1 /* $NetBSD: vfs_subr.c,v 1.484 2020/03/14 20:45:23 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 1997, 1998, 2004, 2005, 2007, 2008, 2019, 2020 5 * The NetBSD Foundation, Inc. 6 * All rights reserved. 7 * 8 * This code is derived from software contributed to The NetBSD Foundation 9 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 10 * NASA Ames Research Center, by Charles M. Hannum, by Andrew Doran, 11 * by Marshall Kirk McKusick and Greg Ganger at the University of Michigan. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 24 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 25 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 26 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 27 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 28 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 29 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 30 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 31 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 32 * POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 /* 36 * Copyright (c) 1989, 1993 37 * The Regents of the University of California. All rights reserved. 38 * (c) UNIX System Laboratories, Inc. 39 * All or some portions of this file are derived from material licensed 40 * to the University of California by American Telephone and Telegraph 41 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 42 * the permission of UNIX System Laboratories, Inc. 43 * 44 * Redistribution and use in source and binary forms, with or without 45 * modification, are permitted provided that the following conditions 46 * are met: 47 * 1. Redistributions of source code must retain the above copyright 48 * notice, this list of conditions and the following disclaimer. 49 * 2. Redistributions in binary form must reproduce the above copyright 50 * notice, this list of conditions and the following disclaimer in the 51 * documentation and/or other materials provided with the distribution. 52 * 3. Neither the name of the University nor the names of its contributors 53 * may be used to endorse or promote products derived from this software 54 * without specific prior written permission. 55 * 56 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 57 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 58 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 59 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 60 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 61 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 62 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 63 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 64 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 65 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 66 * SUCH DAMAGE. 67 * 68 * @(#)vfs_subr.c 8.13 (Berkeley) 4/18/94 69 */ 70 71 #include <sys/cdefs.h> 72 __KERNEL_RCSID(0, "$NetBSD: vfs_subr.c,v 1.484 2020/03/14 20:45:23 ad Exp $"); 73 74 #ifdef _KERNEL_OPT 75 #include "opt_ddb.h" 76 #include "opt_compat_netbsd.h" 77 #include "opt_compat_43.h" 78 #endif 79 80 #include <sys/param.h> 81 #include <sys/systm.h> 82 #include <sys/conf.h> 83 #include <sys/dirent.h> 84 #include <sys/filedesc.h> 85 #include <sys/kernel.h> 86 #include <sys/mount.h> 87 #include <sys/fstrans.h> 88 #include <sys/vnode_impl.h> 89 #include <sys/stat.h> 90 #include <sys/sysctl.h> 91 #include <sys/namei.h> 92 #include <sys/buf.h> 93 #include <sys/errno.h> 94 #include <sys/kmem.h> 95 #include <sys/syscallargs.h> 96 #include <sys/kauth.h> 97 #include <sys/module.h> 98 99 #include <miscfs/genfs/genfs.h> 100 #include <miscfs/specfs/specdev.h> 101 #include <uvm/uvm_ddb.h> 102 103 const enum vtype iftovt_tab[16] = { 104 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 105 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 106 }; 107 const int vttoif_tab[9] = { 108 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 109 S_IFSOCK, S_IFIFO, S_IFMT, 110 }; 111 112 /* 113 * Insq/Remq for the vnode usage lists. 114 */ 115 #define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs) 116 #define bufremvn(bp) { \ 117 LIST_REMOVE(bp, b_vnbufs); \ 118 (bp)->b_vnbufs.le_next = NOLIST; \ 119 } 120 121 int doforce = 1; /* 1 => permit forcible unmounting */ 122 123 extern struct mount *dead_rootmount; 124 125 /* 126 * Local declarations. 127 */ 128 129 static void vn_initialize_syncerd(void); 130 131 /* 132 * Initialize the vnode management data structures. 133 */ 134 void 135 vntblinit(void) 136 { 137 138 vn_initialize_syncerd(); 139 vfs_mount_sysinit(); 140 vfs_vnode_sysinit(); 141 } 142 143 /* 144 * Flush out and invalidate all buffers associated with a vnode. 145 * Called with the underlying vnode locked, which should prevent new dirty 146 * buffers from being queued. 147 */ 148 int 149 vinvalbuf(struct vnode *vp, int flags, kauth_cred_t cred, struct lwp *l, 150 bool catch_p, int slptimeo) 151 { 152 struct buf *bp, *nbp; 153 int error; 154 int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO | 155 (flags & V_SAVE ? PGO_CLEANIT | PGO_RECLAIM : 0); 156 157 /* XXXUBC this doesn't look at flags or slp* */ 158 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 159 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 160 if (error) { 161 return error; 162 } 163 164 if (flags & V_SAVE) { 165 error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0); 166 if (error) 167 return (error); 168 KASSERT(LIST_EMPTY(&vp->v_dirtyblkhd)); 169 } 170 171 mutex_enter(&bufcache_lock); 172 restart: 173 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 174 KASSERT(bp->b_vp == vp); 175 nbp = LIST_NEXT(bp, b_vnbufs); 176 error = bbusy(bp, catch_p, slptimeo, NULL); 