1 /*- 2 * Copyright (c) 1989, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 35 */ 36 37 /* 38 * External virtual filesystem routines 39 */ 40 41 #include <sys/cdefs.h> 42 __FBSDID("$FreeBSD$"); 43 44 #include "opt_ddb.h" 45 46 #include <sys/param.h> 47 #include <sys/systm.h> 48 #include <sys/bio.h> 49 #include <sys/buf.h> 50 #include <sys/condvar.h> 51 #include <sys/conf.h> 52 #include <sys/dirent.h> 53 #include <sys/event.h> 54 #include <sys/eventhandler.h> 55 #include <sys/extattr.h> 56 #include <sys/file.h> 57 #include <sys/fcntl.h> 58 #include <sys/jail.h> 59 #include <sys/kdb.h> 60 #include <sys/kernel.h> 61 #include <sys/kthread.h> 62 #include <sys/lockf.h> 63 #include <sys/malloc.h> 64 #include <sys/mount.h> 65 #include <sys/namei.h> 66 #include <sys/priv.h> 67 #include <sys/reboot.h> 68 #include <sys/sched.h> 69 #include <sys/sleepqueue.h> 70 #include <sys/stat.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/vmmeter.h> 74 #include <sys/vnode.h> 75 76 #include <machine/stdarg.h> 77 78 #include <security/mac/mac_framework.h> 79 80 #include <vm/vm.h> 81 #include <vm/vm_object.h> 82 #include <vm/vm_extern.h> 83 #include <vm/pmap.h> 84 #include <vm/vm_map.h> 85 #include <vm/vm_page.h> 86 #include <vm/vm_kern.h> 87 #include <vm/uma.h> 88 89 #ifdef DDB 90 #include <ddb/ddb.h> 91 #endif 92 93 #define WI_MPSAFEQ 0 94 #define WI_GIANTQ 1 95 96 static MALLOC_DEFINE(M_NETADDR, "subr_export_host", "Export host address structure"); 97 98 static void delmntque(struct vnode *vp); 99 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, 100 int slpflag, int slptimeo); 101 static void syncer_shutdown(void *arg, int howto); 102 static int vtryrecycle(struct vnode *vp); 103 static void vbusy(struct vnode *vp); 104 static void vinactive(struct vnode *, struct thread *); 105 static void v_incr_usecount(struct vnode *); 106 static void v_decr_usecount(struct vnode *); 107 static void v_decr_useonly(struct vnode *); 108 static void v_upgrade_usecount(struct vnode *); 109 static void vfree(struct vnode *); 110 static void vnlru_free(int); 111 static void vgonel(struct vnode *); 112 static void vfs_knllock(void *arg); 113 static void vfs_knlunlock(void *arg); 114 static void vfs_knl_assert_locked(void *arg); 115 static void vfs_knl_assert_unlocked(void *arg); 116 static void destroy_vpollinfo(struct vpollinfo *vi); 117 118 /* 119 * Number of vnodes in existence. Increased whenever getnewvnode() 120 * allocates a new vnode, decreased on vdestroy() called on VI_DOOMed 121 * vnode. 122 */ 123 static unsigned long numvnodes; 124 125 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, 126 "Number of vnodes in existence"); 127 128 /* 129 * Conversion tables for conversion from vnode types to inode formats 130 * and back. 131 */ 132 enum vtype iftovt_tab[16] = { 133 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 134 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 135 }; 136 int vttoif_tab[10] = { 137 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 138 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT 139 }; 140 141 /* 142 * List of vnodes that are ready for recycling. 143 */ 144 static TAILQ_HEAD(freelst, vnode) vnode_free_list; 145 146 /* 147 * Free vnode target. Free vnodes may simply be files which have been stat'd 148 * but not read. This is somewhat common, and a small cache of such files 149 * should be kept to avoid recreation costs. 150 */ 151 static u_long wantfreevnodes; 152 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 153 /* Number of vnodes in the free list. */ 154 static u_long freevnodes; 155 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, 156 "Number of vnodes in the free list"); 157 158 static int vlru_allow_cache_src; 159 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW, 160 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode"); 161 162 /* 163 * Various variables used for debugging the new implementation of 164 * reassignbuf(). 165 * XXX these are probably of (very) limited utility now. 166 */ 167 static int reassignbufcalls; 168 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, 169 "Number of calls to reassignbuf"); 170 171 /* 172 * Cache for the mount type id assigned to NFS. This is used for 173 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. 174 */ 175 int nfs_mount_type = -1; 176 177 /* To keep more than one thread at a time from running vfs_getnewfsid */ 178 static struct mtx mntid_mtx; 179 180 /* 181 * Lock for any access to the following: 182 * vnode_free_list 183 * numvnodes 184 * freevnodes 185 */ 186 static struct mtx vnode_free_list_mtx; 187 188 /* Publicly exported FS */ 189 struct nfs_public nfs_pub; 190 191 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ 192 static uma_zone_t vnode_zone; 193 static uma_zone_t vnodepoll_zone; 194 195 /* 196 * The workitem queue. 197 * 198 * It is useful to delay writes of file data and filesystem metadata 199 * for tens of seconds so that quickly created and deleted files need 200 * not waste disk bandwidth being created and removed. To realize this, 201 * we append vnodes to a "workitem" queue. When running with a soft 202 * updates implementation, most pending metadata dependencies should 203 * not wait for more than a few seconds. Thus, mounted on block devices 204 * are delayed only about a half the time that file data is delayed. 205 * Similarly, directory updates are more critical, so are only delayed 206 * about a third the time that file data is delayed. Thus, there are 207 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 208 * one each second (driven off the filesystem syncer process). The 209 * syncer_delayno variable indicates the next queue that is to be processed. 210 * Items that need to be processed soon are placed in this queue: 211 * 212 * syncer_workitem_pending[syncer_delayno] 213 * 214 * A delay of fifteen seconds is done by placing the request fifteen 215 * entries later in the queue: 216 * 217 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 218 * 219 */ 220 static int syncer_delayno; 221 static long syncer_mask; 222 LIST_HEAD(synclist, bufobj); 223 static struct synclist *syncer_workitem_pending[2]; 224 /* 225 * The sync_mtx protects: 226 * bo->bo_synclist 227 * sync_vnode_count 228 * syncer_delayno 229 * syncer_state 230 * syncer_workitem_pending 231 * syncer_worklist_len 232 * rushjob 233 */ 234 static struct mtx sync_mtx; 235 static struct cv sync_wakeup; 236 237 #define SYNCER_MAXDELAY 32 238 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 239 static int syncdelay = 30; /* max time to delay syncing data */ 240 static int filedelay = 30; /* time to delay syncing files */ 241 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, 242 "Time to delay syncing files (in seconds)"); 243 static int dirdelay = 29; /* time to delay syncing directories */ 244 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, 245 "Time to delay syncing directories (in seconds)"); 246 static int metadelay = 28; /* time to delay syncing metadata */ 247 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, 248 "Time to delay syncing metadata (in seconds)"); 249 static int rushjob; /* number of slots to run ASAP */ 250 static int stat_rush_requests; /* number of times I/O speeded up */ 251 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, 252 "Number of times I/O speeded up (rush requests)"); 253 254 /* 255 * When shutting down the syncer, run it at four times normal speed. 256 */ 257 #define SYNCER_SHUTDOWN_SPEEDUP 4 258 static int sync_vnode_count; 259 static int syncer_worklist_len; 260 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } 261 syncer_state; 262 263 /* 264 * Number of vnodes we want to exist at any one time. This is mostly used 265 * to size hash tables in vnode-related code. It is normally not used in 266 * getnewvnode(), as wantfreevnodes is normally nonzero.) 267 * 268 * XXX desiredvnodes is historical cruft and should not exist. 269 */ 270 int desiredvnodes; 271 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 272 &desiredvnodes, 0, "Maximum number of vnodes"); 273 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 274 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)"); 275 static int vnlru_nowhere; 276 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, 277 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); 278 279 /* 280 * Macros to control when a vnode is freed and recycled. All require 281 * the vnode interlock. 282 */ 283 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 284 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 285 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt) 286 287 288 /* 289 * Initialize the vnode management data structures. 290 * 291 * Reevaluate the following cap on the number of vnodes after the physical 292 * memory size exceeds 512GB. In the limit, as the physical memory size 293 * grows, the ratio of physical pages to vnodes approaches sixteen to one. 294 */ 295 #ifndef MAXVNODES_MAX 296 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16)) 297 #endif 298 static void 299 vntblinit(void *dummy __unused) 300 { 301 int physvnodes, virtvnodes; 302 303 /* 304 * Desiredvnodes is a function of the physical memory size and the 305 * kernel's heap size. Generally speaking, it scales with the 306 * physical memory size. The ratio of desiredvnodes to physical pages 307 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the 308 * marginal ratio of desiredvnodes to physical pages is one to 309 * sixteen. However, desiredvnodes is limited by the kernel's heap 310 * size. The memory required by desiredvnodes vnodes and vm objects 311 * may not exceed one seventh of the kernel's heap size. 312 */ 313 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4, 314 cnt.v_page_count) / 16; 315 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) + 316 sizeof(struct vnode))); 317 desiredvnodes = min(physvnodes, virtvnodes); 318 if (desiredvnodes > MAXVNODES_MAX) { 319 if (bootverbose) 320 printf("Reducing kern.maxvnodes %d -> %d\n", 321 desiredvnodes, MAXVNODES_MAX); 322 desiredvnodes = MAXVNODES_MAX; 323 } 324 wantfreevnodes = desiredvnodes / 4; 325 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); 326 TAILQ_INIT(&vnode_free_list); 327 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); 328 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, 329 NULL, NULL, UMA_ALIGN_PTR, 0); 330 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), 331 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 332 /* 333 * Initialize the filesystem syncer. 334 */ 335 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE, 336 &syncer_mask); 337 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE, 338 &syncer_mask); 339 syncer_maxdelay = syncer_mask + 1; 340 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); 341 cv_init(&sync_wakeup, "syncer"); 342 } 343 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); 344 345 346 /* 347 * Mark a mount point as busy. Used to synchronize access and to delay 348 * unmounting. Eventually, mountlist_mtx is not released on failure. 349 */ 350 int 351 vfs_busy(struct mount *mp, int flags) 352 { 353 354 MPASS((flags & ~MBF_MASK) == 0); 355 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); 356 357 MNT_ILOCK(mp); 358 MNT_REF(mp); 359 /* 360 * If mount point is currenly being unmounted, sleep until the 361 * mount point fate is decided. If thread doing the unmounting fails, 362 * it will clear MNTK_UNMOUNT flag before waking us up, indicating 363 * that this mount point has survived the unmount attempt and vfs_busy 364 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE 365 * flag in addition to MNTK_UNMOUNT, indicating that mount point is 366 * about to be really destroyed. vfs_busy needs to release its 367 * reference on the mount point in this case and return with ENOENT, 368 * telling the caller that mount mount it tried to busy is no longer 369 * valid. 370 */ 371 while (mp->mnt_kern_flag & MNTK_UNMOUNT) { 372 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { 373 MNT_REL(mp); 374 MNT_IUNLOCK(mp); 375 CTR1(KTR_VFS, "%s: failed busying before sleeping", 376 __func__); 377 return (ENOENT); 378 } 379 if (flags & MBF_MNTLSTLOCK) 380 mtx_unlock(&mountlist_mtx); 381 mp->mnt_kern_flag |= MNTK_MWAIT; 382 msleep(mp, MNT_MTX(mp), PVFS, "vfs_busy", 0); 383 if (flags & MBF_MNTLSTLOCK) 384 mtx_lock(&mountlist_mtx); 385 } 386 if (flags & MBF_MNTLSTLOCK) 387 mtx_unlock(&mountlist_mtx); 388 mp->mnt_lockref++; 389 MNT_IUNLOCK(mp); 390 return (0); 391 } 392 393 /* 394 * Free a busy filesystem. 395 */ 396 void 397 vfs_unbusy(struct mount *mp) 398 { 399 400 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 401 MNT_ILOCK(mp); 402 MNT_REL(mp); 403 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref")); 404 mp->mnt_lockref--; 405 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { 406 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); 407 CTR1(KTR_VFS, "%s: waking up waiters", __func__); 408 mp->mnt_kern_flag &= ~MNTK_DRAINING; 409 wakeup(&mp->mnt_lockref); 410 } 411 MNT_IUNLOCK(mp); 412 } 413 414 /* 415 * Lookup a mount point by filesystem identifier. 416 */ 417 struct mount * 418 vfs_getvfs(fsid_t *fsid) 419 { 420 struct mount *mp; 421 422 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 423 mtx_lock(&mountlist_mtx); 424 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 425 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 426 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 427 vfs_ref(mp); 428 mtx_unlock(&mountlist_mtx); 429 return (mp); 430 } 431 } 432 mtx_unlock(&mountlist_mtx); 433 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 434 return ((struct mount *) 0); 435 } 436 437 /* 438 * Lookup a mount point by filesystem identifier, busying it before 439 * returning. 440 */ 441 struct mount * 442 vfs_busyfs(fsid_t *fsid) 443 { 444 struct mount *mp; 445 int error; 446 447 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 448 mtx_lock(&mountlist_mtx); 449 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 450 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 451 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 452 error = vfs_busy(mp, MBF_MNTLSTLOCK); 453 if (error) { 454 mtx_unlock(&mountlist_mtx); 455 return (NULL); 456 } 457 return (mp); 458 } 459 } 460 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 461 mtx_unlock(&mountlist_mtx); 462 return ((struct mount *) 0); 463 } 464 465 /* 466 * Check if a user can access privileged mount options. 467 */ 468 int 469 vfs_suser(struct mount *mp, struct thread *td) 470 { 471 int error; 472 473 /* 474 * If the thread is jailed, but this is not a jail-friendly file 475 * system, deny immediately. 476 */ 477 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred)) 478 return (EPERM); 479 480 /* 481 * If the file system was mounted outside the jail of the calling 482 * thread, deny immediately. 483 */ 484 if (prison_check(td->td_ucred, mp->mnt_cred) != 0) 485 return (EPERM); 486 487 /* 488 * If file system supports delegated administration, we don't check 489 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified 490 * by the file system itself. 491 * If this is not the user that did original mount, we check for 492 * the PRIV_VFS_MOUNT_OWNER privilege. 