1 /* $NetBSD: vfs_subr.c,v 1.502 2024/12/07 02:27:38 riastradh 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 1.502 2024/12/07 02:27:38 riastradh Exp $"); 73 74 #ifdef _KERNEL_OPT 75 #include "opt_compat_43.h" 76 #include "opt_compat_netbsd.h" 77 #include "opt_ddb.h" 78 #endif 79 80 #include <sys/param.h> 81 #include <sys/types.h> 82 83 #include <sys/buf.h> 84 #include <sys/conf.h> 85 #include <sys/dirent.h> 86 #include <sys/errno.h> 87 #include <sys/filedesc.h> 88 #include <sys/fstrans.h> 89 #include <sys/kauth.h> 90 #include <sys/kernel.h> 91 #include <sys/kmem.h> 92 #include <sys/module.h> 93 #include <sys/mount.h> 94 #include <sys/namei.h> 95 #include <sys/sdt.h> 96 #include <sys/stat.h> 97 #include <sys/syscallargs.h> 98 #include <sys/sysctl.h> 99 #include <sys/systm.h> 100 #include <sys/vnode_impl.h> 101 102 #include <miscfs/deadfs/deadfs.h> 103 #include <miscfs/genfs/genfs.h> 104 #include <miscfs/specfs/specdev.h> 105 106 #include <uvm/uvm_ddb.h> 107 108 SDT_PROBE_DEFINE3(vfs, syncer, worklist, vnode__add, 109 "struct vnode *"/*vp*/, 110 "int"/*delayx*/, 111 "int"/*slot*/); 112 SDT_PROBE_DEFINE4(vfs, syncer, worklist, vnode__update, 113 "struct vnode *"/*vp*/, 114 "int"/*delayx*/, 115 "int"/*oslot*/, 116 "int"/*nslot*/); 117 SDT_PROBE_DEFINE1(vfs, syncer, worklist, vnode__remove, 118 "struct vnode *"/*vp*/); 119 120 SDT_PROBE_DEFINE3(vfs, syncer, worklist, mount__add, 121 "struct mount *"/*mp*/, 122 "int"/*vdelay*/, 123 "int"/*slot*/); 124 SDT_PROBE_DEFINE4(vfs, syncer, worklist, mount__update, 125 "struct mount *"/*vp*/, 126 "int"/*vdelay*/, 127 "int"/*oslot*/, 128 "int"/*nslot*/); 129 SDT_PROBE_DEFINE1(vfs, syncer, worklist, mount__remove, 130 "struct mount *"/*mp*/); 131 132 SDT_PROBE_DEFINE1(vfs, syncer, sync, start, 133 "int"/*starttime*/); 134 SDT_PROBE_DEFINE1(vfs, syncer, sync, mount__start, 135 "struct mount *"/*mp*/); 136 SDT_PROBE_DEFINE2(vfs, syncer, sync, mount__done, 137 "struct mount *"/*mp*/, 138 "int"/*error*/); 139 SDT_PROBE_DEFINE1(vfs, syncer, sync, mount__skip, 140 "struct mount *"/*mp*/); 141 SDT_PROBE_DEFINE1(vfs, syncer, sync, vnode__start, 142 "struct vnode *"/*vp*/); 143 SDT_PROBE_DEFINE2(vfs, syncer, sync, vnode__done, 144 "struct vnode *"/*vp*/, 145 "int"/*error*/); 146 SDT_PROBE_DEFINE2(vfs, syncer, sync, vnode__fail__lock, 147 "struct vnode *"/*vp*/, 148 "int"/*error*/); 149 SDT_PROBE_DEFINE2(vfs, syncer, sync, vnode__fail__vget, 150 "struct vnode *"/*vp*/, 151 "int"/*error*/); 152 SDT_PROBE_DEFINE2(vfs, syncer, sync, done, 153 "int"/*starttime*/, 154 "int"/*endtime*/); 155 156 const enum vtype iftovt_tab[16] = { 157 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 158 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 159 }; 160 const int vttoif_tab[9] = { 161 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 162 S_IFSOCK, S_IFIFO, S_IFMT, 163 }; 164 165 /* 166 * Insq/Remq for the vnode usage lists. 167 */ 168 #define bufinsvn(bp, dp) LIST_INSERT_HEAD(dp, bp, b_vnbufs) 169 #define bufremvn(bp) { \ 170 LIST_REMOVE(bp, b_vnbufs); \ 171 (bp)->b_vnbufs.le_next = NOLIST; \ 172 } 173 174 int doforce = 1; /* 1 => permit forcible unmounting */ 175 176 /* 177 * Local declarations. 178 */ 179 180 static void vn_initialize_syncerd(void); 181 182 /* 183 * Initialize the vnode management data structures. 184 */ 185 void 186 vntblinit(void) 187 { 188 189 vn_initialize_syncerd(); 190 vfs_mount_sysinit(); 191 vfs_vnode_sysinit(); 192 } 193 194 /* 195 * Flush out and invalidate all buffers associated with a vnode. 196 * Called with the underlying vnode locked, which should prevent new dirty 197 * buffers from being queued. 198 */ 199 int 200 vinvalbuf(struct vnode *vp, int flags, kauth_cred_t cred, struct lwp *l, 201 bool catch_p, int slptimeo) 202 { 203 struct buf *bp, *nbp; 204 int error; 205 int flushflags = PGO_ALLPAGES | PGO_FREE | PGO_SYNCIO | 206 (flags & V_SAVE ? PGO_CLEANIT | PGO_RECLAIM : 0); 207 208 /* XXXUBC this doesn't look at flags or slp* */ 209 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 210 error = VOP_PUTPAGES(vp, 0, 0, flushflags); 211 if (error) { 212 return error; 213 } 214 215 if (flags & V_SAVE) { 216 error = VOP_FSYNC(vp, cred, FSYNC_WAIT|FSYNC_RECLAIM, 0, 0); 217 if (error) 218 return error; 219 KASSERT(LIST_EMPTY(&vp->v_dirtyblkhd)); 220 } 221 222 mutex_enter(&bufcache_lock); 223 restart: 224 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 225 KASSERT(bp->b_vp == vp); 226 nbp = LIST_NEXT(bp, b_vnbufs); 227 error = bbusy(bp, catch_p, slptimeo, NULL); 228 if (error != 0) { 229 if (error == EPASSTHROUGH) 230 goto restart; 231 mutex_exit(&bufcache_lock); 232 return error; 233 } 234 brelsel(bp, BC_INVAL | BC_VFLUSH); 235 } 236 237 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 238 KASSERT(bp->b_vp == vp); 239 nbp = LIST_NEXT(bp, b_vnbufs); 240 error = bbusy(bp, catch_p, slptimeo, NULL); 241 if (error != 0) { 242 if (error == EPASSTHROUGH) 243 goto restart; 244 mutex_exit(&bufcache_lock); 245 return error; 246 } 247 /* 248 * XXX Since there are no node locks for NFS, I believe 249 * there is a slight chance that a delayed write will 250 * occur while sleeping just above, so check for it. 251 */ 252 if ((bp->b_oflags & BO_DELWRI) && (flags & V_SAVE)) { 253 #ifdef DEBUG 254 printf("buffer still DELWRI\n"); 255 #endif 256 bp->b_cflags |= BC_BUSY | BC_VFLUSH; 257 mutex_exit(&bufcache_lock); 258 VOP_BWRITE(bp->b_vp, bp); 259 mutex_enter(&bufcache_lock); 260 goto restart; 261 } 262 brelsel(bp, BC_INVAL | BC_VFLUSH); 263 } 264 265 #ifdef DIAGNOSTIC 266 if (!LIST_EMPTY(&vp->v_cleanblkhd) || !LIST_EMPTY(&vp->v_dirtyblkhd)) 267 panic("vinvalbuf: flush failed, vp %p", vp); 268 #endif 269 270 mutex_exit(&bufcache_lock); 271 272 return 0; 273 } 274 275 /* 276 * Destroy any in core blocks past the truncation length. 277 * Called with the underlying vnode locked, which should prevent new dirty 278 * buffers from being queued. 