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 #include "opt_watchdog.h" 46 47 #include <sys/param.h> 48 #include <sys/systm.h> 49 #include <sys/bio.h> 50 #include <sys/buf.h> 51 #include <sys/condvar.h> 52 #include <sys/conf.h> 53 #include <sys/dirent.h> 54 #include <sys/event.h> 55 #include <sys/eventhandler.h> 56 #include <sys/extattr.h> 57 #include <sys/file.h> 58 #include <sys/fcntl.h> 59 #include <sys/jail.h> 60 #include <sys/kdb.h> 61 #include <sys/kernel.h> 62 #include <sys/kthread.h> 63 #include <sys/lockf.h> 64 #include <sys/malloc.h> 65 #include <sys/mount.h> 66 #include <sys/namei.h> 67 #include <sys/priv.h> 68 #include <sys/reboot.h> 69 #include <sys/sched.h> 70 #include <sys/sleepqueue.h> 71 #include <sys/stat.h> 72 #include <sys/sysctl.h> 73 #include <sys/syslog.h> 74 #include <sys/vmmeter.h> 75 #include <sys/vnode.h> 76 #ifdef SW_WATCHDOG 77 #include <sys/watchdog.h> 78 #endif 79 80 #include <machine/stdarg.h> 81 82 #include <security/mac/mac_framework.h> 83 84 #include <vm/vm.h> 85 #include <vm/vm_object.h> 86 #include <vm/vm_extern.h> 87 #include <vm/pmap.h> 88 #include <vm/vm_map.h> 89 #include <vm/vm_page.h> 90 #include <vm/vm_kern.h> 91 #include <vm/uma.h> 92 93 #ifdef DDB 94 #include <ddb/ddb.h> 95 #endif 96 97 #define WI_MPSAFEQ 0 98 #define WI_GIANTQ 1 99 100 static void delmntque(struct vnode *vp); 101 static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, 102 int slpflag, int slptimeo); 103 static void syncer_shutdown(void *arg, int howto); 104 static int vtryrecycle(struct vnode *vp); 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 vnlru_free(int); 110 static void vgonel(struct vnode *); 111 static void vfs_knllock(void *arg); 112 static void vfs_knlunlock(void *arg); 113 static void vfs_knl_assert_locked(void *arg); 114 static void vfs_knl_assert_unlocked(void *arg); 115 static void destroy_vpollinfo(struct vpollinfo *vi); 116 117 /* 118 * Number of vnodes in existence. Increased whenever getnewvnode() 119 * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode. 120 */ 121 static unsigned long numvnodes; 122 123 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, 124 "Number of vnodes in existence"); 125 126 /* 127 * Conversion tables for conversion from vnode types to inode formats 128 * and back. 129 */ 130 enum vtype iftovt_tab[16] = { 131 VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, 132 VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD, 133 }; 134 int vttoif_tab[10] = { 135 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, 136 S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT 137 }; 138 139 /* 140 * List of vnodes that are ready for recycling. 141 */ 142 static TAILQ_HEAD(freelst, vnode) vnode_free_list; 143 144 /* 145 * Free vnode target. Free vnodes may simply be files which have been stat'd 146 * but not read. This is somewhat common, and a small cache of such files 147 * should be kept to avoid recreation costs. 148 */ 149 static u_long wantfreevnodes; 150 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, ""); 151 /* Number of vnodes in the free list. */ 152 static u_long freevnodes; 153 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, 154 "Number of vnodes in the free list"); 155 156 static int vlru_allow_cache_src; 157 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW, 158 &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode"); 159 160 /* 161 * Various variables used for debugging the new implementation of 162 * reassignbuf(). 163 * XXX these are probably of (very) limited utility now. 164 */ 165 static int reassignbufcalls; 166 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, 167 "Number of calls to reassignbuf"); 168 169 /* 170 * Cache for the mount type id assigned to NFS. This is used for 171 * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c. 172 */ 173 int nfs_mount_type = -1; 174 175 /* To keep more than one thread at a time from running vfs_getnewfsid */ 176 static struct mtx mntid_mtx; 177 178 /* 179 * Lock for any access to the following: 180 * vnode_free_list 181 * numvnodes 182 * freevnodes 183 */ 184 static struct mtx vnode_free_list_mtx; 185 186 /* Publicly exported FS */ 187 struct nfs_public nfs_pub; 188 189 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ 190 static uma_zone_t vnode_zone; 191 static uma_zone_t vnodepoll_zone; 192 193 /* 194 * The workitem queue. 195 * 196 * It is useful to delay writes of file data and filesystem metadata 197 * for tens of seconds so that quickly created and deleted files need 198 * not waste disk bandwidth being created and removed. To realize this, 199 * we append vnodes to a "workitem" queue. When running with a soft 200 * updates implementation, most pending metadata dependencies should 201 * not wait for more than a few seconds. Thus, mounted on block devices 202 * are delayed only about a half the time that file data is delayed. 203 * Similarly, directory updates are more critical, so are only delayed 204 * about a third the time that file data is delayed. Thus, there are 205 * SYNCER_MAXDELAY queues that are processed round-robin at a rate of 206 * one each second (driven off the filesystem syncer process). The 207 * syncer_delayno variable indicates the next queue that is to be processed. 208 * Items that need to be processed soon are placed in this queue: 209 * 210 * syncer_workitem_pending[syncer_delayno] 211 * 212 * A delay of fifteen seconds is done by placing the request fifteen 213 * entries later in the queue: 214 * 215 * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] 216 * 217 */ 218 static int syncer_delayno; 219 static long syncer_mask; 220 LIST_HEAD(synclist, bufobj); 221 static struct synclist *syncer_workitem_pending[2]; 222 /* 223 * The sync_mtx protects: 224 * bo->bo_synclist 225 * sync_vnode_count 226 * syncer_delayno 227 * syncer_state 228 * syncer_workitem_pending 229 * syncer_worklist_len 230 * rushjob 231 */ 232 static struct mtx sync_mtx; 233 static struct cv sync_wakeup; 234 235 #define SYNCER_MAXDELAY 32 236 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ 237 static int syncdelay = 30; /* max time to delay syncing data */ 238 static int filedelay = 30; /* time to delay syncing files */ 239 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, 240 "Time to delay syncing files (in seconds)"); 241 static int dirdelay = 29; /* time to delay syncing directories */ 242 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, 243 "Time to delay syncing directories (in seconds)"); 244 static int metadelay = 28; /* time to delay syncing metadata */ 245 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, 246 "Time to delay syncing metadata (in seconds)"); 247 static int rushjob; /* number of slots to run ASAP */ 248 static int stat_rush_requests; /* number of times I/O speeded up */ 249 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, 250 "Number of times I/O speeded up (rush requests)"); 251 252 /* 253 * When shutting down the syncer, run it at four times normal speed. 254 */ 255 #define SYNCER_SHUTDOWN_SPEEDUP 4 256 static int sync_vnode_count; 257 static int syncer_worklist_len; 258 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } 259 syncer_state; 260 261 /* 262 * Number of vnodes we want to exist at any one time. This is mostly used 263 * to size hash tables in vnode-related code. It is normally not used in 264 * getnewvnode(), as wantfreevnodes is normally nonzero.) 265 * 266 * XXX desiredvnodes is historical cruft and should not exist. 267 */ 268 int desiredvnodes; 269 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW, 270 &desiredvnodes, 0, "Maximum number of vnodes"); 271 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, 272 &wantfreevnodes, 0, "Minimum number of vnodes (legacy)"); 273 static int vnlru_nowhere; 274 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, 275 &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); 276 277 /* 278 * Macros to control when a vnode is freed and recycled. All require 279 * the vnode interlock. 280 */ 281 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 282 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt) 283 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt) 284 285 286 /* 287 * Initialize the vnode management data structures. 288 * 289 * Reevaluate the following cap on the number of vnodes after the physical 290 * memory size exceeds 512GB. In the limit, as the physical memory size 291 * grows, the ratio of physical pages to vnodes approaches sixteen to one. 292 */ 293 #ifndef MAXVNODES_MAX 294 #define MAXVNODES_MAX (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16)) 295 #endif 296 static void 297 vntblinit(void *dummy __unused) 298 { 299 int physvnodes, virtvnodes; 300 301 /* 302 * Desiredvnodes is a function of the physical memory size and the 303 * kernel's heap size. Generally speaking, it scales with the 304 * physical memory size. The ratio of desiredvnodes to physical pages 305 * is one to four until desiredvnodes exceeds 98,304. Thereafter, the 306 * marginal ratio of desiredvnodes to physical pages is one to 307 * sixteen. However, desiredvnodes is limited by the kernel's heap 308 * size. The memory required by desiredvnodes vnodes and vm objects 309 * may not exceed one seventh of the kernel's heap size. 310 */ 311 physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4, 312 cnt.v_page_count) / 16; 313 virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) + 314 sizeof(struct vnode))); 315 desiredvnodes = min(physvnodes, virtvnodes); 316 if (desiredvnodes > MAXVNODES_MAX) { 317 if (bootverbose) 318 printf("Reducing kern.maxvnodes %d -> %d\n", 319 desiredvnodes, MAXVNODES_MAX); 320 desiredvnodes = MAXVNODES_MAX; 321 } 322 wantfreevnodes = desiredvnodes / 4; 323 mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); 324 TAILQ_INIT(&vnode_free_list); 325 mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF); 326 vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, 327 NULL, NULL, UMA_ALIGN_PTR, 0); 328 vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), 329 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 330 /* 331 * Initialize the filesystem syncer. 332 */ 333 syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE, 334 &syncer_mask); 335 syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE, 336 &syncer_mask); 337 syncer_maxdelay = syncer_mask + 1; 338 mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); 339 cv_init(&sync_wakeup, "syncer"); 340 } 341 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); 342 343 344 /* 345 * Mark a mount point as busy. Used to synchronize access and to delay 346 * unmounting. Eventually, mountlist_mtx is not released on failure. 347 * 348 * vfs_busy() is a custom lock, it can block the caller. 349 * vfs_busy() only sleeps if the unmount is active on the mount point. 350 * For a mountpoint mp, vfs_busy-enforced lock is before lock of any 351 * vnode belonging to mp. 352 * 353 * Lookup uses vfs_busy() to traverse mount points. 354 * root fs var fs 355 * / vnode lock A / vnode lock (/var) D 356 * /var vnode lock B /log vnode lock(/var/log) E 357 * vfs_busy lock C vfs_busy lock F 358 * 359 * Within each file system, the lock order is C->A->B and F->D->E. 360 * 361 * When traversing across mounts, the system follows that lock order: 362 * 363 * C->A->B 364 * | 365 * +->F->D->E 366 * 367 * The lookup() process for namei("/var") illustrates the process: 368 * VOP_LOOKUP() obtains B while A is held 369 * vfs_busy() obtains a shared lock on F while A and B are held 370 * vput() releases lock on B 371 * vput() releases lock on A 372 * VFS_ROOT() obtains lock on D while shared lock on F is held 373 * vfs_unbusy() releases shared lock on F 374 * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A. 375 * Attempt to lock A (instead of vp_crossmp) while D is held would 376 * violate the global order, causing deadlocks. 377 * 378 * dounmount() locks B while F is drained. 379 */ 380 int 381 vfs_busy(struct mount *mp, int flags) 382 { 383 384 MPASS((flags & ~MBF_MASK) == 0); 385 CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); 386 387 MNT_ILOCK(mp); 388 MNT_REF(mp); 389 /* 390 * If mount point is currenly being unmounted, sleep until the 391 * mount point fate is decided. If thread doing the unmounting fails, 392 * it will clear MNTK_UNMOUNT flag before waking us up, indicating 393 * that this mount point has survived the unmount attempt and vfs_busy 394 * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE 395 * flag in addition to MNTK_UNMOUNT, indicating that mount point is 396 * about to be really destroyed. vfs_busy needs to release its 397 * reference on the mount point in this case and return with ENOENT, 398 * telling the caller that mount mount it tried to busy is no longer 399 * valid. 400 */ 401 while (mp->mnt_kern_flag & MNTK_UNMOUNT) { 402 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { 403 MNT_REL(mp); 404 MNT_IUNLOCK(mp); 405 CTR1(KTR_VFS, "%s: failed busying before sleeping", 406 __func__); 407 return (ENOENT); 408 } 409 if (flags & MBF_MNTLSTLOCK) 410 mtx_unlock(&mountlist_mtx); 411 mp->mnt_kern_flag |= MNTK_MWAIT; 412 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0); 413 if (flags & MBF_MNTLSTLOCK) 414 mtx_lock(&mountlist_mtx); 415 MNT_ILOCK(mp); 416 } 417 if (flags & MBF_MNTLSTLOCK) 418 mtx_unlock(&mountlist_mtx); 419 mp->mnt_lockref++; 420 MNT_IUNLOCK(mp); 421 return (0); 422 } 423 424 /* 425 * Free a busy filesystem. 426 */ 427 void 428 vfs_unbusy(struct mount *mp) 429 { 430 431 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 432 MNT_ILOCK(mp); 433 MNT_REL(mp); 434 KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref")); 435 mp->mnt_lockref--; 436 if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { 437 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); 438 CTR1(KTR_VFS, "%s: waking up waiters", __func__); 439 mp->mnt_kern_flag &= ~MNTK_DRAINING; 440 wakeup(&mp->mnt_lockref); 441 } 442 MNT_IUNLOCK(mp); 443 } 444 445 /* 446 * Lookup a mount point by filesystem identifier. 447 */ 448 struct mount * 449 vfs_getvfs(fsid_t *fsid) 450 { 451 struct mount *mp; 452 453 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 454 mtx_lock(&mountlist_mtx); 455 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 456 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 457 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 458 vfs_ref(mp); 459 mtx_unlock(&mountlist_mtx); 460 return (mp); 461 } 462 } 463 mtx_unlock(&mountlist_mtx); 464 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 465 return ((struct mount *) 0); 466 } 467 468 /* 469 * Lookup a mount point by filesystem identifier, busying it before 470 * returning. 471 */ 472 struct mount * 473 vfs_busyfs(fsid_t *fsid) 474 { 475 struct mount *mp; 476 int error; 477 478 CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); 479 mtx_lock(&mountlist_mtx); 480 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 481 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && 482 mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { 483 error = vfs_busy(mp, MBF_MNTLSTLOCK); 484 if (error) { 485 mtx_unlock(&mountlist_mtx); 486 return (NULL); 487 } 488 return (mp); 489 } 490 } 491 CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); 492 mtx_unlock(&mountlist_mtx); 493 return ((struct mount *) 0); 494 } 495 496 /* 497 * Check if a user can access privileged mount options. 498 */ 499 int 500 vfs_suser(struct mount *mp, struct thread *td) 501 { 502 int error; 503 504 /* 505 * If the thread is jailed, but this is not a jail-friendly file 506 * system, deny immediately. 507 */ 508 if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred)) 509 return (EPERM); 510 511 /* 512 * If the file system was mounted outside the jail of the calling 513 * thread, deny immediately. 514 */ 515 if (prison_check(td->td_ucred, mp->mnt_cred) != 0) 516 return (EPERM); 517 518 /* 519 * If file system supports delegated administration, we don't check 520 * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified 521 * by the file system itself. 522 * If this is not the user that did original mount, we check for 523 * the PRIV_VFS_MOUNT_OWNER privilege. 524 */ 525 if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && 526 mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { 527 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) 528 return (error); 529 } 530 return (0); 531 } 532 533 /* 534 * Get a new unique fsid. Try to make its val[0] unique, since this value 535 * will be used to create fake device numbers for stat(). Also try (but 536 * not so hard) make its val[0] unique mod 2^16, since some emulators only 537 * support 16-bit device numbers. We end up with unique val[0]'s for the 538 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. 