1 /* 2 * Copyright (c) 2004,2013-2019 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/spinlock2.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 <vm/vm.h> 86 #include <vm/vm_object.h> 87 88 struct mountscan_info { 89 TAILQ_ENTRY(mountscan_info) msi_entry; 90 int msi_how; 91 struct mount *msi_node; 92 }; 93 94 struct vmntvnodescan_info { 95 TAILQ_ENTRY(vmntvnodescan_info) entry; 96 struct vnode *vp; 97 }; 98 99 struct vnlru_info { 100 int pass; 101 }; 102 103 static int 104 mount_cmp(struct mount *mnt1, struct mount *mnt2) 105 { 106 if (mnt1->mnt_stat.f_fsid.val[0] < mnt2->mnt_stat.f_fsid.val[0]) 107 return -1; 108 if (mnt1->mnt_stat.f_fsid.val[0] > mnt2->mnt_stat.f_fsid.val[0]) 109 return 1; 110 if (mnt1->mnt_stat.f_fsid.val[1] < mnt2->mnt_stat.f_fsid.val[1]) 111 return -1; 112 if (mnt1->mnt_stat.f_fsid.val[1] > mnt2->mnt_stat.f_fsid.val[1]) 113 return 1; 114 return 0; 115 } 116 117 static int 118 mount_fsid_cmp(fsid_t *fsid, struct mount *mnt) 119 { 120 if (fsid->val[0] < mnt->mnt_stat.f_fsid.val[0]) 121 return -1; 122 if (fsid->val[0] > mnt->mnt_stat.f_fsid.val[0]) 123 return 1; 124 if (fsid->val[1] < mnt->mnt_stat.f_fsid.val[1]) 125 return -1; 126 if (fsid->val[1] > mnt->mnt_stat.f_fsid.val[1]) 127 return 1; 128 return 0; 129 } 130 131 RB_HEAD(mount_rb_tree, mount); 132 RB_PROTOTYPEX(mount_rb_tree, FSID, mount, mnt_node, mount_cmp, fsid_t *); 133 RB_GENERATE(mount_rb_tree, mount, mnt_node, mount_cmp); 134 RB_GENERATE_XLOOKUP(mount_rb_tree, FSID, mount, mnt_node, 135 mount_fsid_cmp, fsid_t *); 136 137 static int vnlru_nowhere = 0; 138 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RD, 139 &vnlru_nowhere, 0, 140 "Number of times the vnlru process ran without success"); 141 142 143 static struct lwkt_token mntid_token; 144 static struct mount dummymount; 145 146 /* note: mountlist exported to pstat */ 147 struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist); 148 struct mount_rb_tree mounttree = RB_INITIALIZER(dev_tree_mounttree); 149 static TAILQ_HEAD(,mountscan_info) mountscan_list; 150 static struct lwkt_token mountlist_token; 151 152 static TAILQ_HEAD(,bio_ops) bio_ops_list = TAILQ_HEAD_INITIALIZER(bio_ops_list); 153 154 /* 155 * Called from vfsinit() 156 */ 157 void 158 vfs_mount_init(void) 159 { 160 lwkt_token_init(&mountlist_token, "mntlist"); 161 lwkt_token_init(&mntid_token, "mntid"); 162 TAILQ_INIT(&mountscan_list); 163 mount_init(&dummymount, NULL); 164 dummymount.mnt_flag |= MNT_RDONLY; 165 dummymount.mnt_kern_flag |= MNTK_ALL_MPSAFE; 166 } 167 168 /* 169 * Support function called to remove a vnode from the mountlist and 170 * deal with side effects for scans in progress. 171 * 172 * Target mnt_token is held on call. 173 */ 174 static void 175 vremovevnodemnt(struct vnode *vp) 176 { 177 struct vmntvnodescan_info *info; 178 struct mount *mp = vp->v_mount; 179 180 TAILQ_FOREACH(info, &mp->mnt_vnodescan_list, entry) { 181 if (info->vp == vp) 182 info->vp = TAILQ_NEXT(vp, v_nmntvnodes); 183 } 184 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes); 185 } 186 187 /* 188 * Allocate a new vnode and associate it with a tag, mount point, and 189 * operations vector. 190 * 191 * A VX locked and refd vnode is returned. The caller should setup the 192 * remaining fields and vx_put() or, if he wishes to leave a vref, 193 * vx_unlock() the vnode. Or if he wishes to return a normal locked 194 * vnode, call vx_downgrade(vp); to downgrade the VX lock to a normal 195 * VN lock. 196 */ 197 int 198 getnewvnode(enum vtagtype tag, struct mount *mp, 199 struct vnode **vpp, int lktimeout, int lkflags) 200 { 201 struct vnode *vp; 202 203 KKASSERT(mp != NULL); 204 205 vp = allocvnode(lktimeout, lkflags); 206 vp->v_tag = tag; 207 vp->v_data = NULL; 208 209 /* 210 * By default the vnode is assigned the mount point's normal 211 * operations vector. 212 */ 213 vp->v_ops = &mp->mnt_vn_use_ops; 214 vp->v_pbuf_count = nswbuf_kva / NSWBUF_SPLIT; 215 216 /* 217 * Placing the vnode on the mount point's queue makes it visible. 218 * VNON prevents it from being messed with, however. 