279 */ 280 int 281 vtruncbuf(struct vnode *vp, daddr_t lbn, bool catch_p, int slptimeo) 282 { 283 struct buf *bp, *nbp; 284 int error; 285 voff_t off; 286 287 off = round_page((voff_t)lbn << vp->v_mount->mnt_fs_bshift); 288 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 289 error = VOP_PUTPAGES(vp, off, 0, PGO_FREE | PGO_SYNCIO); 290 if (error) { 291 return error; 292 } 293 294 mutex_enter(&bufcache_lock); 295 restart: 296 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 297 KASSERT(bp->b_vp == vp); 298 nbp = LIST_NEXT(bp, b_vnbufs); 299 if (bp->b_lblkno < lbn) 300 continue; 301 error = bbusy(bp, catch_p, slptimeo, NULL); 302 if (error != 0) { 303 if (error == EPASSTHROUGH) 304 goto restart; 305 mutex_exit(&bufcache_lock); 306 return error; 307 } 308 brelsel(bp, BC_INVAL | BC_VFLUSH); 309 } 310 311 for (bp = LIST_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) { 312 KASSERT(bp->b_vp == vp); 313 nbp = LIST_NEXT(bp, b_vnbufs); 314 if (bp->b_lblkno < lbn) 315 continue; 316 error = bbusy(bp, catch_p, slptimeo, NULL); 317 if (error != 0) { 318 if (error == EPASSTHROUGH) 319 goto restart; 320 mutex_exit(&bufcache_lock); 321 return error; 322 } 323 brelsel(bp, BC_INVAL | BC_VFLUSH); 324 } 325 mutex_exit(&bufcache_lock); 326 327 return 0; 328 } 329 330 /* 331 * Flush all dirty buffers from a vnode. 332 * Called with the underlying vnode locked, which should prevent new dirty 333 * buffers from being queued. 334 */ 335 int 336 vflushbuf(struct vnode *vp, int flags) 337 { 338 struct buf *bp, *nbp; 339 int error, pflags; 340 bool dirty, sync; 341 342 sync = (flags & FSYNC_WAIT) != 0; 343 pflags = PGO_CLEANIT | PGO_ALLPAGES | 344 (sync ? PGO_SYNCIO : 0) | 345 ((flags & FSYNC_LAZY) ? PGO_LAZY : 0); 346 rw_enter(vp->v_uobj.vmobjlock, RW_WRITER); 347 (void) VOP_PUTPAGES(vp, 0, 0, pflags); 348 349 loop: 350 mutex_enter(&bufcache_lock); 351 for (bp = LIST_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) { 352 KASSERT(bp->b_vp == vp); 353 nbp = LIST_NEXT(bp, b_vnbufs); 354 if ((bp->b_cflags & BC_BUSY)) 355 continue; 356 if ((bp->b_oflags & BO_DELWRI) == 0) 357 panic("vflushbuf: not dirty, bp %p", bp); 358 bp->b_cflags |= BC_BUSY | BC_VFLUSH; 359 mutex_exit(&bufcache_lock); 360 /* 361 * Wait for I/O associated with indirect blocks to complete, 362 * since there is no way to quickly wait for them below. 363 */ 364 if (bp->b_vp == vp || !sync) 365 (void) bawrite(bp); 366 else { 367 error = bwrite(bp); 368 if (error) 369 return error; 370 } 371 goto loop; 372 } 373 mutex_exit(&bufcache_lock); 374 375 if (!sync) 376 return 0; 377 378 mutex_enter(vp->v_interlock); 379 while (vp->v_numoutput != 0) 380 cv_wait(&vp->v_cv, vp->v_interlock); 381 dirty = !LIST_EMPTY(&vp->v_dirtyblkhd); 382 mutex_exit(vp->v_interlock); 383 384 if (dirty) { 385 vprint("vflushbuf: dirty", vp); 386 goto loop; 387 } 388 389 return 0; 390 } 391 392 /* 393 * Create a vnode for a block device. 394 * Used for root filesystem and swap areas. 395 * Also used for memory file system special devices. 396 */ 397 int 398 bdevvp(dev_t dev, vnode_t **vpp) 399 { 400 struct vattr va; 401 402 vattr_null(&va); 403 va.va_type = VBLK; 404 va.va_rdev = dev; 405 406 return vcache_new(dead_rootmount, NULL, &va, NOCRED, NULL, vpp); 407 } 408 409 /* 410 * Create a vnode for a character device. 411 * Used for kernfs and some console handling. 412 */ 413 int 414 cdevvp(dev_t dev, vnode_t **vpp) 415 { 416 struct vattr va; 417 418 vattr_null(&va); 419 va.va_type = VCHR; 420 va.va_rdev = dev; 421 422 return vcache_new(dead_rootmount, NULL, &va, NOCRED, NULL, vpp); 423 } 424 425 /* 426 * Associate a buffer with a vnode. There must already be a hold on 427 * the vnode. 428 */ 429 void 430 bgetvp(struct vnode *vp, struct buf *bp) 431 { 432 433 KASSERT(bp->b_vp == NULL); 434 KASSERT(bp->b_objlock == &buffer_lock); 435 KASSERT(mutex_owned(vp->v_interlock)); 436 KASSERT(mutex_owned(&bufcache_lock)); 437 KASSERT((bp->b_cflags & BC_BUSY) != 0); 438 KASSERT(!cv_has_waiters(&bp->b_done)); 439 440 vholdl(vp); 441 bp->b_vp = vp; 442 if (vp->v_type == VBLK || vp->v_type == VCHR) 443 bp->b_dev = vp->v_rdev; 444 else 445 bp->b_dev = NODEV; 446 447 /* 448 * Insert onto list for new vnode. 449 */ 450 bufinsvn(bp, &vp->v_cleanblkhd); 451 bp->b_objlock = vp->v_interlock; 452 } 453 454 /* 455 * Disassociate a buffer from a vnode. 456 */ 457 void 458 brelvp(struct buf *bp) 459 { 460 struct vnode *vp = bp->b_vp; 461 462 KASSERT(vp != NULL); 463 KASSERT(bp->b_objlock == vp->v_interlock); 464 KASSERT(mutex_owned(vp->v_interlock)); 465 KASSERT(mutex_owned(&bufcache_lock)); 466 KASSERT((bp->b_cflags & BC_BUSY) != 0); 467 KASSERT(!cv_has_waiters(&bp->b_done)); 468 469 /* 470 * Delete from old vnode list, if on one. 471 */ 472 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 473 bufremvn(bp); 474 475 if ((vp->v_iflag & (VI_ONWORKLST | VI_PAGES)) == VI_ONWORKLST && 476 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) 477 vn_syncer_remove_from_worklist(vp); 478 479 bp->b_objlock = &buffer_lock; 480 bp->b_vp = NULL; 481 holdrelel(vp); 482 } 483 484 /* 485 * Reassign a buffer from one vnode list to another. 486 * The list reassignment must be within the same vnode. 487 * Used to assign file specific control information 488 * (indirect blocks) to the list to which they belong. 489 */ 490 void 491 reassignbuf(struct buf *bp, struct vnode *vp) 492 { 493 struct buflists *listheadp; 494 int delayx; 495 496 KASSERT(mutex_owned(&bufcache_lock)); 497 KASSERT(bp->b_objlock == vp->v_interlock); 498 KASSERT(mutex_owned(vp->v_interlock)); 499 KASSERT((bp->b_cflags & BC_BUSY) != 0); 500 501 /* 502 * Delete from old vnode list, if on one. 503 */ 504 if (LIST_NEXT(bp, b_vnbufs) != NOLIST) 505 bufremvn(bp); 506 507 /* 508 * If dirty, put on list of dirty buffers; 509 * otherwise insert onto list of clean buffers. 