1 /*- 2 * Copyright (c) 2005, 2006 The NetBSD Foundation, Inc. 3 * All rights reserved. 4 * 5 * This code is derived from software contributed to The NetBSD Foundation 6 * by Julio M. Merino Vidal, developed as part of Google's Summer of Code 7 * 2005 program. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 19 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 20 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 21 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 22 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 23 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 24 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 25 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 26 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 27 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 28 * POSSIBILITY OF SUCH DAMAGE. 29 * 30 * $NetBSD: tmpfs_vnops.c,v 1.39 2007/07/23 15:41:01 jmmv Exp $ 31 */ 32 33 /* 34 * tmpfs vnode interface. 35 */ 36 37 #include <sys/kernel.h> 38 #include <sys/kern_syscall.h> 39 #include <sys/param.h> 40 #include <sys/fcntl.h> 41 #include <sys/lockf.h> 42 #include <sys/priv.h> 43 #include <sys/proc.h> 44 #include <sys/resourcevar.h> 45 #include <sys/sched.h> 46 #include <sys/stat.h> 47 #include <sys/systm.h> 48 #include <sys/unistd.h> 49 #include <sys/vfsops.h> 50 #include <sys/vnode.h> 51 #include <sys/mountctl.h> 52 53 #include <vm/vm.h> 54 #include <vm/vm_extern.h> 55 #include <vm/vm_object.h> 56 #include <vm/vm_page.h> 57 #include <vm/vm_pageout.h> 58 #include <vm/vm_pager.h> 59 #include <vm/swap_pager.h> 60 61 #include <sys/buf2.h> 62 #include <vm/vm_page2.h> 63 64 #include <vfs/fifofs/fifo.h> 65 #include <vfs/tmpfs/tmpfs_vnops.h> 66 #include "tmpfs.h" 67 68 static void tmpfs_strategy_done(struct bio *bio); 69 70 static __inline 71 void 72 tmpfs_knote(struct vnode *vp, int flags) 73 { 74 if (flags) 75 KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags); 76 } 77 78 79 /* --------------------------------------------------------------------- */ 80 81 static int 82 tmpfs_nresolve(struct vop_nresolve_args *v) 83 { 84 struct vnode *dvp = v->a_dvp; 85 struct vnode *vp = NULL; 86 struct namecache *ncp = v->a_nch->ncp; 87 struct tmpfs_node *tnode; 88 struct mount *mp; 89 struct tmpfs_dirent *de; 90 struct tmpfs_node *dnode; 91 int error; 92 93 mp = dvp->v_mount; 94 95 dnode = VP_TO_TMPFS_DIR(dvp); 96 97 TMPFS_NODE_LOCK_SH(dnode); 98 de = tmpfs_dir_lookup(dnode, NULL, ncp); 99 if (de == NULL) { 100 error = ENOENT; 101 } else { 102 /* 103 * Allocate a vnode for the node we found. 104 */ 105 tnode = de->td_node; 106 error = tmpfs_alloc_vp(dvp->v_mount, tnode, 107 LK_EXCLUSIVE | LK_RETRY, &vp); 108 if (error) 109 goto out; 110 KKASSERT(vp); 111 } 112 113 out: 114 TMPFS_NODE_UNLOCK(dnode); 115 116 if ((dnode->tn_status & TMPFS_NODE_ACCESSED) == 0) { 117 TMPFS_NODE_LOCK(dnode); 118 dnode->tn_status |= TMPFS_NODE_ACCESSED; 119 TMPFS_NODE_UNLOCK(dnode); 120 } 121 122 /* 123 * Store the result of this lookup in the cache. Avoid this if the 124 * request was for creation, as it does not improve timings on 125 * emprical tests. 126 */ 127 if (vp) { 128 vn_unlock(vp); 129 cache_setvp(v->a_nch, vp); 130 vrele(vp); 131 } else if (error == ENOENT) { 132 cache_setvp(v->a_nch, NULL); 133 } 134 return (error); 135 } 136 137 static int 138 tmpfs_nlookupdotdot(struct vop_nlookupdotdot_args *v) 139 { 140 struct vnode *dvp = v->a_dvp; 141 struct vnode **vpp = v->a_vpp; 142 struct tmpfs_node *dnode = VP_TO_TMPFS_NODE(dvp); 143 struct ucred *cred = v->a_cred; 144 struct mount *mp; 145 int error; 146 147 *vpp = NULL; 148 149 mp = dvp->v_mount; 150 151 /* Check accessibility of requested node as a first step. */ 152 error = VOP_ACCESS(dvp, VEXEC, cred); 153 if (error != 0) 154 return error; 155 156 if (dnode->tn_dir.tn_parent != NULL) { 157 /* Allocate a new vnode on the matching entry. */ 158 error = tmpfs_alloc_vp(dvp->v_mount, dnode->tn_dir.tn_parent, 159 LK_EXCLUSIVE | LK_RETRY, vpp); 160 161 if (*vpp) 162 vn_unlock(*vpp); 163 } 164 return (*vpp == NULL) ? ENOENT : 0; 165 } 166 167 /* --------------------------------------------------------------------- */ 168 169 static int 170 tmpfs_ncreate(struct vop_ncreate_args *v) 171 { 172 struct vnode *dvp = v->a_dvp; 173 struct vnode **vpp = v->a_vpp; 174 struct namecache *ncp = v->a_nch->ncp; 175 struct vattr *vap = v->a_vap; 176 struct ucred *cred = v->a_cred; 177 struct mount *mp; 178 int error; 179 180 mp = dvp->v_mount; 181 182 KKASSERT(vap->va_type == VREG || vap->va_type == VSOCK); 183 184 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); 185 if (error == 0) { 186 cache_setunresolved(v->a_nch); 187 cache_setvp(v->a_nch, *vpp); 188 tmpfs_knote(dvp, NOTE_WRITE); 189 } 190 return (error); 191 } 192 /* --------------------------------------------------------------------- */ 193 194 static int 195 tmpfs_nmknod(struct vop_nmknod_args *v) 196 { 197 struct vnode *dvp = v->a_dvp; 198 struct vnode **vpp = v->a_vpp; 199 struct namecache *ncp = v->a_nch->ncp; 200 struct vattr *vap = v->a_vap; 201 struct ucred *cred = v->a_cred; 202 int error; 203 204 if (vap->va_type != VBLK && vap->va_type != VCHR && 205 vap->va_type != VFIFO) { 206 return (EINVAL); 207 } 208 209 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); 210 if (error == 0) { 211 cache_setunresolved(v->a_nch); 212 cache_setvp(v->a_nch, *vpp); 213 tmpfs_knote(dvp, NOTE_WRITE); 214 } 215 return error; 216 } 217 218 /* --------------------------------------------------------------------- */ 219 220 static int 221 tmpfs_open(struct vop_open_args *v) 222 { 223 struct vnode *vp = v->a_vp; 224 int mode = v->a_mode; 225 struct tmpfs_node *node; 226 int error; 227 228 node = VP_TO_TMPFS_NODE(vp); 229 230 #if 0 231 /* The file is still active but all its names have been removed 232 * (e.g. by a "rmdir $(pwd)"). It cannot be opened any more as 233 * it is about to die. */ 234 if (node->tn_links < 1) 235 return (ENOENT); 236 #endif 237 238 /* If the file is marked append-only, deny write requests. */ 239 if ((node->tn_flags & APPEND) && 240 (mode & (FWRITE | O_APPEND)) == FWRITE) { 241 error = EPERM; 242 } else { 243 error = (vop_stdopen(v)); 244 } 245 246 return (error); 247 } 248 249 /* --------------------------------------------------------------------- */ 250 251 static int 252 tmpfs_close(struct vop_close_args *v) 253 { 254 struct vnode *vp = v->a_vp; 255 struct tmpfs_node *node; 256 int error; 257 258 node = VP_TO_TMPFS_NODE(vp); 259 260 if (node->tn_links > 0) { 261 /* 262 * Update node times. No need to do it if the node has 263 * been deleted, because it will vanish after we return. 