219 */ 220 insmntque(vp, mp); 221 222 /* 223 * A VX locked & refd vnode is returned. 224 */ 225 *vpp = vp; 226 return (0); 227 } 228 229 /* 230 * This function creates vnodes with special operations vectors. The 231 * mount point is optional. 232 * 233 * This routine is being phased out but is still used by vfs_conf to 234 * create vnodes for devices prior to the root mount (with mp == NULL). 235 */ 236 int 237 getspecialvnode(enum vtagtype tag, struct mount *mp, 238 struct vop_ops **ops, 239 struct vnode **vpp, int lktimeout, int lkflags) 240 { 241 struct vnode *vp; 242 243 vp = allocvnode(lktimeout, lkflags); 244 vp->v_tag = tag; 245 vp->v_data = NULL; 246 vp->v_ops = ops; 247 248 if (mp == NULL) 249 mp = &dummymount; 250 251 /* 252 * Placing the vnode on the mount point's queue makes it visible. 253 * VNON prevents it from being messed with, however. 254 */ 255 insmntque(vp, mp); 256 257 /* 258 * A VX locked & refd vnode is returned. 259 */ 260 *vpp = vp; 261 return (0); 262 } 263 264 /* 265 * Interlock against an unmount, return 0 on success, non-zero on failure. 266 * 267 * The passed flag may be 0 or LK_NOWAIT and is only used if an unmount 268 * is in-progress. 269 * 270 * If no unmount is in-progress LK_NOWAIT is ignored. No other flag bits 271 * are used. A shared locked will be obtained and the filesystem will not 272 * be unmountable until the lock is released. 273 */ 274 int 275 vfs_busy(struct mount *mp, int flags) 276 { 277 int lkflags; 278 279 atomic_add_int(&mp->mnt_refs, 1); 280 lwkt_gettoken(&mp->mnt_token); 281 if (mp->mnt_kern_flag & MNTK_UNMOUNT) { 282 if (flags & LK_NOWAIT) { 283 lwkt_reltoken(&mp->mnt_token); 284 atomic_add_int(&mp->mnt_refs, -1); 285 return (ENOENT); 286 } 287 /* XXX not MP safe */ 288 mp->mnt_kern_flag |= MNTK_MWAIT; 289 290 /* 291 * Since all busy locks are shared except the exclusive 292 * lock granted when unmounting, the only place that a 293 * wakeup needs to be done is at the release of the 294 * exclusive lock at the end of dounmount. 295 * 296 * WARNING! mp can potentially go away once we release 297 * our ref. 298 */ 299 tsleep((caddr_t)mp, 0, "vfs_busy", 0); 300 lwkt_reltoken(&mp->mnt_token); 301 atomic_add_int(&mp->mnt_refs, -1); 302 return (ENOENT); 303 } 304 lkflags = LK_SHARED; 305 if (lockmgr(&mp->mnt_lock, lkflags)) 306 panic("vfs_busy: unexpected lock failure"); 307 lwkt_reltoken(&mp->mnt_token); 308 return (0); 309 } 310 311 /* 312 * Free a busy filesystem. 313 * 314 * Once refs is decremented the mount point can potentially get ripped 315 * out from under us, but we want to clean up our refs before unlocking 316 * so do a hold/drop around the whole mess. 317 * 318 * This is not in the critical path (I hope). 319 */ 320 void 321 vfs_unbusy(struct mount *mp) 322 { 323 mount_hold(mp); 324 atomic_add_int(&mp->mnt_refs, -1); 325 lockmgr(&mp->mnt_lock, LK_RELEASE); 326 mount_drop(mp); 327 } 328 329 /* 330 * Lookup a filesystem type, and if found allocate and initialize 331 * a mount structure for it. 332 * 333 * Devname is usually updated by mount(8) after booting. 334 */ 335 int 336 vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp) 337 { 338 struct vfsconf *vfsp; 339 struct mount *mp; 340 341 if (fstypename == NULL) 342 return (ENODEV); 343 344 vfsp = vfsconf_find_by_name(fstypename); 345 if (vfsp == NULL) 346 return (ENODEV); 347 mp = kmalloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO); 348 mount_init(mp, vfsp->vfc_vfsops); 349 lockinit(&mp->mnt_lock, "vfslock", VLKTIMEOUT, 0); 350 351 vfs_busy(mp, 0); 352 mp->mnt_vfc = vfsp; 353 mp->mnt_pbuf_count = nswbuf_kva / NSWBUF_SPLIT; 354 vfsp->vfc_refcount++; 355 mp->mnt_stat.f_type = vfsp->vfc_typenum; 356 mp->mnt_flag |= MNT_RDONLY; 357 mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK; 358 strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN); 359 copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0); 360 361 /* 362 * Pre-set MPSAFE flags for VFS_MOUNT() call. 363 */ 364 if (vfsp->vfc_flags & VFCF_MPSAFE) 365 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; 366 367 *mpp = mp; 368 369 return (0); 370 } 371 372 /* 373 * Basic