539 * 540 * Keep in mind that several mounts may be running in parallel. Starting 541 * the search one past where the previous search terminated is both a 542 * micro-optimization and a defense against returning the same fsid to 543 * different mounts. 544 */ 545 void 546 vfs_getnewfsid(struct mount *mp) 547 { 548 static uint16_t mntid_base; 549 struct mount *nmp; 550 fsid_t tfsid; 551 int mtype; 552 553 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 554 mtx_lock(&mntid_mtx); 555 mtype = mp->mnt_vfc->vfc_typenum; 556 tfsid.val[1] = mtype; 557 mtype = (mtype & 0xFF) << 24; 558 for (;;) { 559 tfsid.val[0] = makedev(255, 560 mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); 561 mntid_base++; 562 if ((nmp = vfs_getvfs(&tfsid)) == NULL) 563 break; 564 vfs_rel(nmp); 565 } 566 mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; 567 mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; 568 mtx_unlock(&mntid_mtx); 569 } 570 571 /* 572 * Knob to control the precision of file timestamps: 573 * 574 * 0 = seconds only; nanoseconds zeroed. 575 * 1 = seconds and nanoseconds, accurate within 1/HZ. 576 * 2 = seconds and nanoseconds, truncated to microseconds. 577 * >=3 = seconds and nanoseconds, maximum precision. 578 */ 579 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; 580 581 static int timestamp_precision = TSP_SEC; 582 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, 583 ×tamp_precision, 0, "File timestamp precision (0: seconds, " 584 "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, " 585 "3+: sec + ns (max. precision))"); 586 587 /* 588 * Get a current timestamp. 589 */ 590 void 591 vfs_timestamp(struct timespec *tsp) 592 { 593 struct timeval tv; 594 595 switch (timestamp_precision) { 596 case TSP_SEC: 597 tsp->tv_sec = time_second; 598 tsp->tv_nsec = 0; 599 break; 600 case TSP_HZ: 601 getnanotime(tsp); 602 break; 603 case TSP_USEC: 604 microtime(&tv); 605 TIMEVAL_TO_TIMESPEC(&tv, tsp); 606 break; 607 case TSP_NSEC: 608 default: 609 nanotime(tsp); 610 break; 611 } 612 } 613 614 /* 615 * Set vnode attributes to VNOVAL 616 */ 617 void 618 vattr_null(struct vattr *vap) 619 { 620 621 vap->va_type = VNON; 622 vap->va_size = VNOVAL; 623 vap->va_bytes = VNOVAL; 624 vap->va_mode = VNOVAL; 625 vap->va_nlink = VNOVAL; 626 vap->va_uid = VNOVAL; 627 vap->va_gid = VNOVAL; 628 vap->va_fsid = VNOVAL; 629 vap->va_fileid = VNOVAL; 630 vap->va_blocksize = VNOVAL; 631 vap->va_rdev = VNOVAL; 632 vap->va_atime.tv_sec = VNOVAL; 633 vap->va_atime.tv_nsec = VNOVAL; 634 vap->va_mtime.tv_sec = VNOVAL; 635 vap->va_mtime.tv_nsec = VNOVAL; 636 vap->va_ctime.tv_sec = VNOVAL; 637 vap->va_ctime.tv_nsec = VNOVAL; 638 vap->va_birthtime.tv_sec = VNOVAL; 639 vap->va_birthtime.tv_nsec = VNOVAL; 640 vap->va_flags = VNOVAL; 641 vap->va_gen = VNOVAL; 642 vap->va_vaflags = 0; 643 } 644 645 /* 646 * This routine is called when we have too many vnodes. It attempts 647 * to free <count> vnodes and will potentially free vnodes that still 648 * have VM backing store (VM backing store is typically the cause 649 * of a vnode blowout so we want to do this). Therefore, this operation 650 * is not considered cheap. 651 * 652 * A number of conditions may prevent a vnode from being reclaimed. 653 * the buffer cache may have references on the vnode, a directory 654 * vnode may still have references due to the namei cache representing 655 * underlying files, or the vnode may be in active use. It is not 656 * desireable to reuse such vnodes. These conditions may cause the 657 * number of vnodes to reach some minimum value regardless of what 658 * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. 659 */ 660 static int 661 vlrureclaim(struct mount *mp) 662 { 663 struct vnode *vp; 664 int done; 665 int trigger; 666 int usevnodes; 667 int count; 668 669 /* 670 * Calculate the trigger point, don't allow user 671 * screwups to blow us up. This prevents us from 672 * recycling vnodes with lots of resident pages. We 673 * aren't trying to free memory, we are trying to 674 * free vnodes. 675 */ 676 usevnodes = desiredvnodes; 677 if (usevnodes <= 0) 678 usevnodes = 1; 679 trigger = cnt.v_page_count * 2 / usevnodes; 680 done = 0; 681 vn_start_write(NULL, &mp, V_WAIT); 682 MNT_ILOCK(mp); 683 count = mp->mnt_nvnodelistsize / 10 + 1; 684 while (count != 0) { 685 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 686 while (vp != NULL && vp->v_type == VMARKER) 687 vp = TAILQ_NEXT(vp, v_nmntvnodes); 688 if (vp == NULL) 689 break; 690 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 691 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 692 --count; 693 if (!VI_TRYLOCK(vp)) 694 goto next_iter; 695 /* 696 * If it's been deconstructed already, it's still 697 * referenced, or it exceeds the trigger, skip it. 698 */ 699 if (vp->v_usecount || 700 (!vlru_allow_cache_src && 701 !LIST_EMPTY(&(vp)->v_cache_src)) || 702 (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL && 703 vp->v_object->resident_page_count > trigger)) { 704 VI_UNLOCK(vp); 705 goto next_iter; 706 } 707 MNT_IUNLOCK(mp); 708 vholdl(vp); 709 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) { 710 vdrop(vp); 711 goto next_iter_mntunlocked; 712 } 713 VI_LOCK(vp); 714 /* 715 * v_usecount may have been bumped after VOP_LOCK() dropped 716 * the vnode interlock and before it was locked again. 717 * 718 * It is not necessary to recheck VI_DOOMED because it can 719 * only be set by another thread that holds both the vnode 720 * lock and vnode interlock. If another thread has the 721 * vnode lock before we get to VOP_LOCK() and obtains the 722 * vnode interlock after VOP_LOCK() drops the vnode 723 * interlock, the other thread will be unable to drop the 724 * vnode lock before our VOP_LOCK() call fails. 725 */ 726 if (vp->v_usecount || 727 (!vlru_allow_cache_src && 728 !LIST_EMPTY(&(vp)->v_cache_src)) || 729 (vp->v_object != NULL && 730 vp->v_object->resident_page_count > trigger)) { 731 VOP_UNLOCK(vp, LK_INTERLOCK); 732 goto next_iter_mntunlocked; 733 } 734 KASSERT((vp->v_iflag & VI_DOOMED) == 0, 735 ("VI_DOOMED unexpectedly detected in vlrureclaim()")); 736 vgonel(vp); 737 VOP_UNLOCK(vp, 0); 738 vdropl(vp); 739 done++; 740 next_iter_mntunlocked: 741 if (!should_yield()) 742 goto relock_mnt; 743 goto yield; 744 next_iter: 745 if (!should_yield()) 746 continue; 747 MNT_IUNLOCK(mp); 748 yield: 749 kern_yield(PRI_UNCHANGED); 750 relock_mnt: 751 MNT_ILOCK(mp); 752 } 753 MNT_IUNLOCK(mp); 754 vn_finished_write(mp); 755 return done; 756 } 757 758 /* 759 * Attempt to keep the free list at wantfreevnodes length. 760 */ 761 static void 762 vnlru_free(int count) 763 { 764 struct vnode *vp; 765 int vfslocked; 766 767 mtx_assert(&vnode_free_list_mtx, MA_OWNED); 768 for (; count > 0; count--) { 769 vp = TAILQ_FIRST(&vnode_free_list); 770 /* 771 * The list can be modified while the free_list_mtx 772 * has been dropped and vp could be NULL here. 773 */ 774 if (!vp) 775 break; 776 VNASSERT(vp->v_op != NULL, vp, 777 ("vnlru_free: vnode already reclaimed.")); 778 KASSERT((vp->v_iflag & VI_FREE) != 0, 779 ("Removing vnode not on freelist")); 780 KASSERT((vp->v_iflag & VI_ACTIVE) == 0, 781 ("Mangling active vnode")); 782 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist); 783 /* 784 * Don't recycle if we can't get the interlock. 785 */ 786 if (!VI_TRYLOCK(vp)) { 787 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist); 788 continue; 789 } 790 VNASSERT(VCANRECYCLE(vp), vp, 791 ("vp inconsistent on freelist")); 792 freevnodes--; 793 vp->v_iflag &= ~VI_FREE; 794 vholdl(vp); 795 mtx_unlock(&vnode_free_list_mtx); 796 VI_UNLOCK(vp); 797 vfslocked = VFS_LOCK_GIANT(vp->v_mount); 798 vtryrecycle(vp); 799 VFS_UNLOCK_GIANT(vfslocked); 800 /* 801 * If the recycled succeeded this vdrop will actually free 802 * the vnode. If not it will simply place it back on 803 * the free list. 804 */ 805 vdrop(vp); 806 mtx_lock(&vnode_free_list_mtx); 807 } 808 } 809 /* 810 * Attempt to recycle vnodes in a context that is always safe to block. 811 * Calling vlrurecycle() from the bowels of filesystem code has some 812 * interesting deadlock problems. 813 */ 814 static struct proc *vnlruproc; 815 static int vnlruproc_sig; 816 817 static void 818 vnlru_proc(void) 819 { 820 struct mount *mp, *nmp; 821 int done, vfslocked; 822 struct proc *p = vnlruproc; 823 824 EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p, 825 SHUTDOWN_PRI_FIRST); 826 827 for (;;) { 828 kproc_suspend_check(p); 829 mtx_lock(&vnode_free_list_mtx); 830 if (freevnodes > wantfreevnodes) 831 vnlru_free(freevnodes - wantfreevnodes); 832 if (numvnodes <= desiredvnodes * 9 / 10) { 833 vnlruproc_sig = 0; 834 wakeup(&vnlruproc_sig); 835 msleep(vnlruproc, &vnode_free_list_mtx, 836 PVFS|PDROP, "vlruwt", hz); 837 continue; 838 } 839 mtx_unlock(&vnode_free_list_mtx); 840 done = 0; 841 mtx_lock(&mountlist_mtx); 842 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 843 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) { 844 nmp = TAILQ_NEXT(mp, mnt_list); 845 continue; 846 } 847 vfslocked = VFS_LOCK_GIANT(mp); 848 done += vlrureclaim(mp); 849 VFS_UNLOCK_GIANT(vfslocked); 850 mtx_lock(&mountlist_mtx); 851 nmp = TAILQ_NEXT(mp, mnt_list); 852 vfs_unbusy(mp); 853 } 854 mtx_unlock(&mountlist_mtx); 855 if (done == 0) { 856 #if 0 857 /* These messages are temporary debugging aids */ 858 if (vnlru_nowhere < 5) 859 printf("vnlru process getting nowhere..\n"); 860 else if (vnlru_nowhere == 5) 861 printf("vnlru process messages stopped.\n"); 862 #endif 863 vnlru_nowhere++; 864 tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); 865 } else 866 kern_yield(PRI_UNCHANGED); 867 } 868 } 869 870 static struct kproc_desc vnlru_kp = { 871 "vnlru", 872 vnlru_proc, 873 &vnlruproc 874 }; 875 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, 876 &vnlru_kp); 877 878 /* 879 * Routines having to do with the management of the vnode table. 880 */ 881 882 /* 883 * Try to recycle a freed vnode. We abort if anyone picks up a reference 884 * before we actually vgone(). This function must be called with the vnode 885 * held to prevent the vnode from being returned to the free list midway 886 * through vgone(). 887 */ 888 static int 889 vtryrecycle(struct vnode *vp) 890 { 891 struct mount *vnmp; 892 893 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 894 VNASSERT(vp->v_holdcnt, vp, 895 ("vtryrecycle: Recycling vp %p without a reference.", vp)); 896 /* 897 * This vnode may found and locked via some other list, if so we 898 * can't recycle it yet. 899 */ 900 if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { 901 CTR2(KTR_VFS, 902 "%s: impossible to recycle, vp %p lock is already held", 903 __func__, vp); 904 return (EWOULDBLOCK); 905 } 906 /* 907 * Don't recycle if its filesystem is being suspended. 908 */ 909 if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { 910 VOP_UNLOCK(vp, 0); 911 CTR2(KTR_VFS, 912 "%s: impossible to recycle, cannot start the write for %p", 913 __func__, vp); 914 return (EBUSY); 915 } 916 /* 917 * If we got this far, we need to acquire the interlock and see if 918 * anyone picked up this vnode from another list. If not, we will 919 * mark it with DOOMED via vgonel() so that anyone who does find it 920 * will skip over it. 921 */ 922 VI_LOCK(vp); 923 if (vp->v_usecount) { 924 VOP_UNLOCK(vp, LK_INTERLOCK); 925 vn_finished_write(vnmp); 926 CTR2(KTR_VFS, 927 "%s: impossible to recycle, %p is already referenced", 928 __func__, vp); 929 return (EBUSY); 930 } 931 if ((vp->v_iflag & VI_DOOMED) == 0) 932 vgonel(vp); 933 VOP_UNLOCK(vp, LK_INTERLOCK); 934 vn_finished_write(vnmp); 935 return (0); 936 } 937 938 /* 939 * Return the next vnode from the free list. 940 */ 941 int 942 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, 943 struct vnode **vpp) 944 { 945 struct vnode *vp = NULL; 946 struct bufobj *bo; 947 948 CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); 949 mtx_lock(&vnode_free_list_mtx); 950 /* 951 * Lend our context to reclaim vnodes if they've exceeded the max. 952 */ 953 if (freevnodes > wantfreevnodes) 954 vnlru_free(1); 955 /* 956 * Wait for available vnodes. 957 */ 958 if (numvnodes > desiredvnodes) { 959 if (mp != NULL && (mp->mnt_kern_flag & MNTK_SUSPEND)) { 960 /* 961 * File system is beeing suspended, we cannot risk a 962 * deadlock here, so allocate new vnode anyway. 963 */ 964 if (freevnodes > wantfreevnodes) 965 vnlru_free(freevnodes - wantfreevnodes); 966 goto alloc; 967 } 968 if (vnlruproc_sig == 0) { 969 vnlruproc_sig = 1; /* avoid unnecessary wakeups */ 970 wakeup(vnlruproc); 971 } 972 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS, 973 "vlruwk", hz); 974 #if 0 /* XXX Not all VFS_VGET/ffs_vget callers check returns. */ 975 if (numvnodes > desiredvnodes) { 976 mtx_unlock(&vnode_free_list_mtx); 977 return (ENFILE); 978 } 979 #endif 980 } 981 alloc: 982 numvnodes++; 983 mtx_unlock(&vnode_free_list_mtx); 984 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO); 985 /* 986 * Setup locks. 987 */ 988 vp->v_vnlock = &vp->v_lock; 989 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); 990 /* 991 * By default, don't allow shared locks unless filesystems 992 * opt-in. 993 */ 994 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE); 995 /* 996 * Initialize bufobj. 997 */ 998 bo = &vp->v_bufobj; 999 bo->__bo_vnode = vp; 1000 mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF); 1001 bo->bo_ops = &buf_ops_bio; 1002 bo->bo_private = vp; 1003 TAILQ_INIT(&bo->bo_clean.bv_hd); 1004 TAILQ_INIT(&bo->bo_dirty.bv_hd); 1005 /* 1006 * Initialize namecache. 1007 */ 1008 LIST_INIT(&vp->v_cache_src); 1009 TAILQ_INIT(&vp->v_cache_dst); 1010 /* 1011 * Finalize various vnode identity bits. 1012 */ 1013 vp->v_type = VNON; 1014 vp->v_tag = tag; 1015 vp->v_op = vops; 1016 v_incr_usecount(vp); 1017 vp->v_data = NULL; 1018 #ifdef MAC 1019 mac_vnode_init(vp); 1020 if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) 1021 mac_vnode_associate_singlelabel(mp, vp); 1022 else if (mp == NULL && vops != &dead_vnodeops) 1023 printf("NULL mp in getnewvnode()\n"); 1024 #endif 1025 if (mp != NULL) { 1026 bo->bo_bsize = mp->mnt_stat.f_iosize; 1027 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) 1028 vp->v_vflag |= VV_NOKNOTE; 1029 } 1030 1031 *vpp = vp; 1032 return (0); 1033 } 1034 1035 /* 1036 * Delete from old mount point vnode list, if on one. 1037 */ 1038 static void 1039 delmntque(struct vnode *vp) 1040 { 1041 struct mount *mp; 1042 int active; 1043 1044 mp = vp->v_mount; 1045 if (mp == NULL) 1046 return; 1047 MNT_ILOCK(mp); 1048 VI_LOCK(vp); 1049 KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize, 1050 ("Active vnode list size %d > Vnode list size %d", 1051 mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize)); 1052 active = vp->v_iflag & VI_ACTIVE; 1053 vp->v_iflag &= ~VI_ACTIVE; 1054 if (active) { 1055 mtx_lock(&vnode_free_list_mtx); 1056 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist); 1057 mp->mnt_activevnodelistsize--; 1058 mtx_unlock(&vnode_free_list_mtx); 1059 } 1060 vp->v_mount = NULL; 1061 VI_UNLOCK(vp); 1062 VNASSERT(mp->mnt_nvnodelistsize > 0, vp, 1063 ("bad mount point vnode list size")); 1064 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1065 mp->mnt_nvnodelistsize--; 1066 MNT_REL(mp); 1067 MNT_IUNLOCK(mp); 1068 } 1069 1070 static void 1071 insmntque_stddtr(struct vnode *vp, void *dtr_arg) 1072 { 1073 1074 vp->v_data = NULL; 1075 vp->v_op = &dead_vnodeops; 1076 /* XXX non mp-safe fs may still call insmntque with vnode 1077 unlocked */ 1078 if (!VOP_ISLOCKED(vp)) 1079 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1080 vgone(vp); 1081 vput(vp); 1082 } 1083 1084 /* 1085 * Insert into list of vnodes for the new mount point, if available. 1086 */ 1087 int 1088 insmntque1(struct vnode *vp, struct mount *mp, 1089 void (*dtr)(struct vnode *, void *), void *dtr_arg) 1090 { 1091 int locked; 1092 1093 KASSERT(vp->v_mount == NULL, 1094 ("insmntque: vnode already on per mount vnode list")); 1095 VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); 1096 #ifdef DEBUG_VFS_LOCKS 1097 if (!VFS_NEEDSGIANT(mp)) 1098 ASSERT_VOP_ELOCKED(vp, 1099 "insmntque: mp-safe fs and non-locked vp"); 1100 #endif 1101 /* 1102 * We acquire the vnode interlock early to ensure that the 1103 * vnode cannot be recycled by another process releasing a 1104 * holdcnt on it before we get it on both the vnode list 1105 * and the active vnode list. The mount mutex protects only 1106 * manipulation of the vnode list and the vnode freelist 1107 * mutex protects only manipulation of the active vnode list. 1108 * Hence the need to hold the vnode interlock throughout. 