi_node == mp) { 573 if (msi->msi_how & MNTSCAN_FORWARD) 574 msi->msi_node = TAILQ_NEXT(mp, mnt_list); 575 else 576 msi->msi_node = TAILQ_PREV(mp, mntlist, mnt_list); 577 } 578 } 579 TAILQ_REMOVE(&mountlist, mp, mnt_list); 580 lwkt_reltoken(&mountlist_token); 581 } 582 583 /* 584 * mountlist_exists (MP SAFE) 585 * 586 * Checks if a node exists in the mountlist. 587 * This function is mainly used by VFS quota code to check if a 588 * cached nullfs struct mount pointer is still valid at use time 589 * 590 * FIXME: there is no warranty the mp passed to that function 591 * will be the same one used by VFS_ACCOUNT() later 592 */ 593 int 594 mountlist_exists(struct mount *mp) 595 { 596 int node_exists = 0; 597 struct mount* lmp; 598 599 lwkt_gettoken(&mountlist_token); 600 TAILQ_FOREACH(lmp, &mountlist, mnt_list) { 601 if (lmp == mp) { 602 node_exists = 1; 603 break; 604 } 605 } 606 lwkt_reltoken(&mountlist_token); 607 return(node_exists); 608 } 609 610 /* 611 * mountlist_scan (MP SAFE) 612 * 613 * Safely scan the mount points on the mount list. Unless otherwise 614 * specified each mount point will be busied prior to the callback and 615 * unbusied afterwords. The callback may safely remove any mount point 616 * without interfering with the scan. If the current callback 617 * mount is removed the scanner will not attempt to unbusy it. 618 * 619 * If a mount node cannot be busied it is silently skipped. 620 * 621 * The callback return value is aggregated and a total is returned. A return 622 * value of < 0 is not aggregated and will terminate the scan. 623 * 624 * MNTSCAN_FORWARD - the mountlist is scanned in the forward direction 625 * MNTSCAN_REVERSE - the mountlist is scanned in reverse 626 * MNTSCAN_NOBUSY - the scanner will make the callback without busying 627 * the mount node. 628 * 629 * NOTE: mount_hold()/mount_drop() sequence primarily helps us avoid 630 * confusion for the unbusy check, particularly if a kfree/kmalloc 631 * occurs quickly (lots of processes mounting and unmounting at the 632 * same time). 633 */ 634 int 635 mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how) 636 { 637 struct mountscan_info info; 638 struct mount *mp; 639 int count; 640 int res; 641 642 lwkt_gettoken(&mountlist_token); 643 644 info.msi_how = how; 645 info.msi_node = NULL; /* paranoia */ 646 TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry); 647 648 res = 0; 649 650 if (how & MNTSCAN_FORWARD) { 651 info.msi_node = TAILQ_FIRST(&mountlist); 652 while ((mp = info.msi_node) != NULL) { 653 mount_hold(mp); 654 if (how & MNTSCAN_NOBUSY) { 655 count = callback(mp, data); 656 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 657 count = callback(mp, data); 658 if (mp == info.msi_node) 659 vfs_unbusy(mp); 660 } else { 661 count = 0; 662 } 663 mount_drop(mp); 664 if (count < 0) 665 break; 666 res += count; 667 if (mp == info.msi_node) 668 info.msi_node = TAILQ_NEXT(mp, mnt_list); 669 } 670 } else if (how & MNTSCAN_REVERSE) { 671 info.msi_node = TAILQ_LAST(&mountlist, mntlist); 672 while ((mp = info.msi_node) != NULL) { 673 mount_hold(mp); 674 if (how & MNTSCAN_NOBUSY) { 675 count = callback(mp, data); 676 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 677 count = callback(mp, data); 678 if (mp == info.msi_node) 679 vfs_unbusy(mp); 680 } else { 681 count = 0; 682 } 683 mount_drop(mp); 684 if (count < 0) 685 break; 686 res += count; 687 if (mp == info.msi_node) 688 info.msi_node = TAILQ_PREV(mp, mntlist, mnt_list); 689 } 690 } 691 TAILQ_REMOVE(&mountscan_list, &info, msi_entry); 692 lwkt_reltoken(&mountlist_token); 693 return(res); 694 } 695 696 /* 697 * MOUNT RELATED VNODE FUNCTIONS 698 */ 699 700 static struct kproc_desc vnlru_kp = { 701 "vnlru", 702 vnlru_proc, 703 &vnlruthread 704 }; 705 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); 706 707 /* 708 * Move a vnode from one mount queue to another. 