mount structure initialization 374 */ 375 void 376 mount_init(struct mount *mp, struct vfsops *ops) 377 { 378 lockinit(&mp->mnt_lock, "vfslock", hz*5, 0); 379 lwkt_token_init(&mp->mnt_token, "permnt"); 380 381 TAILQ_INIT(&mp->mnt_vnodescan_list); 382 TAILQ_INIT(&mp->mnt_nvnodelist); 383 TAILQ_INIT(&mp->mnt_reservedvnlist); 384 TAILQ_INIT(&mp->mnt_jlist); 385 mp->mnt_nvnodelistsize = 0; 386 mp->mnt_flag = 0; 387 mp->mnt_hold = 1; /* hold for umount last drop */ 388 mp->mnt_iosize_max = MAXPHYS; 389 mp->mnt_op = ops; 390 if (ops == NULL || (ops->vfs_flags & VFSOPSF_NOSYNCERTHR) == 0) 391 vn_syncer_thr_create(mp); 392 } 393 394 void 395 mount_hold(struct mount *mp) 396 { 397 atomic_add_int(&mp->mnt_hold, 1); 398 } 399 400 void 401 mount_drop(struct mount *mp) 402 { 403 if (atomic_fetchadd_int(&mp->mnt_hold, -1) == 1) { 404 KKASSERT(mp->mnt_refs == 0); 405 kfree(mp, M_MOUNT); 406 } 407 } 408 409 /* 410 * Lookup a mount point by filesystem identifier. 411 * 412 * If not NULL, the returned mp is held and the caller is expected to drop 413 * it via mount_drop(). 414 */ 415 struct mount * 416 vfs_getvfs(fsid_t *fsid) 417 { 418 struct mount *mp; 419 420 lwkt_gettoken_shared(&mountlist_token); 421 mp = mount_rb_tree_RB_LOOKUP_FSID(&mounttree, fsid); 422 if (mp) 423 mount_hold(mp); 424 lwkt_reltoken(&mountlist_token); 425 return (mp); 426 } 427 428 /* 429 * Generate a FSID based on the mountpt. The FSID will be adjusted to avoid 430 * collisions when the mount is added to mountlist. 431 * 432 * May only be called prior to the mount succeeding. 433 * 434 * OLD: 435 * 436 * Get a new unique fsid. Try to make its val[0] unique, since this value 437 * will be used to create fake device numbers for stat(). Also try (but 438 * not so hard) make its val[0] unique mod 2^16, since some emulators only 439 * support 16-bit device numbers. We end up with unique val[0]'s for the 440 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 441 */ 442 void 443 vfs_getnewfsid(struct mount *mp) 444 { 445 fsid_t tfsid; 446 int mtype; 447 int error; 448 char *retbuf; 449 char *freebuf; 450 451 mtype = mp->mnt_vfc->vfc_typenum; 452 tfsid.val[1] = mtype; 453 error = cache_fullpath(NULL, &mp->mnt_ncmounton, NULL, 454 &retbuf, &freebuf, 0); 455 if (error) { 456 tfsid.val[0] = makeudev(255, 0); 457 } else { 458 tfsid.val[0] = makeudev(255, 459 iscsi_crc32(retbuf, strlen(retbuf)) & 460 ~makeudev(255, 0)); 461 kfree(freebuf, M_TEMP); 462 } 463 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 464 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 465 } 466 467 /* 468 * Set the FSID for a new mount point to the template. 469 * 470 * The FSID will be adjusted to avoid collisions when the mount is 471 * added to mountlist. 472 * 473 * May only be called prior to the mount succeeding. 474 */ 475 void 476 vfs_setfsid(struct mount *mp, fsid_t *template) 477 { 478 bzero(&mp->mnt_stat.f_fsid, sizeof(mp->mnt_stat.f_fsid)); 479 480 #if 0 481 struct mount *mptmp; 482 483 lwkt_gettoken(&mntid_token); 484 for (;;) { 485 mptmp = vfs_getvfs(template); 486 if (mptmp == NULL) 487 break; 488 mount_drop(mptmp); 489 ++template->val[1]; 490 } 491 lwkt_reltoken(&mntid_token); 492 #endif 493 mp->mnt_stat.f_fsid = *template; 494 } 495 496 /* 497 * This routine is called when we have too many vnodes. It attempts 498 * to free <count> vnodes and will potentially free vnodes that still 499 * have VM backing store (VM backing store is typically the cause 500 * of a vnode blowout so we want to do this). Therefore, this operation 501 * is not considered cheap. 502 * 503 * A number of conditions may prevent a vnode from being reclaimed. 504 * the buffer cache may have references on the vnode, a directory 505 * vnode may still have references due to the namei cache representing 506 * underlying files, or the vnode may be in active use. It is not 507 * desireable to reuse such vnodes. These conditions may cause the 508 * number of vnodes to reach some minimum value regardless of what 509 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 510 */ 511 512 /* 513 * Attempt to recycle vnodes in a context that is always safe to block. 