177 if (error != 0) { 178 if (error == EPASSTHROUGH) 179 goto restart; 180 mutex_exit(&bufcache_lock); 181 return (error); 182 } 183 brelsel(bp, BC_INVAL | BC_VFLUSH); 184 } 185 186 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 187 KASSERT(bp->b_vp == vp); 188 nbp = LIST_NEXT(bp, b_vnbufs); 189 error = bbusy(bp, catch_p, slptimeo, NULL); 190 if (error != 0) { 191 if (error == EPASSTHROUGH) 192 goto restart; 193 mutex_exit(&bufcache_lock); 194 return (error); 195 } 196 /* 197 * XXX Since there are no node locks for NFS, I believe 198 * there is a slight chance that a delayed write will 199 * occur while sleeping just above, so check for it. 200 */ 201 if ((bp->b_oflags & BO_DELWRI) && (flags & V_SAVE)) { 202 #ifdef DEBUG 203 printf("buffer still DELWRI\n"); 204 #endif 205 bp->b_cflags |= BC_BUSY | BC_VFLUSH; 206 mutex_exit(&bufcache_lock); 207 VOP_BWRITE(bp->b_vp, bp); 208 mutex_enter(&bufcache_lock); 209 goto restart; 210 } 211 brelsel(bp, BC_INVAL | BC_VFLUSH); 212 } 213 214 #ifdef DIAGNOSTIC 215 if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd)) 216 panic("vinvalbuf: flush failed, vp %p", vp); 217 #endif 218 219 mutex_exit(&bufcache_lock); 220 221 return (0); 222 } 223 224 /* 225 * Destroy any in core blocks past the truncation length. 226 * Called with the underlying vnode locked, which should prevent new dirty 227 * buffers from being queued. 228 */ 229 int 230 vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch_p, int slptimeo) 231 { 232 struct buf *bp, *nbp; 233 int error; 234 voff_t off; 235 236 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 237 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 238 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 239 if (error) { 240 return error; 241 } 242 243 mutex_enter(&bufcache_lock); 244 restart: 245 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 246 KASSERT(bp->b_vp == vp); 247 nbp = LIST_NEXT(bp, b_vnbufs); 248 if (bp->b_lblkno < lbn) 249 continue; 250 error = bbusy(bp, catch_p, slptimeo, NULL); 251 if (error != 0) { 252 if (error == EPASSTHROUGH) 253 goto restart; 254 mutex_exit(&bufcache_lock); 255 return (error); 256 } 257 brelsel(bp, BC_INVAL | BC_VFLUSH); 258 } 259 260 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 261 KASSERT(bp->b_vp == vp); 262 nbp = LIST_NEXT(bp, b_vnbufs); 263 if (bp->b_lblkno < lbn) 264 continue; 265 error = bbusy(bp, catch_p, slptimeo, NULL); 266 if (error != 0) { 267 if (error == EPASSTHROUGH) 268 goto restart; 269 mutex_exit(&bufcache_lock); 270 return (error); 271 } 272 brelsel(bp, BC_INVAL | BC_VFLUSH); 273 } 274 mutex_exit(&bufcache_lock); 275 276 return (0); 277 } 278 279 /* 280 * Flush all dirty buffers from a vnode. 281 * Called with the underlying vnode locked, which should prevent new dirty 282 * buffers from being queued. 283 */ 284 int 285 vflushbuf(struct vnode *vp, int flags) 286 { 287 struct buf *bp, *nbp; 288 int error, pflags; 289 bool dirty, sync; 290 291 sync = (flags & FSYNC_WAIT) != 0; 292 pflags = PGO_CLEANIT | PGO_ALLPAGES | 293 (sync ? PGO_SYNCIO : 0) | 294 ((flags & FSYNC_LAZY) ? PGO_LAZY : 0); 295 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 296 (void) VOP_PUTPAGES(vp, 0, 0, pflags); 297 298 loop: 299 mutex_enter(&bufcache_lock); 300 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 301 KASSERT(bp->b_vp == vp); 302 nbp = LIST_NEXT(bp, b_vnbufs); 303 if ((bp->b_cflags & BC_BUSY)) 304 continue; 305 if ((bp->b_oflags & BO_DELWRI) == 0) 306 panic("vflushbuf: not dirty, bp %p", bp); 307 bp->b_cflags |= BC_BUSY | BC_VFLUSH; 308 mutex_exit(&bufcache_lock); 309 /* 310 * Wait for I/O associated with indirect blocks to complete, 311 * since there is no way to quickly wait for them below. 312 */ 313 if (bp->b_vp == vp || !sync) 314 (void) bawrite(bp); 315 else { 316 error = bwrite(bp); 317 if (error) 318 return error; 319 } 320 goto loop; 321 } 322 mutex_exit(&bufcache_lock); 323 324 if (!sync) 325 return 0; 326 327 mutex_enter(vp->v_interlock); 328 while (vp->v_numoutput != 0) 329 cv_wait(&vp->v_cv, vp->v_interlock); 330 dirty = !LIST_EMPTY(&vp->v_dirtyblkhd); 331 mutex_exit(vp->v_interlock); 332 333 if (dirty) { 334 vprint("vflushbuf: dirty", vp); 335 goto loop; 336 } 337 338 return 0; 339 } 340 341 /* 342 * Create a vnode for a block device. 343 * Used for root filesystem and swap areas. 344 * Also used for memory file system special devices. 345 */ 346 int 347 bdevvp(dev_t dev, vnode_t **vpp) 348 { 349 struct vattr va; 350 351 vattr_null(&va); 352 va.va_type = VBLK; 353 va.va_rdev = dev; 354 355 return vcache_new(dead_rootmount, NULL, &va, NOCRED, NULL, vpp); 356 } 357 358 /* 359 * Create a vnode for a character device. 360 * Used for kernfs and some console handling. 361 */ 362 int 363 cdevvp(dev_t dev, vnode_t **vpp) 364 { 365 struct vattr va; 366 367 vattr_null(&va); 368 va.va_type = VCHR; 369 va.va_rdev = dev; 370 371 return vcache_new(dead_rootmount, NULL, &va, NOCRED, NULL, vpp); 372 } 373 374 /* 375 * Associate a buffer with a vnode. There must already be a hold on 376 * the vnode. 377 */ 378 void 379 bgetvp(struct vnode *vp, struct buf *bp) 380 { 381 382 KASSERT(bp->b_vp == NULL); 383 KASSERT(bp->b_objlock == &buffer_lock); 384 KASSERT(mutex_owned(vp->v_interlock)); 385 KASSERT(mutex_owned(&bufcache_lock)); 386 KASSERT((bp->b_cflags & BC_BUSY) != 0); 387 KASSERT(!cv_has_waiters(&bp->b_done)); 388 389 vholdl(vp); 390 bp->b_vp = vp; 391 if (vp->v_type == VBLK || vp->v_type == VCHR) 392 bp->b_dev = vp->v_rdev; 393 else 394 bp->b_dev = NODEV; 395 396 /* 397 * Insert onto list for new vnode. 398 */ 399 bufinsvn(bp, &vp->v_cleanblkhd); 400 bp->b_objlock = vp->v_interlock; 401 } 402 403 /* 404 * Disassociate a buffer from a vnode. 