264 */ 265 tmpfs_update(vp); 266 } 267 268 error = vop_stdclose(v); 269 270 return (error); 271 } 272 273 /* --------------------------------------------------------------------- */ 274 275 int 276 tmpfs_access(struct vop_access_args *v) 277 { 278 struct vnode *vp = v->a_vp; 279 int error; 280 struct tmpfs_node *node; 281 282 node = VP_TO_TMPFS_NODE(vp); 283 284 switch (vp->v_type) { 285 case VDIR: 286 /* FALLTHROUGH */ 287 case VLNK: 288 /* FALLTHROUGH */ 289 case VREG: 290 if ((v->a_mode & VWRITE) && 291 (vp->v_mount->mnt_flag & MNT_RDONLY)) { 292 error = EROFS; 293 goto out; 294 } 295 break; 296 297 case VBLK: 298 /* FALLTHROUGH */ 299 case VCHR: 300 /* FALLTHROUGH */ 301 case VSOCK: 302 /* FALLTHROUGH */ 303 case VFIFO: 304 break; 305 306 default: 307 error = EINVAL; 308 goto out; 309 } 310 311 if ((v->a_mode & VWRITE) && (node->tn_flags & IMMUTABLE)) { 312 error = EPERM; 313 goto out; 314 } 315 316 error = vop_helper_access(v, node->tn_uid, node->tn_gid, 317 node->tn_mode, 0); 318 out: 319 return error; 320 } 321 322 /* --------------------------------------------------------------------- */ 323 324 int 325 tmpfs_getattr(struct vop_getattr_args *v) 326 { 327 struct vnode *vp = v->a_vp; 328 struct vattr *vap = v->a_vap; 329 struct tmpfs_node *node; 330 331 node = VP_TO_TMPFS_NODE(vp); 332 333 tmpfs_update(vp); 334 335 TMPFS_NODE_LOCK_SH(node); 336 vap->va_type = vp->v_type; 337 vap->va_mode = node->tn_mode; 338 vap->va_nlink = node->tn_links; 339 vap->va_uid = node->tn_uid; 340 vap->va_gid = node->tn_gid; 341 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; 342 vap->va_fileid = node->tn_id; 343 vap->va_size = node->tn_size; 344 vap->va_blocksize = PAGE_SIZE; 345 vap->va_atime.tv_sec = node->tn_atime; 346 vap->va_atime.tv_nsec = node->tn_atimensec; 347 vap->va_mtime.tv_sec = node->tn_mtime; 348 vap->va_mtime.tv_nsec = node->tn_mtimensec; 349 vap->va_ctime.tv_sec = node->tn_ctime; 350 vap->va_ctime.tv_nsec = node->tn_ctimensec; 351 vap->va_gen = node->tn_gen; 352 vap->va_flags = node->tn_flags; 353 if (vp->v_type == VBLK || vp->v_type == VCHR) { 354 vap->va_rmajor = umajor(node->tn_rdev); 355 vap->va_rminor = uminor(node->tn_rdev); 356 } 357 vap->va_bytes = round_page(node->tn_size); 358 vap->va_filerev = 0; 359 TMPFS_NODE_UNLOCK(node); 360 361 return 0; 362 } 363 364 /* --------------------------------------------------------------------- */ 365 366 int 367 tmpfs_setattr(struct vop_setattr_args *v) 368 { 369 struct vnode *vp = v->a_vp; 370 struct vattr *vap = v->a_vap; 371 struct ucred *cred = v->a_cred; 372 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp); 373 int error = 0; 374 int kflags = 0; 375 376 TMPFS_NODE_LOCK(node); 377 if (error == 0 && (vap->va_flags != VNOVAL)) { 378 error = tmpfs_chflags(vp, vap->va_flags, cred); 379 kflags |= NOTE_ATTRIB; 380 } 381 382 if (error == 0 && (vap->va_size != VNOVAL)) { 383 if (vap->va_size > node->tn_size) 384 kflags |= NOTE_WRITE | NOTE_EXTEND; 385 else 386 kflags |= NOTE_WRITE; 387 error = tmpfs_chsize(vp, vap->va_size, cred); 388 } 389 390 if (error == 0 && (vap->va_uid != (uid_t)VNOVAL || 391 vap->va_gid != (gid_t)VNOVAL)) { 392 error = tmpfs_chown(vp, vap->va_uid, vap->va_gid, cred); 393 kflags |= NOTE_ATTRIB; 394 } 395 396 if (error == 0 && (vap->va_mode != (mode_t)VNOVAL)) { 397 error = tmpfs_chmod(vp, vap->va_mode, cred); 398 kflags |= NOTE_ATTRIB; 399 } 400 401 if (error == 0 && ((vap->va_atime.tv_sec != VNOVAL && 402 vap->va_atime.tv_nsec != VNOVAL) || 403 (vap->va_mtime.tv_sec != VNOVAL && 404 vap->va_mtime.tv_nsec != VNOVAL) )) { 405 error = tmpfs_chtimes(vp, &vap->va_atime, &vap->va_mtime, 406 vap->va_vaflags, cred); 407 kflags |= NOTE_ATTRIB; 408 } 409 410 /* 411 * Update the node times. We give preference to the error codes 412 * generated by this function rather than the ones that may arise 413 * from tmpfs_update. 414 */ 415 tmpfs_update(vp); 416 TMPFS_NODE_UNLOCK(node); 417 tmpfs_knote(vp, kflags); 418 419 return (error); 420 } 421 422 /* --------------------------------------------------------------------- */ 423 424 /* 425 * fsync is usually a NOP, but we must take action when unmounting or 426 * when recycling. 427 */ 428 static int 429 tmpfs_fsync(struct vop_fsync_args *v) 430 { 431 struct tmpfs_node *node; 432 struct vnode *vp = v->a_vp; 433 434 node = VP_TO_TMPFS_NODE(vp); 435 436 tmpfs_update(vp); 437 if (vp->v_type == VREG) { 438 if (vp->v_flag & VRECLAIMED) { 439 if (node->tn_links == 0) 440 tmpfs_truncate(vp, 0); 441 else 442 vfsync(v->a_vp, v->a_waitfor, 1, NULL, NULL); 443 } 444 } 445 return 0; 446 } 447 448 /* --------------------------------------------------------------------- */ 449 450 static int 451 tmpfs_read (struct vop_read_args *ap) 452 { 453 struct buf *bp; 454 struct vnode *vp = ap->a_vp; 455 struct uio *uio = ap->a_uio; 456 struct tmpfs_node *node; 457 off_t base_offset; 458 size_t offset; 459 size_t len; 460 size_t resid; 461 int error; 462 463 /* 464 * Check the basics 465 */ 466 if (uio->uio_offset < 0) 467 return (EINVAL); 468 if (vp->v_type != VREG) 469 return (EINVAL); 470 471 /* 472 * Extract node, try to shortcut the operation through 473 * the VM page cache, allowing us to avoid buffer cache 474 * overheads. 475 */ 476 node = VP_TO_TMPFS_NODE(vp); 477 resid = uio->uio_resid; 478 error = vop_helper_read_shortcut(ap); 479 if (error) 480 return error; 481 if (uio->uio_resid == 0) { 482 if (resid) 483 goto finished; 484 return error; 485 } 486 487 /* 488 * Fall-through to our normal read code. 489 */ 490 while (uio->uio_resid > 0 && uio->uio_offset < node->tn_size) { 491 /* 492 * Use buffer cache I/O (via tmpfs_strategy) 493 */ 494 offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64; 495 base_offset = (off_t)uio->uio_offset - offset; 496 bp = getcacheblk(vp, base_offset, TMPFS_BLKSIZE, 0); 497 if (bp == NULL) { 498 error = bread(vp, base_offset, TMPFS_BLKSIZE, &bp); 499 if (error) { 500 brelse(bp); 501 kprintf("tmpfs_read bread error %d\n", error); 502 break; 503 } 504 505 /* 506 * tmpfs pretty much fiddles directly with the VM 507 * system, don't let it exhaust it or we won't play 508 * nice with other processes. 