493 */ 494 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && 495 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { 496 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) 497 return (error); 498 } 499 return (0); 500 } 501 502 /* 503 * Get a new unique fsid. Try to make its val[0] unique, since this value 504 * will be used to create fake device numbers for stat(). Also try (but 505 * not so hard) make its val[0] unique mod 2^16, since some emulators only 506 * support 16-bit device numbers. We end up with unique val[0]'s for the 507 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 508 * 509 * Keep in mind that several mounts may be running in parallel. Starting 510 * the search one past where the previous search terminated is both a 511 * micro-optimization and a defense against returning the same fsid to 512 * different mounts. 513 */ 514 void 515 vfs_getnewfsid(struct mount *mp) 516 { 517 static uint16_t mntid_base; 518 struct mount *nmp; 519 fsid_t tfsid; 520 int mtype; 521 522 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 523 mtx_lock(&mntid_mtx); 524 mtype = mp->mnt_vfc->vfc_typenum; 525 tfsid.val[1] = mtype; 526 mtype = (mtype & 0xFF) << 24; 527 for (;;) { 528 tfsid.val[0] = makedev(255, 529 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 530 mntid_base++; 531 if ((nmp = vfs_getvfs(&tfsid)) == NULL) 532 break; 533 vfs_rel(nmp); 534 } 535 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 536 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 537 mtx_unlock(&mntid_mtx); 538 } 539 540 /* 541 * Knob to control the precision of file timestamps: 542 * 543 * 0 = seconds only; nanoseconds zeroed. 544 * 1 = seconds and nanoseconds, accurate within 1/HZ. 545 * 2 = seconds and nanoseconds, truncated to microseconds. 546 * >=3 = seconds and nanoseconds, maximum precision. 547 */ 548 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 549 550 static int timestamp_precision = TSP_SEC; 551 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 552 ×tamp_precision, 0, "File timestamp precision (0: seconds, " 553 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, " 554 "3+: sec + ns (max. precision))"); 555 556 /* 557 * Get a current timestamp. 558 */ 559 void 560 vfs_timestamp(struct timespec *tsp) 561 { 562 struct timeval tv; 563 564 switch (timestamp_precision) { 565 case TSP_SEC: 566 tsp->tv_sec = time_second; 567 tsp->tv_nsec = 0; 568 break; 569 case TSP_HZ: 570 getnanotime(tsp); 571 break; 572 case TSP_USEC: 573 microtime(&tv); 574 TIMEVAL_TO_TIMESPEC(&tv, tsp); 575 break; 576 case TSP_NSEC: 577 default: 578 nanotime(tsp); 579 break; 580 } 581 } 582 583 /* 584 * Set vnode attributes to VNOVAL 585 */ 586 void 587 vattr_null(struct vattr *vap) 588 { 589 590 vap->va_type = VNON; 591 vap->va_size = VNOVAL; 592 vap->va_bytes = VNOVAL; 593 vap->va_mode = VNOVAL; 594 vap->va_nlink = VNOVAL; 595 vap->va_uid = VNOVAL; 596 vap->va_gid = VNOVAL; 597 vap->va_fsid = VNOVAL; 598 vap->va_fileid = VNOVAL; 599 vap->va_blocksize = VNOVAL; 600 vap->va_rdev = VNOVAL; 601 vap->va_atime.tv_sec = VNOVAL; 602 vap->va_atime.tv_nsec = VNOVAL; 603 vap->va_mtime.tv_sec = VNOVAL; 604 vap->va_mtime.tv_nsec = VNOVAL; 605 vap->va_ctime.tv_sec = VNOVAL; 606 vap->va_ctime.tv_nsec = VNOVAL; 607 vap->va_birthtime.tv_sec = VNOVAL; 608 vap->va_birthtime.tv_nsec = VNOVAL; 609 vap->va_flags = VNOVAL; 610 vap->va_gen = VNOVAL; 611 vap->va_vaflags = 0; 612 } 613 614 /* 615 * This routine is called when we have too many vnodes. It attempts 616 * to free <count> vnodes and will potentially free vnodes that still 617 * have VM backing store (VM backing store is typically the cause 618 * of a vnode blowout so we want to do this). Therefore, this operation 619 * is not considered cheap. 620 * 621 * A number of conditions may prevent a vnode from being reclaimed. 622 * the buffer cache may have references on the vnode, a directory 623 * vnode may still have references due to the namei cache representing 624 * underlying files, or the vnode may be in active use. It is not 625 * desireable to reuse such vnodes. These conditions may cause the 626 * number of vnodes to reach some minimum value regardless of what 627 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 628 */ 629 static int 630 vlrureclaim(struct mount *mp) 631 { 632 struct vnode *vp; 633 int done; 634 int trigger; 635 int usevnodes; 636 int count; 637 638 /* 639 * Calculate the trigger point, don't allow user 640 * screwups to blow us up. This prevents us from 641 * recycling vnodes with lots of resident pages. We 642 * aren't trying to free memory, we are trying to 643 * free vnodes. 644 */ 645 usevnodes = desiredvnodes; 646 if (usevnodes <= 0) 647 usevnodes = 1; 648 trigger = cnt.v_page_count * 2 / usevnodes; 649 done = 0; 650 vn_start_write(NULL, &mp, V_WAIT); 651 MNT_ILOCK(mp); 652 count = mp->mnt_nvnodelistsize / 10 + 1; 653 while (count != 0) { 654 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 655 while (vp != NULL && vp->v_type == VMARKER) 656 vp = TAILQ_NEXT(vp, v_nmntvnodes); 657 if (vp == NULL) 658 break; 659 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 660 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 661 --count; 662 if (!VI_TRYLOCK(vp)) 663 goto next_iter; 664 /* 665 * If it's been deconstructed already, it's still 666 * referenced, or it exceeds the trigger, skip it. 667 */ 668 if (vp->v_usecount || 669 (!vlru_allow_cache_src && 670 !LIST_EMPTY(&(vp)->v_cache_src)) || 671 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL && 672 vp->v_object->resident_page_count > trigger)) { 673 VI_UNLOCK(vp); 674 goto next_iter; 675 } 676 MNT_IUNLOCK(mp); 677 vholdl(vp); 678 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) { 679 vdrop(vp); 680 goto next_iter_mntunlocked; 681 } 682 VI_LOCK(vp); 683 /* 684 * v_usecount may have been bumped after VOP_LOCK() dropped 685 * the vnode interlock and before it was locked again. 686 * 687 * It is not necessary to recheck VI_DOOMED because it can 688 * only be set by another thread that holds both the vnode 689 * lock and vnode interlock. If another thread has the 690 * vnode lock before we get to VOP_LOCK() and obtains the 691 * vnode interlock after VOP_LOCK() drops the vnode 692 * interlock, the other thread will be unable to drop the 693 * vnode lock before our VOP_LOCK() call fails. 694 */ 695 if (vp->v_usecount || 696 (!vlru_allow_cache_src && 697 !LIST_EMPTY(&(vp)->v_cache_src)) || 698 (vp->v_object != NULL && 699 vp->v_object->resident_page_count > trigger)) { 700 VOP_UNLOCK(vp, LK_INTERLOCK); 701 goto next_iter_mntunlocked; 702 } 703 KASSERT((vp->v_iflag & VI_DOOMED) == 0, 704 ("VI_DOOMED unexpectedly detected in vlrureclaim()")); 705 vgonel(vp); 706 VOP_UNLOCK(vp, 0); 707 vdropl(vp); 708 done++; 709 next_iter_mntunlocked: 710 if (!should_yield()) 711 goto relock_mnt; 712 goto yield; 713 next_iter: 714 if (!should_yield()) 715 continue; 716 MNT_IUNLOCK(mp); 717 yield: 718 kern_yield(-1); 719 relock_mnt: 720 MNT_ILOCK(mp); 721 } 722 MNT_IUNLOCK(mp); 723 vn_finished_write(mp); 724 return done; 725 } 726 727 /* 728 * Attempt to keep the free list at wantfreevnodes length. 729 */ 730 static void 731 vnlru_free(int count) 732 { 733 struct vnode *vp; 734 int vfslocked; 735 736 mtx_assert(&vnode_free_list_mtx, MA_OWNED); 737 for (; count > 0; count--) { 738 vp = TAILQ_FIRST(&vnode_free_list); 739 /* 740 * The list can be modified while the free_list_mtx 741 * has been dropped and vp could be NULL here. 742 */ 743 if (!vp) 744 break; 745 VNASSERT(vp->v_op != NULL, vp, 746 ("vnlru_free: vnode already reclaimed.")); 747 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 748 /* 749 * Don't recycle if we can't get the interlock. 750 */ 751 if (!VI_TRYLOCK(vp)) { 752 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 753 continue; 754 } 755 VNASSERT(VCANRECYCLE(vp), vp, 756 ("vp inconsistent on freelist")); 757 freevnodes--; 758 vp->v_iflag &= ~VI_FREE; 759 vholdl(vp); 760 mtx_unlock(&vnode_free_list_mtx); 761 VI_UNLOCK(vp); 762 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 763 vtryrecycle(vp); 764 VFS_UNLOCK_GIANT(vfslocked); 765 /* 766 * If the recycled succeeded this vdrop will actually free 767 * the vnode. If not it will simply place it back on 768 * the free list. 769 */ 770 vdrop(vp); 771 mtx_lock(&vnode_free_list_mtx); 772 } 773 } 774 /* 775 * Attempt to recycle vnodes in a context that is always safe to block. 776 * Calling vlrurecycle() from the bowels of filesystem code has some 777 * interesting deadlock problems. 778 */ 779 static struct proc *vnlruproc; 780 static int vnlruproc_sig; 781 782 static void 783 vnlru_proc(void) 784 { 785 struct mount *mp, *nmp; 786 int done, vfslocked; 787 struct proc *p = vnlruproc; 788 789 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p, 790 SHUTDOWN_PRI_FIRST); 791 792 for (;;) { 793 kproc_suspend_check(p); 794 mtx_lock(&vnode_free_list_mtx); 795 if (freevnodes > wantfreevnodes) 796 vnlru_free(freevnodes - wantfreevnodes); 797 if (numvnodes <= desiredvnodes * 9 / 10) { 798 vnlruproc_sig = 0; 799 wakeup(&vnlruproc_sig); 800 msleep(vnlruproc, &vnode_free_list_mtx, 801 PVFS|PDROP, "vlruwt", hz); 802 continue; 803 } 804 mtx_unlock(&vnode_free_list_mtx); 805 done = 0; 806 mtx_lock(&mountlist_mtx); 807 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 808 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { 809 nmp = TAILQ_NEXT(mp, mnt_list); 810 continue; 811 } 812 vfslocked = VFS_LOCK_GIANT(mp); 813 done += vlrureclaim(mp); 814 VFS_UNLOCK_GIANT(vfslocked); 815 mtx_lock(&mountlist_mtx); 816 nmp = TAILQ_NEXT(mp, mnt_list); 817 vfs_unbusy(mp); 818 } 819 mtx_unlock(&mountlist_mtx); 820 if (done == 0) { 821 #if 0 822 /* These messages are temporary debugging aids */ 823 if (vnlru_nowhere < 5) 824 printf("vnlru process getting nowhere..\n"); 825 else if (vnlru_nowhere == 5) 826 printf("vnlru process messages stopped.\n"); 827 #endif 828 vnlru_nowhere++; 829 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 830 } else 831 kern_yield(-1); 832 } 833 } 834 835 static struct kproc_desc vnlru_kp = { 836 "vnlru", 837 vnlru_proc, 838 &vnlruproc 839 }; 840 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, 841 &vnlru_kp); 842 843 /* 844 * Routines having to do with the management of the vnode table. 845 */ 846 847 void 848 vdestroy(struct vnode *vp) 849 { 850 struct bufobj *bo; 851 852 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 853 mtx_lock(&vnode_free_list_mtx); 854 numvnodes--; 855 mtx_unlock(&vnode_free_list_mtx); 856 bo = &vp->v_bufobj; 857 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, 858 ("cleaned vnode still on the free list.")); 859 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); 860 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); 861 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); 862 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); 863 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); 864 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); 865 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL")); 866 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); 867 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL")); 868 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); 869 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); 870 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); 871 VI_UNLOCK(vp); 872 #ifdef MAC 873 mac_vnode_destroy(vp); 874 #endif 875 if (vp->v_pollinfo != NULL) 876 destroy_vpollinfo(vp->v_pollinfo); 877 #ifdef INVARIANTS 878 /* XXX Elsewhere we can detect an already freed vnode via NULL v_op. */ 879 vp->v_op = NULL; 880 #endif 881 lockdestroy(vp->v_vnlock); 882 mtx_destroy(&vp->v_interlock); 883 mtx_destroy(BO_MTX(bo)); 884 uma_zfree(vnode_zone, vp); 885 } 886 887 /* 888 * Try to recycle a freed vnode. We abort if anyone picks up a reference 889 * before we actually vgone(). This function must be called with the vnode 890 * held to prevent the vnode from being returned to the free list midway 891 * through vgone(). 892 */ 893 static int 894 vtryrecycle(struct vnode *vp) 895 { 896 struct mount *vnmp; 897 898 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 899 VNASSERT(vp->v_holdcnt, vp, 900 ("vtryrecycle: Recycling vp %p without a reference.", vp)); 901 /* 902 * This vnode may found and locked via some other list, if so we 903 * can't recycle it yet. 904 */ 905 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 906 CTR2(KTR_VFS, 907 "%s: impossible to recycle, vp %p lock is already held", 908 __func__, vp); 909 return (EWOULDBLOCK); 910 } 911 /* 912 * Don't recycle if its filesystem is being suspended. 913 */ 914 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { 915 VOP_UNLOCK(vp, 0); 916 CTR2(KTR_VFS, 917 "%s: impossible to recycle, cannot start the write for %p", 918 __func__, vp); 919 return (EBUSY); 920 } 921 /* 922 * If we got this far, we need to acquire the interlock and see if 923 * anyone picked up this vnode from another list. If not, we will 924 * mark it with DOOMED via vgonel() so that anyone who does find it 925 * will skip over it. 926 */ 927 VI_LOCK(vp); 928 if (vp->v_usecount) { 929 VOP_UNLOCK(vp, LK_INTERLOCK); 930 vn_finished_write(vnmp); 931 CTR2(KTR_VFS, 932 "%s: impossible to recycle, %p is already referenced", 933 __func__, vp); 934 return (EBUSY); 935 } 936 if ((vp->v_iflag & VI_DOOMED) == 0) 937 vgonel(vp); 938 VOP_UNLOCK(vp, LK_INTERLOCK); 939 vn_finished_write(vnmp); 940 return (0); 941 } 942 943 /* 944 * Return the next vnode from the free list. 945 */ 946 int 947 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, 948 struct vnode **vpp) 949 { 950 struct vnode *vp = NULL; 951 struct bufobj *bo; 952 953 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); 954 mtx_lock(&vnode_free_list_mtx); 955 /* 956 * Lend our context to reclaim vnodes if they've exceeded the max. 957 */ 958 if (freevnodes > wantfreevnodes) 959 vnlru_free(1); 960 /* 961 * Wait for available vnodes. 962 */ 963 if (numvnodes > desiredvnodes) { 964 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) { 965 /* 966 * File system is beeing suspended, we cannot risk a 967 * deadlock here, so allocate new vnode anyway. 968 */ 969 if (freevnodes > wantfreevnodes) 970 vnlru_free(freevnodes - wantfreevnodes); 971 goto alloc; 972 } 973 if (vnlruproc_sig == 0) { 974 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 975 wakeup(vnlruproc); 976 } 977 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS, 978 "vlruwk", hz); 979 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */ 980 if (numvnodes > desiredvnodes) { 981 mtx_unlock(&vnode_free_list_mtx); 982 return (ENFILE); 983 } 984 #endif 985 } 986 alloc: 987 numvnodes++; 988 mtx_unlock(&vnode_free_list_mtx); 989 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO); 990 /* 991 * Setup locks. 992 */ 993 vp->v_vnlock = &vp->v_lock; 994 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); 995 /* 996 * By default, don't allow shared locks unless filesystems 997 * opt-in. 998 */ 999 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE); 1000 /* 1001 * Initialize bufobj. 1002 */ 1003 bo = &vp->v_bufobj; 1004 bo->__bo_vnode = vp; 1005 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF); 1006 bo->bo_ops = &buf_ops_bio; 1007 bo->bo_private = vp; 1008 TAILQ_INIT(&bo->bo_clean.bv_hd); 1009 TAILQ_INIT(&bo->bo_dirty.bv_hd); 1010 /* 1011 * Initialize namecache. 