405 */ 406 void 407 brelvp(struct buf *bp) 408 { 409 struct vnode *vp = bp->b_vp; 410 411 KASSERT(vp != NULL); 412 KASSERT(bp->b_objlock == vp->v_interlock); 413 KASSERT(mutex_owned(vp->v_interlock)); 414 KASSERT(mutex_owned(&bufcache_lock)); 415 KASSERT((bp->b_cflags & BC_BUSY) != 0); 416 KASSERT(!cv_has_waiters(&bp->b_done)); 417 418 /* 419 * Delete from old vnode list, if on one. 420 */ 421 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 422 bufremvn(bp); 423 424 if ((vp->v_iflag & (VI_ONWORKLST | VI_PAGES)) == VI_ONWORKLST && 425 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) 426 vn_syncer_remove_from_worklist(vp); 427 428 bp->b_objlock = &buffer_lock; 429 bp->b_vp = NULL; 430 holdrelel(vp); 431 } 432 433 /* 434 * Reassign a buffer from one vnode list to another. 435 * The list reassignment must be within the same vnode. 436 * Used to assign file specific control information 437 * (indirect blocks) to the list to which they belong. 438 */ 439 void 440 reassignbuf(struct buf *bp, struct vnode *vp) 441 { 442 struct buflists *listheadp; 443 int delayx; 444 445 KASSERT(mutex_owned(&bufcache_lock)); 446 KASSERT(bp->b_objlock == vp->v_interlock); 447 KASSERT(mutex_owned(vp->v_interlock)); 448 KASSERT((bp->b_cflags & BC_BUSY) != 0); 449 450 /* 451 * Delete from old vnode list, if on one. 452 */ 453 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 454 bufremvn(bp); 455 456 /* 457 * If dirty, put on list of dirty buffers; 458 * otherwise insert onto list of clean buffers. 459 */ 460 if ((bp->b_oflags & BO_DELWRI) == 0) { 461 listheadp = &vp->v_cleanblkhd; 462 if ((vp->v_iflag & (VI_ONWORKLST | VI_PAGES)) == 463 VI_ONWORKLST && 464 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) 465 vn_syncer_remove_from_worklist(vp); 466 } else { 467 listheadp = &vp->v_dirtyblkhd; 468 if ((vp->v_iflag & VI_ONWORKLST) == 0) { 469 switch (vp->v_type) { 470 case VDIR: 471 delayx = dirdelay; 472 break; 473 case VBLK: 474 if (spec_node_getmountedfs(vp) != NULL) { 475 delayx = metadelay; 476 break; 477 } 478 /* fall through */ 479 default: 480 delayx = filedelay; 481 break; 482 } 483 if (!vp->v_mount || 484 (vp->v_mount->mnt_flag & MNT_ASYNC) == 0) 485 vn_syncer_add_to_worklist(vp, delayx); 486 } 487 } 488 bufinsvn(bp, listheadp); 489 } 490 491 /* 492 * Lookup a vnode by device number and return it referenced. 493 */ 494 int 495 vfinddev(dev_t dev, enum vtype type, vnode_t **vpp) 496 { 497 498 return (spec_node_lookup_by_dev(type, dev, vpp) == 0); 499 } 500 501 /* 502 * Revoke all the vnodes corresponding to the specified minor number 503 * range (endpoints inclusive) of the specified major. 504 */ 505 void 506 vdevgone(int maj, int minl, int minh, enum vtype type) 507 { 508 vnode_t *vp; 509 dev_t dev; 510 int mn; 511 512 for (mn = minl; mn <= minh; mn++) { 513 dev = makedev(maj, mn); 514 while (spec_node_lookup_by_dev(type, dev, &vp) == 0) { 515 VOP_REVOKE(vp, REVOKEALL); 516 vrele(vp); 517 } 518 } 519 } 520 521 /* 522 * The filesystem synchronizer mechanism - syncer. 523 * 524 * It is useful to delay writes of file data and filesystem metadata for 525 * a certain amount of time so that quickly created and deleted files need 526 * not waste disk bandwidth being created and removed. To implement this, 527 * vnodes are appended to a "workitem" queue. 528 * 529 * Most pending metadata should not wait for more than ten seconds. Thus, 530 * mounted on block devices are delayed only about a half the time that file 531 * data is delayed. Similarly, directory updates are more critical, so are 532 * only delayed about a third the time that file data is delayed. 533 * 534 * There are SYNCER_MAXDELAY queues that are processed in a round-robin 535 * manner at a rate of one each second (driven off the filesystem syner 536 * thread). The syncer_delayno variable indicates the next queue that is 537 * to be processed. Items that need to be processed soon are placed in 538 * this queue: 539 * 540 * syncer_workitem_pending[syncer_delayno] 541 * 542 * A delay of e.g. fifteen seconds is done by placing the request fifteen 543 * entries later in the queue: 544 * 545 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 546 * 547 * Flag VI_ONWORKLST indicates that vnode is added into the queue. 548 */ 549 550 #define SYNCER_MAXDELAY 32 551 552 typedef TAILQ_HEAD(synclist, vnode_impl) synclist_t; 553 554 static void vn_syncer_add1(struct vnode *, int); 555 static void sysctl_vfs_syncfs_setup(struct sysctllog **); 556 557 /* 558 * Defines and variables for the syncer process. 559 */ 560 int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 561 time_t syncdelay = 30; /* max time to delay syncing data */ 562 time_t filedelay = 30; /* time to delay syncing files */ 563 time_t dirdelay = 15; /* time to delay syncing directories */ 564 time_t metadelay = 10; /* time to delay syncing metadata */ 565 time_t lockdelay = 1; /* time to delay if locking fails */ 566 567 static kmutex_t syncer_data_lock; /* short term lock on data structs */ 568 569 static int syncer_delayno = 0; 570 static long syncer_last; 571 static synclist_t * syncer_workitem_pending; 572 573 static void 574 vn_initialize_syncerd(void) 575 { 576 int i; 577 578 syncer_last = SYNCER_MAXDELAY + 2; 579 580 sysctl_vfs_syncfs_setup(NULL); 581 582 syncer_workitem_pending = 583 kmem_alloc(syncer_last * sizeof (struct synclist), KM_SLEEP); 584 585 for (i = 0; i < syncer_last; i++) 586 TAILQ_INIT(&syncer_workitem_pending[i]); 587 588 mutex_init(&syncer_data_lock, MUTEX_DEFAULT, IPL_NONE); 589 } 590 591 /* 592 * Return delay factor appropriate for the given file system. For 593 * WAPBL we use the sync vnode to burst out metadata updates: sync 594 * those file systems more frequently. 