510 */ 511 if ((bp->b_oflags & BO_DELWRI) == 0) { 512 listheadp = &vp->v_cleanblkhd; 513 if ((vp->v_iflag & (VI_ONWORKLST | VI_PAGES)) == 514 VI_ONWORKLST && 515 LIST_FIRST(&vp->v_dirtyblkhd) == NULL) 516 vn_syncer_remove_from_worklist(vp); 517 } else { 518 listheadp = &vp->v_dirtyblkhd; 519 if ((vp->v_iflag & VI_ONWORKLST) == 0) { 520 switch (vp->v_type) { 521 case VDIR: 522 delayx = dirdelay; 523 break; 524 case VBLK: 525 if (spec_node_getmountedfs(vp) != NULL) { 526 delayx = metadelay; 527 break; 528 } 529 /* fall through */ 530 default: 531 delayx = filedelay; 532 break; 533 } 534 if (!vp->v_mount || 535 (vp->v_mount->mnt_flag & MNT_ASYNC) == 0) 536 vn_syncer_add_to_worklist(vp, delayx); 537 } 538 } 539 bufinsvn(bp, listheadp); 540 } 541 542 /* 543 * Lookup a vnode by device number and return it referenced. 544 */ 545 int 546 vfinddev(dev_t dev, enum vtype type, vnode_t **vpp) 547 { 548 549 return (spec_node_lookup_by_dev(type, dev, VDEAD_NOWAIT, vpp) == 0); 550 } 551 552 /* 553 * Revoke all the vnodes corresponding to the specified minor number 554 * range (endpoints inclusive) of the specified major. 555 */ 556 void 557 vdevgone(int maj, int minl, int minh, enum vtype type) 558 { 559 vnode_t *vp; 560 dev_t dev; 561 int mn; 562 563 for (mn = minl; mn <= minh; mn++) { 564 dev = makedev(maj, mn); 565 /* 566 * Notify anyone trying to get at this device that it 567 * has been detached, and then revoke it. 568 */ 569 switch (type) { 570 case VBLK: 571 bdev_detached(dev); 572 break; 573 case VCHR: 574 cdev_detached(dev); 575 break; 576 default: 577 panic("invalid specnode type: %d", type); 578 } 579 /* 580 * Passing 0 as flags, instead of VDEAD_NOWAIT, means 581 * spec_node_lookup_by_dev will wait for vnodes it 582 * finds concurrently being revoked before returning. 583 */ 584 while (spec_node_lookup_by_dev(type, dev, 0, &vp) == 0) { 585 VOP_REVOKE(vp, REVOKEALL); 586 vrele(vp); 587 } 588 } 589 } 590 591 /* 592 * The filesystem synchronizer mechanism - syncer. 593 * 594 * It is useful to delay writes of file data and filesystem metadata for 595 * a certain amount of time so that quickly created and deleted files need 596 * not waste disk bandwidth being created and removed. To implement this, 597 * vnodes are appended to a "workitem" queue. 598 * 599 * Most pending metadata should not wait for more than ten seconds. Thus, 600 * mounted on block devices are delayed only about a half the time that file 601 * data is delayed. Similarly, directory updates are more critical, so are 602 * only delayed about a third the time that file data is delayed. 603 * 604 * There are SYNCER_MAXDELAY queues that are processed in a round-robin 605 * manner at a rate of one each second (driven off the filesystem syner 606 * thread). The syncer_delayno variable indicates the next queue that is 607 * to be processed. Items that need to be processed soon are placed in 608 * this queue: 609 * 610 * syncer_workitem_pending[syncer_delayno] 611 * 612 * A delay of e.g. fifteen seconds is done by placing the request fifteen 613 * entries later in the queue: 614 * 615 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 616 * 617 * Flag VI_ONWORKLST indicates that vnode is added into the queue. 618 */ 619 620 #define SYNCER_MAXDELAY 32 621 622 typedef TAILQ_HEAD(synclist, vnode_impl) synclist_t; 623 624 static void vn_syncer_add1(struct vnode *, int); 625 static void sysctl_vfs_syncfs_setup(struct sysctllog **); 626 627 /* 628 * Defines and variables for the syncer process. 629 */ 630 int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 631 time_t syncdelay = 30; /* max time to delay syncing data */ 632 time_t filedelay = 30; /* time to delay syncing files */ 633 time_t dirdelay = 15; /* time to delay syncing directories */ 634 time_t metadelay = 10; /* time to delay syncing metadata */ 635 time_t lockdelay = 1; /* time to delay if locking fails */ 636 637 static kmutex_t syncer_data_lock; /* short term lock on data structs */ 638 639 static int syncer_delayno = 0; 640 static long syncer_last; 641 static synclist_t * syncer_workitem_pending; 642 643 static void 644 vn_initialize_syncerd(void) 645 { 646 int i; 647 648 syncer_last = SYNCER_MAXDELAY + 2; 649 650 sysctl_vfs_syncfs_setup(NULL); 651 652 syncer_workitem_pending = 653 kmem_alloc(syncer_last * sizeof (struct synclist), KM_SLEEP); 654 655 for (i = 0; i < syncer_last; i++) 656 TAILQ_INIT(&syncer_workitem_pending[i]); 657 658 mutex_init(&syncer_data_lock, MUTEX_DEFAULT, IPL_NONE); 659 } 660 661 /* 662 * Return delay factor appropriate for the given file system. For 663 * WAPBL we use the sync vnode to burst out metadata updates: sync 664 * those file systems more frequently. 665 */ 666 static inline int 667 sync_delay(struct mount *mp) 668 { 669 670 return mp->mnt_wapbl != NULL ? metadelay : syncdelay; 671 } 672 673 /* 674 * Compute the next slot index from delay. 675 */ 676 static inline int 677 sync_delay_slot(int delayx) 678 { 679 680 if (delayx > syncer_maxdelay - 2) 681 delayx = syncer_maxdelay - 2; 682 return (syncer_delayno + delayx) % syncer_last; 683 } 684 685 /* 686 * Add an item to the syncer work queue. 687 */ 688 static void 689 vn_syncer_add1(struct vnode *vp, int delayx) 690 { 691 synclist_t *slp; 692 vnode_impl_t *vip = VNODE_TO_VIMPL(vp); 693 694 KASSERT(mutex_owned(&syncer_data_lock)); 695 696 if (vp->v_iflag & VI_ONWORKLST) { 697 /* 698 * Remove in order to adjust the position of the vnode. 699 * Note: called from sched_sync(), which will not hold 700 * interlock, therefore we cannot modify v_iflag here. 701 */ 702 slp = &syncer_workitem_pending[vip->vi_synclist_slot]; 703 TAILQ_REMOVE(slp, vip, vi_synclist); 704 } else { 705 KASSERT(mutex_owned(vp->v_interlock)); 706 vp->v_iflag |= VI_ONWORKLST; 707 } 708 709 vip->vi_synclist_slot = sync_delay_slot(delayx); 710 711 slp = &syncer_workitem_pending[vip->vi_synclist_slot]; 712 TAILQ_INSERT_TAIL(slp, vip, vi_synclist); 713 } 714 715 void 716 vn_syncer_add_to_worklist(struct vnode *vp, int delayx) 717 { 718 vnode_impl_t *vip = VNODE_TO_VIMPL(vp); 719 720 KASSERT(mutex_owned(vp->v_interlock)); 721 722 mutex_enter(&syncer_data_lock); 723 vn_syncer_add1(vp, delayx); 724 SDT_PROBE3(vfs, syncer, worklist, vnode__add, 725 vp, delayx, vip->vi_synclist_slot); 726 mutex_exit(&syncer_data_lock); 727 } 728 729 /* 730 * Remove an item from the syncer work queue. 