1109 */ 1110 MNT_ILOCK(mp); 1111 VI_LOCK(vp); 1112 if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 && 1113 ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || 1114 mp->mnt_nvnodelistsize == 0)) { 1115 locked = VOP_ISLOCKED(vp); 1116 if (!locked || (locked == LK_EXCLUSIVE && 1117 (vp->v_vflag & VV_FORCEINSMQ) == 0)) { 1118 VI_UNLOCK(vp); 1119 MNT_IUNLOCK(mp); 1120 if (dtr != NULL) 1121 dtr(vp, dtr_arg); 1122 return (EBUSY); 1123 } 1124 } 1125 vp->v_mount = mp; 1126 MNT_REF(mp); 1127 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); 1128 VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, 1129 ("neg mount point vnode list size")); 1130 mp->mnt_nvnodelistsize++; 1131 KASSERT((vp->v_iflag & VI_ACTIVE) == 0, 1132 ("Activating already active vnode")); 1133 vp->v_iflag |= VI_ACTIVE; 1134 mtx_lock(&vnode_free_list_mtx); 1135 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist); 1136 mp->mnt_activevnodelistsize++; 1137 mtx_unlock(&vnode_free_list_mtx); 1138 VI_UNLOCK(vp); 1139 MNT_IUNLOCK(mp); 1140 return (0); 1141 } 1142 1143 int 1144 insmntque(struct vnode *vp, struct mount *mp) 1145 { 1146 1147 return (insmntque1(vp, mp, insmntque_stddtr, NULL)); 1148 } 1149 1150 /* 1151 * Flush out and invalidate all buffers associated with a bufobj 1152 * Called with the underlying object locked. 1153 */ 1154 int 1155 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) 1156 { 1157 int error; 1158 1159 BO_LOCK(bo); 1160 if (flags & V_SAVE) { 1161 error = bufobj_wwait(bo, slpflag, slptimeo); 1162 if (error) { 1163 BO_UNLOCK(bo); 1164 return (error); 1165 } 1166 if (bo->bo_dirty.bv_cnt > 0) { 1167 BO_UNLOCK(bo); 1168 if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) 1169 return (error); 1170 /* 1171 * XXX We could save a lock/unlock if this was only 1172 * enabled under INVARIANTS 1173 */ 1174 BO_LOCK(bo); 1175 if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) 1176 panic("vinvalbuf: dirty bufs"); 1177 } 1178 } 1179 /* 1180 * If you alter this loop please notice that interlock is dropped and 1181 * reacquired in flushbuflist. Special care is needed to ensure that 1182 * no race conditions occur from this. 1183 */ 1184 do { 1185 error = flushbuflist(&bo->bo_clean, 1186 flags, bo, slpflag, slptimeo); 1187 if (error == 0 && !(flags & V_CLEANONLY)) 1188 error = flushbuflist(&bo->bo_dirty, 1189 flags, bo, slpflag, slptimeo); 1190 if (error != 0 && error != EAGAIN) { 1191 BO_UNLOCK(bo); 1192 return (error); 1193 } 1194 } while (error != 0); 1195 1196 /* 1197 * Wait for I/O to complete. XXX needs cleaning up. The vnode can 1198 * have write I/O in-progress but if there is a VM object then the 1199 * VM object can also have read-I/O in-progress. 1200 */ 1201 do { 1202 bufobj_wwait(bo, 0, 0); 1203 BO_UNLOCK(bo); 1204 if (bo->bo_object != NULL) { 1205 VM_OBJECT_LOCK(bo->bo_object); 1206 vm_object_pip_wait(bo->bo_object, "bovlbx"); 1207 VM_OBJECT_UNLOCK(bo->bo_object); 1208 } 1209 BO_LOCK(bo); 1210 } while (bo->bo_numoutput > 0); 1211 BO_UNLOCK(bo); 1212 1213 /* 1214 * Destroy the copy in the VM cache, too. 1215 */ 1216 if (bo->bo_object != NULL && 1217 (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) { 1218 VM_OBJECT_LOCK(bo->bo_object); 1219 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ? 1220 OBJPR_CLEANONLY : 0); 1221 VM_OBJECT_UNLOCK(bo->bo_object); 1222 } 1223 1224 #ifdef INVARIANTS 1225 BO_LOCK(bo); 1226 if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 && 1227 (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) 1228 panic("vinvalbuf: flush failed"); 1229 BO_UNLOCK(bo); 1230 #endif 1231 return (0); 1232 } 1233 1234 /* 1235 * Flush out and invalidate all buffers associated with a vnode. 1236 * Called with the underlying object locked. 1237 */ 1238 int 1239 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) 1240 { 1241 1242 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 1243 ASSERT_VOP_LOCKED(vp, "vinvalbuf"); 1244 return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); 1245 } 1246 1247 /* 1248 * Flush out buffers on the specified list. 1249 * 1250 */ 1251 static int 1252 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, 1253 int slptimeo) 1254 { 1255 struct buf *bp, *nbp; 1256 int retval, error; 1257 daddr_t lblkno; 1258 b_xflags_t xflags; 1259 1260 ASSERT_BO_LOCKED(bo); 1261 1262 retval = 0; 1263 TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { 1264 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) || 1265 ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) { 1266 continue; 1267 } 1268 lblkno = 0; 1269 xflags = 0; 1270 if (nbp != NULL) { 1271 lblkno = nbp->b_lblkno; 1272 xflags = nbp->b_xflags & 1273 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN); 1274 } 1275 retval = EAGAIN; 1276 error = BUF_TIMELOCK(bp, 1277 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo), 1278 "flushbuf", slpflag, slptimeo); 1279 if (error) { 1280 BO_LOCK(bo); 1281 return (error != ENOLCK ? error : EAGAIN); 1282 } 1283 KASSERT(bp->b_bufobj == bo, 1284 ("bp %p wrong b_bufobj %p should be %p", 1285 bp, bp->b_bufobj, bo)); 1286 if (bp->b_bufobj != bo) { /* XXX: necessary ? */ 1287 BUF_UNLOCK(bp); 1288 BO_LOCK(bo); 1289 return (EAGAIN); 1290 } 1291 /* 1292 * XXX Since there are no node locks for NFS, I 1293 * believe there is a slight chance that a delayed 1294 * write will occur while sleeping just above, so 1295 * check for it. 1296 */ 1297 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && 1298 (flags & V_SAVE)) { 1299 BO_LOCK(bo); 1300 bremfree(bp); 1301 BO_UNLOCK(bo); 1302 bp->b_flags |= B_ASYNC; 1303 bwrite(bp); 1304 BO_LOCK(bo); 1305 return (EAGAIN); /* XXX: why not loop ? */ 1306 } 1307 BO_LOCK(bo); 1308 bremfree(bp); 1309 BO_UNLOCK(bo); 1310 bp->b_flags |= (B_INVAL | B_RELBUF); 1311 bp->b_flags &= ~B_ASYNC; 1312 brelse(bp); 1313 BO_LOCK(bo); 1314 if (nbp != NULL && 1315 (nbp->b_bufobj != bo || 1316 nbp->b_lblkno != lblkno || 1317 (nbp->b_xflags & 1318 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags)) 1319 break; /* nbp invalid */ 1320 } 1321 return (retval); 1322 } 1323 1324 /* 1325 * Truncate a file's buffer and pages to a specified length. This 1326 * is in lieu of the old vinvalbuf mechanism, which performed unneeded 1327 * sync activity. 1328 */ 1329 int 1330 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize) 1331 { 1332 struct buf *bp, *nbp; 1333 int anyfreed; 1334 int trunclbn; 1335 struct bufobj *bo; 1336 1337 CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__, 1338 vp, cred, blksize, (uintmax_t)length); 1339 1340 /* 1341 * Round up to the *next* lbn. 1342 */ 1343 trunclbn = (length + blksize - 1) / blksize; 1344 1345 ASSERT_VOP_LOCKED(vp, "vtruncbuf"); 1346 restart: 1347 bo = &vp->v_bufobj; 1348 BO_LOCK(bo); 1349 anyfreed = 1; 1350 for (;anyfreed;) { 1351 anyfreed = 0; 1352 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { 1353 if (bp->b_lblkno < trunclbn) 1354 continue; 1355 if (BUF_LOCK(bp, 1356 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1357 BO_MTX(bo)) == ENOLCK) 1358 goto restart; 1359 1360 BO_LOCK(bo); 1361 bremfree(bp); 1362 BO_UNLOCK(bo); 1363 bp->b_flags |= (B_INVAL | B_RELBUF); 1364 bp->b_flags &= ~B_ASYNC; 1365 brelse(bp); 1366 anyfreed = 1; 1367 1368 BO_LOCK(bo); 1369 if (nbp != NULL && 1370 (((nbp->b_xflags & BX_VNCLEAN) == 0) || 1371 (nbp->b_vp != vp) || 1372 (nbp->b_flags & B_DELWRI))) { 1373 BO_UNLOCK(bo); 1374 goto restart; 1375 } 1376 } 1377 1378 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1379 if (bp->b_lblkno < trunclbn) 1380 continue; 1381 if (BUF_LOCK(bp, 1382 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1383 BO_MTX(bo)) == ENOLCK) 1384 goto restart; 1385 BO_LOCK(bo); 1386 bremfree(bp); 1387 BO_UNLOCK(bo); 1388 bp->b_flags |= (B_INVAL | B_RELBUF); 1389 bp->b_flags &= ~B_ASYNC; 1390 brelse(bp); 1391 anyfreed = 1; 1392 1393 BO_LOCK(bo); 1394 if (nbp != NULL && 1395 (((nbp->b_xflags & BX_VNDIRTY) == 0) || 1396 (nbp->b_vp != vp) || 1397 (nbp->b_flags & B_DELWRI) == 0)) { 1398 BO_UNLOCK(bo); 1399 goto restart; 1400 } 1401 } 1402 } 1403 1404 if (length > 0) { 1405 restartsync: 1406 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { 1407 if (bp->b_lblkno > 0) 1408 continue; 1409 /* 1410 * Since we hold the vnode lock this should only 1411 * fail if we're racing with the buf daemon. 1412 */ 1413 if (BUF_LOCK(bp, 1414 LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, 1415 BO_MTX(bo)) == ENOLCK) { 1416 goto restart; 1417 } 1418 VNASSERT((bp->b_flags & B_DELWRI), vp, 1419 ("buf(%p) on dirty queue without DELWRI", bp)); 1420 1421 BO_LOCK(bo); 1422 bremfree(bp); 1423 BO_UNLOCK(bo); 1424 bawrite(bp); 1425 BO_LOCK(bo); 1426 goto restartsync; 1427 } 1428 } 1429 1430 bufobj_wwait(bo, 0, 0); 1431 BO_UNLOCK(bo); 1432 vnode_pager_setsize(vp, length); 1433 1434 return (0); 1435 } 1436 1437 /* 1438 * buf_splay() - splay tree core for the clean/dirty list of buffers in 1439 * a vnode. 1440 * 1441 * NOTE: We have to deal with the special case of a background bitmap 1442 * buffer, a situation where two buffers will have the same logical 1443 * block offset. We want (1) only the foreground buffer to be accessed 1444 * in a lookup and (2) must differentiate between the foreground and 1445 * background buffer in the splay tree algorithm because the splay 1446 * tree cannot normally handle multiple entities with the same 'index'. 1447 * We accomplish this by adding differentiating flags to the splay tree's 1448 * numerical domain. 1449 */ 1450 static 1451 struct buf * 1452 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root) 1453 { 1454 struct buf dummy; 1455 struct buf *lefttreemax, *righttreemin, *y; 1456 1457 if (root == NULL) 1458 return (NULL); 1459 lefttreemax = righttreemin = &dummy; 1460 for (;;) { 1461 if (lblkno < root->b_lblkno || 1462 (lblkno == root->b_lblkno && 1463 (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1464 if ((y = root->b_left) == NULL) 1465 break; 1466 if (lblkno < y->b_lblkno) { 1467 /* Rotate right. */ 1468 root->b_left = y->b_right; 1469 y->b_right = root; 1470 root = y; 1471 if ((y = root->b_left) == NULL) 1472 break; 1473 } 1474 /* Link into the new root's right tree. */ 1475 righttreemin->b_left = root; 1476 righttreemin = root; 1477 } else if (lblkno > root->b_lblkno || 1478 (lblkno == root->b_lblkno && 1479 (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) { 1480 if ((y = root->b_right) == NULL) 1481 break; 1482 if (lblkno > y->b_lblkno) { 1483 /* Rotate left. */ 1484 root->b_right = y->b_left; 1485 y->b_left = root; 1486 root = y; 1487 if ((y = root->b_right) == NULL) 1488 break; 1489 } 1490 /* Link into the new root's left tree. */ 1491 lefttreemax->b_right = root; 1492 lefttreemax = root; 1493 } else { 1494 break; 1495 } 1496 root = y; 1497 } 1498 /* Assemble the new root. */ 1499 lefttreemax->b_right = root->b_left; 1500 righttreemin->b_left = root->b_right; 1501 root->b_left = dummy.b_right; 1502 root->b_right = dummy.b_left; 1503 return (root); 1504 } 1505 1506 static void 1507 buf_vlist_remove(struct buf *bp) 1508 { 1509 struct buf *root; 1510 struct bufv *bv; 1511 1512 KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); 1513 ASSERT_BO_LOCKED(bp->b_bufobj); 1514 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != 1515 (BX_VNDIRTY|BX_VNCLEAN), 1516 ("buf_vlist_remove: Buf %p is on two lists", bp)); 1517 if (bp->b_xflags & BX_VNDIRTY) 1518 bv = &bp->b_bufobj->bo_dirty; 1519 else 1520 bv = &bp->b_bufobj->bo_clean; 1521 if (bp != bv->bv_root) { 1522 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1523 KASSERT(root == bp, ("splay lookup failed in remove")); 1524 } 1525 if (bp->b_left == NULL) { 1526 root = bp->b_right; 1527 } else { 1528 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left); 1529 root->b_right = bp->b_right; 1530 } 1531 bv->bv_root = root; 1532 TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); 1533 bv->bv_cnt--; 1534 bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); 1535 } 1536 1537 /* 1538 * Add the buffer to the sorted clean or dirty block list using a 1539 * splay tree algorithm. 1540 * 1541 * NOTE: xflags is passed as a constant, optimizing this inline function! 1542 */ 1543 static void 1544 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) 1545 { 1546 struct buf *root; 1547 struct bufv *bv; 1548 1549 ASSERT_BO_LOCKED(bo); 1550 KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, 1551 ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); 1552 bp->b_xflags |= xflags; 1553 if (xflags & BX_VNDIRTY) 1554 bv = &bo->bo_dirty; 1555 else 1556 bv = &bo->bo_clean; 1557 1558 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root); 1559 if (root == NULL) { 1560 bp->b_left = NULL; 1561 bp->b_right = NULL; 1562 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); 1563 } else if (bp->b_lblkno < root->b_lblkno || 1564 (bp->b_lblkno == root->b_lblkno && 1565 (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) { 1566 bp->b_left = root->b_left; 1567 bp->b_right = root; 1568 root->b_left = NULL; 1569 TAILQ_INSERT_BEFORE(root, bp, b_bobufs); 1570 } else { 1571 bp->b_right = root->b_right; 1572 bp->b_left = root; 1573 root->b_right = NULL; 1574 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs); 1575 } 1576 bv->bv_cnt++; 1577 bv->bv_root = bp; 1578 } 1579 1580 /* 1581 * Lookup a buffer using the splay tree. Note that we specifically avoid 1582 * shadow buffers used in background bitmap writes. 1583 * 1584 * This code isn't quite efficient as it could be because we are maintaining 1585 * two sorted lists and do not know which list the block resides in. 1586 * 1587 * During a "make buildworld" the desired buffer is found at one of 1588 * the roots more than 60% of the time. Thus, checking both roots 1589 * before performing either splay eliminates unnecessary splays on the 1590 * first tree splayed. 1591 */ 1592 struct buf * 1593 gbincore(struct bufobj *bo, daddr_t lblkno) 1594 { 1595 struct buf *bp; 1596 1597 ASSERT_BO_LOCKED(bo); 1598 if ((bp = bo->bo_clean.bv_root) != NULL && 1599 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1600 return (bp); 1601 if ((bp = bo->bo_dirty.bv_root) != NULL && 1602 bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1603 return (bp); 1604 if ((bp = bo->bo_clean.bv_root) != NULL) { 1605 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp); 1606 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1607 return (bp); 1608 } 1609 if ((bp = bo->bo_dirty.bv_root) != NULL) { 1610 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp); 1611 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER)) 1612 return (bp); 1613 } 1614 return (NULL); 1615 } 1616 1617 /* 1618 * Associate a buffer with a vnode. 1619 */ 1620 void 1621 bgetvp(struct vnode *vp, struct buf *bp) 1622 { 1623 struct bufobj *bo; 1624 1625 bo = &vp->v_bufobj; 1626 ASSERT_BO_LOCKED(bo); 1627 VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); 1628 1629 CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); 1630 VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, 1631 ("bgetvp: bp already attached! %p", bp)); 1632 1633 vhold(vp); 1634 if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT) 1635 bp->b_flags |= B_NEEDSGIANT; 1636 bp->b_vp = vp; 1637 bp->b_bufobj = bo; 1638 /* 1639 * Insert onto list for new vnode. 1640 */ 1641 buf_vlist_add(bp, bo, BX_VNCLEAN); 1642 } 1643 1644 /* 1645 * Disassociate a buffer from a vnode. 1646 */ 1647 void 1648 brelvp(struct buf *bp) 1649 { 1650 struct bufobj *bo; 1651 struct vnode *vp; 1652 1653 CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); 1654 KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); 1655 1656 /* 1657 * Delete from old vnode list, if on one. 1658 */ 1659 vp = bp->b_vp; /* XXX */ 1660 bo = bp->b_bufobj; 1661 BO_LOCK(bo); 1662 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 1663 buf_vlist_remove(bp); 1664 else 1665 panic("brelvp: Buffer %p not on queue.", bp); 1666 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 1667 bo->bo_flag &= ~BO_ONWORKLST; 1668 mtx_lock(&sync_mtx); 1669 LIST_REMOVE(bo, bo_synclist); 1670 syncer_worklist_len--; 1671 mtx_unlock(&sync_mtx); 1672 } 1673 bp->b_flags &= ~B_NEEDSGIANT; 1674 bp->b_vp = NULL; 1675 bp->b_bufobj = NULL; 1676 BO_UNLOCK(bo); 1677 vdrop(vp); 1678 } 1679 1680 /* 1681 * Add an item to the syncer work queue. 