514 * Calling vlrurecycle() from the bowels of file system code has some 515 * interesting deadlock problems. 516 */ 517 static struct thread *vnlruthread; 518 519 static void 520 vnlru_proc(void) 521 { 522 struct thread *td = curthread; 523 524 EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td, 525 SHUTDOWN_PRI_FIRST); 526 527 for (;;) { 528 int ncachedandinactive; 529 530 kproc_suspend_loop(); 531 532 /* 533 * Try to free some vnodes if we have too many. Trigger based 534 * on potentially freeable vnodes but calculate the count 535 * based on total vnodes. 536 * 537 * (long) -> deal with 64 bit machines, intermediate overflow 538 */ 539 synchronizevnodecount(); 540 ncachedandinactive = countcachedandinactivevnodes(); 541 if (numvnodes >= maxvnodes * 9 / 10 && 542 ncachedandinactive >= maxvnodes * 5 / 10) { 543 int count = numvnodes - maxvnodes * 9 / 10; 544 545 if (count > (ncachedandinactive) / 100) 546 count = (ncachedandinactive) / 100; 547 if (count < 5) 548 count = 5; 549 freesomevnodes(count); 550 } 551 552 /* 553 * Do non-critical-path (more robust) cache cleaning, 554 * even if vnode counts are nominal, to try to avoid 555 * having to do it in the critical path. 556 */ 557 cache_hysteresis(0); 558 559 /* 560 * Nothing to do if most of our vnodes are already on 561 * the free list. 562 */ 563 synchronizevnodecount(); 564 ncachedandinactive = countcachedandinactivevnodes(); 565 if (numvnodes <= maxvnodes * 9 / 10 || 566 ncachedandinactive <= maxvnodes * 5 / 10) { 567 tsleep(vnlruthread, 0, "vlruwt", hz); 568 continue; 569 } 570 } 571 } 572 573 /* 574 * MOUNTLIST FUNCTIONS 575 */ 576 577 /* 578 * mountlist_insert (MP SAFE) 579 * 580 * Add a new mount point to the mount list. Filesystem should attempt to 581 * supply a unique fsid but if a duplicate occurs adjust the fsid to ensure 582 * uniqueness. 583 */ 584 void 585 mountlist_insert(struct mount *mp, int how) 586 { 587 int lim = 0x01000000; 588 589 lwkt_gettoken(&mountlist_token); 590 if (how == MNTINS_FIRST) 591 TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list); 592 else 593 TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list); 594 while (mount_rb_tree_RB_INSERT(&mounttree, mp)) { 595 int32_t val; 596 597 /* 598 * minor device mask: 0xFFFF00FF 599 */ 600 val = mp->mnt_stat.f_fsid.val[0]; 601 val = ((val & 0xFFFF0000) >> 8) | (val & 0x000000FF); 602 ++val; 603 val = ((val << 8) & 0xFFFF0000) | (val & 0x000000FF); 604 mp->mnt_stat.f_fsid.val[0] = val; 605 if (--lim == 0) { 606 lim = 0x01000000; 607 mp->mnt_stat.f_fsid.val[1] += 0x0100; 608 kprintf("mountlist_insert: fsid collision, " 609 "too many mounts\n"); 610 } 611 } 612 lwkt_reltoken(&mountlist_token); 613 } 614 615 /* 616 * mountlist_interlock (MP SAFE) 617 * 618 * Execute the specified interlock function with the mountlist token 619 * held. The function will be called in a serialized fashion verses 620 * other functions called through this mechanism. 621 * 622 * The function is expected to be very short-lived. 623 */ 624 int 625 mountlist_interlock(int (*callback)(struct mount *), struct mount *mp) 626 { 627 int error; 628 629 lwkt_gettoken(&mountlist_token); 630 error = callback(mp); 631 lwkt_reltoken(&mountlist_token); 632 return (error); 633 } 634 635 /* 636 * mountlist_boot_getfirst (DURING BOOT ONLY) 637 * 638 * This function returns the first mount on the mountlist, which is 639 * expected to be the root mount. Since no interlocks are obtained 640 * this function is only safe to use during booting. 641 */ 642 643 struct mount * 644 mountlist_boot_getfirst(void) 645 { 646 return(TAILQ_FIRST(&mountlist)); 647 } 648 649 /* 650 * mountlist_remove (MP SAFE) 651 * 652 * Remove a node from the mountlist. If this node is the next scan node 653 * for any active mountlist scans, the active mountlist scan will be 654 * adjusted to skip the node, thus allowing removals during mountlist 655 * scans. 