731 */ 732 void 733 vn_syncer_remove_from_worklist(struct vnode *vp) 734 { 735 synclist_t *slp; 736 vnode_impl_t *vip = VNODE_TO_VIMPL(vp); 737 738 KASSERT(mutex_owned(vp->v_interlock)); 739 740 if (vp->v_iflag & VI_ONWORKLST) { 741 mutex_enter(&syncer_data_lock); 742 SDT_PROBE1(vfs, syncer, worklist, vnode__remove, vp); 743 vp->v_iflag &= ~VI_ONWORKLST; 744 slp = &syncer_workitem_pending[vip->vi_synclist_slot]; 745 TAILQ_REMOVE(slp, vip, vi_synclist); 746 mutex_exit(&syncer_data_lock); 747 } 748 } 749 750 /* 751 * Add this mount point to the syncer. 752 */ 753 void 754 vfs_syncer_add_to_worklist(struct mount *mp) 755 { 756 static int start, incr, next; 757 int vdelay; 758 759 KASSERT(mutex_owned(mp->mnt_updating)); 760 KASSERT((mp->mnt_iflag & IMNT_ONWORKLIST) == 0); 761 762 /* 763 * We attempt to scatter the mount points on the list 764 * so that they will go off at evenly distributed times 765 * even if all the filesystems are mounted at once. 766 */ 767 768 next += incr; 769 if (next == 0 || next > syncer_maxdelay) { 770 start /= 2; 771 incr /= 2; 772 if (start == 0) { 773 start = syncer_maxdelay / 2; 774 incr = syncer_maxdelay; 775 } 776 next = start; 777 } 778 mp->mnt_iflag |= IMNT_ONWORKLIST; 779 vdelay = sync_delay(mp); 780 mp->mnt_synclist_slot = vdelay > 0 ? next % vdelay : 0; 781 SDT_PROBE3(vfs, syncer, worklist, mount__add, 782 mp, vdelay, mp->mnt_synclist_slot); 783 } 784 785 /* 786 * Remove the mount point from the syncer. 787 */ 788 void 789 vfs_syncer_remove_from_worklist(struct mount *mp) 790 { 791 792 KASSERT(mutex_owned(mp->mnt_updating)); 793 KASSERT((mp->mnt_iflag & IMNT_ONWORKLIST) != 0); 794 795 SDT_PROBE1(vfs, syncer, worklist, mount__remove, mp); 796 mp->mnt_iflag &= ~IMNT_ONWORKLIST; 797 } 798 799 /* 800 * Try lazy sync, return true on success. 801 */ 802 static bool 803 lazy_sync_vnode(struct vnode *vp) 804 { 805 bool synced; 806 int error; 807 808 KASSERT(mutex_owned(&syncer_data_lock)); 809 810 synced = false; 811 if ((error = vcache_tryvget(vp)) == 0) { 812 mutex_exit(&syncer_data_lock); 813 if ((error = vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT)) == 0) { 814 synced = true; 815 SDT_PROBE1(vfs, syncer, sync, vnode__start, vp); 816 error = VOP_FSYNC(vp, curlwp->l_cred, 817 FSYNC_LAZY, 0, 0); 818 SDT_PROBE2(vfs, syncer, sync, vnode__done, vp, error); 819 vput(vp); 820 } else { 821 SDT_PROBE2(vfs, syncer, sync, vnode__fail__lock, 822 vp, error); 823 vrele(vp); 824 } 825 mutex_enter(&syncer_data_lock); 826 } else { 827 SDT_PROBE2(vfs, syncer, sync, vnode__fail__vget, vp, error); 828 } 829 return synced; 830 } 831 832 /* 833 * System filesystem synchronizer daemon. 834 */ 835 void 836 sched_sync(void *arg) 837 { 838 mount_iterator_t *iter; 839 synclist_t *slp; 840 struct vnode_impl *vi; 841 struct vnode *vp; 842 struct mount *mp; 843 time_t starttime, endtime; 844 int vdelay, oslot, nslot, delayx; 845 bool synced; 846 int error; 847 848 for (;;) { 849 starttime = time_second; 850 SDT_PROBE1(vfs, syncer, sync, start, starttime); 851 852 /* 853 * Sync mounts whose dirty time has expired. 854 */ 855 mountlist_iterator_init(&iter); 856 while ((mp = mountlist_iterator_trynext(iter)) != NULL) { 857 if ((mp->mnt_iflag & IMNT_ONWORKLIST) == 0 || 858 mp->mnt_synclist_slot != syncer_delayno) { 859 SDT_PROBE1(vfs, syncer, sync, mount__skip, 860 mp); 861 continue; 862 } 863 864 vdelay = sync_delay(mp); 865 oslot = mp->mnt_synclist_slot; 866 nslot = sync_delay_slot(vdelay); 867 mp->mnt_synclist_slot = nslot; 868 SDT_PROBE4(vfs, syncer, worklist, mount__update, 869 mp, vdelay, oslot, nslot); 870 871 SDT_PROBE1(vfs, syncer, sync, mount__start, mp); 872 error = VFS_SYNC(mp, MNT_LAZY, curlwp->l_cred); 873 SDT_PROBE2(vfs, syncer, sync, mount__done, 874 mp, error); 875 } 876 mountlist_iterator_destroy(iter); 877 878 mutex_enter(&syncer_data_lock); 879 880 /* 881 * Push files whose dirty time has expired. 882 */ 883 slp = &syncer_workitem_pending[syncer_delayno]; 884 syncer_delayno += 1; 885 if (syncer_delayno >= syncer_last) 886 syncer_delayno = 0; 887 888 while ((vi = TAILQ_FIRST(slp)) != NULL) { 889 vp = VIMPL_TO_VNODE(vi); 890 synced = lazy_sync_vnode(vp); 891 892 /* 893 * XXX The vnode may have been recycled, in which 894 * case it may have a new identity. 895 */ 896 vi = TAILQ_FIRST(slp); 897 if (vi != NULL && VIMPL_TO_VNODE(vi) == vp) { 898 /* 899 * Put us back on the worklist. The worklist 900 * routine will remove us from our current 901 * position and then add us back in at a later 902 * position. 903 * 904 * Try again sooner rather than later if 905 * we were unable to lock the vnode. Lock 906 * failure should not prevent us from doing 907 * the sync "soon". 908 * 909 * If we locked it yet arrive here, it's 910 * likely that lazy sync is in progress and 911 * so the vnode still has dirty metadata. 912 * syncdelay is mainly to get this vnode out 913 * of the way so we do not consider it again 914 * "soon" in this loop, so the delay time is 915 * not critical as long as it is not "soon". 916 * While write-back strategy is the file 917 * system's domain, we expect write-back to 918 * occur no later than syncdelay seconds 919 * into the future. 920 */ 921 delayx = synced ? syncdelay : lockdelay; 922 oslot = vi->vi_synclist_slot; 923 vn_syncer_add1(vp, delayx); 924 nslot = vi->vi_synclist_slot; 925 SDT_PROBE4(vfs, syncer, worklist, 926 vnode__update, 927 vp, delayx, oslot, nslot); 928 } 929 } 930 931 endtime = time_second; 932 933 SDT_PROBE2(vfs, syncer, sync, done, starttime, endtime); 934 935 /* 936 * If it has taken us less than a second to process the 937 * current work, then wait. Otherwise start right over 938 * again. We can still lose time if any single round 939 * takes more than two seconds, but it does not really 940 * matter as we are just trying to generally pace the 941 * filesystem activity. 