1682 */ 1683 static void 1684 vn_syncer_add_to_worklist(struct bufobj *bo, int delay) 1685 { 1686 int queue, slot; 1687 1688 ASSERT_BO_LOCKED(bo); 1689 1690 mtx_lock(&sync_mtx); 1691 if (bo->bo_flag & BO_ONWORKLST) 1692 LIST_REMOVE(bo, bo_synclist); 1693 else { 1694 bo->bo_flag |= BO_ONWORKLST; 1695 syncer_worklist_len++; 1696 } 1697 1698 if (delay > syncer_maxdelay - 2) 1699 delay = syncer_maxdelay - 2; 1700 slot = (syncer_delayno + delay) & syncer_mask; 1701 1702 queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ : 1703 WI_MPSAFEQ; 1704 LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo, 1705 bo_synclist); 1706 mtx_unlock(&sync_mtx); 1707 } 1708 1709 static int 1710 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) 1711 { 1712 int error, len; 1713 1714 mtx_lock(&sync_mtx); 1715 len = syncer_worklist_len - sync_vnode_count; 1716 mtx_unlock(&sync_mtx); 1717 error = SYSCTL_OUT(req, &len, sizeof(len)); 1718 return (error); 1719 } 1720 1721 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0, 1722 sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); 1723 1724 static struct proc *updateproc; 1725 static void sched_sync(void); 1726 static struct kproc_desc up_kp = { 1727 "syncer", 1728 sched_sync, 1729 &updateproc 1730 }; 1731 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); 1732 1733 static int 1734 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) 1735 { 1736 struct vnode *vp; 1737 struct mount *mp; 1738 1739 *bo = LIST_FIRST(slp); 1740 if (*bo == NULL) 1741 return (0); 1742 vp = (*bo)->__bo_vnode; /* XXX */ 1743 if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) 1744 return (1); 1745 /* 1746 * We use vhold in case the vnode does not 1747 * successfully sync. vhold prevents the vnode from 1748 * going away when we unlock the sync_mtx so that 1749 * we can acquire the vnode interlock. 1750 */ 1751 vholdl(vp); 1752 mtx_unlock(&sync_mtx); 1753 VI_UNLOCK(vp); 1754 if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { 1755 vdrop(vp); 1756 mtx_lock(&sync_mtx); 1757 return (*bo == LIST_FIRST(slp)); 1758 } 1759 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1760 (void) VOP_FSYNC(vp, MNT_LAZY, td); 1761 VOP_UNLOCK(vp, 0); 1762 vn_finished_write(mp); 1763 BO_LOCK(*bo); 1764 if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { 1765 /* 1766 * Put us back on the worklist. The worklist 1767 * routine will remove us from our current 1768 * position and then add us back in at a later 1769 * position. 1770 */ 1771 vn_syncer_add_to_worklist(*bo, syncdelay); 1772 } 1773 BO_UNLOCK(*bo); 1774 vdrop(vp); 1775 mtx_lock(&sync_mtx); 1776 return (0); 1777 } 1778 1779 /* 1780 * System filesystem synchronizer daemon. 1781 */ 1782 static void 1783 sched_sync(void) 1784 { 1785 struct synclist *gnext, *next; 1786 struct synclist *gslp, *slp; 1787 struct bufobj *bo; 1788 long starttime; 1789 struct thread *td = curthread; 1790 int last_work_seen; 1791 int net_worklist_len; 1792 int syncer_final_iter; 1793 int first_printf; 1794 int error; 1795 1796 last_work_seen = 0; 1797 syncer_final_iter = 0; 1798 first_printf = 1; 1799 syncer_state = SYNCER_RUNNING; 1800 starttime = time_uptime; 1801 td->td_pflags |= TDP_NORUNNINGBUF; 1802 1803 EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, 1804 SHUTDOWN_PRI_LAST); 1805 1806 mtx_lock(&sync_mtx); 1807 for (;;) { 1808 if (syncer_state == SYNCER_FINAL_DELAY && 1809 syncer_final_iter == 0) { 1810 mtx_unlock(&sync_mtx); 1811 kproc_suspend_check(td->td_proc); 1812 mtx_lock(&sync_mtx); 1813 } 1814 net_worklist_len = syncer_worklist_len - sync_vnode_count; 1815 if (syncer_state != SYNCER_RUNNING && 1816 starttime != time_uptime) { 1817 if (first_printf) { 1818 printf("\nSyncing disks, vnodes remaining..."); 1819 first_printf = 0; 1820 } 1821 printf("%d ", net_worklist_len); 1822 } 1823 starttime = time_uptime; 1824 1825 /* 1826 * Push files whose dirty time has expired. Be careful 1827 * of interrupt race on slp queue. 1828 * 1829 * Skip over empty worklist slots when shutting down. 1830 */ 1831 do { 1832 slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1833 gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1834 syncer_delayno += 1; 1835 if (syncer_delayno == syncer_maxdelay) 1836 syncer_delayno = 0; 1837 next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno]; 1838 gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno]; 1839 /* 1840 * If the worklist has wrapped since the 1841 * it was emptied of all but syncer vnodes, 1842 * switch to the FINAL_DELAY state and run 1843 * for one more second. 1844 */ 1845 if (syncer_state == SYNCER_SHUTTING_DOWN && 1846 net_worklist_len == 0 && 1847 last_work_seen == syncer_delayno) { 1848 syncer_state = SYNCER_FINAL_DELAY; 1849 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; 1850 } 1851 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && 1852 LIST_EMPTY(gslp) && syncer_worklist_len > 0); 1853 1854 /* 1855 * Keep track of the last time there was anything 1856 * on the worklist other than syncer vnodes. 1857 * Return to the SHUTTING_DOWN state if any 1858 * new work appears. 1859 */ 1860 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) 1861 last_work_seen = syncer_delayno; 1862 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) 1863 syncer_state = SYNCER_SHUTTING_DOWN; 1864 while (!LIST_EMPTY(slp)) { 1865 error = sync_vnode(slp, &bo, td); 1866 if (error == 1) { 1867 LIST_REMOVE(bo, bo_synclist); 1868 LIST_INSERT_HEAD(next, bo, bo_synclist); 1869 continue; 1870 } 1871 #ifdef SW_WATCHDOG 1872 if (first_printf == 0) 1873 wdog_kern_pat(WD_LASTVAL); 1874 #endif 1875 } 1876 if (!LIST_EMPTY(gslp)) { 1877 mtx_unlock(&sync_mtx); 1878 mtx_lock(&Giant); 1879 mtx_lock(&sync_mtx); 1880 while (!LIST_EMPTY(gslp)) { 1881 error = sync_vnode(gslp, &bo, td); 1882 if (error == 1) { 1883 LIST_REMOVE(bo, bo_synclist); 1884 LIST_INSERT_HEAD(gnext, bo, 1885 bo_synclist); 1886 continue; 1887 } 1888 } 1889 mtx_unlock(&Giant); 1890 } 1891 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) 1892 syncer_final_iter--; 1893 /* 1894 * The variable rushjob allows the kernel to speed up the 1895 * processing of the filesystem syncer process. A rushjob 1896 * value of N tells the filesystem syncer to process the next 1897 * N seconds worth of work on its queue ASAP. Currently rushjob 1898 * is used by the soft update code to speed up the filesystem 1899 * syncer process when the incore state is getting so far 1900 * ahead of the disk that the kernel memory pool is being 1901 * threatened with exhaustion. 1902 */ 1903 if (rushjob > 0) { 1904 rushjob -= 1; 1905 continue; 1906 } 1907 /* 1908 * Just sleep for a short period of time between 1909 * iterations when shutting down to allow some I/O 1910 * to happen. 1911 * 1912 * If it has taken us less than a second to process the 1913 * current work, then wait. Otherwise start right over 1914 * again. We can still lose time if any single round 1915 * takes more than two seconds, but it does not really 1916 * matter as we are just trying to generally pace the 1917 * filesystem activity. 1918 */ 1919 if (syncer_state != SYNCER_RUNNING || 1920 time_uptime == starttime) { 1921 thread_lock(td); 1922 sched_prio(td, PPAUSE); 1923 thread_unlock(td); 1924 } 1925 if (syncer_state != SYNCER_RUNNING) 1926 cv_timedwait(&sync_wakeup, &sync_mtx, 1927 hz / SYNCER_SHUTDOWN_SPEEDUP); 1928 else if (time_uptime == starttime) 1929 cv_timedwait(&sync_wakeup, &sync_mtx, hz); 1930 } 1931 } 1932 1933 /* 1934 * Request the syncer daemon to speed up its work. 1935 * We never push it to speed up more than half of its 1936 * normal turn time, otherwise it could take over the cpu. 1937 */ 1938 int 1939 speedup_syncer(void) 1940 { 1941 int ret = 0; 1942 1943 mtx_lock(&sync_mtx); 1944 if (rushjob < syncdelay / 2) { 1945 rushjob += 1; 1946 stat_rush_requests += 1; 1947 ret = 1; 1948 } 1949 mtx_unlock(&sync_mtx); 1950 cv_broadcast(&sync_wakeup); 1951 return (ret); 1952 } 1953 1954 /* 1955 * Tell the syncer to speed up its work and run though its work 1956 * list several times, then tell it to shut down. 1957 */ 1958 static void 1959 syncer_shutdown(void *arg, int howto) 1960 { 1961 1962 if (howto & RB_NOSYNC) 1963 return; 1964 mtx_lock(&sync_mtx); 1965 syncer_state = SYNCER_SHUTTING_DOWN; 1966 rushjob = 0; 1967 mtx_unlock(&sync_mtx); 1968 cv_broadcast(&sync_wakeup); 1969 kproc_shutdown(arg, howto); 1970 } 1971 1972 /* 1973 * Reassign a buffer from one vnode to another. 1974 * Used to assign file specific control information 1975 * (indirect blocks) to the vnode to which they belong. 1976 */ 1977 void 1978 reassignbuf(struct buf *bp) 1979 { 1980 struct vnode *vp; 1981 struct bufobj *bo; 1982 int delay; 1983 #ifdef INVARIANTS 1984 struct bufv *bv; 1985 #endif 1986 1987 vp = bp->b_vp; 1988 bo = bp->b_bufobj; 1989 ++reassignbufcalls; 1990 1991 CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", 1992 bp, bp->b_vp, bp->b_flags); 1993 /* 1994 * B_PAGING flagged buffers cannot be reassigned because their vp 1995 * is not fully linked in. 1996 */ 1997 if (bp->b_flags & B_PAGING) 1998 panic("cannot reassign paging buffer"); 1999 2000 /* 2001 * Delete from old vnode list, if on one. 2002 */ 2003 BO_LOCK(bo); 2004 if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) 2005 buf_vlist_remove(bp); 2006 else 2007 panic("reassignbuf: Buffer %p not on queue.", bp); 2008 /* 2009 * If dirty, put on list of dirty buffers; otherwise insert onto list 2010 * of clean buffers. 2011 */ 2012 if (bp->b_flags & B_DELWRI) { 2013 if ((bo->bo_flag & BO_ONWORKLST) == 0) { 2014 switch (vp->v_type) { 2015 case VDIR: 2016 delay = dirdelay; 2017 break; 2018 case VCHR: 2019 delay = metadelay; 2020 break; 2021 default: 2022 delay = filedelay; 2023 } 2024 vn_syncer_add_to_worklist(bo, delay); 2025 } 2026 buf_vlist_add(bp, bo, BX_VNDIRTY); 2027 } else { 2028 buf_vlist_add(bp, bo, BX_VNCLEAN); 2029 2030 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { 2031 mtx_lock(&sync_mtx); 2032 LIST_REMOVE(bo, bo_synclist); 2033 syncer_worklist_len--; 2034 mtx_unlock(&sync_mtx); 2035 bo->bo_flag &= ~BO_ONWORKLST; 2036 } 2037 } 2038 #ifdef INVARIANTS 2039 bv = &bo->bo_clean; 2040 bp = TAILQ_FIRST(&bv->bv_hd); 2041 KASSERT(bp == NULL || bp->b_bufobj == bo, 2042 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2043 bp = TAILQ_LAST(&bv->bv_hd, buflists); 2044 KASSERT(bp == NULL || bp->b_bufobj == bo, 2045 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2046 bv = &bo->bo_dirty; 2047 bp = TAILQ_FIRST(&bv->bv_hd); 2048 KASSERT(bp == NULL || bp->b_bufobj == bo, 2049 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2050 bp = TAILQ_LAST(&bv->bv_hd, buflists); 2051 KASSERT(bp == NULL || bp->b_bufobj == bo, 2052 ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); 2053 #endif 2054 BO_UNLOCK(bo); 2055 } 2056 2057 /* 2058 * Increment the use and hold counts on the vnode, taking care to reference 2059 * the driver's usecount if this is a chardev. The vholdl() will remove 2060 * the vnode from the free list if it is presently free. Requires the 2061 * vnode interlock and returns with it held. 2062 */ 2063 static void 2064 v_incr_usecount(struct vnode *vp) 2065 { 2066 2067 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2068 vp->v_usecount++; 2069 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2070 dev_lock(); 2071 vp->v_rdev->si_usecount++; 2072 dev_unlock(); 2073 } 2074 vholdl(vp); 2075 } 2076 2077 /* 2078 * Turn a holdcnt into a use+holdcnt such that only one call to 2079 * v_decr_usecount is needed. 2080 */ 2081 static void 2082 v_upgrade_usecount(struct vnode *vp) 2083 { 2084 2085 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2086 vp->v_usecount++; 2087 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2088 dev_lock(); 2089 vp->v_rdev->si_usecount++; 2090 dev_unlock(); 2091 } 2092 } 2093 2094 /* 2095 * Decrement the vnode use and hold count along with the driver's usecount 2096 * if this is a chardev. The vdropl() below releases the vnode interlock 2097 * as it may free the vnode. 2098 */ 2099 static void 2100 v_decr_usecount(struct vnode *vp) 2101 { 2102 2103 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2104 VNASSERT(vp->v_usecount > 0, vp, 2105 ("v_decr_usecount: negative usecount")); 2106 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2107 vp->v_usecount--; 2108 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2109 dev_lock(); 2110 vp->v_rdev->si_usecount--; 2111 dev_unlock(); 2112 } 2113 vdropl(vp); 2114 } 2115 2116 /* 2117 * Decrement only the use count and driver use count. This is intended to 2118 * be paired with a follow on vdropl() to release the remaining hold count. 2119 * In this way we may vgone() a vnode with a 0 usecount without risk of 2120 * having it end up on a free list because the hold count is kept above 0. 2121 */ 2122 static void 2123 v_decr_useonly(struct vnode *vp) 2124 { 2125 2126 ASSERT_VI_LOCKED(vp, __FUNCTION__); 2127 VNASSERT(vp->v_usecount > 0, vp, 2128 ("v_decr_useonly: negative usecount")); 2129 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2130 vp->v_usecount--; 2131 if (vp->v_type == VCHR && vp->v_rdev != NULL) { 2132 dev_lock(); 2133 vp->v_rdev->si_usecount--; 2134 dev_unlock(); 2135 } 2136 } 2137 2138 /* 2139 * Grab a particular vnode from the free list, increment its 2140 * reference count and lock it. VI_DOOMED is set if the vnode 2141 * is being destroyed. Only callers who specify LK_RETRY will 2142 * see doomed vnodes. If inactive processing was delayed in 2143 * vput try to do it here. 2144 */ 2145 int 2146 vget(struct vnode *vp, int flags, struct thread *td) 2147 { 2148 int error; 2149 2150 error = 0; 2151 VFS_ASSERT_GIANT(vp->v_mount); 2152 VNASSERT((flags & LK_TYPE_MASK) != 0, vp, 2153 ("vget: invalid lock operation")); 2154 CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); 2155 2156 if ((flags & LK_INTERLOCK) == 0) 2157 VI_LOCK(vp); 2158 vholdl(vp); 2159 if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) { 2160 vdrop(vp); 2161 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, 2162 vp); 2163 return (error); 2164 } 2165 if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) 2166 panic("vget: vn_lock failed to return ENOENT\n"); 2167 VI_LOCK(vp); 2168 /* Upgrade our holdcnt to a usecount. */ 2169 v_upgrade_usecount(vp); 2170 /* 2171 * We don't guarantee that any particular close will 2172 * trigger inactive processing so just make a best effort 2173 * here at preventing a reference to a removed file. If 2174 * we don't succeed no harm is done. 2175 */ 2176 if (vp->v_iflag & VI_OWEINACT) { 2177 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE && 2178 (flags & LK_NOWAIT) == 0) 2179 vinactive(vp, td); 2180 vp->v_iflag &= ~VI_OWEINACT; 2181 } 2182 VI_UNLOCK(vp); 2183 return (0); 2184 } 2185 2186 /* 2187 * Increase the reference count of a vnode. 2188 */ 2189 void 2190 vref(struct vnode *vp) 2191 { 2192 2193 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2194 VI_LOCK(vp); 2195 v_incr_usecount(vp); 2196 VI_UNLOCK(vp); 2197 } 2198 2199 /* 2200 * Return reference count of a vnode. 2201 * 2202 * The results of this call are only guaranteed when some mechanism other 2203 * than the VI lock is used to stop other processes from gaining references 2204 * to the vnode. This may be the case if the caller holds the only reference. 2205 * This is also useful when stale data is acceptable as race conditions may 2206 * be accounted for by some other means. 2207 */ 2208 int 2209 vrefcnt(struct vnode *vp) 2210 { 2211 int usecnt; 2212 2213 VI_LOCK(vp); 2214 usecnt = vp->v_usecount; 2215 VI_UNLOCK(vp); 2216 2217 return (usecnt); 2218 } 2219 2220 #define VPUTX_VRELE 1 2221 #define VPUTX_VPUT 2 2222 #define VPUTX_VUNREF 3 2223 2224 static void 2225 vputx(struct vnode *vp, int func) 2226 { 2227 int error; 2228 2229 KASSERT(vp != NULL, ("vputx: null vp")); 2230 if (func == VPUTX_VUNREF) 2231 ASSERT_VOP_LOCKED(vp, "vunref"); 2232 else if (func == VPUTX_VPUT) 2233 ASSERT_VOP_LOCKED(vp, "vput"); 2234 else 2235 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func")); 2236 VFS_ASSERT_GIANT(vp->v_mount); 2237 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2238 VI_LOCK(vp); 2239 2240 /* Skip this v_writecount check if we're going to panic below. */ 2241 VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp, 2242 ("vputx: missed vn_close")); 2243 error = 0; 2244 2245 if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) && 2246 vp->v_usecount == 1)) { 2247 if (func == VPUTX_VPUT) 2248 VOP_UNLOCK(vp, 0); 2249 v_decr_usecount(vp); 2250 return; 2251 } 2252 2253 if (vp->v_usecount != 1) { 2254 vprint("vputx: negative ref count", vp); 2255 panic("vputx: negative ref cnt"); 2256 } 2257 CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp); 2258 /* 2259 * We want to hold the vnode until the inactive finishes to 2260 * prevent vgone() races. We drop the use count here and the 2261 * hold count below when we're done. 2262 */ 2263 v_decr_useonly(vp); 2264 /* 2265 * We must call VOP_INACTIVE with the node locked. Mark 2266 * as VI_DOINGINACT to avoid recursion. 2267 */ 2268 vp->v_iflag |= VI_OWEINACT; 2269 switch (func) { 2270 case VPUTX_VRELE: 2271 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); 2272 VI_LOCK(vp); 2273 break; 2274 case VPUTX_VPUT: 2275 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { 2276 error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | 2277 LK_NOWAIT); 2278 VI_LOCK(vp); 2279 } 2280 break; 2281 case VPUTX_VUNREF: 2282 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 2283 error = EBUSY; 2284 break; 2285 } 2286 if (vp->v_usecount > 0) 2287 vp->v_iflag &= ~VI_OWEINACT; 2288 if (error == 0) { 2289 if (vp->v_iflag & VI_OWEINACT) 2290 vinactive(vp, curthread); 2291 if (func != VPUTX_VUNREF) 2292 VOP_UNLOCK(vp, 0); 2293 } 2294 vdropl(vp); 2295 } 2296 2297 /* 2298 * Vnode put/release. 