656 */ 657 void 658 mountlist_remove(struct mount *mp) 659 { 660 struct mountscan_info *msi; 661 662 lwkt_gettoken(&mountlist_token); 663 TAILQ_FOREACH(msi, &mountscan_list, msi_entry) { 664 if (msi->msi_node == mp) { 665 if (msi->msi_how & MNTSCAN_FORWARD) 666 msi->msi_node = TAILQ_NEXT(mp, mnt_list); 667 else 668 msi->msi_node = TAILQ_PREV(mp, mntlist, 669 mnt_list); 670 } 671 } 672 TAILQ_REMOVE(&mountlist, mp, mnt_list); 673 mount_rb_tree_RB_REMOVE(&mounttree, mp); 674 lwkt_reltoken(&mountlist_token); 675 } 676 677 /* 678 * mountlist_exists (MP SAFE) 679 * 680 * Checks if a node exists in the mountlist. 681 * This function is mainly used by VFS quota code to check if a 682 * cached nullfs struct mount pointer is still valid at use time 683 * 684 * FIXME: there is no warranty the mp passed to that function 685 * will be the same one used by VFS_ACCOUNT() later 686 */ 687 int 688 mountlist_exists(struct mount *mp) 689 { 690 int node_exists = 0; 691 struct mount* lmp; 692 693 lwkt_gettoken_shared(&mountlist_token); 694 TAILQ_FOREACH(lmp, &mountlist, mnt_list) { 695 if (lmp == mp) { 696 node_exists = 1; 697 break; 698 } 699 } 700 lwkt_reltoken(&mountlist_token); 701 702 return(node_exists); 703 } 704 705 /* 706 * mountlist_scan 707 * 708 * Safely scan the mount points on the mount list. Each mountpoint 709 * is held across the callback. The callback is responsible for 710 * acquiring any further tokens or locks. 711 * 712 * Unless otherwise specified each mount point will be busied prior to the 713 * callback and unbusied afterwords. The callback may safely remove any 714 * mount point without interfering with the scan. If the current callback 715 * mount is removed the scanner will not attempt to unbusy it. 716 * 717 * If a mount node cannot be busied it is silently skipped. 718 * 719 * The callback return value is aggregated and a total is returned. A return 720 * value of < 0 is not aggregated and will terminate the scan. 721 * 722 * MNTSCAN_FORWARD - the mountlist is scanned in the forward direction 723 * MNTSCAN_REVERSE - the mountlist is scanned in reverse 724 * MNTSCAN_NOBUSY - the scanner will make the callback without busying 725 * the mount node. 726 * MNTSCAN_NOUNLOCK - Do not unlock mountlist_token across callback 727 * 728 * NOTE: mountlist_token is not held across the callback. 729 */ 730 int 731 mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how) 732 { 733 struct mountscan_info info; 734 struct mount *mp; 735 int count; 736 int res; 737 int dounlock = ((how & MNTSCAN_NOUNLOCK) == 0); 738 739 lwkt_gettoken(&mountlist_token); 740 info.msi_how = how; 741 info.msi_node = NULL; /* paranoia */ 742 TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry); 743 lwkt_reltoken(&mountlist_token); 744 745 res = 0; 746 lwkt_gettoken_shared(&mountlist_token); 747 748 if (how & MNTSCAN_FORWARD) { 749 info.msi_node = TAILQ_FIRST(&mountlist); 750 while ((mp = info.msi_node) != NULL) { 751 mount_hold(mp); 752 if (how & MNTSCAN_NOBUSY) { 753 if (dounlock) 754 lwkt_reltoken(&mountlist_token); 755 count = callback(mp, data); 756 if (dounlock) 757 lwkt_gettoken_shared(&mountlist_token); 758 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 759 if (dounlock) 760 lwkt_reltoken(&mountlist_token); 761 count = callback(mp, data); 762 if (dounlock) 763 lwkt_gettoken_shared(&mountlist_token); 764 if (mp == info.msi_node) 765 vfs_unbusy(mp); 766 } else { 767 count = 0; 768 } 769 mount_drop(mp); 770 if (count < 0) 771 break; 772 res += count; 773 if (mp == info.msi_node) 774 info.msi_node = TAILQ_NEXT(mp, mnt_list); 775 } 776 } else if (how & MNTSCAN_REVERSE) { 777 info.msi_node = TAILQ_LAST(&mountlist, mntlist); 778 while ((mp = info.msi_node) != NULL) { 779 mount_hold(mp); 780 if (how & MNTSCAN_NOBUSY) { 781 if (dounlock) 782 lwkt_reltoken(&mountlist_token); 783 count = callback(mp, data); 784 if (dounlock) 785 lwkt_gettoken_shared(&mountlist_token); 786 } else if (vfs_busy(mp, LK_NOWAIT) == 0) { 787 if (dounlock) 788 lwkt_reltoken(&mountlist_token); 789 count = callback(mp, data); 790 if (dounlock) 791 lwkt_gettoken_shared(&mountlist_token); 792 if (mp == info.msi_node) 793 vfs_unbusy(mp); 794 } else { 795 count = 0; 796 } 797 mount_drop(mp); 798 if (count < 0) 799 break; 800 res += count; 801 if (mp == info.msi_node) 802 info.msi_node = TAILQ_PREV(mp, mntlist, 803 mnt_list); 804 } 805 } 806 lwkt_reltoken(&mountlist_token); 807 808 lwkt_gettoken(&mountlist_token); 809 TAILQ_REMOVE(&mountscan_list, &info, msi_entry); 810 lwkt_reltoken(&mountlist_token); 811 812 return(res); 813 } 814 815 /* 816 * MOUNT RELATED VNODE FUNCTIONS 817 */ 818 819 static struct kproc_desc vnlru_kp = { 820 "vnlru", 821 vnlru_proc, 822 &vnlruthread 823 }; 824 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); 825 826 /* 827 * Move a vnode from one mount queue to another. 