1012 */ 1013 LIST_INIT(&vp->v_cache_src); 1014 TAILQ_INIT(&vp->v_cache_dst); 1015 /* 1016 * Finalize various vnode identity bits. 1017 */ 1018 vp->v_type = VNON; 1019 vp->v_tag = tag; 1020 vp->v_op = vops; 1021 v_incr_usecount(vp); 1022 vp->v_data = 0; 1023 #ifdef MAC 1024 mac_vnode_init(vp); 1025 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) 1026 mac_vnode_associate_singlelabel(mp, vp); 1027 else if (mp == NULL && vops != &dead_vnodeops) 1028 printf("NULL mp in getnewvnode()\n"); 1029 #endif 1030 if (mp != NULL) { 1031 bo->bo_bsize = mp->mnt_stat.f_iosize; 1032 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) 1033 vp->v_vflag |= VV_NOKNOTE; 1034 } 1035 1036 *vpp = vp; 1037 return (0); 1038 } 1039 1040 /* 1041 * Delete from old mount point vnode list, if on one. 1042 */ 1043 static void 1044 delmntque(struct vnode *vp) 1045 { 1046 struct mount *mp; 1047 1048 mp = vp->v_mount; 1049 if (mp == NULL) 1050 return; 1051 MNT_ILOCK(mp); 1052 vp->v_mount = NULL; 1053 VNASSERT(mp->mnt_nvnodelistsize > 0, vp, 1054 ("bad mount point vnode list size")); 1055 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1056 mp->mnt_nvnodelistsize--; 1057 MNT_REL(mp); 1058 MNT_IUNLOCK(mp); 1059 } 1060 1061 static void 1062 insmntque_stddtr(struct vnode *vp, void *dtr_arg) 1063 { 1064 1065 vp->v_data = NULL; 1066 vp->v_op = &dead_vnodeops; 1067 /* XXX non mp-safe fs may still call insmntque with vnode 1068 unlocked */ 1069 if (!VOP_ISLOCKED(vp)) 1070 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1071 vgone(vp); 1072 vput(vp); 1073 } 1074 1075 /* 1076 * Insert into list of vnodes for the new mount point, if available. 1077 */ 1078 int 1079 insmntque1(struct vnode *vp, struct mount *mp, 1080 void (*dtr)(struct vnode *, void *), void *dtr_arg) 1081 { 1082 int locked; 1083 1084 KASSERT(vp->v_mount == NULL, 1085 ("insmntque: vnode already on per mount vnode list")); 1086 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); 1087 #ifdef DEBUG_VFS_LOCKS 1088 if (!VFS_NEEDSGIANT(mp)) 1089 ASSERT_VOP_ELOCKED(vp, 1090 "insmntque: mp-safe fs and non-locked vp"); 1091 #endif 1092 MNT_ILOCK(mp); 1093 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 && 1094 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || 1095 mp->mnt_nvnodelistsize == 0)) { 1096 locked = VOP_ISLOCKED(vp); 1097 if (!locked || (locked == LK_EXCLUSIVE && 1098 (vp->v_vflag & VV_FORCEINSMQ) == 0)) { 1099 MNT_IUNLOCK(mp); 1100 if (dtr != NULL) 1101 dtr(vp, dtr_arg); 1102 return (EBUSY); 1103 } 1104 } 1105 vp->v_mount = mp; 1106 MNT_REF(mp); 1107 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1108 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, 1109 ("neg mount point vnode list size")); 1110 mp->mnt_nvnodelistsize++; 1111 MNT_IUNLOCK(mp); 1112 return (0); 1113 } 1114 1115 int 1116 insmntque(struct vnode *vp, struct mount *mp) 1117 { 1118 1119 return (insmntque1(vp, mp, insmntque_stddtr, NULL)); 1120 } 1121 1122 /* 1123 * Flush out and invalidate all buffers associated with a bufobj 1124 * Called with the underlying object locked. 1125 */ 1126 int 1127 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) 1128 { 1129 int error; 1130 1131 BO_LOCK(bo); 1132 if (flags & V_SAVE) { 1133 error = bufobj_wwait(bo, slpflag, slptimeo); 1134 if (error) { 1135 BO_UNLOCK(bo); 1136 return (error); 1137 } 1138 if (bo->bo_dirty.bv_cnt > 0) { 1139 BO_UNLOCK(bo); 1140 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) 1141 return (error); 1142 /* 1143 * XXX We could save a lock/unlock if this was only 1144 * enabled under INVARIANTS 1145 */ 1146 BO_LOCK(bo); 1147 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) 1148 panic("vinvalbuf: dirty bufs"); 1149 } 1150 } 1151 /* 1152 * If you alter this loop please notice that interlock is dropped and 1153 * reacquired in flushbuflist. Special care is needed to ensure that 1154 * no race conditions occur from this. 1155 */ 1156 do { 1157 error = flushbuflist(&bo->bo_clean, 1158 flags, bo, slpflag, slptimeo); 1159 if (error == 0) 1160 error = flushbuflist(&bo->bo_dirty, 1161 flags, bo, slpflag, slptimeo); 1162 if (error != 0 && error != EAGAIN) { 1163 BO_UNLOCK(bo); 1164 return (error); 1165 } 1166 } while (error != 0); 1167 1168 /* 1169 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 1170 * have write I/O in-progress but if there is a VM object then the 1171 * VM object can also have read-I/O in-progress. 1172 */ 1173 do { 1174 bufobj_wwait(bo, 0, 0); 1175 BO_UNLOCK(bo); 1176 if (bo->bo_object != NULL) { 1177 VM_OBJECT_LOCK(bo->bo_object); 1178 vm_object_pip_wait(bo->bo_object, "bovlbx"); 1179 VM_OBJECT_UNLOCK(bo->bo_object); 1180 } 1181 BO_LOCK(bo); 1182 } while (bo->bo_numoutput > 0); 1183 BO_UNLOCK(bo); 1184 1185 /* 1186 * Destroy the copy in the VM cache, too. 1187 */ 1188 if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL)) == 0) { 1189 VM_OBJECT_LOCK(bo->bo_object); 1190 vm_object_page_remove(bo->bo_object, 0, 0, 1191 (flags & V_SAVE) ? TRUE : FALSE); 1192 VM_OBJECT_UNLOCK(bo->bo_object); 1193 } 1194 1195 #ifdef INVARIANTS 1196 BO_LOCK(bo); 1197 if ((flags & (V_ALT | V_NORMAL)) == 0 && 1198 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) 1199 panic("vinvalbuf: flush failed"); 1200 BO_UNLOCK(bo); 1201 #endif 1202 return (0); 1203 } 1204 1205 /* 1206 * Flush out and invalidate all buffers associated with a vnode. 1207 * Called with the underlying object locked. 1208 */ 1209 int 1210 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 1211 { 1212 1213 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 1214 ASSERT_VOP_LOCKED(vp, "vinvalbuf"); 1215 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); 1216 } 1217 1218 /* 1219 * Flush out buffers on the specified list. 1220 * 1221 */ 1222 static int 1223 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, 1224 int slptimeo) 1225 { 1226 struct buf *bp, *nbp; 1227 int retval, error; 1228 daddr_t lblkno; 1229 b_xflags_t xflags; 1230 1231 ASSERT_BO_LOCKED(bo); 1232 1233 retval = 0; 1234 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { 1235 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) || 1236 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) { 1237 continue; 1238 } 1239 lblkno = 0; 1240 xflags = 0; 1241 if (nbp != NULL) { 1242 lblkno = nbp->b_lblkno; 1243 xflags = nbp->b_xflags & 1244 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN); 1245 } 1246 retval = EAGAIN; 1247 error = BUF_TIMELOCK(bp, 1248 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo), 1249 "flushbuf", slpflag, slptimeo); 1250 if (error) { 1251 BO_LOCK(bo); 1252 return (error != ENOLCK ? error : EAGAIN); 1253 } 1254 KASSERT(bp->b_bufobj == bo, 1255 ("bp %p wrong b_bufobj %p should be %p", 1256 bp, bp->b_bufobj, bo)); 1257 if (bp->b_bufobj != bo) { /* XXX: necessary ? */ 1258 BUF_UNLOCK(bp); 1259 BO_LOCK(bo); 1260 return (EAGAIN); 1261 } 1262 /* 1263 * XXX Since there are no node locks for NFS, I 1264 * believe there is a slight chance that a delayed 1265 * write will occur while sleeping just above, so 1266 * check for it. 1267 */ 1268 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 1269 (flags & V_SAVE)) { 1270 BO_LOCK(bo); 1271 bremfree(bp); 1272 BO_UNLOCK(bo); 1273 bp->b_flags |= B_ASYNC; 1274 bwrite(bp); 1275 BO_LOCK(bo); 1276 return (EAGAIN); /* XXX: why not loop ? */ 1277 } 1278 BO_LOCK(bo); 1279 bremfree(bp); 1280 BO_UNLOCK(bo); 1281 bp->b_flags |= (B_INVAL | B_RELBUF); 1282 bp->b_flags &= ~B_ASYNC; 1283 brelse(bp); 1284 BO_LOCK(bo); 1285 if (nbp != NULL && 1286 (nbp->b_bufobj != bo || 1287 nbp->b_lblkno != lblkno || 1288 (nbp->b_xflags & 1289 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags)) 1290 break; /* nbp invalid */ 1291 } 1292 return (retval); 1293 } 1294 1295 /* 1296 * Truncate a file's buffer and pages to a specified length. This 1297 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 1298 * sync activity. 1299 */ 1300 int 1301 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td, 1302 off_t length, int blksize) 1303 { 1304 struct buf *bp, *nbp; 1305 int anyfreed; 1306 int trunclbn; 1307 struct bufobj *bo; 1308 1309 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__, 1310 vp, cred, blksize, (uintmax_t)length); 1311 1312 /* 1313 * Round up to the *next* lbn. 1314 */ 1315 trunclbn = (length + blksize - 1) / blksize; 1316 1317 ASSERT_VOP_LOCKED(vp, "vtruncbuf"); 1318 restart: 1319 bo = &vp->v_bufobj; 1320 BO_LOCK(bo); 1321 anyfreed = 1; 1322 for (;anyfreed;) { 1323 anyfreed = 0; 1324 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { 1325 if (bp->b_lblkno < trunclbn) 1326 continue; 1327 if (BUF_LOCK(bp, 1328 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1329 BO_MTX(bo)) == ENOLCK) 1330 goto restart; 1331 1332 BO_LOCK(bo); 1333 bremfree(bp); 1334 BO_UNLOCK(bo); 1335 bp->b_flags |= (B_INVAL | B_RELBUF); 1336 bp->b_flags &= ~B_ASYNC; 1337 brelse(bp); 1338 anyfreed = 1; 1339 1340 BO_LOCK(bo); 1341 if (nbp != NULL && 1342 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 1343 (nbp->b_vp != vp) || 1344 (nbp->b_flags & B_DELWRI))) { 1345 BO_UNLOCK(bo); 1346 goto restart; 1347 } 1348 } 1349 1350 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1351 if (bp->b_lblkno < trunclbn) 1352 continue; 1353 if (BUF_LOCK(bp, 1354 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1355 BO_MTX(bo)) == ENOLCK) 1356 goto restart; 1357 BO_LOCK(bo); 1358 bremfree(bp); 1359 BO_UNLOCK(bo); 1360 bp->b_flags |= (B_INVAL | B_RELBUF); 1361 bp->b_flags &= ~B_ASYNC; 1362 brelse(bp); 1363 anyfreed = 1; 1364 1365 BO_LOCK(bo); 1366 if (nbp != NULL && 1367 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 1368 (nbp->b_vp != vp) || 1369 (nbp->b_flags & B_DELWRI) == 0)) { 1370 BO_UNLOCK(bo); 1371 goto restart; 1372 } 1373 } 1374 } 1375 1376 if (length > 0) { 1377 restartsync: 1378 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1379 if (bp->b_lblkno > 0) 1380 continue; 1381 /* 1382 * Since we hold the vnode lock this should only 1383 * fail if we're racing with the buf daemon. 1384 */ 1385 if (BUF_LOCK(bp, 1386 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1387 BO_MTX(bo)) == ENOLCK) { 1388 goto restart; 1389 } 1390 VNASSERT((bp->b_flags & B_DELWRI), vp, 1391 ("buf(%p) on dirty queue without DELWRI", bp)); 1392 1393 BO_LOCK(bo); 1394 bremfree(bp); 1395 BO_UNLOCK(bo); 1396 bawrite(bp); 1397 BO_LOCK(bo); 1398 goto restartsync; 1399 } 1400 } 1401 1402 bufobj_wwait(bo, 0, 0); 1403 BO_UNLOCK(bo); 1404 vnode_pager_setsize(vp, length); 1405 1406 return (0); 1407 } 1408 1409 /* 1410 * buf_splay() - splay tree core for the clean/dirty list of buffers in 1411 * a vnode. 1412 * 1413 * NOTE: We have to deal with the special case of a background bitmap 1414 * buffer, a situation where two buffers will have the same logical 1415 * block offset. We want (1) only the foreground buffer to be accessed 1416 * in a lookup and (2) must differentiate between the foreground and 1417 * background buffer in the splay tree algorithm because the splay 1418 * tree cannot normally handle multiple entities with the same 'index'. 1419 * We accomplish this by adding differentiating flags to the splay tree's 1420 * numerical domain. 1421 */ 1422 static 1423 struct buf * 1424 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root) 1425 { 1426 struct buf dummy; 1427 struct buf *lefttreemax, *righttreemin, *y; 1428 1429 if (root == NULL) 1430 return (NULL); 1431 lefttreemax = righttreemin = &dummy; 1432 for (;;) { 1433 if (lblkno < root->b_lblkno || 1434 (lblkno == root->b_lblkno && 1435 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1436 if ((y = root->b_left) == NULL) 1437 break; 1438 if (lblkno < y->b_lblkno) { 1439 /* Rotate right. */ 1440 root->b_left = y->b_right; 1441 y->b_right = root; 1442 root = y; 1443 if ((y = root->b_left) == NULL) 1444 break; 1445 } 1446 /* Link into the new root's right tree. */ 1447 righttreemin->b_left = root; 1448 righttreemin = root; 1449 } else if (lblkno > root->b_lblkno || 1450 (lblkno == root->b_lblkno && 1451 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) { 1452 if ((y = root->b_right) == NULL) 1453 break; 1454 if (lblkno > y->b_lblkno) { 1455 /* Rotate left. */ 1456 root->b_right = y->b_left; 1457 y->b_left = root; 1458 root = y; 1459 if ((y = root->b_right) == NULL) 1460 break; 1461 } 1462 /* Link into the new root's left tree. */ 1463 lefttreemax->b_right = root; 1464 lefttreemax = root; 1465 } else { 1466 break; 1467 } 1468 root = y; 1469 } 1470 /* Assemble the new root. */ 1471 lefttreemax->b_right = root->b_left; 1472 righttreemin->b_left = root->b_right; 1473 root->b_left = dummy.b_right; 1474 root->b_right = dummy.b_left; 1475 return (root); 1476 } 1477 1478 static void 1479 buf_vlist_remove(struct buf *bp) 1480 { 1481 struct buf *root; 1482 struct bufv *bv; 1483 1484 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); 1485 ASSERT_BO_LOCKED(bp->b_bufobj); 1486 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != 1487 (BX_VNDIRTY|BX_VNCLEAN), 1488 ("buf_vlist_remove: Buf %p is on two lists", bp)); 1489 if (bp->b_xflags & BX_VNDIRTY) 1490 bv = &bp->b_bufobj->bo_dirty; 1491 else 1492 bv = &bp->b_bufobj->bo_clean; 1493 if (bp != bv->bv_root) { 1494 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1495 KASSERT(root == bp, ("splay lookup failed in remove")); 1496 } 1497 if (bp->b_left == NULL) { 1498 root = bp->b_right; 1499 } else { 1500 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left); 1501 root->b_right = bp->b_right; 1502 } 1503 bv->bv_root = root; 1504 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); 1505 bv->bv_cnt--; 1506 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1507 } 1508 1509 /* 1510 * Add the buffer to the sorted clean or dirty block list using a 1511 * splay tree algorithm. 1512 * 1513 * NOTE: xflags is passed as a constant, optimizing this inline function! 1514 */ 1515 static void 1516 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) 1517 { 1518 struct buf *root; 1519 struct bufv *bv; 1520 1521 ASSERT_BO_LOCKED(bo); 1522 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, 1523 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); 1524 bp->b_xflags |= xflags; 1525 if (xflags & BX_VNDIRTY) 1526 bv = &bo->bo_dirty; 1527 else 1528 bv = &bo->bo_clean; 1529 1530 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1531 if (root == NULL) { 1532 bp->b_left = NULL; 1533 bp->b_right = NULL; 1534 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); 1535 } else if (bp->b_lblkno < root->b_lblkno || 1536 (bp->b_lblkno == root->b_lblkno && 1537 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1538 bp->b_left = root->b_left; 1539 bp->b_right = root; 1540 root->b_left = NULL; 1541 TAILQ_INSERT_BEFORE(root, bp, b_bobufs); 1542 } else { 1543 bp->b_right = root->b_right; 1544 bp->b_left = root; 1545 root->b_right = NULL; 1546 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs); 1547 } 1548 bv->bv_cnt++; 1549 bv->bv_root = bp; 1550 } 1551 1552 /* 1553 * Lookup a buffer using the splay tree. Note that we specifically avoid 1554 * shadow buffers used in background bitmap writes. 1555 * 1556 * This code isn't quite efficient as it could be because we are maintaining 1557 * two sorted lists and do not know which list the block resides in. 1558 * 1559 * During a "make buildworld" the desired buffer is found at one of 1560 * the roots more than 60% of the time. Thus, checking both roots 1561 * before performing either splay eliminates unnecessary splays on the 1562 * first tree splayed. 1563 */ 1564 struct buf * 1565 gbincore(struct bufobj *bo, daddr_t lblkno) 1566 { 1567 struct buf *bp; 1568 1569 ASSERT_BO_LOCKED(bo); 1570 if ((bp = bo->bo_clean.bv_root) != NULL && 1571 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1572 return (bp); 1573 if ((bp = bo->bo_dirty.bv_root) != NULL && 1574 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1575 return (bp); 1576 if ((bp = bo->bo_clean.bv_root) != NULL) { 1577 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp); 1578 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1579 return (bp); 1580 } 1581 if ((bp = bo->bo_dirty.bv_root) != NULL) { 1582 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp); 1583 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1584 return (bp); 1585 } 1586 return (NULL); 1587 } 1588 1589 /* 1590 * Associate a buffer with a vnode. 