509 * 510 * Only do this if the VOP is coming from a normal 511 * read/write. The VM system handles the case for 512 * UIO_NOCOPY. 513 */ 514 if (uio->uio_segflg != UIO_NOCOPY) 515 vm_wait_nominal(); 516 } 517 bp->b_flags |= B_CLUSTEROK; 518 519 /* 520 * Figure out how many bytes we can actually copy this loop. 521 */ 522 len = TMPFS_BLKSIZE - offset; 523 if (len > uio->uio_resid) 524 len = uio->uio_resid; 525 if (len > node->tn_size - uio->uio_offset) 526 len = (size_t)(node->tn_size - uio->uio_offset); 527 528 error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio); 529 bqrelse(bp); 530 if (error) { 531 kprintf("tmpfs_read uiomove error %d\n", error); 532 break; 533 } 534 } 535 536 finished: 537 if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) { 538 TMPFS_NODE_LOCK(node); 539 node->tn_status |= TMPFS_NODE_ACCESSED; 540 TMPFS_NODE_UNLOCK(node); 541 } 542 return (error); 543 } 544 545 static int 546 tmpfs_write (struct vop_write_args *ap) 547 { 548 struct buf *bp; 549 struct vnode *vp = ap->a_vp; 550 struct uio *uio = ap->a_uio; 551 struct thread *td = uio->uio_td; 552 struct tmpfs_node *node; 553 boolean_t extended; 554 off_t oldsize; 555 int error; 556 off_t base_offset; 557 size_t offset; 558 size_t len; 559 struct rlimit limit; 560 int trivial = 0; 561 int kflags = 0; 562 int seqcount; 563 564 error = 0; 565 if (uio->uio_resid == 0) { 566 return error; 567 } 568 569 node = VP_TO_TMPFS_NODE(vp); 570 571 if (vp->v_type != VREG) 572 return (EINVAL); 573 seqcount = ap->a_ioflag >> 16; 574 575 TMPFS_NODE_LOCK(node); 576 577 oldsize = node->tn_size; 578 if (ap->a_ioflag & IO_APPEND) 579 uio->uio_offset = node->tn_size; 580 581 /* 582 * Check for illegal write offsets. 583 */ 584 if (uio->uio_offset + uio->uio_resid > 585 VFS_TO_TMPFS(vp->v_mount)->tm_maxfilesize) { 586 error = EFBIG; 587 goto done; 588 } 589 590 /* 591 * NOTE: Ignore if UIO does not come from a user thread (e.g. VN). 592 */ 593 if (vp->v_type == VREG && td != NULL && td->td_lwp != NULL) { 594 error = kern_getrlimit(RLIMIT_FSIZE, &limit); 595 if (error) 596 goto done; 597 if (uio->uio_offset + uio->uio_resid > limit.rlim_cur) { 598 ksignal(td->td_proc, SIGXFSZ); 599 error = EFBIG; 600 goto done; 601 } 602 } 603 604 /* 605 * Extend the file's size if necessary 606 */ 607 extended = ((uio->uio_offset + uio->uio_resid) > node->tn_size); 608 609 while (uio->uio_resid > 0) { 610 /* 611 * Don't completely blow out running buffer I/O 612 * when being hit from the pageout daemon. 613 */ 614 if (uio->uio_segflg == UIO_NOCOPY && 615 (ap->a_ioflag & IO_RECURSE) == 0) { 616 bwillwrite(TMPFS_BLKSIZE); 617 } 618 619 /* 620 * Use buffer cache I/O (via tmpfs_strategy) 621 */ 622 offset = (size_t)uio->uio_offset & TMPFS_BLKMASK64; 623 base_offset = (off_t)uio->uio_offset - offset; 624 len = TMPFS_BLKSIZE - offset; 625 if (len > uio->uio_resid) 626 len = uio->uio_resid; 627 628 if ((uio->uio_offset + len) > node->tn_size) { 629 trivial = (uio->uio_offset <= node->tn_size); 630 error = tmpfs_reg_resize(vp, uio->uio_offset + len, 631 trivial); 632 if (error) 633 break; 634 } 635 636 /* 637 * Read to fill in any gaps. Theoretically we could 638 * optimize this if the write covers the entire buffer 639 * and is not a UIO_NOCOPY write, however this can lead 640 * to a security violation exposing random kernel memory 641 * (whatever junk was in the backing VM pages before). 642 * 643 * So just use bread() to do the right thing. 644 */ 645 error = bread(vp, base_offset, TMPFS_BLKSIZE, &bp); 646 error = uiomovebp(bp, (char *)bp->b_data + offset, len, uio); 647 if (error) { 648 kprintf("tmpfs_write uiomove error %d\n", error); 649 brelse(bp); 650 break; 651 } 652 653 if (uio->uio_offset > node->tn_size) { 654 node->tn_size = uio->uio_offset; 655 kflags |= NOTE_EXTEND; 656 } 657 kflags |= NOTE_WRITE; 658 659 /* 660 * Always try to flush the page in the UIO_NOCOPY case. This 661 * can come from the pageout daemon or during vnode eviction. 662 * It is not necessarily going to be marked IO_ASYNC/IO_SYNC. 663 * 664 * For the normal case we buwrite(), dirtying the underlying 665 * VM pages instead of dirtying the buffer and releasing the 666 * buffer as a clean buffer. This allows tmpfs to use 667 * essentially all available memory to cache file data. 668 * If we used bdwrite() the buffer cache would wind up 669 * flushing the data to swap too quickly. 670 * 671 * But because tmpfs can seriously load the VM system we 672 * fall-back to using bdwrite() when free memory starts 673 * to get low. This shifts the load away from the VM system 674 * and makes tmpfs act more like a normal filesystem with 675 * regards to disk activity. 676 * 677 * tmpfs pretty much fiddles directly with the VM 678 * system, don't let it exhaust it or we won't play 679 * nice with other processes. Only do this if the 680 * VOP is coming from a normal read/write. The VM system 681 * handles the case for UIO_NOCOPY. 682 */ 683 bp->b_flags |= B_CLUSTEROK; 684 if (uio->uio_segflg == UIO_NOCOPY) { 685 /* 686 * Flush from the pageout daemon, deal with 687 * potentially very heavy tmpfs write activity 688 * causing long stalls in the pageout daemon 689 * before pages get to free/cache. 690 * 691 * (a) Under severe pressure setting B_DIRECT will 692 * cause a buffer release to try to free the 693 * underlying pages. 694 * 695 * (b) Under modest memory pressure the B_RELBUF 696 * alone is sufficient to get the pages moved 697 * to the cache. We could also force this by 698 * setting B_NOTMETA but that might have other 699 * unintended side-effects (e.g. setting 700 * PG_NOTMETA on the VM page). 701 * 702 * Hopefully this will unblock the VM system more 703 * quickly under extreme tmpfs write load. 704 */ 705 if (vm_page_count_min(vm_page_free_hysteresis)) 706 bp->b_flags |= B_DIRECT; 707 bp->b_flags |= B_AGE | B_RELBUF; 708 bp->b_act_count = 0; /* buffer->deactivate pgs */ 709 cluster_awrite(bp); 710 } else if (vm_page_count_target()) { 711 /* 712 * Normal (userland) write but we are low on memory, 713 * run the buffer the buffer cache. 714 */ 715 bp->b_act_count = 0; /* buffer->deactivate pgs */ 716 bdwrite(bp); 717 } else { 718 /* 719 * Otherwise run the buffer directly through to the 720 * backing VM store. 