2299 * If count drops to zero, call inactive routine and return to freelist. 2300 */ 2301 void 2302 vrele(struct vnode *vp) 2303 { 2304 2305 vputx(vp, VPUTX_VRELE); 2306 } 2307 2308 /* 2309 * Release an already locked vnode. This give the same effects as 2310 * unlock+vrele(), but takes less time and avoids releasing and 2311 * re-aquiring the lock (as vrele() acquires the lock internally.) 2312 */ 2313 void 2314 vput(struct vnode *vp) 2315 { 2316 2317 vputx(vp, VPUTX_VPUT); 2318 } 2319 2320 /* 2321 * Release an exclusively locked vnode. Do not unlock the vnode lock. 2322 */ 2323 void 2324 vunref(struct vnode *vp) 2325 { 2326 2327 vputx(vp, VPUTX_VUNREF); 2328 } 2329 2330 /* 2331 * Somebody doesn't want the vnode recycled. 2332 */ 2333 void 2334 vhold(struct vnode *vp) 2335 { 2336 2337 VI_LOCK(vp); 2338 vholdl(vp); 2339 VI_UNLOCK(vp); 2340 } 2341 2342 /* 2343 * Increase the hold count and activate if this is the first reference. 2344 */ 2345 void 2346 vholdl(struct vnode *vp) 2347 { 2348 struct mount *mp; 2349 2350 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2351 vp->v_holdcnt++; 2352 if (!VSHOULDBUSY(vp)) 2353 return; 2354 ASSERT_VI_LOCKED(vp, "vholdl"); 2355 VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free")); 2356 VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed.")); 2357 /* 2358 * Remove a vnode from the free list, mark it as in use, 2359 * and put it on the active list. 2360 */ 2361 mtx_lock(&vnode_free_list_mtx); 2362 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist); 2363 freevnodes--; 2364 vp->v_iflag &= ~(VI_FREE|VI_AGE); 2365 KASSERT((vp->v_iflag & VI_ACTIVE) == 0, 2366 ("Activating already active vnode")); 2367 vp->v_iflag |= VI_ACTIVE; 2368 mp = vp->v_mount; 2369 TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist); 2370 mp->mnt_activevnodelistsize++; 2371 mtx_unlock(&vnode_free_list_mtx); 2372 } 2373 2374 /* 2375 * Note that there is one less who cares about this vnode. 2376 * vdrop() is the opposite of vhold(). 2377 */ 2378 void 2379 vdrop(struct vnode *vp) 2380 { 2381 2382 VI_LOCK(vp); 2383 vdropl(vp); 2384 } 2385 2386 /* 2387 * Drop the hold count of the vnode. If this is the last reference to 2388 * the vnode we place it on the free list unless it has been vgone'd 2389 * (marked VI_DOOMED) in which case we will free it. 2390 */ 2391 void 2392 vdropl(struct vnode *vp) 2393 { 2394 struct bufobj *bo; 2395 struct mount *mp; 2396 int active; 2397 2398 ASSERT_VI_LOCKED(vp, "vdropl"); 2399 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2400 if (vp->v_holdcnt <= 0) 2401 panic("vdrop: holdcnt %d", vp->v_holdcnt); 2402 vp->v_holdcnt--; 2403 if (vp->v_holdcnt > 0) { 2404 VI_UNLOCK(vp); 2405 return; 2406 } 2407 if ((vp->v_iflag & VI_DOOMED) == 0) { 2408 /* 2409 * Mark a vnode as free: remove it from its active list 2410 * and put it up for recycling on the freelist. 2411 */ 2412 VNASSERT(vp->v_op != NULL, vp, 2413 ("vdropl: vnode already reclaimed.")); 2414 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, 2415 ("vnode already free")); 2416 VNASSERT(VSHOULDFREE(vp), vp, 2417 ("vdropl: freeing when we shouldn't")); 2418 active = vp->v_iflag & VI_ACTIVE; 2419 vp->v_iflag &= ~VI_ACTIVE; 2420 mp = vp->v_mount; 2421 mtx_lock(&vnode_free_list_mtx); 2422 if (active) { 2423 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, 2424 v_actfreelist); 2425 mp->mnt_activevnodelistsize--; 2426 } 2427 if (vp->v_iflag & VI_AGE) { 2428 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist); 2429 } else { 2430 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist); 2431 } 2432 freevnodes++; 2433 vp->v_iflag &= ~VI_AGE; 2434 vp->v_iflag |= VI_FREE; 2435 mtx_unlock(&vnode_free_list_mtx); 2436 VI_UNLOCK(vp); 2437 return; 2438 } 2439 /* 2440 * The vnode has been marked for destruction, so free it. 2441 */ 2442 CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp); 2443 mtx_lock(&vnode_free_list_mtx); 2444 numvnodes--; 2445 mtx_unlock(&vnode_free_list_mtx); 2446 bo = &vp->v_bufobj; 2447 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp, 2448 ("cleaned vnode still on the free list.")); 2449 VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); 2450 VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); 2451 VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); 2452 VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); 2453 VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); 2454 VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); 2455 VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL")); 2456 VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); 2457 VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL")); 2458 VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); 2459 VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); 2460 VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); 2461 VI_UNLOCK(vp); 2462 #ifdef MAC 2463 mac_vnode_destroy(vp); 2464 #endif 2465 if (vp->v_pollinfo != NULL) 2466 destroy_vpollinfo(vp->v_pollinfo); 2467 #ifdef INVARIANTS 2468 /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */ 2469 vp->v_op = NULL; 2470 #endif 2471 lockdestroy(vp->v_vnlock); 2472 mtx_destroy(&vp->v_interlock); 2473 mtx_destroy(BO_MTX(bo)); 2474 uma_zfree(vnode_zone, vp); 2475 } 2476 2477 /* 2478 * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT 2479 * flags. DOINGINACT prevents us from recursing in calls to vinactive. 2480 * OWEINACT tracks whether a vnode missed a call to inactive due to a 2481 * failed lock upgrade. 2482 */ 2483 void 2484 vinactive(struct vnode *vp, struct thread *td) 2485 { 2486 struct vm_object *obj; 2487 2488 ASSERT_VOP_ELOCKED(vp, "vinactive"); 2489 ASSERT_VI_LOCKED(vp, "vinactive"); 2490 VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, 2491 ("vinactive: recursed on VI_DOINGINACT")); 2492 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2493 vp->v_iflag |= VI_DOINGINACT; 2494 vp->v_iflag &= ~VI_OWEINACT; 2495 VI_UNLOCK(vp); 2496 /* 2497 * Before moving off the active list, we must be sure that any 2498 * modified pages are on the vnode's dirty list since these will 2499 * no longer be checked once the vnode is on the inactive list. 2500 */ 2501 obj = vp->v_object; 2502 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) { 2503 VM_OBJECT_LOCK(obj); 2504 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC); 2505 VM_OBJECT_UNLOCK(obj); 2506 } 2507 VOP_INACTIVE(vp, td); 2508 VI_LOCK(vp); 2509 VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, 2510 ("vinactive: lost VI_DOINGINACT")); 2511 vp->v_iflag &= ~VI_DOINGINACT; 2512 } 2513 2514 /* 2515 * Remove any vnodes in the vnode table belonging to mount point mp. 2516 * 2517 * If FORCECLOSE is not specified, there should not be any active ones, 2518 * return error if any are found (nb: this is a user error, not a 2519 * system error). If FORCECLOSE is specified, detach any active vnodes 2520 * that are found. 2521 * 2522 * If WRITECLOSE is set, only flush out regular file vnodes open for 2523 * writing. 2524 * 2525 * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. 2526 * 2527 * `rootrefs' specifies the base reference count for the root vnode 2528 * of this filesystem. The root vnode is considered busy if its 2529 * v_usecount exceeds this value. On a successful return, vflush(, td) 2530 * will call vrele() on the root vnode exactly rootrefs times. 2531 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must 2532 * be zero. 2533 */ 2534 #ifdef DIAGNOSTIC 2535 static int busyprt = 0; /* print out busy vnodes */ 2536 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes"); 2537 #endif 2538 2539 int 2540 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td) 2541 { 2542 struct vnode *vp, *mvp, *rootvp = NULL; 2543 struct vattr vattr; 2544 int busy = 0, error; 2545 2546 CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, 2547 rootrefs, flags); 2548 if (rootrefs > 0) { 2549 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, 2550 ("vflush: bad args")); 2551 /* 2552 * Get the filesystem root vnode. We can vput() it 2553 * immediately, since with rootrefs > 0, it won't go away. 2554 */ 2555 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) { 2556 CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", 2557 __func__, error); 2558 return (error); 2559 } 2560 vput(rootvp); 2561 } 2562 loop: 2563 MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { 2564 vholdl(vp); 2565 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); 2566 if (error) { 2567 vdrop(vp); 2568 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 2569 goto loop; 2570 } 2571 /* 2572 * Skip over a vnodes marked VV_SYSTEM. 2573 */ 2574 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { 2575 VOP_UNLOCK(vp, 0); 2576 vdrop(vp); 2577 continue; 2578 } 2579 /* 2580 * If WRITECLOSE is set, flush out unlinked but still open 2581 * files (even if open only for reading) and regular file 2582 * vnodes open for writing. 2583 */ 2584 if (flags & WRITECLOSE) { 2585 if (vp->v_object != NULL) { 2586 VM_OBJECT_LOCK(vp->v_object); 2587 vm_object_page_clean(vp->v_object, 0, 0, 0); 2588 VM_OBJECT_UNLOCK(vp->v_object); 2589 } 2590 error = VOP_FSYNC(vp, MNT_WAIT, td); 2591 if (error != 0) { 2592 VOP_UNLOCK(vp, 0); 2593 vdrop(vp); 2594 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); 2595 return (error); 2596 } 2597 error = VOP_GETATTR(vp, &vattr, td->td_ucred); 2598 VI_LOCK(vp); 2599 2600 if ((vp->v_type == VNON || 2601 (error == 0 && vattr.va_nlink > 0)) && 2602 (vp->v_writecount == 0 || vp->v_type != VREG)) { 2603 VOP_UNLOCK(vp, 0); 2604 vdropl(vp); 2605 continue; 2606 } 2607 } else 2608 VI_LOCK(vp); 2609 /* 2610 * With v_usecount == 0, all we need to do is clear out the 2611 * vnode data structures and we are done. 2612 * 2613 * If FORCECLOSE is set, forcibly close the vnode. 2614 */ 2615 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { 2616 VNASSERT(vp->v_usecount == 0 || 2617 (vp->v_type != VCHR && vp->v_type != VBLK), vp, 2618 ("device VNODE %p is FORCECLOSED", vp)); 2619 vgonel(vp); 2620 } else { 2621 busy++; 2622 #ifdef DIAGNOSTIC 2623 if (busyprt) 2624 vprint("vflush: busy vnode", vp); 2625 #endif 2626 } 2627 VOP_UNLOCK(vp, 0); 2628 vdropl(vp); 2629 } 2630 if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { 2631 /* 2632 * If just the root vnode is busy, and if its refcount 2633 * is equal to `rootrefs', then go ahead and kill it. 2634 */ 2635 VI_LOCK(rootvp); 2636 KASSERT(busy > 0, ("vflush: not busy")); 2637 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, 2638 ("vflush: usecount %d < rootrefs %d", 2639 rootvp->v_usecount, rootrefs)); 2640 if (busy == 1 && rootvp->v_usecount == rootrefs) { 2641 VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); 2642 vgone(rootvp); 2643 VOP_UNLOCK(rootvp, 0); 2644 busy = 0; 2645 } else 2646 VI_UNLOCK(rootvp); 2647 } 2648 if (busy) { 2649 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, 2650 busy); 2651 return (EBUSY); 2652 } 2653 for (; rootrefs > 0; rootrefs--) 2654 vrele(rootvp); 2655 return (0); 2656 } 2657 2658 /* 2659 * Recycle an unused vnode to the front of the free list. 2660 */ 2661 int 2662 vrecycle(struct vnode *vp) 2663 { 2664 int recycled; 2665 2666 ASSERT_VOP_ELOCKED(vp, "vrecycle"); 2667 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2668 recycled = 0; 2669 VI_LOCK(vp); 2670 if (vp->v_usecount == 0) { 2671 recycled = 1; 2672 vgonel(vp); 2673 } 2674 VI_UNLOCK(vp); 2675 return (recycled); 2676 } 2677 2678 /* 2679 * Eliminate all activity associated with a vnode 2680 * in preparation for reuse. 2681 */ 2682 void 2683 vgone(struct vnode *vp) 2684 { 2685 VI_LOCK(vp); 2686 vgonel(vp); 2687 VI_UNLOCK(vp); 2688 } 2689 2690 /* 2691 * vgone, with the vp interlock held. 2692 */ 2693 void 2694 vgonel(struct vnode *vp) 2695 { 2696 struct thread *td; 2697 int oweinact; 2698 int active; 2699 struct mount *mp; 2700 2701 ASSERT_VOP_ELOCKED(vp, "vgonel"); 2702 ASSERT_VI_LOCKED(vp, "vgonel"); 2703 VNASSERT(vp->v_holdcnt, vp, 2704 ("vgonel: vp %p has no reference.", vp)); 2705 CTR2(KTR_VFS, "%s: vp %p", __func__, vp); 2706 td = curthread; 2707 2708 /* 2709 * Don't vgonel if we're already doomed. 2710 */ 2711 if (vp->v_iflag & VI_DOOMED) 2712 return; 2713 vp->v_iflag |= VI_DOOMED; 2714 /* 2715 * Check to see if the vnode is in use. If so, we have to call 2716 * VOP_CLOSE() and VOP_INACTIVE(). 2717 */ 2718 active = vp->v_usecount; 2719 oweinact = (vp->v_iflag & VI_OWEINACT); 2720 VI_UNLOCK(vp); 2721 /* 2722 * Clean out any buffers associated with the vnode. 2723 * If the flush fails, just toss the buffers. 2724 */ 2725 mp = NULL; 2726 if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) 2727 (void) vn_start_secondary_write(vp, &mp, V_WAIT); 2728 if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) 2729 vinvalbuf(vp, 0, 0, 0); 2730 2731 /* 2732 * If purging an active vnode, it must be closed and 2733 * deactivated before being reclaimed. 2734 */ 2735 if (active) 2736 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); 2737 if (oweinact || active) { 2738 VI_LOCK(vp); 2739 if ((vp->v_iflag & VI_DOINGINACT) == 0) 2740 vinactive(vp, td); 2741 VI_UNLOCK(vp); 2742 } 2743 if (vp->v_type == VSOCK) 2744 vfs_unp_reclaim(vp); 2745 /* 2746 * Reclaim the vnode. 2747 */ 2748 if (VOP_RECLAIM(vp, td)) 2749 panic("vgone: cannot reclaim"); 2750 if (mp != NULL) 2751 vn_finished_secondary_write(mp); 2752 VNASSERT(vp->v_object == NULL, vp, 2753 ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag)); 2754 /* 2755 * Clear the advisory locks and wake up waiting threads. 2756 */ 2757 (void)VOP_ADVLOCKPURGE(vp); 2758 /* 2759 * Delete from old mount point vnode list. 2760 */ 2761 delmntque(vp); 2762 cache_purge(vp); 2763 /* 2764 * Done with purge, reset to the standard lock and invalidate 2765 * the vnode. 2766 */ 2767 VI_LOCK(vp); 2768 vp->v_vnlock = &vp->v_lock; 2769 vp->v_op = &dead_vnodeops; 2770 vp->v_tag = "none"; 2771 vp->v_type = VBAD; 2772 } 2773 2774 /* 2775 * Calculate the total number of references to a special device. 2776 */ 2777 int 2778 vcount(struct vnode *vp) 2779 { 2780 int count; 2781 2782 dev_lock(); 2783 count = vp->v_rdev->si_usecount; 2784 dev_unlock(); 2785 return (count); 2786 } 2787 2788 /* 2789 * Same as above, but using the struct cdev *as argument 2790 */ 2791 int 2792 count_dev(struct cdev *dev) 2793 { 2794 int count; 2795 2796 dev_lock(); 2797 count = dev->si_usecount; 2798 dev_unlock(); 2799 return(count); 2800 } 2801 2802 /* 2803 * Print out a description of a vnode. 2804 */ 2805 static char *typename[] = 2806 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", 2807 "VMARKER"}; 2808 2809 void 2810 vn_printf(struct vnode *vp, const char *fmt, ...) 2811 { 2812 va_list ap; 2813 char buf[256], buf2[16]; 2814 u_long flags; 2815 2816 va_start(ap, fmt); 2817 vprintf(fmt, ap); 2818 va_end(ap); 2819 printf("%p: ", (void *)vp); 2820 printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]); 2821 printf(" usecount %d, writecount %d, refcount %d mountedhere %p\n", 2822 vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere); 2823 buf[0] = '\0'; 2824 buf[1] = '\0'; 2825 if (vp->v_vflag & VV_ROOT) 2826 strlcat(buf, "|VV_ROOT", sizeof(buf)); 2827 if (vp->v_vflag & VV_ISTTY) 2828 strlcat(buf, "|VV_ISTTY", sizeof(buf)); 2829 if (vp->v_vflag & VV_NOSYNC) 2830 strlcat(buf, "|VV_NOSYNC", sizeof(buf)); 2831 if (vp->v_vflag & VV_CACHEDLABEL) 2832 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); 2833 if (vp->v_vflag & VV_TEXT) 2834 strlcat(buf, "|VV_TEXT", sizeof(buf)); 2835 if (vp->v_vflag & VV_COPYONWRITE) 2836 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); 2837 if (vp->v_vflag & VV_SYSTEM) 2838 strlcat(buf, "|VV_SYSTEM", sizeof(buf)); 2839 if (vp->v_vflag & VV_PROCDEP) 2840 strlcat(buf, "|VV_PROCDEP", sizeof(buf)); 2841 if (vp->v_vflag & VV_NOKNOTE) 2842 strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); 2843 if (vp->v_vflag & VV_DELETED) 2844 strlcat(buf, "|VV_DELETED", sizeof(buf)); 2845 if (vp->v_vflag & VV_MD) 2846 strlcat(buf, "|VV_MD", sizeof(buf)); 2847 flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | 2848 VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | 2849 VV_NOKNOTE | VV_DELETED | VV_MD); 2850 if (flags != 0) { 2851 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); 2852 strlcat(buf, buf2, sizeof(buf)); 2853 } 2854 if (vp->v_iflag & VI_MOUNT) 2855 strlcat(buf, "|VI_MOUNT", sizeof(buf)); 2856 if (vp->v_iflag & VI_AGE) 2857 strlcat(buf, "|VI_AGE", sizeof(buf)); 2858 if (vp->v_iflag & VI_DOOMED) 2859 strlcat(buf, "|VI_DOOMED", sizeof(buf)); 2860 if (vp->v_iflag & VI_FREE) 2861 strlcat(buf, "|VI_FREE", sizeof(buf)); 2862 if (vp->v_iflag & VI_DOINGINACT) 2863 strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); 2864 if (vp->v_iflag & VI_OWEINACT) 2865 strlcat(buf, "|VI_OWEINACT", sizeof(buf)); 2866 flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE | 2867 VI_DOINGINACT | VI_OWEINACT); 2868 if (flags != 0) { 2869 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); 2870 strlcat(buf, buf2, sizeof(buf)); 2871 } 2872 printf(" flags (%s)\n", buf + 1); 2873 if (mtx_owned(VI_MTX(vp))) 2874 printf(" VI_LOCKed"); 2875 if (vp->v_object != NULL) 2876 printf(" v_object %p ref %d pages %d\n", 2877 vp->v_object, vp->v_object->ref_count, 2878 vp->v_object->resident_page_count); 2879 printf(" "); 2880 lockmgr_printinfo(vp->v_vnlock); 2881 if (vp->v_data != NULL) 2882 VOP_PRINT(vp); 2883 } 2884 2885 #ifdef DDB 2886 /* 2887 * List all of the locked vnodes in the system. 