1591 */ 1592 void 1593 bgetvp(struct vnode *vp, struct buf *bp) 1594 { 1595 struct bufobj *bo; 1596 1597 bo = &vp->v_bufobj; 1598 ASSERT_BO_LOCKED(bo); 1599 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); 1600 1601 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); 1602 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, 1603 ("bgetvp: bp already attached! %p", bp)); 1604 1605 vhold(vp); 1606 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT) 1607 bp->b_flags |= B_NEEDSGIANT; 1608 bp->b_vp = vp; 1609 bp->b_bufobj = bo; 1610 /* 1611 * Insert onto list for new vnode. 1612 */ 1613 buf_vlist_add(bp, bo, BX_VNCLEAN); 1614 } 1615 1616 /* 1617 * Disassociate a buffer from a vnode. 1618 */ 1619 void 1620 brelvp(struct buf *bp) 1621 { 1622 struct bufobj *bo; 1623 struct vnode *vp; 1624 1625 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1626 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1627 1628 /* 1629 * Delete from old vnode list, if on one. 1630 */ 1631 vp = bp->b_vp; /* XXX */ 1632 bo = bp->b_bufobj; 1633 BO_LOCK(bo); 1634 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1635 buf_vlist_remove(bp); 1636 else 1637 panic("brelvp: Buffer %p not on queue.", bp); 1638 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1639 bo->bo_flag &= ~BO_ONWORKLST; 1640 mtx_lock(&sync_mtx); 1641 LIST_REMOVE(bo, bo_synclist); 1642 syncer_worklist_len--; 1643 mtx_unlock(&sync_mtx); 1644 } 1645 bp->b_flags &= ~B_NEEDSGIANT; 1646 bp->b_vp = NULL; 1647 bp->b_bufobj = NULL; 1648 BO_UNLOCK(bo); 1649 vdrop(vp); 1650 } 1651 1652 /* 1653 * Add an item to the syncer work queue. 1654 */ 1655 static void 1656 vn_syncer_add_to_worklist(struct bufobj *bo, int delay) 1657 { 1658 int queue, slot; 1659 1660 ASSERT_BO_LOCKED(bo); 1661 1662 mtx_lock(&sync_mtx); 1663 if (bo->bo_flag & BO_ONWORKLST) 1664 LIST_REMOVE(bo, bo_synclist); 1665 else { 1666 bo->bo_flag |= BO_ONWORKLST; 1667 syncer_worklist_len++; 1668 } 1669 1670 if (delay > syncer_maxdelay - 2) 1671 delay = syncer_maxdelay - 2; 1672 slot = (syncer_delayno + delay) & syncer_mask; 1673 1674 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ : 1675 WI_MPSAFEQ; 1676 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo, 1677 bo_synclist); 1678 mtx_unlock(&sync_mtx); 1679 } 1680 1681 static int 1682 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) 1683 { 1684 int error, len; 1685 1686 mtx_lock(&sync_mtx); 1687 len = syncer_worklist_len - sync_vnode_count; 1688 mtx_unlock(&sync_mtx); 1689 error = SYSCTL_OUT(req, &len, sizeof(len)); 1690 return (error); 1691 } 1692 1693 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, 1694 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); 1695 1696 static struct proc *updateproc; 1697 static void sched_sync(void); 1698 static struct kproc_desc up_kp = { 1699 "syncer", 1700 sched_sync, 1701 &updateproc 1702 }; 1703 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); 1704 1705 static int 1706 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) 1707 { 1708 struct vnode *vp; 1709 struct mount *mp; 1710 1711 *bo = LIST_FIRST(slp); 1712 if (*bo == NULL) 1713 return (0); 1714 vp = (*bo)->__bo_vnode; /* XXX */ 1715 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) 1716 return (1); 1717 /* 1718 * We use vhold in case the vnode does not 1719 * successfully sync. vhold prevents the vnode from 1720 * going away when we unlock the sync_mtx so that 1721 * we can acquire the vnode interlock. 1722 */ 1723 vholdl(vp); 1724 mtx_unlock(&sync_mtx); 1725 VI_UNLOCK(vp); 1726 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { 1727 vdrop(vp); 1728 mtx_lock(&sync_mtx); 1729 return (*bo == LIST_FIRST(slp)); 1730 } 1731 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1732 (void) VOP_FSYNC(vp, MNT_LAZY, td); 1733 VOP_UNLOCK(vp, 0); 1734 vn_finished_write(mp); 1735 BO_LOCK(*bo); 1736 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { 1737 /* 1738 * Put us back on the worklist. The worklist 1739 * routine will remove us from our current 1740 * position and then add us back in at a later 1741 * position. 1742 */ 1743 vn_syncer_add_to_worklist(*bo, syncdelay); 1744 } 1745 BO_UNLOCK(*bo); 1746 vdrop(vp); 1747 mtx_lock(&sync_mtx); 1748 return (0); 1749 } 1750 1751 /* 1752 * System filesystem synchronizer daemon. 1753 */ 1754 static void 1755 sched_sync(void) 1756 { 1757 struct synclist *gnext, *next; 1758 struct synclist *gslp, *slp; 1759 struct bufobj *bo; 1760 long starttime; 1761 struct thread *td = curthread; 1762 int last_work_seen; 1763 int net_worklist_len; 1764 int syncer_final_iter; 1765 int first_printf; 1766 int error; 1767 1768 last_work_seen = 0; 1769 syncer_final_iter = 0; 1770 first_printf = 1; 1771 syncer_state = SYNCER_RUNNING; 1772 starttime = time_uptime; 1773 td->td_pflags |= TDP_NORUNNINGBUF; 1774 1775 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, 1776 SHUTDOWN_PRI_LAST); 1777 1778 mtx_lock(&sync_mtx); 1779 for (;;) { 1780 if (syncer_state == SYNCER_FINAL_DELAY && 1781 syncer_final_iter == 0) { 1782 mtx_unlock(&sync_mtx); 1783 kproc_suspend_check(td->td_proc); 1784 mtx_lock(&sync_mtx); 1785 } 1786 net_worklist_len = syncer_worklist_len - sync_vnode_count; 1787 if (syncer_state != SYNCER_RUNNING && 1788 starttime != time_uptime) { 1789 if (first_printf) { 1790 printf("\nSyncing disks, vnodes remaining..."); 1791 first_printf = 0; 1792 } 1793 printf("%d ", net_worklist_len); 1794 } 1795 starttime = time_uptime; 1796 1797 /* 1798 * Push files whose dirty time has expired. Be careful 1799 * of interrupt race on slp queue. 1800 * 1801 * Skip over empty worklist slots when shutting down. 1802 */ 1803 do { 1804 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1805 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1806 syncer_delayno += 1; 1807 if (syncer_delayno == syncer_maxdelay) 1808 syncer_delayno = 0; 1809 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1810 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1811 /* 1812 * If the worklist has wrapped since the 1813 * it was emptied of all but syncer vnodes, 1814 * switch to the FINAL_DELAY state and run 1815 * for one more second. 1816 */ 1817 if (syncer_state == SYNCER_SHUTTING_DOWN && 1818 net_worklist_len == 0 && 1819 last_work_seen == syncer_delayno) { 1820 syncer_state = SYNCER_FINAL_DELAY; 1821 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; 1822 } 1823 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && 1824 LIST_EMPTY(gslp) && syncer_worklist_len > 0); 1825 1826 /* 1827 * Keep track of the last time there was anything 1828 * on the worklist other than syncer vnodes. 1829 * Return to the SHUTTING_DOWN state if any 1830 * new work appears. 1831 */ 1832 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) 1833 last_work_seen = syncer_delayno; 1834 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) 1835 syncer_state = SYNCER_SHUTTING_DOWN; 1836 while (!LIST_EMPTY(slp)) { 1837 error = sync_vnode(slp, &bo, td); 1838 if (error == 1) { 1839 LIST_REMOVE(bo, bo_synclist); 1840 LIST_INSERT_HEAD(next, bo, bo_synclist); 1841 continue; 1842 } 1843 } 1844 if (!LIST_EMPTY(gslp)) { 1845 mtx_unlock(&sync_mtx); 1846 mtx_lock(&Giant); 1847 mtx_lock(&sync_mtx); 1848 while (!LIST_EMPTY(gslp)) { 1849 error = sync_vnode(gslp, &bo, td); 1850 if (error == 1) { 1851 LIST_REMOVE(bo, bo_synclist); 1852 LIST_INSERT_HEAD(gnext, bo, 1853 bo_synclist); 1854 continue; 1855 } 1856 } 1857 mtx_unlock(&Giant); 1858 } 1859 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) 1860 syncer_final_iter--; 1861 /* 1862 * The variable rushjob allows the kernel to speed up the 1863 * processing of the filesystem syncer process. A rushjob 1864 * value of N tells the filesystem syncer to process the next 1865 * N seconds worth of work on its queue ASAP. Currently rushjob 1866 * is used by the soft update code to speed up the filesystem 1867 * syncer process when the incore state is getting so far 1868 * ahead of the disk that the kernel memory pool is being 1869 * threatened with exhaustion. 1870 */ 1871 if (rushjob > 0) { 1872 rushjob -= 1; 1873 continue; 1874 } 1875 /* 1876 * Just sleep for a short period of time between 1877 * iterations when shutting down to allow some I/O 1878 * to happen. 1879 * 1880 * If it has taken us less than a second to process the 1881 * current work, then wait. Otherwise start right over 1882 * again. We can still lose time if any single round 1883 * takes more than two seconds, but it does not really 1884 * matter as we are just trying to generally pace the 1885 * filesystem activity. 1886 */ 1887 if (syncer_state != SYNCER_RUNNING || 1888 time_uptime == starttime) { 1889 thread_lock(td); 1890 sched_prio(td, PPAUSE); 1891 thread_unlock(td); 1892 } 1893 if (syncer_state != SYNCER_RUNNING) 1894 cv_timedwait(&sync_wakeup, &sync_mtx, 1895 hz / SYNCER_SHUTDOWN_SPEEDUP); 1896 else if (time_uptime == starttime) 1897 cv_timedwait(&sync_wakeup, &sync_mtx, hz); 1898 } 1899 } 1900 1901 /* 1902 * Request the syncer daemon to speed up its work. 1903 * We never push it to speed up more than half of its 1904 * normal turn time, otherwise it could take over the cpu. 1905 */ 1906 int 1907 speedup_syncer(void) 1908 { 1909 int ret = 0; 1910 1911 mtx_lock(&sync_mtx); 1912 if (rushjob < syncdelay / 2) { 1913 rushjob += 1; 1914 stat_rush_requests += 1; 1915 ret = 1; 1916 } 1917 mtx_unlock(&sync_mtx); 1918 cv_broadcast(&sync_wakeup); 1919 return (ret); 1920 } 1921 1922 /* 1923 * Tell the syncer to speed up its work and run though its work 1924 * list several times, then tell it to shut down. 1925 */ 1926 static void 1927 syncer_shutdown(void *arg, int howto) 1928 { 1929 1930 if (howto & RB_NOSYNC) 1931 return; 1932 mtx_lock(&sync_mtx); 1933 syncer_state = SYNCER_SHUTTING_DOWN; 1934 rushjob = 0; 1935 mtx_unlock(&sync_mtx); 1936 cv_broadcast(&sync_wakeup); 1937 kproc_shutdown(arg, howto); 1938 } 1939 1940 /* 1941 * Reassign a buffer from one vnode to another. 1942 * Used to assign file specific control information 1943 * (indirect blocks) to the vnode to which they belong. 1944 */ 1945 void 1946 reassignbuf(struct buf *bp) 1947 { 1948 struct vnode *vp; 1949 struct bufobj *bo; 1950 int delay; 1951 #ifdef INVARIANTS 1952 struct bufv *bv; 1953 #endif 1954 1955 vp = bp->b_vp; 1956 bo = bp->b_bufobj; 1957 ++reassignbufcalls; 1958 1959 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", 1960 bp, bp->b_vp, bp->b_flags); 1961 /* 1962 * B_PAGING flagged buffers cannot be reassigned because their vp 1963 * is not fully linked in. 1964 */ 1965 if (bp->b_flags & B_PAGING) 1966 panic("cannot reassign paging buffer"); 1967 1968 /* 1969 * Delete from old vnode list, if on one. 1970 */ 1971 BO_LOCK(bo); 1972 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1973 buf_vlist_remove(bp); 1974 else 1975 panic("reassignbuf: Buffer %p not on queue.", bp); 1976 /* 1977 * If dirty, put on list of dirty buffers; otherwise insert onto list 1978 * of clean buffers. 1979 */ 1980 if (bp->b_flags & B_DELWRI) { 1981 if ((bo->bo_flag & BO_ONWORKLST) == 0) { 1982 switch (vp->v_type) { 1983 case VDIR: 1984 delay = dirdelay; 1985 break; 1986 case VCHR: 1987 delay = metadelay; 1988 break; 1989 default: 1990 delay = filedelay; 1991 } 1992 vn_syncer_add_to_worklist(bo, delay); 1993 } 1994 buf_vlist_add(bp, bo, BX_VNDIRTY); 1995 } else { 1996 buf_vlist_add(bp, bo, BX_VNCLEAN); 1997 1998 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1999 mtx_lock(&sync_mtx); 2000 LIST_REMOVE(bo, bo_synclist); 2001 syncer_worklist_len--; 2002 mtx_unlock(&sync_mtx); 2003 bo->bo_flag &= ~BO_ONWORKLST; 2004 } 2005 } 2006 #ifdef INVARIANTS 2007 bv = &bo->bo_clean; 2008 bp = TAILQ_FIRST(&bv->bv_hd); 2009 KASSERT(bp == NULL || bp->b_bufobj == bo, 2010 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2011 bp = TAILQ_LAST(&bv->bv_hd, buflists); 2012 KASSERT(bp == NULL || bp->b_bufobj == bo, 2013 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2014 bv = &bo->bo_dirty; 2015 bp = TAILQ_FIRST(&bv->bv_hd); 2016 KASSERT(bp == NULL || bp->b_bufobj == bo, 2017 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2018 bp = TAILQ_LAST(&bv->bv_hd, buflists); 2019 KASSERT(bp == NULL || bp->b_bufobj == bo, 2020 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2021 #endif 2022 BO_UNLOCK(bo); 2023 } 2024 2025 /* 2026 * Increment the use and hold counts on the vnode, taking care to reference 2027 * the driver's usecount if this is a chardev. The vholdl() will remove 2028 * the vnode from the free list if it is presently free. Requires the 2029 * vnode interlock and returns with it held. 2030 */ 2031 static void 2032 v_incr_usecount(struct vnode *vp) 2033 { 2034 2035 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2036 vp->v_usecount++; 2037 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2038 dev_lock(); 2039 vp->v_rdev->si_usecount++; 2040 dev_unlock(); 2041 } 2042 vholdl(vp); 2043 } 2044 2045 /* 2046 * Turn a holdcnt into a use+holdcnt such that only one call to 2047 * v_decr_usecount is needed. 2048 */ 2049 static void 2050 v_upgrade_usecount(struct vnode *vp) 2051 { 2052 2053 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2054 vp->v_usecount++; 2055 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2056 dev_lock(); 2057 vp->v_rdev->si_usecount++; 2058 dev_unlock(); 2059 } 2060 } 2061 2062 /* 2063 * Decrement the vnode use and hold count along with the driver's usecount 2064 * if this is a chardev. The vdropl() below releases the vnode interlock 2065 * as it may free the vnode. 2066 */ 2067 static void 2068 v_decr_usecount(struct vnode *vp) 2069 { 2070 2071 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2072 VNASSERT(vp->v_usecount > 0, vp, 2073 ("v_decr_usecount: negative usecount")); 2074 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2075 vp->v_usecount--; 2076 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2077 dev_lock(); 2078 vp->v_rdev->si_usecount--; 2079 dev_unlock(); 2080 } 2081 vdropl(vp); 2082 } 2083 2084 /* 2085 * Decrement only the use count and driver use count. This is intended to 2086 * be paired with a follow on vdropl() to release the remaining hold count. 2087 * In this way we may vgone() a vnode with a 0 usecount without risk of 2088 * having it end up on a free list because the hold count is kept above 0. 2089 */ 2090 static void 2091 v_decr_useonly(struct vnode *vp) 2092 { 2093 2094 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2095 VNASSERT(vp->v_usecount > 0, vp, 2096 ("v_decr_useonly: negative usecount")); 2097 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2098 vp->v_usecount--; 2099 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2100 dev_lock(); 2101 vp->v_rdev->si_usecount--; 2102 dev_unlock(); 2103 } 2104 } 2105 2106 /* 2107 * Grab a particular vnode from the free list, increment its 2108 * reference count and lock it. VI_DOOMED is set if the vnode 2109 * is being destroyed. Only callers who specify LK_RETRY will 2110 * see doomed vnodes. If inactive processing was delayed in 2111 * vput try to do it here. 2112 */ 2113 int 2114 vget(struct vnode *vp, int flags, struct thread *td) 2115 { 2116 int error; 2117 2118 error = 0; 2119 VFS_ASSERT_GIANT(vp->v_mount); 2120 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 2121 ("vget: invalid lock operation")); 2122 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 2123 2124 if ((flags & LK_INTERLOCK) == 0) 2125 VI_LOCK(vp); 2126 vholdl(vp); 2127 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) { 2128 vdrop(vp); 2129 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, 2130 vp); 2131 return (error); 2132 } 2133 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) 2134 panic("vget: vn_lock failed to return ENOENT\n"); 2135 VI_LOCK(vp); 2136 /* Upgrade our holdcnt to a usecount. */ 2137 v_upgrade_usecount(vp); 2138 /* 2139 * We don't guarantee that any particular close will 2140 * trigger inactive processing so just make a best effort 2141 * here at preventing a reference to a removed file. If 2142 * we don't succeed no harm is done. 2143 */ 2144 if (vp->v_iflag & VI_OWEINACT) { 2145 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && 2146 (flags & LK_NOWAIT) == 0) 2147 vinactive(vp, td); 2148 vp->v_iflag &= ~VI_OWEINACT; 2149 } 2150 VI_UNLOCK(vp); 2151 return (0); 2152 } 2153 2154 /* 2155 * Increase the reference count of a vnode. 