721 */ 722 buwrite(bp); 723 /*vm_wait_nominal();*/ 724 } 725 726 if (bp->b_error) { 727 kprintf("tmpfs_write bwrite error %d\n", bp->b_error); 728 break; 729 } 730 } 731 732 if (error) { 733 if (extended) { 734 (void)tmpfs_reg_resize(vp, oldsize, trivial); 735 kflags &= ~NOTE_EXTEND; 736 } 737 goto done; 738 } 739 740 /* 741 * Currently we don't set the mtime on files modified via mmap() 742 * because we can't tell the difference between those modifications 743 * and an attempt by the pageout daemon to flush tmpfs pages to 744 * swap. 745 * 746 * This is because in order to defer flushes as long as possible 747 * buwrite() works by marking the underlying VM pages dirty in 748 * order to be able to dispose of the buffer cache buffer without 749 * flushing it. 750 */ 751 if (uio->uio_segflg != UIO_NOCOPY) 752 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_MODIFIED; 753 if (extended) 754 node->tn_status |= TMPFS_NODE_CHANGED; 755 756 if (node->tn_mode & (S_ISUID | S_ISGID)) { 757 if (priv_check_cred(ap->a_cred, PRIV_VFS_RETAINSUGID, 0)) 758 node->tn_mode &= ~(S_ISUID | S_ISGID); 759 } 760 done: 761 TMPFS_NODE_UNLOCK(node); 762 if (kflags) 763 tmpfs_knote(vp, kflags); 764 765 return(error); 766 } 767 768 static int 769 tmpfs_advlock (struct vop_advlock_args *ap) 770 { 771 struct tmpfs_node *node; 772 struct vnode *vp = ap->a_vp; 773 int error; 774 775 node = VP_TO_TMPFS_NODE(vp); 776 error = (lf_advlock(ap, &node->tn_advlock, node->tn_size)); 777 778 return (error); 779 } 780 781 /* 782 * The strategy function is typically only called when memory pressure 783 * forces the system to attempt to pageout pages. It can also be called 784 * by [n]vtruncbuf() when a truncation cuts a page in half. Normal write 785 * operations 786 */ 787 static int 788 tmpfs_strategy(struct vop_strategy_args *ap) 789 { 790 struct bio *bio = ap->a_bio; 791 struct bio *nbio; 792 struct buf *bp = bio->bio_buf; 793 struct vnode *vp = ap->a_vp; 794 struct tmpfs_node *node; 795 vm_object_t uobj; 796 vm_page_t m; 797 int i; 798 799 if (vp->v_type != VREG) { 800 bp->b_resid = bp->b_bcount; 801 bp->b_flags |= B_ERROR | B_INVAL; 802 bp->b_error = EINVAL; 803 biodone(bio); 804 return(0); 805 } 806 807 node = VP_TO_TMPFS_NODE(vp); 808 809 uobj = node->tn_reg.tn_aobj; 810 811 /* 812 * Don't bother flushing to swap if there is no swap, just 813 * ensure that the pages are marked as needing a commit (still). 814 */ 815 if (bp->b_cmd == BUF_CMD_WRITE && vm_swap_size == 0) { 816 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 817 m = bp->b_xio.xio_pages[i]; 818 vm_page_need_commit(m); 819 } 820 bp->b_resid = 0; 821 bp->b_error = 0; 822 biodone(bio); 823 } else { 824 nbio = push_bio(bio); 825 nbio->bio_done = tmpfs_strategy_done; 826 nbio->bio_offset = bio->bio_offset; 827 swap_pager_strategy(uobj, nbio); 828 } 829 return 0; 830 } 831 832 /* 833 * If we were unable to commit the pages to swap make sure they are marked 834 * as needing a commit (again). If we were, clear the flag to allow the 835 * pages to be freed. 836 */ 837 static void 838 tmpfs_strategy_done(struct bio *bio) 839 { 840 struct buf *bp; 841 vm_page_t m; 842 int i; 843 844 bp = bio->bio_buf; 845 846 if (bp->b_flags & B_ERROR) { 847 bp->b_flags &= ~B_ERROR; 848 bp->b_error = 0; 849 bp->b_resid = 0; 850 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 851 m = bp->b_xio.xio_pages[i]; 852 vm_page_need_commit(m); 853 } 854 } else { 855 for (i = 0; i < bp->b_xio.xio_npages; ++i) { 856 m = bp->b_xio.xio_pages[i]; 857 vm_page_clear_commit(m); 858 } 859 } 860 bio = pop_bio(bio); 861 biodone(bio); 862 } 863 864 static int 865 tmpfs_bmap(struct vop_bmap_args *ap) 866 { 867 if (ap->a_doffsetp != NULL) 868 *ap->a_doffsetp = ap->a_loffset; 869 if (ap->a_runp != NULL) 870 *ap->a_runp = 0; 871 if (ap->a_runb != NULL) 872 *ap->a_runb = 0; 873 874 return 0; 875 } 876 877 /* --------------------------------------------------------------------- */ 878 879 static int 880 tmpfs_nremove(struct vop_nremove_args *v) 881 { 882 struct vnode *dvp = v->a_dvp; 883 struct namecache *ncp = v->a_nch->ncp; 884 struct vnode *vp; 885 int error; 886 struct tmpfs_dirent *de; 887 struct tmpfs_mount *tmp; 888 struct tmpfs_node *dnode; 889 struct tmpfs_node *node; 890 struct mount *mp; 891 892 mp = dvp->v_mount; 893 894 /* 895 * We have to acquire the vp from v->a_nch because we will likely 896 * unresolve the namecache entry, and a vrele/vput is needed to 897 * trigger the tmpfs_inactive/tmpfs_reclaim sequence. 898 * 899 * We have to use vget to clear any inactive state on the vnode, 900 * otherwise the vnode may remain inactive and thus tmpfs_inactive 901 * will not get called when we release it. 902 */ 903 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp); 904 KKASSERT(vp->v_mount == dvp->v_mount); 905 KKASSERT(error == 0); 906 vn_unlock(vp); 907 908 if (vp->v_type == VDIR) { 909 error = EISDIR; 910 goto out2; 911 } 912 913 dnode = VP_TO_TMPFS_DIR(dvp); 914 node = VP_TO_TMPFS_NODE(vp); 915 tmp = VFS_TO_TMPFS(vp->v_mount); 916 917 TMPFS_NODE_LOCK(dnode); 918 de = tmpfs_dir_lookup(dnode, node, ncp); 919 if (de == NULL) { 920 error = ENOENT; 921 goto out; 922 } 923 924 /* Files marked as immutable or append-only cannot be deleted. */ 925 if ((node->tn_flags & (IMMUTABLE | APPEND | NOUNLINK)) || 926 (dnode->tn_flags & APPEND)) { 927 error = EPERM; 928 goto out; 929 } 930 931 /* Remove the entry from the directory; as it is a file, we do not 932 * have to change the number of hard links of the directory. */ 933 tmpfs_dir_detach(dnode, de); 934 935 /* Free the directory entry we just deleted. Note that the node 936 * referred by it will not be removed until the vnode is really 937 * reclaimed. */ 938 tmpfs_free_dirent(tmp, de); 939 940 if (node->tn_links > 0) { 941 TMPFS_NODE_LOCK(node); 942 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | \ 943 TMPFS_NODE_MODIFIED; 944 TMPFS_NODE_UNLOCK(node); 945 } 946 947 cache_unlink(v->a_nch); 948 tmpfs_knote(vp, NOTE_DELETE); 949 error = 0; 950 951 out: 952 TMPFS_NODE_UNLOCK(dnode); 953 if (error == 0) 954 tmpfs_knote(dvp, NOTE_WRITE); 955 out2: 956 vrele(vp); 957 958 return error; 959 } 960 961 /* --------------------------------------------------------------------- */ 962 963 static int 964 tmpfs_nlink(struct vop_nlink_args *v) 965 { 966 struct vnode *dvp = v->a_dvp; 967 struct vnode *vp = v->a_vp; 968 struct namecache *ncp = v->a_nch->ncp; 969 struct tmpfs_dirent *de; 970 struct tmpfs_node *node; 971 struct tmpfs_node *dnode; 972 struct mount *mp; 973 int error; 974 975 mp = dvp->v_mount; 976 977 KKASSERT(dvp != vp); /* XXX When can this be false? */ 978 979 node = VP_TO_TMPFS_NODE(vp); 980 dnode = VP_TO_TMPFS_NODE(dvp); 981 TMPFS_NODE_LOCK(dnode); 982 983 /* XXX: Why aren't the following two tests done by the caller? */ 984 985 /* Hard links of directories are forbidden. */ 986 if (vp->v_type == VDIR) { 987 error = EPERM; 988 goto out; 989 } 990 991 /* Cannot create cross-device links. */ 992 if (dvp->v_mount != vp->v_mount) { 993 error = EXDEV; 994 goto out; 995 } 996 997 /* Ensure that we do not overflow the maximum number of links imposed 998 * by the system. */ 999 KKASSERT(node->tn_links <= LINK_MAX); 1000 if (node->tn_links >= LINK_MAX) { 1001 error = EMLINK; 1002 goto out; 1003 } 1004 1005 /* We cannot create links of files marked immutable or append-only. */ 1006 if (node->tn_flags & (IMMUTABLE | APPEND)) { 1007 error = EPERM; 1008 goto out; 1009 } 1010 1011 /* Allocate a new directory entry to represent the node. */ 1012 error = tmpfs_alloc_dirent(VFS_TO_TMPFS(vp->v_mount), node, 1013 ncp->nc_name, ncp->nc_nlen, &de); 1014 if (error != 0) 1015 goto out; 1016 1017 /* Insert the new directory entry into the appropriate directory. */ 1018 tmpfs_dir_attach(dnode, de); 1019 1020 /* vp link count has changed, so update node times. */ 1021 1022 TMPFS_NODE_LOCK(node); 1023 node->tn_status |= TMPFS_NODE_CHANGED; 1024 TMPFS_NODE_UNLOCK(node); 1025 tmpfs_update(vp); 1026 1027 tmpfs_knote(vp, NOTE_LINK); 1028 cache_setunresolved(v->a_nch); 1029 cache_setvp(v->a_nch, vp); 1030 error = 0; 1031 1032 out: 1033 TMPFS_NODE_UNLOCK(dnode); 1034 if (error == 0) 1035 tmpfs_knote(dvp, NOTE_WRITE); 1036 return error; 1037 } 1038 1039 /* --------------------------------------------------------------------- */ 1040 1041 static int 1042 tmpfs_nrename(struct vop_nrename_args *v) 1043 { 1044 struct vnode *fdvp = v->a_fdvp; 1045 struct namecache *fncp = v->a_fnch->ncp; 1046 struct vnode *fvp = fncp->nc_vp; 1047 struct vnode *tdvp = v->a_tdvp; 1048 struct namecache *tncp = v->a_tnch->ncp; 1049 struct vnode *tvp; 1050 struct tmpfs_dirent *de, *tde; 1051 struct tmpfs_mount *tmp; 1052 struct tmpfs_node *fdnode; 1053 struct tmpfs_node *fnode; 1054 struct tmpfs_node *tnode; 1055 struct tmpfs_node *tdnode; 1056 struct mount *mp; 1057 char *newname; 1058 char *oldname; 1059 int error; 1060 1061 mp = fdvp->v_mount; 1062 KKASSERT(fdvp->v_mount == fvp->v_mount); 1063 1064 /* 1065 * Because tvp can get overwritten we have to vget it instead of 1066 * just vref or use it, otherwise it's VINACTIVE flag may not get 1067 * cleared and the node won't get destroyed. 1068 */ 1069 error = cache_vget(v->a_tnch, v->a_cred, LK_SHARED, &tvp); 1070 if (error == 0) { 1071 tnode = VP_TO_TMPFS_NODE(tvp); 1072 vn_unlock(tvp); 1073 } else { 1074 tnode = NULL; 1075 } 1076 1077 /* Disallow cross-device renames. 1078 * XXX Why isn't this done by the caller? */ 1079 if (fvp->v_mount != tdvp->v_mount || 1080 (tvp != NULL && fvp->v_mount != tvp->v_mount)) { 1081 error = EXDEV; 1082 goto out; 1083 } 1084 1085 tmp = VFS_TO_TMPFS(tdvp->v_mount); 1086 tdnode = VP_TO_TMPFS_DIR(tdvp); 1087 1088 /* If source and target are the same file, there is nothing to do. */ 1089 if (fvp == tvp) { 1090 error = 0; 1091 goto out; 1092 } 1093 1094 fdnode = VP_TO_TMPFS_DIR(fdvp); 1095 fnode = VP_TO_TMPFS_NODE(fvp); 1096 TMPFS_NODE_LOCK(fdnode); 1097 de = tmpfs_dir_lookup(fdnode, fnode, fncp); 1098 TMPFS_NODE_UNLOCK(fdnode); /* XXX depend on namecache lock */ 1099 1100 /* Avoid manipulating '.' and '..' entries. */ 1101 if (de == NULL) { 1102 error = ENOENT; 1103 goto out_locked; 1104 } 1105 KKASSERT(de->td_node == fnode); 1106 1107 /* 1108 * If replacing an entry in the target directory and that entry 1109 * is a directory, it must be empty. 1110 * 1111 * Kern_rename gurantees the destination to be a directory 1112 * if the source is one (it does?). 1113 */ 1114 if (tvp != NULL) { 1115 KKASSERT(tnode != NULL); 1116 1117 if ((tnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 1118 (tdnode->tn_flags & (APPEND | IMMUTABLE))) { 1119 error = EPERM; 1120 goto out_locked; 1121 } 1122 1123 if (fnode->tn_type == VDIR && tnode->tn_type == VDIR) { 1124 if (tnode->tn_size > 0) { 1125 error = ENOTEMPTY; 1126 goto out_locked; 1127 } 1128 } else if (fnode->tn_type == VDIR && tnode->tn_type != VDIR) { 1129 error = ENOTDIR; 1130 goto out_locked; 1131 } else if (fnode->tn_type != VDIR && tnode->tn_type == VDIR) { 1132 error = EISDIR; 1133 goto out_locked; 1134 } else { 1135 KKASSERT(fnode->tn_type != VDIR && 1136 tnode->tn_type != VDIR); 1137 } 1138 } 1139 1140 if ((fnode->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) || 1141 (fdnode->tn_flags & (APPEND | IMMUTABLE))) { 1142 error = EPERM; 1143 goto out_locked; 1144 } 1145 1146 /* 1147 * Ensure that we have enough memory to hold the new name, if it 1148 * has to be changed. 1149 */ 1150 if (fncp->nc_nlen != tncp->nc_nlen || 1151 bcmp(fncp->nc_name, tncp->nc_name, fncp->nc_nlen) != 0) { 1152 newname = kmalloc(tncp->nc_nlen + 1, tmp->tm_name_zone, 1153 M_WAITOK | M_NULLOK); 1154 if (newname == NULL) { 1155 error = ENOSPC; 1156 goto out_locked; 1157 } 1158 bcopy(tncp->nc_name, newname, tncp->nc_nlen); 1159 newname[tncp->nc_nlen] = '\0'; 1160 } else { 1161 newname = NULL; 1162 } 1163 1164 /* 1165 * Unlink entry from source directory. Note that the kernel has 1166 * already checked for illegal recursion cases (renaming a directory 1167 * into a subdirectory of itself). 1168 */ 1169 if (fdnode != tdnode) { 1170 tmpfs_dir_detach(fdnode, de); 1171 } else { 1172 /* XXX depend on namecache lock */ 1173 TMPFS_NODE_LOCK(fdnode); 1174 KKASSERT(de == tmpfs_dir_lookup(fdnode, fnode, fncp)); 1175 RB_REMOVE(tmpfs_dirtree, &fdnode->tn_dir.tn_dirtree, de); 1176 RB_REMOVE(tmpfs_dirtree_cookie, 1177 &fdnode->tn_dir.tn_cookietree, de); 1178 TMPFS_NODE_UNLOCK(fdnode); 1179 } 1180 1181 /* 1182 * Handle any name change. Swap with newname, we will 1183 * deallocate it at the end. 1184 */ 1185 if (newname != NULL) { 1186 #if 0 1187 TMPFS_NODE_LOCK(fnode); 1188 fnode->tn_status |= TMPFS_NODE_CHANGED; 1189 TMPFS_NODE_UNLOCK(fnode); 1190 #endif 1191 oldname = de->td_name; 1192 de->td_name = newname; 1193 de->td_namelen = (uint16_t)tncp->nc_nlen; 1194 newname = oldname; 1195 } 1196 1197 /* 1198 * If we are overwriting an entry, we have to remove the old one 1199 * from the target directory. 