595 */ 596 static inline int 597 sync_delay(struct mount *mp) 598 { 599 600 return mp->mnt_wapbl != NULL ? metadelay : syncdelay; 601 } 602 603 /* 604 * Compute the next slot index from delay. 605 */ 606 static inline int 607 sync_delay_slot(int delayx) 608 { 609 610 if (delayx > syncer_maxdelay - 2) 611 delayx = syncer_maxdelay - 2; 612 return (syncer_delayno + delayx) % syncer_last; 613 } 614 615 /* 616 * Add an item to the syncer work queue. 617 */ 618 static void 619 vn_syncer_add1(struct vnode *vp, int delayx) 620 { 621 synclist_t *slp; 622 vnode_impl_t *vip = VNODE_TO_VIMPL(vp); 623 624 KASSERT(mutex_owned(&syncer_data_lock)); 625 626 if (vp->v_iflag & VI_ONWORKLST) { 627 /* 628 * Remove in order to adjust the position of the vnode. 629 * Note: called from sched_sync(), which will not hold 630 * interlock, therefore we cannot modify v_iflag here. 631 */ 632 slp = &syncer_workitem_pending[vip->vi_synclist_slot]; 633 TAILQ_REMOVE(slp, vip, vi_synclist); 634 } else { 635 KASSERT(mutex_owned(vp->v_interlock)); 636 vp->v_iflag |= VI_ONWORKLST; 637 } 638 639 vip->vi_synclist_slot = sync_delay_slot(delayx); 640 641 slp = &syncer_workitem_pending[vip->vi_synclist_slot]; 642 TAILQ_INSERT_TAIL(slp, vip, vi_synclist); 643 } 644 645 void 646 vn_syncer_add_to_worklist(struct vnode *vp, int delayx) 647 { 648 649 KASSERT(mutex_owned(vp->v_interlock)); 650 651 mutex_enter(&syncer_data_lock); 652 vn_syncer_add1(vp, delayx); 653 mutex_exit(&syncer_data_lock); 654 } 655 656 /* 657 * Remove an item from the syncer work queue. 658 */ 659 void 660 vn_syncer_remove_from_worklist(struct vnode *vp) 661 { 662 synclist_t *slp; 663 vnode_impl_t *vip = VNODE_TO_VIMPL(vp); 664 665 KASSERT(mutex_owned(vp->v_interlock)); 666 667 if (vp->v_iflag & VI_ONWORKLST) { 668 mutex_enter(&syncer_data_lock); 669 vp->v_iflag &= ~VI_ONWORKLST; 670 slp = &syncer_workitem_pending[vip->vi_synclist_slot]; 671 TAILQ_REMOVE(slp, vip, vi_synclist); 672 mutex_exit(&syncer_data_lock); 673 } 674 } 675 676 /* 677 * Add this mount point to the syncer. 678 */ 679 void 680 vfs_syncer_add_to_worklist(struct mount *mp) 681 { 682 static int start, incr, next; 683 int vdelay; 684 685 KASSERT(mutex_owned(mp->mnt_updating)); 686 KASSERT((mp->mnt_iflag & IMNT_ONWORKLIST) == 0); 687 688 /* 689 * We attempt to scatter the mount points on the list 690 * so that they will go off at evenly distributed times 691 * even if all the filesystems are mounted at once. 692 */ 693 694 next += incr; 695 if (next == 0 || next > syncer_maxdelay) { 696 start /= 2; 697 incr /= 2; 698 if (start == 0) { 699 start = syncer_maxdelay / 2; 700 incr = syncer_maxdelay; 701 } 702 next = start; 703 } 704 mp->mnt_iflag |= IMNT_ONWORKLIST; 705 vdelay = sync_delay(mp); 706 mp->mnt_synclist_slot = vdelay > 0 ? next % vdelay : 0; 707 } 708 709 /* 710 * Remove the mount point from the syncer. 711 */ 712 void 713 vfs_syncer_remove_from_worklist(struct mount *mp) 714 { 715 716 KASSERT(mutex_owned(mp->mnt_updating)); 717 KASSERT((mp->mnt_iflag & IMNT_ONWORKLIST) != 0); 718 719 mp->mnt_iflag &= ~IMNT_ONWORKLIST; 720 } 721 722 /* 723 * Try lazy sync, return true on success. 724 */ 725 static bool 726 lazy_sync_vnode(struct vnode *vp) 727 { 728 bool synced; 729 730 KASSERT(mutex_owned(&syncer_data_lock)); 731 732 synced = false; 733 /* We are locking in the wrong direction. */ 734 if (mutex_tryenter(vp->v_interlock)) { 735 mutex_exit(&syncer_data_lock); 736 if (vcache_tryvget(vp) == 0) { 737 if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT) == 0) { 738 synced = true; 739 (void) VOP_FSYNC(vp, curlwp->l_cred, 740 FSYNC_LAZY, 0, 0); 741 vput(vp); 742 } else 743 vrele(vp); 744 } 745 mutex_enter(&syncer_data_lock); 746 } 747 return synced; 748 } 749 750 /* 751 * System filesystem synchronizer daemon. 752 */ 753 void 754 sched_sync(void *arg) 755 { 756 mount_iterator_t *iter; 757 synclist_t *slp; 758 struct vnode_impl *vi; 759 struct vnode *vp; 760 struct mount *mp; 761 time_t starttime; 762 bool synced; 763 764 for (;;) { 765 starttime = time_second; 766 767 /* 768 * Sync mounts whose dirty time has expired. 769 */ 770 mountlist_iterator_init(&iter); 771 while ((mp = mountlist_iterator_trynext(iter)) != NULL) { 772 if ((mp->mnt_iflag & IMNT_ONWORKLIST) == 0 || 773 mp->mnt_synclist_slot != syncer_delayno) { 774 continue; 775 } 776 mp->mnt_synclist_slot = sync_delay_slot(sync_delay(mp)); 777 VFS_SYNC(mp, MNT_LAZY, curlwp->l_cred); 778 } 779 mountlist_iterator_destroy(iter); 780 781 mutex_enter(&syncer_data_lock); 782 783 /* 784 * Push files whose dirty time has expired. 785 */ 786 slp = &syncer_workitem_pending[syncer_delayno]; 787 syncer_delayno += 1; 788 if (syncer_delayno >= syncer_last) 789 syncer_delayno = 0; 790 791 while ((vi = TAILQ_FIRST(slp)) != NULL) { 792 vp = VIMPL_TO_VNODE(vi); 793 synced = lazy_sync_vnode(vp); 794 795 /* 796 * XXX The vnode may have been recycled, in which 797 * case it may have a new identity. 798 */ 799 vi = TAILQ_FIRST(slp); 800 if (vi != NULL && VIMPL_TO_VNODE(vi) == vp) { 801 /* 802 * Put us back on the worklist. The worklist 803 * routine will remove us from our current 804 * position and then add us back in at a later 805 * position. 806 * 807 * Try again sooner rather than later if 808 * we were unable to lock the vnode. Lock 809 * failure should not prevent us from doing 810 * the sync "soon". 811 * 812 * If we locked it yet arrive here, it's 813 * likely that lazy sync is in progress and 814 * so the vnode still has dirty metadata. 