942 */ 943 if (endtime == starttime) { 944 kpause("syncer", false, hz, &syncer_data_lock); 945 } 946 mutex_exit(&syncer_data_lock); 947 } 948 } 949 950 static void 951 sysctl_vfs_syncfs_setup(struct sysctllog **clog) 952 { 953 const struct sysctlnode *rnode, *cnode; 954 955 sysctl_createv(clog, 0, NULL, &rnode, 956 CTLFLAG_PERMANENT, 957 CTLTYPE_NODE, "sync", 958 SYSCTL_DESCR("syncer options"), 959 NULL, 0, NULL, 0, 960 CTL_VFS, CTL_CREATE, CTL_EOL); 961 962 sysctl_createv(clog, 0, &rnode, &cnode, 963 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 964 CTLTYPE_QUAD, "delay", 965 SYSCTL_DESCR("max time to delay syncing data"), 966 NULL, 0, &syncdelay, 0, 967 CTL_CREATE, CTL_EOL); 968 969 sysctl_createv(clog, 0, &rnode, &cnode, 970 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 971 CTLTYPE_QUAD, "filedelay", 972 SYSCTL_DESCR("time to delay syncing files"), 973 NULL, 0, &filedelay, 0, 974 CTL_CREATE, CTL_EOL); 975 976 sysctl_createv(clog, 0, &rnode, &cnode, 977 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 978 CTLTYPE_QUAD, "dirdelay", 979 SYSCTL_DESCR("time to delay syncing directories"), 980 NULL, 0, &dirdelay, 0, 981 CTL_CREATE, CTL_EOL); 982 983 sysctl_createv(clog, 0, &rnode, &cnode, 984 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 985 CTLTYPE_QUAD, "metadelay", 986 SYSCTL_DESCR("time to delay syncing metadata"), 987 NULL, 0, &metadelay, 0, 988 CTL_CREATE, CTL_EOL); 989 } 990 991 /* 992 * sysctl helper routine to return list of supported fstypes 993 */ 994 int 995 sysctl_vfs_generic_fstypes(SYSCTLFN_ARGS) 996 { 997 char bf[sizeof(((struct statvfs *)NULL)->f_fstypename)]; 998 char *where = oldp; 999 struct vfsops *v; 1000 size_t needed, left, slen; 1001 int error, first; 1002 1003 if (newp != NULL) 1004 return SET_ERROR(EPERM); 1005 if (namelen != 0) 1006 return SET_ERROR(EINVAL); 1007 1008 first = 1; 1009 error = 0; 1010 needed = 0; 1011 left = *oldlenp; 1012 1013 sysctl_unlock(); 1014 mutex_enter(&vfs_list_lock); 1015 LIST_FOREACH(v, &vfs_list, vfs_list) { 1016 if (where == NULL) 1017 needed += strlen(v->vfs_name) + 1; 1018 else { 1019 memset(bf, 0, sizeof(bf)); 1020 if (first) { 1021 strncpy(bf, v->vfs_name, sizeof(bf)); 1022 first = 0; 1023 } else { 1024 bf[0] = ' '; 1025 strncpy(bf + 1, v->vfs_name, sizeof(bf) - 1); 1026 } 1027 bf[sizeof(bf)-1] = '\0'; 1028 slen = strlen(bf); 1029 if (left < slen + 1) 1030 break; 1031 v->vfs_refcount++; 1032 mutex_exit(&vfs_list_lock); 1033 /* +1 to copy out the trailing NUL byte */ 1034 error = copyout(bf, where, slen + 1); 1035 mutex_enter(&vfs_list_lock); 1036 v->vfs_refcount--; 1037 if (error) 1038 break; 1039 where += slen; 1040 needed += slen; 1041 left -= slen; 1042 } 1043 } 1044 mutex_exit(&vfs_list_lock); 1045 sysctl_relock(); 1046 *oldlenp = needed; 1047 return error; 1048 } 1049 1050 int kinfo_vdebug = 1; 1051 int kinfo_vgetfailed; 1052 1053 #define KINFO_VNODESLOP 10 1054 1055 /* 1056 * Dump vnode list (via sysctl). 1057 * Copyout address of vnode followed by vnode. 1058 */ 1059 int 1060 sysctl_kern_vnode(SYSCTLFN_ARGS) 1061 { 1062 char *where = oldp; 1063 size_t *sizep = oldlenp; 1064 struct mount *mp; 1065 vnode_t *vp, vbuf; 1066 mount_iterator_t *iter; 1067 struct vnode_iterator *marker; 1068 char *bp = where; 1069 char *ewhere; 1070 int error; 1071 1072 if (namelen != 0) 1073 return SET_ERROR(EOPNOTSUPP); 1074 if (newp != NULL) 1075 return SET_ERROR(EPERM); 1076 1077 #define VPTRSZ sizeof(vnode_t *) 1078 #define VNODESZ sizeof(vnode_t) 1079 if (where == NULL) { 1080 *sizep = (numvnodes + KINFO_VNODESLOP) * (VPTRSZ + VNODESZ); 1081 return 0; 1082 } 1083 ewhere = where + *sizep; 1084 1085 sysctl_unlock(); 1086 mountlist_iterator_init(&iter); 1087 while ((mp = mountlist_iterator_next(iter)) != NULL) { 1088 vfs_vnode_iterator_init(mp, &marker); 1089 while ((vp = vfs_vnode_iterator_next(marker, NULL, NULL))) { 1090 if (bp + VPTRSZ + VNODESZ > ewhere) { 1091 vrele(vp); 1092 vfs_vnode_iterator_destroy(marker); 1093 mountlist_iterator_destroy(iter); 1094 sysctl_relock(); 1095 *sizep = bp - where; 1096 return SET_ERROR(ENOMEM); 1097 } 1098 memcpy(&vbuf, vp, VNODESZ); 1099 if ((error = copyout(&vp, bp, VPTRSZ)) || 1100 (error = copyout(&vbuf, bp + VPTRSZ, VNODESZ))) { 1101 vrele(vp); 1102 vfs_vnode_iterator_destroy(marker); 1103 mountlist_iterator_destroy(iter); 1104 sysctl_relock(); 1105 return error; 1106 } 1107 vrele(vp); 1108 bp += VPTRSZ + VNODESZ; 1109 } 1110 vfs_vnode_iterator_destroy(marker); 1111 } 1112 mountlist_iterator_destroy(iter); 1113 sysctl_relock(); 1114 1115 *sizep = bp - where; 1116 return 0; 1117 } 1118 1119 /* 1120 * Set vnode attributes to VNOVAL 1121 */ 1122 void 1123 vattr_null(struct vattr *vap) 1124 { 1125 1126 memset(vap, 0, sizeof(*vap)); 1127 1128 vap->va_type = VNON; 1129 1130 /* 1131 * Assign individually so that it is safe even if size and 1132 * sign of each member are varied. 1133 */ 1134 vap->va_mode = VNOVAL; 1135 vap->va_nlink = VNOVAL; 1136 vap->va_uid = VNOVAL; 1137 vap->va_gid = VNOVAL; 1138 vap->va_fsid = VNOVAL; 1139 vap->va_fileid = VNOVAL; 1140 vap->va_size = VNOVAL; 1141 vap->va_blocksize = VNOVAL; 1142 vap->va_atime.tv_sec = 1143 vap->va_mtime.tv_sec = 1144 vap->va_ctime.tv_sec = 1145 vap->va_birthtime.tv_sec = VNOVAL; 1146 vap->va_atime.tv_nsec = 1147 vap->va_mtime.tv_nsec = 1148 vap->va_ctime.tv_nsec = 1149 vap->va_birthtime.tv_nsec = VNOVAL; 1150 vap->va_gen = VNOVAL; 1151 vap->va_flags = VNOVAL; 1152 vap->va_rdev = VNOVAL; 1153 vap->va_bytes = VNOVAL; 1154 } 1155 1156 /* 1157 * Vnode state to string. 1158 */ 1159 const char * 1160 vstate_name(enum vnode_state state) 1161 { 1162 1163 switch (state) { 1164 case VS_ACTIVE: 1165 return "ACTIVE"; 1166 case VS_MARKER: 1167 return "MARKER"; 1168 case VS_LOADING: 1169 return "LOADING"; 1170 case VS_LOADED: 1171 return "LOADED"; 1172 case VS_BLOCKED: 1173 return "BLOCKED"; 1174 case VS_RECLAIMING: 1175 return "RECLAIMING"; 1176 case VS_RECLAIMED: 1177 return "RECLAIMED"; 1178 default: 1179 return "ILLEGAL"; 1180 } 1181 } 1182 1183 /* 1184 * Print a description of a vnode (common part). 