2888 * Called when debugging the kernel. 2889 */ 2890 DB_SHOW_COMMAND(lockedvnods, lockedvnodes) 2891 { 2892 struct mount *mp, *nmp; 2893 struct vnode *vp; 2894 2895 /* 2896 * Note: because this is DDB, we can't obey the locking semantics 2897 * for these structures, which means we could catch an inconsistent 2898 * state and dereference a nasty pointer. Not much to be done 2899 * about that. 2900 */ 2901 db_printf("Locked vnodes\n"); 2902 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) { 2903 nmp = TAILQ_NEXT(mp, mnt_list); 2904 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 2905 if (vp->v_type != VMARKER && 2906 VOP_ISLOCKED(vp)) 2907 vprint("", vp); 2908 } 2909 nmp = TAILQ_NEXT(mp, mnt_list); 2910 } 2911 } 2912 2913 /* 2914 * Show details about the given vnode. 2915 */ 2916 DB_SHOW_COMMAND(vnode, db_show_vnode) 2917 { 2918 struct vnode *vp; 2919 2920 if (!have_addr) 2921 return; 2922 vp = (struct vnode *)addr; 2923 vn_printf(vp, "vnode "); 2924 } 2925 2926 /* 2927 * Show details about the given mount point. 2928 */ 2929 DB_SHOW_COMMAND(mount, db_show_mount) 2930 { 2931 struct mount *mp; 2932 struct vfsopt *opt; 2933 struct statfs *sp; 2934 struct vnode *vp; 2935 char buf[512]; 2936 uint64_t mflags; 2937 u_int flags; 2938 2939 if (!have_addr) { 2940 /* No address given, print short info about all mount points. */ 2941 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 2942 db_printf("%p %s on %s (%s)\n", mp, 2943 mp->mnt_stat.f_mntfromname, 2944 mp->mnt_stat.f_mntonname, 2945 mp->mnt_stat.f_fstypename); 2946 if (db_pager_quit) 2947 break; 2948 } 2949 db_printf("\nMore info: show mount <addr>\n"); 2950 return; 2951 } 2952 2953 mp = (struct mount *)addr; 2954 db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, 2955 mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); 2956 2957 buf[0] = '\0'; 2958 mflags = mp->mnt_flag; 2959 #define MNT_FLAG(flag) do { \ 2960 if (mflags & (flag)) { \ 2961 if (buf[0] != '\0') \ 2962 strlcat(buf, ", ", sizeof(buf)); \ 2963 strlcat(buf, (#flag) + 4, sizeof(buf)); \ 2964 mflags &= ~(flag); \ 2965 } \ 2966 } while (0) 2967 MNT_FLAG(MNT_RDONLY); 2968 MNT_FLAG(MNT_SYNCHRONOUS); 2969 MNT_FLAG(MNT_NOEXEC); 2970 MNT_FLAG(MNT_NOSUID); 2971 MNT_FLAG(MNT_UNION); 2972 MNT_FLAG(MNT_ASYNC); 2973 MNT_FLAG(MNT_SUIDDIR); 2974 MNT_FLAG(MNT_SOFTDEP); 2975 MNT_FLAG(MNT_SUJ); 2976 MNT_FLAG(MNT_NOSYMFOLLOW); 2977 MNT_FLAG(MNT_GJOURNAL); 2978 MNT_FLAG(MNT_MULTILABEL); 2979 MNT_FLAG(MNT_ACLS); 2980 MNT_FLAG(MNT_NOATIME); 2981 MNT_FLAG(MNT_NOCLUSTERR); 2982 MNT_FLAG(MNT_NOCLUSTERW); 2983 MNT_FLAG(MNT_NFS4ACLS); 2984 MNT_FLAG(MNT_EXRDONLY); 2985 MNT_FLAG(MNT_EXPORTED); 2986 MNT_FLAG(MNT_DEFEXPORTED); 2987 MNT_FLAG(MNT_EXPORTANON); 2988 MNT_FLAG(MNT_EXKERB); 2989 MNT_FLAG(MNT_EXPUBLIC); 2990 MNT_FLAG(MNT_LOCAL); 2991 MNT_FLAG(MNT_QUOTA); 2992 MNT_FLAG(MNT_ROOTFS); 2993 MNT_FLAG(MNT_USER); 2994 MNT_FLAG(MNT_IGNORE); 2995 MNT_FLAG(MNT_UPDATE); 2996 MNT_FLAG(MNT_DELEXPORT); 2997 MNT_FLAG(MNT_RELOAD); 2998 MNT_FLAG(MNT_FORCE); 2999 MNT_FLAG(MNT_SNAPSHOT); 3000 MNT_FLAG(MNT_BYFSID); 3001 #undef MNT_FLAG 3002 if (mflags != 0) { 3003 if (buf[0] != '\0') 3004 strlcat(buf, ", ", sizeof(buf)); 3005 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 3006 "0x%016jx", mflags); 3007 } 3008 db_printf(" mnt_flag = %s\n", buf); 3009 3010 buf[0] = '\0'; 3011 flags = mp->mnt_kern_flag; 3012 #define MNT_KERN_FLAG(flag) do { \ 3013 if (flags & (flag)) { \ 3014 if (buf[0] != '\0') \ 3015 strlcat(buf, ", ", sizeof(buf)); \ 3016 strlcat(buf, (#flag) + 5, sizeof(buf)); \ 3017 flags &= ~(flag); \ 3018 } \ 3019 } while (0) 3020 MNT_KERN_FLAG(MNTK_UNMOUNTF); 3021 MNT_KERN_FLAG(MNTK_ASYNC); 3022 MNT_KERN_FLAG(MNTK_SOFTDEP); 3023 MNT_KERN_FLAG(MNTK_NOINSMNTQ); 3024 MNT_KERN_FLAG(MNTK_DRAINING); 3025 MNT_KERN_FLAG(MNTK_REFEXPIRE); 3026 MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); 3027 MNT_KERN_FLAG(MNTK_SHARED_WRITES); 3028 MNT_KERN_FLAG(MNTK_NOASYNC); 3029 MNT_KERN_FLAG(MNTK_UNMOUNT); 3030 MNT_KERN_FLAG(MNTK_MWAIT); 3031 MNT_KERN_FLAG(MNTK_SUSPEND); 3032 MNT_KERN_FLAG(MNTK_SUSPEND2); 3033 MNT_KERN_FLAG(MNTK_SUSPENDED); 3034 MNT_KERN_FLAG(MNTK_MPSAFE); 3035 MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); 3036 MNT_KERN_FLAG(MNTK_NOKNOTE); 3037 #undef MNT_KERN_FLAG 3038 if (flags != 0) { 3039 if (buf[0] != '\0') 3040 strlcat(buf, ", ", sizeof(buf)); 3041 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), 3042 "0x%08x", flags); 3043 } 3044 db_printf(" mnt_kern_flag = %s\n", buf); 3045 3046 db_printf(" mnt_opt = "); 3047 opt = TAILQ_FIRST(mp->mnt_opt); 3048 if (opt != NULL) { 3049 db_printf("%s", opt->name); 3050 opt = TAILQ_NEXT(opt, link); 3051 while (opt != NULL) { 3052 db_printf(", %s", opt->name); 3053 opt = TAILQ_NEXT(opt, link); 3054 } 3055 } 3056 db_printf("\n"); 3057 3058 sp = &mp->mnt_stat; 3059 db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " 3060 "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " 3061 "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " 3062 "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", 3063 (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, 3064 (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, 3065 (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, 3066 (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, 3067 (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, 3068 (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, 3069 (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, 3070 (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); 3071 3072 db_printf(" mnt_cred = { uid=%u ruid=%u", 3073 (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); 3074 if (jailed(mp->mnt_cred)) 3075 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); 3076 db_printf(" }\n"); 3077 db_printf(" mnt_ref = %d\n", mp->mnt_ref); 3078 db_printf(" mnt_gen = %d\n", mp->mnt_gen); 3079 db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); 3080 db_printf(" mnt_activevnodelistsize = %d\n", 3081 mp->mnt_activevnodelistsize); 3082 db_printf(" mnt_writeopcount = %d\n", mp->mnt_writeopcount); 3083 db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); 3084 db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); 3085 db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); 3086 db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); 3087 db_printf(" mnt_secondary_accwrites = %d\n", 3088 mp->mnt_secondary_accwrites); 3089 db_printf(" mnt_gjprovider = %s\n", 3090 mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); 3091 3092 db_printf("\n\nList of active vnodes\n"); 3093 TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) { 3094 if (vp->v_type != VMARKER) { 3095 vn_printf(vp, "vnode "); 3096 if (db_pager_quit) 3097 break; 3098 } 3099 } 3100 db_printf("\n\nList of inactive vnodes\n"); 3101 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 3102 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) { 3103 vn_printf(vp, "vnode "); 3104 if (db_pager_quit) 3105 break; 3106 } 3107 } 3108 } 3109 #endif /* DDB */ 3110 3111 /* 3112 * Fill in a struct xvfsconf based on a struct vfsconf. 3113 */ 3114 static void 3115 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp) 3116 { 3117 3118 strcpy(xvfsp->vfc_name, vfsp->vfc_name); 3119 xvfsp->vfc_typenum = vfsp->vfc_typenum; 3120 xvfsp->vfc_refcount = vfsp->vfc_refcount; 3121 xvfsp->vfc_flags = vfsp->vfc_flags; 3122 /* 3123 * These are unused in userland, we keep them 3124 * to not break binary compatibility. 3125 */ 3126 xvfsp->vfc_vfsops = NULL; 3127 xvfsp->vfc_next = NULL; 3128 } 3129 3130 /* 3131 * Top level filesystem related information gathering. 3132 */ 3133 static int 3134 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) 3135 { 3136 struct vfsconf *vfsp; 3137 struct xvfsconf xvfsp; 3138 int error; 3139 3140 error = 0; 3141 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 3142 bzero(&xvfsp, sizeof(xvfsp)); 3143 vfsconf2x(vfsp, &xvfsp); 3144 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp); 3145 if (error) 3146 break; 3147 } 3148 return (error); 3149 } 3150 3151 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD, 3152 NULL, 0, sysctl_vfs_conflist, 3153 "S,xvfsconf", "List of all configured filesystems"); 3154 3155 #ifndef BURN_BRIDGES 3156 static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); 3157 3158 static int 3159 vfs_sysctl(SYSCTL_HANDLER_ARGS) 3160 { 3161 int *name = (int *)arg1 - 1; /* XXX */ 3162 u_int namelen = arg2 + 1; /* XXX */ 3163 struct vfsconf *vfsp; 3164 struct xvfsconf xvfsp; 3165 3166 log(LOG_WARNING, "userland calling deprecated sysctl, " 3167 "please rebuild world\n"); 3168 3169 #if 1 || defined(COMPAT_PRELITE2) 3170 /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ 3171 if (namelen == 1) 3172 return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); 3173 #endif 3174 3175 switch (name[1]) { 3176 case VFS_MAXTYPENUM: 3177 if (namelen != 2) 3178 return (ENOTDIR); 3179 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); 3180 case VFS_CONF: 3181 if (namelen != 3) 3182 return (ENOTDIR); /* overloaded */ 3183 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) 3184 if (vfsp->vfc_typenum == name[2]) 3185 break; 3186 if (vfsp == NULL) 3187 return (EOPNOTSUPP); 3188 bzero(&xvfsp, sizeof(xvfsp)); 3189 vfsconf2x(vfsp, &xvfsp); 3190 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); 3191 } 3192 return (EOPNOTSUPP); 3193 } 3194 3195 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, 3196 vfs_sysctl, "Generic filesystem"); 3197 3198 #if 1 || defined(COMPAT_PRELITE2) 3199 3200 static int 3201 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) 3202 { 3203 int error; 3204 struct vfsconf *vfsp; 3205 struct ovfsconf ovfs; 3206 3207 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { 3208 bzero(&ovfs, sizeof(ovfs)); 3209 ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ 3210 strcpy(ovfs.vfc_name, vfsp->vfc_name); 3211 ovfs.vfc_index = vfsp->vfc_typenum; 3212 ovfs.vfc_refcount = vfsp->vfc_refcount; 3213 ovfs.vfc_flags = vfsp->vfc_flags; 3214 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); 3215 if (error) 3216 return error; 3217 } 3218 return 0; 3219 } 3220 3221 #endif /* 1 || COMPAT_PRELITE2 */ 3222 #endif /* !BURN_BRIDGES */ 3223 3224 #define KINFO_VNODESLOP 10 3225 #ifdef notyet 3226 /* 3227 * Dump vnode list (via sysctl). 3228 */ 3229 /* ARGSUSED */ 3230 static int 3231 sysctl_vnode(SYSCTL_HANDLER_ARGS) 3232 { 3233 struct xvnode *xvn; 3234 struct mount *mp; 3235 struct vnode *vp; 3236 int error, len, n; 3237 3238 /* 3239 * Stale numvnodes access is not fatal here. 3240 */ 3241 req->lock = 0; 3242 len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; 3243 if (!req->oldptr) 3244 /* Make an estimate */ 3245 return (SYSCTL_OUT(req, 0, len)); 3246 3247 error = sysctl_wire_old_buffer(req, 0); 3248 if (error != 0) 3249 return (error); 3250 xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); 3251 n = 0; 3252 mtx_lock(&mountlist_mtx); 3253 TAILQ_FOREACH(mp, &mountlist, mnt_list) { 3254 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) 3255 continue; 3256 MNT_ILOCK(mp); 3257 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { 3258 if (n == len) 3259 break; 3260 vref(vp); 3261 xvn[n].xv_size = sizeof *xvn; 3262 xvn[n].xv_vnode = vp; 3263 xvn[n].xv_id = 0; /* XXX compat */ 3264 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field 3265 XV_COPY(usecount); 3266 XV_COPY(writecount); 3267 XV_COPY(holdcnt); 3268 XV_COPY(mount); 3269 XV_COPY(numoutput); 3270 XV_COPY(type); 3271 #undef XV_COPY 3272 xvn[n].xv_flag = vp->v_vflag; 3273 3274 switch (vp->v_type) { 3275 case VREG: 3276 case VDIR: 3277 case VLNK: 3278 break; 3279 case VBLK: 3280 case VCHR: 3281 if (vp->v_rdev == NULL) { 3282 vrele(vp); 3283 continue; 3284 } 3285 xvn[n].xv_dev = dev2udev(vp->v_rdev); 3286 break; 3287 case VSOCK: 3288 xvn[n].xv_socket = vp->v_socket; 3289 break; 3290 case VFIFO: 3291 xvn[n].xv_fifo = vp->v_fifoinfo; 3292 break; 3293 case VNON: 3294 case VBAD: 3295 default: 3296 /* shouldn't happen? */ 3297 vrele(vp); 3298 continue; 3299 } 3300 vrele(vp); 3301 ++n; 3302 } 3303 MNT_IUNLOCK(mp); 3304 mtx_lock(&mountlist_mtx); 3305 vfs_unbusy(mp); 3306 if (n == len) 3307 break; 3308 } 3309 mtx_unlock(&mountlist_mtx); 3310 3311 error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); 3312 free(xvn, M_TEMP); 3313 return (error); 3314 } 3315 3316 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD, 3317 0, 0, sysctl_vnode, "S,xvnode", ""); 3318 #endif 3319 3320 /* 3321 * Unmount all filesystems. The list is traversed in reverse order 3322 * of mounting to avoid dependencies. 3323 */ 3324 void 3325 vfs_unmountall(void) 3326 { 3327 struct mount *mp; 3328 struct thread *td; 3329 int error; 3330 3331 KASSERT(curthread != NULL, ("vfs_unmountall: NULL curthread")); 3332 CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); 3333 td = curthread; 3334 3335 /* 3336 * Since this only runs when rebooting, it is not interlocked. 3337 */ 3338 while(!TAILQ_EMPTY(&mountlist)) { 3339 mp = TAILQ_LAST(&mountlist, mntlist); 3340 error = dounmount(mp, MNT_FORCE, td); 3341 if (error) { 3342 TAILQ_REMOVE(&mountlist, mp, mnt_list); 3343 /* 3344 * XXX: Due to the way in which we mount the root 3345 * file system off of devfs, devfs will generate a 3346 * "busy" warning when we try to unmount it before 3347 * the root. Don't print a warning as a result in 3348 * order to avoid false positive errors that may 3349 * cause needless upset. 3350 */ 3351 if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) { 3352 printf("unmount of %s failed (", 3353 mp->mnt_stat.f_mntonname); 3354 if (error == EBUSY) 3355 printf("BUSY)\n"); 3356 else 3357 printf("%d)\n", error); 3358 } 3359 } else { 3360 /* The unmount has removed mp from the mountlist */ 3361 } 3362 } 3363 } 3364 3365 /* 3366 * perform msync on all vnodes under a mount point 3367 * the mount point must be locked. 3368 */ 3369 void 3370 vfs_msync(struct mount *mp, int flags) 3371 { 3372 struct vnode *vp, *mvp; 3373 struct vm_object *obj; 3374 3375 CTR2(KTR_VFS, "%s: mp %p", __func__, mp); 3376 MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) { 3377 obj = vp->v_object; 3378 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 && 3379 (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) { 3380 if (!vget(vp, 3381 LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, 3382 curthread)) { 3383 if (vp->v_vflag & VV_NOSYNC) { /* unlinked */ 3384 vput(vp); 3385 continue; 3386 } 3387 3388 obj = vp->v_object; 3389 if (obj != NULL) { 3390 VM_OBJECT_LOCK(obj); 3391 vm_object_page_clean(obj, 0, 0, 3392 flags == MNT_WAIT ? 3393 OBJPC_SYNC : OBJPC_NOSYNC); 3394 VM_OBJECT_UNLOCK(obj); 3395 } 3396 vput(vp); 3397 } 3398 } else 3399 VI_UNLOCK(vp); 3400 } 3401 } 3402 3403 static void 3404 destroy_vpollinfo(struct vpollinfo *vi) 3405 { 3406 seldrain(&vi->vpi_selinfo); 3407 knlist_destroy(&vi->vpi_selinfo.si_note); 3408 mtx_destroy(&vi->vpi_lock); 3409 uma_zfree(vnodepoll_zone, vi); 3410 } 3411 3412 /* 3413 * Initalize per-vnode helper structure to hold poll-related state. 3414 */ 3415 void 3416 v_addpollinfo(struct vnode *vp) 3417 { 3418 struct vpollinfo *vi; 3419 3420 if (vp->v_pollinfo != NULL) 3421 return; 3422 vi = uma_zalloc(vnodepoll_zone, M_WAITOK); 3423 mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); 3424 knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, 3425 vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked); 3426 VI_LOCK(vp); 3427 if (vp->v_pollinfo != NULL) { 3428 VI_UNLOCK(vp); 3429 destroy_vpollinfo(vi); 3430 return; 3431 } 3432 vp->v_pollinfo = vi; 3433 VI_UNLOCK(vp); 3434 } 3435 3436 /* 3437 * Record a process's interest in events which might happen to 3438 * a vnode. Because poll uses the historic select-style interface 3439 * internally, this routine serves as both the ``check for any 3440 * pending events'' and the ``record my interest in future events'' 3441 * functions. (These are done together, while the lock is held, 3442 * to avoid race conditions.) 