2156 */ 2157 void 2158 vref(struct vnode *vp) 2159 { 2160 2161 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2162 VI_LOCK(vp); 2163 v_incr_usecount(vp); 2164 VI_UNLOCK(vp); 2165 } 2166 2167 /* 2168 * Return reference count of a vnode. 2169 * 2170 * The results of this call are only guaranteed when some mechanism other 2171 * than the VI lock is used to stop other processes from gaining references 2172 * to the vnode. This may be the case if the caller holds the only reference. 2173 * This is also useful when stale data is acceptable as race conditions may 2174 * be accounted for by some other means. 2175 */ 2176 int 2177 vrefcnt(struct vnode *vp) 2178 { 2179 int usecnt; 2180 2181 VI_LOCK(vp); 2182 usecnt = vp->v_usecount; 2183 VI_UNLOCK(vp); 2184 2185 return (usecnt); 2186 } 2187 2188 #define VPUTX_VRELE 1 2189 #define VPUTX_VPUT 2 2190 #define VPUTX_VUNREF 3 2191 2192 static void 2193 vputx(struct vnode *vp, int func) 2194 { 2195 int error; 2196 2197 KASSERT(vp != NULL, ("vputx: null vp")); 2198 if (func == VPUTX_VUNREF) 2199 ASSERT_VOP_LOCKED(vp, "vunref"); 2200 else if (func == VPUTX_VPUT) 2201 ASSERT_VOP_LOCKED(vp, "vput"); 2202 else 2203 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func")); 2204 VFS_ASSERT_GIANT(vp->v_mount); 2205 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2206 VI_LOCK(vp); 2207 2208 /* Skip this v_writecount check if we're going to panic below. */ 2209 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp, 2210 ("vputx: missed vn_close")); 2211 error = 0; 2212 2213 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2214 vp->v_usecount == 1)) { 2215 if (func == VPUTX_VPUT) 2216 VOP_UNLOCK(vp, 0); 2217 v_decr_usecount(vp); 2218 return; 2219 } 2220 2221 if (vp->v_usecount != 1) { 2222 vprint("vputx: negative ref count", vp); 2223 panic("vputx: negative ref cnt"); 2224 } 2225 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp); 2226 /* 2227 * We want to hold the vnode until the inactive finishes to 2228 * prevent vgone() races. We drop the use count here and the 2229 * hold count below when we're done. 2230 */ 2231 v_decr_useonly(vp); 2232 /* 2233 * We must call VOP_INACTIVE with the node locked. Mark 2234 * as VI_DOINGINACT to avoid recursion. 2235 */ 2236 vp->v_iflag |= VI_OWEINACT; 2237 switch (func) { 2238 case VPUTX_VRELE: 2239 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2240 VI_LOCK(vp); 2241 break; 2242 case VPUTX_VPUT: 2243 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 2244 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | 2245 LK_NOWAIT); 2246 VI_LOCK(vp); 2247 } 2248 break; 2249 case VPUTX_VUNREF: 2250 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 2251 error = EBUSY; 2252 break; 2253 } 2254 if (vp->v_usecount > 0) 2255 vp->v_iflag &= ~VI_OWEINACT; 2256 if (error == 0) { 2257 if (vp->v_iflag & VI_OWEINACT) 2258 vinactive(vp, curthread); 2259 if (func != VPUTX_VUNREF) 2260 VOP_UNLOCK(vp, 0); 2261 } 2262 vdropl(vp); 2263 } 2264 2265 /* 2266 * Vnode put/release. 2267 * If count drops to zero, call inactive routine and return to freelist. 2268 */ 2269 void 2270 vrele(struct vnode *vp) 2271 { 2272 2273 vputx(vp, VPUTX_VRELE); 2274 } 2275 2276 /* 2277 * Release an already locked vnode. This give the same effects as 2278 * unlock+vrele(), but takes less time and avoids releasing and 2279 * re-aquiring the lock (as vrele() acquires the lock internally.) 2280 */ 2281 void 2282 vput(struct vnode *vp) 2283 { 2284 2285 vputx(vp, VPUTX_VPUT); 2286 } 2287 2288 /* 2289 * Release an exclusively locked vnode. Do not unlock the vnode lock. 2290 */ 2291 void 2292 vunref(struct vnode *vp) 2293 { 2294 2295 vputx(vp, VPUTX_VUNREF); 2296 } 2297 2298 /* 2299 * Somebody doesn't want the vnode recycled. 2300 */ 2301 void 2302 vhold(struct vnode *vp) 2303 { 2304 2305 VI_LOCK(vp); 2306 vholdl(vp); 2307 VI_UNLOCK(vp); 2308 } 2309 2310 void 2311 vholdl(struct vnode *vp) 2312 { 2313 2314 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2315 vp->v_holdcnt++; 2316 if (VSHOULDBUSY(vp)) 2317 vbusy(vp); 2318 } 2319 2320 /* 2321 * Note that there is one less who cares about this vnode. vdrop() is the 2322 * opposite of vhold(). 2323 */ 2324 void 2325 vdrop(struct vnode *vp) 2326 { 2327 2328 VI_LOCK(vp); 2329 vdropl(vp); 2330 } 2331 2332 /* 2333 * Drop the hold count of the vnode. If this is the last reference to 2334 * the vnode we will free it if it has been vgone'd otherwise it is 2335 * placed on the free list. 2336 */ 2337 void 2338 vdropl(struct vnode *vp) 2339 { 2340 2341 ASSERT_VI_LOCKED(vp, "vdropl"); 2342 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2343 if (vp->v_holdcnt <= 0) 2344 panic("vdrop: holdcnt %d", vp->v_holdcnt); 2345 vp->v_holdcnt--; 2346 if (vp->v_holdcnt == 0) { 2347 if (vp->v_iflag & VI_DOOMED) { 2348 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, 2349 vp); 2350 vdestroy(vp); 2351 return; 2352 } else 2353 vfree(vp); 2354 } 2355 VI_UNLOCK(vp); 2356 } 2357 2358 /* 2359 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT 2360 * flags. DOINGINACT prevents us from recursing in calls to vinactive. 2361 * OWEINACT tracks whether a vnode missed a call to inactive due to a 2362 * failed lock upgrade. 2363 */ 2364 static void 2365 vinactive(struct vnode *vp, struct thread *td) 2366 { 2367 2368 ASSERT_VOP_ELOCKED(vp, "vinactive"); 2369 ASSERT_VI_LOCKED(vp, "vinactive"); 2370 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, 2371 ("vinactive: recursed on VI_DOINGINACT")); 2372 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2373 vp->v_iflag |= VI_DOINGINACT; 2374 vp->v_iflag &= ~VI_OWEINACT; 2375 VI_UNLOCK(vp); 2376 VOP_INACTIVE(vp, td); 2377 VI_LOCK(vp); 2378 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, 2379 ("vinactive: lost VI_DOINGINACT")); 2380 vp->v_iflag &= ~VI_DOINGINACT; 2381 } 2382 2383 /* 2384 * Remove any vnodes in the vnode table belonging to mount point mp. 2385 * 2386 * If FORCECLOSE is not specified, there should not be any active ones, 2387 * return error if any are found (nb: this is a user error, not a 2388 * system error). If FORCECLOSE is specified, detach any active vnodes 2389 * that are found. 2390 * 2391 * If WRITECLOSE is set, only flush out regular file vnodes open for 2392 * writing. 2393 * 2394 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. 2395 * 2396 * `rootrefs' specifies the base reference count for the root vnode 2397 * of this filesystem. The root vnode is considered busy if its 2398 * v_usecount exceeds this value. On a successful return, vflush(, td) 2399 * will call vrele() on the root vnode exactly rootrefs times. 2400 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 2401 * be zero. 2402 */ 2403 #ifdef DIAGNOSTIC 2404 static int busyprt = 0; /* print out busy vnodes */ 2405 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes"); 2406 #endif 2407 2408 int 2409 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td) 2410 { 2411 struct vnode *vp, *mvp, *rootvp = NULL; 2412 struct vattr vattr; 2413 int busy = 0, error; 2414 2415 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, 2416 rootrefs, flags); 2417 if (rootrefs > 0) { 2418 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 2419 ("vflush: bad args")); 2420 /* 2421 * Get the filesystem root vnode. We can vput() it 2422 * immediately, since with rootrefs > 0, it won't go away. 2423 */ 2424 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) { 2425 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", 2426 __func__, error); 2427 return (error); 2428 } 2429 vput(rootvp); 2430 } 2431 MNT_ILOCK(mp); 2432 loop: 2433 MNT_VNODE_FOREACH(vp, mp, mvp) { 2434 VI_LOCK(vp); 2435 vholdl(vp); 2436 MNT_IUNLOCK(mp); 2437 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); 2438 if (error) { 2439 vdrop(vp); 2440 MNT_ILOCK(mp); 2441 MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp); 2442 goto loop; 2443 } 2444 /* 2445 * Skip over a vnodes marked VV_SYSTEM. 2446 */ 2447 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 2448 VOP_UNLOCK(vp, 0); 2449 vdrop(vp); 2450 MNT_ILOCK(mp); 2451 continue; 2452 } 2453 /* 2454 * If WRITECLOSE is set, flush out unlinked but still open 2455 * files (even if open only for reading) and regular file 2456 * vnodes open for writing. 2457 */ 2458 if (flags & WRITECLOSE) { 2459 error = VOP_GETATTR(vp, &vattr, td->td_ucred); 2460 VI_LOCK(vp); 2461 2462 if ((vp->v_type == VNON || 2463 (error == 0 && vattr.va_nlink > 0)) && 2464 (vp->v_writecount == 0 || vp->v_type != VREG)) { 2465 VOP_UNLOCK(vp, 0); 2466 vdropl(vp); 2467 MNT_ILOCK(mp); 2468 continue; 2469 } 2470 } else 2471 VI_LOCK(vp); 2472 /* 2473 * With v_usecount == 0, all we need to do is clear out the 2474 * vnode data structures and we are done. 2475 * 2476 * If FORCECLOSE is set, forcibly close the vnode. 2477 */ 2478 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { 2479 VNASSERT(vp->v_usecount == 0 || 2480 (vp->v_type != VCHR && vp->v_type != VBLK), vp, 2481 ("device VNODE %p is FORCECLOSED", vp)); 2482 vgonel(vp); 2483 } else { 2484 busy++; 2485 #ifdef DIAGNOSTIC 2486 if (busyprt) 2487 vprint("vflush: busy vnode", vp); 2488 #endif 2489 } 2490 VOP_UNLOCK(vp, 0); 2491 vdropl(vp); 2492 MNT_ILOCK(mp); 2493 } 2494 MNT_IUNLOCK(mp); 2495 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 2496 /* 2497 * If just the root vnode is busy, and if its refcount 2498 * is equal to `rootrefs', then go ahead and kill it. 2499 */ 2500 VI_LOCK(rootvp); 2501 KASSERT(busy > 0, ("vflush: not busy")); 2502 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, 2503 ("vflush: usecount %d < rootrefs %d", 2504 rootvp->v_usecount, rootrefs)); 2505 if (busy == 1 && rootvp->v_usecount == rootrefs) { 2506 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); 2507 vgone(rootvp); 2508 VOP_UNLOCK(rootvp, 0); 2509 busy = 0; 2510 } else 2511 VI_UNLOCK(rootvp); 2512 } 2513 if (busy) { 2514 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, 2515 busy); 2516 return (EBUSY); 2517 } 2518 for (; rootrefs > 0; rootrefs--) 2519 vrele(rootvp); 2520 return (0); 2521 } 2522 2523 /* 2524 * Recycle an unused vnode to the front of the free list. 2525 */ 2526 int 2527 vrecycle(struct vnode *vp, struct thread *td) 2528 { 2529 int recycled; 2530 2531 ASSERT_VOP_ELOCKED(vp, "vrecycle"); 2532 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2533 recycled = 0; 2534 VI_LOCK(vp); 2535 if (vp->v_usecount == 0) { 2536 recycled = 1; 2537 vgonel(vp); 2538 } 2539 VI_UNLOCK(vp); 2540 return (recycled); 2541 } 2542 2543 /* 2544 * Eliminate all activity associated with a vnode 2545 * in preparation for reuse. 2546 */ 2547 void 2548 vgone(struct vnode *vp) 2549 { 2550 VI_LOCK(vp); 2551 vgonel(vp); 2552 VI_UNLOCK(vp); 2553 } 2554 2555 /* 2556 * vgone, with the vp interlock held. 2557 */ 2558 void 2559 vgonel(struct vnode *vp) 2560 { 2561 struct thread *td; 2562 int oweinact; 2563 int active; 2564 struct mount *mp; 2565 2566 ASSERT_VOP_ELOCKED(vp, "vgonel"); 2567 ASSERT_VI_LOCKED(vp, "vgonel"); 2568 VNASSERT(vp->v_holdcnt, vp, 2569 ("vgonel: vp %p has no reference.", vp)); 2570 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2571 td = curthread; 2572 2573 /* 2574 * Don't vgonel if we're already doomed. 2575 */ 2576 if (vp->v_iflag & VI_DOOMED) 2577 return; 2578 vp->v_iflag |= VI_DOOMED; 2579 /* 2580 * Check to see if the vnode is in use. If so, we have to call 2581 * VOP_CLOSE() and VOP_INACTIVE(). 2582 */ 2583 active = vp->v_usecount; 2584 oweinact = (vp->v_iflag & VI_OWEINACT); 2585 VI_UNLOCK(vp); 2586 /* 2587 * Clean out any buffers associated with the vnode. 2588 * If the flush fails, just toss the buffers. 2589 */ 2590 mp = NULL; 2591 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) 2592 (void) vn_start_secondary_write(vp, &mp, V_WAIT); 2593 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) 2594 vinvalbuf(vp, 0, 0, 0); 2595 2596 /* 2597 * If purging an active vnode, it must be closed and 2598 * deactivated before being reclaimed. 2599 */ 2600 if (active) 2601 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); 2602 if (oweinact || active) { 2603 VI_LOCK(vp); 2604 if ((vp->v_iflag & VI_DOINGINACT) == 0) 2605 vinactive(vp, td); 2606 VI_UNLOCK(vp); 2607 } 2608 /* 2609 * Reclaim the vnode. 2610 */ 2611 if (VOP_RECLAIM(vp, td)) 2612 panic("vgone: cannot reclaim"); 2613 if (mp != NULL) 2614 vn_finished_secondary_write(mp); 2615 VNASSERT(vp->v_object == NULL, vp, 2616 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag)); 2617 /* 2618 * Clear the advisory locks and wake up waiting threads. 2619 */ 2620 (void)VOP_ADVLOCKPURGE(vp); 2621 /* 2622 * Delete from old mount point vnode list. 2623 */ 2624 delmntque(vp); 2625 cache_purge(vp); 2626 /* 2627 * Done with purge, reset to the standard lock and invalidate 2628 * the vnode. 2629 */ 2630 VI_LOCK(vp); 2631 vp->v_vnlock = &vp->v_lock; 2632 vp->v_op = &dead_vnodeops; 2633 vp->v_tag = "none"; 2634 vp->v_type = VBAD; 2635 } 2636 2637 /* 2638 * Calculate the total number of references to a special device. 2639 */ 2640 int 2641 vcount(struct vnode *vp) 2642 { 2643 int count; 2644 2645 dev_lock(); 2646 count = vp->v_rdev->si_usecount; 2647 dev_unlock(); 2648 return (count); 2649 } 2650 2651 /* 2652 * Same as above, but using the struct cdev *as argument 2653 */ 2654 int 2655 count_dev(struct cdev *dev) 2656 { 2657 int count; 2658 2659 dev_lock(); 2660 count = dev->si_usecount; 2661 dev_unlock(); 2662 return(count); 2663 } 2664 2665 /* 2666 * Print out a description of a vnode. 2667 */ 2668 static char *typename[] = 2669 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", 2670 "VMARKER"}; 2671 2672 void 2673 vn_printf(struct vnode *vp, const char *fmt, ...) 2674 { 2675 va_list ap; 2676 char buf[256], buf2[16]; 2677 u_long flags; 2678 2679 va_start(ap, fmt); 2680 vprintf(fmt, ap); 2681 va_end(ap); 2682 printf("%p: ", (void *)vp); 2683 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]); 2684 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n", 2685 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere); 2686 buf[0] = '\0'; 2687 buf[1] = '\0'; 2688 if (vp->v_vflag & VV_ROOT) 2689 strlcat(buf, "|VV_ROOT", sizeof(buf)); 2690 if (vp->v_vflag & VV_ISTTY) 2691 strlcat(buf, "|VV_ISTTY", sizeof(buf)); 2692 if (vp->v_vflag & VV_NOSYNC) 2693 strlcat(buf, "|VV_NOSYNC", sizeof(buf)); 2694 if (vp->v_vflag & VV_CACHEDLABEL) 2695 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); 2696 if (vp->v_vflag & VV_TEXT) 2697 strlcat(buf, "|VV_TEXT", sizeof(buf)); 2698 if (vp->v_vflag & VV_COPYONWRITE) 2699 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); 2700 if (vp->v_vflag & VV_SYSTEM) 2701 strlcat(buf, "|VV_SYSTEM", sizeof(buf)); 2702 if (vp->v_vflag & VV_PROCDEP) 2703 strlcat(buf, "|VV_PROCDEP", sizeof(buf)); 2704 if (vp->v_vflag & VV_NOKNOTE) 2705 strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); 2706 if (vp->v_vflag & VV_DELETED) 2707 strlcat(buf, "|VV_DELETED", sizeof(buf)); 2708 if (vp->v_vflag & VV_MD) 2709 strlcat(buf, "|VV_MD", sizeof(buf)); 2710 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | 2711 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | 2712 VV_NOKNOTE | VV_DELETED | VV_MD); 2713 if (flags != 0) { 2714 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); 2715 strlcat(buf, buf2, sizeof(buf)); 2716 } 2717 if (vp->v_iflag & VI_MOUNT) 2718 strlcat(buf, "|VI_MOUNT", sizeof(buf)); 2719 if (vp->v_iflag & VI_AGE) 2720 strlcat(buf, "|VI_AGE", sizeof(buf)); 2721 if (vp->v_iflag & VI_DOOMED) 2722 strlcat(buf, "|VI_DOOMED", sizeof(buf)); 2723 if (vp->v_iflag & VI_FREE) 2724 strlcat(buf, "|VI_FREE", sizeof(buf)); 2725 if (vp->v_iflag & VI_DOINGINACT) 2726 strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); 2727 if (vp->v_iflag & VI_OWEINACT) 2728 strlcat(buf, "|VI_OWEINACT", sizeof(buf)); 2729 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE | 2730 VI_DOINGINACT | VI_OWEINACT); 2731 if (flags != 0) { 2732 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); 2733 strlcat(buf, buf2, sizeof(buf)); 2734 } 2735 printf(" flags (%s)\n", buf + 1); 2736 if (mtx_owned(VI_MTX(vp))) 2737 printf(" VI_LOCKed"); 2738 if (vp->v_object != NULL) 2739 printf(" v_object %p ref %d pages %d\n", 2740 vp->v_object, vp->v_object->ref_count, 2741 vp->v_object->resident_page_count); 2742 printf(" "); 2743 lockmgr_printinfo(vp->v_vnlock); 2744 if (vp->v_data != NULL) 2745 VOP_PRINT(vp); 2746 } 2747 2748 #ifdef DDB 2749 /* 2750 * List all of the locked vnodes in the system. 