815 * syncdelay is mainly to get this vnode out 816 * of the way so we do not consider it again 817 * "soon" in this loop, so the delay time is 818 * not critical as long as it is not "soon". 819 * While write-back strategy is the file 820 * system's domain, we expect write-back to 821 * occur no later than syncdelay seconds 822 * into the future. 823 */ 824 vn_syncer_add1(vp, 825 synced ? syncdelay : lockdelay); 826 } 827 } 828 829 /* 830 * If it has taken us less than a second to process the 831 * current work, then wait. Otherwise start right over 832 * again. We can still lose time if any single round 833 * takes more than two seconds, but it does not really 834 * matter as we are just trying to generally pace the 835 * filesystem activity. 836 */ 837 if (time_second == starttime) { 838 kpause("syncer", false, hz, &syncer_data_lock); 839 } 840 mutex_exit(&syncer_data_lock); 841 } 842 } 843 844 static void 845 sysctl_vfs_syncfs_setup(struct sysctllog **clog) 846 { 847 const struct sysctlnode *rnode, *cnode; 848 849 sysctl_createv(clog, 0, NULL, &rnode, 850 CTLFLAG_PERMANENT, 851 CTLTYPE_NODE, "sync", 852 SYSCTL_DESCR("syncer options"), 853 NULL, 0, NULL, 0, 854 CTL_VFS, CTL_CREATE, CTL_EOL); 855 856 sysctl_createv(clog, 0, &rnode, &cnode, 857 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 858 CTLTYPE_QUAD, "delay", 859 SYSCTL_DESCR("max time to delay syncing data"), 860 NULL, 0, &syncdelay, 0, 861 CTL_CREATE, CTL_EOL); 862 863 sysctl_createv(clog, 0, &rnode, &cnode, 864 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 865 CTLTYPE_QUAD, "filedelay", 866 SYSCTL_DESCR("time to delay syncing files"), 867 NULL, 0, &filedelay, 0, 868 CTL_CREATE, CTL_EOL); 869 870 sysctl_createv(clog, 0, &rnode, &cnode, 871 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 872 CTLTYPE_QUAD, "dirdelay", 873 SYSCTL_DESCR("time to delay syncing directories"), 874 NULL, 0, &dirdelay, 0, 875 CTL_CREATE, CTL_EOL); 876 877 sysctl_createv(clog, 0, &rnode, &cnode, 878 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 879 CTLTYPE_QUAD, "metadelay", 880 SYSCTL_DESCR("time to delay syncing metadata"), 881 NULL, 0, &metadelay, 0, 882 CTL_CREATE, CTL_EOL); 883 } 884 885 /* 886 * sysctl helper routine to return list of supported fstypes 887 */ 888 int 889 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS) 890 { 891 char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)]; 892 char *where = oldp; 893 struct vfsops *v; 894 size_t needed, left, slen; 895 int error, first; 896 897 if (newp != NULL) 898 return (EPERM); 899 if (namelen != 0) 900 return (EINVAL); 901 902 first = 1; 903 error = 0; 904 needed = 0; 905 left = *oldlenp; 906 907 sysctl_unlock(); 908 mutex_enter(&vfs_list_lock); 909 LIST_FOREACH(v, &vfs_list, vfs_list) { 910 if (where == NULL) 911 needed += strlen(v->vfs_name) + 1; 912 else { 913 memset(bf, 0, sizeof(bf)); 914 if (first) { 915 strncpy(bf, v->vfs_name, sizeof(bf)); 916 first = 0; 917 } else { 918 bf[0] = ' '; 919 strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1); 920 } 921 bf[sizeof(bf)-1] = '\0'; 922 slen = strlen(bf); 923 if (left < slen + 1) 924 break; 925 v->vfs_refcount++; 926 mutex_exit(&vfs_list_lock); 927 /* +1 to copy out the trailing NUL byte */ 928 error = copyout(bf, where, slen + 1); 929 mutex_enter(&vfs_list_lock); 930 v->vfs_refcount--; 931 if (error) 932 break; 933 where += slen; 934 needed += slen; 935 left -= slen; 936 } 937 } 938 mutex_exit(&vfs_list_lock); 939 sysctl_relock(); 940 *oldlenp = needed; 941 return (error); 942 } 943 944 int kinfo_vdebug = 1; 945 int kinfo_vgetfailed; 946 947 #define KINFO_VNODESLOP 10 948 949 /* 950 * Dump vnode list (via sysctl). 951 * Copyout address of vnode followed by vnode. 952 */ 953 int 954 sysctl_kern_vnode(SYSCTLFN_ARGS) 955 { 956 char *where = oldp; 957 size_t *sizep = oldlenp; 958 struct mount *mp; 959 vnode_t *vp, vbuf; 960 mount_iterator_t *iter; 961 struct vnode_iterator *marker; 962 char *bp = where; 963 char *ewhere; 964 int error; 965 966 if (namelen != 0) 967 return (EOPNOTSUPP); 968 if (newp != NULL) 969 return (EPERM); 970 971 #define VPTRSZ sizeof(vnode_t *) 972 #define VNODESZ sizeof(vnode_t) 973 if (where == NULL) { 974 *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ); 975 return (0); 976 } 977 ewhere = where + *sizep; 978 979 sysctl_unlock(); 980 mountlist_iterator_init(&iter); 981 while ((mp = mountlist_iterator_next(iter)) != NULL) { 982 vfs_vnode_iterator_init(mp, &marker); 983 while ((vp = vfs_vnode_iterator_next(marker, NULL, NULL))) { 984 if (bp + VPTRSZ + VNODESZ > ewhere) { 985 vrele(vp); 986 vfs_vnode_iterator_destroy(marker); 987 mountlist_iterator_destroy(iter); 988 sysctl_relock(); 989 *sizep = bp - where; 990 return (ENOMEM); 991 } 992 memcpy(&vbuf, vp, VNODESZ); 993 if ((error = copyout(&vp, bp, VPTRSZ)) || 994 (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) { 995 vrele(vp); 996 vfs_vnode_iterator_destroy(marker); 997 mountlist_iterator_destroy(iter); 998 sysctl_relock(); 999 return (error); 1000 } 1001 vrele(vp); 1002 bp += VPTRSZ + VNODESZ; 1003 } 1004 vfs_vnode_iterator_destroy(marker); 1005 } 1006 mountlist_iterator_destroy(iter); 1007 sysctl_relock(); 1008 1009 *sizep = bp - where; 1010 return (0); 1011 } 1012 1013 /* 1014 * Set vnode attributes to VNOVAL 1015 */ 1016 void 1017 vattr_null(struct vattr *vap) 1018 { 1019 1020 memset(vap, 0, sizeof(*vap)); 1021 1022 vap->va_type = VNON; 1023 1024 /* 1025 * Assign individually so that it is safe even if size and 1026 * sign of each member are varied. 