709 */ 710 void 711 insmntque(struct vnode *vp, struct mount *mp) 712 { 713 struct mount *omp; 714 715 /* 716 * Delete from old mount point vnode list, if on one. 717 */ 718 if ((omp = vp->v_mount) != NULL) { 719 lwkt_gettoken(&omp->mnt_token); 720 KKASSERT(omp == vp->v_mount); 721 KASSERT(omp->mnt_nvnodelistsize > 0, 722 ("bad mount point vnode list size")); 723 vremovevnodemnt(vp); 724 omp->mnt_nvnodelistsize--; 725 lwkt_reltoken(&omp->mnt_token); 726 } 727 728 /* 729 * Insert into list of vnodes for the new mount point, if available. 730 * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer. 731 */ 732 if (mp == NULL) { 733 vp->v_mount = NULL; 734 return; 735 } 736 lwkt_gettoken(&mp->mnt_token); 737 vp->v_mount = mp; 738 if (mp->mnt_syncer) { 739 TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes); 740 } else { 741 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 742 } 743 mp->mnt_nvnodelistsize++; 744 lwkt_reltoken(&mp->mnt_token); 745 } 746 747 748 /* 749 * Scan the vnodes under a mount point and issue appropriate callbacks. 750 * 751 * The fastfunc() callback is called with just the mountlist token held 752 * (no vnode lock). It may not block and the vnode may be undergoing 753 * modifications while the caller is processing it. The vnode will 754 * not be entirely destroyed, however, due to the fact that the mountlist 755 * token is held. A return value < 0 skips to the next vnode without calling 756 * the slowfunc(), a return value > 0 terminates the loop. 757 * 758 * WARNING! The fastfunc() should not indirect through vp->v_object, the vp 759 * data structure is unstable when called from fastfunc(). 760 * 761 * The slowfunc() callback is called after the vnode has been successfully 762 * locked based on passed flags. The vnode is skipped if it gets rearranged 763 * or destroyed while blocking on the lock. A non-zero return value from 764 * the slow function terminates the loop. The slow function is allowed to 765 * arbitrarily block. The scanning code guarentees consistency of operation 766 * even if the slow function deletes or moves the node, or blocks and some 767 * other thread deletes or moves the node. 768 */ 769 int 770 vmntvnodescan( 771 struct mount *mp, 772 int flags, 773 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data), 774 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data), 775 void *data 776 ) { 777 struct vmntvnodescan_info info; 778 struct vnode *vp; 779 int r = 0; 780 int maxcount = mp->mnt_nvnodelistsize * 2; 781 int stopcount = 0; 782 int count = 0; 783 784 lwkt_gettoken(&mp->mnt_token); 785 786 /* 787 * If asked to do one pass stop after iterating available vnodes. 788 * Under heavy loads new vnodes can be added while we are scanning, 789 * so this isn't perfect. Create a slop factor of 2x. 790 */ 791 if (flags & VMSC_ONEPASS) 792 stopcount = mp->mnt_nvnodelistsize; 793 794 info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 795 TAILQ_INSERT_TAIL(&mp->mnt_vnodescan_list, &info, entry); 796 797 while ((vp = info.vp) != NULL) { 798 if (--maxcount == 0) { 799 kprintf("Warning: excessive fssync iteration\n"); 800 maxcount = mp->mnt_nvnodelistsize * 2; 801 } 802 803 /* 804 * Skip if visible but not ready, or special (e.g. 805 * mp->mnt_syncer) 806 */ 807 if (vp->v_type == VNON) 808 goto next; 809 KKASSERT(vp->v_mount == mp); 810 811 /* 812 * Quick test. A negative return continues the loop without 813 * calling the slow test. 0 continues onto the slow test. 814 * A positive number aborts the loop. 815 */ 816 if (fastfunc) { 817 if ((r = fastfunc(mp, vp, data)) < 0) { 818 r = 0; 819 goto next; 820 } 821 if (r) 822 break; 823 } 824 825 /* 826 * Get a vxlock on the vnode, retry if it has moved or isn't 827 * in the