1185 */ 1186 static void 1187 vprint_common(struct vnode *vp, const char *prefix, 1188 void (*pr)(const char *, ...) __printflike(1, 2)) 1189 { 1190 int n; 1191 char bf[96]; 1192 const uint8_t *cp; 1193 vnode_impl_t *vip; 1194 const char * const vnode_tags[] = { VNODE_TAGS }; 1195 const char * const vnode_types[] = { VNODE_TYPES }; 1196 const char vnode_flagbits[] = VNODE_FLAGBITS; 1197 1198 #define ARRAY_SIZE(arr) (sizeof(arr) / sizeof(arr[0])) 1199 #define ARRAY_PRINT(idx, arr) \ 1200 ((unsigned int)(idx) < ARRAY_SIZE(arr) ? (arr)[(idx)] : "UNKNOWN") 1201 1202 vip = VNODE_TO_VIMPL(vp); 1203 1204 snprintb(bf, sizeof(bf), 1205 vnode_flagbits, vp->v_iflag | vp->v_vflag | vp->v_uflag); 1206 1207 (*pr)("vnode %p flags %s\n", vp, bf); 1208 (*pr)("%stag %s(%d) type %s(%d) mount %p typedata %p\n", prefix, 1209 ARRAY_PRINT(vp->v_tag, vnode_tags), vp->v_tag, 1210 ARRAY_PRINT(vp->v_type, vnode_types), vp->v_type, 1211 vp->v_mount, vp->v_mountedhere); 1212 (*pr)("%susecount %d writecount %d holdcount %d\n", prefix, 1213 vrefcnt(vp), vp->v_writecount, vp->v_holdcnt); 1214 (*pr)("%ssize %" PRIx64 " writesize %" PRIx64 " numoutput %d\n", 1215 prefix, vp->v_size, vp->v_writesize, vp->v_numoutput); 1216 (*pr)("%sdata %p lock %p\n", prefix, vp->v_data, &vip->vi_lock); 1217 1218 (*pr)("%sstate %s key(%p %zd)", prefix, vstate_name(vip->vi_state), 1219 vip->vi_key.vk_mount, vip->vi_key.vk_key_len); 1220 n = vip->vi_key.vk_key_len; 1221 cp = vip->vi_key.vk_key; 1222 while (n-- > 0) 1223 (*pr)(" %02x", *cp++); 1224 (*pr)("\n"); 1225 (*pr)("%slrulisthd %p\n", prefix, vip->vi_lrulisthd); 1226 1227 #undef ARRAY_PRINT 1228 #undef ARRAY_SIZE 1229 } 1230 1231 /* 1232 * Print out a description of a vnode. 1233 */ 1234 void 1235 vprint(const char *label, struct vnode *vp) 1236 { 1237 1238 if (label != NULL) 1239 printf("%s: ", label); 1240 vprint_common(vp, "\t", printf); 1241 if (vp->v_data != NULL) { 1242 printf("\t"); 1243 VOP_PRINT(vp); 1244 } 1245 } 1246 1247 /* 1248 * Given a file system name, look up the vfsops for that 1249 * file system, or return NULL if file system isn't present 1250 * in the kernel. 1251 */ 1252 struct vfsops * 1253 vfs_getopsbyname(const char *name) 1254 { 1255 struct vfsops *v; 1256 1257 mutex_enter(&vfs_list_lock); 1258 LIST_FOREACH(v, &vfs_list, vfs_list) { 1259 if (strcmp(v->vfs_name, name) == 0) 1260 break; 1261 } 1262 if (v != NULL) 1263 v->vfs_refcount++; 1264 mutex_exit(&vfs_list_lock); 1265 1266 return v; 1267 } 1268 1269 void 1270 copy_statvfs_info(struct statvfs *sbp, const struct mount *mp) 1271 { 1272 const struct statvfs *mbp; 1273 1274 if (sbp == (mbp = &mp->mnt_stat)) 1275 return; 1276 1277 (void)memcpy(&sbp->f_fsidx, &mbp->f_fsidx, sizeof(sbp->f_fsidx)); 1278 sbp->f_fsid = mbp->f_fsid; 1279 sbp->f_owner = mbp->f_owner; 1280 sbp->f_flag = mbp->f_flag; 1281 sbp->f_syncwrites = mbp->f_syncwrites; 1282 sbp->f_asyncwrites = mbp->f_asyncwrites; 1283 sbp->f_syncreads = mbp->f_syncreads; 1284 sbp->f_asyncreads = mbp->f_asyncreads; 1285 (void)memcpy(sbp->f_spare, mbp->f_spare, sizeof(mbp->f_spare)); 1286 (void)memcpy(sbp->f_fstypename, mbp->f_fstypename, 1287 sizeof(sbp->f_fstypename)); 1288 (void)memcpy(sbp->f_mntonname, mbp->f_mntonname, 1289 sizeof(sbp->f_mntonname)); 1290 (void)memcpy(sbp->f_mntfromname, mp->mnt_stat.f_mntfromname, 1291 sizeof(sbp->f_mntfromname)); 1292 (void)memcpy(sbp->f_mntfromlabel, mp->mnt_stat.f_mntfromlabel, 1293 sizeof(sbp->f_mntfromlabel)); 1294 sbp->f_namemax = mbp->f_namemax; 1295 } 1296 1297 int 1298 set_statvfs_info(const char *onp, int ukon, const char *fromp, int ukfrom, 1299 const char *vfsname, struct mount *mp, struct lwp *l) 1300 { 1301 int error; 1302 size_t size; 1303 struct statvfs *sfs = &mp->mnt_stat; 1304 int (*fun)(const void *, void *, size_t, size_t *); 1305 1306 (void)strlcpy(mp->mnt_stat.f_fstypename, vfsname, 1307 sizeof(mp->mnt_stat.f_fstypename)); 1308 1309 if (onp) { 1310 struct cwdinfo *cwdi = l->l_proc->p_cwdi; 1311 fun = (ukon == UIO_SYSSPACE) ? copystr : copyinstr; 1312 if (cwdi->cwdi_rdir != NULL) { 1313 size_t len; 1314 char *bp; 1315 char *path = PNBUF_GET(); 1316 1317 bp = path + MAXPATHLEN; 1318 *--bp = '\0'; 1319 rw_enter(&cwdi->cwdi_lock, RW_READER); 1320 error = getcwd_common(cwdi->cwdi_rdir, rootvnode, &bp, 1321 path, MAXPATHLEN / 2, 0, l); 1322 rw_exit(&cwdi->cwdi_lock); 1323 if (error) { 1324 PNBUF_PUT(path); 1325 return error; 1326 } 1327 1328 len = strlen(bp); 1329 if (len > sizeof(sfs->f_mntonname) - 1) 1330 len = sizeof(sfs->f_mntonname) - 1; 1331 (void)strncpy(sfs->f_mntonname, bp, len); 1332 PNBUF_PUT(path); 1333 1334 if (len < sizeof(sfs->f_mntonname) - 1) { 1335 error = (*fun)(onp, &sfs->f_mntonname[len], 1336 sizeof(sfs->f_mntonname) - len - 1, &size); 1337 if (error) 1338 return error; 1339 size += len; 1340 } else { 1341 size = len; 1342 } 1343 } else { 1344 error = (*fun)(onp, &sfs->f_mntonname, 1345 sizeof(sfs->f_mntonname) - 1, &size); 1346 if (error) 1347 return error; 1348 } 1349 (void)memset(sfs->f_mntonname + size, 0, 1350 sizeof(sfs->f_mntonname) - size); 1351 } 1352 1353 if (fromp) { 1354 fun = (ukfrom == UIO_SYSSPACE) ? copystr : copyinstr; 1355 error = (*fun)(fromp, sfs->f_mntfromname, 1356 sizeof(sfs->f_mntfromname) - 1, &size); 1357 if (error) 1358 return error; 1359 (void)memset(sfs->f_mntfromname + size, 0, 1360 sizeof(sfs->f_mntfromname) - size); 1361 } 1362 return 0; 1363 } 1364 1365 /* 1366 * Knob to control the precision of file timestamps: 1367 * 1368 * 0 = seconds only; nanoseconds zeroed. 1369 * 1 = seconds and nanoseconds, accurate within 1/HZ. 1370 * 2 = seconds and nanoseconds, truncated to microseconds. 1371 * >=3 = seconds and nanoseconds, maximum precision. 1372 */ 1373 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 1374 1375 int vfs_timestamp_precision __read_mostly = TSP_NSEC; 1376 1377 void 1378 vfs_timestamp(struct timespec *tsp) 1379 { 1380 struct timeval tv; 1381 1382 switch (vfs_timestamp_precision) { 1383 case TSP_SEC: 1384 tsp->tv_sec = time_second; 1385 tsp->tv_nsec = 0; 1386 break; 1387 case TSP_HZ: 1388 getnanotime(tsp); 1389 break; 1390 case TSP_USEC: 1391 microtime(&tv); 1392 TIMEVAL_TO_TIMESPEC(&tv, tsp); 1393 break; 1394 case TSP_NSEC: 1395 default: 1396 nanotime(tsp); 1397 break; 1398 } 1399 } 1400 1401 /* 1402 * The purpose of this routine is to remove granularity from accmode_t, 1403 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, 1404 * VADMIN and VAPPEND. 