1027 */ 1028 vap->va_mode = VNOVAL; 1029 vap->va_nlink = VNOVAL; 1030 vap->va_uid = VNOVAL; 1031 vap->va_gid = VNOVAL; 1032 vap->va_fsid = VNOVAL; 1033 vap->va_fileid = VNOVAL; 1034 vap->va_size = VNOVAL; 1035 vap->va_blocksize = VNOVAL; 1036 vap->va_atime.tv_sec = 1037 vap->va_mtime.tv_sec = 1038 vap->va_ctime.tv_sec = 1039 vap->va_birthtime.tv_sec = VNOVAL; 1040 vap->va_atime.tv_nsec = 1041 vap->va_mtime.tv_nsec = 1042 vap->va_ctime.tv_nsec = 1043 vap->va_birthtime.tv_nsec = VNOVAL; 1044 vap->va_gen = VNOVAL; 1045 vap->va_flags = VNOVAL; 1046 vap->va_rdev = VNOVAL; 1047 vap->va_bytes = VNOVAL; 1048 } 1049 1050 /* 1051 * Vnode state to string. 1052 */ 1053 const char * 1054 vstate_name(enum vnode_state state) 1055 { 1056 1057 switch (state) { 1058 case VS_ACTIVE: 1059 return "ACTIVE"; 1060 case VS_MARKER: 1061 return "MARKER"; 1062 case VS_LOADING: 1063 return "LOADING"; 1064 case VS_LOADED: 1065 return "LOADED"; 1066 case VS_BLOCKED: 1067 return "BLOCKED"; 1068 case VS_RECLAIMING: 1069 return "RECLAIMING"; 1070 case VS_RECLAIMED: 1071 return "RECLAIMED"; 1072 default: 1073 return "ILLEGAL"; 1074 } 1075 } 1076 1077 /* 1078 * Print a description of a vnode (common part). 1079 */ 1080 static void 1081 vprint_common(struct vnode *vp, const char *prefix, 1082 void (*pr)(const char *, ...) __printflike(1, 2)) 1083 { 1084 int n; 1085 char bf[96]; 1086 const uint8_t *cp; 1087 vnode_impl_t *vip; 1088 const char * const vnode_tags[] = { VNODE_TAGS }; 1089 const char * const vnode_types[] = { VNODE_TYPES }; 1090 const char vnode_flagbits[] = VNODE_FLAGBITS; 1091 1092 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0])) 1093 #define ARRAY_PRINT(idx, arr) \ 1094 ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN") 1095 1096 vip = VNODE_TO_VIMPL(vp); 1097 1098 snprintb(bf, sizeof(bf), 1099 vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag); 1100 1101 (*pr)("vnode %p flags %s\n", vp, bf); 1102 (*pr)("%stag %s(%d) type %s(%d) mount %p typedata %p\n", prefix, 1103 ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag, 1104 ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type, 1105 vp->v_mount, vp->v_mountedhere); 1106 (*pr)("%susecount %d writecount %d holdcount %d\n", prefix, 1107 vp->v_usecount, vp->v_writecount, vp->v_holdcnt); 1108 (*pr)("%ssize %" PRIx64 " writesize %" PRIx64 " numoutput %d\n", 1109 prefix, vp->v_size, vp->v_writesize, vp->v_numoutput); 1110 (*pr)("%sdata %p lock %p\n", prefix, vp->v_data, &vip->vi_lock); 1111 1112 (*pr)("%sstate %s key(%p %zd)", prefix, vstate_name(vip->vi_state), 1113 vip->vi_key.vk_mount, vip->vi_key.vk_key_len); 1114 n = vip->vi_key.vk_key_len; 1115 cp = vip->vi_key.vk_key; 1116 while (n-- > 0) 1117 (*pr)(" %02x", *cp++); 1118 (*pr)("\n"); 1119 (*pr)("%slrulisthd %p\n", prefix, vip->vi_lrulisthd); 1120 1121 #undef ARRAY_PRINT 1122 #undef ARRAY_SIZE 1123 } 1124 1125 /* 1126 * Print out a description of a vnode. 1127 */ 1128 void 1129 vprint(const char *label, struct vnode *vp) 1130 { 1131 1132 if (label != NULL) 1133 printf("%s: ", label); 1134 vprint_common(vp, "\t", printf); 1135 if (vp->v_data != NULL) { 1136 printf("\t"); 1137 VOP_PRINT(vp); 1138 } 1139 } 1140 1141 /* Deprecated. Kept for KPI compatibility. */ 1142 int 1143 vaccess(enum vtype type, mode_t file_mode, uid_t uid, gid_t gid, 1144 mode_t acc_mode, kauth_cred_t cred) 1145 { 1146 1147 #ifdef DIAGNOSTIC 1148 printf("vaccess: deprecated interface used.\n"); 1149 #endif /* DIAGNOSTIC */ 1150 1151 return kauth_authorize_vnode(cred, KAUTH_ACCESS_ACTION(acc_mode, 1152 type, file_mode), NULL /* This may panic. */, NULL, 1153 genfs_can_access(type, file_mode, uid, gid, acc_mode, cred)); 1154 } 1155 1156 /* 1157 * Given a file system name, look up the vfsops for that 1158 * file system, or return NULL if file system isn't present 1159 * in the kernel. 1160 */ 1161 struct vfsops * 1162 vfs_getopsbyname(const char *name) 1163 { 1164 struct vfsops *v; 1165 1166 mutex_enter(&vfs_list_lock); 1167 LIST_FOREACH(v, &vfs_list, vfs_list) { 1168 if (strcmp(v->vfs_name, name) == 0) 1169 break; 1170 } 1171 if (v != NULL) 1172 v->vfs_refcount++; 1173 mutex_exit(&vfs_list_lock); 1174 1175 return (v); 1176 } 1177 1178 void 1179 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp) 1180 { 1181 const struct statvfs *mbp; 1182 1183 if (sbp == (mbp = &mp->mnt_stat)) 1184 return; 1185 1186 (void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx)); 1187 sbp->f_fsid = mbp->f_fsid; 1188 sbp->f_owner = mbp->f_owner; 1189 sbp->f_flag = mbp->f_flag; 1190 sbp->f_syncwrites = mbp->f_syncwrites; 1191 sbp->f_asyncwrites = mbp->f_asyncwrites; 1192 sbp->f_syncreads = mbp->f_syncreads; 1193 sbp->f_asyncreads = mbp->f_asyncreads; 1194 (void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare)); 1195 (void)memcpy(sbp->f_fstypename, mbp->f_fstypename, 1196 sizeof(sbp->f_fstypename)); 1197 (void)memcpy(sbp->f_mntonname, mbp->f_mntonname, 1198 sizeof(sbp->f_mntonname)); 1199 (void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, 1200 sizeof(sbp->f_mntfromname)); 1201 (void)memcpy(sbp->f_mntfromlabel, mp->mnt_stat.f_mntfromlabel, 1202 sizeof(sbp->f_mntfromlabel)); 1203 sbp->f_namemax = mbp->f_namemax; 1204 } 1205 1206 int 1207 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom, 1208 const char *vfsname, struct mount *mp, struct lwp *l) 1209 { 1210 int error; 1211 size_t size; 1212 struct statvfs *sfs = &mp->mnt_stat; 1213 int (*fun)(const void *, void *, size_t, size_t *); 1214 struct vnode *rvp; 1215 1216 (void)strlcpy(mp->mnt_stat.f_fstypename, vfsname, 1217 sizeof(mp->mnt_stat.f_fstypename)); 1218 1219 if (onp) { 1220 fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr; 1221 KASSERT(l == curlwp); 1222 rvp = cwdrdir(); 1223 if (rvp != NULL) { 1224 size_t len; 1225 char *bp; 1226 char *path = PNBUF_GET(); 1227 1228 bp = path + MAXPATHLEN; 1229 *--bp = '\0'; 1230 error = getcwd_common(rvp, rootvnode, &bp, 