mountlist where we expect it. 828 */ 829 if (slowfunc) { 830 int error; 831 832 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 833 case VMSC_GETVP: 834 error = vget(vp, LK_EXCLUSIVE); 835 break; 836 case VMSC_GETVP|VMSC_NOWAIT: 837 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT); 838 break; 839 case VMSC_GETVX: 840 vx_get(vp); 841 error = 0; 842 break; 843 default: 844 error = 0; 845 break; 846 } 847 if (error) 848 goto next; 849 /* 850 * Do not call the slow function if the vnode is 851 * invalid or if it was ripped out from under us 852 * while we (potentially) blocked. 853 */ 854 if (info.vp == vp && vp->v_type != VNON) 855 r = slowfunc(mp, vp, data); 856 857 /* 858 * Cleanup 859 */ 860 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 861 case VMSC_GETVP: 862 case VMSC_GETVP|VMSC_NOWAIT: 863 vput(vp); 864 break; 865 case VMSC_GETVX: 866 vx_put(vp); 867 break; 868 default: 869 break; 870 } 871 if (r != 0) 872 break; 873 } 874 875 next: 876 /* 877 * Yield after some processing. Depending on the number 878 * of vnodes, we might wind up running for a long time. 879 * Because threads are not preemptable, time critical 880 * userland processes might starve. Give them a chance 881 * now and then. 882 */ 883 if (++count == 10000) { 884 /* 885 * We really want to yield a bit, so we simply 886 * sleep a tick 887 */ 888 tsleep(mp, 0, "vnodescn", 1); 889 count = 0; 890 } 891 892 /* 893 * If doing one pass this decrements to zero. If it starts 894 * at zero it is effectively unlimited for the purposes of 895 * this loop. 896 */ 897 if (--stopcount == 0) 898 break; 899 900 /* 901 * Iterate. If the vnode was ripped out from under us 902 * info.vp will already point to the next vnode, otherwise 903 * we have to obtain the next valid vnode ourselves. 904 */ 905 if (info.vp == vp) 906 info.vp = TAILQ_NEXT(vp, v_nmntvnodes); 907 } 908 909 TAILQ_REMOVE(&mp->mnt_vnodescan_list, &info, entry); 910 lwkt_reltoken(&mp->mnt_token); 911 return(r); 912 } 913 914 /* 915 * Remove any vnodes in the vnode table belonging to mount point mp. 916 * 917 * If FORCECLOSE is not specified, there should not be any active ones, 918 * return error if any are found (nb: this is a user error, not a 919 * system error). If FORCECLOSE is specified, detach any active vnodes 920 * that are found. 921 * 922 * If WRITECLOSE is set, only flush out regular file vnodes open for 923 * writing. 924 * 925 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped. 926 * 927 * `rootrefs' specifies the base reference count for the root vnode 928 * of this filesystem. The root vnode is considered busy if its 929 * v_refcnt exceeds this value. On a successful return, vflush() 930 * will call vrele() on the root vnode exactly rootrefs times. 931 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 932 * be zero. 933 */ 934 #ifdef DIAGNOSTIC 935 static int busyprt = 0; /* print out busy vnodes */ 936 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, ""); 937 #endif 938 939 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data); 940 941 struct vflush_info { 942 int flags; 943 int busy; 944 thread_t td; 945 }; 946 947 int 948 vflush(struct mount *mp, int rootrefs, int flags) 949 { 950 struct thread *td = curthread; /* XXX */ 951 struct vnode *rootvp = NULL; 952 int error; 953 struct vflush_info vflush_info; 954 955 if (rootrefs > 0) { 956 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 957 ("vflush: bad args")); 958 /* 959 * Get the filesystem root vnode. We can vput() it 960 * immediately, since with rootrefs > 0, it won't go away. 961 */ 962 if ((error = VFS_ROOT(mp, &rootvp)) != 0) { 963 if ((flags & FORCECLOSE) == 0) 964 return (error); 965 rootrefs = 0; 966 /* continue anyway */ 967 } 968 if (rootrefs) 969 vput(rootvp); 970 } 971 972 vflush_info.busy = 0; 973 