1200 */ 1201 if (tvp != NULL) { 1202 /* Remove the old entry from the target directory. */ 1203 TMPFS_NODE_LOCK(tdnode); 1204 tde = tmpfs_dir_lookup(tdnode, tnode, tncp); 1205 tmpfs_dir_detach(tdnode, tde); 1206 TMPFS_NODE_UNLOCK(tdnode); 1207 tmpfs_knote(tdnode->tn_vnode, NOTE_DELETE); 1208 1209 /* 1210 * Free the directory entry we just deleted. Note that the 1211 * node referred by it will not be removed until the vnode is 1212 * really reclaimed. 1213 */ 1214 tmpfs_free_dirent(VFS_TO_TMPFS(tvp->v_mount), tde); 1215 /*cache_inval_vp(tvp, CINV_DESTROY);*/ 1216 } 1217 1218 /* 1219 * Link entry to target directory. If the entry 1220 * represents a directory move the parent linkage 1221 * as well. 1222 */ 1223 if (fdnode != tdnode) { 1224 if (de->td_node->tn_type == VDIR) { 1225 TMPFS_VALIDATE_DIR(fnode); 1226 } 1227 tmpfs_dir_attach(tdnode, de); 1228 } else { 1229 TMPFS_NODE_LOCK(tdnode); 1230 tdnode->tn_status |= TMPFS_NODE_MODIFIED; 1231 RB_INSERT(tmpfs_dirtree, &tdnode->tn_dir.tn_dirtree, de); 1232 RB_INSERT(tmpfs_dirtree_cookie, 1233 &tdnode->tn_dir.tn_cookietree, de); 1234 TMPFS_NODE_UNLOCK(tdnode); 1235 } 1236 1237 /* 1238 * Finish up 1239 */ 1240 if (newname) { 1241 kfree(newname, tmp->tm_name_zone); 1242 newname = NULL; 1243 } 1244 cache_rename(v->a_fnch, v->a_tnch); 1245 tmpfs_knote(v->a_fdvp, NOTE_WRITE); 1246 tmpfs_knote(v->a_tdvp, NOTE_WRITE); 1247 if (fnode->tn_vnode) 1248 tmpfs_knote(fnode->tn_vnode, NOTE_RENAME); 1249 error = 0; 1250 1251 out_locked: 1252 ; 1253 out: 1254 if (tvp) 1255 vrele(tvp); 1256 return error; 1257 } 1258 1259 /* --------------------------------------------------------------------- */ 1260 1261 static int 1262 tmpfs_nmkdir(struct vop_nmkdir_args *v) 1263 { 1264 struct vnode *dvp = v->a_dvp; 1265 struct vnode **vpp = v->a_vpp; 1266 struct namecache *ncp = v->a_nch->ncp; 1267 struct vattr *vap = v->a_vap; 1268 struct ucred *cred = v->a_cred; 1269 struct mount *mp; 1270 int error; 1271 1272 mp = dvp->v_mount; 1273 1274 KKASSERT(vap->va_type == VDIR); 1275 1276 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, NULL); 1277 if (error == 0) { 1278 cache_setunresolved(v->a_nch); 1279 cache_setvp(v->a_nch, *vpp); 1280 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK); 1281 } 1282 return error; 1283 } 1284 1285 /* --------------------------------------------------------------------- */ 1286 1287 static int 1288 tmpfs_nrmdir(struct vop_nrmdir_args *v) 1289 { 1290 struct vnode *dvp = v->a_dvp; 1291 struct namecache *ncp = v->a_nch->ncp; 1292 struct vnode *vp; 1293 struct tmpfs_dirent *de; 1294 struct tmpfs_mount *tmp; 1295 struct tmpfs_node *dnode; 1296 struct tmpfs_node *node; 1297 struct mount *mp; 1298 int error; 1299 1300 mp = dvp->v_mount; 1301 1302 /* 1303 * We have to acquire the vp from v->a_nch because we will likely 1304 * unresolve the namecache entry, and a vrele/vput is needed to 1305 * trigger the tmpfs_inactive/tmpfs_reclaim sequence. 1306 * 1307 * We have to use vget to clear any inactive state on the vnode, 1308 * otherwise the vnode may remain inactive and thus tmpfs_inactive 1309 * will not get called when we release it. 1310 */ 1311 error = cache_vget(v->a_nch, v->a_cred, LK_SHARED, &vp); 1312 KKASSERT(error == 0); 1313 vn_unlock(vp); 1314 1315 /* 1316 * Prevalidate so we don't hit an assertion later 1317 */ 1318 if (vp->v_type != VDIR) { 1319 error = ENOTDIR; 1320 goto out; 1321 } 1322 1323 tmp = VFS_TO_TMPFS(dvp->v_mount); 1324 dnode = VP_TO_TMPFS_DIR(dvp); 1325 node = VP_TO_TMPFS_DIR(vp); 1326 1327 /* 1328 * Directories with more than two entries ('.' and '..') cannot 1329 * be removed. 1330 */ 1331 if (node->tn_size > 0) { 1332 error = ENOTEMPTY; 1333 goto out; 1334 } 1335 1336 if ((dnode->tn_flags & APPEND) 1337 || (node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND))) { 1338 error = EPERM; 1339 goto out; 1340 } 1341 1342 /* 1343 * This invariant holds only if we are not trying to 1344 * remove "..". We checked for that above so this is safe now. 1345 */ 1346 KKASSERT(node->tn_dir.tn_parent == dnode); 1347 1348 /* 1349 * Get the directory entry associated with node (vp). This 1350 * was filled by tmpfs_lookup while looking up the entry. 1351 */ 1352 TMPFS_NODE_LOCK(dnode); 1353 de = tmpfs_dir_lookup(dnode, node, ncp); 1354 KKASSERT(TMPFS_DIRENT_MATCHES(de, ncp->nc_name, ncp->nc_nlen)); 1355 1356 /* Check flags to see if we are allowed to remove the directory. */ 1357 if ((dnode->tn_flags & APPEND) || 1358 node->tn_flags & (NOUNLINK | IMMUTABLE | APPEND)) { 1359 error = EPERM; 1360 TMPFS_NODE_UNLOCK(dnode); 1361 goto out; 1362 } 1363 1364 /* Detach the directory entry from the directory (dnode). */ 1365 tmpfs_dir_detach(dnode, de); 1366 TMPFS_NODE_UNLOCK(dnode); 1367 1368 /* No vnode should be allocated for this entry from this point */ 1369 TMPFS_NODE_LOCK(dnode); 1370 TMPFS_ASSERT_ELOCKED(dnode); 1371 TMPFS_NODE_LOCK(node); 1372 TMPFS_ASSERT_ELOCKED(node); 1373 1374 /* 1375 * Must set parent linkage to NULL (tested by ncreate to disallow 1376 * the creation of new files/dirs in a deleted directory) 1377 */ 1378 node->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 1379 TMPFS_NODE_MODIFIED; 1380 1381 dnode->tn_status |= TMPFS_NODE_ACCESSED | TMPFS_NODE_CHANGED | 1382 TMPFS_NODE_MODIFIED; 1383 1384 TMPFS_NODE_UNLOCK(node); 1385 TMPFS_NODE_UNLOCK(dnode); 1386 1387 /* Free the directory entry we just deleted. Note that the node 1388 * referred by it will not be removed until the vnode is really 1389 * reclaimed. */ 1390 tmpfs_free_dirent(tmp, de); 1391 1392 /* Release the deleted vnode (will destroy the node, notify 1393 * interested parties and clean it from the cache). */ 1394 1395 TMPFS_NODE_LOCK(dnode); 1396 dnode->tn_status |= TMPFS_NODE_CHANGED; 1397 TMPFS_NODE_UNLOCK(dnode); 1398 tmpfs_update(dvp); 1399 1400 cache_unlink(v->a_nch); 1401 tmpfs_knote(dvp, NOTE_WRITE | NOTE_LINK); 1402 error = 0; 1403 1404 out: 1405 vrele(vp); 1406 1407 return error; 1408 } 1409 1410 /* --------------------------------------------------------------------- */ 1411 1412 static int 1413 tmpfs_nsymlink(struct vop_nsymlink_args *v) 1414 { 1415 struct vnode *dvp = v->a_dvp; 1416 struct vnode **vpp = v->a_vpp; 1417 struct namecache *ncp = v->a_nch->ncp; 1418 struct vattr *vap = v->a_vap; 1419 struct ucred *cred = v->a_cred; 1420 char *target = v->a_target; 1421 int error; 1422 1423 vap->va_type = VLNK; 1424 error = tmpfs_alloc_file(dvp, vpp, vap, ncp, cred, target); 1425 if (error == 0) { 1426 tmpfs_knote(*vpp, NOTE_WRITE); 1427 cache_setunresolved(v->a_nch); 1428 cache_setvp(v->a_nch, *vpp); 1429 } 1430 return error; 1431 } 1432 1433 /* --------------------------------------------------------------------- */ 1434 1435 static int 1436 tmpfs_readdir(struct vop_readdir_args *v) 1437 { 1438 struct vnode *vp = v->a_vp; 1439 struct uio *uio = v->a_uio; 1440 int *eofflag = v->a_eofflag; 1441 off_t **cookies = v->a_cookies; 1442 int *ncookies = v->a_ncookies; 1443 struct tmpfs_mount *tmp; 1444 int error; 1445 off_t startoff; 1446 off_t cnt = 0; 1447 struct tmpfs_node *node; 1448 1449 /* This operation only makes sense on directory nodes. */ 1450 if (vp->v_type != VDIR) { 1451 return ENOTDIR; 1452 } 1453 1454 tmp = VFS_TO_TMPFS(vp->v_mount); 1455 node = VP_TO_TMPFS_DIR(vp); 1456 startoff = uio->uio_offset; 1457 1458 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOT) { 1459 error = tmpfs_dir_getdotdent(node, uio); 1460 if (error != 0) { 1461 TMPFS_NODE_LOCK_SH(node); 1462 goto outok; 1463 } 1464 cnt++; 1465 } 1466 1467 if (uio->uio_offset == TMPFS_DIRCOOKIE_DOTDOT) { 1468 /* may lock parent, cannot hold node lock */ 1469 error = tmpfs_dir_getdotdotdent(tmp, node, uio); 1470 if (error != 0) { 1471 TMPFS_NODE_LOCK_SH(node); 1472 goto outok; 1473 } 1474 cnt++; 1475 } 1476 1477 TMPFS_NODE_LOCK_SH(node); 1478 error = tmpfs_dir_getdents(node, uio, &cnt); 1479 1480 outok: 1481 KKASSERT(error >= -1); 1482 1483 if (error == -1) 1484 error = 0; 1485 1486 if (eofflag != NULL) 1487 *eofflag = 1488 (error == 0 && uio->uio_offset == TMPFS_DIRCOOKIE_EOF); 1489 1490 /* Update NFS-related variables. */ 1491 if (error == 0 && cookies != NULL && ncookies != NULL) { 1492 off_t i; 1493 off_t off = startoff; 1494 struct tmpfs_dirent *de = NULL; 1495 1496 *ncookies = cnt; 1497 *cookies = kmalloc(cnt * sizeof(off_t), M_TEMP, M_WAITOK); 1498 1499 for (i = 0; i < cnt; i++) { 1500 KKASSERT(off != TMPFS_DIRCOOKIE_EOF); 1501 if (off == TMPFS_DIRCOOKIE_DOT) { 1502 off = TMPFS_DIRCOOKIE_DOTDOT; 1503 } else { 1504 if (off == TMPFS_DIRCOOKIE_DOTDOT) { 1505 de = RB_MIN(tmpfs_dirtree_cookie, 1506 &node->tn_dir.tn_cookietree); 1507 } else if (de != NULL) { 1508 de = RB_NEXT(tmpfs_dirtree_cookie, 1509 &node->tn_dir.tn_cookietree, de); 1510 } else { 1511 de = tmpfs_dir_lookupbycookie(node, 1512 off); 1513 KKASSERT(de != NULL); 1514 de = RB_NEXT(tmpfs_dirtree_cookie, 1515 &node->tn_dir.tn_cookietree, de); 1516 } 1517 if (de == NULL) 1518 off = TMPFS_DIRCOOKIE_EOF; 1519 else 1520 off = tmpfs_dircookie(de); 1521 } 1522 (*cookies)[i] = off; 1523 } 1524 KKASSERT(uio->uio_offset == off); 1525 } 1526 TMPFS_NODE_UNLOCK(node); 1527 1528 if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) { 1529 TMPFS_NODE_LOCK(node); 1530 node->tn_status |= TMPFS_NODE_ACCESSED; 1531 TMPFS_NODE_UNLOCK(node); 1532 } 1533 return error; 1534 } 1535 1536 /* --------------------------------------------------------------------- */ 1537 1538 static int 1539 tmpfs_readlink(struct vop_readlink_args *v) 1540 { 1541 struct vnode *vp = v->a_vp; 1542 struct uio *uio = v->a_uio; 1543 int error; 1544 struct tmpfs_node *node; 1545 1546 KKASSERT(uio->uio_offset == 0); 1547 KKASSERT(vp->v_type == VLNK); 1548 1549 node = VP_TO_TMPFS_NODE(vp); 1550 TMPFS_NODE_LOCK_SH(node); 1551 error = uiomove(node->tn_link, 1552 MIN(node->tn_size, uio->uio_resid), uio); 1553 TMPFS_NODE_UNLOCK(node); 1554 if ((node->tn_status & TMPFS_NODE_ACCESSED) == 0) { 1555 TMPFS_NODE_LOCK(node); 1556 node->tn_status |= TMPFS_NODE_ACCESSED; 1557 TMPFS_NODE_UNLOCK(node); 1558 } 1559 return error; 1560 } 1561 1562 /* --------------------------------------------------------------------- */ 1563 1564 static int 1565 tmpfs_inactive(struct vop_inactive_args *v) 1566 { 1567 struct vnode *vp = v->a_vp; 1568 struct tmpfs_node *node; 1569 struct mount *mp; 1570 1571 mp = vp->v_mount; 1572 lwkt_gettoken(&mp->mnt_token); 1573 node = VP_TO_TMPFS_NODE(vp); 1574 1575 /* 1576 * Degenerate case 1577 */ 1578 if (node == NULL) { 1579 vrecycle(vp); 1580 lwkt_reltoken(&mp->mnt_token); 1581 return(0); 1582 } 1583 1584 /* 1585 * Get rid of unreferenced deleted vnodes sooner rather than 1586 * later so the data memory can be recovered immediately. 1587 * 1588 * We must truncate the vnode to prevent the normal reclamation 1589 * path from flushing the data for the removed file to disk. 1590 */ 1591 TMPFS_NODE_LOCK(node); 1592 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 && 1593 node->tn_links == 0) 1594 { 1595 node->tn_vpstate = TMPFS_VNODE_DOOMED; 1596 TMPFS_NODE_UNLOCK(node); 1597 if (node->tn_type == VREG) 1598 tmpfs_truncate(vp, 0); 1599 vrecycle(vp); 1600 } else { 1601 TMPFS_NODE_UNLOCK(node); 1602 } 1603 lwkt_reltoken(&mp->mnt_token); 1604 1605 return 0; 1606 } 1607 1608 /* --------------------------------------------------------------------- */ 1609 1610 int 1611 tmpfs_reclaim(struct vop_reclaim_args *v) 1612 { 1613 struct vnode *vp = v->a_vp; 1614 struct tmpfs_mount *tmp; 1615 struct tmpfs_node *node; 1616 struct mount *mp; 1617 1618 mp = vp->v_mount; 1619 lwkt_gettoken(&mp->mnt_token); 1620 1621 node = VP_TO_TMPFS_NODE(vp); 1622 tmp = VFS_TO_TMPFS(vp->v_mount); 1623 KKASSERT(mp == tmp->tm_mount); 1624 1625 tmpfs_free_vp(vp); 1626 1627 /* 1628 * If the node referenced by this vnode was deleted by the 1629 * user, we must free its associated data structures now that 1630 * the vnode is being reclaimed. 