1231 path, MAXPATHLEN / 2, 0, l); 1232 vrele(rvp); 1233 if (error) { 1234 PNBUF_PUT(path); 1235 return error; 1236 } 1237 1238 len = strlen(bp); 1239 if (len > sizeof(sfs->f_mntonname) - 1) 1240 len = sizeof(sfs->f_mntonname) - 1; 1241 (void)strncpy(sfs->f_mntonname, bp, len); 1242 PNBUF_PUT(path); 1243 1244 if (len < sizeof(sfs->f_mntonname) - 1) { 1245 error = (*fun)(onp, &sfs->f_mntonname[len], 1246 sizeof(sfs->f_mntonname) - len - 1, &size); 1247 if (error) 1248 return error; 1249 size += len; 1250 } else { 1251 size = len; 1252 } 1253 } else { 1254 error = (*fun)(onp, &sfs->f_mntonname, 1255 sizeof(sfs->f_mntonname) - 1, &size); 1256 if (error) 1257 return error; 1258 } 1259 (void)memset(sfs->f_mntonname + size, 0, 1260 sizeof(sfs->f_mntonname) - size); 1261 } 1262 1263 if (fromp) { 1264 fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr; 1265 error = (*fun)(fromp, sfs->f_mntfromname, 1266 sizeof(sfs->f_mntfromname) - 1, &size); 1267 if (error) 1268 return error; 1269 (void)memset(sfs->f_mntfromname + size, 0, 1270 sizeof(sfs->f_mntfromname) - size); 1271 } 1272 return 0; 1273 } 1274 1275 void 1276 vfs_timestamp(struct timespec *ts) 1277 { 1278 1279 nanotime(ts); 1280 } 1281 1282 time_t rootfstime; /* recorded root fs time, if known */ 1283 void 1284 setrootfstime(time_t t) 1285 { 1286 rootfstime = t; 1287 } 1288 1289 static const uint8_t vttodt_tab[ ] = { 1290 [VNON] = DT_UNKNOWN, 1291 [VREG] = DT_REG, 1292 [VDIR] = DT_DIR, 1293 [VBLK] = DT_BLK, 1294 [VCHR] = DT_CHR, 1295 [VLNK] = DT_LNK, 1296 [VSOCK] = DT_SOCK, 1297 [VFIFO] = DT_FIFO, 1298 [VBAD] = DT_UNKNOWN 1299 }; 1300 1301 uint8_t 1302 vtype2dt(enum vtype vt) 1303 { 1304 1305 CTASSERT(VBAD == __arraycount(vttodt_tab) - 1); 1306 return vttodt_tab[vt]; 1307 } 1308 1309 int 1310 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c) 1311 { 1312 int error; 1313 1314 KERNEL_LOCK(1, NULL); 1315 error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c); 1316 KERNEL_UNLOCK_ONE(NULL); 1317 1318 return error; 1319 } 1320 1321 int 1322 VFS_START(struct mount *mp, int a) 1323 { 1324 int error; 1325 1326 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1327 KERNEL_LOCK(1, NULL); 1328 } 1329 error = (*(mp->mnt_op->vfs_start))(mp, a); 1330 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1331 KERNEL_UNLOCK_ONE(NULL); 1332 } 1333 1334 return error; 1335 } 1336 1337 int 1338 VFS_UNMOUNT(struct mount *mp, int a) 1339 { 1340 int error; 1341 1342 KERNEL_LOCK(1, NULL); 1343 error = (*(mp->mnt_op->vfs_unmount))(mp, a); 1344 KERNEL_UNLOCK_ONE(NULL); 1345 1346 return error; 1347 } 1348 1349 int 1350 VFS_ROOT(struct mount *mp, int lktype, struct vnode **a) 1351 { 1352 int error; 1353 1354 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1355 KERNEL_LOCK(1, NULL); 1356 } 1357 error = (*(mp->mnt_op->vfs_root))(mp, lktype, a); 1358 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1359 KERNEL_UNLOCK_ONE(NULL); 1360 } 1361 1362 return error; 1363 } 1364 1365 int 1366 VFS_QUOTACTL(struct mount *mp, struct quotactl_args *args) 1367 { 1368 int error; 1369 1370 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1371 KERNEL_LOCK(1, NULL); 1372 } 1373 error = (*(mp->mnt_op->vfs_quotactl))(mp, args); 1374 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1375 KERNEL_UNLOCK_ONE(NULL); 1376 } 1377 1378 return error; 1379 } 1380 1381 int 1382 VFS_STATVFS(struct mount *mp, struct statvfs *a) 1383 { 1384 int error; 1385 1386 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1387 KERNEL_LOCK(1, NULL); 1388 } 1389 error = (*(mp->mnt_op->vfs_statvfs))(mp, a); 1390 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1391 KERNEL_UNLOCK_ONE(NULL); 1392 } 1393 1394 return error; 1395 } 1396 1397 int 1398 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b) 1399 { 1400 int error; 1401 1402 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1403 KERNEL_LOCK(1, NULL); 1404 } 1405 error = (*(mp->mnt_op->vfs_sync))(mp, a, b); 1406 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1407 KERNEL_UNLOCK_ONE(NULL); 1408 } 1409 1410 return error; 1411 } 1412 1413 int 1414 VFS_FHTOVP(struct mount *mp, struct fid *a, int b, struct vnode **c) 1415 { 1416 int error; 1417 1418 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1419 KERNEL_LOCK(1, NULL); 1420 } 1421 error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b, c); 1422 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1423 KERNEL_UNLOCK_ONE(NULL); 1424 } 1425 1426 return error; 1427 } 1428 1429 int 1430 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b) 1431 { 1432 int error; 1433 1434 if ((vp->v_vflag & VV_MPSAFE) == 0) { 1435 KERNEL_LOCK(1, NULL); 1436 } 1437 error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b); 1438 if ((vp->v_vflag & VV_MPSAFE) == 0) { 1439 KERNEL_UNLOCK_ONE(NULL); 1440 } 1441 1442 return error; 1443 } 1444 1445 int 1446 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b) 1447 { 1448 int error; 1449 1450 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1451 KERNEL_LOCK(1, NULL); 1452 } 1453 error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b); 1454 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1455 KERNEL_UNLOCK_ONE(NULL); 1456 } 1457 1458 return error; 1459 } 1460 1461 int 1462 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d) 1463 { 1464 int error; 1465 1466 KERNEL_LOCK(1, NULL); /* XXXSMP check ffs */ 1467 error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d); 1468 KERNEL_UNLOCK_ONE(NULL); /* XXX */ 1469 1470 return error; 1471 } 1472 1473 int 1474 VFS_SUSPENDCTL(struct mount *mp, int a) 1475 { 1476 int error; 1477 1478 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1479 KERNEL_LOCK(1, NULL); 1480 } 1481 error = (*(mp->mnt_op->vfs_suspendctl))(mp, a); 1482 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1483 KERNEL_UNLOCK_ONE(NULL); 1484 } 1485 1486 return error; 1487 } 1488 1489 #if defined(DDB) || defined(DEBUGPRINT) 1490 static const char buf_flagbits[] = BUF_FLAGBITS; 1491 1492 void 1493 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...)) 