3443 */ 3444 int 3445 vn_pollrecord(struct vnode *vp, struct thread *td, int events) 3446 { 3447 3448 v_addpollinfo(vp); 3449 mtx_lock(&vp->v_pollinfo->vpi_lock); 3450 if (vp->v_pollinfo->vpi_revents & events) { 3451 /* 3452 * This leaves events we are not interested 3453 * in available for the other process which 3454 * which presumably had requested them 3455 * (otherwise they would never have been 3456 * recorded). 3457 */ 3458 events &= vp->v_pollinfo->vpi_revents; 3459 vp->v_pollinfo->vpi_revents &= ~events; 3460 3461 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3462 return (events); 3463 } 3464 vp->v_pollinfo->vpi_events |= events; 3465 selrecord(td, &vp->v_pollinfo->vpi_selinfo); 3466 mtx_unlock(&vp->v_pollinfo->vpi_lock); 3467 return (0); 3468 } 3469 3470 /* 3471 * Routine to create and manage a filesystem syncer vnode. 3472 */ 3473 #define sync_close ((int (*)(struct vop_close_args *))nullop) 3474 static int sync_fsync(struct vop_fsync_args *); 3475 static int sync_inactive(struct vop_inactive_args *); 3476 static int sync_reclaim(struct vop_reclaim_args *); 3477 3478 static struct vop_vector sync_vnodeops = { 3479 .vop_bypass = VOP_EOPNOTSUPP, 3480 .vop_close = sync_close, /* close */ 3481 .vop_fsync = sync_fsync, /* fsync */ 3482 .vop_inactive = sync_inactive, /* inactive */ 3483 .vop_reclaim = sync_reclaim, /* reclaim */ 3484 .vop_lock1 = vop_stdlock, /* lock */ 3485 .vop_unlock = vop_stdunlock, /* unlock */ 3486 .vop_islocked = vop_stdislocked, /* islocked */ 3487 }; 3488 3489 /* 3490 * Create a new filesystem syncer vnode for the specified mount point. 3491 */ 3492 void 3493 vfs_allocate_syncvnode(struct mount *mp) 3494 { 3495 struct vnode *vp; 3496 struct bufobj *bo; 3497 static long start, incr, next; 3498 int error; 3499 3500 /* Allocate a new vnode */ 3501 error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); 3502 if (error != 0) 3503 panic("vfs_allocate_syncvnode: getnewvnode() failed"); 3504 vp->v_type = VNON; 3505 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3506 vp->v_vflag |= VV_FORCEINSMQ; 3507 error = insmntque(vp, mp); 3508 if (error != 0) 3509 panic("vfs_allocate_syncvnode: insmntque() failed"); 3510 vp->v_vflag &= ~VV_FORCEINSMQ; 3511 VOP_UNLOCK(vp, 0); 3512 /* 3513 * Place the vnode onto the syncer worklist. We attempt to 3514 * scatter them about on the list so that they will go off 3515 * at evenly distributed times even if all the filesystems 3516 * are mounted at once. 3517 */ 3518 next += incr; 3519 if (next == 0 || next > syncer_maxdelay) { 3520 start /= 2; 3521 incr /= 2; 3522 if (start == 0) { 3523 start = syncer_maxdelay / 2; 3524 incr = syncer_maxdelay; 3525 } 3526 next = start; 3527 } 3528 bo = &vp->v_bufobj; 3529 BO_LOCK(bo); 3530 vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); 3531 /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ 3532 mtx_lock(&sync_mtx); 3533 sync_vnode_count++; 3534 if (mp->mnt_syncer == NULL) { 3535 mp->mnt_syncer = vp; 3536 vp = NULL; 3537 } 3538 mtx_unlock(&sync_mtx); 3539 BO_UNLOCK(bo); 3540 if (vp != NULL) { 3541 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 3542 vgone(vp); 3543 vput(vp); 3544 } 3545 } 3546 3547 void 3548 vfs_deallocate_syncvnode(struct mount *mp) 3549 { 3550 struct vnode *vp; 3551 3552 mtx_lock(&sync_mtx); 3553 vp = mp->mnt_syncer; 3554 if (vp != NULL) 3555 mp->mnt_syncer = NULL; 3556 mtx_unlock(&sync_mtx); 3557 if (vp != NULL) 3558 vrele(vp); 3559 } 3560 3561 /* 3562 * Do a lazy sync of the filesystem. 3563 */ 3564 static int 3565 sync_fsync(struct vop_fsync_args *ap) 3566 { 3567 struct vnode *syncvp = ap->a_vp; 3568 struct mount *mp = syncvp->v_mount; 3569 int error, save; 3570 struct bufobj *bo; 3571 3572 /* 3573 * We only need to do something if this is a lazy evaluation. 3574 */ 3575 if (ap->a_waitfor != MNT_LAZY) 3576 return (0); 3577 3578 /* 3579 * Move ourselves to the back of the sync list. 3580 */ 3581 bo = &syncvp->v_bufobj; 3582 BO_LOCK(bo); 3583 vn_syncer_add_to_worklist(bo, syncdelay); 3584 BO_UNLOCK(bo); 3585 3586 /* 3587 * Walk the list of vnodes pushing all that are dirty and 3588 * not already on the sync list. 3589 */ 3590 mtx_lock(&mountlist_mtx); 3591 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) { 3592 mtx_unlock(&mountlist_mtx); 3593 return (0); 3594 } 3595 if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { 3596 vfs_unbusy(mp); 3597 return (0); 3598 } 3599 save = curthread_pflags_set(TDP_SYNCIO); 3600 vfs_msync(mp, MNT_NOWAIT); 3601 error = VFS_SYNC(mp, MNT_LAZY); 3602 curthread_pflags_restore(save); 3603 vn_finished_write(mp); 3604 vfs_unbusy(mp); 3605 return (error); 3606 } 3607 3608 /* 3609 * The syncer vnode is no referenced. 3610 */ 3611 static int 3612 sync_inactive(struct vop_inactive_args *ap) 3613 { 3614 3615 vgone(ap->a_vp); 3616 return (0); 3617 } 3618 3619 /* 3620 * The syncer vnode is no longer needed and is being decommissioned. 3621 * 3622 * Modifications to the worklist must be protected by sync_mtx. 3623 */ 3624 static int 3625 sync_reclaim(struct vop_reclaim_args *ap) 3626 { 3627 struct vnode *vp = ap->a_vp; 3628 struct bufobj *bo; 3629 3630 bo = &vp->v_bufobj; 3631 BO_LOCK(bo); 3632 mtx_lock(&sync_mtx); 3633 if (vp->v_mount->mnt_syncer == vp) 3634 vp->v_mount->mnt_syncer = NULL; 3635 if (bo->bo_flag & BO_ONWORKLST) { 3636 LIST_REMOVE(bo, bo_synclist); 3637 syncer_worklist_len--; 3638 sync_vnode_count--; 3639 bo->bo_flag &= ~BO_ONWORKLST; 3640 } 3641 mtx_unlock(&sync_mtx); 3642 BO_UNLOCK(bo); 3643 3644 return (0); 3645 } 3646 3647 /* 3648 * Check if vnode represents a disk device 3649 */ 3650 int 3651 vn_isdisk(struct vnode *vp, int *errp) 3652 { 3653 int error; 3654 3655 error = 0; 3656 dev_lock(); 3657 if (vp->v_type != VCHR) 3658 error = ENOTBLK; 3659 else if (vp->v_rdev == NULL) 3660 error = ENXIO; 3661 else if (vp->v_rdev->si_devsw == NULL) 3662 error = ENXIO; 3663 else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) 3664 error = ENOTBLK; 3665 dev_unlock(); 3666 if (errp != NULL) 3667 *errp = error; 3668 return (error == 0); 3669 } 3670 3671 /* 3672 * Common filesystem object access control check routine. Accepts a 3673 * vnode's type, "mode", uid and gid, requested access mode, credentials, 3674 * and optional call-by-reference privused argument allowing vaccess() 3675 * to indicate to the caller whether privilege was used to satisfy the 3676 * request (obsoleted). Returns 0 on success, or an errno on failure. 3677 */ 3678 int 3679 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, 3680 accmode_t accmode, struct ucred *cred, int *privused) 3681 { 3682 accmode_t dac_granted; 3683 accmode_t priv_granted; 3684 3685 KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, 3686 ("invalid bit in accmode")); 3687 KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), 3688 ("VAPPEND without VWRITE")); 3689 3690 /* 3691 * Look for a normal, non-privileged way to access the file/directory 3692 * as requested. If it exists, go with that. 3693 */ 3694 3695 if (privused != NULL) 3696 *privused = 0; 3697 3698 dac_granted = 0; 3699 3700 /* Check the owner. */ 3701 if (cred->cr_uid == file_uid) { 3702 dac_granted |= VADMIN; 3703 if (file_mode & S_IXUSR) 3704 dac_granted |= VEXEC; 3705 if (file_mode & S_IRUSR) 3706 dac_granted |= VREAD; 3707 if (file_mode & S_IWUSR) 3708 dac_granted |= (VWRITE | VAPPEND); 3709 3710 if ((accmode & dac_granted) == accmode) 3711 return (0); 3712 3713 goto privcheck; 3714 } 3715 3716 /* Otherwise, check the groups (first match) */ 3717 if (groupmember(file_gid, cred)) { 3718 if (file_mode & S_IXGRP) 3719 dac_granted |= VEXEC; 3720 if (file_mode & S_IRGRP) 3721 dac_granted |= VREAD; 3722 if (file_mode & S_IWGRP) 3723 dac_granted |= (VWRITE | VAPPEND); 3724 3725 if ((accmode & dac_granted) == accmode) 3726 return (0); 3727 3728 goto privcheck; 3729 } 3730 3731 /* Otherwise, check everyone else. */ 3732 if (file_mode & S_IXOTH) 3733 dac_granted |= VEXEC; 3734 if (file_mode & S_IROTH) 3735 dac_granted |= VREAD; 3736 if (file_mode & S_IWOTH) 3737 dac_granted |= (VWRITE | VAPPEND); 3738 if ((accmode & dac_granted) == accmode) 3739 return (0); 3740 3741 privcheck: 3742 /* 3743 * Build a privilege mask to determine if the set of privileges 3744 * satisfies the requirements when combined with the granted mask 3745 * from above. For each privilege, if the privilege is required, 3746 * bitwise or the request type onto the priv_granted mask. 3747 */ 3748 priv_granted = 0; 3749 3750 if (type == VDIR) { 3751 /* 3752 * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC 3753 * requests, instead of PRIV_VFS_EXEC. 3754 */ 3755 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3756 !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0)) 3757 priv_granted |= VEXEC; 3758 } else { 3759 /* 3760 * Ensure that at least one execute bit is on. Otherwise, 3761 * a privileged user will always succeed, and we don't want 3762 * this to happen unless the file really is executable. 3763 */ 3764 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && 3765 (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && 3766 !priv_check_cred(cred, PRIV_VFS_EXEC, 0)) 3767 priv_granted |= VEXEC; 3768 } 3769 3770 if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && 3771 !priv_check_cred(cred, PRIV_VFS_READ, 0)) 3772 priv_granted |= VREAD; 3773 3774 if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && 3775 !priv_check_cred(cred, PRIV_VFS_WRITE, 0)) 3776 priv_granted |= (VWRITE | VAPPEND); 3777 3778 if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && 3779 !priv_check_cred(cred, PRIV_VFS_ADMIN, 0)) 3780 priv_granted |= VADMIN; 3781 3782 if ((accmode & (priv_granted | dac_granted)) == accmode) { 3783 /* XXX audit: privilege used */ 3784 if (privused != NULL) 3785 *privused = 1; 3786 return (0); 3787 } 3788 3789 return ((accmode & VADMIN) ? EPERM : EACCES); 3790 } 3791 3792 /* 3793 * Credential check based on process requesting service, and per-attribute 3794 * permissions. 3795 */ 3796 int 3797 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, 3798 struct thread *td, accmode_t accmode) 3799 { 3800 3801 /* 3802 * Kernel-invoked always succeeds. 3803 */ 3804 if (cred == NOCRED) 3805 return (0); 3806 3807 /* 3808 * Do not allow privileged processes in jail to directly manipulate 3809 * system attributes. 3810 */ 3811 switch (attrnamespace) { 3812 case EXTATTR_NAMESPACE_SYSTEM: 3813 /* Potentially should be: return (EPERM); */ 3814 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0)); 3815 case EXTATTR_NAMESPACE_USER: 3816 return (VOP_ACCESS(vp, accmode, cred, td)); 3817 default: 3818 return (EPERM); 3819 } 3820 } 3821 3822 #ifdef DEBUG_VFS_LOCKS 3823 /* 3824 * This only exists to supress warnings from unlocked specfs accesses. It is 3825 * no longer ok to have an unlocked VFS. 3826 */ 3827 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL || \ 3828 (vp)->v_type == VCHR || (vp)->v_type == VBAD) 3829 3830 int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ 3831 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, 3832 "Drop into debugger on lock violation"); 3833 3834 int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ 3835 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 3836 0, "Check for interlock across VOPs"); 3837 3838 int vfs_badlock_print = 1; /* Print lock violations. */ 3839 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 3840 0, "Print lock violations"); 3841 3842 #ifdef KDB 3843 int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ 3844 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, 3845 &vfs_badlock_backtrace, 0, "Print backtrace at lock violations"); 3846 #endif 3847 3848 static void 3849 vfs_badlock(const char *msg, const char *str, struct vnode *vp) 3850 { 3851 3852 #ifdef KDB 3853 if (vfs_badlock_backtrace) 3854 kdb_backtrace(); 3855 #endif 3856 if (vfs_badlock_print) 3857 printf("%s: %p %s\n", str, (void *)vp, msg); 3858 if (vfs_badlock_ddb) 3859 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3860 } 3861 3862 void 3863 assert_vi_locked(struct vnode *vp, const char *str) 3864 { 3865 3866 if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) 3867 vfs_badlock("interlock is not locked but should be", str, vp); 3868 } 3869 3870 void 3871 assert_vi_unlocked(struct vnode *vp, const char *str) 3872 { 3873 3874 if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) 3875 vfs_badlock("interlock is locked but should not be", str, vp); 3876 } 3877 3878 void 3879 assert_vop_locked(struct vnode *vp, const char *str) 3880 { 3881 3882 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == 0) 3883 vfs_badlock("is not locked but should be", str, vp); 3884 } 3885 3886 void 3887 assert_vop_unlocked(struct vnode *vp, const char *str) 3888 { 3889 3890 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) 3891 vfs_badlock("is locked but should not be", str, vp); 3892 } 3893 3894 void 3895 assert_vop_elocked(struct vnode *vp, const char *str) 3896 { 3897 3898 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) 3899 vfs_badlock("is not exclusive locked but should be", str, vp); 3900 } 3901 3902 #if 0 3903 void 3904 assert_vop_elocked_other(struct vnode *vp, const char *str) 3905 { 3906 3907 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER) 3908 vfs_badlock("is not exclusive locked by another thread", 3909 str, vp); 3910 } 3911 3912 void 3913 assert_vop_slocked(struct vnode *vp, const char *str) 3914 { 3915 3916 if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED) 3917 vfs_badlock("is not locked shared but should be", str, vp); 3918 } 3919 #endif /* 0 */ 3920 #endif /* DEBUG_VFS_LOCKS */ 3921 3922 void 3923 vop_rename_fail(struct vop_rename_args *ap) 3924 { 3925 3926 if (ap->a_tvp != NULL) 3927 vput(ap->a_tvp); 3928 if (ap->a_tdvp == ap->a_tvp) 3929 vrele(ap->a_tdvp); 3930 else 3931 vput(ap->a_tdvp); 3932 vrele(ap->a_fdvp); 3933 vrele(ap->a_fvp); 3934 } 3935 3936 void 3937 vop_rename_pre(void *ap) 3938 { 3939 struct vop_rename_args *a = ap; 3940 3941 #ifdef DEBUG_VFS_LOCKS 3942 if (a->a_tvp) 3943 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); 3944 ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); 3945 ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); 3946 ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); 3947 3948 /* Check the source (from). */ 3949 if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && 3950 (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) 3951 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); 3952 if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) 3953 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); 3954 3955 /* Check the target. */ 3956 if (a->a_tvp) 3957 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); 3958 ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); 3959 #endif 3960 if (a->a_tdvp != a->a_fdvp) 3961 vhold(a->a_fdvp); 3962 if (a->a_tvp != a->a_fvp) 3963 vhold(a->a_fvp); 3964 vhold(a->a_tdvp); 3965 if (a->a_tvp) 3966 vhold(a->a_tvp); 3967 } 3968 3969 void 3970 vop_strategy_pre(void *ap) 3971 { 3972 #ifdef DEBUG_VFS_LOCKS 3973 struct vop_strategy_args *a; 3974 struct buf *bp; 3975 3976 a = ap; 3977 bp = a->a_bp; 3978 3979 /* 3980 * Cluster ops lock their component buffers but not the IO container. 