vflush_info.flags = flags; 974 vflush_info.td = td; 975 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info); 976 977 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 978 /* 979 * If just the root vnode is busy, and if its refcount 980 * is equal to `rootrefs', then go ahead and kill it. 981 */ 982 KASSERT(vflush_info.busy > 0, ("vflush: not busy")); 983 KASSERT(VREFCNT(rootvp) >= rootrefs, ("vflush: rootrefs")); 984 if (vflush_info.busy == 1 && VREFCNT(rootvp) == rootrefs) { 985 vx_lock(rootvp); 986 vgone_vxlocked(rootvp); 987 vx_unlock(rootvp); 988 vflush_info.busy = 0; 989 } 990 } 991 if (vflush_info.busy) 992 return (EBUSY); 993 for (; rootrefs > 0; rootrefs--) 994 vrele(rootvp); 995 return (0); 996 } 997 998 /* 999 * The scan callback is made with an VX locked vnode. 1000 */ 1001 static int 1002 vflush_scan(struct mount *mp, struct vnode *vp, void *data) 1003 { 1004 struct vflush_info *info = data; 1005 struct vattr vattr; 1006 int flags = info->flags; 1007 1008 /* 1009 * Generally speaking try to deactivate on 0 refs (catch-all) 1010 */ 1011 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1012 1013 /* 1014 * Skip over a vnodes marked VSYSTEM. 1015 */ 1016 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1017 return(0); 1018 } 1019 1020 /* 1021 * Do not force-close VCHR or VBLK vnodes 1022 */ 1023 if (vp->v_type == VCHR || vp->v_type == VBLK) 1024 flags &= ~(WRITECLOSE|FORCECLOSE); 1025 1026 /* 1027 * If WRITECLOSE is set, flush out unlinked but still open 1028 * files (even if open only for reading) and regular file 1029 * vnodes open for writing. 1030 */ 1031 if ((flags & WRITECLOSE) && 1032 (vp->v_type == VNON || 1033 (VOP_GETATTR(vp, &vattr) == 0 && 1034 vattr.va_nlink > 0)) && 1035 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1036 return(0); 1037 } 1038 1039 /* 1040 * If we are the only holder (refcnt of 1) or the vnode is in 1041 * termination (refcnt < 0), we can vgone the vnode. 1042 */ 1043 if (VREFCNT(vp) <= 1) { 1044 vgone_vxlocked(vp); 1045 return(0); 1046 } 1047 1048 /* 1049 * If FORCECLOSE is set, forcibly destroy the vnode and then move 1050 * it to a dummymount structure so vop_*() functions don't deref 1051 * a NULL pointer. 1052 */ 1053 if (flags & FORCECLOSE) { 1054 vhold(vp); 1055 vgone_vxlocked(vp); 1056 if (vp->v_mount == NULL) 1057 insmntque(vp, &dummymount); 1058 vdrop(vp); 1059 return(0); 1060 } 1061 if (vp->v_type == VCHR || vp->v_type == VBLK) 1062 kprintf("vflush: Warning, cannot destroy busy device vnode\n"); 1063 #ifdef DIAGNOSTIC 1064 if (busyprt) 1065 vprint("vflush: busy vnode", vp); 1066 #endif 1067 ++info->busy; 1068 return(0); 1069 } 1070 1071 void 1072 add_bio_ops(struct bio_ops *ops) 1073 { 1074 TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry); 1075 } 1076 1077 void 1078 rem_bio_ops(struct bio_ops *ops) 1079 { 1080 TAILQ_REMOVE(&bio_ops_list, ops, entry); 1081 } 1082 1083 /* 1084 * This calls the bio_ops io_sync function either for a mount point 1085 * or generally. 1086 * 1087 * WARNING: softdeps is weirdly coded and just isn't happy unless 1088 * io_sync is called with a NULL mount from the general syncing code. 1089 */ 1090 void 1091 bio_ops_sync(struct mount *mp) 1092 { 1093 struct bio_ops *ops; 1094 1095 if (mp) { 1096 if ((ops = mp->mnt_bioops) != NULL) 1097 ops->io_sync(mp); 1098 } else { 1099 TAILQ_FOREACH(ops, &bio_ops_list, entry) { 1100 ops->io_sync(NULL); 1101 } 1102 } 1103 } 1104 1105 /* 1106 * Lookup a mount point by nch 1107 */ 1108 struct mount * 1109 mount_get_by_nc(struct namecache *ncp) 1110 { 1111 struct mount *mp = NULL; 1112 1113 lwkt_gettoken(&mountlist_token); 1114 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1115 if (ncp == mp->mnt_ncmountpt.ncp) 1116 break; 1117 } 1118 lwkt_reltoken(&mountlist_token); 1119 return (mp); 1120 } 1121 1122