2751 * Called when debugging the kernel. 2752 */ 2753 DB_SHOW_COMMAND(lockedvnods, lockedvnodes) 2754 { 2755 struct mount *mp, *nmp; 2756 struct vnode *vp; 2757 2758 /* 2759 * Note: because this is DDB, we can't obey the locking semantics 2760 * for these structures, which means we could catch an inconsistent 2761 * state and dereference a nasty pointer. Not much to be done 2762 * about that. 2763 */ 2764 db_printf("Locked vnodes\n"); 2765 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2766 nmp = TAILQ_NEXT(mp, mnt_list); 2767 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2768 if (vp->v_type != VMARKER && 2769 VOP_ISLOCKED(vp)) 2770 vprint("", vp); 2771 } 2772 nmp = TAILQ_NEXT(mp, mnt_list); 2773 } 2774 } 2775 2776 /* 2777 * Show details about the given vnode. 2778 */ 2779 DB_SHOW_COMMAND(vnode, db_show_vnode) 2780 { 2781 struct vnode *vp; 2782 2783 if (!have_addr) 2784 return; 2785 vp = (struct vnode *)addr; 2786 vn_printf(vp, "vnode "); 2787 } 2788 2789 /* 2790 * Show details about the given mount point. 2791 */ 2792 DB_SHOW_COMMAND(mount, db_show_mount) 2793 { 2794 struct mount *mp; 2795 struct vfsopt *opt; 2796 struct statfs *sp; 2797 struct vnode *vp; 2798 char buf[512]; 2799 u_int flags; 2800 2801 if (!have_addr) { 2802 /* No address given, print short info about all mount points. */ 2803 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2804 db_printf("%p %s on %s (%s)\n", mp, 2805 mp->mnt_stat.f_mntfromname, 2806 mp->mnt_stat.f_mntonname, 2807 mp->mnt_stat.f_fstypename); 2808 if (db_pager_quit) 2809 break; 2810 } 2811 db_printf("\nMore info: show mount <addr>\n"); 2812 return; 2813 } 2814 2815 mp = (struct mount *)addr; 2816 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, 2817 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); 2818 2819 buf[0] = '\0'; 2820 flags = mp->mnt_flag; 2821 #define MNT_FLAG(flag) do { \ 2822 if (flags & (flag)) { \ 2823 if (buf[0] != '\0') \ 2824 strlcat(buf, ", ", sizeof(buf)); \ 2825 strlcat(buf, (#flag) + 4, sizeof(buf)); \ 2826 flags &= ~(flag); \ 2827 } \ 2828 } while (0) 2829 MNT_FLAG(MNT_RDONLY); 2830 MNT_FLAG(MNT_SYNCHRONOUS); 2831 MNT_FLAG(MNT_NOEXEC); 2832 MNT_FLAG(MNT_NOSUID); 2833 MNT_FLAG(MNT_UNION); 2834 MNT_FLAG(MNT_ASYNC); 2835 MNT_FLAG(MNT_SUIDDIR); 2836 MNT_FLAG(MNT_SOFTDEP); 2837 MNT_FLAG(MNT_NOSYMFOLLOW); 2838 MNT_FLAG(MNT_GJOURNAL); 2839 MNT_FLAG(MNT_MULTILABEL); 2840 MNT_FLAG(MNT_ACLS); 2841 MNT_FLAG(MNT_NOATIME); 2842 MNT_FLAG(MNT_NOCLUSTERR); 2843 MNT_FLAG(MNT_NOCLUSTERW); 2844 MNT_FLAG(MNT_NFS4ACLS); 2845 MNT_FLAG(MNT_EXRDONLY); 2846 MNT_FLAG(MNT_EXPORTED); 2847 MNT_FLAG(MNT_DEFEXPORTED); 2848 MNT_FLAG(MNT_EXPORTANON); 2849 MNT_FLAG(MNT_EXKERB); 2850 MNT_FLAG(MNT_EXPUBLIC); 2851 MNT_FLAG(MNT_LOCAL); 2852 MNT_FLAG(MNT_QUOTA); 2853 MNT_FLAG(MNT_ROOTFS); 2854 MNT_FLAG(MNT_USER); 2855 MNT_FLAG(MNT_IGNORE); 2856 MNT_FLAG(MNT_UPDATE); 2857 MNT_FLAG(MNT_DELEXPORT); 2858 MNT_FLAG(MNT_RELOAD); 2859 MNT_FLAG(MNT_FORCE); 2860 MNT_FLAG(MNT_SNAPSHOT); 2861 MNT_FLAG(MNT_BYFSID); 2862 MNT_FLAG(MNT_SOFTDEP); 2863 #undef MNT_FLAG 2864 if (flags != 0) { 2865 if (buf[0] != '\0') 2866 strlcat(buf, ", ", sizeof(buf)); 2867 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 2868 "0x%08x", flags); 2869 } 2870 db_printf(" mnt_flag = %s\n", buf); 2871 2872 buf[0] = '\0'; 2873 flags = mp->mnt_kern_flag; 2874 #define MNT_KERN_FLAG(flag) do { \ 2875 if (flags & (flag)) { \ 2876 if (buf[0] != '\0') \ 2877 strlcat(buf, ", ", sizeof(buf)); \ 2878 strlcat(buf, (#flag) + 5, sizeof(buf)); \ 2879 flags &= ~(flag); \ 2880 } \ 2881 } while (0) 2882 MNT_KERN_FLAG(MNTK_UNMOUNTF); 2883 MNT_KERN_FLAG(MNTK_ASYNC); 2884 MNT_KERN_FLAG(MNTK_SOFTDEP); 2885 MNT_KERN_FLAG(MNTK_NOINSMNTQ); 2886 MNT_KERN_FLAG(MNTK_DRAINING); 2887 MNT_KERN_FLAG(MNTK_REFEXPIRE); 2888 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); 2889 MNT_KERN_FLAG(MNTK_SHARED_WRITES); 2890 MNT_KERN_FLAG(MNTK_SUJ); 2891 MNT_KERN_FLAG(MNTK_UNMOUNT); 2892 MNT_KERN_FLAG(MNTK_MWAIT); 2893 MNT_KERN_FLAG(MNTK_SUSPEND); 2894 MNT_KERN_FLAG(MNTK_SUSPEND2); 2895 MNT_KERN_FLAG(MNTK_SUSPENDED); 2896 MNT_KERN_FLAG(MNTK_MPSAFE); 2897 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); 2898 MNT_KERN_FLAG(MNTK_NOKNOTE); 2899 #undef MNT_KERN_FLAG 2900 if (flags != 0) { 2901 if (buf[0] != '\0') 2902 strlcat(buf, ", ", sizeof(buf)); 2903 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 2904 "0x%08x", flags); 2905 } 2906 db_printf(" mnt_kern_flag = %s\n", buf); 2907 2908 db_printf(" mnt_opt = "); 2909 opt = TAILQ_FIRST(mp->mnt_opt); 2910 if (opt != NULL) { 2911 db_printf("%s", opt->name); 2912 opt = TAILQ_NEXT(opt, link); 2913 while (opt != NULL) { 2914 db_printf(", %s", opt->name); 2915 opt = TAILQ_NEXT(opt, link); 2916 } 2917 } 2918 db_printf("\n"); 2919 2920 sp = &mp->mnt_stat; 2921 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " 2922 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " 2923 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " 2924 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", 2925 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, 2926 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, 2927 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, 2928 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, 2929 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, 2930 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, 2931 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, 2932 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); 2933 2934 db_printf(" mnt_cred = { uid=%u ruid=%u", 2935 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); 2936 if (jailed(mp->mnt_cred)) 2937 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); 2938 db_printf(" }\n"); 2939 db_printf(" mnt_ref = %d\n", mp->mnt_ref); 2940 db_printf(" mnt_gen = %d\n", mp->mnt_gen); 2941 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); 2942 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount); 2943 db_printf(" mnt_noasync = %u\n", mp->mnt_noasync); 2944 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); 2945 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); 2946 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); 2947 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); 2948 db_printf(" mnt_secondary_accwrites = %d\n", 2949 mp->mnt_secondary_accwrites); 2950 db_printf(" mnt_gjprovider = %s\n", 2951 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); 2952 db_printf("\n"); 2953 2954 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2955 if (vp->v_type != VMARKER) { 2956 vn_printf(vp, "vnode "); 2957 if (db_pager_quit) 2958 break; 2959 } 2960 } 2961 } 2962 #endif /* DDB */ 2963 2964 /* 2965 * Fill in a struct xvfsconf based on a struct vfsconf. 2966 */ 2967 static void 2968 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp) 2969 { 2970 2971 strcpy(xvfsp->vfc_name, vfsp->vfc_name); 2972 xvfsp->vfc_typenum = vfsp->vfc_typenum; 2973 xvfsp->vfc_refcount = vfsp->vfc_refcount; 2974 xvfsp->vfc_flags = vfsp->vfc_flags; 2975 /* 2976 * These are unused in userland, we keep them 2977 * to not break binary compatibility. 2978 */ 2979 xvfsp->vfc_vfsops = NULL; 2980 xvfsp->vfc_next = NULL; 2981 } 2982 2983 /* 2984 * Top level filesystem related information gathering. 2985 */ 2986 static int 2987 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) 2988 { 2989 struct vfsconf *vfsp; 2990 struct xvfsconf xvfsp; 2991 int error; 2992 2993 error = 0; 2994 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 2995 bzero(&xvfsp, sizeof(xvfsp)); 2996 vfsconf2x(vfsp, &xvfsp); 2997 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp); 2998 if (error) 2999 break; 3000 } 3001 return (error); 3002 } 3003 3004 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD, 3005 NULL, 0, sysctl_vfs_conflist, 3006 "S,xvfsconf", "List of all configured filesystems"); 3007 3008 #ifndef BURN_BRIDGES 3009 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); 3010 3011 static int 3012 vfs_sysctl(SYSCTL_HANDLER_ARGS) 3013 { 3014 int *name = (int *)arg1 - 1; /* XXX */ 3015 u_int namelen = arg2 + 1; /* XXX */ 3016 struct vfsconf *vfsp; 3017 struct xvfsconf xvfsp; 3018 3019 printf("WARNING: userland calling deprecated sysctl, " 3020 "please rebuild world\n"); 3021 3022 #if 1 || defined(COMPAT_PRELITE2) 3023 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 3024 if (namelen == 1) 3025 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 3026 #endif 3027 3028 switch (name[1]) { 3029 case VFS_MAXTYPENUM: 3030 if (namelen != 2) 3031 return (ENOTDIR); 3032 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 3033 case VFS_CONF: 3034 if (namelen != 3) 3035 return (ENOTDIR); /* overloaded */ 3036 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) 3037 if (vfsp->vfc_typenum == name[2]) 3038 break; 3039 if (vfsp == NULL) 3040 return (EOPNOTSUPP); 3041 bzero(&xvfsp, sizeof(xvfsp)); 3042 vfsconf2x(vfsp, &xvfsp); 3043 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); 3044 } 3045 return (EOPNOTSUPP); 3046 } 3047 3048 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, 3049 vfs_sysctl, "Generic filesystem"); 3050 3051 #if 1 || defined(COMPAT_PRELITE2) 3052 3053 static int 3054 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 3055 { 3056 int error; 3057 struct vfsconf *vfsp; 3058 struct ovfsconf ovfs; 3059 3060 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 3061 bzero(&ovfs, sizeof(ovfs)); 3062 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 3063 strcpy(ovfs.vfc_name, vfsp->vfc_name); 3064 ovfs.vfc_index = vfsp->vfc_typenum; 3065 ovfs.vfc_refcount = vfsp->vfc_refcount; 3066 ovfs.vfc_flags = vfsp->vfc_flags; 3067 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 3068 if (error) 3069 return error; 3070 } 3071 return 0; 3072 } 3073 3074 #endif /* 1 || COMPAT_PRELITE2 */ 3075 #endif /* !BURN_BRIDGES */ 3076 3077 #define KINFO_VNODESLOP 10 3078 #ifdef notyet 3079 /* 3080 * Dump vnode list (via sysctl). 3081 */ 3082 /* ARGSUSED */ 3083 static int 3084 sysctl_vnode(SYSCTL_HANDLER_ARGS) 3085 { 3086 struct xvnode *xvn; 3087 struct mount *mp; 3088 struct vnode *vp; 3089 int error, len, n; 3090 3091 /* 3092 * Stale numvnodes access is not fatal here. 3093 */ 3094 req->lock = 0; 3095 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; 3096 if (!req->oldptr) 3097 /* Make an estimate */ 3098 return (SYSCTL_OUT(req, 0, len)); 3099 3100 error = sysctl_wire_old_buffer(req, 0); 3101 if (error != 0) 3102 return (error); 3103 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); 3104 n = 0; 3105 mtx_lock(&mountlist_mtx); 3106 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3107 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) 3108 continue; 3109 MNT_ILOCK(mp); 3110 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 3111 if (n == len) 3112 break; 3113 vref(vp); 3114 xvn[n].xv_size = sizeof *xvn; 3115 xvn[n].xv_vnode = vp; 3116 xvn[n].xv_id = 0; /* XXX compat */ 3117 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field 3118 XV_COPY(usecount); 3119 XV_COPY(writecount); 3120 XV_COPY(holdcnt); 3121 XV_COPY(mount); 3122 XV_COPY(numoutput); 3123 XV_COPY(type); 3124 #undef XV_COPY 3125 xvn[n].xv_flag = vp->v_vflag; 3126 3127 switch (vp->v_type) { 3128 case VREG: 3129 case VDIR: 3130 case VLNK: 3131 break; 3132 case VBLK: 3133 case VCHR: 3134 if (vp->v_rdev == NULL) { 3135 vrele(vp); 3136 continue; 3137 } 3138 xvn[n].xv_dev = dev2udev(vp->v_rdev); 3139 break; 3140 case VSOCK: 3141 xvn[n].xv_socket = vp->v_socket; 3142 break; 3143 case VFIFO: 3144 xvn[n].xv_fifo = vp->v_fifoinfo; 3145 break; 3146 case VNON: 3147 case VBAD: 3148 default: 3149 /* shouldn't happen? */ 3150 vrele(vp); 3151 continue; 3152 } 3153 vrele(vp); 3154 ++n; 3155 } 3156 MNT_IUNLOCK(mp); 3157 mtx_lock(&mountlist_mtx); 3158 vfs_unbusy(mp); 3159 if (n == len) 3160 break; 3161 } 3162 mtx_unlock(&mountlist_mtx); 3163 3164 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); 3165 free(xvn, M_TEMP); 3166 return (error); 3167 } 3168 3169 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 3170 0, 0, sysctl_vnode, "S,xvnode", ""); 3171 #endif 3172 3173 /* 3174 * Unmount all filesystems. The list is traversed in reverse order 3175 * of mounting to avoid dependencies. 3176 */ 3177 void 3178 vfs_unmountall(void) 3179 { 3180 struct mount *mp; 3181 struct thread *td; 3182 int error; 3183 3184 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread")); 3185 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); 3186 td = curthread; 3187 3188 /* 3189 * Since this only runs when rebooting, it is not interlocked. 3190 */ 3191 while(!TAILQ_EMPTY(&mountlist)) { 3192 mp = TAILQ_LAST(&mountlist, mntlist); 3193 error = dounmount(mp, MNT_FORCE, td); 3194 if (error) { 3195 TAILQ_REMOVE(&mountlist, mp, mnt_list); 3196 /* 3197 * XXX: Due to the way in which we mount the root 3198 * file system off of devfs, devfs will generate a 3199 * "busy" warning when we try to unmount it before 3200 * the root. Don't print a warning as a result in 3201 * order to avoid false positive errors that may 3202 * cause needless upset. 3203 */ 3204 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) { 3205 printf("unmount of %s failed (", 3206 mp->mnt_stat.f_mntonname); 3207 if (error == EBUSY) 3208 printf("BUSY)\n"); 3209 else 3210 printf("%d)\n", error); 3211 } 3212 } else { 3213 /* The unmount has removed mp from the mountlist */ 3214 } 3215 } 3216 } 3217 3218 /* 3219 * perform msync on all vnodes under a mount point 3220 * the mount point must be locked. 3221 */ 3222 void 3223 vfs_msync(struct mount *mp, int flags) 3224 { 3225 struct vnode *vp, *mvp; 3226 struct vm_object *obj; 3227 3228 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 3229 MNT_ILOCK(mp); 3230 MNT_VNODE_FOREACH(vp, mp, mvp) { 3231 VI_LOCK(vp); 3232 obj = vp->v_object; 3233 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 && 3234 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) { 3235 MNT_IUNLOCK(mp); 3236 if (!vget(vp, 3237 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 3238 curthread)) { 3239 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ 3240 vput(vp); 3241 MNT_ILOCK(mp); 3242 continue; 3243 } 3244 3245 obj = vp->v_object; 3246 if (obj != NULL) { 3247 VM_OBJECT_LOCK(obj); 3248 vm_object_page_clean(obj, 0, 0, 3249 flags == MNT_WAIT ? 3250 OBJPC_SYNC : OBJPC_NOSYNC); 3251 VM_OBJECT_UNLOCK(obj); 3252 } 3253 vput(vp); 3254 } 3255 MNT_ILOCK(mp); 3256 } else 3257 VI_UNLOCK(vp); 3258 } 3259 MNT_IUNLOCK(mp); 3260 } 3261 3262 /* 3263 * Mark a vnode as free, putting it up for recycling. 