1494 { 1495 char bf[1024]; 1496 1497 (*pr)(" vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%" 1498 PRIx64 " dev 0x%x\n", 1499 bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev); 1500 1501 snprintb(bf, sizeof(bf), 1502 buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags); 1503 (*pr)(" error %d flags %s\n", bp->b_error, bf); 1504 1505 (*pr)(" bufsize 0x%lx bcount 0x%lx resid 0x%lx\n", 1506 bp->b_bufsize, bp->b_bcount, bp->b_resid); 1507 (*pr)(" data %p saveaddr %p\n", 1508 bp->b_data, bp->b_saveaddr); 1509 (*pr)(" iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock); 1510 } 1511 1512 void 1513 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...)) 1514 { 1515 1516 uvm_object_printit(&vp->v_uobj, full, pr); 1517 (*pr)("\n"); 1518 vprint_common(vp, "", printf); 1519 if (full) { 1520 struct buf *bp; 1521 1522 (*pr)("clean bufs:\n"); 1523 LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) { 1524 (*pr)(" bp %p\n", bp); 1525 vfs_buf_print(bp, full, pr); 1526 } 1527 1528 (*pr)("dirty bufs:\n"); 1529 LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) { 1530 (*pr)(" bp %p\n", bp); 1531 vfs_buf_print(bp, full, pr); 1532 } 1533 } 1534 } 1535 1536 void 1537 vfs_vnode_lock_print(void *vlock, int full, void (*pr)(const char *, ...)) 1538 { 1539 struct mount *mp; 1540 vnode_impl_t *vip; 1541 1542 for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp)) { 1543 TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) { 1544 if (&vip->vi_lock == vlock || 1545 VIMPL_TO_VNODE(vip)->v_interlock == vlock) 1546 vfs_vnode_print(VIMPL_TO_VNODE(vip), full, pr); 1547 } 1548 } 1549 } 1550 1551 void 1552 vfs_mount_print_all(int full, void (*pr)(const char *, ...)) 1553 { 1554 struct mount *mp; 1555 for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp)) 1556 vfs_mount_print(mp, full, pr); 1557 } 1558 1559 void 1560 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...)) 1561 { 1562 char sbuf[256]; 1563 1564 (*pr)("vnodecovered = %p data = %p\n", 1565 mp->mnt_vnodecovered,mp->mnt_data); 1566 1567 (*pr)("fs_bshift %d dev_bshift = %d\n", 1568 mp->mnt_fs_bshift,mp->mnt_dev_bshift); 1569 1570 snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag); 1571 (*pr)("flag = %s\n", sbuf); 1572 1573 snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag); 1574 (*pr)("iflag = %s\n", sbuf); 1575 1576 (*pr)("refcnt = %d updating @ %p\n", mp->mnt_refcnt, mp->mnt_updating); 1577 1578 (*pr)("statvfs cache:\n"); 1579 (*pr)("\tbsize = %lu\n",mp->mnt_stat.f_bsize); 1580 (*pr)("\tfrsize = %lu\n",mp->mnt_stat.f_frsize); 1581 (*pr)("\tiosize = %lu\n",mp->mnt_stat.f_iosize); 1582 1583 (*pr)("\tblocks = %"PRIu64"\n",mp->mnt_stat.f_blocks); 1584 (*pr)("\tbfree = %"PRIu64"\n",mp->mnt_stat.f_bfree); 1585 (*pr)("\tbavail = %"PRIu64"\n",mp->mnt_stat.f_bavail); 1586 (*pr)("\tbresvd = %"PRIu64"\n",mp->mnt_stat.f_bresvd); 1587 1588 (*pr)("\tfiles = %"PRIu64"\n",mp->mnt_stat.f_files); 1589 (*pr)("\tffree = %"PRIu64"\n",mp->mnt_stat.f_ffree); 1590 (*pr)("\tfavail = %"PRIu64"\n",mp->mnt_stat.f_favail); 1591 (*pr)("\tfresvd = %"PRIu64"\n",mp->mnt_stat.f_fresvd); 1592 1593 (*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n", 1594 mp->mnt_stat.f_fsidx.__fsid_val[0], 1595 mp->mnt_stat.f_fsidx.__fsid_val[1]); 1596 1597 (*pr)("\towner = %"PRIu32"\n",mp->mnt_stat.f_owner); 1598 (*pr)("\tnamemax = %lu\n",mp->mnt_stat.f_namemax); 1599 1600 snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag); 1601 1602 (*pr)("\tflag = %s\n",sbuf); 1603 (*pr)("\tsyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_syncwrites); 1604 (*pr)("\tasyncwrites = %" PRIu64 "\n",mp->mnt_stat.f_asyncwrites); 1605 (*pr)("\tsyncreads = %" PRIu64 "\n",mp->mnt_stat.f_syncreads); 1606 (*pr)("\tasyncreads = %" PRIu64 "\n",mp->mnt_stat.f_asyncreads); 1607 (*pr)("\tfstypename = %s\n",mp->mnt_stat.f_fstypename); 1608 (*pr)("\tmntonname = %s\n",mp->mnt_stat.f_mntonname); 1609 (*pr)("\tmntfromname = %s\n",mp->mnt_stat.f_mntfromname); 1610 1611 { 1612 int cnt = 0; 1613 vnode_t *vp; 1614 vnode_impl_t *vip; 1615 (*pr)("locked vnodes ="); 1616 TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) { 1617 vp = VIMPL_TO_VNODE(vip); 1618 if (VOP_ISLOCKED(vp)) { 1619 if ((++cnt % 6) == 0) { 1620 (*pr)(" %p,\n\t", vp); 1621 } else { 1622 (*pr)(" %p,", vp); 1623 } 1624 } 1625 } 1626 (*pr)("\n"); 1627 } 1628 1629 if (full) { 1630 int cnt = 0; 1631 vnode_t *vp; 1632 vnode_impl_t *vip; 1633 (*pr)("all vnodes ="); 1634 TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) { 1635 vp = VIMPL_TO_VNODE(vip); 1636 if (!TAILQ_NEXT(vip, vi_mntvnodes)) { 1637 (*pr)(" %p", vp); 1638 } else if ((++cnt % 6) == 0) { 1639 (*pr)(" %p,\n\t", vp); 1640 } else { 1641 (*pr)(" %p,", vp); 1642 } 1643 } 1644 (*pr)("\n"); 1645 } 1646 } 1647 1648 /* 1649 * List all of the locked vnodes in the system. 1650 */ 1651 void printlockedvnodes(void); 1652 1653 void 1654 printlockedvnodes(void) 1655 { 1656 struct mount *mp; 1657 vnode_t *vp; 1658 vnode_impl_t *vip; 1659 1660 printf("Locked vnodes\n"); 1661 for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp)) { 1662 TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) { 1663 vp = VIMPL_TO_VNODE(vip); 1664 if (VOP_ISLOCKED(vp)) 1665 vprint(NULL, vp); 1666 } 1667 } 1668 } 1669 1670 #endif /* DDB || DEBUGPRINT */ 1671