828 */ 829 void 830 insmntque(struct vnode *vp, struct mount *mp) 831 { 832 struct mount *omp; 833 834 /* 835 * Delete from old mount point vnode list, if on one. 836 */ 837 if ((omp = vp->v_mount) != NULL) { 838 lwkt_gettoken(&omp->mnt_token); 839 KKASSERT(omp == vp->v_mount); 840 KASSERT(omp->mnt_nvnodelistsize > 0, 841 ("bad mount point vnode list size")); 842 vremovevnodemnt(vp); 843 omp->mnt_nvnodelistsize--; 844 lwkt_reltoken(&omp->mnt_token); 845 } 846 847 /* 848 * Insert into list of vnodes for the new mount point, if available. 849 * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer. 850 */ 851 if (mp == NULL) { 852 vp->v_mount = NULL; 853 return; 854 } 855 lwkt_gettoken(&mp->mnt_token); 856 vp->v_mount = mp; 857 if (mp->mnt_syncer) { 858 TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes); 859 } else { 860 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 861 } 862 mp->mnt_nvnodelistsize++; 863 lwkt_reltoken(&mp->mnt_token); 864 } 865 866 867 /* 868 * Scan the vnodes under a mount point and issue appropriate callbacks. 869 * 870 * The fastfunc() callback is called with just the mountlist token held 871 * (no vnode lock). It may not block and the vnode may be undergoing 872 * modifications while the caller is processing it. The vnode will 873 * not be entirely destroyed, however, due to the fact that the mountlist 874 * token is held. A return value < 0 skips to the next vnode without calling 875 * the slowfunc(), a return value > 0 terminates the loop. 876 * 877 * WARNING! The fastfunc() should not indirect through vp->v_object, the vp 878 * data structure is unstable when called from fastfunc(). 879 * 880 * The slowfunc() callback is called after the vnode has been successfully 881 * locked based on passed flags. The vnode is skipped if it gets rearranged 882 * or destroyed while blocking on the lock. A non-zero return value from 883 * the slow function terminates the loop. The slow function is allowed to 884 * arbitrarily block. The scanning code guarentees consistency of operation 885 * even if the slow function deletes or moves the node, or blocks and some 886 * other thread deletes or moves the node. 887 */ 888 int 889 vmntvnodescan( 890 struct mount *mp, 891 int flags, 892 int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data), 893 int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data), 894 void *data 895 ) { 896 struct vmntvnodescan_info info; 897 struct vnode *vp; 898 int r = 0; 899 int maxcount = mp->mnt_nvnodelistsize * 2; 900 int stopcount = 0; 901 int count = 0; 902 903 lwkt_gettoken(&mp->mnt_token); 904 905 /* 906 * If asked to do one pass stop after iterating available vnodes. 907 * Under heavy loads new vnodes can be added while we are scanning, 908 * so this isn't perfect. Create a slop factor of 2x. 909 */ 910 if (flags & VMSC_ONEPASS) 911 stopcount = mp->mnt_nvnodelistsize; 912 913 info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 914 TAILQ_INSERT_TAIL(&mp->mnt_vnodescan_list, &info, entry); 915 916 while ((vp = info.vp) != NULL) { 917 if (--maxcount == 0) { 918 kprintf("Warning: excessive fssync iteration\n"); 919 maxcount = mp->mnt_nvnodelistsize * 2; 920 } 921 922 /* 923 * Skip if visible but not ready, or special (e.g. 924 * mp->mnt_syncer) 925 */ 926 if (vp->v_type == VNON) 927 goto next; 928 KKASSERT(vp->v_mount == mp); 929 930 /* 931 * Quick test. A negative return continues the loop without 932 * calling the slow test. 0 continues onto the slow test. 933 * A positive number aborts