3264 */ 3265 static void 3266 vfree(struct vnode *vp) 3267 { 3268 3269 ASSERT_VI_LOCKED(vp, "vfree"); 3270 mtx_lock(&vnode_free_list_mtx); 3271 VNASSERT(vp->v_op != NULL, vp, ("vfree: vnode already reclaimed.")); 3272 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, ("vnode already free")); 3273 VNASSERT(VSHOULDFREE(vp), vp, ("vfree: freeing when we shouldn't")); 3274 VNASSERT((vp->v_iflag & VI_DOOMED) == 0, vp, 3275 ("vfree: Freeing doomed vnode")); 3276 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 3277 if (vp->v_iflag & VI_AGE) { 3278 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist); 3279 } else { 3280 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist); 3281 } 3282 freevnodes++; 3283 vp->v_iflag &= ~VI_AGE; 3284 vp->v_iflag |= VI_FREE; 3285 mtx_unlock(&vnode_free_list_mtx); 3286 } 3287 3288 /* 3289 * Opposite of vfree() - mark a vnode as in use. 3290 */ 3291 static void 3292 vbusy(struct vnode *vp) 3293 { 3294 ASSERT_VI_LOCKED(vp, "vbusy"); 3295 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free")); 3296 VNASSERT(vp->v_op != NULL, vp, ("vbusy: vnode already reclaimed.")); 3297 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 3298 3299 mtx_lock(&vnode_free_list_mtx); 3300 TAILQ_REMOVE(&vnode_free_list, vp, v_freelist); 3301 freevnodes--; 3302 vp->v_iflag &= ~(VI_FREE|VI_AGE); 3303 mtx_unlock(&vnode_free_list_mtx); 3304 } 3305 3306 static void 3307 destroy_vpollinfo(struct vpollinfo *vi) 3308 { 3309 knlist_destroy(&vi->vpi_selinfo.si_note); 3310 mtx_destroy(&vi->vpi_lock); 3311 uma_zfree(vnodepoll_zone, vi); 3312 } 3313 3314 /* 3315 * Initalize per-vnode helper structure to hold poll-related state. 3316 */ 3317 void 3318 v_addpollinfo(struct vnode *vp) 3319 { 3320 struct vpollinfo *vi; 3321 3322 if (vp->v_pollinfo != NULL) 3323 return; 3324 vi = uma_zalloc(vnodepoll_zone, M_WAITOK); 3325 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); 3326 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, 3327 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked); 3328 VI_LOCK(vp); 3329 if (vp->v_pollinfo != NULL) { 3330 VI_UNLOCK(vp); 3331 destroy_vpollinfo(vi); 3332 return; 3333 } 3334 vp->v_pollinfo = vi; 3335 VI_UNLOCK(vp); 3336 } 3337 3338 /* 3339 * Record a process's interest in events which might happen to 3340 * a vnode. Because poll uses the historic select-style interface 3341 * internally, this routine serves as both the ``check for any 3342 * pending events'' and the ``record my interest in future events'' 3343 * functions. (These are done together, while the lock is held, 3344 * to avoid race conditions.) 3345 */ 3346 int 3347 vn_pollrecord(struct vnode *vp, struct thread *td, int events) 3348 { 3349 3350 v_addpollinfo(vp); 3351 mtx_lock(&vp->v_pollinfo->vpi_lock); 3352 if (vp->v_pollinfo->vpi_revents & events) { 3353 /* 3354 * This leaves events we are not interested 3355 * in available for the other process which 3356 * which presumably had requested them 3357 * (otherwise they would never have been 3358 * recorded). 3359 */ 3360 events &= vp->v_pollinfo->vpi_revents; 3361 vp->v_pollinfo->vpi_revents &= ~events; 3362 3363 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3364 return (events); 3365 } 3366 vp->v_pollinfo->vpi_events |= events; 3367 selrecord(td, &vp->v_pollinfo->vpi_selinfo); 3368 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3369 return (0); 3370 } 3371 3372 /* 3373 * Routine to create and manage a filesystem syncer vnode. 3374 */ 3375 #define sync_close ((int (*)(struct vop_close_args *))nullop) 3376 static int sync_fsync(struct vop_fsync_args *); 3377 static int sync_inactive(struct vop_inactive_args *); 3378 static int sync_reclaim(struct vop_reclaim_args *); 3379 3380 static struct vop_vector sync_vnodeops = { 3381 .vop_bypass = VOP_EOPNOTSUPP, 3382 .vop_close = sync_close, /* close */ 3383 .vop_fsync = sync_fsync, /* fsync */ 3384 .vop_inactive = sync_inactive, /* inactive */ 3385 .vop_reclaim = sync_reclaim, /* reclaim */ 3386 .vop_lock1 = vop_stdlock, /* lock */ 3387 .vop_unlock = vop_stdunlock, /* unlock */ 3388 .vop_islocked = vop_stdislocked, /* islocked */ 3389 }; 3390 3391 /* 3392 * Create a new filesystem syncer vnode for the specified mount point. 3393 */ 3394 void 3395 vfs_allocate_syncvnode(struct mount *mp) 3396 { 3397 struct vnode *vp; 3398 struct bufobj *bo; 3399 static long start, incr, next; 3400 int error; 3401 3402 /* Allocate a new vnode */ 3403 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); 3404 if (error != 0) 3405 panic("vfs_allocate_syncvnode: getnewvnode() failed"); 3406 vp->v_type = VNON; 3407 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3408 vp->v_vflag |= VV_FORCEINSMQ; 3409 error = insmntque(vp, mp); 3410 if (error != 0) 3411 panic("vfs_allocate_syncvnode: insmntque() failed"); 3412 vp->v_vflag &= ~VV_FORCEINSMQ; 3413 VOP_UNLOCK(vp, 0); 3414 /* 3415 * Place the vnode onto the syncer worklist. We attempt to 3416 * scatter them about on the list so that they will go off 3417 * at evenly distributed times even if all the filesystems 3418 * are mounted at once. 3419 */ 3420 next += incr; 3421 if (next == 0 || next > syncer_maxdelay) { 3422 start /= 2; 3423 incr /= 2; 3424 if (start == 0) { 3425 start = syncer_maxdelay / 2; 3426 incr = syncer_maxdelay; 3427 } 3428 next = start; 3429 } 3430 bo = &vp->v_bufobj; 3431 BO_LOCK(bo); 3432 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); 3433 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ 3434 mtx_lock(&sync_mtx); 3435 sync_vnode_count++; 3436 if (mp->mnt_syncer == NULL) { 3437 mp->mnt_syncer = vp; 3438 vp = NULL; 3439 } 3440 mtx_unlock(&sync_mtx); 3441 BO_UNLOCK(bo); 3442 if (vp != NULL) { 3443 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3444 vgone(vp); 3445 vput(vp); 3446 } 3447 } 3448 3449 void 3450 vfs_deallocate_syncvnode(struct mount *mp) 3451 { 3452 struct vnode *vp; 3453 3454 mtx_lock(&sync_mtx); 3455 vp = mp->mnt_syncer; 3456 if (vp != NULL) 3457 mp->mnt_syncer = NULL; 3458 mtx_unlock(&sync_mtx); 3459 if (vp != NULL) 3460 vrele(vp); 3461 } 3462 3463 /* 3464 * Do a lazy sync of the filesystem. 3465 */ 3466 static int 3467 sync_fsync(struct vop_fsync_args *ap) 3468 { 3469 struct vnode *syncvp = ap->a_vp; 3470 struct mount *mp = syncvp->v_mount; 3471 int error; 3472 struct bufobj *bo; 3473 3474 /* 3475 * We only need to do something if this is a lazy evaluation. 3476 */ 3477 if (ap->a_waitfor != MNT_LAZY) 3478 return (0); 3479 3480 /* 3481 * Move ourselves to the back of the sync list. 3482 */ 3483 bo = &syncvp->v_bufobj; 3484 BO_LOCK(bo); 3485 vn_syncer_add_to_worklist(bo, syncdelay); 3486 BO_UNLOCK(bo); 3487 3488 /* 3489 * Walk the list of vnodes pushing all that are dirty and 3490 * not already on the sync list. 3491 */ 3492 mtx_lock(&mountlist_mtx); 3493 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) { 3494 mtx_unlock(&mountlist_mtx); 3495 return (0); 3496 } 3497 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 3498 vfs_unbusy(mp); 3499 return (0); 3500 } 3501 MNT_ILOCK(mp); 3502 mp->mnt_noasync++; 3503 mp->mnt_kern_flag &= ~MNTK_ASYNC; 3504 MNT_IUNLOCK(mp); 3505 vfs_msync(mp, MNT_NOWAIT); 3506 error = VFS_SYNC(mp, MNT_LAZY); 3507 MNT_ILOCK(mp); 3508 mp->mnt_noasync--; 3509 if ((mp->mnt_flag & MNT_ASYNC) != 0 && mp->mnt_noasync == 0) 3510 mp->mnt_kern_flag |= MNTK_ASYNC; 3511 MNT_IUNLOCK(mp); 3512 vn_finished_write(mp); 3513 vfs_unbusy(mp); 3514 return (error); 3515 } 3516 3517 /* 3518 * The syncer vnode is no referenced. 3519 */ 3520 static int 3521 sync_inactive(struct vop_inactive_args *ap) 3522 { 3523 3524 vgone(ap->a_vp); 3525 return (0); 3526 } 3527 3528 /* 3529 * The syncer vnode is no longer needed and is being decommissioned. 3530 * 3531 * Modifications to the worklist must be protected by sync_mtx. 3532 */ 3533 static int 3534 sync_reclaim(struct vop_reclaim_args *ap) 3535 { 3536 struct vnode *vp = ap->a_vp; 3537 struct bufobj *bo; 3538 3539 bo = &vp->v_bufobj; 3540 BO_LOCK(bo); 3541 mtx_lock(&sync_mtx); 3542 if (vp->v_mount->mnt_syncer == vp) 3543 vp->v_mount->mnt_syncer = NULL; 3544 if (bo->bo_flag & BO_ONWORKLST) { 3545 LIST_REMOVE(bo, bo_synclist); 3546 syncer_worklist_len--; 3547 sync_vnode_count--; 3548 bo->bo_flag &= ~BO_ONWORKLST; 3549 } 3550 mtx_unlock(&sync_mtx); 3551 BO_UNLOCK(bo); 3552 3553 return (0); 3554 } 3555 3556 /* 3557 * Check if vnode represents a disk device 3558 */ 3559 int 3560 vn_isdisk(struct vnode *vp, int *errp) 3561 { 3562 int error; 3563 3564 error = 0; 3565 dev_lock(); 3566 if (vp->v_type != VCHR) 3567 error = ENOTBLK; 3568 else if (vp->v_rdev == NULL) 3569 error = ENXIO; 3570 else if (vp->v_rdev->si_devsw == NULL) 3571 error = ENXIO; 3572 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) 3573 error = ENOTBLK; 3574 dev_unlock(); 3575 if (errp != NULL) 3576 *errp = error; 3577 return (error == 0); 3578 } 3579 3580 /* 3581 * Common filesystem object access control check routine. Accepts a 3582 * vnode's type, "mode", uid and gid, requested access mode, credentials, 3583 * and optional call-by-reference privused argument allowing vaccess() 3584 * to indicate to the caller whether privilege was used to satisfy the 3585 * request (obsoleted). Returns 0 on success, or an errno on failure. 3586 * 3587 * The ifdef'd CAPABILITIES version is here for reference, but is not 3588 * actually used. 3589 */ 3590 int 3591 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, 3592 accmode_t accmode, struct ucred *cred, int *privused) 3593 { 3594 accmode_t dac_granted; 3595 accmode_t priv_granted; 3596 3597 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, 3598 ("invalid bit in accmode")); 3599 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), 3600 ("VAPPEND without VWRITE")); 3601 3602 /* 3603 * Look for a normal, non-privileged way to access the file/directory 3604 * as requested. If it exists, go with that. 3605 */ 3606 3607 if (privused != NULL) 3608 *privused = 0; 3609 3610 dac_granted = 0; 3611 3612 /* Check the owner. */ 3613 if (cred->cr_uid == file_uid) { 3614 dac_granted |= VADMIN; 3615 if (file_mode & S_IXUSR) 3616 dac_granted |= VEXEC; 3617 if (file_mode & S_IRUSR) 3618 dac_granted |= VREAD; 3619 if (file_mode & S_IWUSR) 3620 dac_granted |= (VWRITE | VAPPEND); 3621 3622 if ((accmode & dac_granted) == accmode) 3623 return (0); 3624 3625 goto privcheck; 3626 } 3627 3628 /* Otherwise, check the groups (first match) */ 3629 if (groupmember(file_gid, cred)) { 3630 if (file_mode & S_IXGRP) 3631 dac_granted |= VEXEC; 3632 if (file_mode & S_IRGRP) 3633 dac_granted |= VREAD; 3634 if (file_mode & S_IWGRP) 3635 dac_granted |= (VWRITE | VAPPEND); 3636 3637 if ((accmode & dac_granted) == accmode) 3638 return (0); 3639 3640 goto privcheck; 3641 } 3642 3643 /* Otherwise, check everyone else. */ 3644 if (file_mode & S_IXOTH) 3645 dac_granted |= VEXEC; 3646 if (file_mode & S_IROTH) 3647 dac_granted |= VREAD; 3648 if (file_mode & S_IWOTH) 3649 dac_granted |= (VWRITE | VAPPEND); 3650 if ((accmode & dac_granted) == accmode) 3651 return (0); 3652 3653 privcheck: 3654 /* 3655 * Build a privilege mask to determine if the set of privileges 3656 * satisfies the requirements when combined with the granted mask 3657 * from above. For each privilege, if the privilege is required, 3658 * bitwise or the request type onto the priv_granted mask. 3659 */ 3660 priv_granted = 0; 3661 3662 if (type == VDIR) { 3663 /* 3664 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC 3665 * requests, instead of PRIV_VFS_EXEC. 3666 */ 3667 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3668 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0)) 3669 priv_granted |= VEXEC; 3670 } else { 3671 /* 3672 * Ensure that at least one execute bit is on. Otherwise, 3673 * a privileged user will always succeed, and we don't want 3674 * this to happen unless the file really is executable. 3675 */ 3676 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3677 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && 3678 !priv_check_cred(cred, PRIV_VFS_EXEC, 0)) 3679 priv_granted |= VEXEC; 3680 } 3681 3682 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && 3683 !priv_check_cred(cred, PRIV_VFS_READ, 0)) 3684 priv_granted |= VREAD; 3685 3686 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && 3687 !priv_check_cred(cred, PRIV_VFS_WRITE, 0)) 3688 priv_granted |= (VWRITE | VAPPEND); 3689 3690 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && 3691 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0)) 3692 priv_granted |= VADMIN; 3693 3694 if ((accmode & (priv_granted | dac_granted)) == accmode) { 3695 /* XXX audit: privilege used */ 3696 if (privused != NULL) 3697 *privused = 1; 3698 return (0); 3699 } 3700 3701 return ((accmode & VADMIN) ? EPERM : EACCES); 3702 } 3703 3704 /* 3705 * Credential check based on process requesting service, and per-attribute 3706 * permissions. 3707 */ 3708 int 3709 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, 3710 struct thread *td, accmode_t accmode) 3711 { 3712 3713 /* 3714 * Kernel-invoked always succeeds. 3715 */ 3716 if (cred == NOCRED) 3717 return (0); 3718 3719 /* 3720 * Do not allow privileged processes in jail to directly manipulate 3721 * system attributes. 3722 */ 3723 switch (attrnamespace) { 3724 case EXTATTR_NAMESPACE_SYSTEM: 3725 /* Potentially should be: return (EPERM); */ 3726 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0)); 3727 case EXTATTR_NAMESPACE_USER: 3728 return (VOP_ACCESS(vp, accmode, cred, td)); 3729 default: 3730 return (EPERM); 3731 } 3732 } 3733 3734 #ifdef DEBUG_VFS_LOCKS 3735 /* 3736 * This only exists to supress warnings from unlocked specfs accesses. It is 3737 * no longer ok to have an unlocked VFS. 3738 */ 3739 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \ 3740 (vp)->v_type == VCHR || (vp)->v_type == VBAD) 3741 3742 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ 3743 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, 3744 "Drop into debugger on lock violation"); 3745 3746 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ 3747 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 3748 0, "Check for interlock across VOPs"); 3749 3750 int vfs_badlock_print = 1; /* Print lock violations. */ 3751 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 3752 0, "Print lock violations"); 3753 3754 #ifdef KDB 3755 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ 3756 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, 3757 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations"); 3758 #endif 3759 3760 static void 3761 vfs_badlock(const char *msg, const char *str, struct vnode *vp) 3762 { 3763 3764 #ifdef KDB 3765 if (vfs_badlock_backtrace) 3766 kdb_backtrace(); 3767 #endif 3768 if (vfs_badlock_print) 3769 printf("%s: %p %s\n", str, (void *)vp, msg); 3770 if (vfs_badlock_ddb) 3771 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3772 } 3773 3774 void 3775 assert_vi_locked(struct vnode *vp, const char *str) 3776 { 3777 3778 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) 3779 vfs_badlock("interlock is not locked but should be", str, vp); 3780 } 3781 3782 void 3783 assert_vi_unlocked(struct vnode *vp, const char *str) 3784 { 3785 3786 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) 3787 vfs_badlock("interlock is locked but should not be", str, vp); 3788 } 3789 3790 void 3791 assert_vop_locked(struct vnode *vp, const char *str) 3792 { 3793 3794 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0) 3795 vfs_badlock("is not locked but should be", str, vp); 3796 } 3797 3798 void 3799 assert_vop_unlocked(struct vnode *vp, const char *str) 3800 { 3801 3802 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) 3803 vfs_badlock("is locked but should not be", str, vp); 3804 } 3805 3806 