3981 */ 3982 if ((bp->b_flags & B_CLUSTER) != 0) 3983 return; 3984 3985 if (panicstr == NULL && !BUF_ISLOCKED(bp)) { 3986 if (vfs_badlock_print) 3987 printf( 3988 "VOP_STRATEGY: bp is not locked but should be\n"); 3989 if (vfs_badlock_ddb) 3990 kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); 3991 } 3992 #endif 3993 } 3994 3995 void 3996 vop_lookup_pre(void *ap) 3997 { 3998 #ifdef DEBUG_VFS_LOCKS 3999 struct vop_lookup_args *a; 4000 struct vnode *dvp; 4001 4002 a = ap; 4003 dvp = a->a_dvp; 4004 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 4005 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 4006 #endif 4007 } 4008 4009 void 4010 vop_lookup_post(void *ap, int rc) 4011 { 4012 #ifdef DEBUG_VFS_LOCKS 4013 struct vop_lookup_args *a; 4014 struct vnode *dvp; 4015 struct vnode *vp; 4016 4017 a = ap; 4018 dvp = a->a_dvp; 4019 vp = *(a->a_vpp); 4020 4021 ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP"); 4022 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP"); 4023 4024 if (!rc) 4025 ASSERT_VOP_LOCKED(vp, "VOP_LOOKUP (child)"); 4026 #endif 4027 } 4028 4029 void 4030 vop_lock_pre(void *ap) 4031 { 4032 #ifdef DEBUG_VFS_LOCKS 4033 struct vop_lock1_args *a = ap; 4034 4035 if ((a->a_flags & LK_INTERLOCK) == 0) 4036 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 4037 else 4038 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); 4039 #endif 4040 } 4041 4042 void 4043 vop_lock_post(void *ap, int rc) 4044 { 4045 #ifdef DEBUG_VFS_LOCKS 4046 struct vop_lock1_args *a = ap; 4047 4048 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); 4049 if (rc == 0) 4050 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); 4051 #endif 4052 } 4053 4054 void 4055 vop_unlock_pre(void *ap) 4056 { 4057 #ifdef DEBUG_VFS_LOCKS 4058 struct vop_unlock_args *a = ap; 4059 4060 if (a->a_flags & LK_INTERLOCK) 4061 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK"); 4062 ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); 4063 #endif 4064 } 4065 4066 void 4067 vop_unlock_post(void *ap, int rc) 4068 { 4069 #ifdef DEBUG_VFS_LOCKS 4070 struct vop_unlock_args *a = ap; 4071 4072 if (a->a_flags & LK_INTERLOCK) 4073 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK"); 4074 #endif 4075 } 4076 4077 void 4078 vop_create_post(void *ap, int rc) 4079 { 4080 struct vop_create_args *a = ap; 4081 4082 if (!rc) 4083 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4084 } 4085 4086 void 4087 vop_deleteextattr_post(void *ap, int rc) 4088 { 4089 struct vop_deleteextattr_args *a = ap; 4090 4091 if (!rc) 4092 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 4093 } 4094 4095 void 4096 vop_link_post(void *ap, int rc) 4097 { 4098 struct vop_link_args *a = ap; 4099 4100 if (!rc) { 4101 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); 4102 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); 4103 } 4104 } 4105 4106 void 4107 vop_mkdir_post(void *ap, int rc) 4108 { 4109 struct vop_mkdir_args *a = ap; 4110 4111 if (!rc) 4112 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 4113 } 4114 4115 void 4116 vop_mknod_post(void *ap, int rc) 4117 { 4118 struct vop_mknod_args *a = ap; 4119 4120 if (!rc) 4121 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4122 } 4123 4124 void 4125 vop_remove_post(void *ap, int rc) 4126 { 4127 struct vop_remove_args *a = ap; 4128 4129 if (!rc) { 4130 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4131 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 4132 } 4133 } 4134 4135 void 4136 vop_rename_post(void *ap, int rc) 4137 { 4138 struct vop_rename_args *a = ap; 4139 4140 if (!rc) { 4141 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE); 4142 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE); 4143 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); 4144 if (a->a_tvp) 4145 VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); 4146 } 4147 if (a->a_tdvp != a->a_fdvp) 4148 vdrop(a->a_fdvp); 4149 if (a->a_tvp != a->a_fvp) 4150 vdrop(a->a_fvp); 4151 vdrop(a->a_tdvp); 4152 if (a->a_tvp) 4153 vdrop(a->a_tvp); 4154 } 4155 4156 void 4157 vop_rmdir_post(void *ap, int rc) 4158 { 4159 struct vop_rmdir_args *a = ap; 4160 4161 if (!rc) { 4162 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); 4163 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); 4164 } 4165 } 4166 4167 void 4168 vop_setattr_post(void *ap, int rc) 4169 { 4170 struct vop_setattr_args *a = ap; 4171 4172 if (!rc) 4173 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 4174 } 4175 4176 void 4177 vop_setextattr_post(void *ap, int rc) 4178 { 4179 struct vop_setextattr_args *a = ap; 4180 4181 if (!rc) 4182 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); 4183 } 4184 4185 void 4186 vop_symlink_post(void *ap, int rc) 4187 { 4188 struct vop_symlink_args *a = ap; 4189 4190 if (!rc) 4191 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); 4192 } 4193 4194 static struct knlist fs_knlist; 4195 4196 static void 4197 vfs_event_init(void *arg) 4198 { 4199 knlist_init_mtx(&fs_knlist, NULL); 4200 } 4201 /* XXX - correct order? */ 4202 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); 4203 4204 void 4205 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) 4206 { 4207 4208 KNOTE_UNLOCKED(&fs_knlist, event); 4209 } 4210 4211 static int filt_fsattach(struct knote *kn); 4212 static void filt_fsdetach(struct knote *kn); 4213 static int filt_fsevent(struct knote *kn, long hint); 4214 4215 struct filterops fs_filtops = { 4216 .f_isfd = 0, 4217 .f_attach = filt_fsattach, 4218 .f_detach = filt_fsdetach, 4219 .f_event = filt_fsevent 4220 }; 4221 4222 static int 4223 filt_fsattach(struct knote *kn) 4224 { 4225 4226 kn->kn_flags |= EV_CLEAR; 4227 knlist_add(&fs_knlist, kn, 0); 4228 return (0); 4229 } 4230 4231 static void 4232 filt_fsdetach(struct knote *kn) 4233 { 4234 4235 knlist_remove(&fs_knlist, kn, 0); 4236 } 4237 4238 static int 4239 filt_fsevent(struct knote *kn, long hint) 4240 { 4241 4242 kn->kn_fflags |= hint; 4243 return (kn->kn_fflags != 0); 4244 } 4245 4246 static int 4247 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) 4248 { 4249 struct vfsidctl vc; 4250 int error; 4251 struct mount *mp; 4252 4253 error = SYSCTL_IN(req, &vc, sizeof(vc)); 4254 if (error) 4255 return (error); 4256 if (vc.vc_vers != VFS_CTL_VERS1) 4257 return (EINVAL); 4258 mp = vfs_getvfs(&vc.vc_fsid); 4259 if (mp == NULL) 4260 return (ENOENT); 4261 /* ensure that a specific sysctl goes to the right filesystem. */ 4262 if (strcmp(vc.vc_fstypename, "*") != 0 && 4263 strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { 4264 vfs_rel(mp); 4265 return (EINVAL); 4266 } 4267 VCTLTOREQ(&vc, req); 4268 error = VFS_SYSCTL(mp, vc.vc_op, req); 4269 vfs_rel(mp); 4270 return (error); 4271 } 4272 4273 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR, 4274 NULL, 0, sysctl_vfs_ctl, "", 4275 "Sysctl by fsid"); 4276 4277 /* 4278 * Function to initialize a va_filerev field sensibly. 4279 * XXX: Wouldn't a random number make a lot more sense ?? 4280 */ 4281 u_quad_t 4282 init_va_filerev(void) 4283 { 4284 struct bintime bt; 4285 4286 getbinuptime(&bt); 4287 return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); 4288 } 4289 4290 static int filt_vfsread(struct knote *kn, long hint); 4291 static int filt_vfswrite(struct knote *kn, long hint); 4292 static int filt_vfsvnode(struct knote *kn, long hint); 4293 static void filt_vfsdetach(struct knote *kn); 4294 static struct filterops vfsread_filtops = { 4295 .f_isfd = 1, 4296 .f_detach = filt_vfsdetach, 4297 .f_event = filt_vfsread 4298 }; 4299 static struct filterops vfswrite_filtops = { 4300 .f_isfd = 1, 4301 .f_detach = filt_vfsdetach, 4302 .f_event = filt_vfswrite 4303 }; 4304 static struct filterops vfsvnode_filtops = { 4305 .f_isfd = 1, 4306 .f_detach = filt_vfsdetach, 4307 .f_event = filt_vfsvnode 4308 }; 4309 4310 static void 4311 vfs_knllock(void *arg) 4312 { 4313 struct vnode *vp = arg; 4314 4315 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 4316 } 4317 4318 static void 4319 vfs_knlunlock(void *arg) 4320 { 4321 struct vnode *vp = arg; 4322 4323 VOP_UNLOCK(vp, 0); 4324 } 4325 4326 static void 4327 vfs_knl_assert_locked(void *arg) 4328 { 4329 #ifdef DEBUG_VFS_LOCKS 4330 struct vnode *vp = arg; 4331 4332 ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); 4333 #endif 4334 } 4335 4336 static void 4337 vfs_knl_assert_unlocked(void *arg) 4338 { 4339 #ifdef DEBUG_VFS_LOCKS 4340 struct vnode *vp = arg; 4341 4342 ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); 4343 #endif 4344 } 4345 4346 int 4347 vfs_kqfilter(struct vop_kqfilter_args *ap) 4348 { 4349 struct vnode *vp = ap->a_vp; 4350 struct knote *kn = ap->a_kn; 4351 struct knlist *knl; 4352 4353 switch (kn->kn_filter) { 4354 case EVFILT_READ: 4355 kn->kn_fop = &vfsread_filtops; 4356 break; 4357 case EVFILT_WRITE: 4358 kn->kn_fop = &vfswrite_filtops; 4359 break; 4360 case EVFILT_VNODE: 4361 kn->kn_fop = &vfsvnode_filtops; 4362 break; 4363 default: 4364 return (EINVAL); 4365 } 4366 4367 kn->kn_hook = (caddr_t)vp; 4368 4369 v_addpollinfo(vp); 4370 if (vp->v_pollinfo == NULL) 4371 return (ENOMEM); 4372 knl = &vp->v_pollinfo->vpi_selinfo.si_note; 4373 knlist_add(knl, kn, 0); 4374 4375 return (0); 4376 } 4377 4378 /* 4379 * Detach knote from vnode 4380 */ 4381 static void 4382 filt_vfsdetach(struct knote *kn) 4383 { 4384 struct vnode *vp = (struct vnode *)kn->kn_hook; 4385 4386 KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); 4387 knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); 4388 } 4389 4390 /*ARGSUSED*/ 4391 static int 4392 filt_vfsread(struct knote *kn, long hint) 4393 { 4394 struct vnode *vp = (struct vnode *)kn->kn_hook; 4395 struct vattr va; 4396 int res; 4397 4398 /* 4399 * filesystem is gone, so set the EOF flag and schedule 4400 * the knote for deletion. 4401 */ 4402 if (hint == NOTE_REVOKE) { 4403 VI_LOCK(vp); 4404 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4405 VI_UNLOCK(vp); 4406 return (1); 4407 } 4408 4409 if (VOP_GETATTR(vp, &va, curthread->td_ucred)) 4410 return (0); 4411 4412 VI_LOCK(vp); 4413 kn->kn_data = va.va_size - kn->kn_fp->f_offset; 4414 res = (kn->kn_data != 0); 4415 VI_UNLOCK(vp); 4416 return (res); 4417 } 4418 4419 /*ARGSUSED*/ 4420 static int 4421 filt_vfswrite(struct knote *kn, long hint) 4422 { 4423 struct vnode *vp = (struct vnode *)kn->kn_hook; 4424 4425 VI_LOCK(vp); 4426 4427 /* 4428 * filesystem is gone, so set the EOF flag and schedule 4429 * the knote for deletion. 4430 */ 4431 if (hint == NOTE_REVOKE) 4432 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 4433 4434 kn->kn_data = 0; 4435 VI_UNLOCK(vp); 4436 return (1); 4437 } 4438 4439 static int 4440 filt_vfsvnode(struct knote *kn, long hint) 4441 { 4442 struct vnode *vp = (struct vnode *)kn->kn_hook; 4443 int res; 4444 4445 VI_LOCK(vp); 4446 if (kn->kn_sfflags & hint) 4447 kn->kn_fflags |= hint; 4448 if (hint == NOTE_REVOKE) { 4449 kn->kn_flags |= EV_EOF; 4450 VI_UNLOCK(vp); 4451 return (1); 4452 } 4453 res = (kn->kn_fflags != 0); 4454 VI_UNLOCK(vp); 4455 return (res); 4456 } 4457 4458 int 4459 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) 4460 { 4461 int error; 4462 4463 if (dp->d_reclen > ap->a_uio->uio_resid) 4464 return (ENAMETOOLONG); 4465 error = uiomove(dp, dp->d_reclen, ap->a_uio); 4466 if (error) { 4467 if (ap->a_ncookies != NULL) { 4468 if (ap->a_cookies != NULL) 4469 free(ap->a_cookies, M_TEMP); 4470 ap->a_cookies = NULL; 4471 *ap->a_ncookies = 0; 4472 } 4473 return (error); 4474 } 4475 if (ap->a_ncookies == NULL) 4476 return (0); 4477 4478 KASSERT(ap->a_cookies, 4479 ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); 4480 4481 *ap->a_cookies = realloc(*ap->a_cookies, 4482 (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); 4483 (*ap->a_cookies)[*ap->a_ncookies] = off; 4484 return (0); 4485 } 4486 4487 /* 4488 * Mark for update the access time of the file if the filesystem 4489 * supports VOP_MARKATIME. This functionality is used by execve and 4490 * mmap, so we want to avoid the I/O implied by directly setting 4491 * va_atime for the sake of efficiency. 4492 */ 4493 void 4494 vfs_mark_atime(struct vnode *vp, struct ucred *cred) 4495 { 4496 struct mount *mp; 4497 4498 mp = vp->v_mount; 4499 VFS_ASSERT_GIANT(mp); 4500 ASSERT_VOP_LOCKED(vp, "vfs_mark_atime"); 4501 if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0) 4502 (void)VOP_MARKATIME(vp); 4503 } 4504 4505 /* 4506 * The purpose of this routine is to remove granularity from accmode_t, 4507 * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, 4508 * VADMIN and VAPPEND. 4509 * 4510 * If it returns 0, the caller is supposed to continue with the usual 4511 * access checks using 'accmode' as modified by this routine. If it 4512 * returns nonzero value, the caller is supposed to return that value 4513 * as errno. 4514 * 4515 * Note that after this routine runs, accmode may be zero. 4516 */ 4517 int 4518 vfs_unixify_accmode(accmode_t *accmode) 4519 { 4520 /* 4521 * There is no way to specify explicit "deny" rule using 4522 * file mode or POSIX.1e ACLs. 4523 */ 4524 if (*accmode & VEXPLICIT_DENY) { 4525 *accmode = 0; 4526 return (0); 4527 } 4528 4529 /* 4530 * None of these can be translated into usual access bits. 4531 * Also, the common case for NFSv4 ACLs is to not contain 4532 * either of these bits. Caller should check for VWRITE 4533 * on the containing directory instead. 4534 */ 4535 if (*accmode & (VDELETE_CHILD | VDELETE)) 4536 return (EPERM); 4537 4538 if (*accmode & VADMIN_PERMS) { 4539 *accmode &= ~VADMIN_PERMS; 4540 *accmode |= VADMIN; 4541 } 4542 4543 /* 4544 * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL 4545 * or VSYNCHRONIZE using file mode or POSIX.1e ACL. 4546 */ 4547 *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); 4548 4549 return (0); 4550 } 4551 4552 /* 4553 * These are helper functions for filesystems to traverse all 4554 * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h. 4555 * 4556 * This interface replaces MNT_VNODE_FOREACH. 4557 */ 4558 4559 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); 4560 4561 struct vnode * 4562 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp) 4563 { 4564 struct vnode *vp; 4565 4566 if (should_yield()) 4567 kern_yield(PRI_UNCHANGED); 4568 MNT_ILOCK(mp); 4569 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 4570 vp = TAILQ_NEXT(*mvp, v_nmntvnodes); 4571 while (vp != NULL && (vp->v_type == VMARKER || 4572 (vp->v_iflag & VI_DOOMED) != 0)) 4573 vp = TAILQ_NEXT(vp, v_nmntvnodes); 4574 4575 /* Check if we are done */ 4576 if (vp == NULL) { 4577 __mnt_vnode_markerfree_all(mvp, mp); 4578 /* MNT_IUNLOCK(mp); -- done in above function */ 4579 mtx_assert(MNT_MTX(mp), MA_NOTOWNED); 4580 return (NULL); 4581 } 4582 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 4583 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 4584 VI_LOCK(vp); 4585 MNT_IUNLOCK(mp); 4586 return (vp); 4587 } 4588 4589 struct vnode * 4590 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp) 4591 { 4592 struct vnode *vp; 4593 4594 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); 4595 MNT_ILOCK(mp); 4596 MNT_REF(mp); 4597 (*mvp)->v_type = VMARKER; 4598 4599 vp = TAILQ_FIRST(&mp->mnt_nvnodelist); 4600 while (vp != NULL && (vp->v_type == VMARKER || 4601 (vp->v_iflag & VI_DOOMED) != 0)) 4602 vp = TAILQ_NEXT(vp, v_nmntvnodes); 4603 4604 /* Check if we are done */ 4605 if (vp == NULL) { 4606 MNT_REL(mp); 4607 MNT_IUNLOCK(mp); 4608 free(*mvp, M_VNODE_MARKER); 4609 *mvp = NULL; 4610 return (NULL); 4611 } 4612 (*mvp)->v_mount = mp; 4613 TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); 4614 VI_LOCK(vp); 4615 MNT_IUNLOCK(mp); 4616 return (vp); 4617 } 4618 4619 4620 void 4621 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp) 4622 { 4623 4624 if (*mvp == NULL) { 4625 MNT_IUNLOCK(mp); 4626 return; 4627 } 4628 4629 mtx_assert(MNT_MTX(mp), MA_OWNED); 4630 4631 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 4632 TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); 4633 MNT_REL(mp); 4634 MNT_IUNLOCK(mp); 4635 free(*mvp, M_VNODE_MARKER); 4636 *mvp = NULL; 4637 } 4638 4639 /* 4640 * These are helper functions for filesystems to traverse their 4641 * active vnodes. See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h 4642 */ 4643 struct vnode * 4644 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp) 4645 { 4646 struct vnode *vp, *nvp; 4647 4648 if (should_yield()) 4649 kern_yield(PRI_UNCHANGED); 4650 MNT_ILOCK(mp); 4651 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 4652 vp = TAILQ_NEXT(*mvp, v_actfreelist); 4653 while (vp != NULL) { 4654 VI_LOCK(vp); 4655 if (vp->v_mount == mp && vp->v_type != VMARKER && 4656 (vp->v_iflag & VI_DOOMED) == 0) 4657 break; 4658 nvp = TAILQ_NEXT(vp, v_actfreelist); 4659 VI_UNLOCK(vp); 4660 vp = nvp; 4661 } 4662 4663 /* Check if we are done */ 4664 if (vp == NULL) { 4665 __mnt_vnode_markerfree_active(mvp, mp); 4666 /* MNT_IUNLOCK(mp); -- done in above function */ 4667 mtx_assert(MNT_MTX(mp), MA_NOTOWNED); 4668 return (NULL); 4669 } 4670 mtx_lock(&vnode_free_list_mtx); 4671 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist); 4672 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist); 4673 mtx_unlock(&vnode_free_list_mtx); 4674 MNT_IUNLOCK(mp); 4675 return (vp); 4676 } 4677 4678 struct vnode * 4679 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp) 4680 { 4681 struct vnode *vp, *nvp; 4682 4683 *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); 4684 MNT_ILOCK(mp); 4685 MNT_REF(mp); 4686 (*mvp)->v_type = VMARKER; 4687 4688 vp = TAILQ_NEXT(*mvp, v_actfreelist); 4689 while (vp != NULL) { 4690 VI_LOCK(vp); 4691 if (vp->v_mount == mp && vp->v_type != VMARKER && 4692 (vp->v_iflag & VI_DOOMED) == 0) 4693 break; 4694 nvp = TAILQ_NEXT(vp, v_actfreelist); 4695 VI_UNLOCK(vp); 4696 vp = nvp; 4697 } 4698 4699 /* Check if we are done */ 4700 if (vp == NULL) { 4701 MNT_REL(mp); 4702 MNT_IUNLOCK(mp); 4703 free(*mvp, M_VNODE_MARKER); 4704 *mvp = NULL; 4705 return (NULL); 4706 } 4707 (*mvp)->v_mount = mp; 4708 mtx_lock(&vnode_free_list_mtx); 4709 TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist); 4710 mtx_unlock(&vnode_free_list_mtx); 4711 MNT_IUNLOCK(mp); 4712 return (vp); 4713 } 4714 4715 void 4716 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp) 4717 { 4718 4719 if (*mvp == NULL) { 4720 MNT_IUNLOCK(mp); 4721 return; 4722 } 4723 4724 mtx_assert(MNT_MTX(mp), MA_OWNED); 4725 4726 KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); 4727 mtx_lock(&vnode_free_list_mtx); 4728 TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist); 4729 mtx_unlock(&vnode_free_list_mtx); 4730 MNT_REL(mp); 4731 MNT_IUNLOCK(mp); 4732 free(*mvp, M_VNODE_MARKER); 4733 *mvp = NULL; 4734 } 4735