1405 * 1406 * If it returns 0, the caller is supposed to continue with the usual 1407 * access checks using 'accmode' as modified by this routine. If it 1408 * returns nonzero value, the caller is supposed to return that value 1409 * as errno. 1410 * 1411 * Note that after this routine runs, accmode may be zero. 1412 */ 1413 int 1414 vfs_unixify_accmode(accmode_t *accmode) 1415 { 1416 1417 /* 1418 * There is no way to specify explicit "deny" rule using 1419 * file mode or POSIX.1e ACLs. 1420 */ 1421 if (*accmode & VEXPLICIT_DENY) { 1422 *accmode = 0; 1423 return 0; 1424 } 1425 1426 /* 1427 * None of these can be translated into usual access bits. 1428 * Also, the common case for NFSv4 ACLs is to not contain 1429 * either of these bits. Caller should check for VWRITE 1430 * on the containing directory instead. 1431 */ 1432 if (*accmode & (VDELETE_CHILD | VDELETE)) 1433 return SET_ERROR(EPERM); 1434 1435 if (*accmode & VADMIN_PERMS) { 1436 *accmode &= ~VADMIN_PERMS; 1437 *accmode |= VADMIN; 1438 } 1439 1440 /* 1441 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL 1442 * or VSYNCHRONIZE using file mode or POSIX.1e ACL. 1443 */ 1444 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); 1445 1446 return 0; 1447 } 1448 1449 time_t rootfstime; /* recorded root fs time, if known */ 1450 void 1451 setrootfstime(time_t t) 1452 { 1453 1454 rootfstime = t; 1455 } 1456 1457 static const uint8_t vttodt_tab[] = { 1458 [VNON] = DT_UNKNOWN, 1459 [VREG] = DT_REG, 1460 [VDIR] = DT_DIR, 1461 [VBLK] = DT_BLK, 1462 [VCHR] = DT_CHR, 1463 [VLNK] = DT_LNK, 1464 [VSOCK] = DT_SOCK, 1465 [VFIFO] = DT_FIFO, 1466 [VBAD] = DT_UNKNOWN 1467 }; 1468 1469 uint8_t 1470 vtype2dt(enum vtype vt) 1471 { 1472 1473 CTASSERT(VBAD == __arraycount(vttodt_tab) - 1); 1474 return vttodt_tab[vt]; 1475 } 1476 1477 int 1478 VFS_MOUNT(struct mount *mp, const char *a, void *b, size_t *c) 1479 { 1480 int mpsafe = mp->mnt_iflag & IMNT_MPSAFE; 1481 int error; 1482 1483 /* 1484 * Note: The first time through, the vfs_mount function may set 1485 * IMNT_MPSAFE, so we have to cache it on entry in order to 1486 * avoid leaking a kernel lock. 1487 * 1488 * XXX Maybe the MPSAFE bit should be set in struct vfsops and 1489 * not in struct mount. 1490 */ 1491 if (mpsafe) { 1492 KERNEL_LOCK(1, NULL); 1493 } 1494 error = (*(mp->mnt_op->vfs_mount))(mp, a, b, c); 1495 if (mpsafe) { 1496 KERNEL_UNLOCK_ONE(NULL); 1497 } 1498 1499 return error; 1500 } 1501 1502 int 1503 VFS_START(struct mount *mp, int a) 1504 { 1505 int error; 1506 1507 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1508 KERNEL_LOCK(1, NULL); 1509 } 1510 error = (*(mp->mnt_op->vfs_start))(mp, a); 1511 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1512 KERNEL_UNLOCK_ONE(NULL); 1513 } 1514 1515 return error; 1516 } 1517 1518 int 1519 VFS_UNMOUNT(struct mount *mp, int a) 1520 { 1521 int error; 1522 1523 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1524 KERNEL_LOCK(1, NULL); 1525 } 1526 error = (*(mp->mnt_op->vfs_unmount))(mp, a); 1527 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1528 KERNEL_UNLOCK_ONE(NULL); 1529 } 1530 1531 return error; 1532 } 1533 1534 int 1535 VFS_ROOT(struct mount *mp, int lktype, struct vnode **a) 1536 { 1537 int error; 1538 1539 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1540 KERNEL_LOCK(1, NULL); 1541 } 1542 error = (*(mp->mnt_op->vfs_root))(mp, lktype, a); 1543 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1544 KERNEL_UNLOCK_ONE(NULL); 1545 } 1546 1547 return error; 1548 } 1549 1550 int 1551 VFS_QUOTACTL(struct mount *mp, struct quotactl_args *args) 1552 { 1553 int error; 1554 1555 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1556 KERNEL_LOCK(1, NULL); 1557 } 1558 error = (*(mp->mnt_op->vfs_quotactl))(mp, args); 1559 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1560 KERNEL_UNLOCK_ONE(NULL); 1561 } 1562 1563 return error; 1564 } 1565 1566 int 1567 VFS_STATVFS(struct mount *mp, struct statvfs *a) 1568 { 1569 int error; 1570 1571 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1572 KERNEL_LOCK(1, NULL); 1573 } 1574 error = (*(mp->mnt_op->vfs_statvfs))(mp, a); 1575 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1576 KERNEL_UNLOCK_ONE(NULL); 1577 } 1578 1579 return error; 1580 } 1581 1582 int 1583 VFS_SYNC(struct mount *mp, int a, struct kauth_cred *b) 1584 { 1585 int error; 1586 1587 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1588 KERNEL_LOCK(1, NULL); 1589 } 1590 error = (*(mp->mnt_op->vfs_sync))(mp, a, b); 1591 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1592 KERNEL_UNLOCK_ONE(NULL); 1593 } 1594 1595 return error; 1596 } 1597 1598 int 1599 VFS_FHTOVP(struct mount *mp, struct fid *a, int b, struct vnode **c) 1600 { 1601 int error; 1602 1603 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1604 KERNEL_LOCK(1, NULL); 1605 } 1606 error = (*(mp->mnt_op->vfs_fhtovp))(mp, a, b, c); 1607 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1608 KERNEL_UNLOCK_ONE(NULL); 1609 } 1610 1611 return error; 1612 } 1613 1614 int 1615 VFS_VPTOFH(struct vnode *vp, struct fid *a, size_t *b) 1616 { 1617 int error; 1618 1619 if ((vp->v_vflag & VV_MPSAFE) == 0) { 1620 KERNEL_LOCK(1, NULL); 1621 } 1622 error = (*(vp->v_mount->mnt_op->vfs_vptofh))(vp, a, b); 1623 if ((vp->v_vflag & VV_MPSAFE) == 0) { 1624 KERNEL_UNLOCK_ONE(NULL); 1625 } 1626 1627 return error; 1628 } 1629 1630 int 1631 VFS_SNAPSHOT(struct mount *mp, struct vnode *a, struct timespec *b) 1632 { 1633 int error; 1634 1635 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1636 KERNEL_LOCK(1, NULL); 1637 } 1638 error = (*(mp->mnt_op->vfs_snapshot))(mp, a, b); 1639 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1640 KERNEL_UNLOCK_ONE(NULL); 1641 } 1642 1643 return error; 1644 } 1645 1646 int 1647 VFS_EXTATTRCTL(struct mount *mp, int a, struct vnode *b, int c, const char *d) 1648 { 1649 int error; 1650 1651 KERNEL_LOCK(1, NULL); /* XXXSMP check ffs */ 1652 error = (*(mp->mnt_op->vfs_extattrctl))(mp, a, b, c, d); 1653 KERNEL_UNLOCK_ONE(NULL); /* XXX */ 1654 1655 return error; 1656 } 1657 1658 int 1659 VFS_SUSPENDCTL(struct mount *mp, int a) 1660 { 1661 int error; 1662 1663 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1664 KERNEL_LOCK(1, NULL); 1665 } 1666 error = (*(mp->mnt_op->vfs_suspendctl))(mp, a); 1667 if ((mp->mnt_iflag & IMNT_MPSAFE) == 0) { 1668 KERNEL_UNLOCK_ONE(NULL); 1669 } 1670 1671 return error; 1672 } 1673 1674 #if defined(DDB) || defined(DEBUGPRINT) 1675 static const char buf_flagbits[] = BUF_FLAGBITS; 1676 1677 void 1678 vfs_buf_print(struct buf *bp, int full, void (*pr)(const char *, ...)) 