the loop. 934 */ 935 if (fastfunc) { 936 if ((r = fastfunc(mp, vp, data)) < 0) { 937 r = 0; 938 goto next; 939 } 940 if (r) 941 break; 942 } 943 944 /* 945 * Get a vxlock on the vnode, retry if it has moved or isn't 946 * in the mountlist where we expect it. 947 */ 948 if (slowfunc) { 949 int error; 950 951 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 952 case VMSC_GETVP: 953 error = vget(vp, LK_EXCLUSIVE); 954 break; 955 case VMSC_GETVP|VMSC_NOWAIT: 956 error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT); 957 break; 958 case VMSC_GETVX: 959 vx_get(vp); 960 error = 0; 961 break; 962 default: 963 error = 0; 964 break; 965 } 966 if (error) 967 goto next; 968 /* 969 * Do not call the slow function if the vnode is 970 * invalid or if it was ripped out from under us 971 * while we (potentially) blocked. 972 */ 973 if (info.vp == vp && vp->v_type != VNON) 974 r = slowfunc(mp, vp, data); 975 976 /* 977 * Cleanup 978 */ 979 switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) { 980 case VMSC_GETVP: 981 case VMSC_GETVP|VMSC_NOWAIT: 982 vput(vp); 983 break; 984 case VMSC_GETVX: 985 vx_put(vp); 986 break; 987 default: 988 break; 989 } 990 if (r != 0) 991 break; 992 } 993 994 next: 995 /* 996 * Yield after some processing. Depending on the number 997 * of vnodes, we might wind up running for a long time. 998 * Because threads are not preemptable, time critical 999 * userland processes might starve. Give them a chance 1000 * now and then. 1001 */ 1002 if (++count == 10000) { 1003 /* 1004 * We really want to yield a bit, so we simply 1005 * sleep a tick 1006 */ 1007 tsleep(mp, 0, "vnodescn", 1); 1008 count = 0; 1009 } 1010 1011 /* 1012 * If doing one pass this decrements to zero. If it starts 1013 * at zero it is effectively unlimited for the purposes of 1014 * this loop. 1015 */ 1016 if (--stopcount == 0) 1017 break; 1018 1019 /* 1020 * Iterate. If the vnode was ripped out from under us 1021 * info.vp will already point to the next vnode, otherwise 1022 * we have to obtain the next valid vnode ourselves. 1023 */ 1024 if (info.vp == vp) 1025 info.vp = TAILQ_NEXT(vp, v_nmntvnodes); 1026 } 1027 1028 TAILQ_REMOVE(&mp->mnt_vnodescan_list, &info, entry); 1029 lwkt_reltoken(&mp->mnt_token); 1030 return(r); 1031 } 1032 1033 /* 1034 * Remove any vnodes in the vnode table belonging to mount point mp. 1035 * 1036 * If FORCECLOSE is not specified, there should not be any active ones, 1037 * return error if any are found (nb: this is a user error, not a 1038 * system error). If FORCECLOSE is specified, detach any active vnodes 1039 * that are found. 1040 * 1041 * If WRITECLOSE is set, only flush out regular file vnodes open for 1042 * writing. 1043 * 1044 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped. 1045 * 1046 * `rootrefs' specifies the base reference count for the root vnode 1047 * of this filesystem. The root vnode is considered busy if its 1048 * v_refcnt exceeds this value. On a successful return, vflush() 1049 * will call vrele() on the root vnode exactly rootrefs times. 1050 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 1051 * be zero. 1052 */ 1053 static int debug_busyprt = 0; /* print out busy vnodes */ 1054 SYSCTL_INT(_vfs, OID_AUTO, debug_busyprt, CTLFLAG_RW, &debug_busyprt, 0, ""); 1055 1056 static int vflush_scan(struct mount *mp, struct vnode *vp, void *data); 1057 1058 struct vflush_info { 1059 int flags; 1060 int busy; 1061 thread_t td; 1062 }; 1063 1064 int 1065 vflush(struct mount *mp, int rootrefs, int flags) 1066 { 1067 struct thread *td = curthread; /* XXX */ 1068 struct vnode *rootvp = NULL; 1069 int error; 1070 struct vflush_info vflush_info; 1071 1072 if (rootrefs > 0) { 1073 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 1074 ("vflush: bad args")); 1075 /* 1076 * Get the filesystem root vnode. We can vput() it 1077 * immediately, since with rootrefs > 0, it won't go away. 1078 */ 1079 if ((error = VFS_ROOT(mp, &rootvp)) != 0) { 1080 if ((flags & FORCECLOSE) == 0) 1081 return (error); 1082 rootrefs = 0; 1083 /* continue anyway */ 1084 } 1085 if (rootrefs) 1086 vput(rootvp); 