void 3807 assert_vop_elocked(struct vnode *vp, const char *str) 3808 { 3809 3810 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 3811 vfs_badlock("is not exclusive locked but should be", str, vp); 3812 } 3813 3814 #if 0 3815 void 3816 assert_vop_elocked_other(struct vnode *vp, const char *str) 3817 { 3818 3819 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER) 3820 vfs_badlock("is not exclusive locked by another thread", 3821 str, vp); 3822 } 3823 3824 void 3825 assert_vop_slocked(struct vnode *vp, const char *str) 3826 { 3827 3828 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED) 3829 vfs_badlock("is not locked shared but should be", str, vp); 3830 } 3831 #endif /* 0 */ 3832 #endif /* DEBUG_VFS_LOCKS */ 3833 3834 void 3835 vop_rename_fail(struct vop_rename_args *ap) 3836 { 3837 3838 if (ap->a_tvp != NULL) 3839 vput(ap->a_tvp); 3840 if (ap->a_tdvp == ap->a_tvp) 3841 vrele(ap->a_tdvp); 3842 else 3843 vput(ap->a_tdvp); 3844 vrele(ap->a_fdvp); 3845 vrele(ap->a_fvp); 3846 } 3847 3848 void 3849 vop_rename_pre(void *ap) 3850 { 3851 struct vop_rename_args *a = ap; 3852 3853 #ifdef DEBUG_VFS_LOCKS 3854 if (a->a_tvp) 3855 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); 3856 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); 3857 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); 3858 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); 3859 3860 /* Check the source (from). */ 3861 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && 3862 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) 3863 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); 3864 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) 3865 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); 3866 3867 /* Check the target. */ 3868 if (a->a_tvp) 3869 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); 3870 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); 3871 #endif 3872 if (a->a_tdvp != a->a_fdvp) 3873 vhold(a->a_fdvp); 3874 if (a->a_tvp != a->a_fvp) 3875 vhold(a->a_fvp); 3876 vhold(a->a_tdvp); 3877 if (a->a_tvp) 3878 vhold(a->a_tvp); 3879 } 3880 3881 void 3882 vop_strategy_pre(void *ap) 3883 { 3884 #ifdef DEBUG_VFS_LOCKS 3885 struct vop_strategy_args *a; 3886 struct buf *bp; 3887 3888 a = ap; 3889 bp = a->a_bp; 3890 3891 /* 3892 * Cluster ops lock their component buffers but not the IO container. 3893 */ 3894 if ((bp->b_flags & B_CLUSTER) != 0) 3895 return; 3896 3897 if (panicstr == NULL && !BUF_ISLOCKED(bp)) { 3898 if (vfs_badlock_print) 3899 printf( 3900 "VOP_STRATEGY: bp is not locked but should be\n"); 3901 if (vfs_badlock_ddb) 3902 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3903 } 3904 #endif 3905 } 3906 3907 void 3908 vop_lookup_pre(void *ap) 3909 { 3910 #ifdef DEBUG_VFS_LOCKS 3911 struct vop_lookup_args *a; 3912 struct vnode *dvp; 3913 3914 a = ap; 3915 dvp = a->a_dvp; 3916 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3917 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3918 #endif 3919 } 3920 3921 void 3922 vop_lookup_post(void *ap, int rc) 3923 { 3924 #ifdef DEBUG_VFS_LOCKS 3925 struct vop_lookup_args *a; 3926 struct vnode *dvp; 3927 struct vnode *vp; 3928 3929 a = ap; 3930 dvp = a->a_dvp; 3931 vp = *(a->a_vpp); 3932 3933 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 3934 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 3935 3936 if (!rc) 3937 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)"); 3938 #endif 3939 } 3940 3941 void 3942 vop_lock_pre(void *ap) 3943 { 3944 #ifdef DEBUG_VFS_LOCKS 3945 struct vop_lock1_args *a = ap; 3946 3947 if ((a->a_flags & LK_INTERLOCK) == 0) 3948 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3949 else 3950 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); 3951 #endif 3952 } 3953 3954 void 3955 vop_lock_post(void *ap, int rc) 3956 { 3957 #ifdef DEBUG_VFS_LOCKS 3958 struct vop_lock1_args *a = ap; 3959 3960 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 3961 if (rc == 0) 3962 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); 3963 #endif 3964 } 3965 3966 void 3967 vop_unlock_pre(void *ap) 3968 { 3969 #ifdef DEBUG_VFS_LOCKS 3970 struct vop_unlock_args *a = ap; 3971 3972 if (a->a_flags & LK_INTERLOCK) 3973 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); 3974 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); 3975 #endif 3976 } 3977 3978 void 3979 vop_unlock_post(void *ap, int rc) 3980 { 3981 #ifdef DEBUG_VFS_LOCKS 3982 struct vop_unlock_args *a = ap; 3983 3984 if (a->a_flags & LK_INTERLOCK) 3985 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); 3986 #endif 3987 } 3988 3989 void 3990 vop_create_post(void *ap, int rc) 3991 { 3992 struct vop_create_args *a = ap; 3993 3994 if (!rc) 3995 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 3996 } 3997 3998 void 3999 vop_link_post(void *ap, int rc) 4000 { 4001 struct vop_link_args *a = ap; 4002 4003 if (!rc) { 4004 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); 4005 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); 4006 } 4007 } 4008 4009 void 4010 vop_mkdir_post(void *ap, int rc) 4011 { 4012 struct vop_mkdir_args *a = ap; 4013 4014 if (!rc) 4015 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 4016 } 4017 4018 void 4019 vop_mknod_post(void *ap, int rc) 4020 { 4021 struct vop_mknod_args *a = ap; 4022 4023 if (!rc) 4024 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4025 } 4026 4027 void 4028 vop_remove_post(void *ap, int rc) 4029 { 4030 struct vop_remove_args *a = ap; 4031 4032 if (!rc) { 4033 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4034 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 4035 } 4036 } 4037 4038 void 4039 vop_rename_post(void *ap, int rc) 4040 { 4041 struct vop_rename_args *a = ap; 4042 4043 if (!rc) { 4044 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE); 4045 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE); 4046 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); 4047 if (a->a_tvp) 4048 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); 4049 } 4050 if (a->a_tdvp != a->a_fdvp) 4051 vdrop(a->a_fdvp); 4052 if (a->a_tvp != a->a_fvp) 4053 vdrop(a->a_fvp); 4054 vdrop(a->a_tdvp); 4055 if (a->a_tvp) 4056 vdrop(a->a_tvp); 4057 } 4058 4059 void 4060 vop_rmdir_post(void *ap, int rc) 4061 { 4062 struct vop_rmdir_args *a = ap; 4063 4064 if (!rc) { 4065 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 4066 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 4067 } 4068 } 4069 4070 void 4071 vop_setattr_post(void *ap, int rc) 4072 { 4073 struct vop_setattr_args *a = ap; 4074 4075 if (!rc) 4076 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 4077 } 4078 4079 void 4080 vop_symlink_post(void *ap, int rc) 4081 { 4082 struct vop_symlink_args *a = ap; 4083 4084 if (!rc) 4085 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4086 } 4087 4088 static struct knlist fs_knlist; 4089 4090 static void 4091 vfs_event_init(void *arg) 4092 { 4093 knlist_init_mtx(&fs_knlist, NULL); 4094 } 4095 /* XXX - correct order? */ 4096 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); 4097 4098 void 4099 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) 4100 { 4101 4102 KNOTE_UNLOCKED(&fs_knlist, event); 4103 } 4104 4105 static int filt_fsattach(struct knote *kn); 4106 static void filt_fsdetach(struct knote *kn); 4107 static int filt_fsevent(struct knote *kn, long hint); 4108 4109 struct filterops fs_filtops = { 4110 .f_isfd = 0, 4111 .f_attach = filt_fsattach, 4112 .f_detach = filt_fsdetach, 4113 .f_event = filt_fsevent 4114 }; 4115 4116 static int 4117 filt_fsattach(struct knote *kn) 4118 { 4119 4120 kn->kn_flags |= EV_CLEAR; 4121 knlist_add(&fs_knlist, kn, 0); 4122 return (0); 4123 } 4124 4125 static void 4126 filt_fsdetach(struct knote *kn) 4127 { 4128 4129 knlist_remove(&fs_knlist, kn, 0); 4130 } 4131 4132 static int 4133 filt_fsevent(struct knote *kn, long hint) 4134 { 4135 4136 kn->kn_fflags |= hint; 4137 return (kn->kn_fflags != 0); 4138 } 4139 4140 static int 4141 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) 4142 { 4143 struct vfsidctl vc; 4144 int error; 4145 struct mount *mp; 4146 4147 error = SYSCTL_IN(req, &vc, sizeof(vc)); 4148 if (error) 4149 return (error); 4150 if (vc.vc_vers != VFS_CTL_VERS1) 4151 return (EINVAL); 4152 mp = vfs_getvfs(&vc.vc_fsid); 4153 if (mp == NULL) 4154 return (ENOENT); 4155 /* ensure that a specific sysctl goes to the right filesystem. */ 4156 if (strcmp(vc.vc_fstypename, "*") != 0 && 4157 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { 4158 vfs_rel(mp); 4159 return (EINVAL); 4160 } 4161 VCTLTOREQ(&vc, req); 4162 error = VFS_SYSCTL(mp, vc.vc_op, req); 4163 vfs_rel(mp); 4164 return (error); 4165 } 4166 4167 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR, 4168 NULL, 0, sysctl_vfs_ctl, "", 4169 "Sysctl by fsid"); 4170 4171 /* 4172 * Function to initialize a va_filerev field sensibly. 4173 * XXX: Wouldn't a random number make a lot more sense ?? 4174 */ 4175 u_quad_t 4176 init_va_filerev(void) 4177 { 4178 struct bintime bt; 4179 4180 getbinuptime(&bt); 4181 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); 4182 } 4183 4184 static int filt_vfsread(struct knote *kn, long hint); 4185 static int filt_vfswrite(struct knote *kn, long hint); 4186 static int filt_vfsvnode(struct knote *kn, long hint); 4187 static void filt_vfsdetach(struct knote *kn); 4188 static struct filterops vfsread_filtops = { 4189 .f_isfd = 1, 4190 .f_detach = filt_vfsdetach, 4191 .f_event = filt_vfsread 4192 }; 4193 static struct filterops vfswrite_filtops = { 4194 .f_isfd = 1, 4195 .f_detach = filt_vfsdetach, 4196 .f_event = filt_vfswrite 4197 }; 4198 static struct filterops vfsvnode_filtops = { 4199 .f_isfd = 1, 4200 .f_detach = filt_vfsdetach, 4201 .f_event = filt_vfsvnode 4202 }; 4203 4204 static void 4205 vfs_knllock(void *arg) 4206 { 4207 struct vnode *vp = arg; 4208 4209 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4210 } 4211 4212 static void 4213 vfs_knlunlock(void *arg) 4214 { 4215 struct vnode *vp = arg; 4216 4217 VOP_UNLOCK(vp, 0); 4218 } 4219 4220 static void 4221 vfs_knl_assert_locked(void *arg) 4222 { 4223 #ifdef DEBUG_VFS_LOCKS 4224 struct vnode *vp = arg; 4225 4226 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); 4227 #endif 4228 } 4229 4230 static void 4231 vfs_knl_assert_unlocked(void *arg) 4232 { 4233 #ifdef DEBUG_VFS_LOCKS 4234 struct vnode *vp = arg; 4235 4236 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); 4237 #endif 4238 } 4239 4240 int 4241 vfs_kqfilter(struct vop_kqfilter_args *ap) 4242 { 4243 struct vnode *vp = ap->a_vp; 4244 struct knote *kn = ap->a_kn; 4245 struct knlist *knl; 4246 4247 switch (kn->kn_filter) { 4248 case EVFILT_READ: 4249 kn->kn_fop = &vfsread_filtops; 4250 break; 4251 case EVFILT_WRITE: 4252 kn->kn_fop = &vfswrite_filtops; 4253 break; 4254 case EVFILT_VNODE: 4255 kn->kn_fop = &vfsvnode_filtops; 4256 break; 4257 default: 4258 return (EINVAL); 4259 } 4260 4261 kn->kn_hook = (caddr_t)vp; 4262 4263 v_addpollinfo(vp); 4264 if (vp->v_pollinfo == NULL) 4265 return (ENOMEM); 4266 knl = &vp->v_pollinfo->vpi_selinfo.si_note; 4267 knlist_add(knl, kn, 0); 4268 4269 return (0); 4270 } 4271 4272 /* 4273 * Detach knote from vnode 4274 */ 4275 static void 4276 filt_vfsdetach(struct knote *kn) 4277 { 4278 struct vnode *vp = (struct vnode *)kn->kn_hook; 4279 4280 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); 4281 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); 4282 } 4283 4284 /*ARGSUSED*/ 4285 static int 4286 filt_vfsread(struct knote *kn, long hint) 4287 { 4288 struct vnode *vp = (struct vnode *)kn->kn_hook; 4289 struct vattr va; 4290 int res; 4291 4292 /* 4293 * filesystem is gone, so set the EOF flag and schedule 4294 * the knote for deletion. 4295 */ 4296 if (hint == NOTE_REVOKE) { 4297 VI_LOCK(vp); 4298 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4299 VI_UNLOCK(vp); 4300 return (1); 4301 } 4302 4303 if (VOP_GETATTR(vp, &va, curthread->td_ucred)) 4304 return (0); 4305 4306 VI_LOCK(vp); 4307 kn->kn_data = va.va_size - kn->kn_fp->f_offset; 4308 res = (kn->kn_data != 0); 4309 VI_UNLOCK(vp); 4310 return (res); 4311 } 4312 4313 /*ARGSUSED*/ 4314 static int 4315 filt_vfswrite(struct knote *kn, long hint) 4316 { 4317 struct vnode *vp = (struct vnode *)kn->kn_hook; 4318 4319 VI_LOCK(vp); 4320 4321 /* 4322 * filesystem is gone, so set the EOF flag and schedule 4323 * the knote for deletion. 4324 */ 4325 if (hint == NOTE_REVOKE) 4326 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4327 4328 kn->kn_data = 0; 4329 VI_UNLOCK(vp); 4330 return (1); 4331 } 4332 4333 static int 4334 filt_vfsvnode(struct knote *kn, long hint) 4335 { 4336 struct vnode *vp = (struct vnode *)kn->kn_hook; 4337 int res; 4338 4339 VI_LOCK(vp); 4340 if (kn->kn_sfflags & hint) 4341 kn->kn_fflags |= hint; 4342 if (hint == NOTE_REVOKE) { 4343 kn->kn_flags |= EV_EOF; 4344 VI_UNLOCK(vp); 4345 return (1); 4346 } 4347 res = (kn->kn_fflags != 0); 4348 VI_UNLOCK(vp); 4349 return (res); 4350 } 4351 4352 int 4353 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) 4354 { 4355 int error; 4356 4357 if (dp->d_reclen > ap->a_uio->uio_resid) 4358 return (ENAMETOOLONG); 4359 error = uiomove(dp, dp->d_reclen, ap->a_uio); 4360 if (error) { 4361 if (ap->a_ncookies != NULL) { 4362 if (ap->a_cookies != NULL) 4363 free(ap->a_cookies, M_TEMP); 4364 ap->a_cookies = NULL; 4365 *ap->a_ncookies = 0; 4366 } 4367 return (error); 4368 } 4369 if (ap->a_ncookies == NULL) 4370 return (0); 4371 4372 KASSERT(ap->a_cookies, 4373 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); 4374 4375 *ap->a_cookies = realloc(*ap->a_cookies, 4376 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); 4377 (*ap->a_cookies)[*ap->a_ncookies] = off; 4378 return (0); 4379 } 4380 4381 /* 4382 * Mark for update the access time of the file if the filesystem 4383 * supports VOP_MARKATIME. This functionality is used by execve and 4384 * mmap, so we want to avoid the I/O implied by directly setting 4385 * va_atime for the sake of efficiency. 4386 */ 4387 void 4388 vfs_mark_atime(struct vnode *vp, struct ucred *cred) 4389 { 4390 struct mount *mp; 4391 4392 mp = vp->v_mount; 4393 VFS_ASSERT_GIANT(mp); 4394 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime"); 4395 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 4396 (void)VOP_MARKATIME(vp); 4397 } 4398 4399 /* 4400 * The purpose of this routine is to remove granularity from accmode_t, 4401 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, 4402 * VADMIN and VAPPEND. 4403 * 4404 * If it returns 0, the caller is supposed to continue with the usual 4405 * access checks using 'accmode' as modified by this routine. If it 4406 * returns nonzero value, the caller is supposed to return that value 4407 * as errno. 4408 * 4409 * Note that after this routine runs, accmode may be zero. 4410 */ 4411 int 4412 vfs_unixify_accmode(accmode_t *accmode) 4413 { 4414 /* 4415 * There is no way to specify explicit "deny" rule using 4416 * file mode or POSIX.1e ACLs. 4417 */ 4418 if (*accmode & VEXPLICIT_DENY) { 4419 *accmode = 0; 4420 return (0); 4421 } 4422 4423 /* 4424 * None of these can be translated into usual access bits. 4425 * Also, the common case for NFSv4 ACLs is to not contain 4426 * either of these bits. Caller should check for VWRITE 4427 * on the containing directory instead. 4428 */ 4429 if (*accmode & (VDELETE_CHILD | VDELETE)) 4430 return (EPERM); 4431 4432 if (*accmode & VADMIN_PERMS) { 4433 *accmode &= ~VADMIN_PERMS; 4434 *accmode |= VADMIN; 4435 } 4436 4437 /* 4438 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL 4439 * or VSYNCHRONIZE using file mode or POSIX.1e ACL. 4440 */ 4441 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); 4442 4443 return (0); 4444 } 4445