1679 { 1680 char bf[1024]; 1681 1682 (*pr)(" vp %p lblkno 0x%"PRIx64" blkno 0x%"PRIx64" rawblkno 0x%" 1683 PRIx64 " dev 0x%x\n", 1684 bp->b_vp, bp->b_lblkno, bp->b_blkno, bp->b_rawblkno, bp->b_dev); 1685 1686 snprintb(bf, sizeof(bf), 1687 buf_flagbits, bp->b_flags | bp->b_oflags | bp->b_cflags); 1688 (*pr)(" error %d flags %s\n", bp->b_error, bf); 1689 1690 (*pr)(" bufsize 0x%lx bcount 0x%lx resid 0x%lx\n", 1691 bp->b_bufsize, bp->b_bcount, bp->b_resid); 1692 (*pr)(" data %p saveaddr %p\n", 1693 bp->b_data, bp->b_saveaddr); 1694 (*pr)(" iodone %p objlock %p\n", bp->b_iodone, bp->b_objlock); 1695 } 1696 1697 void 1698 vfs_vnode_print(struct vnode *vp, int full, void (*pr)(const char *, ...)) 1699 { 1700 1701 uvm_object_printit(&vp->v_uobj, full, pr); 1702 (*pr)("\n"); 1703 vprint_common(vp, "", pr); 1704 if (full) { 1705 struct buf *bp; 1706 1707 (*pr)("clean bufs:\n"); 1708 LIST_FOREACH(bp, &vp->v_cleanblkhd, b_vnbufs) { 1709 (*pr)(" bp %p\n", bp); 1710 vfs_buf_print(bp, full, pr); 1711 } 1712 1713 (*pr)("dirty bufs:\n"); 1714 LIST_FOREACH(bp, &vp->v_dirtyblkhd, b_vnbufs) { 1715 (*pr)(" bp %p\n", bp); 1716 vfs_buf_print(bp, full, pr); 1717 } 1718 } 1719 } 1720 1721 void 1722 vfs_vnode_lock_print(void *vlock, int full, void (*pr)(const char *, ...)) 1723 { 1724 struct mount *mp; 1725 vnode_impl_t *vip; 1726 1727 for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp)) { 1728 TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) { 1729 if (&vip->vi_lock == vlock || 1730 VIMPL_TO_VNODE(vip)->v_interlock == vlock) 1731 vfs_vnode_print(VIMPL_TO_VNODE(vip), full, pr); 1732 } 1733 } 1734 } 1735 1736 void 1737 vfs_mount_print_all(int full, void (*pr)(const char *, ...)) 1738 { 1739 struct mount *mp; 1740 for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp)) 1741 vfs_mount_print(mp, full, pr); 1742 } 1743 1744 void 1745 vfs_mount_print(struct mount *mp, int full, void (*pr)(const char *, ...)) 1746 { 1747 char sbuf[256]; 1748 1749 (*pr)("vnodecovered = %p data = %p\n", 1750 mp->mnt_vnodecovered, mp->mnt_data); 1751 1752 (*pr)("fs_bshift %d dev_bshift = %d\n", 1753 mp->mnt_fs_bshift, mp->mnt_dev_bshift); 1754 1755 snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_flag); 1756 (*pr)("flag = %s\n", sbuf); 1757 1758 snprintb(sbuf, sizeof(sbuf), __IMNT_FLAG_BITS, mp->mnt_iflag); 1759 (*pr)("iflag = %s\n", sbuf); 1760 1761 (*pr)("refcnt = %d updating @ %p\n", mp->mnt_refcnt, mp->mnt_updating); 1762 1763 (*pr)("statvfs cache:\n"); 1764 (*pr)("\tbsize = %lu\n", mp->mnt_stat.f_bsize); 1765 (*pr)("\tfrsize = %lu\n", mp->mnt_stat.f_frsize); 1766 (*pr)("\tiosize = %lu\n", mp->mnt_stat.f_iosize); 1767 1768 (*pr)("\tblocks = %"PRIu64"\n", mp->mnt_stat.f_blocks); 1769 (*pr)("\tbfree = %"PRIu64"\n", mp->mnt_stat.f_bfree); 1770 (*pr)("\tbavail = %"PRIu64"\n", mp->mnt_stat.f_bavail); 1771 (*pr)("\tbresvd = %"PRIu64"\n", mp->mnt_stat.f_bresvd); 1772 1773 (*pr)("\tfiles = %"PRIu64"\n", mp->mnt_stat.f_files); 1774 (*pr)("\tffree = %"PRIu64"\n", mp->mnt_stat.f_ffree); 1775 (*pr)("\tfavail = %"PRIu64"\n", mp->mnt_stat.f_favail); 1776 (*pr)("\tfresvd = %"PRIu64"\n", mp->mnt_stat.f_fresvd); 1777 1778 (*pr)("\tf_fsidx = { 0x%"PRIx32", 0x%"PRIx32" }\n", 1779 mp->mnt_stat.f_fsidx.__fsid_val[0], 1780 mp->mnt_stat.f_fsidx.__fsid_val[1]); 1781 1782 (*pr)("\towner = %"PRIu32"\n", mp->mnt_stat.f_owner); 1783 (*pr)("\tnamemax = %lu\n", mp->mnt_stat.f_namemax); 1784 1785 snprintb(sbuf, sizeof(sbuf), __MNT_FLAG_BITS, mp->mnt_stat.f_flag); 1786 1787 (*pr)("\tflag = %s\n", sbuf); 1788 (*pr)("\tsyncwrites = %" PRIu64 "\n", mp->mnt_stat.f_syncwrites); 1789 (*pr)("\tasyncwrites = %" PRIu64 "\n", mp->mnt_stat.f_asyncwrites); 1790 (*pr)("\tsyncreads = %" PRIu64 "\n", mp->mnt_stat.f_syncreads); 1791 (*pr)("\tasyncreads = %" PRIu64 "\n", mp->mnt_stat.f_asyncreads); 1792 (*pr)("\tfstypename = %s\n", mp->mnt_stat.f_fstypename); 1793 (*pr)("\tmntonname = %s\n", mp->mnt_stat.f_mntonname); 1794 (*pr)("\tmntfromname = %s\n", mp->mnt_stat.f_mntfromname); 1795 1796 { 1797 int cnt = 0; 1798 vnode_t *vp; 1799 vnode_impl_t *vip; 1800 (*pr)("locked vnodes ="); 1801 TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) { 1802 vp = VIMPL_TO_VNODE(vip); 1803 if (VOP_ISLOCKED(vp)) { 1804 if ((++cnt % 6) == 0) { 1805 (*pr)(" %p,\n\t", vp); 1806 } else { 1807 (*pr)(" %p,", vp); 1808 } 1809 } 1810 } 1811 (*pr)("\n"); 1812 } 1813 1814 if (full) { 1815 int cnt = 0; 1816 vnode_t *vp; 1817 vnode_impl_t *vip; 1818 1819 (*pr)("all vnodes ="); 1820 TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) { 1821 vp = VIMPL_TO_VNODE(vip); 1822 if (!TAILQ_NEXT(vip, vi_mntvnodes)) { 1823 (*pr)(" %p", vp); 1824 } else if ((++cnt % 6) == 0) { 1825 (*pr)(" %p,\n\t", vp); 1826 } else { 1827 (*pr)(" %p,", vp); 1828 } 1829 } 1830 (*pr)("\n"); 1831 } 1832 } 1833 1834 /* 1835 * List all of the locked vnodes in the system. 1836 */ 1837 void printlockedvnodes(void); 1838 1839 void 1840 printlockedvnodes(void) 1841 { 1842 struct mount *mp; 1843 vnode_t *vp; 1844 vnode_impl_t *vip; 1845 1846 printf("Locked vnodes\n"); 1847 for (mp = _mountlist_next(NULL); mp; mp = _mountlist_next(mp)) { 1848 TAILQ_FOREACH(vip, &mp->mnt_vnodelist, vi_mntvnodes) { 1849 vp = VIMPL_TO_VNODE(vip); 1850 if (VOP_ISLOCKED(vp)) 1851 vprint(NULL, vp); 1852 } 1853 } 1854 } 1855 1856 #endif /* DDB || DEBUGPRINT */ 1857