1087 } 1088 1089 vflush_info.busy = 0; 1090 vflush_info.flags = flags; 1091 vflush_info.td = td; 1092 vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info); 1093 1094 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 1095 /* 1096 * If just the root vnode is busy, and if its refcount 1097 * is equal to `rootrefs', then go ahead and kill it. 1098 */ 1099 KASSERT(vflush_info.busy > 0, ("vflush: not busy")); 1100 KASSERT(VREFCNT(rootvp) >= rootrefs, ("vflush: rootrefs")); 1101 if (vflush_info.busy == 1 && VREFCNT(rootvp) == rootrefs) { 1102 vx_lock(rootvp); 1103 vgone_vxlocked(rootvp); 1104 vx_unlock(rootvp); 1105 vflush_info.busy = 0; 1106 } 1107 } 1108 if (vflush_info.busy) 1109 return (EBUSY); 1110 for (; rootrefs > 0; rootrefs--) 1111 vrele(rootvp); 1112 return (0); 1113 } 1114 1115 /* 1116 * The scan callback is made with an VX locked vnode. 1117 */ 1118 static int 1119 vflush_scan(struct mount *mp, struct vnode *vp, void *data) 1120 { 1121 struct vflush_info *info = data; 1122 struct vattr vattr; 1123 int flags = info->flags; 1124 1125 /* 1126 * Generally speaking try to deactivate on 0 refs (catch-all) 1127 */ 1128 atomic_set_int(&vp->v_refcnt, VREF_FINALIZE); 1129 1130 /* 1131 * Skip over a vnodes marked VSYSTEM. 1132 */ 1133 if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) { 1134 return(0); 1135 } 1136 1137 /* 1138 * Do not force-close VCHR or VBLK vnodes 1139 */ 1140 if (vp->v_type == VCHR || vp->v_type == VBLK) 1141 flags &= ~(WRITECLOSE|FORCECLOSE); 1142 1143 /* 1144 * If WRITECLOSE is set, flush out unlinked but still open 1145 * files (even if open only for reading) and regular file 1146 * vnodes open for writing. 1147 */ 1148 if ((flags & WRITECLOSE) && 1149 (vp->v_type == VNON || 1150 (VOP_GETATTR(vp, &vattr) == 0 && 1151 vattr.va_nlink > 0)) && 1152 (vp->v_writecount == 0 || vp->v_type != VREG)) { 1153 return(0); 1154 } 1155 1156 /* 1157 * If we are the only holder (refcnt of 1) or the vnode is in 1158 * termination (refcnt < 0), we can vgone the vnode. 1159 */ 1160 if (VREFCNT(vp) <= 1) { 1161 vgone_vxlocked(vp); 1162 return(0); 1163 } 1164 1165 /* 1166 * If FORCECLOSE is set, forcibly destroy the vnode and then move 1167 * it to a dummymount structure so vop_*() functions don't deref 1168 * a NULL pointer. 1169 */ 1170 if (flags & FORCECLOSE) { 1171 vhold(vp); 1172 vgone_vxlocked(vp); 1173 if (vp->v_mount == NULL) 1174 insmntque(vp, &dummymount); 1175 vdrop(vp); 1176 return(0); 1177 } 1178 if (vp->v_type == VCHR || vp->v_type == VBLK) 1179 kprintf("vflush: Warning, cannot destroy busy device vnode\n"); 1180 if (debug_busyprt) { 1181 const char *filename; 1182 1183 spin_lock(&vp->v_spin); 1184 filename = TAILQ_FIRST(&vp->v_namecache) ? 1185 TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"; 1186 spin_unlock(&vp->v_spin); 1187 kprintf("vflush: busy vnode (%p) %s\n", vp, filename); 1188 } 1189 ++info->busy; 1190 return(0); 1191 } 1192 1193 void 1194 add_bio_ops(struct bio_ops *ops) 1195 { 1196 TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry); 1197 } 1198 1199 void 1200 rem_bio_ops(struct bio_ops *ops) 1201 { 1202 TAILQ_REMOVE(&bio_ops_list, ops, entry); 1203 } 1204 1205 /* 1206 * This calls the bio_ops io_sync function either for a mount point 1207 * or generally. 1208 * 1209 * WARNING: softdeps is weirdly coded and just isn't happy unless 1210 * io_sync is called with a NULL mount from the general syncing code. 1211 */ 1212 void 1213 bio_ops_sync(struct mount *mp) 1214 { 1215 struct bio_ops *ops; 1216 1217 if (mp) { 1218 if ((ops = mp->mnt_bioops) != NULL) 1219 ops->io_sync(mp); 1220 } else { 1221 TAILQ_FOREACH(ops, &bio_ops_list, entry) { 1222 ops->io_sync(NULL); 1223 } 1224 } 1225 } 1226 1227 /* 1228 * Lookup a mount point by nch 1229 */ 1230 struct mount * 1231 mount_get_by_nc(struct namecache *ncp) 1232 { 1233 struct mount *mp = NULL; 1234 1235 lwkt_gettoken_shared(&mountlist_token); 1236 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 1237 if (ncp == mp->mnt_ncmountpt.ncp) 1238 break; 1239 } 1240 lwkt_reltoken(&mountlist_token); 1241 1242 return (mp); 1243 } 1244 1245