1631 * 1632 * Directories have an extra link ref. 1633 */ 1634 TMPFS_NODE_LOCK(node); 1635 if ((node->tn_vpstate & TMPFS_VNODE_ALLOCATING) == 0 && 1636 node->tn_links == 0) { 1637 node->tn_vpstate = TMPFS_VNODE_DOOMED; 1638 tmpfs_free_node(tmp, node); 1639 /* eats the lock */ 1640 } else { 1641 TMPFS_NODE_UNLOCK(node); 1642 } 1643 lwkt_reltoken(&mp->mnt_token); 1644 1645 KKASSERT(vp->v_data == NULL); 1646 return 0; 1647 } 1648 1649 /* --------------------------------------------------------------------- */ 1650 1651 static int 1652 tmpfs_mountctl(struct vop_mountctl_args *ap) 1653 { 1654 struct tmpfs_mount *tmp; 1655 struct mount *mp; 1656 int rc; 1657 1658 mp = ap->a_head.a_ops->head.vv_mount; 1659 lwkt_gettoken(&mp->mnt_token); 1660 1661 switch (ap->a_op) { 1662 case (MOUNTCTL_SET_EXPORT): 1663 tmp = (struct tmpfs_mount *) mp->mnt_data; 1664 1665 if (ap->a_ctllen != sizeof(struct export_args)) 1666 rc = (EINVAL); 1667 else 1668 rc = vfs_export(mp, &tmp->tm_export, 1669 (const struct export_args *) ap->a_ctl); 1670 break; 1671 default: 1672 rc = vop_stdmountctl(ap); 1673 break; 1674 } 1675 1676 lwkt_reltoken(&mp->mnt_token); 1677 return (rc); 1678 } 1679 1680 /* --------------------------------------------------------------------- */ 1681 1682 static int 1683 tmpfs_print(struct vop_print_args *v) 1684 { 1685 struct vnode *vp = v->a_vp; 1686 1687 struct tmpfs_node *node; 1688 1689 node = VP_TO_TMPFS_NODE(vp); 1690 1691 kprintf("tag VT_TMPFS, tmpfs_node %p, flags 0x%x, links %d\n", 1692 node, node->tn_flags, node->tn_links); 1693 kprintf("\tmode 0%o, owner %d, group %d, size %ju, status 0x%x\n", 1694 node->tn_mode, node->tn_uid, node->tn_gid, 1695 (uintmax_t)node->tn_size, node->tn_status); 1696 1697 if (vp->v_type == VFIFO) 1698 fifo_printinfo(vp); 1699 1700 kprintf("\n"); 1701 1702 return 0; 1703 } 1704 1705 /* --------------------------------------------------------------------- */ 1706 1707 static int 1708 tmpfs_pathconf(struct vop_pathconf_args *v) 1709 { 1710 int name = v->a_name; 1711 register_t *retval = v->a_retval; 1712 1713 int error; 1714 1715 error = 0; 1716 1717 switch (name) { 1718 case _PC_LINK_MAX: 1719 *retval = LINK_MAX; 1720 break; 1721 1722 case _PC_NAME_MAX: 1723 *retval = NAME_MAX; 1724 break; 1725 1726 case _PC_PATH_MAX: 1727 *retval = PATH_MAX; 1728 break; 1729 1730 case _PC_PIPE_BUF: 1731 *retval = PIPE_BUF; 1732 break; 1733 1734 case _PC_CHOWN_RESTRICTED: 1735 *retval = 1; 1736 break; 1737 1738 case _PC_NO_TRUNC: 1739 *retval = 1; 1740 break; 1741 1742 case _PC_SYNC_IO: 1743 *retval = 1; 1744 break; 1745 1746 case _PC_FILESIZEBITS: 1747 *retval = 0; /* XXX Don't know which value should I return. */ 1748 break; 1749 1750 default: 1751 error = EINVAL; 1752 } 1753 1754 return error; 1755 } 1756 1757 /************************************************************************ 1758 * KQFILTER OPS * 1759 ************************************************************************/ 1760 1761 static void filt_tmpfsdetach(struct knote *kn); 1762 static int filt_tmpfsread(struct knote *kn, long hint); 1763 static int filt_tmpfswrite(struct knote *kn, long hint); 1764 static int filt_tmpfsvnode(struct knote *kn, long hint); 1765 1766 static struct filterops tmpfsread_filtops = 1767 { FILTEROP_ISFD | FILTEROP_MPSAFE, 1768 NULL, filt_tmpfsdetach, filt_tmpfsread }; 1769 static struct filterops tmpfswrite_filtops = 1770 { FILTEROP_ISFD | FILTEROP_MPSAFE, 1771 NULL, filt_tmpfsdetach, filt_tmpfswrite }; 1772 static struct filterops tmpfsvnode_filtops = 1773 { FILTEROP_ISFD | FILTEROP_MPSAFE, 1774 NULL, filt_tmpfsdetach, filt_tmpfsvnode }; 1775 1776 static int 1777 tmpfs_kqfilter (struct vop_kqfilter_args *ap) 1778 { 1779 struct vnode *vp = ap->a_vp; 1780 struct knote *kn = ap->a_kn; 1781 1782 switch (kn->kn_filter) { 1783 case EVFILT_READ: 1784 kn->kn_fop = &tmpfsread_filtops; 1785 break; 1786 case EVFILT_WRITE: 1787 kn->kn_fop = &tmpfswrite_filtops; 1788 break; 1789 case EVFILT_VNODE: 1790 kn->kn_fop = &tmpfsvnode_filtops; 1791 break; 1792 default: 1793 return (EOPNOTSUPP); 1794 } 1795 1796 kn->kn_hook = (caddr_t)vp; 1797 1798 knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 1799 1800 return(0); 1801 } 1802 1803 static void 1804 filt_tmpfsdetach(struct knote *kn) 1805 { 1806 struct vnode *vp = (void *)kn->kn_hook; 1807 1808 knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn); 1809 } 1810 1811 static int 1812 filt_tmpfsread(struct knote *kn, long hint) 1813 { 1814 struct vnode *vp = (void *)kn->kn_hook; 1815 struct tmpfs_node *node = VP_TO_TMPFS_NODE(vp); 1816 off_t off; 1817 1818 if (hint == NOTE_REVOKE) { 1819 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 1820 return(1); 1821 } 1822 1823 /* 1824 * Interlock against MP races when performing this function. 1825 */ 1826 TMPFS_NODE_LOCK_SH(node); 1827 off = node->tn_size - kn->kn_fp->f_offset; 1828 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 1829 if (kn->kn_sfflags & NOTE_OLDAPI) { 1830 TMPFS_NODE_UNLOCK(node); 1831 return(1); 1832 } 1833 if (kn->kn_data == 0) { 1834 kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX; 1835 } 1836 TMPFS_NODE_UNLOCK(node); 1837 return (kn->kn_data != 0); 1838 } 1839 1840 static int 1841 filt_tmpfswrite(struct knote *kn, long hint) 1842 { 1843 if (hint == NOTE_REVOKE) 1844 kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT); 1845 kn->kn_data = 0; 1846 return (1); 1847 } 1848 1849 static int 1850 filt_tmpfsvnode(struct knote *kn, long hint) 1851 { 1852 if (kn->kn_sfflags & hint) 1853 kn->kn_fflags |= hint; 1854 if (hint == NOTE_REVOKE) { 1855 kn->kn_flags |= (EV_EOF | EV_NODATA); 1856 return (1); 1857 } 1858 return (kn->kn_fflags != 0); 1859 } 1860 1861 1862 /* --------------------------------------------------------------------- */ 1863 1864 /* 1865 * vnode operations vector used for files stored in a tmpfs file system. 1866 */ 1867 struct vop_ops tmpfs_vnode_vops = { 1868 .vop_default = vop_defaultop, 1869 .vop_getpages = vop_stdgetpages, 1870 .vop_putpages = vop_stdputpages, 1871 .vop_ncreate = tmpfs_ncreate, 1872 .vop_nresolve = tmpfs_nresolve, 1873 .vop_nlookupdotdot = tmpfs_nlookupdotdot, 1874 .vop_nmknod = tmpfs_nmknod, 1875 .vop_open = tmpfs_open, 1876 .vop_close = tmpfs_close, 1877 .vop_access = tmpfs_access, 1878 .vop_getattr = tmpfs_getattr, 1879 .vop_setattr = tmpfs_setattr, 1880 .vop_read = tmpfs_read, 1881 .vop_write = tmpfs_write, 1882 .vop_fsync = tmpfs_fsync, 1883 .vop_mountctl = tmpfs_mountctl, 1884 .vop_nremove = tmpfs_nremove, 1885 .vop_nlink = tmpfs_nlink, 1886 .vop_nrename = tmpfs_nrename, 1887 .vop_nmkdir = tmpfs_nmkdir, 1888 .vop_nrmdir = tmpfs_nrmdir, 1889 .vop_nsymlink = tmpfs_nsymlink, 1890 .vop_readdir = tmpfs_readdir, 1891 .vop_readlink = tmpfs_readlink, 1892 .vop_inactive = tmpfs_inactive, 1893 .vop_reclaim = tmpfs_reclaim, 1894 .vop_print = tmpfs_print, 1895 .vop_pathconf = tmpfs_pathconf, 1896 .vop_bmap = tmpfs_bmap, 1897 .vop_strategy = tmpfs_strategy, 1898 .vop_advlock = tmpfs_advlock, 1899 .vop_kqfilter = tmpfs_kqfilter 1900 }; 1901